sidedata: enable sidedata computers to optionally rewrite flags

Sidedata computers may want to influence the flags of the revision they touch.
For example, the computer for changelog-based copytracing can add a flag to
signify that this revision might affect copytracing, inversely removing said
flag if the information is no longer applicable.

See inline documentation in `storageutil` for more details.

Differential Revision: https://phab.mercurial-scm.org/D10344

mercurial/cext/revlog.c

@@ -1,3028 +1,3029 b''
 /*
  parsers.c - efficient content parsing
 
  Copyright 2008 Olivia Mackall <[e-mail]> and others
 
  This software may be used and distributed according to the terms of
  the GNU General Public License, incorporated herein by reference.
 */
 
 #define PY_SSIZE_T_CLEAN
 #include <Python.h>
 #include <assert.h>
 #include <ctype.h>
 #include <limits.h>
 #include <stddef.h>
 #include <stdlib.h>
 #include <string.h>
 #include <structmember.h>
 
 #include "bitmanipulation.h"
 #include "charencode.h"
 #include "compat.h"
 #include "revlog.h"
 #include "util.h"
 
 #ifdef IS_PY3K
 /* The mapping of Python types is meant to be temporary to get Python
  * 3 to compile. We should remove this once Python 3 support is fully
  * supported and proper types are used in the extensions themselves. */
 #define PyInt_Check PyLong_Check
 #define PyInt_FromLong PyLong_FromLong
 #define PyInt_FromSsize_t PyLong_FromSsize_t
 #define PyInt_AsLong PyLong_AsLong
 #endif
 
 typedef struct indexObjectStruct indexObject;
 
 typedef struct {
 	int children[16];
 } nodetreenode;
 
 typedef struct {
 	int abi_version;
 	Py_ssize_t (*index_length)(const indexObject *);
 	const char *(*index_node)(indexObject *, Py_ssize_t);
 	int (*index_parents)(PyObject *, int, int *);
 } Revlog_CAPI;
 
 /*
  * A base-16 trie for fast node->rev mapping.
  *
  * Positive value is index of the next node in the trie
  * Negative value is a leaf: -(rev + 2)
  * Zero is empty
  */
 typedef struct {
 	indexObject *index;
 	nodetreenode *nodes;
 	Py_ssize_t nodelen;
 	size_t length;   /* # nodes in use */
 	size_t capacity; /* # nodes allocated */
 	int depth;       /* maximum depth of tree */
 	int splits;      /* # splits performed */
 } nodetree;
 
 typedef struct {
 	PyObject_HEAD /* ; */
 	    nodetree nt;
 } nodetreeObject;
 
 /*
  * This class has two behaviors.
  *
  * When used in a list-like way (with integer keys), we decode an
  * entry in a RevlogNG index file on demand. We have limited support for
  * integer-keyed insert and delete, only at elements right before the
  * end.
  *
  * With string keys, we lazily perform a reverse mapping from node to
  * rev, using a base-16 trie.
  */
 struct indexObjectStruct {
 	PyObject_HEAD
 	    /* Type-specific fields go here. */
 	    PyObject *data;     /* raw bytes of index */
 	Py_ssize_t nodelen;     /* digest size of the hash, 20 for SHA-1 */
 	PyObject *nullentry;    /* fast path for references to null */
 	Py_buffer buf;          /* buffer of data */
 	const char **offsets;   /* populated on demand */
 	Py_ssize_t length;      /* current on-disk number of elements */
 	unsigned new_length;    /* number of added elements */
 	unsigned added_length;  /* space reserved for added elements */
 	char *added;            /* populated on demand */
 	PyObject *headrevs;     /* cache, invalidated on changes */
 	PyObject *filteredrevs; /* filtered revs set */
 	nodetree nt;            /* base-16 trie */
 	int ntinitialized;      /* 0 or 1 */
 	int ntrev;              /* last rev scanned */
 	int ntlookups;          /* # lookups */
 	int ntmisses;           /* # lookups that miss the cache */
 	int inlined;
 	long hdrsize; /* size of index headers. Differs in v1 v.s. v2 format */
 };
 
 static Py_ssize_t index_length(const indexObject *self)
 {
 	return self->length + self->new_length;
 }
 
 static const char nullid[32] = {0};
 static const Py_ssize_t nullrev = -1;
 
 static Py_ssize_t inline_scan(indexObject *self, const char **offsets);
 
 static int index_find_node(indexObject *self, const char *node);
 
 #if LONG_MAX == 0x7fffffffL
 static const char *const v1_tuple_format = PY23("Kiiiiiis#", "Kiiiiiiy#");
 static const char *const v2_tuple_format = PY23("Kiiiiiis#Ki", "Kiiiiiiy#Ki");
 #else
 static const char *const v1_tuple_format = PY23("kiiiiiis#", "kiiiiiiy#");
 static const char *const v2_tuple_format = PY23("kiiiiiis#ki", "kiiiiiiy#ki");
 #endif
 
 /* A RevlogNG v1 index entry is 64 bytes long. */
 static const long v1_hdrsize = 64;
 
 /* A Revlogv2 index entry is 96 bytes long. */
 static const long v2_hdrsize = 96;
 
 static void raise_revlog_error(void)
 {
 	PyObject *mod = NULL, *dict = NULL, *errclass = NULL;
 
 	mod = PyImport_ImportModule("mercurial.error");
 	if (mod == NULL) {
 		goto cleanup;
 	}
 
 	dict = PyModule_GetDict(mod);
 	if (dict == NULL) {
 		goto cleanup;
 	}
 	Py_INCREF(dict);
 
 	errclass = PyDict_GetItemString(dict, "RevlogError");
 	if (errclass == NULL) {
 		PyErr_SetString(PyExc_SystemError,
 		                "could not find RevlogError");
 		goto cleanup;
 	}
 
 	/* value of exception is ignored by callers */
 	PyErr_SetString(errclass, "RevlogError");
 
 cleanup:
 	Py_XDECREF(dict);
 	Py_XDECREF(mod);
 }
 
 /*
  * Return a pointer to the beginning of a RevlogNG record.
  */
 static const char *index_deref(indexObject *self, Py_ssize_t pos)
 {
 	if (pos >= self->length)
 		return self->added + (pos - self->length) * self->hdrsize;
 
 	if (self->inlined && pos > 0) {
 		if (self->offsets == NULL) {
 			Py_ssize_t ret;
 			self->offsets =
 			    PyMem_Malloc(self->length * sizeof(*self->offsets));
 			if (self->offsets == NULL)
 				return (const char *)PyErr_NoMemory();
 			ret = inline_scan(self, self->offsets);
 			if (ret == -1) {
 				return NULL;
 			};
 		}
 		return self->offsets[pos];
 	}
 
 	return (const char *)(self->buf.buf) + pos * self->hdrsize;
 }
 
 /*
  * Get parents of the given rev.
  *
  * The specified rev must be valid and must not be nullrev. A returned
  * parent revision may be nullrev, but is guaranteed to be in valid range.
  */
 static inline int index_get_parents(indexObject *self, Py_ssize_t rev, int *ps,
                                     int maxrev)
 {
 	const char *data = index_deref(self, rev);
 
 	ps[0] = getbe32(data + 24);
 	ps[1] = getbe32(data + 28);
 
 	/* If index file is corrupted, ps[] may point to invalid revisions. So
 	 * there is a risk of buffer overflow to trust them unconditionally. */
 	if (ps[0] < -1 || ps[0] > maxrev || ps[1] < -1 || ps[1] > maxrev) {
 		PyErr_SetString(PyExc_ValueError, "parent out of range");
 		return -1;
 	}
 	return 0;
 }
 
 /*
  * Get parents of the given rev.
  *
  * If the specified rev is out of range, IndexError will be raised. If the
  * revlog entry is corrupted, ValueError may be raised.
  *
  * Returns 0 on success or -1 on failure.
  */
 static int HgRevlogIndex_GetParents(PyObject *op, int rev, int *ps)
 {
 	int tiprev;
 	if (!op || !HgRevlogIndex_Check(op) || !ps) {
 		PyErr_BadInternalCall();
 		return -1;
 	}
 	tiprev = (int)index_length((indexObject *)op) - 1;
 	if (rev < -1 || rev > tiprev) {
 		PyErr_Format(PyExc_IndexError, "rev out of range: %d", rev);
 		return -1;
 	} else if (rev == -1) {
 		ps[0] = ps[1] = -1;
 		return 0;
 	} else {
 		return index_get_parents((indexObject *)op, rev, ps, tiprev);
 	}
 }
 
 static inline int64_t index_get_start(indexObject *self, Py_ssize_t rev)
 {
 	const char *data;
 	uint64_t offset;
 
 	if (rev == nullrev)
 		return 0;
 
 	data = index_deref(self, rev);
 	offset = getbe32(data + 4);
 	if (rev == 0) {
 		/* mask out version number for the first entry */
 		offset &= 0xFFFF;
 	} else {
 		uint32_t offset_high = getbe32(data);
 		offset |= ((uint64_t)offset_high) << 32;
 	}
 	return (int64_t)(offset >> 16);
 }
 
 static inline int index_get_length(indexObject *self, Py_ssize_t rev)
 {
 	const char *data;
 	int tmp;
 
 	if (rev == nullrev)
 		return 0;
 
 	data = index_deref(self, rev);
 
 	tmp = (int)getbe32(data + 8);
 	if (tmp < 0) {
 		PyErr_Format(PyExc_OverflowError,
 		             "revlog entry size out of bound (%d)", tmp);
 		return -1;
 	}
 	return tmp;
 }
 
 /*
  * RevlogNG format (all in big endian, data may be inlined):
  *    6 bytes: offset
  *    2 bytes: flags
  *    4 bytes: compressed length
  *    4 bytes: uncompressed length
  *    4 bytes: base revision
  *    4 bytes: link revision
  *    4 bytes: parent 1 revision
  *    4 bytes: parent 2 revision
  *   32 bytes: nodeid (only 20 bytes used with SHA-1)
  */
 static PyObject *index_get(indexObject *self, Py_ssize_t pos)
 {
 	uint64_t offset_flags, sidedata_offset;
 	int comp_len, uncomp_len, base_rev, link_rev, parent_1, parent_2,
 	    sidedata_comp_len;
 	const char *c_node_id;
 	const char *data;
 	Py_ssize_t length = index_length(self);
 
 	if (pos == nullrev) {
 		Py_INCREF(self->nullentry);
 		return self->nullentry;
 	}
 
 	if (pos < 0 || pos >= length) {
 		PyErr_SetString(PyExc_IndexError, "revlog index out of range");
 		return NULL;
 	}
 
 	data = index_deref(self, pos);
 	if (data == NULL)
 		return NULL;
 
 	offset_flags = getbe32(data + 4);
 	/*
 	 * The first entry on-disk needs the version number masked out,
 	 * but this doesn't apply if entries are added to an empty index.
 	 */
 	if (self->length && pos == 0)
 		offset_flags &= 0xFFFF;
 	else {
 		uint32_t offset_high = getbe32(data);
 		offset_flags |= ((uint64_t)offset_high) << 32;
 	}
 
 	comp_len = getbe32(data + 8);
 	uncomp_len = getbe32(data + 12);
 	base_rev = getbe32(data + 16);
 	link_rev = getbe32(data + 20);
 	parent_1 = getbe32(data + 24);
 	parent_2 = getbe32(data + 28);
 	c_node_id = data + 32;
 
 	if (self->hdrsize == v1_hdrsize) {
 		return Py_BuildValue(v1_tuple_format, offset_flags, comp_len,
 		                     uncomp_len, base_rev, link_rev, parent_1,
 		                     parent_2, c_node_id, self->nodelen);
 	} else {
 		sidedata_offset = getbe64(data + 64);
 		sidedata_comp_len = getbe32(data + 72);
 
 		return Py_BuildValue(v2_tuple_format, offset_flags, comp_len,
 		                     uncomp_len, base_rev, link_rev, parent_1,
 		                     parent_2, c_node_id, self->nodelen,
 		                     sidedata_offset, sidedata_comp_len);
 	}
 }
 /*
  * Pack header information in binary
  */
 static PyObject *index_pack_header(indexObject *self, PyObject *args)
 {
 	int header;
 	char out[4];
 	if (!PyArg_ParseTuple(args, "I", &header)) {
 		return NULL;
 	}
 	putbe32(header, out);
 	return PyBytes_FromStringAndSize(out, 4);
 }
 /*
  * Return the raw binary string representing a revision
  */
 static PyObject *index_entry_binary(indexObject *self, PyObject *value)
 {
 	long rev;
 	const char *data;
 	Py_ssize_t length = index_length(self);
 
 	if (!pylong_to_long(value, &rev)) {
 		return NULL;
 	}
 	if (rev < 0 || rev >= length) {
 		PyErr_Format(PyExc_ValueError, "revlog index out of range: %ld",
 		             rev);
 		return NULL;
 	};
 
 	data = index_deref(self, rev);
 	if (data == NULL)
 		return NULL;
 	if (rev == 0) {
 		/* the header is eating the start of the first entry */
 		return PyBytes_FromStringAndSize(data + 4, self->hdrsize - 4);
 	}
 	return PyBytes_FromStringAndSize(data, self->hdrsize);
 }
 
 /*
  * Return the hash of node corresponding to the given rev.
  */
 static const char *index_node(indexObject *self, Py_ssize_t pos)
 {
 	Py_ssize_t length = index_length(self);
 	const char *data;
 
 	if (pos == nullrev)
 		return nullid;
 
 	if (pos >= length)
 		return NULL;
 
 	data = index_deref(self, pos);
 	return data ? data + 32 : NULL;
 }
 
 /*
  * Return the hash of the node corresponding to the given rev. The
  * rev is assumed to be existing. If not, an exception is set.
  */
 static const char *index_node_existing(indexObject *self, Py_ssize_t pos)
 {
 	const char *node = index_node(self, pos);
 	if (node == NULL) {
 		PyErr_Format(PyExc_IndexError, "could not access rev %d",
 		             (int)pos);
 	}
 	return node;
 }
 
 static int nt_insert(nodetree *self, const char *node, int rev);
 
 static int node_check(Py_ssize_t nodelen, PyObject *obj, char **node)
 {
 	Py_ssize_t thisnodelen;
 	if (PyBytes_AsStringAndSize(obj, node, &thisnodelen) == -1)
 		return -1;
 	if (nodelen == thisnodelen)
 		return 0;
 	PyErr_Format(PyExc_ValueError, "node len %zd != expected node len %zd",
 	             thisnodelen, nodelen);
 	return -1;
 }
 
 static PyObject *index_append(indexObject *self, PyObject *obj)
 {
 	uint64_t offset_flags, sidedata_offset;
 	int rev, comp_len, uncomp_len, base_rev, link_rev, parent_1, parent_2;
 	Py_ssize_t c_node_id_len, sidedata_comp_len;
 	const char *c_node_id;
 	char *data;
 
 	if (self->hdrsize == v1_hdrsize) {
 		if (!PyArg_ParseTuple(obj, v1_tuple_format, &offset_flags,
 		                      &comp_len, &uncomp_len, &base_rev,
 		                      &link_rev, &parent_1, &parent_2,
 		                      &c_node_id, &c_node_id_len)) {
 			PyErr_SetString(PyExc_TypeError, "8-tuple required");
 			return NULL;
 		}
 	} else {
 		if (!PyArg_ParseTuple(obj, v2_tuple_format, &offset_flags,
 		                      &comp_len, &uncomp_len, &base_rev,
 		                      &link_rev, &parent_1, &parent_2,
 		                      &c_node_id, &c_node_id_len,
 		                      &sidedata_offset, &sidedata_comp_len)) {
 			PyErr_SetString(PyExc_TypeError, "10-tuple required");
 			return NULL;
 		}
 	}
 
 	if (c_node_id_len != self->nodelen) {
 		PyErr_SetString(PyExc_TypeError, "invalid node");
 		return NULL;
 	}
 
 	if (self->new_length == self->added_length) {
 		size_t new_added_length =
 		    self->added_length ? self->added_length * 2 : 4096;
 		void *new_added = PyMem_Realloc(self->added, new_added_length *
 		                                                 self->hdrsize);
 		if (!new_added)
 			return PyErr_NoMemory();
 		self->added = new_added;
 		self->added_length = new_added_length;
 	}
 	rev = self->length + self->new_length;
 	data = self->added + self->hdrsize * self->new_length++;
 	putbe32(offset_flags >> 32, data);
 	putbe32(offset_flags & 0xffffffffU, data + 4);
 	putbe32(comp_len, data + 8);
 	putbe32(uncomp_len, data + 12);
 	putbe32(base_rev, data + 16);
 	putbe32(link_rev, data + 20);
 	putbe32(parent_1, data + 24);
 	putbe32(parent_2, data + 28);
 	memcpy(data + 32, c_node_id, c_node_id_len);
 	/* Padding since SHA-1 is only 20 bytes for now */
 	memset(data + 32 + c_node_id_len, 0, 32 - c_node_id_len);
 	if (self->hdrsize != v1_hdrsize) {
 		putbe64(sidedata_offset, data + 64);
 		putbe32(sidedata_comp_len, data + 72);
 		/* Padding for 96 bytes alignment */
 		memset(data + 76, 0, self->hdrsize - 76);
 	}
 
 	if (self->ntinitialized)
 		nt_insert(&self->nt, c_node_id, rev);
 
 	Py_CLEAR(self->headrevs);
 	Py_RETURN_NONE;
 }
 
 /* Replace an existing index entry's sidedata offset and length with new ones.
    This cannot be used outside of the context of sidedata rewriting,
    inside the transaction that creates the given revision. */
 static PyObject *index_replace_sidedata_info(indexObject *self, PyObject *args)
 {
-	uint64_t sidedata_offset;
+	uint64_t offset_flags, sidedata_offset;
 	int rev;
 	Py_ssize_t sidedata_comp_len;
 	char *data;
 #if LONG_MAX == 0x7fffffffL
-	const char *const sidedata_format = PY23("nKi", "nKi");
+	const char *const sidedata_format = PY23("nKiK", "nKiK");
 #else
-	const char *const sidedata_format = PY23("nki", "nki");
+	const char *const sidedata_format = PY23("nkik", "nkik");
 #endif
 
 	if (self->hdrsize == v1_hdrsize || self->inlined) {
 		/*
 		 There is a bug in the transaction handling when going from an
 	   inline revlog to a separate index and data file. Turn it off until
 	   it's fixed, since v2 revlogs sometimes get rewritten on exchange.
 	   See issue6485.
 	  */
 		raise_revlog_error();
 		return NULL;
 	}
 
 	if (!PyArg_ParseTuple(args, sidedata_format, &rev, &sidedata_offset,
-	                      &sidedata_comp_len))
+	                      &sidedata_comp_len, &offset_flags))
 		return NULL;
 
 	if (rev < 0 || rev >= index_length(self)) {
 		PyErr_SetString(PyExc_IndexError, "revision outside index");
 		return NULL;
 	}
 	if (rev < self->length) {
 		PyErr_SetString(
 		    PyExc_IndexError,
 		    "cannot rewrite entries outside of this transaction");
 		return NULL;
 	}
 
 	/* Find the newly added node, offset from the "already on-disk" length
 	 */
 	data = self->added + self->hdrsize * (rev - self->length);
+	putbe64(offset_flags, data);
 	putbe64(sidedata_offset, data + 64);
 	putbe32(sidedata_comp_len, data + 72);
 
 	Py_RETURN_NONE;
 }
 
 static PyObject *index_stats(indexObject *self)
 {
 	PyObject *obj = PyDict_New();
 	PyObject *s = NULL;
 	PyObject *t = NULL;
 
 	if (obj == NULL)
 		return NULL;
 
 #define istat(__n, __d)                                                        \
 	do {                                                                   \
 		s = PyBytes_FromString(__d);                                   \
 		t = PyInt_FromSsize_t(self->__n);                              \
 		if (!s || !t)                                                  \
 			goto bail;                                             \
 		if (PyDict_SetItem(obj, s, t) == -1)                           \
 			goto bail;                                             \
 		Py_CLEAR(s);                                                   \
 		Py_CLEAR(t);                                                   \
 	} while (0)
 
 	if (self->added_length)
 		istat(new_length, "index entries added");
 	istat(length, "revs in memory");
 	istat(ntlookups, "node trie lookups");
 	istat(ntmisses, "node trie misses");
 	istat(ntrev, "node trie last rev scanned");
 	if (self->ntinitialized) {
 		istat(nt.capacity, "node trie capacity");
 		istat(nt.depth, "node trie depth");
 		istat(nt.length, "node trie count");
 		istat(nt.splits, "node trie splits");
 	}
 
 #undef istat
 
 	return obj;
 
 bail:
 	Py_XDECREF(obj);
 	Py_XDECREF(s);
 	Py_XDECREF(t);
 	return NULL;
 }
 
 /*
  * When we cache a list, we want to be sure the caller can't mutate
  * the cached copy.
  */
 static PyObject *list_copy(PyObject *list)
 {
 	Py_ssize_t len = PyList_GET_SIZE(list);
 	PyObject *newlist = PyList_New(len);
 	Py_ssize_t i;
 
 	if (newlist == NULL)
 		return NULL;
 
 	for (i = 0; i < len; i++) {
 		PyObject *obj = PyList_GET_ITEM(list, i);
 		Py_INCREF(obj);
 		PyList_SET_ITEM(newlist, i, obj);
 	}
 
 	return newlist;
 }
 
 static int check_filter(PyObject *filter, Py_ssize_t arg)
 {
 	if (filter) {
 		PyObject *arglist, *result;
 		int isfiltered;
 
 		arglist = Py_BuildValue("(n)", arg);
 		if (!arglist) {
 			return -1;
 		}
 
 		result = PyObject_Call(filter, arglist, NULL);
 		Py_DECREF(arglist);
 		if (!result) {
 			return -1;
 		}
 
 		/* PyObject_IsTrue returns 1 if true, 0 if false, -1 if error,
 		 * same as this function, so we can just return it directly.*/
 		isfiltered = PyObject_IsTrue(result);
 		Py_DECREF(result);
 		return isfiltered;
 	} else {
 		return 0;
 	}
 }
 
 static inline void set_phase_from_parents(char *phases, int parent_1,
                                           int parent_2, Py_ssize_t i)
 {
 	if (parent_1 >= 0 && phases[parent_1] > phases[i])
 		phases[i] = phases[parent_1];
 	if (parent_2 >= 0 && phases[parent_2] > phases[i])
 		phases[i] = phases[parent_2];
 }
 
 static PyObject *reachableroots2(indexObject *self, PyObject *args)
 {
 
 	/* Input */
 	long minroot;
 	PyObject *includepatharg = NULL;
 	int includepath = 0;
 	/* heads and roots are lists */
 	PyObject *heads = NULL;
 	PyObject *roots = NULL;
 	PyObject *reachable = NULL;
 
 	PyObject *val;
 	Py_ssize_t len = index_length(self);
 	long revnum;
 	Py_ssize_t k;
 	Py_ssize_t i;
 	Py_ssize_t l;
 	int r;
 	int parents[2];
 
 	/* Internal data structure:
 	 * tovisit: array of length len+1 (all revs + nullrev), filled upto
 	 * lentovisit
 	 *
 	 * revstates: array of length len+1 (all revs + nullrev) */
 	int *tovisit = NULL;
 	long lentovisit = 0;
 	enum { RS_SEEN = 1, RS_ROOT = 2, RS_REACHABLE = 4 };
 	char *revstates = NULL;
 
 	/* Get arguments */
 	if (!PyArg_ParseTuple(args, "lO!O!O!", &minroot, &PyList_Type, &heads,
 	                      &PyList_Type, &roots, &PyBool_Type,
 	                      &includepatharg))
 		goto bail;
 
 	if (includepatharg == Py_True)
 		includepath = 1;
 
 	/* Initialize return set */
 	reachable = PyList_New(0);
 	if (reachable == NULL)
 		goto bail;
 
 	/* Initialize internal datastructures */
 	tovisit = (int *)malloc((len + 1) * sizeof(int));
 	if (tovisit == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	revstates = (char *)calloc(len + 1, 1);
 	if (revstates == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	l = PyList_GET_SIZE(roots);
 	for (i = 0; i < l; i++) {
 		revnum = PyInt_AsLong(PyList_GET_ITEM(roots, i));
 		if (revnum == -1 && PyErr_Occurred())
 			goto bail;
 		/* If root is out of range, e.g. wdir(), it must be unreachable
 		 * from heads. So we can just ignore it. */
 		if (revnum + 1 < 0 || revnum + 1 >= len + 1)
 			continue;
 		revstates[revnum + 1] |= RS_ROOT;
 	}
 
 	/* Populate tovisit with all the heads */
 	l = PyList_GET_SIZE(heads);
 	for (i = 0; i < l; i++) {
 		revnum = PyInt_AsLong(PyList_GET_ITEM(heads, i));
 		if (revnum == -1 && PyErr_Occurred())
 			goto bail;
 		if (revnum + 1 < 0 || revnum + 1 >= len + 1) {
 			PyErr_SetString(PyExc_IndexError, "head out of range");
 			goto bail;
 		}
 		if (!(revstates[revnum + 1] & RS_SEEN)) {
 			tovisit[lentovisit++] = (int)revnum;
 			revstates[revnum + 1] |= RS_SEEN;
 		}
 	}
 
 	/* Visit the tovisit list and find the reachable roots */
 	k = 0;
 	while (k < lentovisit) {
 		/* Add the node to reachable if it is a root*/
 		revnum = tovisit[k++];
 		if (revstates[revnum + 1] & RS_ROOT) {
 			revstates[revnum + 1] |= RS_REACHABLE;
 			val = PyInt_FromLong(revnum);
 			if (val == NULL)
 				goto bail;
 			r = PyList_Append(reachable, val);
 			Py_DECREF(val);
 			if (r < 0)
 				goto bail;
 			if (includepath == 0)
 				continue;
 		}
 
 		/* Add its parents to the list of nodes to visit */
 		if (revnum == nullrev)
 			continue;
 		r = index_get_parents(self, revnum, parents, (int)len - 1);
 		if (r < 0)
 			goto bail;
 		for (i = 0; i < 2; i++) {
 			if (!(revstates[parents[i] + 1] & RS_SEEN) &&
 			    parents[i] >= minroot) {
 				tovisit[lentovisit++] = parents[i];
 				revstates[parents[i] + 1] |= RS_SEEN;
 			}
 		}
 	}
 
 	/* Find all the nodes in between the roots we found and the heads
 	 * and add them to the reachable set */
 	if (includepath == 1) {
 		long minidx = minroot;
 		if (minidx < 0)
 			minidx = 0;
 		for (i = minidx; i < len; i++) {
 			if (!(revstates[i + 1] & RS_SEEN))
 				continue;
 			r = index_get_parents(self, i, parents, (int)len - 1);
 			/* Corrupted index file, error is set from
 			 * index_get_parents */
 			if (r < 0)
 				goto bail;
 			if (((revstates[parents[0] + 1] |
 			      revstates[parents[1] + 1]) &
 			     RS_REACHABLE) &&
 			    !(revstates[i + 1] & RS_REACHABLE)) {
 				revstates[i + 1] |= RS_REACHABLE;
 				val = PyInt_FromSsize_t(i);
 				if (val == NULL)
 					goto bail;
 				r = PyList_Append(reachable, val);
 				Py_DECREF(val);
 				if (r < 0)
 					goto bail;
 			}
 		}
 	}
 
 	free(revstates);
 	free(tovisit);
 	return reachable;
 bail:
 	Py_XDECREF(reachable);
 	free(revstates);
 	free(tovisit);
 	return NULL;
 }
 
 static int add_roots_get_min(indexObject *self, PyObject *roots, char *phases,
                              char phase)
 {
 	Py_ssize_t len = index_length(self);
 	PyObject *item;
 	PyObject *iterator;
 	int rev, minrev = -1;
 	char *node;
 
 	if (!PySet_Check(roots)) {
 		PyErr_SetString(PyExc_TypeError,
 		                "roots must be a set of nodes");
 		return -2;
 	}
 	iterator = PyObject_GetIter(roots);
 	if (iterator == NULL)
 		return -2;
 	while ((item = PyIter_Next(iterator))) {
 		if (node_check(self->nodelen, item, &node) == -1)
 			goto failed;
 		rev = index_find_node(self, node);
 		/* null is implicitly public, so negative is invalid */
 		if (rev < 0 || rev >= len)
 			goto failed;
 		phases[rev] = phase;
 		if (minrev == -1 || minrev > rev)
 			minrev = rev;
 		Py_DECREF(item);
 	}
 	Py_DECREF(iterator);
 	return minrev;
 failed:
 	Py_DECREF(iterator);
 	Py_DECREF(item);
 	return -2;
 }
 
 static PyObject *compute_phases_map_sets(indexObject *self, PyObject *args)
 {
 	/* 0: public (untracked), 1: draft, 2: secret, 32: archive,
 	   96: internal */
 	static const char trackedphases[] = {1, 2, 32, 96};
 	PyObject *roots = Py_None;
 	PyObject *phasesetsdict = NULL;
 	PyObject *phasesets[4] = {NULL, NULL, NULL, NULL};
 	Py_ssize_t len = index_length(self);
 	char *phases = NULL;
 	int minphaserev = -1, rev, i;
 	const int numphases = (int)(sizeof(phasesets) / sizeof(phasesets[0]));
 
 	if (!PyArg_ParseTuple(args, "O", &roots))
 		return NULL;
 	if (roots == NULL || !PyDict_Check(roots)) {
 		PyErr_SetString(PyExc_TypeError, "roots must be a dictionary");
 		return NULL;
 	}
 
 	phases = calloc(len, 1);
 	if (phases == NULL) {
 		PyErr_NoMemory();
 		return NULL;
 	}
 
 	for (i = 0; i < numphases; ++i) {
 		PyObject *pyphase = PyInt_FromLong(trackedphases[i]);
 		PyObject *phaseroots = NULL;
 		if (pyphase == NULL)
 			goto release;
 		phaseroots = PyDict_GetItem(roots, pyphase);
 		Py_DECREF(pyphase);
 		if (phaseroots == NULL)
 			continue;
 		rev = add_roots_get_min(self, phaseroots, phases,
 		                        trackedphases[i]);
 		if (rev == -2)
 			goto release;
 		if (rev != -1 && (minphaserev == -1 || rev < minphaserev))
 			minphaserev = rev;
 	}
 
 	for (i = 0; i < numphases; ++i) {
 		phasesets[i] = PySet_New(NULL);
 		if (phasesets[i] == NULL)
 			goto release;
 	}
 
 	if (minphaserev == -1)
 		minphaserev = len;
 	for (rev = minphaserev; rev < len; ++rev) {
 		PyObject *pyphase = NULL;
 		PyObject *pyrev = NULL;
 		int parents[2];
 		/*
 		 * The parent lookup could be skipped for phaseroots, but
 		 * phase --force would historically not recompute them
 		 * correctly, leaving descendents with a lower phase around.
 		 * As such, unconditionally recompute the phase.
 		 */
 		if (index_get_parents(self, rev, parents, (int)len - 1) < 0)
 			goto release;
 		set_phase_from_parents(phases, parents[0], parents[1], rev);
 		switch (phases[rev]) {
 		case 0:
 			continue;
 		case 1:
 			pyphase = phasesets[0];
 			break;
 		case 2:
 			pyphase = phasesets[1];
 			break;
 		case 32:
 			pyphase = phasesets[2];
 			break;
 		case 96:
 			pyphase = phasesets[3];
 			break;
 		default:
 			/* this should never happen since the phase number is
 			 * specified by this function. */
 			PyErr_SetString(PyExc_SystemError,
 			                "bad phase number in internal list");
 			goto release;
 		}
 		pyrev = PyInt_FromLong(rev);
 		if (pyrev == NULL)
 			goto release;
 		if (PySet_Add(pyphase, pyrev) == -1) {
 			Py_DECREF(pyrev);
 			goto release;
 		}
 		Py_DECREF(pyrev);
 	}
 
 	phasesetsdict = _dict_new_presized(numphases);
 	if (phasesetsdict == NULL)
 		goto release;
 	for (i = 0; i < numphases; ++i) {
 		PyObject *pyphase = PyInt_FromLong(trackedphases[i]);
 		if (pyphase == NULL)
 			goto release;
 		if (PyDict_SetItem(phasesetsdict, pyphase, phasesets[i]) ==
 		    -1) {
 			Py_DECREF(pyphase);
 			goto release;
 		}
 		Py_DECREF(phasesets[i]);
 		phasesets[i] = NULL;
 	}
 
 	return Py_BuildValue("nN", len, phasesetsdict);
 
 release:
 	for (i = 0; i < numphases; ++i)
 		Py_XDECREF(phasesets[i]);
 	Py_XDECREF(phasesetsdict);
 
 	free(phases);
 	return NULL;
 }
 
 static PyObject *index_headrevs(indexObject *self, PyObject *args)
 {
 	Py_ssize_t i, j, len;
 	char *nothead = NULL;
 	PyObject *heads = NULL;
 	PyObject *filter = NULL;
 	PyObject *filteredrevs = Py_None;
 
 	if (!PyArg_ParseTuple(args, "|O", &filteredrevs)) {
 		return NULL;
 	}
 
 	if (self->headrevs && filteredrevs == self->filteredrevs)
 		return list_copy(self->headrevs);
 
 	Py_DECREF(self->filteredrevs);
 	self->filteredrevs = filteredrevs;
 	Py_INCREF(filteredrevs);
 
 	if (filteredrevs != Py_None) {
 		filter = PyObject_GetAttrString(filteredrevs, "__contains__");
 		if (!filter) {
 			PyErr_SetString(
 			    PyExc_TypeError,
 			    "filteredrevs has no attribute __contains__");
 			goto bail;
 		}
 	}
 
 	len = index_length(self);
 	heads = PyList_New(0);
 	if (heads == NULL)
 		goto bail;
 	if (len == 0) {
 		PyObject *nullid = PyInt_FromLong(-1);
 		if (nullid == NULL || PyList_Append(heads, nullid) == -1) {
 			Py_XDECREF(nullid);
 			goto bail;
 		}
 		goto done;
 	}
 
 	nothead = calloc(len, 1);
 	if (nothead == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	for (i = len - 1; i >= 0; i--) {
 		int isfiltered;
 		int parents[2];
 
 		/* If nothead[i] == 1, it means we've seen an unfiltered child
 		 * of this node already, and therefore this node is not
 		 * filtered. So we can skip the expensive check_filter step.
 		 */
 		if (nothead[i] != 1) {
 			isfiltered = check_filter(filter, i);
 			if (isfiltered == -1) {
 				PyErr_SetString(PyExc_TypeError,
 				                "unable to check filter");
 				goto bail;
 			}
 
 			if (isfiltered) {
 				nothead[i] = 1;
 				continue;
 			}
 		}
 
 		if (index_get_parents(self, i, parents, (int)len - 1) < 0)
 			goto bail;
 		for (j = 0; j < 2; j++) {
 			if (parents[j] >= 0)
 				nothead[parents[j]] = 1;
 		}
 	}
 
 	for (i = 0; i < len; i++) {
 		PyObject *head;
 
 		if (nothead[i])
 			continue;
 		head = PyInt_FromSsize_t(i);
 		if (head == NULL || PyList_Append(heads, head) == -1) {
 			Py_XDECREF(head);
 			goto bail;
 		}
 	}
 
 done:
 	self->headrevs = heads;
 	Py_XDECREF(filter);
 	free(nothead);
 	return list_copy(self->headrevs);
 bail:
 	Py_XDECREF(filter);
 	Py_XDECREF(heads);
 	free(nothead);
 	return NULL;
 }
 
 /**
  * Obtain the base revision index entry.
  *
  * Callers must ensure that rev >= 0 or illegal memory access may occur.
  */
 static inline int index_baserev(indexObject *self, int rev)
 {
 	const char *data;
 	int result;
 
 	data = index_deref(self, rev);
 	if (data == NULL)
 		return -2;
 	result = getbe32(data + 16);
 
 	if (result > rev) {
 		PyErr_Format(
 		    PyExc_ValueError,
 		    "corrupted revlog, revision base above revision: %d, %d",
 		    rev, result);
 		return -2;
 	}
 	if (result < -1) {
 		PyErr_Format(
 		    PyExc_ValueError,
 		    "corrupted revlog, revision base out of range: %d, %d", rev,
 		    result);
 		return -2;
 	}
 	return result;
 }
 
 /**
  * Find if a revision is a snapshot or not
  *
  * Only relevant for sparse-revlog case.
  * Callers must ensure that rev is in a valid range.
  */
 static int index_issnapshotrev(indexObject *self, Py_ssize_t rev)
 {
 	int ps[2];
 	Py_ssize_t base;
 	while (rev >= 0) {
 		base = (Py_ssize_t)index_baserev(self, rev);
 		if (base == rev) {
 			base = -1;
 		}
 		if (base == -2) {
 			assert(PyErr_Occurred());
 			return -1;
 		}
 		if (base == -1) {
 			return 1;
 		}
 		if (index_get_parents(self, rev, ps, (int)rev) < 0) {
 			assert(PyErr_Occurred());
 			return -1;
 		};
 		if (base == ps[0] || base == ps[1]) {
 			return 0;
 		}
 		rev = base;
 	}
 	return rev == -1;
 }
 
 static PyObject *index_issnapshot(indexObject *self, PyObject *value)
 {
 	long rev;
 	int issnap;
 	Py_ssize_t length = index_length(self);
 
 	if (!pylong_to_long(value, &rev)) {
 		return NULL;
 	}
 	if (rev < -1 || rev >= length) {
 		PyErr_Format(PyExc_ValueError, "revlog index out of range: %ld",
 		             rev);
 		return NULL;
 	};
 	issnap = index_issnapshotrev(self, (Py_ssize_t)rev);
 	if (issnap < 0) {
 		return NULL;
 	};
 	return PyBool_FromLong((long)issnap);
 }
 
 static PyObject *index_findsnapshots(indexObject *self, PyObject *args)
 {
 	Py_ssize_t start_rev;
 	PyObject *cache;
 	Py_ssize_t base;
 	Py_ssize_t rev;
 	PyObject *key = NULL;
 	PyObject *value = NULL;
 	const Py_ssize_t length = index_length(self);
 	if (!PyArg_ParseTuple(args, "O!n", &PyDict_Type, &cache, &start_rev)) {
 		return NULL;
 	}
 	for (rev = start_rev; rev < length; rev++) {
 		int issnap;
 		PyObject *allvalues = NULL;
 		issnap = index_issnapshotrev(self, rev);
 		if (issnap < 0) {
 			goto bail;
 		}
 		if (issnap == 0) {
 			continue;
 		}
 		base = (Py_ssize_t)index_baserev(self, rev);
 		if (base == rev) {
 			base = -1;
 		}
 		if (base == -2) {
 			assert(PyErr_Occurred());
 			goto bail;
 		}
 		key = PyInt_FromSsize_t(base);
 		allvalues = PyDict_GetItem(cache, key);
 		if (allvalues == NULL && PyErr_Occurred()) {
 			goto bail;
 		}
 		if (allvalues == NULL) {
 			int r;
 			allvalues = PyList_New(0);
 			if (!allvalues) {
 				goto bail;
 			}
 			r = PyDict_SetItem(cache, key, allvalues);
 			Py_DECREF(allvalues);
 			if (r < 0) {
 				goto bail;
 			}
 		}
 		value = PyInt_FromSsize_t(rev);
 		if (PyList_Append(allvalues, value)) {
 			goto bail;
 		}
 		Py_CLEAR(key);
 		Py_CLEAR(value);
 	}
 	Py_RETURN_NONE;
 bail:
 	Py_XDECREF(key);
 	Py_XDECREF(value);
 	return NULL;
 }
 
 static PyObject *index_deltachain(indexObject *self, PyObject *args)
 {
 	int rev, generaldelta;
 	PyObject *stoparg;
 	int stoprev, iterrev, baserev = -1;
 	int stopped;
 	PyObject *chain = NULL, *result = NULL;
 	const Py_ssize_t length = index_length(self);
 
 	if (!PyArg_ParseTuple(args, "iOi", &rev, &stoparg, &generaldelta)) {
 		return NULL;
 	}
 
 	if (PyInt_Check(stoparg)) {
 		stoprev = (int)PyInt_AsLong(stoparg);
 		if (stoprev == -1 && PyErr_Occurred()) {
 			return NULL;
 		}
 	} else if (stoparg == Py_None) {
 		stoprev = -2;
 	} else {
 		PyErr_SetString(PyExc_ValueError,
 		                "stoprev must be integer or None");
 		return NULL;
 	}
 
 	if (rev < 0 || rev >= length) {
 		PyErr_SetString(PyExc_ValueError, "revlog index out of range");
 		return NULL;
 	}
 
 	chain = PyList_New(0);
 	if (chain == NULL) {
 		return NULL;
 	}
 
 	baserev = index_baserev(self, rev);
 
 	/* This should never happen. */
 	if (baserev <= -2) {
 		/* Error should be set by index_deref() */
 		assert(PyErr_Occurred());
 		goto bail;
 	}
 
 	iterrev = rev;
 
 	while (iterrev != baserev && iterrev != stoprev) {
 		PyObject *value = PyInt_FromLong(iterrev);
 		if (value == NULL) {
 			goto bail;
 		}
 		if (PyList_Append(chain, value)) {
 			Py_DECREF(value);
 			goto bail;
 		}
 		Py_DECREF(value);
 
 		if (generaldelta) {
 			iterrev = baserev;
 		} else {
 			iterrev--;
 		}
 
 		if (iterrev < 0) {
 			break;
 		}
 
 		if (iterrev >= length) {
 			PyErr_SetString(PyExc_IndexError,
 			                "revision outside index");
 			return NULL;
 		}
 
 		baserev = index_baserev(self, iterrev);
 
 		/* This should never happen. */
 		if (baserev <= -2) {
 			/* Error should be set by index_deref() */
 			assert(PyErr_Occurred());
 			goto bail;
 		}
 	}
 
 	if (iterrev == stoprev) {
 		stopped = 1;
 	} else {
 		PyObject *value = PyInt_FromLong(iterrev);
 		if (value == NULL) {
 			goto bail;
 		}
 		if (PyList_Append(chain, value)) {
 			Py_DECREF(value);
 			goto bail;
 		}
 		Py_DECREF(value);
 
 		stopped = 0;
 	}
 
 	if (PyList_Reverse(chain)) {
 		goto bail;
 	}
 
 	result = Py_BuildValue("OO", chain, stopped ? Py_True : Py_False);
 	Py_DECREF(chain);
 	return result;
 
 bail:
 	Py_DECREF(chain);
 	return NULL;
 }
 
 static inline int64_t
 index_segment_span(indexObject *self, Py_ssize_t start_rev, Py_ssize_t end_rev)
 {
 	int64_t start_offset;
 	int64_t end_offset;
 	int end_size;
 	start_offset = index_get_start(self, start_rev);
 	if (start_offset < 0) {
 		return -1;
 	}
 	end_offset = index_get_start(self, end_rev);
 	if (end_offset < 0) {
 		return -1;
 	}
 	end_size = index_get_length(self, end_rev);
 	if (end_size < 0) {
 		return -1;
 	}
 	if (end_offset < start_offset) {
 		PyErr_Format(PyExc_ValueError,
 		             "corrupted revlog index: inconsistent offset "
 		             "between revisions (%zd) and (%zd)",
 		             start_rev, end_rev);
 		return -1;
 	}
 	return (end_offset - start_offset) + (int64_t)end_size;
 }
 
 /* returns endidx so that revs[startidx:endidx] has no empty trailing revs */
 static Py_ssize_t trim_endidx(indexObject *self, const Py_ssize_t *revs,
                               Py_ssize_t startidx, Py_ssize_t endidx)
 {
 	int length;
 	while (endidx > 1 && endidx > startidx) {
 		length = index_get_length(self, revs[endidx - 1]);
 		if (length < 0) {
 			return -1;
 		}
 		if (length != 0) {
 			break;
 		}
 		endidx -= 1;
 	}
 	return endidx;
 }
 
 struct Gap {
 	int64_t size;
 	Py_ssize_t idx;
 };
 
 static int gap_compare(const void *left, const void *right)
 {
 	const struct Gap *l_left = ((const struct Gap *)left);
 	const struct Gap *l_right = ((const struct Gap *)right);
 	if (l_left->size < l_right->size) {
 		return -1;
 	} else if (l_left->size > l_right->size) {
 		return 1;
 	}
 	return 0;
 }
 static int Py_ssize_t_compare(const void *left, const void *right)
 {
 	const Py_ssize_t l_left = *(const Py_ssize_t *)left;
 	const Py_ssize_t l_right = *(const Py_ssize_t *)right;
 	if (l_left < l_right) {
 		return -1;
 	} else if (l_left > l_right) {
 		return 1;
 	}
 	return 0;
 }
 
 static PyObject *index_slicechunktodensity(indexObject *self, PyObject *args)
 {
 	/* method arguments */
 	PyObject *list_revs = NULL; /* revisions in the chain */
 	double targetdensity = 0;   /* min density to achieve */
 	Py_ssize_t mingapsize = 0;  /* threshold to ignore gaps */
 
 	/* other core variables */
 	Py_ssize_t idxlen = index_length(self);
 	Py_ssize_t i;            /* used for various iteration */
 	PyObject *result = NULL; /* the final return of the function */
 
 	/* generic information about the delta chain being slice */
 	Py_ssize_t num_revs = 0;    /* size of the full delta chain */
 	Py_ssize_t *revs = NULL;    /* native array of revision in the chain */
 	int64_t chainpayload = 0;   /* sum of all delta in the chain */
 	int64_t deltachainspan = 0; /* distance from first byte to last byte */
 
 	/* variable used for slicing the delta chain */
 	int64_t readdata = 0; /* amount of data currently planned to be read */
 	double density = 0;   /* ration of payload data compared to read ones */
 	int64_t previous_end;
 	struct Gap *gaps = NULL; /* array of notable gap in the chain */
 	Py_ssize_t num_gaps =
 	    0; /* total number of notable gap recorded so far */
 	Py_ssize_t *selected_indices = NULL; /* indices of gap skipped over */
 	Py_ssize_t num_selected = 0;         /* number of gaps skipped */
 	PyObject *chunk = NULL;              /* individual slice */
 	PyObject *allchunks = NULL;          /* all slices */
 	Py_ssize_t previdx;
 
 	/* parsing argument */
 	if (!PyArg_ParseTuple(args, "O!dn", &PyList_Type, &list_revs,
 	                      &targetdensity, &mingapsize)) {
 		goto bail;
 	}
 
 	/* If the delta chain contains a single element, we do not need slicing
 	 */
 	num_revs = PyList_GET_SIZE(list_revs);
 	if (num_revs <= 1) {
 		result = PyTuple_Pack(1, list_revs);
 		goto done;
 	}
 
 	/* Turn the python list into a native integer array (for efficiency) */
 	revs = (Py_ssize_t *)calloc(num_revs, sizeof(Py_ssize_t));
 	if (revs == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 	for (i = 0; i < num_revs; i++) {
 		Py_ssize_t revnum = PyInt_AsLong(PyList_GET_ITEM(list_revs, i));
 		if (revnum == -1 && PyErr_Occurred()) {
 			goto bail;
 		}
 		if (revnum < nullrev || revnum >= idxlen) {
 			PyErr_Format(PyExc_IndexError,
 			             "index out of range: %zd", revnum);
 			goto bail;
 		}
 		revs[i] = revnum;
 	}
 
 	/* Compute and check various property of the unsliced delta chain */
 	deltachainspan = index_segment_span(self, revs[0], revs[num_revs - 1]);
 	if (deltachainspan < 0) {
 		goto bail;
 	}
 
 	if (deltachainspan <= mingapsize) {
 		result = PyTuple_Pack(1, list_revs);
 		goto done;
 	}
 	chainpayload = 0;
 	for (i = 0; i < num_revs; i++) {
 		int tmp = index_get_length(self, revs[i]);
 		if (tmp < 0) {
 			goto bail;
 		}
 		chainpayload += tmp;
 	}
 
 	readdata = deltachainspan;
 	density = 1.0;
 
 	if (0 < deltachainspan) {
 		density = (double)chainpayload / (double)deltachainspan;
 	}
 
 	if (density >= targetdensity) {
 		result = PyTuple_Pack(1, list_revs);
 		goto done;
 	}
 
 	/* if chain is too sparse, look for relevant gaps */
 	gaps = (struct Gap *)calloc(num_revs, sizeof(struct Gap));
 	if (gaps == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	previous_end = -1;
 	for (i = 0; i < num_revs; i++) {
 		int64_t revstart;
 		int revsize;
 		revstart = index_get_start(self, revs[i]);
 		if (revstart < 0) {
 			goto bail;
 		};
 		revsize = index_get_length(self, revs[i]);
 		if (revsize < 0) {
 			goto bail;
 		};
 		if (revsize == 0) {
 			continue;
 		}
 		if (previous_end >= 0) {
 			int64_t gapsize = revstart - previous_end;
 			if (gapsize > mingapsize) {
 				gaps[num_gaps].size = gapsize;
 				gaps[num_gaps].idx = i;
 				num_gaps += 1;
 			}
 		}
 		previous_end = revstart + revsize;
 	}
 	if (num_gaps == 0) {
 		result = PyTuple_Pack(1, list_revs);
 		goto done;
 	}
 	qsort(gaps, num_gaps, sizeof(struct Gap), &gap_compare);
 
 	/* Slice the largest gap first, they improve the density the most */
 	selected_indices =
 	    (Py_ssize_t *)malloc((num_gaps + 1) * sizeof(Py_ssize_t));
 	if (selected_indices == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	for (i = num_gaps - 1; i >= 0; i--) {
 		selected_indices[num_selected] = gaps[i].idx;
 		readdata -= gaps[i].size;
 		num_selected += 1;
 		if (readdata <= 0) {
 			density = 1.0;
 		} else {
 			density = (double)chainpayload / (double)readdata;
 		}
 		if (density >= targetdensity) {
 			break;
 		}
 	}
 	qsort(selected_indices, num_selected, sizeof(Py_ssize_t),
 	      &Py_ssize_t_compare);
 
 	/* create the resulting slice */
 	allchunks = PyList_New(0);
 	if (allchunks == NULL) {
 		goto bail;
 	}
 	previdx = 0;
 	selected_indices[num_selected] = num_revs;
 	for (i = 0; i <= num_selected; i++) {
 		Py_ssize_t idx = selected_indices[i];
 		Py_ssize_t endidx = trim_endidx(self, revs, previdx, idx);
 		if (endidx < 0) {
 			goto bail;
 		}
 		if (previdx < endidx) {
 			chunk = PyList_GetSlice(list_revs, previdx, endidx);
 			if (chunk == NULL) {
 				goto bail;
 			}
 			if (PyList_Append(allchunks, chunk) == -1) {
 				goto bail;
 			}
 			Py_DECREF(chunk);
 			chunk = NULL;
 		}
 		previdx = idx;
 	}
 	result = allchunks;
 	goto done;
 
 bail:
 	Py_XDECREF(allchunks);
 	Py_XDECREF(chunk);
 done:
 	free(revs);
 	free(gaps);
 	free(selected_indices);
 	return result;
 }
 
 static inline int nt_level(const char *node, Py_ssize_t level)
 {
 	int v = node[level >> 1];
 	if (!(level & 1))
 		v >>= 4;
 	return v & 0xf;
 }
 
 /*
  * Return values:
  *
  *   -4: match is ambiguous (multiple candidates)
  *   -2: not found
  * rest: valid rev
  */
 static int nt_find(nodetree *self, const char *node, Py_ssize_t nodelen,
                    int hex)
 {
 	int (*getnybble)(const char *, Py_ssize_t) = hex ? hexdigit : nt_level;
 	int level, maxlevel, off;
 
 	/* If the input is binary, do a fast check for the nullid first. */
 	if (!hex && nodelen == self->nodelen && node[0] == '\0' &&
 	    node[1] == '\0' && memcmp(node, nullid, self->nodelen) == 0)
 		return -1;
 
 	if (hex)
 		maxlevel = nodelen;
 	else
 		maxlevel = 2 * nodelen;
 	if (maxlevel > 2 * self->nodelen)
 		maxlevel = 2 * self->nodelen;
 
 	for (level = off = 0; level < maxlevel; level++) {
 		int k = getnybble(node, level);
 		nodetreenode *n = &self->nodes[off];
 		int v = n->children[k];
 
 		if (v < 0) {
 			const char *n;
 			Py_ssize_t i;
 
 			v = -(v + 2);
 			n = index_node(self->index, v);
 			if (n == NULL)
 				return -2;
 			for (i = level; i < maxlevel; i++)
 				if (getnybble(node, i) != nt_level(n, i))
 					return -2;
 			return v;
 		}
 		if (v == 0)
 			return -2;
 		off = v;
 	}
 	/* multiple matches against an ambiguous prefix */
 	return -4;
 }
 
 static int nt_new(nodetree *self)
 {
 	if (self->length == self->capacity) {
 		size_t newcapacity;
 		nodetreenode *newnodes;
 		newcapacity = self->capacity * 2;
 		if (newcapacity >= SIZE_MAX / sizeof(nodetreenode)) {
 			PyErr_SetString(PyExc_MemoryError,
 			                "overflow in nt_new");
 			return -1;
 		}
 		newnodes =
 		    realloc(self->nodes, newcapacity * sizeof(nodetreenode));
 		if (newnodes == NULL) {
 			PyErr_SetString(PyExc_MemoryError, "out of memory");
 			return -1;
 		}
 		self->capacity = newcapacity;
 		self->nodes = newnodes;
 		memset(&self->nodes[self->length], 0,
 		       sizeof(nodetreenode) * (self->capacity - self->length));
 	}
 	return self->length++;
 }
 
 static int nt_insert(nodetree *self, const char *node, int rev)
 {
 	int level = 0;
 	int off = 0;
 
 	while (level < 2 * self->nodelen) {
 		int k = nt_level(node, level);
 		nodetreenode *n;
 		int v;
 
 		n = &self->nodes[off];
 		v = n->children[k];
 
 		if (v == 0) {
 			n->children[k] = -rev - 2;
 			return 0;
 		}
 		if (v < 0) {
 			const char *oldnode =
 			    index_node_existing(self->index, -(v + 2));
 			int noff;
 
 			if (oldnode == NULL)
 				return -1;
 			if (!memcmp(oldnode, node, self->nodelen)) {
 				n->children[k] = -rev - 2;
 				return 0;
 			}
 			noff = nt_new(self);
 			if (noff == -1)
 				return -1;
 			/* self->nodes may have been changed by realloc */
 			self->nodes[off].children[k] = noff;
 			off = noff;
 			n = &self->nodes[off];
 			n->children[nt_level(oldnode, ++level)] = v;
 			if (level > self->depth)
 				self->depth = level;
 			self->splits += 1;
 		} else {
 			level += 1;
 			off = v;
 		}
 	}
 
 	return -1;
 }
 
 static PyObject *ntobj_insert(nodetreeObject *self, PyObject *args)
 {
 	Py_ssize_t rev;
 	const char *node;
 	Py_ssize_t length;
 	if (!PyArg_ParseTuple(args, "n", &rev))
 		return NULL;
 	length = index_length(self->nt.index);
 	if (rev < 0 || rev >= length) {
 		PyErr_SetString(PyExc_ValueError, "revlog index out of range");
 		return NULL;
 	}
 	node = index_node_existing(self->nt.index, rev);
 	if (nt_insert(&self->nt, node, (int)rev) == -1)
 		return NULL;
 	Py_RETURN_NONE;
 }
 
 static int nt_delete_node(nodetree *self, const char *node)
 {
 	/* rev==-2 happens to get encoded as 0, which is interpreted as not set
 	 */
 	return nt_insert(self, node, -2);
 }
 
 static int nt_init(nodetree *self, indexObject *index, unsigned capacity)
 {
 	/* Initialize before overflow-checking to avoid nt_dealloc() crash. */
 	self->nodes = NULL;
 
 	self->index = index;
 	/* The input capacity is in terms of revisions, while the field is in
 	 * terms of nodetree nodes. */
 	self->capacity = (capacity < 4 ? 4 : capacity / 2);
 	self->nodelen = index->nodelen;
 	self->depth = 0;
 	self->splits = 0;
 	if (self->capacity > SIZE_MAX / sizeof(nodetreenode)) {
 		PyErr_SetString(PyExc_ValueError, "overflow in init_nt");
 		return -1;
 	}
 	self->nodes = calloc(self->capacity, sizeof(nodetreenode));
 	if (self->nodes == NULL) {
 		PyErr_NoMemory();
 		return -1;
 	}
 	self->length = 1;
 	return 0;
 }
 
 static int ntobj_init(nodetreeObject *self, PyObject *args)
 {
 	PyObject *index;
 	unsigned capacity;
 	if (!PyArg_ParseTuple(args, "O!I", &HgRevlogIndex_Type, &index,
 	                      &capacity))
 		return -1;
 	Py_INCREF(index);
 	return nt_init(&self->nt, (indexObject *)index, capacity);
 }
 
 static int nt_partialmatch(nodetree *self, const char *node, Py_ssize_t nodelen)
 {
 	return nt_find(self, node, nodelen, 1);
 }
 
 /*
  * Find the length of the shortest unique prefix of node.
  *
  * Return values:
  *
  *   -3: error (exception set)
  *   -2: not found (no exception set)
  * rest: length of shortest prefix
  */
 static int nt_shortest(nodetree *self, const char *node)
 {
 	int level, off;
 
 	for (level = off = 0; level < 2 * self->nodelen; level++) {
 		int k, v;
 		nodetreenode *n = &self->nodes[off];
 		k = nt_level(node, level);
 		v = n->children[k];
 		if (v < 0) {
 			const char *n;
 			v = -(v + 2);
 			n = index_node_existing(self->index, v);
 			if (n == NULL)
 				return -3;
 			if (memcmp(node, n, self->nodelen) != 0)
 				/*
 				 * Found a unique prefix, but it wasn't for the
 				 * requested node (i.e the requested node does
 				 * not exist).
 				 */
 				return -2;
 			return level + 1;
 		}
 		if (v == 0)
 			return -2;
 		off = v;
 	}
 	/*
 	 * The node was still not unique after 40 hex digits, so this won't
 	 * happen. Also, if we get here, then there's a programming error in
 	 * this file that made us insert a node longer than 40 hex digits.
 	 */
 	PyErr_SetString(PyExc_Exception, "broken node tree");
 	return -3;
 }
 
 static PyObject *ntobj_shortest(nodetreeObject *self, PyObject *args)
 {
 	PyObject *val;
 	char *node;
 	int length;
 
 	if (!PyArg_ParseTuple(args, "O", &val))
 		return NULL;
 	if (node_check(self->nt.nodelen, val, &node) == -1)
 		return NULL;
 
 	length = nt_shortest(&self->nt, node);
 	if (length == -3)
 		return NULL;
 	if (length == -2) {
 		raise_revlog_error();
 		return NULL;
 	}
 	return PyInt_FromLong(length);
 }
 
 static void nt_dealloc(nodetree *self)
 {
 	free(self->nodes);
 	self->nodes = NULL;
 }
 
 static void ntobj_dealloc(nodetreeObject *self)
 {
 	Py_XDECREF(self->nt.index);
 	nt_dealloc(&self->nt);
 	PyObject_Del(self);
 }
 
 static PyMethodDef ntobj_methods[] = {
     {"insert", (PyCFunction)ntobj_insert, METH_VARARGS,
      "insert an index entry"},
     {"shortest", (PyCFunction)ntobj_shortest, METH_VARARGS,
      "find length of shortest hex nodeid of a binary ID"},
     {NULL} /* Sentinel */
 };
 
 static PyTypeObject nodetreeType = {
     PyVarObject_HEAD_INIT(NULL, 0) /* header */
     "parsers.nodetree",            /* tp_name */
     sizeof(nodetreeObject),        /* tp_basicsize */
     0,                             /* tp_itemsize */
     (destructor)ntobj_dealloc,     /* tp_dealloc */
     0,                             /* tp_print */
     0,                             /* tp_getattr */
     0,                             /* tp_setattr */
     0,                             /* tp_compare */
     0,                             /* tp_repr */
     0,                             /* tp_as_number */
     0,                             /* tp_as_sequence */
     0,                             /* tp_as_mapping */
     0,                             /* tp_hash */
     0,                             /* tp_call */
     0,                             /* tp_str */
     0,                             /* tp_getattro */
     0,                             /* tp_setattro */
     0,                             /* tp_as_buffer */
     Py_TPFLAGS_DEFAULT,            /* tp_flags */
     "nodetree",                    /* tp_doc */
     0,                             /* tp_traverse */
     0,                             /* tp_clear */
     0,                             /* tp_richcompare */
     0,                             /* tp_weaklistoffset */
     0,                             /* tp_iter */
     0,                             /* tp_iternext */
     ntobj_methods,                 /* tp_methods */
     0,                             /* tp_members */
     0,                             /* tp_getset */
     0,                             /* tp_base */
     0,                             /* tp_dict */
     0,                             /* tp_descr_get */
     0,                             /* tp_descr_set */
     0,                             /* tp_dictoffset */
     (initproc)ntobj_init,          /* tp_init */
     0,                             /* tp_alloc */
 };
 
 static int index_init_nt(indexObject *self)
 {
 	if (!self->ntinitialized) {
 		if (nt_init(&self->nt, self, (int)self->length) == -1) {
 			nt_dealloc(&self->nt);
 			return -1;
 		}
 		if (nt_insert(&self->nt, nullid, -1) == -1) {
 			nt_dealloc(&self->nt);
 			return -1;
 		}
 		self->ntinitialized = 1;
 		self->ntrev = (int)index_length(self);
 		self->ntlookups = 1;
 		self->ntmisses = 0;
 	}
 	return 0;
 }
 
 /*
  * Return values:
  *
  *   -3: error (exception set)
  *   -2: not found (no exception set)
  * rest: valid rev
  */
 static int index_find_node(indexObject *self, const char *node)
 {
 	int rev;
 
 	if (index_init_nt(self) == -1)
 		return -3;
 
 	self->ntlookups++;
 	rev = nt_find(&self->nt, node, self->nodelen, 0);
 	if (rev >= -1)
 		return rev;
 
 	/*
 	 * For the first handful of lookups, we scan the entire index,
 	 * and cache only the matching nodes. This optimizes for cases
 	 * like "hg tip", where only a few nodes are accessed.
 	 *
 	 * After that, we cache every node we visit, using a single
 	 * scan amortized over multiple lookups.  This gives the best
 	 * bulk performance, e.g. for "hg log".
 	 */
 	if (self->ntmisses++ < 4) {
 		for (rev = self->ntrev - 1; rev >= 0; rev--) {
 			const char *n = index_node_existing(self, rev);
 			if (n == NULL)
 				return -3;
 			if (memcmp(node, n, self->nodelen) == 0) {
 				if (nt_insert(&self->nt, n, rev) == -1)
 					return -3;
 				break;
 			}
 		}
 	} else {
 		for (rev = self->ntrev - 1; rev >= 0; rev--) {
 			const char *n = index_node_existing(self, rev);
 			if (n == NULL)
 				return -3;
 			if (nt_insert(&self->nt, n, rev) == -1) {
 				self->ntrev = rev + 1;
 				return -3;
 			}
 			if (memcmp(node, n, self->nodelen) == 0) {
 				break;
 			}
 		}
 		self->ntrev = rev;
 	}
 
 	if (rev >= 0)
 		return rev;
 	return -2;
 }
 
 static PyObject *index_getitem(indexObject *self, PyObject *value)
 {
 	char *node;
 	int rev;
 
 	if (PyInt_Check(value)) {
 		long idx;
 		if (!pylong_to_long(value, &idx)) {
 			return NULL;
 		}
 		return index_get(self, idx);
 	}
 
 	if (node_check(self->nodelen, value, &node) == -1)
 		return NULL;
 	rev = index_find_node(self, node);
 	if (rev >= -1)
 		return PyInt_FromLong(rev);
 	if (rev == -2)
 		raise_revlog_error();
 	return NULL;
 }
 
 /*
  * Fully populate the radix tree.
  */
 static int index_populate_nt(indexObject *self)
 {
 	int rev;
 	if (self->ntrev > 0) {
 		for (rev = self->ntrev - 1; rev >= 0; rev--) {
 			const char *n = index_node_existing(self, rev);
 			if (n == NULL)
 				return -1;
 			if (nt_insert(&self->nt, n, rev) == -1)
 				return -1;
 		}
 		self->ntrev = -1;
 	}
 	return 0;
 }
 
 static PyObject *index_partialmatch(indexObject *self, PyObject *args)
 {
 	const char *fullnode;
 	Py_ssize_t nodelen;
 	char *node;
 	int rev, i;
 
 	if (!PyArg_ParseTuple(args, PY23("s#", "y#"), &node, &nodelen))
 		return NULL;
 
 	if (nodelen < 1) {
 		PyErr_SetString(PyExc_ValueError, "key too short");
 		return NULL;
 	}
 
 	if (nodelen > 2 * self->nodelen) {
 		PyErr_SetString(PyExc_ValueError, "key too long");
 		return NULL;
 	}
 
 	for (i = 0; i < nodelen; i++)
 		hexdigit(node, i);
 	if (PyErr_Occurred()) {
 		/* input contains non-hex characters */
 		PyErr_Clear();
 		Py_RETURN_NONE;
 	}
 
 	if (index_init_nt(self) == -1)
 		return NULL;
 	if (index_populate_nt(self) == -1)
 		return NULL;
 	rev = nt_partialmatch(&self->nt, node, nodelen);
 
 	switch (rev) {
 	case -4:
 		raise_revlog_error();
 		return NULL;
 	case -2:
 		Py_RETURN_NONE;
 	case -1:
 		return PyBytes_FromStringAndSize(nullid, self->nodelen);
 	}
 
 	fullnode = index_node_existing(self, rev);
 	if (fullnode == NULL) {
 		return NULL;
 	}
 	return PyBytes_FromStringAndSize(fullnode, self->nodelen);
 }
 
 static PyObject *index_shortest(indexObject *self, PyObject *args)
 {
 	PyObject *val;
 	char *node;
 	int length;
 
 	if (!PyArg_ParseTuple(args, "O", &val))
 		return NULL;
 	if (node_check(self->nodelen, val, &node) == -1)
 		return NULL;
 
 	self->ntlookups++;
 	if (index_init_nt(self) == -1)
 		return NULL;
 	if (index_populate_nt(self) == -1)
 		return NULL;
 	length = nt_shortest(&self->nt, node);
 	if (length == -3)
 		return NULL;
 	if (length == -2) {
 		raise_revlog_error();
 		return NULL;
 	}
 	return PyInt_FromLong(length);
 }
 
 static PyObject *index_m_get(indexObject *self, PyObject *args)
 {
 	PyObject *val;
 	char *node;
 	int rev;
 
 	if (!PyArg_ParseTuple(args, "O", &val))
 		return NULL;
 	if (node_check(self->nodelen, val, &node) == -1)
 		return NULL;
 	rev = index_find_node(self, node);
 	if (rev == -3)
 		return NULL;
 	if (rev == -2)
 		Py_RETURN_NONE;
 	return PyInt_FromLong(rev);
 }
 
 static int index_contains(indexObject *self, PyObject *value)
 {
 	char *node;
 
 	if (PyInt_Check(value)) {
 		long rev;
 		if (!pylong_to_long(value, &rev)) {
 			return -1;
 		}
 		return rev >= -1 && rev < index_length(self);
 	}
 
 	if (node_check(self->nodelen, value, &node) == -1)
 		return -1;
 
 	switch (index_find_node(self, node)) {
 	case -3:
 		return -1;
 	case -2:
 		return 0;
 	default:
 		return 1;
 	}
 }
 
 static PyObject *index_m_has_node(indexObject *self, PyObject *args)
 {
 	int ret = index_contains(self, args);
 	if (ret < 0)
 		return NULL;
 	return PyBool_FromLong((long)ret);
 }
 
 static PyObject *index_m_rev(indexObject *self, PyObject *val)
 {
 	char *node;
 	int rev;
 
 	if (node_check(self->nodelen, val, &node) == -1)
 		return NULL;
 	rev = index_find_node(self, node);
 	if (rev >= -1)
 		return PyInt_FromLong(rev);
 	if (rev == -2)
 		raise_revlog_error();
 	return NULL;
 }
 
 typedef uint64_t bitmask;
 
 /*
  * Given a disjoint set of revs, return all candidates for the
  * greatest common ancestor. In revset notation, this is the set
  * "heads(::a and ::b and ...)"
  */
 static PyObject *find_gca_candidates(indexObject *self, const int *revs,
                                      int revcount)
 {
 	const bitmask allseen = (1ull << revcount) - 1;
 	const bitmask poison = 1ull << revcount;
 	PyObject *gca = PyList_New(0);
 	int i, v, interesting;
 	int maxrev = -1;
 	bitmask sp;
 	bitmask *seen;
 
 	if (gca == NULL)
 		return PyErr_NoMemory();
 
 	for (i = 0; i < revcount; i++) {
 		if (revs[i] > maxrev)
 			maxrev = revs[i];
 	}
 
 	seen = calloc(sizeof(*seen), maxrev + 1);
 	if (seen == NULL) {
 		Py_DECREF(gca);
 		return PyErr_NoMemory();
 	}
 
 	for (i = 0; i < revcount; i++)
 		seen[revs[i]] = 1ull << i;
 
 	interesting = revcount;
 
 	for (v = maxrev; v >= 0 && interesting; v--) {
 		bitmask sv = seen[v];
 		int parents[2];
 
 		if (!sv)
 			continue;
 
 		if (sv < poison) {
 			interesting -= 1;
 			if (sv == allseen) {
 				PyObject *obj = PyInt_FromLong(v);
 				if (obj == NULL)
 					goto bail;
 				if (PyList_Append(gca, obj) == -1) {
 					Py_DECREF(obj);
 					goto bail;
 				}
 				sv |= poison;
 				for (i = 0; i < revcount; i++) {
 					if (revs[i] == v)
 						goto done;
 				}
 			}
 		}
 		if (index_get_parents(self, v, parents, maxrev) < 0)
 			goto bail;
 
 		for (i = 0; i < 2; i++) {
 			int p = parents[i];
 			if (p == -1)
 				continue;
 			sp = seen[p];
 			if (sv < poison) {
 				if (sp == 0) {
 					seen[p] = sv;
 					interesting++;
 				} else if (sp != sv)
 					seen[p] |= sv;
 			} else {
 				if (sp && sp < poison)
 					interesting--;
 				seen[p] = sv;
 			}
 		}
 	}
 
 done:
 	free(seen);
 	return gca;
 bail:
 	free(seen);
 	Py_XDECREF(gca);
 	return NULL;
 }
 
 /*
  * Given a disjoint set of revs, return the subset with the longest
  * path to the root.
  */
 static PyObject *find_deepest(indexObject *self, PyObject *revs)
 {
 	const Py_ssize_t revcount = PyList_GET_SIZE(revs);
 	static const Py_ssize_t capacity = 24;
 	int *depth, *interesting = NULL;
 	int i, j, v, ninteresting;
 	PyObject *dict = NULL, *keys = NULL;
 	long *seen = NULL;
 	int maxrev = -1;
 	long final;
 
 	if (revcount > capacity) {
 		PyErr_Format(PyExc_OverflowError,
 		             "bitset size (%ld) > capacity (%ld)",
 		             (long)revcount, (long)capacity);
 		return NULL;
 	}
 
 	for (i = 0; i < revcount; i++) {
 		int n = (int)PyInt_AsLong(PyList_GET_ITEM(revs, i));
 		if (n > maxrev)
 			maxrev = n;
 	}
 
 	depth = calloc(sizeof(*depth), maxrev + 1);
 	if (depth == NULL)
 		return PyErr_NoMemory();
 
 	seen = calloc(sizeof(*seen), maxrev + 1);
 	if (seen == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	interesting = calloc(sizeof(*interesting), ((size_t)1) << revcount);
 	if (interesting == NULL) {
 		PyErr_NoMemory();
 		goto bail;
 	}
 
 	if (PyList_Sort(revs) == -1)
 		goto bail;
 
 	for (i = 0; i < revcount; i++) {
 		int n = (int)PyInt_AsLong(PyList_GET_ITEM(revs, i));
 		long b = 1l << i;
 		depth[n] = 1;
 		seen[n] = b;
 		interesting[b] = 1;
 	}
 
 	/* invariant: ninteresting is the number of non-zero entries in
 	 * interesting. */
 	ninteresting = (int)revcount;
 
 	for (v = maxrev; v >= 0 && ninteresting > 1; v--) {
 		int dv = depth[v];
 		int parents[2];
 		long sv;
 
 		if (dv == 0)
 			continue;
 
 		sv = seen[v];
 		if (index_get_parents(self, v, parents, maxrev) < 0)
 			goto bail;
 
 		for (i = 0; i < 2; i++) {
 			int p = parents[i];
 			long sp;
 			int dp;
 
 			if (p == -1)
 				continue;
 
 			dp = depth[p];
 			sp = seen[p];
 			if (dp <= dv) {
 				depth[p] = dv + 1;
 				if (sp != sv) {
 					interesting[sv] += 1;
 					seen[p] = sv;
 					if (sp) {
 						interesting[sp] -= 1;
 						if (interesting[sp] == 0)
 							ninteresting -= 1;
 					}
 				}
 			} else if (dv == dp - 1) {
 				long nsp = sp | sv;
 				if (nsp == sp)
 					continue;
 				seen[p] = nsp;
 				interesting[sp] -= 1;
 				if (interesting[sp] == 0)
 					ninteresting -= 1;
 				if (interesting[nsp] == 0)
 					ninteresting += 1;
 				interesting[nsp] += 1;
 			}
 		}
 		interesting[sv] -= 1;
 		if (interesting[sv] == 0)
 			ninteresting -= 1;
 	}
 
 	final = 0;
 	j = ninteresting;
 	for (i = 0; i < (int)(2 << revcount) && j > 0; i++) {
 		if (interesting[i] == 0)
 			continue;
 		final |= i;
 		j -= 1;
 	}
 	if (final == 0) {
 		keys = PyList_New(0);
 		goto bail;
 	}
 
 	dict = PyDict_New();
 	if (dict == NULL)
 		goto bail;
 
 	for (i = 0; i < revcount; i++) {
 		PyObject *key;
 
 		if ((final & (1 << i)) == 0)
 			continue;
 
 		key = PyList_GET_ITEM(revs, i);
 		Py_INCREF(key);
 		Py_INCREF(Py_None);
 		if (PyDict_SetItem(dict, key, Py_None) == -1) {
 			Py_DECREF(key);
 			Py_DECREF(Py_None);
 			goto bail;
 		}
 	}
 
 	keys = PyDict_Keys(dict);
 
 bail:
 	free(depth);
 	free(seen);
 	free(interesting);
 	Py_XDECREF(dict);
 
 	return keys;
 }
 
 /*
  * Given a (possibly overlapping) set of revs, return all the
  * common ancestors heads: heads(::args[0] and ::a[1] and ...)
  */
 static PyObject *index_commonancestorsheads(indexObject *self, PyObject *args)
 {
 	PyObject *ret = NULL;
 	Py_ssize_t argcount, i, len;
 	bitmask repeat = 0;
 	int revcount = 0;
 	int *revs;
 
 	argcount = PySequence_Length(args);
 	revs = PyMem_Malloc(argcount * sizeof(*revs));
 	if (argcount > 0 && revs == NULL)
 		return PyErr_NoMemory();
 	len = index_length(self);
 
 	for (i = 0; i < argcount; i++) {
 		static const int capacity = 24;
 		PyObject *obj = PySequence_GetItem(args, i);
 		bitmask x;
 		long val;
 
 		if (!PyInt_Check(obj)) {
 			PyErr_SetString(PyExc_TypeError,
 			                "arguments must all be ints");
 			Py_DECREF(obj);
 			goto bail;
 		}
 		val = PyInt_AsLong(obj);
 		Py_DECREF(obj);
 		if (val == -1) {
 			ret = PyList_New(0);
 			goto done;
 		}
 		if (val < 0 || val >= len) {
 			PyErr_SetString(PyExc_IndexError, "index out of range");
 			goto bail;
 		}
 		/* this cheesy bloom filter lets us avoid some more
 		 * expensive duplicate checks in the common set-is-disjoint
 		 * case */
 		x = 1ull << (val & 0x3f);
 		if (repeat & x) {
 			int k;
 			for (k = 0; k < revcount; k++) {
 				if (val == revs[k])
 					goto duplicate;
 			}
 		} else
 			repeat |= x;
 		if (revcount >= capacity) {
 			PyErr_Format(PyExc_OverflowError,
 			             "bitset size (%d) > capacity (%d)",
 			             revcount, capacity);
 			goto bail;
 		}
 		revs[revcount++] = (int)val;
 	duplicate:;
 	}
 
 	if (revcount == 0) {
 		ret = PyList_New(0);
 		goto done;
 	}
 	if (revcount == 1) {
 		PyObject *obj;
 		ret = PyList_New(1);
 		if (ret == NULL)
 			goto bail;
 		obj = PyInt_FromLong(revs[0]);
 		if (obj == NULL)
 			goto bail;
 		PyList_SET_ITEM(ret, 0, obj);
 		goto done;
 	}
 
 	ret = find_gca_candidates(self, revs, revcount);
 	if (ret == NULL)
 		goto bail;
 
 done:
 	PyMem_Free(revs);
 	return ret;
 
 bail:
 	PyMem_Free(revs);
 	Py_XDECREF(ret);
 	return NULL;
 }
 
 /*
  * Given a (possibly overlapping) set of revs, return the greatest
  * common ancestors: those with the longest path to the root.
  */
 static PyObject *index_ancestors(indexObject *self, PyObject *args)
 {
 	PyObject *ret;
 	PyObject *gca = index_commonancestorsheads(self, args);
 	if (gca == NULL)
 		return NULL;
 
 	if (PyList_GET_SIZE(gca) <= 1) {
 		return gca;
 	}
 
 	ret = find_deepest(self, gca);
 	Py_DECREF(gca);
 	return ret;
 }
 
 /*
  * Invalidate any trie entries introduced by added revs.
  */
 static void index_invalidate_added(indexObject *self, Py_ssize_t start)
 {
 	Py_ssize_t i, len;
 
 	len = self->length + self->new_length;
 	i = start - self->length;
 	if (i < 0)
 		return;
 
 	for (i = start; i < len; i++)
 		nt_delete_node(&self->nt, index_deref(self, i) + 32);
 
 	self->new_length = start - self->length;
 }
 
 /*
  * Delete a numeric range of revs, which must be at the end of the
  * range.
  */
 static int index_slice_del(indexObject *self, PyObject *item)
 {
 	Py_ssize_t start, stop, step, slicelength;
 	Py_ssize_t length = index_length(self) + 1;
 	int ret = 0;
 
 /* Argument changed from PySliceObject* to PyObject* in Python 3. */
 #ifdef IS_PY3K
 	if (PySlice_GetIndicesEx(item, length, &start, &stop, &step,
 	                         &slicelength) < 0)
 #else
 	if (PySlice_GetIndicesEx((PySliceObject *)item, length, &start, &stop,
 	                         &step, &slicelength) < 0)
 #endif
 		return -1;
 
 	if (slicelength <= 0)
 		return 0;
 
 	if ((step < 0 && start < stop) || (step > 0 && start > stop))
 		stop = start;
 
 	if (step < 0) {
 		stop = start + 1;
 		start = stop + step * (slicelength - 1) - 1;
 		step = -step;
 	}
 
 	if (step != 1) {
 		PyErr_SetString(PyExc_ValueError,
 		                "revlog index delete requires step size of 1");
 		return -1;
 	}
 
 	if (stop != length - 1) {
 		PyErr_SetString(PyExc_IndexError,
 		                "revlog index deletion indices are invalid");
 		return -1;
 	}
 
 	if (start < self->length) {
 		if (self->ntinitialized) {
 			Py_ssize_t i;
 
 			for (i = start; i < self->length; i++) {
 				const char *node = index_node_existing(self, i);
 				if (node == NULL)
 					return -1;
 
 				nt_delete_node(&self->nt, node);
 			}
 			if (self->new_length)
 				index_invalidate_added(self, self->length);
 			if (self->ntrev > start)
 				self->ntrev = (int)start;
 		} else if (self->new_length) {
 			self->new_length = 0;
 		}
 
 		self->length = start;
 		goto done;
 	}
 
 	if (self->ntinitialized) {
 		index_invalidate_added(self, start);
 		if (self->ntrev > start)
 			self->ntrev = (int)start;
 	} else {
 		self->new_length = start - self->length;
 	}
 done:
 	Py_CLEAR(self->headrevs);
 	return ret;
 }
 
 /*
  * Supported ops:
  *
  * slice deletion
  * string assignment (extend node->rev mapping)
  * string deletion (shrink node->rev mapping)
  */
 static int index_assign_subscript(indexObject *self, PyObject *item,
                                   PyObject *value)
 {
 	char *node;
 	long rev;
 
 	if (PySlice_Check(item) && value == NULL)
 		return index_slice_del(self, item);
 
 	if (node_check(self->nodelen, item, &node) == -1)
 		return -1;
 
 	if (value == NULL)
 		return self->ntinitialized ? nt_delete_node(&self->nt, node)
 		                           : 0;
 	rev = PyInt_AsLong(value);
 	if (rev > INT_MAX || rev < 0) {
 		if (!PyErr_Occurred())
 			PyErr_SetString(PyExc_ValueError, "rev out of range");
 		return -1;
 	}
 
 	if (index_init_nt(self) == -1)
 		return -1;
 	return nt_insert(&self->nt, node, (int)rev);
 }
 
 /*
  * Find all RevlogNG entries in an index that has inline data. Update
  * the optional "offsets" table with those entries.
  */
 static Py_ssize_t inline_scan(indexObject *self, const char **offsets)
 {
 	const char *data = (const char *)self->buf.buf;
 	Py_ssize_t pos = 0;
 	Py_ssize_t end = self->buf.len;
 	long incr = self->hdrsize;
 	Py_ssize_t len = 0;
 
 	while (pos + self->hdrsize <= end && pos >= 0) {
 		uint32_t comp_len, sidedata_comp_len = 0;
 		/* 3rd element of header is length of compressed inline data */
 		comp_len = getbe32(data + pos + 8);
 		if (self->hdrsize == v2_hdrsize) {
 			sidedata_comp_len = getbe32(data + pos + 72);
 		}
 		incr = self->hdrsize + comp_len + sidedata_comp_len;
 		if (offsets)
 			offsets[len] = data + pos;
 		len++;
 		pos += incr;
 	}
 
 	if (pos != end) {
 		if (!PyErr_Occurred())
 			PyErr_SetString(PyExc_ValueError, "corrupt index file");
 		return -1;
 	}
 
 	return len;
 }
 
 static int index_init(indexObject *self, PyObject *args, PyObject *kwargs)
 {
 	PyObject *data_obj, *inlined_obj, *revlogv2;
 	Py_ssize_t size;
 
 	static char *kwlist[] = {"data", "inlined", "revlogv2", NULL};
 
 	/* Initialize before argument-checking to avoid index_dealloc() crash.
 	 */
 	self->added = NULL;
 	self->new_length = 0;
 	self->added_length = 0;
 	self->data = NULL;
 	memset(&self->buf, 0, sizeof(self->buf));
 	self->headrevs = NULL;
 	self->filteredrevs = Py_None;
 	Py_INCREF(Py_None);
 	self->ntinitialized = 0;
 	self->offsets = NULL;
 	self->nodelen = 20;
 	self->nullentry = NULL;
 
 	revlogv2 = NULL;
 	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "OO|O", kwlist,
 	                                 &data_obj, &inlined_obj, &revlogv2))
 		return -1;
 	if (!PyObject_CheckBuffer(data_obj)) {
 		PyErr_SetString(PyExc_TypeError,
 		                "data does not support buffer interface");
 		return -1;
 	}
 	if (self->nodelen < 20 || self->nodelen > (Py_ssize_t)sizeof(nullid)) {
 		PyErr_SetString(PyExc_RuntimeError, "unsupported node size");
 		return -1;
 	}
 
 	if (revlogv2 && PyObject_IsTrue(revlogv2)) {
 		self->hdrsize = v2_hdrsize;
 	} else {
 		self->hdrsize = v1_hdrsize;
 	}
 
 	if (self->hdrsize == v1_hdrsize) {
 		self->nullentry =
 		    Py_BuildValue(PY23("iiiiiiis#", "iiiiiiiy#"), 0, 0, 0, -1,
 		                  -1, -1, -1, nullid, self->nodelen);
 	} else {
 		self->nullentry =
 		    Py_BuildValue(PY23("iiiiiiis#ii", "iiiiiiiy#ii"), 0, 0, 0,
 		                  -1, -1, -1, -1, nullid, self->nodelen, 0, 0);
 	}
 
 	if (!self->nullentry)
 		return -1;
 	PyObject_GC_UnTrack(self->nullentry);
 
 	if (PyObject_GetBuffer(data_obj, &self->buf, PyBUF_SIMPLE) == -1)
 		return -1;
 	size = self->buf.len;
 
 	self->inlined = inlined_obj && PyObject_IsTrue(inlined_obj);
 	self->data = data_obj;
 
 	self->ntlookups = self->ntmisses = 0;
 	self->ntrev = -1;
 	Py_INCREF(self->data);
 
 	if (self->inlined) {
 		Py_ssize_t len = inline_scan(self, NULL);
 		if (len == -1)
 			goto bail;
 		self->length = len;
 	} else {
 		if (size % self->hdrsize) {
 			PyErr_SetString(PyExc_ValueError, "corrupt index file");
 			goto bail;
 		}
 		self->length = size / self->hdrsize;
 	}
 
 	return 0;
 bail:
 	return -1;
 }
 
 static PyObject *index_nodemap(indexObject *self)
 {
 	Py_INCREF(self);
 	return (PyObject *)self;
 }
 
 static void _index_clearcaches(indexObject *self)
 {
 	if (self->offsets) {
 		PyMem_Free((void *)self->offsets);
 		self->offsets = NULL;
 	}
 	if (self->ntinitialized) {
 		nt_dealloc(&self->nt);
 	}
 	self->ntinitialized = 0;
 	Py_CLEAR(self->headrevs);
 }
 
 static PyObject *index_clearcaches(indexObject *self)
 {
 	_index_clearcaches(self);
 	self->ntrev = -1;
 	self->ntlookups = self->ntmisses = 0;
 	Py_RETURN_NONE;
 }
 
 static void index_dealloc(indexObject *self)
 {
 	_index_clearcaches(self);
 	Py_XDECREF(self->filteredrevs);
 	if (self->buf.buf) {
 		PyBuffer_Release(&self->buf);
 		memset(&self->buf, 0, sizeof(self->buf));
 	}
 	Py_XDECREF(self->data);
 	PyMem_Free(self->added);
 	Py_XDECREF(self->nullentry);
 	PyObject_Del(self);
 }
 
 static PySequenceMethods index_sequence_methods = {
     (lenfunc)index_length,      /* sq_length */
     0,                          /* sq_concat */
     0,                          /* sq_repeat */
     (ssizeargfunc)index_get,    /* sq_item */
     0,                          /* sq_slice */
     0,                          /* sq_ass_item */
     0,                          /* sq_ass_slice */
     (objobjproc)index_contains, /* sq_contains */
 };
 
 static PyMappingMethods index_mapping_methods = {
     (lenfunc)index_length,                 /* mp_length */
     (binaryfunc)index_getitem,             /* mp_subscript */
     (objobjargproc)index_assign_subscript, /* mp_ass_subscript */
 };
 
 static PyMethodDef index_methods[] = {
     {"ancestors", (PyCFunction)index_ancestors, METH_VARARGS,
      "return the gca set of the given revs"},
     {"commonancestorsheads", (PyCFunction)index_commonancestorsheads,
      METH_VARARGS,
      "return the heads of the common ancestors of the given revs"},
     {"clearcaches", (PyCFunction)index_clearcaches, METH_NOARGS,
      "clear the index caches"},
     {"get", (PyCFunction)index_m_get, METH_VARARGS, "get an index entry"},
     {"get_rev", (PyCFunction)index_m_get, METH_VARARGS,
      "return `rev` associated with a node or None"},
     {"has_node", (PyCFunction)index_m_has_node, METH_O,
      "return True if the node exist in the index"},
     {"rev", (PyCFunction)index_m_rev, METH_O,
      "return `rev` associated with a node or raise RevlogError"},
     {"computephasesmapsets", (PyCFunction)compute_phases_map_sets, METH_VARARGS,
      "compute phases"},
     {"reachableroots2", (PyCFunction)reachableroots2, METH_VARARGS,
      "reachableroots"},
     {"replace_sidedata_info", (PyCFunction)index_replace_sidedata_info,
      METH_VARARGS, "replace an existing index entry with a new value"},
     {"headrevs", (PyCFunction)index_headrevs, METH_VARARGS,
      "get head revisions"}, /* Can do filtering since 3.2 */
     {"headrevsfiltered", (PyCFunction)index_headrevs, METH_VARARGS,
      "get filtered head revisions"}, /* Can always do filtering */
     {"issnapshot", (PyCFunction)index_issnapshot, METH_O,
      "True if the object is a snapshot"},
     {"findsnapshots", (PyCFunction)index_findsnapshots, METH_VARARGS,
      "Gather snapshot data in a cache dict"},
     {"deltachain", (PyCFunction)index_deltachain, METH_VARARGS,
      "determine revisions with deltas to reconstruct fulltext"},
     {"slicechunktodensity", (PyCFunction)index_slicechunktodensity,
      METH_VARARGS, "determine revisions with deltas to reconstruct fulltext"},
     {"append", (PyCFunction)index_append, METH_O, "append an index entry"},
     {"partialmatch", (PyCFunction)index_partialmatch, METH_VARARGS,
      "match a potentially ambiguous node ID"},
     {"shortest", (PyCFunction)index_shortest, METH_VARARGS,
      "find length of shortest hex nodeid of a binary ID"},
     {"stats", (PyCFunction)index_stats, METH_NOARGS, "stats for the index"},
     {"entry_binary", (PyCFunction)index_entry_binary, METH_O,
      "return an entry in binary form"},
     {"pack_header", (PyCFunction)index_pack_header, METH_VARARGS,
      "pack the revlog header information into binary"},
     {NULL} /* Sentinel */
 };
 
 static PyGetSetDef index_getset[] = {
     {"nodemap", (getter)index_nodemap, NULL, "nodemap", NULL},
     {NULL} /* Sentinel */
 };
 
 static PyMemberDef index_members[] = {
     {"entry_size", T_LONG, offsetof(indexObject, hdrsize), 0,
      "size of an index entry"},
     {NULL} /* Sentinel */
 };
 
 PyTypeObject HgRevlogIndex_Type = {
     PyVarObject_HEAD_INIT(NULL, 0) /* header */
     "parsers.index",               /* tp_name */
     sizeof(indexObject),           /* tp_basicsize */
     0,                             /* tp_itemsize */
     (destructor)index_dealloc,     /* tp_dealloc */
     0,                             /* tp_print */
     0,                             /* tp_getattr */
     0,                             /* tp_setattr */
     0,                             /* tp_compare */
     0,                             /* tp_repr */
     0,                             /* tp_as_number */
     &index_sequence_methods,       /* tp_as_sequence */
     &index_mapping_methods,        /* tp_as_mapping */
     0,                             /* tp_hash */
     0,                             /* tp_call */
     0,                             /* tp_str */
     0,                             /* tp_getattro */
     0,                             /* tp_setattro */
     0,                             /* tp_as_buffer */
     Py_TPFLAGS_DEFAULT,            /* tp_flags */
     "revlog index",                /* tp_doc */
     0,                             /* tp_traverse */
     0,                             /* tp_clear */
     0,                             /* tp_richcompare */
     0,                             /* tp_weaklistoffset */
     0,                             /* tp_iter */
     0,                             /* tp_iternext */
     index_methods,                 /* tp_methods */
     index_members,                 /* tp_members */
     index_getset,                  /* tp_getset */
     0,                             /* tp_base */
     0,                             /* tp_dict */
     0,                             /* tp_descr_get */
     0,                             /* tp_descr_set */
     0,                             /* tp_dictoffset */
     (initproc)index_init,          /* tp_init */
     0,                             /* tp_alloc */
 };
 
 /*
  * returns a tuple of the form (index, cache) with elements as
  * follows:
  *
  * index: an index object that lazily parses Revlog (v1 or v2) records
  * cache: if data is inlined, a tuple (0, index_file_content), else None
  *        index_file_content could be a string, or a buffer
  *
  * added complications are for backwards compatibility
  */
 PyObject *parse_index2(PyObject *self, PyObject *args, PyObject *kwargs)
 {
 	PyObject *cache = NULL;
 	indexObject *idx;
 	int ret;
 
 	idx = PyObject_New(indexObject, &HgRevlogIndex_Type);
 	if (idx == NULL)
 		goto bail;
 
 	ret = index_init(idx, args, kwargs);
 	if (ret == -1)
 		goto bail;
 
 	if (idx->inlined) {
 		cache = Py_BuildValue("iO", 0, idx->data);
 		if (cache == NULL)
 			goto bail;
 	} else {
 		cache = Py_None;
 		Py_INCREF(cache);
 	}
 
 	return Py_BuildValue("NN", idx, cache);
 
 bail:
 	Py_XDECREF(idx);
 	Py_XDECREF(cache);
 	return NULL;
 }
 
 static Revlog_CAPI CAPI = {
     /* increment the abi_version field upon each change in the Revlog_CAPI
        struct or in the ABI of the listed functions */
     2,
     index_length,
     index_node,
     HgRevlogIndex_GetParents,
 };
 
 void revlog_module_init(PyObject *mod)
 {
 	PyObject *caps = NULL;
 	HgRevlogIndex_Type.tp_new = PyType_GenericNew;
 	if (PyType_Ready(&HgRevlogIndex_Type) < 0)
 		return;
 	Py_INCREF(&HgRevlogIndex_Type);
 	PyModule_AddObject(mod, "index", (PyObject *)&HgRevlogIndex_Type);
 
 	nodetreeType.tp_new = PyType_GenericNew;
 	if (PyType_Ready(&nodetreeType) < 0)
 		return;
 	Py_INCREF(&nodetreeType);
 	PyModule_AddObject(mod, "nodetree", (PyObject *)&nodetreeType);
 
 	caps = PyCapsule_New(&CAPI, "mercurial.cext.parsers.revlog_CAPI", NULL);
 	if (caps != NULL)
 		PyModule_AddObject(mod, "revlog_CAPI", caps);
 }

mercurial/interfaces/repository.py

@@ -1,2028 +1,2028 b''
 # repository.py - Interfaces and base classes for repositories and peers.
 #
 # Copyright 2017 Gregory Szorc <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 from ..i18n import _
 from .. import error
 from . import util as interfaceutil
 
 # Local repository feature string.
 
 # Revlogs are being used for file storage.
 REPO_FEATURE_REVLOG_FILE_STORAGE = b'revlogfilestorage'
 # The storage part of the repository is shared from an external source.
 REPO_FEATURE_SHARED_STORAGE = b'sharedstore'
 # LFS supported for backing file storage.
 REPO_FEATURE_LFS = b'lfs'
 # Repository supports being stream cloned.
 REPO_FEATURE_STREAM_CLONE = b'streamclone'
 # Files storage may lack data for all ancestors.
 REPO_FEATURE_SHALLOW_FILE_STORAGE = b'shallowfilestorage'
 
 REVISION_FLAG_CENSORED = 1 << 15
 REVISION_FLAG_ELLIPSIS = 1 << 14
 REVISION_FLAG_EXTSTORED = 1 << 13
 REVISION_FLAG_HASCOPIESINFO = 1 << 12
 
 REVISION_FLAGS_KNOWN = (
     REVISION_FLAG_CENSORED
     | REVISION_FLAG_ELLIPSIS
     | REVISION_FLAG_EXTSTORED
     | REVISION_FLAG_HASCOPIESINFO
 )
 
 CG_DELTAMODE_STD = b'default'
 CG_DELTAMODE_PREV = b'previous'
 CG_DELTAMODE_FULL = b'fulltext'
 CG_DELTAMODE_P1 = b'p1'
 
 
 class ipeerconnection(interfaceutil.Interface):
     """Represents a "connection" to a repository.
 
     This is the base interface for representing a connection to a repository.
     It holds basic properties and methods applicable to all peer types.
 
     This is not a complete interface definition and should not be used
     outside of this module.
     """
 
     ui = interfaceutil.Attribute("""ui.ui instance""")
 
     def url():
         """Returns a URL string representing this peer.
 
         Currently, implementations expose the raw URL used to construct the
         instance. It may contain credentials as part of the URL. The
         expectations of the value aren't well-defined and this could lead to
         data leakage.
 
         TODO audit/clean consumers and more clearly define the contents of this
         value.
         """
 
     def local():
         """Returns a local repository instance.
 
         If the peer represents a local repository, returns an object that
         can be used to interface with it. Otherwise returns ``None``.
         """
 
     def peer():
         """Returns an object conforming to this interface.
 
         Most implementations will ``return self``.
         """
 
     def canpush():
         """Returns a boolean indicating if this peer can be pushed to."""
 
     def close():
         """Close the connection to this peer.
 
         This is called when the peer will no longer be used. Resources
         associated with the peer should be cleaned up.
         """
 
 
 class ipeercapabilities(interfaceutil.Interface):
     """Peer sub-interface related to capabilities."""
 
     def capable(name):
         """Determine support for a named capability.
 
         Returns ``False`` if capability not supported.
 
         Returns ``True`` if boolean capability is supported. Returns a string
         if capability support is non-boolean.
 
         Capability strings may or may not map to wire protocol capabilities.
         """
 
     def requirecap(name, purpose):
         """Require a capability to be present.
 
         Raises a ``CapabilityError`` if the capability isn't present.
         """
 
 
 class ipeercommands(interfaceutil.Interface):
     """Client-side interface for communicating over the wire protocol.
 
     This interface is used as a gateway to the Mercurial wire protocol.
     methods commonly call wire protocol commands of the same name.
     """
 
     def branchmap():
         """Obtain heads in named branches.
 
         Returns a dict mapping branch name to an iterable of nodes that are
         heads on that branch.
         """
 
     def capabilities():
         """Obtain capabilities of the peer.
 
         Returns a set of string capabilities.
         """
 
     def clonebundles():
         """Obtains the clone bundles manifest for the repo.
 
         Returns the manifest as unparsed bytes.
         """
 
     def debugwireargs(one, two, three=None, four=None, five=None):
         """Used to facilitate debugging of arguments passed over the wire."""
 
     def getbundle(source, **kwargs):
         """Obtain remote repository data as a bundle.
 
         This command is how the bulk of repository data is transferred from
         the peer to the local repository
 
         Returns a generator of bundle data.
         """
 
     def heads():
         """Determine all known head revisions in the peer.
 
         Returns an iterable of binary nodes.
         """
 
     def known(nodes):
         """Determine whether multiple nodes are known.
 
         Accepts an iterable of nodes whose presence to check for.
 
         Returns an iterable of booleans indicating of the corresponding node
         at that index is known to the peer.
         """
 
     def listkeys(namespace):
         """Obtain all keys in a pushkey namespace.
 
         Returns an iterable of key names.
         """
 
     def lookup(key):
         """Resolve a value to a known revision.
 
         Returns a binary node of the resolved revision on success.
         """
 
     def pushkey(namespace, key, old, new):
         """Set a value using the ``pushkey`` protocol.
 
         Arguments correspond to the pushkey namespace and key to operate on and
         the old and new values for that key.
 
         Returns a string with the peer result. The value inside varies by the
         namespace.
         """
 
     def stream_out():
         """Obtain streaming clone data.
 
         Successful result should be a generator of data chunks.
         """
 
     def unbundle(bundle, heads, url):
         """Transfer repository data to the peer.
 
         This is how the bulk of data during a push is transferred.
 
         Returns the integer number of heads added to the peer.
         """
 
 
 class ipeerlegacycommands(interfaceutil.Interface):
     """Interface for implementing support for legacy wire protocol commands.
 
     Wire protocol commands transition to legacy status when they are no longer
     used by modern clients. To facilitate identifying which commands are
     legacy, the interfaces are split.
     """
 
     def between(pairs):
         """Obtain nodes between pairs of nodes.
 
         ``pairs`` is an iterable of node pairs.
 
         Returns an iterable of iterables of nodes corresponding to each
         requested pair.
         """
 
     def branches(nodes):
         """Obtain ancestor changesets of specific nodes back to a branch point.
 
         For each requested node, the peer finds the first ancestor node that is
         a DAG root or is a merge.
 
         Returns an iterable of iterables with the resolved values for each node.
         """
 
     def changegroup(nodes, source):
         """Obtain a changegroup with data for descendants of specified nodes."""
 
     def changegroupsubset(bases, heads, source):
         pass
 
 
 class ipeercommandexecutor(interfaceutil.Interface):
     """Represents a mechanism to execute remote commands.
 
     This is the primary interface for requesting that wire protocol commands
     be executed. Instances of this interface are active in a context manager
     and have a well-defined lifetime. When the context manager exits, all
     outstanding requests are waited on.
     """
 
     def callcommand(name, args):
         """Request that a named command be executed.
 
         Receives the command name and a dictionary of command arguments.
 
         Returns a ``concurrent.futures.Future`` that will resolve to the
         result of that command request. That exact value is left up to
         the implementation and possibly varies by command.
 
         Not all commands can coexist with other commands in an executor
         instance: it depends on the underlying wire protocol transport being
         used and the command itself.
 
         Implementations MAY call ``sendcommands()`` automatically if the
         requested command can not coexist with other commands in this executor.
 
         Implementations MAY call ``sendcommands()`` automatically when the
         future's ``result()`` is called. So, consumers using multiple
         commands with an executor MUST ensure that ``result()`` is not called
         until all command requests have been issued.
         """
 
     def sendcommands():
         """Trigger submission of queued command requests.
 
         Not all transports submit commands as soon as they are requested to
         run. When called, this method forces queued command requests to be
         issued. It will no-op if all commands have already been sent.
 
         When called, no more new commands may be issued with this executor.
         """
 
     def close():
         """Signal that this command request is finished.
 
         When called, no more new commands may be issued. All outstanding
         commands that have previously been issued are waited on before
         returning. This not only includes waiting for the futures to resolve,
         but also waiting for all response data to arrive. In other words,
         calling this waits for all on-wire state for issued command requests
         to finish.
 
         When used as a context manager, this method is called when exiting the
         context manager.
 
         This method may call ``sendcommands()`` if there are buffered commands.
         """
 
 
 class ipeerrequests(interfaceutil.Interface):
     """Interface for executing commands on a peer."""
 
     limitedarguments = interfaceutil.Attribute(
         """True if the peer cannot receive large argument value for commands."""
     )
 
     def commandexecutor():
         """A context manager that resolves to an ipeercommandexecutor.
 
         The object this resolves to can be used to issue command requests
         to the peer.
 
         Callers should call its ``callcommand`` method to issue command
         requests.
 
         A new executor should be obtained for each distinct set of commands
         (possibly just a single command) that the consumer wants to execute
         as part of a single operation or round trip. This is because some
         peers are half-duplex and/or don't support persistent connections.
         e.g. in the case of HTTP peers, commands sent to an executor represent
         a single HTTP request. While some peers may support multiple command
         sends over the wire per executor, consumers need to code to the least
         capable peer. So it should be assumed that command executors buffer
         called commands until they are told to send them and that each
         command executor could result in a new connection or wire-level request
         being issued.
         """
 
 
 class ipeerbase(ipeerconnection, ipeercapabilities, ipeerrequests):
     """Unified interface for peer repositories.
 
     All peer instances must conform to this interface.
     """
 
 
 class ipeerv2(ipeerconnection, ipeercapabilities, ipeerrequests):
     """Unified peer interface for wire protocol version 2 peers."""
 
     apidescriptor = interfaceutil.Attribute(
         """Data structure holding description of server API."""
     )
 
 
 @interfaceutil.implementer(ipeerbase)
 class peer(object):
     """Base class for peer repositories."""
 
     limitedarguments = False
 
     def capable(self, name):
         caps = self.capabilities()
         if name in caps:
             return True
 
         name = b'%s=' % name
         for cap in caps:
             if cap.startswith(name):
                 return cap[len(name) :]
 
         return False
 
     def requirecap(self, name, purpose):
         if self.capable(name):
             return
 
         raise error.CapabilityError(
             _(
                 b'cannot %s; remote repository does not support the '
                 b'\'%s\' capability'
             )
             % (purpose, name)
         )
 
 
 class iverifyproblem(interfaceutil.Interface):
     """Represents a problem with the integrity of the repository.
 
     Instances of this interface are emitted to describe an integrity issue
     with a repository (e.g. corrupt storage, missing data, etc).
 
     Instances are essentially messages associated with severity.
     """
 
     warning = interfaceutil.Attribute(
         """Message indicating a non-fatal problem."""
     )
 
     error = interfaceutil.Attribute("""Message indicating a fatal problem.""")
 
     node = interfaceutil.Attribute(
         """Revision encountering the problem.
 
         ``None`` means the problem doesn't apply to a single revision.
         """
     )
 
 
 class irevisiondelta(interfaceutil.Interface):
     """Represents a delta between one revision and another.
 
     Instances convey enough information to allow a revision to be exchanged
     with another repository.
 
     Instances represent the fulltext revision data or a delta against
     another revision. Therefore the ``revision`` and ``delta`` attributes
     are mutually exclusive.
 
     Typically used for changegroup generation.
     """
 
     node = interfaceutil.Attribute("""20 byte node of this revision.""")
 
     p1node = interfaceutil.Attribute(
         """20 byte node of 1st parent of this revision."""
     )
 
     p2node = interfaceutil.Attribute(
         """20 byte node of 2nd parent of this revision."""
     )
 
     linknode = interfaceutil.Attribute(
         """20 byte node of the changelog revision this node is linked to."""
     )
 
     flags = interfaceutil.Attribute(
         """2 bytes of integer flags that apply to this revision.
 
         This is a bitwise composition of the ``REVISION_FLAG_*`` constants.
         """
     )
 
     basenode = interfaceutil.Attribute(
         """20 byte node of the revision this data is a delta against.
 
         ``nullid`` indicates that the revision is a full revision and not
         a delta.
         """
     )
 
     baserevisionsize = interfaceutil.Attribute(
         """Size of base revision this delta is against.
 
         May be ``None`` if ``basenode`` is ``nullid``.
         """
     )
 
     revision = interfaceutil.Attribute(
         """Raw fulltext of revision data for this node."""
     )
 
     delta = interfaceutil.Attribute(
         """Delta between ``basenode`` and ``node``.
 
         Stored in the bdiff delta format.
         """
     )
 
     sidedata = interfaceutil.Attribute(
         """Raw sidedata bytes for the given revision."""
     )
 
     protocol_flags = interfaceutil.Attribute(
         """Single byte of integer flags that can influence the protocol.
 
         This is a bitwise composition of the ``storageutil.CG_FLAG*`` constants.
         """
     )
 
 
 class ifilerevisionssequence(interfaceutil.Interface):
     """Contains index data for all revisions of a file.
 
     Types implementing this behave like lists of tuples. The index
     in the list corresponds to the revision number. The values contain
     index metadata.
 
     The *null* revision (revision number -1) is always the last item
     in the index.
     """
 
     def __len__():
         """The total number of revisions."""
 
     def __getitem__(rev):
         """Returns the object having a specific revision number.
 
         Returns an 8-tuple with the following fields:
 
         offset+flags
            Contains the offset and flags for the revision. 64-bit unsigned
            integer where first 6 bytes are the offset and the next 2 bytes
            are flags. The offset can be 0 if it is not used by the store.
         compressed size
             Size of the revision data in the store. It can be 0 if it isn't
             needed by the store.
         uncompressed size
             Fulltext size. It can be 0 if it isn't needed by the store.
         base revision
             Revision number of revision the delta for storage is encoded
             against. -1 indicates not encoded against a base revision.
         link revision
             Revision number of changelog revision this entry is related to.
         p1 revision
             Revision number of 1st parent. -1 if no 1st parent.
         p2 revision
             Revision number of 2nd parent. -1 if no 1st parent.
         node
             Binary node value for this revision number.
 
         Negative values should index off the end of the sequence. ``-1``
         should return the null revision. ``-2`` should return the most
         recent revision.
         """
 
     def __contains__(rev):
         """Whether a revision number exists."""
 
     def insert(self, i, entry):
         """Add an item to the index at specific revision."""
 
 
 class ifileindex(interfaceutil.Interface):
     """Storage interface for index data of a single file.
 
     File storage data is divided into index metadata and data storage.
     This interface defines the index portion of the interface.
 
     The index logically consists of:
 
     * A mapping between revision numbers and nodes.
     * DAG data (storing and querying the relationship between nodes).
     * Metadata to facilitate storage.
     """
 
     nullid = interfaceutil.Attribute(
         """node for the null revision for use as delta base."""
     )
 
     def __len__():
         """Obtain the number of revisions stored for this file."""
 
     def __iter__():
         """Iterate over revision numbers for this file."""
 
     def hasnode(node):
         """Returns a bool indicating if a node is known to this store.
 
         Implementations must only return True for full, binary node values:
         hex nodes, revision numbers, and partial node matches must be
         rejected.
 
         The null node is never present.
         """
 
     def revs(start=0, stop=None):
         """Iterate over revision numbers for this file, with control."""
 
     def parents(node):
         """Returns a 2-tuple of parent nodes for a revision.
 
         Values will be ``nullid`` if the parent is empty.
         """
 
     def parentrevs(rev):
         """Like parents() but operates on revision numbers."""
 
     def rev(node):
         """Obtain the revision number given a node.
 
         Raises ``error.LookupError`` if the node is not known.
         """
 
     def node(rev):
         """Obtain the node value given a revision number.
 
         Raises ``IndexError`` if the node is not known.
         """
 
     def lookup(node):
         """Attempt to resolve a value to a node.
 
         Value can be a binary node, hex node, revision number, or a string
         that can be converted to an integer.
 
         Raises ``error.LookupError`` if a node could not be resolved.
         """
 
     def linkrev(rev):
         """Obtain the changeset revision number a revision is linked to."""
 
     def iscensored(rev):
         """Return whether a revision's content has been censored."""
 
     def commonancestorsheads(node1, node2):
         """Obtain an iterable of nodes containing heads of common ancestors.
 
         See ``ancestor.commonancestorsheads()``.
         """
 
     def descendants(revs):
         """Obtain descendant revision numbers for a set of revision numbers.
 
         If ``nullrev`` is in the set, this is equivalent to ``revs()``.
         """
 
     def heads(start=None, stop=None):
         """Obtain a list of nodes that are DAG heads, with control.
 
         The set of revisions examined can be limited by specifying
         ``start`` and ``stop``. ``start`` is a node. ``stop`` is an
         iterable of nodes. DAG traversal starts at earlier revision
         ``start`` and iterates forward until any node in ``stop`` is
         encountered.
         """
 
     def children(node):
         """Obtain nodes that are children of a node.
 
         Returns a list of nodes.
         """
 
 
 class ifiledata(interfaceutil.Interface):
     """Storage interface for data storage of a specific file.
 
     This complements ``ifileindex`` and provides an interface for accessing
     data for a tracked file.
     """
 
     def size(rev):
         """Obtain the fulltext size of file data.
 
         Any metadata is excluded from size measurements.
         """
 
     def revision(node, raw=False):
         """Obtain fulltext data for a node.
 
         By default, any storage transformations are applied before the data
         is returned. If ``raw`` is True, non-raw storage transformations
         are not applied.
 
         The fulltext data may contain a header containing metadata. Most
         consumers should use ``read()`` to obtain the actual file data.
         """
 
     def rawdata(node):
         """Obtain raw data for a node."""
 
     def read(node):
         """Resolve file fulltext data.
 
         This is similar to ``revision()`` except any metadata in the data
         headers is stripped.
         """
 
     def renamed(node):
         """Obtain copy metadata for a node.
 
         Returns ``False`` if no copy metadata is stored or a 2-tuple of
         (path, node) from which this revision was copied.
         """
 
     def cmp(node, fulltext):
         """Compare fulltext to another revision.
 
         Returns True if the fulltext is different from what is stored.
 
         This takes copy metadata into account.
 
         TODO better document the copy metadata and censoring logic.
         """
 
     def emitrevisions(
         nodes,
         nodesorder=None,
         revisiondata=False,
         assumehaveparentrevisions=False,
         deltamode=CG_DELTAMODE_STD,
     ):
         """Produce ``irevisiondelta`` for revisions.
 
         Given an iterable of nodes, emits objects conforming to the
         ``irevisiondelta`` interface that describe revisions in storage.
 
         This method is a generator.
 
         The input nodes may be unordered. Implementations must ensure that a
         node's parents are emitted before the node itself. Transitively, this
         means that a node may only be emitted once all its ancestors in
         ``nodes`` have also been emitted.
 
         By default, emits "index" data (the ``node``, ``p1node``, and
         ``p2node`` attributes). If ``revisiondata`` is set, revision data
         will also be present on the emitted objects.
 
         With default argument values, implementations can choose to emit
         either fulltext revision data or a delta. When emitting deltas,
         implementations must consider whether the delta's base revision
         fulltext is available to the receiver.
 
         The base revision fulltext is guaranteed to be available if any of
         the following are met:
 
         * Its fulltext revision was emitted by this method call.
         * A delta for that revision was emitted by this method call.
         * ``assumehaveparentrevisions`` is True and the base revision is a
           parent of the node.
 
         ``nodesorder`` can be used to control the order that revisions are
         emitted. By default, revisions can be reordered as long as they are
         in DAG topological order (see above). If the value is ``nodes``,
         the iteration order from ``nodes`` should be used. If the value is
         ``storage``, then the native order from the backing storage layer
         is used. (Not all storage layers will have strong ordering and behavior
         of this mode is storage-dependent.) ``nodes`` ordering can force
         revisions to be emitted before their ancestors, so consumers should
         use it with care.
 
         The ``linknode`` attribute on the returned ``irevisiondelta`` may not
         be set and it is the caller's responsibility to resolve it, if needed.
 
         If ``deltamode`` is CG_DELTAMODE_PREV and revision data is requested,
         all revision data should be emitted as deltas against the revision
         emitted just prior. The initial revision should be a delta against its
         1st parent.
         """
 
 
 class ifilemutation(interfaceutil.Interface):
     """Storage interface for mutation events of a tracked file."""
 
     def add(filedata, meta, transaction, linkrev, p1, p2):
         """Add a new revision to the store.
 
         Takes file data, dictionary of metadata, a transaction, linkrev,
         and parent nodes.
 
         Returns the node that was added.
 
         May no-op if a revision matching the supplied data is already stored.
         """
 
     def addrevision(
         revisiondata,
         transaction,
         linkrev,
         p1,
         p2,
         node=None,
         flags=0,
         cachedelta=None,
     ):
         """Add a new revision to the store and return its number.
 
         This is similar to ``add()`` except it operates at a lower level.
 
         The data passed in already contains a metadata header, if any.
 
         ``node`` and ``flags`` can be used to define the expected node and
         the flags to use with storage. ``flags`` is a bitwise value composed
         of the various ``REVISION_FLAG_*`` constants.
 
         ``add()`` is usually called when adding files from e.g. the working
         directory. ``addrevision()`` is often called by ``add()`` and for
         scenarios where revision data has already been computed, such as when
         applying raw data from a peer repo.
         """
 
     def addgroup(
         deltas,
         linkmapper,
         transaction,
         addrevisioncb=None,
         duplicaterevisioncb=None,
         maybemissingparents=False,
     ):
         """Process a series of deltas for storage.
 
         ``deltas`` is an iterable of 7-tuples of
         (node, p1, p2, linknode, deltabase, delta, flags) defining revisions
         to add.
 
         The ``delta`` field contains ``mpatch`` data to apply to a base
         revision, identified by ``deltabase``. The base node can be
         ``nullid``, in which case the header from the delta can be ignored
         and the delta used as the fulltext.
 
         ``alwayscache`` instructs the lower layers to cache the content of the
         newly added revision, even if it needs to be explicitly computed.
         This used to be the default when ``addrevisioncb`` was provided up to
         Mercurial 5.8.
 
         ``addrevisioncb`` should be called for each new rev as it is committed.
         ``duplicaterevisioncb`` should be called for all revs with a
         pre-existing node.
 
         ``maybemissingparents`` is a bool indicating whether the incoming
         data may reference parents/ancestor revisions that aren't present.
         This flag is set when receiving data into a "shallow" store that
         doesn't hold all history.
 
         Returns a list of nodes that were processed. A node will be in the list
         even if it existed in the store previously.
         """
 
     def censorrevision(tr, node, tombstone=b''):
         """Remove the content of a single revision.
 
         The specified ``node`` will have its content purged from storage.
         Future attempts to access the revision data for this node will
         result in failure.
 
         A ``tombstone`` message can optionally be stored. This message may be
         displayed to users when they attempt to access the missing revision
         data.
 
         Storage backends may have stored deltas against the previous content
         in this revision. As part of censoring a revision, these storage
         backends are expected to rewrite any internally stored deltas such
         that they no longer reference the deleted content.
         """
 
     def getstrippoint(minlink):
         """Find the minimum revision that must be stripped to strip a linkrev.
 
         Returns a 2-tuple containing the minimum revision number and a set
         of all revisions numbers that would be broken by this strip.
 
         TODO this is highly revlog centric and should be abstracted into
         a higher-level deletion API. ``repair.strip()`` relies on this.
         """
 
     def strip(minlink, transaction):
         """Remove storage of items starting at a linkrev.
 
         This uses ``getstrippoint()`` to determine the first node to remove.
         Then it effectively truncates storage for all revisions after that.
 
         TODO this is highly revlog centric and should be abstracted into a
         higher-level deletion API.
         """
 
 
 class ifilestorage(ifileindex, ifiledata, ifilemutation):
     """Complete storage interface for a single tracked file."""
 
     def files():
         """Obtain paths that are backing storage for this file.
 
         TODO this is used heavily by verify code and there should probably
         be a better API for that.
         """
 
     def storageinfo(
         exclusivefiles=False,
         sharedfiles=False,
         revisionscount=False,
         trackedsize=False,
         storedsize=False,
     ):
         """Obtain information about storage for this file's data.
 
         Returns a dict describing storage for this tracked path. The keys
         in the dict map to arguments of the same. The arguments are bools
         indicating whether to calculate and obtain that data.
 
         exclusivefiles
            Iterable of (vfs, path) describing files that are exclusively
            used to back storage for this tracked path.
 
         sharedfiles
            Iterable of (vfs, path) describing files that are used to back
            storage for this tracked path. Those files may also provide storage
            for other stored entities.
 
         revisionscount
            Number of revisions available for retrieval.
 
         trackedsize
            Total size in bytes of all tracked revisions. This is a sum of the
            length of the fulltext of all revisions.
 
         storedsize
            Total size in bytes used to store data for all tracked revisions.
            This is commonly less than ``trackedsize`` due to internal usage
            of deltas rather than fulltext revisions.
 
         Not all storage backends may support all queries are have a reasonable
         value to use. In that case, the value should be set to ``None`` and
         callers are expected to handle this special value.
         """
 
     def verifyintegrity(state):
         """Verifies the integrity of file storage.
 
         ``state`` is a dict holding state of the verifier process. It can be
         used to communicate data between invocations of multiple storage
         primitives.
 
         If individual revisions cannot have their revision content resolved,
         the method is expected to set the ``skipread`` key to a set of nodes
         that encountered problems.  If set, the method can also add the node(s)
         to ``safe_renamed`` in order to indicate nodes that may perform the
         rename checks with currently accessible data.
 
         The method yields objects conforming to the ``iverifyproblem``
         interface.
         """
 
 
 class idirs(interfaceutil.Interface):
     """Interface representing a collection of directories from paths.
 
     This interface is essentially a derived data structure representing
     directories from a collection of paths.
     """
 
     def addpath(path):
         """Add a path to the collection.
 
         All directories in the path will be added to the collection.
         """
 
     def delpath(path):
         """Remove a path from the collection.
 
         If the removal was the last path in a particular directory, the
         directory is removed from the collection.
         """
 
     def __iter__():
         """Iterate over the directories in this collection of paths."""
 
     def __contains__(path):
         """Whether a specific directory is in this collection."""
 
 
 class imanifestdict(interfaceutil.Interface):
     """Interface representing a manifest data structure.
 
     A manifest is effectively a dict mapping paths to entries. Each entry
     consists of a binary node and extra flags affecting that entry.
     """
 
     def __getitem__(path):
         """Returns the binary node value for a path in the manifest.
 
         Raises ``KeyError`` if the path does not exist in the manifest.
 
         Equivalent to ``self.find(path)[0]``.
         """
 
     def find(path):
         """Returns the entry for a path in the manifest.
 
         Returns a 2-tuple of (node, flags).
 
         Raises ``KeyError`` if the path does not exist in the manifest.
         """
 
     def __len__():
         """Return the number of entries in the manifest."""
 
     def __nonzero__():
         """Returns True if the manifest has entries, False otherwise."""
 
     __bool__ = __nonzero__
 
     def __setitem__(path, node):
         """Define the node value for a path in the manifest.
 
         If the path is already in the manifest, its flags will be copied to
         the new entry.
         """
 
     def __contains__(path):
         """Whether a path exists in the manifest."""
 
     def __delitem__(path):
         """Remove a path from the manifest.
 
         Raises ``KeyError`` if the path is not in the manifest.
         """
 
     def __iter__():
         """Iterate over paths in the manifest."""
 
     def iterkeys():
         """Iterate over paths in the manifest."""
 
     def keys():
         """Obtain a list of paths in the manifest."""
 
     def filesnotin(other, match=None):
         """Obtain the set of paths in this manifest but not in another.
 
         ``match`` is an optional matcher function to be applied to both
         manifests.
 
         Returns a set of paths.
         """
 
     def dirs():
         """Returns an object implementing the ``idirs`` interface."""
 
     def hasdir(dir):
         """Returns a bool indicating if a directory is in this manifest."""
 
     def walk(match):
         """Generator of paths in manifest satisfying a matcher.
 
         If the matcher has explicit files listed and they don't exist in
         the manifest, ``match.bad()`` is called for each missing file.
         """
 
     def diff(other, match=None, clean=False):
         """Find differences between this manifest and another.
 
         This manifest is compared to ``other``.
 
         If ``match`` is provided, the two manifests are filtered against this
         matcher and only entries satisfying the matcher are compared.
 
         If ``clean`` is True, unchanged files are included in the returned
         object.
 
         Returns a dict with paths as keys and values of 2-tuples of 2-tuples of
         the form ``((node1, flag1), (node2, flag2))`` where ``(node1, flag1)``
         represents the node and flags for this manifest and ``(node2, flag2)``
         are the same for the other manifest.
         """
 
     def setflag(path, flag):
         """Set the flag value for a given path.
 
         Raises ``KeyError`` if the path is not already in the manifest.
         """
 
     def get(path, default=None):
         """Obtain the node value for a path or a default value if missing."""
 
     def flags(path):
         """Return the flags value for a path (default: empty bytestring)."""
 
     def copy():
         """Return a copy of this manifest."""
 
     def items():
         """Returns an iterable of (path, node) for items in this manifest."""
 
     def iteritems():
         """Identical to items()."""
 
     def iterentries():
         """Returns an iterable of (path, node, flags) for this manifest.
 
         Similar to ``iteritems()`` except items are a 3-tuple and include
         flags.
         """
 
     def text():
         """Obtain the raw data representation for this manifest.
 
         Result is used to create a manifest revision.
         """
 
     def fastdelta(base, changes):
         """Obtain a delta between this manifest and another given changes.
 
         ``base`` in the raw data representation for another manifest.
 
         ``changes`` is an iterable of ``(path, to_delete)``.
 
         Returns a 2-tuple containing ``bytearray(self.text())`` and the
         delta between ``base`` and this manifest.
 
         If this manifest implementation can't support ``fastdelta()``,
         raise ``mercurial.manifest.FastdeltaUnavailable``.
         """
 
 
 class imanifestrevisionbase(interfaceutil.Interface):
     """Base interface representing a single revision of a manifest.
 
     Should not be used as a primary interface: should always be inherited
     as part of a larger interface.
     """
 
     def copy():
         """Obtain a copy of this manifest instance.
 
         Returns an object conforming to the ``imanifestrevisionwritable``
         interface. The instance will be associated with the same
         ``imanifestlog`` collection as this instance.
         """
 
     def read():
         """Obtain the parsed manifest data structure.
 
         The returned object conforms to the ``imanifestdict`` interface.
         """
 
 
 class imanifestrevisionstored(imanifestrevisionbase):
     """Interface representing a manifest revision committed to storage."""
 
     def node():
         """The binary node for this manifest."""
 
     parents = interfaceutil.Attribute(
         """List of binary nodes that are parents for this manifest revision."""
     )
 
     def readdelta(shallow=False):
         """Obtain the manifest data structure representing changes from parent.
 
         This manifest is compared to its 1st parent. A new manifest representing
         those differences is constructed.
 
         The returned object conforms to the ``imanifestdict`` interface.
         """
 
     def readfast(shallow=False):
         """Calls either ``read()`` or ``readdelta()``.
 
         The faster of the two options is called.
         """
 
     def find(key):
         """Calls self.read().find(key)``.
 
         Returns a 2-tuple of ``(node, flags)`` or raises ``KeyError``.
         """
 
 
 class imanifestrevisionwritable(imanifestrevisionbase):
     """Interface representing a manifest revision that can be committed."""
 
     def write(transaction, linkrev, p1node, p2node, added, removed, match=None):
         """Add this revision to storage.
 
         Takes a transaction object, the changeset revision number it will
         be associated with, its parent nodes, and lists of added and
         removed paths.
 
         If match is provided, storage can choose not to inspect or write out
         items that do not match. Storage is still required to be able to provide
         the full manifest in the future for any directories written (these
         manifests should not be "narrowed on disk").
 
         Returns the binary node of the created revision.
         """
 
 
 class imanifeststorage(interfaceutil.Interface):
     """Storage interface for manifest data."""
 
     nodeconstants = interfaceutil.Attribute(
         """nodeconstants used by the current repository."""
     )
 
     tree = interfaceutil.Attribute(
         """The path to the directory this manifest tracks.
 
         The empty bytestring represents the root manifest.
         """
     )
 
     index = interfaceutil.Attribute(
         """An ``ifilerevisionssequence`` instance."""
     )
 
     indexfile = interfaceutil.Attribute(
         """Path of revlog index file.
 
         TODO this is revlog specific and should not be exposed.
         """
     )
 
     opener = interfaceutil.Attribute(
         """VFS opener to use to access underlying files used for storage.
 
         TODO this is revlog specific and should not be exposed.
         """
     )
 
     version = interfaceutil.Attribute(
         """Revlog version number.
 
         TODO this is revlog specific and should not be exposed.
         """
     )
 
     _generaldelta = interfaceutil.Attribute(
         """Whether generaldelta storage is being used.
 
         TODO this is revlog specific and should not be exposed.
         """
     )
 
     fulltextcache = interfaceutil.Attribute(
         """Dict with cache of fulltexts.
 
         TODO this doesn't feel appropriate for the storage interface.
         """
     )
 
     def __len__():
         """Obtain the number of revisions stored for this manifest."""
 
     def __iter__():
         """Iterate over revision numbers for this manifest."""
 
     def rev(node):
         """Obtain the revision number given a binary node.
 
         Raises ``error.LookupError`` if the node is not known.
         """
 
     def node(rev):
         """Obtain the node value given a revision number.
 
         Raises ``error.LookupError`` if the revision is not known.
         """
 
     def lookup(value):
         """Attempt to resolve a value to a node.
 
         Value can be a binary node, hex node, revision number, or a bytes
         that can be converted to an integer.
 
         Raises ``error.LookupError`` if a ndoe could not be resolved.
         """
 
     def parents(node):
         """Returns a 2-tuple of parent nodes for a node.
 
         Values will be ``nullid`` if the parent is empty.
         """
 
     def parentrevs(rev):
         """Like parents() but operates on revision numbers."""
 
     def linkrev(rev):
         """Obtain the changeset revision number a revision is linked to."""
 
     def revision(node, _df=None, raw=False):
         """Obtain fulltext data for a node."""
 
     def rawdata(node, _df=None):
         """Obtain raw data for a node."""
 
     def revdiff(rev1, rev2):
         """Obtain a delta between two revision numbers.
 
         The returned data is the result of ``bdiff.bdiff()`` on the raw
         revision data.
         """
 
     def cmp(node, fulltext):
         """Compare fulltext to another revision.
 
         Returns True if the fulltext is different from what is stored.
         """
 
     def emitrevisions(
         nodes,
         nodesorder=None,
         revisiondata=False,
         assumehaveparentrevisions=False,
     ):
         """Produce ``irevisiondelta`` describing revisions.
 
         See the documentation for ``ifiledata`` for more.
         """
 
     def addgroup(
         deltas,
         linkmapper,
         transaction,
         addrevisioncb=None,
         duplicaterevisioncb=None,
     ):
         """Process a series of deltas for storage.
 
         See the documentation in ``ifilemutation`` for more.
         """
 
     def rawsize(rev):
         """Obtain the size of tracked data.
 
         Is equivalent to ``len(m.rawdata(node))``.
 
         TODO this method is only used by upgrade code and may be removed.
         """
 
     def getstrippoint(minlink):
         """Find minimum revision that must be stripped to strip a linkrev.
 
         See the documentation in ``ifilemutation`` for more.
         """
 
     def strip(minlink, transaction):
         """Remove storage of items starting at a linkrev.
 
         See the documentation in ``ifilemutation`` for more.
         """
 
     def checksize():
         """Obtain the expected sizes of backing files.
 
         TODO this is used by verify and it should not be part of the interface.
         """
 
     def files():
         """Obtain paths that are backing storage for this manifest.
 
         TODO this is used by verify and there should probably be a better API
         for this functionality.
         """
 
     def deltaparent(rev):
         """Obtain the revision that a revision is delta'd against.
 
         TODO delta encoding is an implementation detail of storage and should
         not be exposed to the storage interface.
         """
 
     def clone(tr, dest, **kwargs):
         """Clone this instance to another."""
 
     def clearcaches(clear_persisted_data=False):
         """Clear any caches associated with this instance."""
 
     def dirlog(d):
         """Obtain a manifest storage instance for a tree."""
 
     def add(
         m, transaction, link, p1, p2, added, removed, readtree=None, match=None
     ):
         """Add a revision to storage.
 
         ``m`` is an object conforming to ``imanifestdict``.
 
         ``link`` is the linkrev revision number.
 
         ``p1`` and ``p2`` are the parent revision numbers.
 
         ``added`` and ``removed`` are iterables of added and removed paths,
         respectively.
 
         ``readtree`` is a function that can be used to read the child tree(s)
         when recursively writing the full tree structure when using
         treemanifets.
 
         ``match`` is a matcher that can be used to hint to storage that not all
         paths must be inspected; this is an optimization and can be safely
         ignored. Note that the storage must still be able to reproduce a full
         manifest including files that did not match.
         """
 
     def storageinfo(
         exclusivefiles=False,
         sharedfiles=False,
         revisionscount=False,
         trackedsize=False,
         storedsize=False,
     ):
         """Obtain information about storage for this manifest's data.
 
         See ``ifilestorage.storageinfo()`` for a description of this method.
         This one behaves the same way, except for manifest data.
         """
 
 
 class imanifestlog(interfaceutil.Interface):
     """Interface representing a collection of manifest snapshots.
 
     Represents the root manifest in a repository.
 
     Also serves as a means to access nested tree manifests and to cache
     tree manifests.
     """
 
     nodeconstants = interfaceutil.Attribute(
         """nodeconstants used by the current repository."""
     )
 
     def __getitem__(node):
         """Obtain a manifest instance for a given binary node.
 
         Equivalent to calling ``self.get('', node)``.
 
         The returned object conforms to the ``imanifestrevisionstored``
         interface.
         """
 
     def get(tree, node, verify=True):
         """Retrieve the manifest instance for a given directory and binary node.
 
         ``node`` always refers to the node of the root manifest (which will be
         the only manifest if flat manifests are being used).
 
         If ``tree`` is the empty string, the root manifest is returned.
         Otherwise the manifest for the specified directory will be returned
         (requires tree manifests).
 
         If ``verify`` is True, ``LookupError`` is raised if the node is not
         known.
 
         The returned object conforms to the ``imanifestrevisionstored``
         interface.
         """
 
     def getstorage(tree):
         """Retrieve an interface to storage for a particular tree.
 
         If ``tree`` is the empty bytestring, storage for the root manifest will
         be returned. Otherwise storage for a tree manifest is returned.
 
         TODO formalize interface for returned object.
         """
 
     def clearcaches():
         """Clear caches associated with this collection."""
 
     def rev(node):
         """Obtain the revision number for a binary node.
 
         Raises ``error.LookupError`` if the node is not known.
         """
 
     def update_caches(transaction):
         """update whatever cache are relevant for the used storage."""
 
 
 class ilocalrepositoryfilestorage(interfaceutil.Interface):
     """Local repository sub-interface providing access to tracked file storage.
 
     This interface defines how a repository accesses storage for a single
     tracked file path.
     """
 
     def file(f):
         """Obtain a filelog for a tracked path.
 
         The returned type conforms to the ``ifilestorage`` interface.
         """
 
 
 class ilocalrepositorymain(interfaceutil.Interface):
     """Main interface for local repositories.
 
     This currently captures the reality of things - not how things should be.
     """
 
     nodeconstants = interfaceutil.Attribute(
         """Constant nodes matching the hash function used by the repository."""
     )
     nullid = interfaceutil.Attribute(
         """null revision for the hash function used by the repository."""
     )
 
     supportedformats = interfaceutil.Attribute(
         """Set of requirements that apply to stream clone.
 
         This is actually a class attribute and is shared among all instances.
         """
     )
 
     supported = interfaceutil.Attribute(
         """Set of requirements that this repo is capable of opening."""
     )
 
     requirements = interfaceutil.Attribute(
         """Set of requirements this repo uses."""
     )
 
     features = interfaceutil.Attribute(
         """Set of "features" this repository supports.
 
         A "feature" is a loosely-defined term. It can refer to a feature
         in the classical sense or can describe an implementation detail
         of the repository. For example, a ``readonly`` feature may denote
         the repository as read-only. Or a ``revlogfilestore`` feature may
         denote that the repository is using revlogs for file storage.
 
         The intent of features is to provide a machine-queryable mechanism
         for repo consumers to test for various repository characteristics.
 
         Features are similar to ``requirements``. The main difference is that
         requirements are stored on-disk and represent requirements to open the
         repository. Features are more run-time capabilities of the repository
         and more granular capabilities (which may be derived from requirements).
         """
     )
 
     filtername = interfaceutil.Attribute(
         """Name of the repoview that is active on this repo."""
     )
 
     wvfs = interfaceutil.Attribute(
         """VFS used to access the working directory."""
     )
 
     vfs = interfaceutil.Attribute(
         """VFS rooted at the .hg directory.
 
         Used to access repository data not in the store.
         """
     )
 
     svfs = interfaceutil.Attribute(
         """VFS rooted at the store.
 
         Used to access repository data in the store. Typically .hg/store.
         But can point elsewhere if the store is shared.
         """
     )
 
     root = interfaceutil.Attribute(
         """Path to the root of the working directory."""
     )
 
     path = interfaceutil.Attribute("""Path to the .hg directory.""")
 
     origroot = interfaceutil.Attribute(
         """The filesystem path that was used to construct the repo."""
     )
 
     auditor = interfaceutil.Attribute(
         """A pathauditor for the working directory.
 
         This checks if a path refers to a nested repository.
 
         Operates on the filesystem.
         """
     )
 
     nofsauditor = interfaceutil.Attribute(
         """A pathauditor for the working directory.
 
         This is like ``auditor`` except it doesn't do filesystem checks.
         """
     )
 
     baseui = interfaceutil.Attribute(
         """Original ui instance passed into constructor."""
     )
 
     ui = interfaceutil.Attribute("""Main ui instance for this instance.""")
 
     sharedpath = interfaceutil.Attribute(
         """Path to the .hg directory of the repo this repo was shared from."""
     )
 
     store = interfaceutil.Attribute("""A store instance.""")
 
     spath = interfaceutil.Attribute("""Path to the store.""")
 
     sjoin = interfaceutil.Attribute("""Alias to self.store.join.""")
 
     cachevfs = interfaceutil.Attribute(
         """A VFS used to access the cache directory.
 
         Typically .hg/cache.
         """
     )
 
     wcachevfs = interfaceutil.Attribute(
         """A VFS used to access the cache directory dedicated to working copy
 
         Typically .hg/wcache.
         """
     )
 
     filteredrevcache = interfaceutil.Attribute(
         """Holds sets of revisions to be filtered."""
     )
 
     names = interfaceutil.Attribute("""A ``namespaces`` instance.""")
 
     filecopiesmode = interfaceutil.Attribute(
         """The way files copies should be dealt with in this repo."""
     )
 
     def close():
         """Close the handle on this repository."""
 
     def peer():
         """Obtain an object conforming to the ``peer`` interface."""
 
     def unfiltered():
         """Obtain an unfiltered/raw view of this repo."""
 
     def filtered(name, visibilityexceptions=None):
         """Obtain a named view of this repository."""
 
     obsstore = interfaceutil.Attribute("""A store of obsolescence data.""")
 
     changelog = interfaceutil.Attribute("""A handle on the changelog revlog.""")
 
     manifestlog = interfaceutil.Attribute(
         """An instance conforming to the ``imanifestlog`` interface.
 
         Provides access to manifests for the repository.
         """
     )
 
     dirstate = interfaceutil.Attribute("""Working directory state.""")
 
     narrowpats = interfaceutil.Attribute(
         """Matcher patterns for this repository's narrowspec."""
     )
 
     def narrowmatch(match=None, includeexact=False):
         """Obtain a matcher for the narrowspec."""
 
     def setnarrowpats(newincludes, newexcludes):
         """Define the narrowspec for this repository."""
 
     def __getitem__(changeid):
         """Try to resolve a changectx."""
 
     def __contains__(changeid):
         """Whether a changeset exists."""
 
     def __nonzero__():
         """Always returns True."""
         return True
 
     __bool__ = __nonzero__
 
     def __len__():
         """Returns the number of changesets in the repo."""
 
     def __iter__():
         """Iterate over revisions in the changelog."""
 
     def revs(expr, *args):
         """Evaluate a revset.
 
         Emits revisions.
         """
 
     def set(expr, *args):
         """Evaluate a revset.
 
         Emits changectx instances.
         """
 
     def anyrevs(specs, user=False, localalias=None):
         """Find revisions matching one of the given revsets."""
 
     def url():
         """Returns a string representing the location of this repo."""
 
     def hook(name, throw=False, **args):
         """Call a hook."""
 
     def tags():
         """Return a mapping of tag to node."""
 
     def tagtype(tagname):
         """Return the type of a given tag."""
 
     def tagslist():
         """Return a list of tags ordered by revision."""
 
     def nodetags(node):
         """Return the tags associated with a node."""
 
     def nodebookmarks(node):
         """Return the list of bookmarks pointing to the specified node."""
 
     def branchmap():
         """Return a mapping of branch to heads in that branch."""
 
     def revbranchcache():
         pass
 
     def register_changeset(rev, changelogrevision):
         """Extension point for caches for new nodes.
 
         Multiple consumers are expected to need parts of the changelogrevision,
         so it is provided as optimization to avoid duplicate lookups. A simple
         cache would be fragile when other revisions are accessed, too."""
         pass
 
     def branchtip(branchtip, ignoremissing=False):
         """Return the tip node for a given branch."""
 
     def lookup(key):
         """Resolve the node for a revision."""
 
     def lookupbranch(key):
         """Look up the branch name of the given revision or branch name."""
 
     def known(nodes):
         """Determine whether a series of nodes is known.
 
         Returns a list of bools.
         """
 
     def local():
         """Whether the repository is local."""
         return True
 
     def publishing():
         """Whether the repository is a publishing repository."""
 
     def cancopy():
         pass
 
     def shared():
         """The type of shared repository or None."""
 
     def wjoin(f, *insidef):
         """Calls self.vfs.reljoin(self.root, f, *insidef)"""
 
     def setparents(p1, p2):
         """Set the parent nodes of the working directory."""
 
     def filectx(path, changeid=None, fileid=None):
         """Obtain a filectx for the given file revision."""
 
     def getcwd():
         """Obtain the current working directory from the dirstate."""
 
     def pathto(f, cwd=None):
         """Obtain the relative path to a file."""
 
     def adddatafilter(name, fltr):
         pass
 
     def wread(filename):
         """Read a file from wvfs, using data filters."""
 
     def wwrite(filename, data, flags, backgroundclose=False, **kwargs):
         """Write data to a file in the wvfs, using data filters."""
 
     def wwritedata(filename, data):
         """Resolve data for writing to the wvfs, using data filters."""
 
     def currenttransaction():
         """Obtain the current transaction instance or None."""
 
     def transaction(desc, report=None):
         """Open a new transaction to write to the repository."""
 
     def undofiles():
         """Returns a list of (vfs, path) for files to undo transactions."""
 
     def recover():
         """Roll back an interrupted transaction."""
 
     def rollback(dryrun=False, force=False):
         """Undo the last transaction.
 
         DANGEROUS.
         """
 
     def updatecaches(tr=None, full=False):
         """Warm repo caches."""
 
     def invalidatecaches():
         """Invalidate cached data due to the repository mutating."""
 
     def invalidatevolatilesets():
         pass
 
     def invalidatedirstate():
         """Invalidate the dirstate."""
 
     def invalidate(clearfilecache=False):
         pass
 
     def invalidateall():
         pass
 
     def lock(wait=True):
         """Lock the repository store and return a lock instance."""
 
     def wlock(wait=True):
         """Lock the non-store parts of the repository."""
 
     def currentwlock():
         """Return the wlock if it's held or None."""
 
     def checkcommitpatterns(wctx, match, status, fail):
         pass
 
     def commit(
         text=b'',
         user=None,
         date=None,
         match=None,
         force=False,
         editor=False,
         extra=None,
     ):
         """Add a new revision to the repository."""
 
     def commitctx(ctx, error=False, origctx=None):
         """Commit a commitctx instance to the repository."""
 
     def destroying():
         """Inform the repository that nodes are about to be destroyed."""
 
     def destroyed():
         """Inform the repository that nodes have been destroyed."""
 
     def status(
         node1=b'.',
         node2=None,
         match=None,
         ignored=False,
         clean=False,
         unknown=False,
         listsubrepos=False,
     ):
         """Convenience method to call repo[x].status()."""
 
     def addpostdsstatus(ps):
         pass
 
     def postdsstatus():
         pass
 
     def clearpostdsstatus():
         pass
 
     def heads(start=None):
         """Obtain list of nodes that are DAG heads."""
 
     def branchheads(branch=None, start=None, closed=False):
         pass
 
     def branches(nodes):
         pass
 
     def between(pairs):
         pass
 
     def checkpush(pushop):
         pass
 
     prepushoutgoinghooks = interfaceutil.Attribute("""util.hooks instance.""")
 
     def pushkey(namespace, key, old, new):
         pass
 
     def listkeys(namespace):
         pass
 
     def debugwireargs(one, two, three=None, four=None, five=None):
         pass
 
     def savecommitmessage(text):
         pass
 
-    def register_sidedata_computer(kind, category, keys, computer):
+    def register_sidedata_computer(kind, category, keys, computer, flags):
         pass
 
     def register_wanted_sidedata(category):
         pass
 
 
 class completelocalrepository(
     ilocalrepositorymain, ilocalrepositoryfilestorage
 ):
     """Complete interface for a local repository."""
 
 
 class iwireprotocolcommandcacher(interfaceutil.Interface):
     """Represents a caching backend for wire protocol commands.
 
     Wire protocol version 2 supports transparent caching of many commands.
     To leverage this caching, servers can activate objects that cache
     command responses. Objects handle both cache writing and reading.
     This interface defines how that response caching mechanism works.
 
     Wire protocol version 2 commands emit a series of objects that are
     serialized and sent to the client. The caching layer exists between
     the invocation of the command function and the sending of its output
     objects to an output layer.
 
     Instances of this interface represent a binding to a cache that
     can serve a response (in place of calling a command function) and/or
     write responses to a cache for subsequent use.
 
     When a command request arrives, the following happens with regards
     to this interface:
 
     1. The server determines whether the command request is cacheable.
     2. If it is, an instance of this interface is spawned.
     3. The cacher is activated in a context manager (``__enter__`` is called).
     4. A cache *key* for that request is derived. This will call the
        instance's ``adjustcachekeystate()`` method so the derivation
        can be influenced.
     5. The cacher is informed of the derived cache key via a call to
        ``setcachekey()``.
     6. The cacher's ``lookup()`` method is called to test for presence of
        the derived key in the cache.
     7. If ``lookup()`` returns a hit, that cached result is used in place
        of invoking the command function. ``__exit__`` is called and the instance
        is discarded.
     8. The command function is invoked.
     9. ``onobject()`` is called for each object emitted by the command
        function.
     10. After the final object is seen, ``onfinished()`` is called.
     11. ``__exit__`` is called to signal the end of use of the instance.
 
     Cache *key* derivation can be influenced by the instance.
 
     Cache keys are initially derived by a deterministic representation of
     the command request. This includes the command name, arguments, protocol
     version, etc. This initial key derivation is performed by CBOR-encoding a
     data structure and feeding that output into a hasher.
 
     Instances of this interface can influence this initial key derivation
     via ``adjustcachekeystate()``.
 
     The instance is informed of the derived cache key via a call to
     ``setcachekey()``. The instance must store the key locally so it can
     be consulted on subsequent operations that may require it.
 
     When constructed, the instance has access to a callable that can be used
     for encoding response objects. This callable receives as its single
     argument an object emitted by a command function. It returns an iterable
     of bytes chunks representing the encoded object. Unless the cacher is
     caching native Python objects in memory or has a way of reconstructing
     the original Python objects, implementations typically call this function
     to produce bytes from the output objects and then store those bytes in
     the cache. When it comes time to re-emit those bytes, they are wrapped
     in a ``wireprototypes.encodedresponse`` instance to tell the output
     layer that they are pre-encoded.
 
     When receiving the objects emitted by the command function, instances
     can choose what to do with those objects. The simplest thing to do is
     re-emit the original objects. They will be forwarded to the output
     layer and will be processed as if the cacher did not exist.
 
     Implementations could also choose to not emit objects - instead locally
     buffering objects or their encoded representation. They could then emit
     a single "coalesced" object when ``onfinished()`` is called. In
     this way, the implementation would function as a filtering layer of
     sorts.
 
     When caching objects, typically the encoded form of the object will
     be stored. Keep in mind that if the original object is forwarded to
     the output layer, it will need to be encoded there as well. For large
     output, this redundant encoding could add overhead. Implementations
     could wrap the encoded object data in ``wireprototypes.encodedresponse``
     instances to avoid this overhead.
     """
 
     def __enter__():
         """Marks the instance as active.
 
         Should return self.
         """
 
     def __exit__(exctype, excvalue, exctb):
         """Called when cacher is no longer used.
 
         This can be used by implementations to perform cleanup actions (e.g.
         disconnecting network sockets, aborting a partially cached response.
         """
 
     def adjustcachekeystate(state):
         """Influences cache key derivation by adjusting state to derive key.
 
         A dict defining the state used to derive the cache key is passed.
 
         Implementations can modify this dict to record additional state that
         is wanted to influence key derivation.
 
         Implementations are *highly* encouraged to not modify or delete
         existing keys.
         """
 
     def setcachekey(key):
         """Record the derived cache key for this request.
 
         Instances may mutate the key for internal usage, as desired. e.g.
         instances may wish to prepend the repo name, introduce path
         components for filesystem or URL addressing, etc. Behavior is up to
         the cache.
 
         Returns a bool indicating if the request is cacheable by this
         instance.
         """
 
     def lookup():
         """Attempt to resolve an entry in the cache.
 
         The instance is instructed to look for the cache key that it was
         informed about via the call to ``setcachekey()``.
 
         If there's no cache hit or the cacher doesn't wish to use the cached
         entry, ``None`` should be returned.
 
         Else, a dict defining the cached result should be returned. The
         dict may have the following keys:
 
         objs
            An iterable of objects that should be sent to the client. That
            iterable of objects is expected to be what the command function
            would return if invoked or an equivalent representation thereof.
         """
 
     def onobject(obj):
         """Called when a new object is emitted from the command function.
 
         Receives as its argument the object that was emitted from the
         command function.
 
         This method returns an iterator of objects to forward to the output
         layer. The easiest implementation is a generator that just
         ``yield obj``.
         """
 
     def onfinished():
         """Called after all objects have been emitted from the command function.
 
         Implementations should return an iterator of objects to forward to
         the output layer.
 
         This method can be a generator.
         """

mercurial/localrepo.py

@@ -1,3763 +1,3763 b''
 # localrepo.py - read/write repository class for mercurial
 #
 # Copyright 2005-2007 Olivia Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 import errno
 import functools
 import os
 import random
 import sys
 import time
 import weakref
 
 from .i18n import _
 from .node import (
     bin,
     hex,
     nullrev,
     sha1nodeconstants,
     short,
 )
 from .pycompat import (
     delattr,
     getattr,
 )
 from . import (
     bookmarks,
     branchmap,
     bundle2,
     bundlecaches,
     changegroup,
     color,
     commit,
     context,
     dirstate,
     dirstateguard,
     discovery,
     encoding,
     error,
     exchange,
     extensions,
     filelog,
     hook,
     lock as lockmod,
     match as matchmod,
     mergestate as mergestatemod,
     mergeutil,
     metadata as metadatamod,
     namespaces,
     narrowspec,
     obsolete,
     pathutil,
     phases,
     pushkey,
     pycompat,
     rcutil,
     repoview,
     requirements as requirementsmod,
     revlog,
     revset,
     revsetlang,
     scmutil,
     sparse,
     store as storemod,
     subrepoutil,
     tags as tagsmod,
     transaction,
     txnutil,
     util,
     vfs as vfsmod,
     wireprototypes,
 )
 
 from .interfaces import (
     repository,
     util as interfaceutil,
 )
 
 from .utils import (
     hashutil,
     procutil,
     stringutil,
     urlutil,
 )
 
 from .revlogutils import (
     concurrency_checker as revlogchecker,
     constants as revlogconst,
 )
 
 release = lockmod.release
 urlerr = util.urlerr
 urlreq = util.urlreq
 
 # set of (path, vfs-location) tuples. vfs-location is:
 # - 'plain for vfs relative paths
 # - '' for svfs relative paths
 _cachedfiles = set()
 
 
 class _basefilecache(scmutil.filecache):
     """All filecache usage on repo are done for logic that should be unfiltered"""
 
     def __get__(self, repo, type=None):
         if repo is None:
             return self
         # proxy to unfiltered __dict__ since filtered repo has no entry
         unfi = repo.unfiltered()
         try:
             return unfi.__dict__[self.sname]
         except KeyError:
             pass
         return super(_basefilecache, self).__get__(unfi, type)
 
     def set(self, repo, value):
         return super(_basefilecache, self).set(repo.unfiltered(), value)
 
 
 class repofilecache(_basefilecache):
     """filecache for files in .hg but outside of .hg/store"""
 
     def __init__(self, *paths):
         super(repofilecache, self).__init__(*paths)
         for path in paths:
             _cachedfiles.add((path, b'plain'))
 
     def join(self, obj, fname):
         return obj.vfs.join(fname)
 
 
 class storecache(_basefilecache):
     """filecache for files in the store"""
 
     def __init__(self, *paths):
         super(storecache, self).__init__(*paths)
         for path in paths:
             _cachedfiles.add((path, b''))
 
     def join(self, obj, fname):
         return obj.sjoin(fname)
 
 
 class mixedrepostorecache(_basefilecache):
     """filecache for a mix files in .hg/store and outside"""
 
     def __init__(self, *pathsandlocations):
         # scmutil.filecache only uses the path for passing back into our
         # join(), so we can safely pass a list of paths and locations
         super(mixedrepostorecache, self).__init__(*pathsandlocations)
         _cachedfiles.update(pathsandlocations)
 
     def join(self, obj, fnameandlocation):
         fname, location = fnameandlocation
         if location == b'plain':
             return obj.vfs.join(fname)
         else:
             if location != b'':
                 raise error.ProgrammingError(
                     b'unexpected location: %s' % location
                 )
             return obj.sjoin(fname)
 
 
 def isfilecached(repo, name):
     """check if a repo has already cached "name" filecache-ed property
 
     This returns (cachedobj-or-None, iscached) tuple.
     """
     cacheentry = repo.unfiltered()._filecache.get(name, None)
     if not cacheentry:
         return None, False
     return cacheentry.obj, True
 
 
 class unfilteredpropertycache(util.propertycache):
     """propertycache that apply to unfiltered repo only"""
 
     def __get__(self, repo, type=None):
         unfi = repo.unfiltered()
         if unfi is repo:
             return super(unfilteredpropertycache, self).__get__(unfi)
         return getattr(unfi, self.name)
 
 
 class filteredpropertycache(util.propertycache):
     """propertycache that must take filtering in account"""
 
     def cachevalue(self, obj, value):
         object.__setattr__(obj, self.name, value)
 
 
 def hasunfilteredcache(repo, name):
     """check if a repo has an unfilteredpropertycache value for <name>"""
     return name in vars(repo.unfiltered())
 
 
 def unfilteredmethod(orig):
     """decorate method that always need to be run on unfiltered version"""
 
     @functools.wraps(orig)
     def wrapper(repo, *args, **kwargs):
         return orig(repo.unfiltered(), *args, **kwargs)
 
     return wrapper
 
 
 moderncaps = {
     b'lookup',
     b'branchmap',
     b'pushkey',
     b'known',
     b'getbundle',
     b'unbundle',
 }
 legacycaps = moderncaps.union({b'changegroupsubset'})
 
 
 @interfaceutil.implementer(repository.ipeercommandexecutor)
 class localcommandexecutor(object):
     def __init__(self, peer):
         self._peer = peer
         self._sent = False
         self._closed = False
 
     def __enter__(self):
         return self
 
     def __exit__(self, exctype, excvalue, exctb):
         self.close()
 
     def callcommand(self, command, args):
         if self._sent:
             raise error.ProgrammingError(
                 b'callcommand() cannot be used after sendcommands()'
             )
 
         if self._closed:
             raise error.ProgrammingError(
                 b'callcommand() cannot be used after close()'
             )
 
         # We don't need to support anything fancy. Just call the named
         # method on the peer and return a resolved future.
         fn = getattr(self._peer, pycompat.sysstr(command))
 
         f = pycompat.futures.Future()
 
         try:
             result = fn(**pycompat.strkwargs(args))
         except Exception:
             pycompat.future_set_exception_info(f, sys.exc_info()[1:])
         else:
             f.set_result(result)
 
         return f
 
     def sendcommands(self):
         self._sent = True
 
     def close(self):
         self._closed = True
 
 
 @interfaceutil.implementer(repository.ipeercommands)
 class localpeer(repository.peer):
     '''peer for a local repo; reflects only the most recent API'''
 
     def __init__(self, repo, caps=None):
         super(localpeer, self).__init__()
 
         if caps is None:
             caps = moderncaps.copy()
         self._repo = repo.filtered(b'served')
         self.ui = repo.ui
 
         if repo._wanted_sidedata:
             formatted = bundle2.format_remote_wanted_sidedata(repo)
             caps.add(b'exp-wanted-sidedata=' + formatted)
 
         self._caps = repo._restrictcapabilities(caps)
 
     # Begin of _basepeer interface.
 
     def url(self):
         return self._repo.url()
 
     def local(self):
         return self._repo
 
     def peer(self):
         return self
 
     def canpush(self):
         return True
 
     def close(self):
         self._repo.close()
 
     # End of _basepeer interface.
 
     # Begin of _basewirecommands interface.
 
     def branchmap(self):
         return self._repo.branchmap()
 
     def capabilities(self):
         return self._caps
 
     def clonebundles(self):
         return self._repo.tryread(bundlecaches.CB_MANIFEST_FILE)
 
     def debugwireargs(self, one, two, three=None, four=None, five=None):
         """Used to test argument passing over the wire"""
         return b"%s %s %s %s %s" % (
             one,
             two,
             pycompat.bytestr(three),
             pycompat.bytestr(four),
             pycompat.bytestr(five),
         )
 
     def getbundle(
         self,
         source,
         heads=None,
         common=None,
         bundlecaps=None,
         remote_sidedata=None,
         **kwargs
     ):
         chunks = exchange.getbundlechunks(
             self._repo,
             source,
             heads=heads,
             common=common,
             bundlecaps=bundlecaps,
             remote_sidedata=remote_sidedata,
             **kwargs
         )[1]
         cb = util.chunkbuffer(chunks)
 
         if exchange.bundle2requested(bundlecaps):
             # When requesting a bundle2, getbundle returns a stream to make the
             # wire level function happier. We need to build a proper object
             # from it in local peer.
             return bundle2.getunbundler(self.ui, cb)
         else:
             return changegroup.getunbundler(b'01', cb, None)
 
     def heads(self):
         return self._repo.heads()
 
     def known(self, nodes):
         return self._repo.known(nodes)
 
     def listkeys(self, namespace):
         return self._repo.listkeys(namespace)
 
     def lookup(self, key):
         return self._repo.lookup(key)
 
     def pushkey(self, namespace, key, old, new):
         return self._repo.pushkey(namespace, key, old, new)
 
     def stream_out(self):
         raise error.Abort(_(b'cannot perform stream clone against local peer'))
 
     def unbundle(self, bundle, heads, url):
         """apply a bundle on a repo
 
         This function handles the repo locking itself."""
         try:
             try:
                 bundle = exchange.readbundle(self.ui, bundle, None)
                 ret = exchange.unbundle(self._repo, bundle, heads, b'push', url)
                 if util.safehasattr(ret, b'getchunks'):
                     # This is a bundle20 object, turn it into an unbundler.
                     # This little dance should be dropped eventually when the
                     # API is finally improved.
                     stream = util.chunkbuffer(ret.getchunks())
                     ret = bundle2.getunbundler(self.ui, stream)
                 return ret
             except Exception as exc:
                 # If the exception contains output salvaged from a bundle2
                 # reply, we need to make sure it is printed before continuing
                 # to fail. So we build a bundle2 with such output and consume
                 # it directly.
                 #
                 # This is not very elegant but allows a "simple" solution for
                 # issue4594
                 output = getattr(exc, '_bundle2salvagedoutput', ())
                 if output:
                     bundler = bundle2.bundle20(self._repo.ui)
                     for out in output:
                         bundler.addpart(out)
                     stream = util.chunkbuffer(bundler.getchunks())
                     b = bundle2.getunbundler(self.ui, stream)
                     bundle2.processbundle(self._repo, b)
                 raise
         except error.PushRaced as exc:
             raise error.ResponseError(
                 _(b'push failed:'), stringutil.forcebytestr(exc)
             )
 
     # End of _basewirecommands interface.
 
     # Begin of peer interface.
 
     def commandexecutor(self):
         return localcommandexecutor(self)
 
     # End of peer interface.
 
 
 @interfaceutil.implementer(repository.ipeerlegacycommands)
 class locallegacypeer(localpeer):
     """peer extension which implements legacy methods too; used for tests with
     restricted capabilities"""
 
     def __init__(self, repo):
         super(locallegacypeer, self).__init__(repo, caps=legacycaps)
 
     # Begin of baselegacywirecommands interface.
 
     def between(self, pairs):
         return self._repo.between(pairs)
 
     def branches(self, nodes):
         return self._repo.branches(nodes)
 
     def changegroup(self, nodes, source):
         outgoing = discovery.outgoing(
             self._repo, missingroots=nodes, ancestorsof=self._repo.heads()
         )
         return changegroup.makechangegroup(self._repo, outgoing, b'01', source)
 
     def changegroupsubset(self, bases, heads, source):
         outgoing = discovery.outgoing(
             self._repo, missingroots=bases, ancestorsof=heads
         )
         return changegroup.makechangegroup(self._repo, outgoing, b'01', source)
 
     # End of baselegacywirecommands interface.
 
 
 # Functions receiving (ui, features) that extensions can register to impact
 # the ability to load repositories with custom requirements. Only
 # functions defined in loaded extensions are called.
 #
 # The function receives a set of requirement strings that the repository
 # is capable of opening. Functions will typically add elements to the
 # set to reflect that the extension knows how to handle that requirements.
 featuresetupfuncs = set()
 
 
 def _getsharedvfs(hgvfs, requirements):
     """returns the vfs object pointing to root of shared source
     repo for a shared repository
 
     hgvfs is vfs pointing at .hg/ of current repo (shared one)
     requirements is a set of requirements of current repo (shared one)
     """
     # The ``shared`` or ``relshared`` requirements indicate the
     # store lives in the path contained in the ``.hg/sharedpath`` file.
     # This is an absolute path for ``shared`` and relative to
     # ``.hg/`` for ``relshared``.
     sharedpath = hgvfs.read(b'sharedpath').rstrip(b'\n')
     if requirementsmod.RELATIVE_SHARED_REQUIREMENT in requirements:
         sharedpath = util.normpath(hgvfs.join(sharedpath))
 
     sharedvfs = vfsmod.vfs(sharedpath, realpath=True)
 
     if not sharedvfs.exists():
         raise error.RepoError(
             _(b'.hg/sharedpath points to nonexistent directory %s')
             % sharedvfs.base
         )
     return sharedvfs
 
 
 def _readrequires(vfs, allowmissing):
     """reads the require file present at root of this vfs
     and return a set of requirements
 
     If allowmissing is True, we suppress ENOENT if raised"""
     # requires file contains a newline-delimited list of
     # features/capabilities the opener (us) must have in order to use
     # the repository. This file was introduced in Mercurial 0.9.2,
     # which means very old repositories may not have one. We assume
     # a missing file translates to no requirements.
     try:
         requirements = set(vfs.read(b'requires').splitlines())
     except IOError as e:
         if not (allowmissing and e.errno == errno.ENOENT):
             raise
         requirements = set()
     return requirements
 
 
 def makelocalrepository(baseui, path, intents=None):
     """Create a local repository object.
 
     Given arguments needed to construct a local repository, this function
     performs various early repository loading functionality (such as
     reading the ``.hg/requires`` and ``.hg/hgrc`` files), validates that
     the repository can be opened, derives a type suitable for representing
     that repository, and returns an instance of it.
 
     The returned object conforms to the ``repository.completelocalrepository``
     interface.
 
     The repository type is derived by calling a series of factory functions
     for each aspect/interface of the final repository. These are defined by
     ``REPO_INTERFACES``.
 
     Each factory function is called to produce a type implementing a specific
     interface. The cumulative list of returned types will be combined into a
     new type and that type will be instantiated to represent the local
     repository.
 
     The factory functions each receive various state that may be consulted
     as part of deriving a type.
 
     Extensions should wrap these factory functions to customize repository type
     creation. Note that an extension's wrapped function may be called even if
     that extension is not loaded for the repo being constructed. Extensions
     should check if their ``__name__`` appears in the
     ``extensionmodulenames`` set passed to the factory function and no-op if
     not.
     """
     ui = baseui.copy()
     # Prevent copying repo configuration.
     ui.copy = baseui.copy
 
     # Working directory VFS rooted at repository root.
     wdirvfs = vfsmod.vfs(path, expandpath=True, realpath=True)
 
     # Main VFS for .hg/ directory.
     hgpath = wdirvfs.join(b'.hg')
     hgvfs = vfsmod.vfs(hgpath, cacheaudited=True)
     # Whether this repository is shared one or not
     shared = False
     # If this repository is shared, vfs pointing to shared repo
     sharedvfs = None
 
     # The .hg/ path should exist and should be a directory. All other
     # cases are errors.
     if not hgvfs.isdir():
         try:
             hgvfs.stat()
         except OSError as e:
             if e.errno != errno.ENOENT:
                 raise
         except ValueError as e:
             # Can be raised on Python 3.8 when path is invalid.
             raise error.Abort(
                 _(b'invalid path %s: %s') % (path, stringutil.forcebytestr(e))
             )
 
         raise error.RepoError(_(b'repository %s not found') % path)
 
     requirements = _readrequires(hgvfs, True)
     shared = (
         requirementsmod.SHARED_REQUIREMENT in requirements
         or requirementsmod.RELATIVE_SHARED_REQUIREMENT in requirements
     )
     storevfs = None
     if shared:
         # This is a shared repo
         sharedvfs = _getsharedvfs(hgvfs, requirements)
         storevfs = vfsmod.vfs(sharedvfs.join(b'store'))
     else:
         storevfs = vfsmod.vfs(hgvfs.join(b'store'))
 
     # if .hg/requires contains the sharesafe requirement, it means
     # there exists a `.hg/store/requires` too and we should read it
     # NOTE: presence of SHARESAFE_REQUIREMENT imply that store requirement
     # is present. We never write SHARESAFE_REQUIREMENT for a repo if store
     # is not present, refer checkrequirementscompat() for that
     #
     # However, if SHARESAFE_REQUIREMENT is not present, it means that the
     # repository was shared the old way. We check the share source .hg/requires
     # for SHARESAFE_REQUIREMENT to detect whether the current repository needs
     # to be reshared
     hint = _(b"see `hg help config.format.use-share-safe` for more information")
     if requirementsmod.SHARESAFE_REQUIREMENT in requirements:
 
         if (
             shared
             and requirementsmod.SHARESAFE_REQUIREMENT
             not in _readrequires(sharedvfs, True)
         ):
             mismatch_warn = ui.configbool(
                 b'share', b'safe-mismatch.source-not-safe.warn'
             )
             mismatch_config = ui.config(
                 b'share', b'safe-mismatch.source-not-safe'
             )
             if mismatch_config in (
                 b'downgrade-allow',
                 b'allow',
                 b'downgrade-abort',
             ):
                 # prevent cyclic import localrepo -> upgrade -> localrepo
                 from . import upgrade
 
                 upgrade.downgrade_share_to_non_safe(
                     ui,
                     hgvfs,
                     sharedvfs,
                     requirements,
                     mismatch_config,
                     mismatch_warn,
                 )
             elif mismatch_config == b'abort':
                 raise error.Abort(
                     _(b"share source does not support share-safe requirement"),
                     hint=hint,
                 )
             else:
                 raise error.Abort(
                     _(
                         b"share-safe mismatch with source.\nUnrecognized"
                         b" value '%s' of `share.safe-mismatch.source-not-safe`"
                         b" set."
                     )
                     % mismatch_config,
                     hint=hint,
                 )
         else:
             requirements |= _readrequires(storevfs, False)
     elif shared:
         sourcerequires = _readrequires(sharedvfs, False)
         if requirementsmod.SHARESAFE_REQUIREMENT in sourcerequires:
             mismatch_config = ui.config(b'share', b'safe-mismatch.source-safe')
             mismatch_warn = ui.configbool(
                 b'share', b'safe-mismatch.source-safe.warn'
             )
             if mismatch_config in (
                 b'upgrade-allow',
                 b'allow',
                 b'upgrade-abort',
             ):
                 # prevent cyclic import localrepo -> upgrade -> localrepo
                 from . import upgrade
 
                 upgrade.upgrade_share_to_safe(
                     ui,
                     hgvfs,
                     storevfs,
                     requirements,
                     mismatch_config,
                     mismatch_warn,
                 )
             elif mismatch_config == b'abort':
                 raise error.Abort(
                     _(
                         b'version mismatch: source uses share-safe'
                         b' functionality while the current share does not'
                     ),
                     hint=hint,
                 )
             else:
                 raise error.Abort(
                     _(
                         b"share-safe mismatch with source.\nUnrecognized"
                         b" value '%s' of `share.safe-mismatch.source-safe` set."
                     )
                     % mismatch_config,
                     hint=hint,
                 )
 
     # The .hg/hgrc file may load extensions or contain config options
     # that influence repository construction. Attempt to load it and
     # process any new extensions that it may have pulled in.
     if loadhgrc(ui, wdirvfs, hgvfs, requirements, sharedvfs):
         afterhgrcload(ui, wdirvfs, hgvfs, requirements)
         extensions.loadall(ui)
         extensions.populateui(ui)
 
     # Set of module names of extensions loaded for this repository.
     extensionmodulenames = {m.__name__ for n, m in extensions.extensions(ui)}
 
     supportedrequirements = gathersupportedrequirements(ui)
 
     # We first validate the requirements are known.
     ensurerequirementsrecognized(requirements, supportedrequirements)
 
     # Then we validate that the known set is reasonable to use together.
     ensurerequirementscompatible(ui, requirements)
 
     # TODO there are unhandled edge cases related to opening repositories with
     # shared storage. If storage is shared, we should also test for requirements
     # compatibility in the pointed-to repo. This entails loading the .hg/hgrc in
     # that repo, as that repo may load extensions needed to open it. This is a
     # bit complicated because we don't want the other hgrc to overwrite settings
     # in this hgrc.
     #
     # This bug is somewhat mitigated by the fact that we copy the .hg/requires
     # file when sharing repos. But if a requirement is added after the share is
     # performed, thereby introducing a new requirement for the opener, we may
     # will not see that and could encounter a run-time error interacting with
     # that shared store since it has an unknown-to-us requirement.
 
     # At this point, we know we should be capable of opening the repository.
     # Now get on with doing that.
 
     features = set()
 
     # The "store" part of the repository holds versioned data. How it is
     # accessed is determined by various requirements. If `shared` or
     # `relshared` requirements are present, this indicates current repository
     # is a share and store exists in path mentioned in `.hg/sharedpath`
     if shared:
         storebasepath = sharedvfs.base
         cachepath = sharedvfs.join(b'cache')
         features.add(repository.REPO_FEATURE_SHARED_STORAGE)
     else:
         storebasepath = hgvfs.base
         cachepath = hgvfs.join(b'cache')
     wcachepath = hgvfs.join(b'wcache')
 
     # The store has changed over time and the exact layout is dictated by
     # requirements. The store interface abstracts differences across all
     # of them.
     store = makestore(
         requirements,
         storebasepath,
         lambda base: vfsmod.vfs(base, cacheaudited=True),
     )
     hgvfs.createmode = store.createmode
 
     storevfs = store.vfs
     storevfs.options = resolvestorevfsoptions(ui, requirements, features)
 
     # The cache vfs is used to manage cache files.
     cachevfs = vfsmod.vfs(cachepath, cacheaudited=True)
     cachevfs.createmode = store.createmode
     # The cache vfs is used to manage cache files related to the working copy
     wcachevfs = vfsmod.vfs(wcachepath, cacheaudited=True)
     wcachevfs.createmode = store.createmode
 
     # Now resolve the type for the repository object. We do this by repeatedly
     # calling a factory function to produces types for specific aspects of the
     # repo's operation. The aggregate returned types are used as base classes
     # for a dynamically-derived type, which will represent our new repository.
 
     bases = []
     extrastate = {}
 
     for iface, fn in REPO_INTERFACES:
         # We pass all potentially useful state to give extensions tons of
         # flexibility.
         typ = fn()(
             ui=ui,
             intents=intents,
             requirements=requirements,
             features=features,
             wdirvfs=wdirvfs,
             hgvfs=hgvfs,
             store=store,
             storevfs=storevfs,
             storeoptions=storevfs.options,
             cachevfs=cachevfs,
             wcachevfs=wcachevfs,
             extensionmodulenames=extensionmodulenames,
             extrastate=extrastate,
             baseclasses=bases,
         )
 
         if not isinstance(typ, type):
             raise error.ProgrammingError(
                 b'unable to construct type for %s' % iface
             )
 
         bases.append(typ)
 
     # type() allows you to use characters in type names that wouldn't be
     # recognized as Python symbols in source code. We abuse that to add
     # rich information about our constructed repo.
     name = pycompat.sysstr(
         b'derivedrepo:%s<%s>' % (wdirvfs.base, b','.join(sorted(requirements)))
     )
 
     cls = type(name, tuple(bases), {})
 
     return cls(
         baseui=baseui,
         ui=ui,
         origroot=path,
         wdirvfs=wdirvfs,
         hgvfs=hgvfs,
         requirements=requirements,
         supportedrequirements=supportedrequirements,
         sharedpath=storebasepath,
         store=store,
         cachevfs=cachevfs,
         wcachevfs=wcachevfs,
         features=features,
         intents=intents,
     )
 
 
 def loadhgrc(ui, wdirvfs, hgvfs, requirements, sharedvfs=None):
     """Load hgrc files/content into a ui instance.
 
     This is called during repository opening to load any additional
     config files or settings relevant to the current repository.
 
     Returns a bool indicating whether any additional configs were loaded.
 
     Extensions should monkeypatch this function to modify how per-repo
     configs are loaded. For example, an extension may wish to pull in
     configs from alternate files or sources.
 
     sharedvfs is vfs object pointing to source repo if the current one is a
     shared one
     """
     if not rcutil.use_repo_hgrc():
         return False
 
     ret = False
     # first load config from shared source if we has to
     if requirementsmod.SHARESAFE_REQUIREMENT in requirements and sharedvfs:
         try:
             ui.readconfig(sharedvfs.join(b'hgrc'), root=sharedvfs.base)
             ret = True
         except IOError:
             pass
 
     try:
         ui.readconfig(hgvfs.join(b'hgrc'), root=wdirvfs.base)
         ret = True
     except IOError:
         pass
 
     try:
         ui.readconfig(hgvfs.join(b'hgrc-not-shared'), root=wdirvfs.base)
         ret = True
     except IOError:
         pass
 
     return ret
 
 
 def afterhgrcload(ui, wdirvfs, hgvfs, requirements):
     """Perform additional actions after .hg/hgrc is loaded.
 
     This function is called during repository loading immediately after
     the .hg/hgrc file is loaded and before per-repo extensions are loaded.
 
     The function can be used to validate configs, automatically add
     options (including extensions) based on requirements, etc.
     """
 
     # Map of requirements to list of extensions to load automatically when
     # requirement is present.
     autoextensions = {
         b'git': [b'git'],
         b'largefiles': [b'largefiles'],
         b'lfs': [b'lfs'],
     }
 
     for requirement, names in sorted(autoextensions.items()):
         if requirement not in requirements:
             continue
 
         for name in names:
             if not ui.hasconfig(b'extensions', name):
                 ui.setconfig(b'extensions', name, b'', source=b'autoload')
 
 
 def gathersupportedrequirements(ui):
     """Determine the complete set of recognized requirements."""
     # Start with all requirements supported by this file.
     supported = set(localrepository._basesupported)
 
     # Execute ``featuresetupfuncs`` entries if they belong to an extension
     # relevant to this ui instance.
     modules = {m.__name__ for n, m in extensions.extensions(ui)}
 
     for fn in featuresetupfuncs:
         if fn.__module__ in modules:
             fn(ui, supported)
 
     # Add derived requirements from registered compression engines.
     for name in util.compengines:
         engine = util.compengines[name]
         if engine.available() and engine.revlogheader():
             supported.add(b'exp-compression-%s' % name)
             if engine.name() == b'zstd':
                 supported.add(b'revlog-compression-zstd')
 
     return supported
 
 
 def ensurerequirementsrecognized(requirements, supported):
     """Validate that a set of local requirements is recognized.
 
     Receives a set of requirements. Raises an ``error.RepoError`` if there
     exists any requirement in that set that currently loaded code doesn't
     recognize.
 
     Returns a set of supported requirements.
     """
     missing = set()
 
     for requirement in requirements:
         if requirement in supported:
             continue
 
         if not requirement or not requirement[0:1].isalnum():
             raise error.RequirementError(_(b'.hg/requires file is corrupt'))
 
         missing.add(requirement)
 
     if missing:
         raise error.RequirementError(
             _(b'repository requires features unknown to this Mercurial: %s')
             % b' '.join(sorted(missing)),
             hint=_(
                 b'see https://mercurial-scm.org/wiki/MissingRequirement '
                 b'for more information'
             ),
         )
 
 
 def ensurerequirementscompatible(ui, requirements):
     """Validates that a set of recognized requirements is mutually compatible.
 
     Some requirements may not be compatible with others or require
     config options that aren't enabled. This function is called during
     repository opening to ensure that the set of requirements needed
     to open a repository is sane and compatible with config options.
 
     Extensions can monkeypatch this function to perform additional
     checking.
 
     ``error.RepoError`` should be raised on failure.
     """
     if (
         requirementsmod.SPARSE_REQUIREMENT in requirements
         and not sparse.enabled
     ):
         raise error.RepoError(
             _(
                 b'repository is using sparse feature but '
                 b'sparse is not enabled; enable the '
                 b'"sparse" extensions to access'
             )
         )
 
 
 def makestore(requirements, path, vfstype):
     """Construct a storage object for a repository."""
     if requirementsmod.STORE_REQUIREMENT in requirements:
         if requirementsmod.FNCACHE_REQUIREMENT in requirements:
             dotencode = requirementsmod.DOTENCODE_REQUIREMENT in requirements
             return storemod.fncachestore(path, vfstype, dotencode)
 
         return storemod.encodedstore(path, vfstype)
 
     return storemod.basicstore(path, vfstype)
 
 
 def resolvestorevfsoptions(ui, requirements, features):
     """Resolve the options to pass to the store vfs opener.
 
     The returned dict is used to influence behavior of the storage layer.
     """
     options = {}
 
     if requirementsmod.TREEMANIFEST_REQUIREMENT in requirements:
         options[b'treemanifest'] = True
 
     # experimental config: format.manifestcachesize
     manifestcachesize = ui.configint(b'format', b'manifestcachesize')
     if manifestcachesize is not None:
         options[b'manifestcachesize'] = manifestcachesize
 
     # In the absence of another requirement superseding a revlog-related
     # requirement, we have to assume the repo is using revlog version 0.
     # This revlog format is super old and we don't bother trying to parse
     # opener options for it because those options wouldn't do anything
     # meaningful on such old repos.
     if (
         requirementsmod.REVLOGV1_REQUIREMENT in requirements
         or requirementsmod.REVLOGV2_REQUIREMENT in requirements
     ):
         options.update(resolverevlogstorevfsoptions(ui, requirements, features))
     else:  # explicitly mark repo as using revlogv0
         options[b'revlogv0'] = True
 
     if requirementsmod.COPIESSDC_REQUIREMENT in requirements:
         options[b'copies-storage'] = b'changeset-sidedata'
     else:
         writecopiesto = ui.config(b'experimental', b'copies.write-to')
         copiesextramode = (b'changeset-only', b'compatibility')
         if writecopiesto in copiesextramode:
             options[b'copies-storage'] = b'extra'
 
     return options
 
 
 def resolverevlogstorevfsoptions(ui, requirements, features):
     """Resolve opener options specific to revlogs."""
 
     options = {}
     options[b'flagprocessors'] = {}
 
     if requirementsmod.REVLOGV1_REQUIREMENT in requirements:
         options[b'revlogv1'] = True
     if requirementsmod.REVLOGV2_REQUIREMENT in requirements:
         options[b'revlogv2'] = True
 
     if requirementsmod.GENERALDELTA_REQUIREMENT in requirements:
         options[b'generaldelta'] = True
 
     # experimental config: format.chunkcachesize
     chunkcachesize = ui.configint(b'format', b'chunkcachesize')
     if chunkcachesize is not None:
         options[b'chunkcachesize'] = chunkcachesize
 
     deltabothparents = ui.configbool(
         b'storage', b'revlog.optimize-delta-parent-choice'
     )
     options[b'deltabothparents'] = deltabothparents
 
     lazydelta = ui.configbool(b'storage', b'revlog.reuse-external-delta')
     lazydeltabase = False
     if lazydelta:
         lazydeltabase = ui.configbool(
             b'storage', b'revlog.reuse-external-delta-parent'
         )
     if lazydeltabase is None:
         lazydeltabase = not scmutil.gddeltaconfig(ui)
     options[b'lazydelta'] = lazydelta
     options[b'lazydeltabase'] = lazydeltabase
 
     chainspan = ui.configbytes(b'experimental', b'maxdeltachainspan')
     if 0 <= chainspan:
         options[b'maxdeltachainspan'] = chainspan
 
     mmapindexthreshold = ui.configbytes(b'experimental', b'mmapindexthreshold')
     if mmapindexthreshold is not None:
         options[b'mmapindexthreshold'] = mmapindexthreshold
 
     withsparseread = ui.configbool(b'experimental', b'sparse-read')
     srdensitythres = float(
         ui.config(b'experimental', b'sparse-read.density-threshold')
     )
     srmingapsize = ui.configbytes(b'experimental', b'sparse-read.min-gap-size')
     options[b'with-sparse-read'] = withsparseread
     options[b'sparse-read-density-threshold'] = srdensitythres
     options[b'sparse-read-min-gap-size'] = srmingapsize
 
     sparserevlog = requirementsmod.SPARSEREVLOG_REQUIREMENT in requirements
     options[b'sparse-revlog'] = sparserevlog
     if sparserevlog:
         options[b'generaldelta'] = True
 
     sidedata = requirementsmod.SIDEDATA_REQUIREMENT in requirements
     options[b'side-data'] = sidedata
 
     maxchainlen = None
     if sparserevlog:
         maxchainlen = revlogconst.SPARSE_REVLOG_MAX_CHAIN_LENGTH
     # experimental config: format.maxchainlen
     maxchainlen = ui.configint(b'format', b'maxchainlen', maxchainlen)
     if maxchainlen is not None:
         options[b'maxchainlen'] = maxchainlen
 
     for r in requirements:
         # we allow multiple compression engine requirement to co-exist because
         # strickly speaking, revlog seems to support mixed compression style.
         #
         # The compression used for new entries will be "the last one"
         prefix = r.startswith
         if prefix(b'revlog-compression-') or prefix(b'exp-compression-'):
             options[b'compengine'] = r.split(b'-', 2)[2]
 
     options[b'zlib.level'] = ui.configint(b'storage', b'revlog.zlib.level')
     if options[b'zlib.level'] is not None:
         if not (0 <= options[b'zlib.level'] <= 9):
             msg = _(b'invalid value for `storage.revlog.zlib.level` config: %d')
             raise error.Abort(msg % options[b'zlib.level'])
     options[b'zstd.level'] = ui.configint(b'storage', b'revlog.zstd.level')
     if options[b'zstd.level'] is not None:
         if not (0 <= options[b'zstd.level'] <= 22):
             msg = _(b'invalid value for `storage.revlog.zstd.level` config: %d')
             raise error.Abort(msg % options[b'zstd.level'])
 
     if requirementsmod.NARROW_REQUIREMENT in requirements:
         options[b'enableellipsis'] = True
 
     if ui.configbool(b'experimental', b'rust.index'):
         options[b'rust.index'] = True
     if requirementsmod.NODEMAP_REQUIREMENT in requirements:
         slow_path = ui.config(
             b'storage', b'revlog.persistent-nodemap.slow-path'
         )
         if slow_path not in (b'allow', b'warn', b'abort'):
             default = ui.config_default(
                 b'storage', b'revlog.persistent-nodemap.slow-path'
             )
             msg = _(
                 b'unknown value for config '
                 b'"storage.revlog.persistent-nodemap.slow-path": "%s"\n'
             )
             ui.warn(msg % slow_path)
             if not ui.quiet:
                 ui.warn(_(b'falling back to default value: %s\n') % default)
             slow_path = default
 
         msg = _(
             b"accessing `persistent-nodemap` repository without associated "
             b"fast implementation."
         )
         hint = _(
             b"check `hg help config.format.use-persistent-nodemap` "
             b"for details"
         )
         if not revlog.HAS_FAST_PERSISTENT_NODEMAP:
             if slow_path == b'warn':
                 msg = b"warning: " + msg + b'\n'
                 ui.warn(msg)
                 if not ui.quiet:
                     hint = b'(' + hint + b')\n'
                     ui.warn(hint)
             if slow_path == b'abort':
                 raise error.Abort(msg, hint=hint)
         options[b'persistent-nodemap'] = True
     if ui.configbool(b'storage', b'revlog.persistent-nodemap.mmap'):
         options[b'persistent-nodemap.mmap'] = True
     if ui.configbool(b'devel', b'persistent-nodemap'):
         options[b'devel-force-nodemap'] = True
 
     return options
 
 
 def makemain(**kwargs):
     """Produce a type conforming to ``ilocalrepositorymain``."""
     return localrepository
 
 
 @interfaceutil.implementer(repository.ilocalrepositoryfilestorage)
 class revlogfilestorage(object):
     """File storage when using revlogs."""
 
     def file(self, path):
         if path.startswith(b'/'):
             path = path[1:]
 
         return filelog.filelog(self.svfs, path)
 
 
 @interfaceutil.implementer(repository.ilocalrepositoryfilestorage)
 class revlognarrowfilestorage(object):
     """File storage when using revlogs and narrow files."""
 
     def file(self, path):
         if path.startswith(b'/'):
             path = path[1:]
 
         return filelog.narrowfilelog(self.svfs, path, self._storenarrowmatch)
 
 
 def makefilestorage(requirements, features, **kwargs):
     """Produce a type conforming to ``ilocalrepositoryfilestorage``."""
     features.add(repository.REPO_FEATURE_REVLOG_FILE_STORAGE)
     features.add(repository.REPO_FEATURE_STREAM_CLONE)
 
     if requirementsmod.NARROW_REQUIREMENT in requirements:
         return revlognarrowfilestorage
     else:
         return revlogfilestorage
 
 
 # List of repository interfaces and factory functions for them. Each
 # will be called in order during ``makelocalrepository()`` to iteratively
 # derive the final type for a local repository instance. We capture the
 # function as a lambda so we don't hold a reference and the module-level
 # functions can be wrapped.
 REPO_INTERFACES = [
     (repository.ilocalrepositorymain, lambda: makemain),
     (repository.ilocalrepositoryfilestorage, lambda: makefilestorage),
 ]
 
 
 @interfaceutil.implementer(repository.ilocalrepositorymain)
 class localrepository(object):
     """Main class for representing local repositories.
 
     All local repositories are instances of this class.
 
     Constructed on its own, instances of this class are not usable as
     repository objects. To obtain a usable repository object, call
     ``hg.repository()``, ``localrepo.instance()``, or
     ``localrepo.makelocalrepository()``. The latter is the lowest-level.
     ``instance()`` adds support for creating new repositories.
     ``hg.repository()`` adds more extension integration, including calling
     ``reposetup()``. Generally speaking, ``hg.repository()`` should be
     used.
     """
 
     # obsolete experimental requirements:
     #  - manifestv2: An experimental new manifest format that allowed
     #    for stem compression of long paths. Experiment ended up not
     #    being successful (repository sizes went up due to worse delta
     #    chains), and the code was deleted in 4.6.
     supportedformats = {
         requirementsmod.REVLOGV1_REQUIREMENT,
         requirementsmod.GENERALDELTA_REQUIREMENT,
         requirementsmod.TREEMANIFEST_REQUIREMENT,
         requirementsmod.COPIESSDC_REQUIREMENT,
         requirementsmod.REVLOGV2_REQUIREMENT,
         requirementsmod.SIDEDATA_REQUIREMENT,
         requirementsmod.SPARSEREVLOG_REQUIREMENT,
         requirementsmod.NODEMAP_REQUIREMENT,
         bookmarks.BOOKMARKS_IN_STORE_REQUIREMENT,
         requirementsmod.SHARESAFE_REQUIREMENT,
     }
     _basesupported = supportedformats | {
         requirementsmod.STORE_REQUIREMENT,
         requirementsmod.FNCACHE_REQUIREMENT,
         requirementsmod.SHARED_REQUIREMENT,
         requirementsmod.RELATIVE_SHARED_REQUIREMENT,
         requirementsmod.DOTENCODE_REQUIREMENT,
         requirementsmod.SPARSE_REQUIREMENT,
         requirementsmod.INTERNAL_PHASE_REQUIREMENT,
     }
 
     # list of prefix for file which can be written without 'wlock'
     # Extensions should extend this list when needed
     _wlockfreeprefix = {
         # We migh consider requiring 'wlock' for the next
         # two, but pretty much all the existing code assume
         # wlock is not needed so we keep them excluded for
         # now.
         b'hgrc',
         b'requires',
         # XXX cache is a complicatged business someone
         # should investigate this in depth at some point
         b'cache/',
         # XXX shouldn't be dirstate covered by the wlock?
         b'dirstate',
         # XXX bisect was still a bit too messy at the time
         # this changeset was introduced. Someone should fix
         # the remainig bit and drop this line
         b'bisect.state',
     }
 
     def __init__(
         self,
         baseui,
         ui,
         origroot,
         wdirvfs,
         hgvfs,
         requirements,
         supportedrequirements,
         sharedpath,
         store,
         cachevfs,
         wcachevfs,
         features,
         intents=None,
     ):
         """Create a new local repository instance.
 
         Most callers should use ``hg.repository()``, ``localrepo.instance()``,
         or ``localrepo.makelocalrepository()`` for obtaining a new repository
         object.
 
         Arguments:
 
         baseui
            ``ui.ui`` instance that ``ui`` argument was based off of.
 
         ui
            ``ui.ui`` instance for use by the repository.
 
         origroot
            ``bytes`` path to working directory root of this repository.
 
         wdirvfs
            ``vfs.vfs`` rooted at the working directory.
 
         hgvfs
            ``vfs.vfs`` rooted at .hg/
 
         requirements
            ``set`` of bytestrings representing repository opening requirements.
 
         supportedrequirements
            ``set`` of bytestrings representing repository requirements that we
            know how to open. May be a supetset of ``requirements``.
 
         sharedpath
            ``bytes`` Defining path to storage base directory. Points to a
            ``.hg/`` directory somewhere.
 
         store
            ``store.basicstore`` (or derived) instance providing access to
            versioned storage.
 
         cachevfs
            ``vfs.vfs`` used for cache files.
 
         wcachevfs
            ``vfs.vfs`` used for cache files related to the working copy.
 
         features
            ``set`` of bytestrings defining features/capabilities of this
            instance.
 
         intents
            ``set`` of system strings indicating what this repo will be used
            for.
         """
         self.baseui = baseui
         self.ui = ui
         self.origroot = origroot
         # vfs rooted at working directory.
         self.wvfs = wdirvfs
         self.root = wdirvfs.base
         # vfs rooted at .hg/. Used to access most non-store paths.
         self.vfs = hgvfs
         self.path = hgvfs.base
         self.requirements = requirements
         self.nodeconstants = sha1nodeconstants
         self.nullid = self.nodeconstants.nullid
         self.supported = supportedrequirements
         self.sharedpath = sharedpath
         self.store = store
         self.cachevfs = cachevfs
         self.wcachevfs = wcachevfs
         self.features = features
 
         self.filtername = None
 
         if self.ui.configbool(b'devel', b'all-warnings') or self.ui.configbool(
             b'devel', b'check-locks'
         ):
             self.vfs.audit = self._getvfsward(self.vfs.audit)
         # A list of callback to shape the phase if no data were found.
         # Callback are in the form: func(repo, roots) --> processed root.
         # This list it to be filled by extension during repo setup
         self._phasedefaults = []
 
         color.setup(self.ui)
 
         self.spath = self.store.path
         self.svfs = self.store.vfs
         self.sjoin = self.store.join
         if self.ui.configbool(b'devel', b'all-warnings') or self.ui.configbool(
             b'devel', b'check-locks'
         ):
             if util.safehasattr(self.svfs, b'vfs'):  # this is filtervfs
                 self.svfs.vfs.audit = self._getsvfsward(self.svfs.vfs.audit)
             else:  # standard vfs
                 self.svfs.audit = self._getsvfsward(self.svfs.audit)
 
         self._dirstatevalidatewarned = False
 
         self._branchcaches = branchmap.BranchMapCache()
         self._revbranchcache = None
         self._filterpats = {}
         self._datafilters = {}
         self._transref = self._lockref = self._wlockref = None
 
         # A cache for various files under .hg/ that tracks file changes,
         # (used by the filecache decorator)
         #
         # Maps a property name to its util.filecacheentry
         self._filecache = {}
 
         # hold sets of revision to be filtered
         # should be cleared when something might have changed the filter value:
         # - new changesets,
         # - phase change,
         # - new obsolescence marker,
         # - working directory parent change,
         # - bookmark changes
         self.filteredrevcache = {}
 
         # post-dirstate-status hooks
         self._postdsstatus = []
 
         # generic mapping between names and nodes
         self.names = namespaces.namespaces()
 
         # Key to signature value.
         self._sparsesignaturecache = {}
         # Signature to cached matcher instance.
         self._sparsematchercache = {}
 
         self._extrafilterid = repoview.extrafilter(ui)
 
         self.filecopiesmode = None
         if requirementsmod.COPIESSDC_REQUIREMENT in self.requirements:
             self.filecopiesmode = b'changeset-sidedata'
 
         self._wanted_sidedata = set()
         self._sidedata_computers = {}
         metadatamod.set_sidedata_spec_for_repo(self)
 
     def _getvfsward(self, origfunc):
         """build a ward for self.vfs"""
         rref = weakref.ref(self)
 
         def checkvfs(path, mode=None):
             ret = origfunc(path, mode=mode)
             repo = rref()
             if (
                 repo is None
                 or not util.safehasattr(repo, b'_wlockref')
                 or not util.safehasattr(repo, b'_lockref')
             ):
                 return
             if mode in (None, b'r', b'rb'):
                 return
             if path.startswith(repo.path):
                 # truncate name relative to the repository (.hg)
                 path = path[len(repo.path) + 1 :]
             if path.startswith(b'cache/'):
                 msg = b'accessing cache with vfs instead of cachevfs: "%s"'
                 repo.ui.develwarn(msg % path, stacklevel=3, config=b"cache-vfs")
             # path prefixes covered by 'lock'
             vfs_path_prefixes = (
                 b'journal.',
                 b'undo.',
                 b'strip-backup/',
                 b'cache/',
             )
             if any(path.startswith(prefix) for prefix in vfs_path_prefixes):
                 if repo._currentlock(repo._lockref) is None:
                     repo.ui.develwarn(
                         b'write with no lock: "%s"' % path,
                         stacklevel=3,
                         config=b'check-locks',
                     )
             elif repo._currentlock(repo._wlockref) is None:
                 # rest of vfs files are covered by 'wlock'
                 #
                 # exclude special files
                 for prefix in self._wlockfreeprefix:
                     if path.startswith(prefix):
                         return
                 repo.ui.develwarn(
                     b'write with no wlock: "%s"' % path,
                     stacklevel=3,
                     config=b'check-locks',
                 )
             return ret
 
         return checkvfs
 
     def _getsvfsward(self, origfunc):
         """build a ward for self.svfs"""
         rref = weakref.ref(self)
 
         def checksvfs(path, mode=None):
             ret = origfunc(path, mode=mode)
             repo = rref()
             if repo is None or not util.safehasattr(repo, b'_lockref'):
                 return
             if mode in (None, b'r', b'rb'):
                 return
             if path.startswith(repo.sharedpath):
                 # truncate name relative to the repository (.hg)
                 path = path[len(repo.sharedpath) + 1 :]
             if repo._currentlock(repo._lockref) is None:
                 repo.ui.develwarn(
                     b'write with no lock: "%s"' % path, stacklevel=4
                 )
             return ret
 
         return checksvfs
 
     def close(self):
         self._writecaches()
 
     def _writecaches(self):
         if self._revbranchcache:
             self._revbranchcache.write()
 
     def _restrictcapabilities(self, caps):
         if self.ui.configbool(b'experimental', b'bundle2-advertise'):
             caps = set(caps)
             capsblob = bundle2.encodecaps(
                 bundle2.getrepocaps(self, role=b'client')
             )
             caps.add(b'bundle2=' + urlreq.quote(capsblob))
         if self.ui.configbool(b'experimental', b'narrow'):
             caps.add(wireprototypes.NARROWCAP)
         return caps
 
     # Don't cache auditor/nofsauditor, or you'll end up with reference cycle:
     # self -> auditor -> self._checknested -> self
 
     @property
     def auditor(self):
         # This is only used by context.workingctx.match in order to
         # detect files in subrepos.
         return pathutil.pathauditor(self.root, callback=self._checknested)
 
     @property
     def nofsauditor(self):
         # This is only used by context.basectx.match in order to detect
         # files in subrepos.
         return pathutil.pathauditor(
             self.root, callback=self._checknested, realfs=False, cached=True
         )
 
     def _checknested(self, path):
         """Determine if path is a legal nested repository."""
         if not path.startswith(self.root):
             return False
         subpath = path[len(self.root) + 1 :]
         normsubpath = util.pconvert(subpath)
 
         # XXX: Checking against the current working copy is wrong in
         # the sense that it can reject things like
         #
         #   $ hg cat -r 10 sub/x.txt
         #
         # if sub/ is no longer a subrepository in the working copy
         # parent revision.
         #
         # However, it can of course also allow things that would have
         # been rejected before, such as the above cat command if sub/
         # is a subrepository now, but was a normal directory before.
         # The old path auditor would have rejected by mistake since it
         # panics when it sees sub/.hg/.
         #
         # All in all, checking against the working copy seems sensible
         # since we want to prevent access to nested repositories on
         # the filesystem *now*.
         ctx = self[None]
         parts = util.splitpath(subpath)
         while parts:
             prefix = b'/'.join(parts)
             if prefix in ctx.substate:
                 if prefix == normsubpath:
                     return True
                 else:
                     sub = ctx.sub(prefix)
                     return sub.checknested(subpath[len(prefix) + 1 :])
             else:
                 parts.pop()
         return False
 
     def peer(self):
         return localpeer(self)  # not cached to avoid reference cycle
 
     def unfiltered(self):
         """Return unfiltered version of the repository
 
         Intended to be overwritten by filtered repo."""
         return self
 
     def filtered(self, name, visibilityexceptions=None):
         """Return a filtered version of a repository
 
         The `name` parameter is the identifier of the requested view. This
         will return a repoview object set "exactly" to the specified view.
 
         This function does not apply recursive filtering to a repository. For
         example calling `repo.filtered("served")` will return a repoview using
         the "served" view, regardless of the initial view used by `repo`.
 
         In other word, there is always only one level of `repoview` "filtering".
         """
         if self._extrafilterid is not None and b'%' not in name:
             name = name + b'%' + self._extrafilterid
 
         cls = repoview.newtype(self.unfiltered().__class__)
         return cls(self, name, visibilityexceptions)
 
     @mixedrepostorecache(
         (b'bookmarks', b'plain'),
         (b'bookmarks.current', b'plain'),
         (b'bookmarks', b''),
         (b'00changelog.i', b''),
     )
     def _bookmarks(self):
         # Since the multiple files involved in the transaction cannot be
         # written atomically (with current repository format), there is a race
         # condition here.
         #
         # 1) changelog content A is read
         # 2) outside transaction update changelog to content B
         # 3) outside transaction update bookmark file referring to content B
         # 4) bookmarks file content is read and filtered against changelog-A
         #
         # When this happens, bookmarks against nodes missing from A are dropped.
         #
         # Having this happening during read is not great, but it become worse
         # when this happen during write because the bookmarks to the "unknown"
         # nodes will be dropped for good. However, writes happen within locks.
         # This locking makes it possible to have a race free consistent read.
         # For this purpose data read from disc before locking  are
         # "invalidated" right after the locks are taken. This invalidations are
         # "light", the `filecache` mechanism keep the data in memory and will
         # reuse them if the underlying files did not changed. Not parsing the
         # same data multiple times helps performances.
         #
         # Unfortunately in the case describe above, the files tracked by the
         # bookmarks file cache might not have changed, but the in-memory
         # content is still "wrong" because we used an older changelog content
         # to process the on-disk data. So after locking, the changelog would be
         # refreshed but `_bookmarks` would be preserved.
         # Adding `00changelog.i` to the list of tracked file is not
         # enough, because at the time we build the content for `_bookmarks` in
         # (4), the changelog file has already diverged from the content used
         # for loading `changelog` in (1)
         #
         # To prevent the issue, we force the changelog to be explicitly
         # reloaded while computing `_bookmarks`. The data race can still happen
         # without the lock (with a narrower window), but it would no longer go
         # undetected during the lock time refresh.
         #
         # The new schedule is as follow
         #
         # 1) filecache logic detect that `_bookmarks` needs to be computed
         # 2) cachestat for `bookmarks` and `changelog` are captured (for book)
         # 3) We force `changelog` filecache to be tested
         # 4) cachestat for `changelog` are captured (for changelog)
         # 5) `_bookmarks` is computed and cached
         #
         # The step in (3) ensure we have a changelog at least as recent as the
         # cache stat computed in (1). As a result at locking time:
         #  * if the changelog did not changed since (1) -> we can reuse the data
         #  * otherwise -> the bookmarks get refreshed.
         self._refreshchangelog()
         return bookmarks.bmstore(self)
 
     def _refreshchangelog(self):
         """make sure the in memory changelog match the on-disk one"""
         if 'changelog' in vars(self) and self.currenttransaction() is None:
             del self.changelog
 
     @property
     def _activebookmark(self):
         return self._bookmarks.active
 
     # _phasesets depend on changelog. what we need is to call
     # _phasecache.invalidate() if '00changelog.i' was changed, but it
     # can't be easily expressed in filecache mechanism.
     @storecache(b'phaseroots', b'00changelog.i')
     def _phasecache(self):
         return phases.phasecache(self, self._phasedefaults)
 
     @storecache(b'obsstore')
     def obsstore(self):
         return obsolete.makestore(self.ui, self)
 
     @storecache(b'00changelog.i')
     def changelog(self):
         # load dirstate before changelog to avoid race see issue6303
         self.dirstate.prefetch_parents()
         return self.store.changelog(
             txnutil.mayhavepending(self.root),
             concurrencychecker=revlogchecker.get_checker(self.ui, b'changelog'),
         )
 
     @storecache(b'00manifest.i')
     def manifestlog(self):
         return self.store.manifestlog(self, self._storenarrowmatch)
 
     @repofilecache(b'dirstate')
     def dirstate(self):
         return self._makedirstate()
 
     def _makedirstate(self):
         """Extension point for wrapping the dirstate per-repo."""
         sparsematchfn = lambda: sparse.matcher(self)
 
         return dirstate.dirstate(
             self.vfs,
             self.ui,
             self.root,
             self._dirstatevalidate,
             sparsematchfn,
             self.nodeconstants,
         )
 
     def _dirstatevalidate(self, node):
         try:
             self.changelog.rev(node)
             return node
         except error.LookupError:
             if not self._dirstatevalidatewarned:
                 self._dirstatevalidatewarned = True
                 self.ui.warn(
                     _(b"warning: ignoring unknown working parent %s!\n")
                     % short(node)
                 )
             return self.nullid
 
     @storecache(narrowspec.FILENAME)
     def narrowpats(self):
         """matcher patterns for this repository's narrowspec
 
         A tuple of (includes, excludes).
         """
         return narrowspec.load(self)
 
     @storecache(narrowspec.FILENAME)
     def _storenarrowmatch(self):
         if requirementsmod.NARROW_REQUIREMENT not in self.requirements:
             return matchmod.always()
         include, exclude = self.narrowpats
         return narrowspec.match(self.root, include=include, exclude=exclude)
 
     @storecache(narrowspec.FILENAME)
     def _narrowmatch(self):
         if requirementsmod.NARROW_REQUIREMENT not in self.requirements:
             return matchmod.always()
         narrowspec.checkworkingcopynarrowspec(self)
         include, exclude = self.narrowpats
         return narrowspec.match(self.root, include=include, exclude=exclude)
 
     def narrowmatch(self, match=None, includeexact=False):
         """matcher corresponding the the repo's narrowspec
 
         If `match` is given, then that will be intersected with the narrow
         matcher.
 
         If `includeexact` is True, then any exact matches from `match` will
         be included even if they're outside the narrowspec.
         """
         if match:
             if includeexact and not self._narrowmatch.always():
                 # do not exclude explicitly-specified paths so that they can
                 # be warned later on
                 em = matchmod.exact(match.files())
                 nm = matchmod.unionmatcher([self._narrowmatch, em])
                 return matchmod.intersectmatchers(match, nm)
             return matchmod.intersectmatchers(match, self._narrowmatch)
         return self._narrowmatch
 
     def setnarrowpats(self, newincludes, newexcludes):
         narrowspec.save(self, newincludes, newexcludes)
         self.invalidate(clearfilecache=True)
 
     @unfilteredpropertycache
     def _quick_access_changeid_null(self):
         return {
             b'null': (nullrev, self.nodeconstants.nullid),
             nullrev: (nullrev, self.nodeconstants.nullid),
             self.nullid: (nullrev, self.nullid),
         }
 
     @unfilteredpropertycache
     def _quick_access_changeid_wc(self):
         # also fast path access to the working copy parents
         # however, only do it for filter that ensure wc is visible.
         quick = self._quick_access_changeid_null.copy()
         cl = self.unfiltered().changelog
         for node in self.dirstate.parents():
             if node == self.nullid:
                 continue
             rev = cl.index.get_rev(node)
             if rev is None:
                 # unknown working copy parent case:
                 #
                 #   skip the fast path and let higher code deal with it
                 continue
             pair = (rev, node)
             quick[rev] = pair
             quick[node] = pair
             # also add the parents of the parents
             for r in cl.parentrevs(rev):
                 if r == nullrev:
                     continue
                 n = cl.node(r)
                 pair = (r, n)
                 quick[r] = pair
                 quick[n] = pair
         p1node = self.dirstate.p1()
         if p1node != self.nullid:
             quick[b'.'] = quick[p1node]
         return quick
 
     @unfilteredmethod
     def _quick_access_changeid_invalidate(self):
         if '_quick_access_changeid_wc' in vars(self):
             del self.__dict__['_quick_access_changeid_wc']
 
     @property
     def _quick_access_changeid(self):
         """an helper dictionnary for __getitem__ calls
 
         This contains a list of symbol we can recognise right away without
         further processing.
         """
         if self.filtername in repoview.filter_has_wc:
             return self._quick_access_changeid_wc
         return self._quick_access_changeid_null
 
     def __getitem__(self, changeid):
         # dealing with special cases
         if changeid is None:
             return context.workingctx(self)
         if isinstance(changeid, context.basectx):
             return changeid
 
         # dealing with multiple revisions
         if isinstance(changeid, slice):
             # wdirrev isn't contiguous so the slice shouldn't include it
             return [
                 self[i]
                 for i in pycompat.xrange(*changeid.indices(len(self)))
                 if i not in self.changelog.filteredrevs
             ]
 
         # dealing with some special values
         quick_access = self._quick_access_changeid.get(changeid)
         if quick_access is not None:
             rev, node = quick_access
             return context.changectx(self, rev, node, maybe_filtered=False)
         if changeid == b'tip':
             node = self.changelog.tip()
             rev = self.changelog.rev(node)
             return context.changectx(self, rev, node)
 
         # dealing with arbitrary values
         try:
             if isinstance(changeid, int):
                 node = self.changelog.node(changeid)
                 rev = changeid
             elif changeid == b'.':
                 # this is a hack to delay/avoid loading obsmarkers
                 # when we know that '.' won't be hidden
                 node = self.dirstate.p1()
                 rev = self.unfiltered().changelog.rev(node)
             elif len(changeid) == self.nodeconstants.nodelen:
                 try:
                     node = changeid
                     rev = self.changelog.rev(changeid)
                 except error.FilteredLookupError:
                     changeid = hex(changeid)  # for the error message
                     raise
                 except LookupError:
                     # check if it might have come from damaged dirstate
                     #
                     # XXX we could avoid the unfiltered if we had a recognizable
                     # exception for filtered changeset access
                     if (
                         self.local()
                         and changeid in self.unfiltered().dirstate.parents()
                     ):
                         msg = _(b"working directory has unknown parent '%s'!")
                         raise error.Abort(msg % short(changeid))
                     changeid = hex(changeid)  # for the error message
                     raise
 
             elif len(changeid) == 2 * self.nodeconstants.nodelen:
                 node = bin(changeid)
                 rev = self.changelog.rev(node)
             else:
                 raise error.ProgrammingError(
                     b"unsupported changeid '%s' of type %s"
                     % (changeid, pycompat.bytestr(type(changeid)))
                 )
 
             return context.changectx(self, rev, node)
 
         except (error.FilteredIndexError, error.FilteredLookupError):
             raise error.FilteredRepoLookupError(
                 _(b"filtered revision '%s'") % pycompat.bytestr(changeid)
             )
         except (IndexError, LookupError):
             raise error.RepoLookupError(
                 _(b"unknown revision '%s'") % pycompat.bytestr(changeid)
             )
         except error.WdirUnsupported:
             return context.workingctx(self)
 
     def __contains__(self, changeid):
         """True if the given changeid exists"""
         try:
             self[changeid]
             return True
         except error.RepoLookupError:
             return False
 
     def __nonzero__(self):
         return True
 
     __bool__ = __nonzero__
 
     def __len__(self):
         # no need to pay the cost of repoview.changelog
         unfi = self.unfiltered()
         return len(unfi.changelog)
 
     def __iter__(self):
         return iter(self.changelog)
 
     def revs(self, expr, *args):
         """Find revisions matching a revset.
 
         The revset is specified as a string ``expr`` that may contain
         %-formatting to escape certain types. See ``revsetlang.formatspec``.
 
         Revset aliases from the configuration are not expanded. To expand
         user aliases, consider calling ``scmutil.revrange()`` or
         ``repo.anyrevs([expr], user=True)``.
 
         Returns a smartset.abstractsmartset, which is a list-like interface
         that contains integer revisions.
         """
         tree = revsetlang.spectree(expr, *args)
         return revset.makematcher(tree)(self)
 
     def set(self, expr, *args):
         """Find revisions matching a revset and emit changectx instances.
 
         This is a convenience wrapper around ``revs()`` that iterates the
         result and is a generator of changectx instances.
 
         Revset aliases from the configuration are not expanded. To expand
         user aliases, consider calling ``scmutil.revrange()``.
         """
         for r in self.revs(expr, *args):
             yield self[r]
 
     def anyrevs(self, specs, user=False, localalias=None):
         """Find revisions matching one of the given revsets.
 
         Revset aliases from the configuration are not expanded by default. To
         expand user aliases, specify ``user=True``. To provide some local
         definitions overriding user aliases, set ``localalias`` to
         ``{name: definitionstring}``.
         """
         if specs == [b'null']:
             return revset.baseset([nullrev])
         if specs == [b'.']:
             quick_data = self._quick_access_changeid.get(b'.')
             if quick_data is not None:
                 return revset.baseset([quick_data[0]])
         if user:
             m = revset.matchany(
                 self.ui,
                 specs,
                 lookup=revset.lookupfn(self),
                 localalias=localalias,
             )
         else:
             m = revset.matchany(None, specs, localalias=localalias)
         return m(self)
 
     def url(self):
         return b'file:' + self.root
 
     def hook(self, name, throw=False, **args):
         """Call a hook, passing this repo instance.
 
         This a convenience method to aid invoking hooks. Extensions likely
         won't call this unless they have registered a custom hook or are
         replacing code that is expected to call a hook.
         """
         return hook.hook(self.ui, self, name, throw, **args)
 
     @filteredpropertycache
     def _tagscache(self):
         """Returns a tagscache object that contains various tags related
         caches."""
 
         # This simplifies its cache management by having one decorated
         # function (this one) and the rest simply fetch things from it.
         class tagscache(object):
             def __init__(self):
                 # These two define the set of tags for this repository. tags
                 # maps tag name to node; tagtypes maps tag name to 'global' or
                 # 'local'. (Global tags are defined by .hgtags across all
                 # heads, and local tags are defined in .hg/localtags.)
                 # They constitute the in-memory cache of tags.
                 self.tags = self.tagtypes = None
 
                 self.nodetagscache = self.tagslist = None
 
         cache = tagscache()
         cache.tags, cache.tagtypes = self._findtags()
 
         return cache
 
     def tags(self):
         '''return a mapping of tag to node'''
         t = {}
         if self.changelog.filteredrevs:
             tags, tt = self._findtags()
         else:
             tags = self._tagscache.tags
         rev = self.changelog.rev
         for k, v in pycompat.iteritems(tags):
             try:
                 # ignore tags to unknown nodes
                 rev(v)
                 t[k] = v
             except (error.LookupError, ValueError):
                 pass
         return t
 
     def _findtags(self):
         """Do the hard work of finding tags.  Return a pair of dicts
         (tags, tagtypes) where tags maps tag name to node, and tagtypes
         maps tag name to a string like \'global\' or \'local\'.
         Subclasses or extensions are free to add their own tags, but
         should be aware that the returned dicts will be retained for the
         duration of the localrepo object."""
 
         # XXX what tagtype should subclasses/extensions use?  Currently
         # mq and bookmarks add tags, but do not set the tagtype at all.
         # Should each extension invent its own tag type?  Should there
         # be one tagtype for all such "virtual" tags?  Or is the status
         # quo fine?
 
         # map tag name to (node, hist)
         alltags = tagsmod.findglobaltags(self.ui, self)
         # map tag name to tag type
         tagtypes = {tag: b'global' for tag in alltags}
 
         tagsmod.readlocaltags(self.ui, self, alltags, tagtypes)
 
         # Build the return dicts.  Have to re-encode tag names because
         # the tags module always uses UTF-8 (in order not to lose info
         # writing to the cache), but the rest of Mercurial wants them in
         # local encoding.
         tags = {}
         for (name, (node, hist)) in pycompat.iteritems(alltags):
             if node != self.nullid:
                 tags[encoding.tolocal(name)] = node
         tags[b'tip'] = self.changelog.tip()
         tagtypes = {
             encoding.tolocal(name): value
             for (name, value) in pycompat.iteritems(tagtypes)
         }
         return (tags, tagtypes)
 
     def tagtype(self, tagname):
         """
         return the type of the given tag. result can be:
 
         'local'  : a local tag
         'global' : a global tag
         None     : tag does not exist
         """
 
         return self._tagscache.tagtypes.get(tagname)
 
     def tagslist(self):
         '''return a list of tags ordered by revision'''
         if not self._tagscache.tagslist:
             l = []
             for t, n in pycompat.iteritems(self.tags()):
                 l.append((self.changelog.rev(n), t, n))
             self._tagscache.tagslist = [(t, n) for r, t, n in sorted(l)]
 
         return self._tagscache.tagslist
 
     def nodetags(self, node):
         '''return the tags associated with a node'''
         if not self._tagscache.nodetagscache:
             nodetagscache = {}
             for t, n in pycompat.iteritems(self._tagscache.tags):
                 nodetagscache.setdefault(n, []).append(t)
             for tags in pycompat.itervalues(nodetagscache):
                 tags.sort()
             self._tagscache.nodetagscache = nodetagscache
         return self._tagscache.nodetagscache.get(node, [])
 
     def nodebookmarks(self, node):
         """return the list of bookmarks pointing to the specified node"""
         return self._bookmarks.names(node)
 
     def branchmap(self):
         """returns a dictionary {branch: [branchheads]} with branchheads
         ordered by increasing revision number"""
         return self._branchcaches[self]
 
     @unfilteredmethod
     def revbranchcache(self):
         if not self._revbranchcache:
             self._revbranchcache = branchmap.revbranchcache(self.unfiltered())
         return self._revbranchcache
 
     def register_changeset(self, rev, changelogrevision):
         self.revbranchcache().setdata(rev, changelogrevision)
 
     def branchtip(self, branch, ignoremissing=False):
         """return the tip node for a given branch
 
         If ignoremissing is True, then this method will not raise an error.
         This is helpful for callers that only expect None for a missing branch
         (e.g. namespace).
 
         """
         try:
             return self.branchmap().branchtip(branch)
         except KeyError:
             if not ignoremissing:
                 raise error.RepoLookupError(_(b"unknown branch '%s'") % branch)
             else:
                 pass
 
     def lookup(self, key):
         node = scmutil.revsymbol(self, key).node()
         if node is None:
             raise error.RepoLookupError(_(b"unknown revision '%s'") % key)
         return node
 
     def lookupbranch(self, key):
         if self.branchmap().hasbranch(key):
             return key
 
         return scmutil.revsymbol(self, key).branch()
 
     def known(self, nodes):
         cl = self.changelog
         get_rev = cl.index.get_rev
         filtered = cl.filteredrevs
         result = []
         for n in nodes:
             r = get_rev(n)
             resp = not (r is None or r in filtered)
             result.append(resp)
         return result
 
     def local(self):
         return self
 
     def publishing(self):
         # it's safe (and desirable) to trust the publish flag unconditionally
         # so that we don't finalize changes shared between users via ssh or nfs
         return self.ui.configbool(b'phases', b'publish', untrusted=True)
 
     def cancopy(self):
         # so statichttprepo's override of local() works
         if not self.local():
             return False
         if not self.publishing():
             return True
         # if publishing we can't copy if there is filtered content
         return not self.filtered(b'visible').changelog.filteredrevs
 
     def shared(self):
         '''the type of shared repository (None if not shared)'''
         if self.sharedpath != self.path:
             return b'store'
         return None
 
     def wjoin(self, f, *insidef):
         return self.vfs.reljoin(self.root, f, *insidef)
 
     def setparents(self, p1, p2=None):
         if p2 is None:
             p2 = self.nullid
         self[None].setparents(p1, p2)
         self._quick_access_changeid_invalidate()
 
     def filectx(self, path, changeid=None, fileid=None, changectx=None):
         """changeid must be a changeset revision, if specified.
         fileid can be a file revision or node."""
         return context.filectx(
             self, path, changeid, fileid, changectx=changectx
         )
 
     def getcwd(self):
         return self.dirstate.getcwd()
 
     def pathto(self, f, cwd=None):
         return self.dirstate.pathto(f, cwd)
 
     def _loadfilter(self, filter):
         if filter not in self._filterpats:
             l = []
             for pat, cmd in self.ui.configitems(filter):
                 if cmd == b'!':
                     continue
                 mf = matchmod.match(self.root, b'', [pat])
                 fn = None
                 params = cmd
                 for name, filterfn in pycompat.iteritems(self._datafilters):
                     if cmd.startswith(name):
                         fn = filterfn
                         params = cmd[len(name) :].lstrip()
                         break
                 if not fn:
                     fn = lambda s, c, **kwargs: procutil.filter(s, c)
                     fn.__name__ = 'commandfilter'
                 # Wrap old filters not supporting keyword arguments
                 if not pycompat.getargspec(fn)[2]:
                     oldfn = fn
                     fn = lambda s, c, oldfn=oldfn, **kwargs: oldfn(s, c)
                     fn.__name__ = 'compat-' + oldfn.__name__
                 l.append((mf, fn, params))
             self._filterpats[filter] = l
         return self._filterpats[filter]
 
     def _filter(self, filterpats, filename, data):
         for mf, fn, cmd in filterpats:
             if mf(filename):
                 self.ui.debug(
                     b"filtering %s through %s\n"
                     % (filename, cmd or pycompat.sysbytes(fn.__name__))
                 )
                 data = fn(data, cmd, ui=self.ui, repo=self, filename=filename)
                 break
 
         return data
 
     @unfilteredpropertycache
     def _encodefilterpats(self):
         return self._loadfilter(b'encode')
 
     @unfilteredpropertycache
     def _decodefilterpats(self):
         return self._loadfilter(b'decode')
 
     def adddatafilter(self, name, filter):
         self._datafilters[name] = filter
 
     def wread(self, filename):
         if self.wvfs.islink(filename):
             data = self.wvfs.readlink(filename)
         else:
             data = self.wvfs.read(filename)
         return self._filter(self._encodefilterpats, filename, data)
 
     def wwrite(self, filename, data, flags, backgroundclose=False, **kwargs):
         """write ``data`` into ``filename`` in the working directory
 
         This returns length of written (maybe decoded) data.
         """
         data = self._filter(self._decodefilterpats, filename, data)
         if b'l' in flags:
             self.wvfs.symlink(data, filename)
         else:
             self.wvfs.write(
                 filename, data, backgroundclose=backgroundclose, **kwargs
             )
             if b'x' in flags:
                 self.wvfs.setflags(filename, False, True)
             else:
                 self.wvfs.setflags(filename, False, False)
         return len(data)
 
     def wwritedata(self, filename, data):
         return self._filter(self._decodefilterpats, filename, data)
 
     def currenttransaction(self):
         """return the current transaction or None if non exists"""
         if self._transref:
             tr = self._transref()
         else:
             tr = None
 
         if tr and tr.running():
             return tr
         return None
 
     def transaction(self, desc, report=None):
         if self.ui.configbool(b'devel', b'all-warnings') or self.ui.configbool(
             b'devel', b'check-locks'
         ):
             if self._currentlock(self._lockref) is None:
                 raise error.ProgrammingError(b'transaction requires locking')
         tr = self.currenttransaction()
         if tr is not None:
             return tr.nest(name=desc)
 
         # abort here if the journal already exists
         if self.svfs.exists(b"journal"):
             raise error.RepoError(
                 _(b"abandoned transaction found"),
                 hint=_(b"run 'hg recover' to clean up transaction"),
             )
 
         idbase = b"%.40f#%f" % (random.random(), time.time())
         ha = hex(hashutil.sha1(idbase).digest())
         txnid = b'TXN:' + ha
         self.hook(b'pretxnopen', throw=True, txnname=desc, txnid=txnid)
 
         self._writejournal(desc)
         renames = [(vfs, x, undoname(x)) for vfs, x in self._journalfiles()]
         if report:
             rp = report
         else:
             rp = self.ui.warn
         vfsmap = {b'plain': self.vfs, b'store': self.svfs}  # root of .hg/
         # we must avoid cyclic reference between repo and transaction.
         reporef = weakref.ref(self)
         # Code to track tag movement
         #
         # Since tags are all handled as file content, it is actually quite hard
         # to track these movement from a code perspective. So we fallback to a
         # tracking at the repository level. One could envision to track changes
         # to the '.hgtags' file through changegroup apply but that fails to
         # cope with case where transaction expose new heads without changegroup
         # being involved (eg: phase movement).
         #
         # For now, We gate the feature behind a flag since this likely comes
         # with performance impacts. The current code run more often than needed
         # and do not use caches as much as it could.  The current focus is on
         # the behavior of the feature so we disable it by default. The flag
         # will be removed when we are happy with the performance impact.
         #
         # Once this feature is no longer experimental move the following
         # documentation to the appropriate help section:
         #
         # The ``HG_TAG_MOVED`` variable will be set if the transaction touched
         # tags (new or changed or deleted tags). In addition the details of
         # these changes are made available in a file at:
         #     ``REPOROOT/.hg/changes/tags.changes``.
         # Make sure you check for HG_TAG_MOVED before reading that file as it
         # might exist from a previous transaction even if no tag were touched
         # in this one. Changes are recorded in a line base format::
         #
         #     <action> <hex-node> <tag-name>\n
         #
         # Actions are defined as follow:
         #   "-R": tag is removed,
         #   "+A": tag is added,
         #   "-M": tag is moved (old value),
         #   "+M": tag is moved (new value),
         tracktags = lambda x: None
         # experimental config: experimental.hook-track-tags
         shouldtracktags = self.ui.configbool(
             b'experimental', b'hook-track-tags'
         )
         if desc != b'strip' and shouldtracktags:
             oldheads = self.changelog.headrevs()
 
             def tracktags(tr2):
                 repo = reporef()
                 assert repo is not None  # help pytype
                 oldfnodes = tagsmod.fnoderevs(repo.ui, repo, oldheads)
                 newheads = repo.changelog.headrevs()
                 newfnodes = tagsmod.fnoderevs(repo.ui, repo, newheads)
                 # notes: we compare lists here.
                 # As we do it only once buiding set would not be cheaper
                 changes = tagsmod.difftags(repo.ui, repo, oldfnodes, newfnodes)
                 if changes:
                     tr2.hookargs[b'tag_moved'] = b'1'
                     with repo.vfs(
                         b'changes/tags.changes', b'w', atomictemp=True
                     ) as changesfile:
                         # note: we do not register the file to the transaction
                         # because we needs it to still exist on the transaction
                         # is close (for txnclose hooks)
                         tagsmod.writediff(changesfile, changes)
 
         def validate(tr2):
             """will run pre-closing hooks"""
             # XXX the transaction API is a bit lacking here so we take a hacky
             # path for now
             #
             # We cannot add this as a "pending" hooks since the 'tr.hookargs'
             # dict is copied before these run. In addition we needs the data
             # available to in memory hooks too.
             #
             # Moreover, we also need to make sure this runs before txnclose
             # hooks and there is no "pending" mechanism that would execute
             # logic only if hooks are about to run.
             #
             # Fixing this limitation of the transaction is also needed to track
             # other families of changes (bookmarks, phases, obsolescence).
             #
             # This will have to be fixed before we remove the experimental
             # gating.
             tracktags(tr2)
             repo = reporef()
             assert repo is not None  # help pytype
 
             singleheadopt = (b'experimental', b'single-head-per-branch')
             singlehead = repo.ui.configbool(*singleheadopt)
             if singlehead:
                 singleheadsub = repo.ui.configsuboptions(*singleheadopt)[1]
                 accountclosed = singleheadsub.get(
                     b"account-closed-heads", False
                 )
                 if singleheadsub.get(b"public-changes-only", False):
                     filtername = b"immutable"
                 else:
                     filtername = b"visible"
                 scmutil.enforcesinglehead(
                     repo, tr2, desc, accountclosed, filtername
                 )
             if hook.hashook(repo.ui, b'pretxnclose-bookmark'):
                 for name, (old, new) in sorted(
                     tr.changes[b'bookmarks'].items()
                 ):
                     args = tr.hookargs.copy()
                     args.update(bookmarks.preparehookargs(name, old, new))
                     repo.hook(
                         b'pretxnclose-bookmark',
                         throw=True,
                         **pycompat.strkwargs(args)
                     )
             if hook.hashook(repo.ui, b'pretxnclose-phase'):
                 cl = repo.unfiltered().changelog
                 for revs, (old, new) in tr.changes[b'phases']:
                     for rev in revs:
                         args = tr.hookargs.copy()
                         node = hex(cl.node(rev))
                         args.update(phases.preparehookargs(node, old, new))
                         repo.hook(
                             b'pretxnclose-phase',
                             throw=True,
                             **pycompat.strkwargs(args)
                         )
 
             repo.hook(
                 b'pretxnclose', throw=True, **pycompat.strkwargs(tr.hookargs)
             )
 
         def releasefn(tr, success):
             repo = reporef()
             if repo is None:
                 # If the repo has been GC'd (and this release function is being
                 # called from transaction.__del__), there's not much we can do,
                 # so just leave the unfinished transaction there and let the
                 # user run `hg recover`.
                 return
             if success:
                 # this should be explicitly invoked here, because
                 # in-memory changes aren't written out at closing
                 # transaction, if tr.addfilegenerator (via
                 # dirstate.write or so) isn't invoked while
                 # transaction running
                 repo.dirstate.write(None)
             else:
                 # discard all changes (including ones already written
                 # out) in this transaction
                 narrowspec.restorebackup(self, b'journal.narrowspec')
                 narrowspec.restorewcbackup(self, b'journal.narrowspec.dirstate')
                 repo.dirstate.restorebackup(None, b'journal.dirstate')
 
                 repo.invalidate(clearfilecache=True)
 
         tr = transaction.transaction(
             rp,
             self.svfs,
             vfsmap,
             b"journal",
             b"undo",
             aftertrans(renames),
             self.store.createmode,
             validator=validate,
             releasefn=releasefn,
             checkambigfiles=_cachedfiles,
             name=desc,
         )
         tr.changes[b'origrepolen'] = len(self)
         tr.changes[b'obsmarkers'] = set()
         tr.changes[b'phases'] = []
         tr.changes[b'bookmarks'] = {}
 
         tr.hookargs[b'txnid'] = txnid
         tr.hookargs[b'txnname'] = desc
         tr.hookargs[b'changes'] = tr.changes
         # note: writing the fncache only during finalize mean that the file is
         # outdated when running hooks. As fncache is used for streaming clone,
         # this is not expected to break anything that happen during the hooks.
         tr.addfinalize(b'flush-fncache', self.store.write)
 
         def txnclosehook(tr2):
             """To be run if transaction is successful, will schedule a hook run"""
             # Don't reference tr2 in hook() so we don't hold a reference.
             # This reduces memory consumption when there are multiple
             # transactions per lock. This can likely go away if issue5045
             # fixes the function accumulation.
             hookargs = tr2.hookargs
 
             def hookfunc(unused_success):
                 repo = reporef()
                 assert repo is not None  # help pytype
 
                 if hook.hashook(repo.ui, b'txnclose-bookmark'):
                     bmchanges = sorted(tr.changes[b'bookmarks'].items())
                     for name, (old, new) in bmchanges:
                         args = tr.hookargs.copy()
                         args.update(bookmarks.preparehookargs(name, old, new))
                         repo.hook(
                             b'txnclose-bookmark',
                             throw=False,
                             **pycompat.strkwargs(args)
                         )
 
                 if hook.hashook(repo.ui, b'txnclose-phase'):
                     cl = repo.unfiltered().changelog
                     phasemv = sorted(
                         tr.changes[b'phases'], key=lambda r: r[0][0]
                     )
                     for revs, (old, new) in phasemv:
                         for rev in revs:
                             args = tr.hookargs.copy()
                             node = hex(cl.node(rev))
                             args.update(phases.preparehookargs(node, old, new))
                             repo.hook(
                                 b'txnclose-phase',
                                 throw=False,
                                 **pycompat.strkwargs(args)
                             )
 
                 repo.hook(
                     b'txnclose', throw=False, **pycompat.strkwargs(hookargs)
                 )
 
             repo = reporef()
             assert repo is not None  # help pytype
             repo._afterlock(hookfunc)
 
         tr.addfinalize(b'txnclose-hook', txnclosehook)
         # Include a leading "-" to make it happen before the transaction summary
         # reports registered via scmutil.registersummarycallback() whose names
         # are 00-txnreport etc. That way, the caches will be warm when the
         # callbacks run.
         tr.addpostclose(b'-warm-cache', self._buildcacheupdater(tr))
 
         def txnaborthook(tr2):
             """To be run if transaction is aborted"""
             repo = reporef()
             assert repo is not None  # help pytype
             repo.hook(
                 b'txnabort', throw=False, **pycompat.strkwargs(tr2.hookargs)
             )
 
         tr.addabort(b'txnabort-hook', txnaborthook)
         # avoid eager cache invalidation. in-memory data should be identical
         # to stored data if transaction has no error.
         tr.addpostclose(b'refresh-filecachestats', self._refreshfilecachestats)
         self._transref = weakref.ref(tr)
         scmutil.registersummarycallback(self, tr, desc)
         return tr
 
     def _journalfiles(self):
         return (
             (self.svfs, b'journal'),
             (self.svfs, b'journal.narrowspec'),
             (self.vfs, b'journal.narrowspec.dirstate'),
             (self.vfs, b'journal.dirstate'),
             (self.vfs, b'journal.branch'),
             (self.vfs, b'journal.desc'),
             (bookmarks.bookmarksvfs(self), b'journal.bookmarks'),
             (self.svfs, b'journal.phaseroots'),
         )
 
     def undofiles(self):
         return [(vfs, undoname(x)) for vfs, x in self._journalfiles()]
 
     @unfilteredmethod
     def _writejournal(self, desc):
         self.dirstate.savebackup(None, b'journal.dirstate')
         narrowspec.savewcbackup(self, b'journal.narrowspec.dirstate')
         narrowspec.savebackup(self, b'journal.narrowspec')
         self.vfs.write(
             b"journal.branch", encoding.fromlocal(self.dirstate.branch())
         )
         self.vfs.write(b"journal.desc", b"%d\n%s\n" % (len(self), desc))
         bookmarksvfs = bookmarks.bookmarksvfs(self)
         bookmarksvfs.write(
             b"journal.bookmarks", bookmarksvfs.tryread(b"bookmarks")
         )
         self.svfs.write(b"journal.phaseroots", self.svfs.tryread(b"phaseroots"))
 
     def recover(self):
         with self.lock():
             if self.svfs.exists(b"journal"):
                 self.ui.status(_(b"rolling back interrupted transaction\n"))
                 vfsmap = {
                     b'': self.svfs,
                     b'plain': self.vfs,
                 }
                 transaction.rollback(
                     self.svfs,
                     vfsmap,
                     b"journal",
                     self.ui.warn,
                     checkambigfiles=_cachedfiles,
                 )
                 self.invalidate()
                 return True
             else:
                 self.ui.warn(_(b"no interrupted transaction available\n"))
                 return False
 
     def rollback(self, dryrun=False, force=False):
         wlock = lock = dsguard = None
         try:
             wlock = self.wlock()
             lock = self.lock()
             if self.svfs.exists(b"undo"):
                 dsguard = dirstateguard.dirstateguard(self, b'rollback')
 
                 return self._rollback(dryrun, force, dsguard)
             else:
                 self.ui.warn(_(b"no rollback information available\n"))
                 return 1
         finally:
             release(dsguard, lock, wlock)
 
     @unfilteredmethod  # Until we get smarter cache management
     def _rollback(self, dryrun, force, dsguard):
         ui = self.ui
         try:
             args = self.vfs.read(b'undo.desc').splitlines()
             (oldlen, desc, detail) = (int(args[0]), args[1], None)
             if len(args) >= 3:
                 detail = args[2]
             oldtip = oldlen - 1
 
             if detail and ui.verbose:
                 msg = _(
                     b'repository tip rolled back to revision %d'
                     b' (undo %s: %s)\n'
                 ) % (oldtip, desc, detail)
             else:
                 msg = _(
                     b'repository tip rolled back to revision %d (undo %s)\n'
                 ) % (oldtip, desc)
         except IOError:
             msg = _(b'rolling back unknown transaction\n')
             desc = None
 
         if not force and self[b'.'] != self[b'tip'] and desc == b'commit':
             raise error.Abort(
                 _(
                     b'rollback of last commit while not checked out '
                     b'may lose data'
                 ),
                 hint=_(b'use -f to force'),
             )
 
         ui.status(msg)
         if dryrun:
             return 0
 
         parents = self.dirstate.parents()
         self.destroying()
         vfsmap = {b'plain': self.vfs, b'': self.svfs}
         transaction.rollback(
             self.svfs, vfsmap, b'undo', ui.warn, checkambigfiles=_cachedfiles
         )
         bookmarksvfs = bookmarks.bookmarksvfs(self)
         if bookmarksvfs.exists(b'undo.bookmarks'):
             bookmarksvfs.rename(
                 b'undo.bookmarks', b'bookmarks', checkambig=True
             )
         if self.svfs.exists(b'undo.phaseroots'):
             self.svfs.rename(b'undo.phaseroots', b'phaseroots', checkambig=True)
         self.invalidate()
 
         has_node = self.changelog.index.has_node
         parentgone = any(not has_node(p) for p in parents)
         if parentgone:
             # prevent dirstateguard from overwriting already restored one
             dsguard.close()
 
             narrowspec.restorebackup(self, b'undo.narrowspec')
             narrowspec.restorewcbackup(self, b'undo.narrowspec.dirstate')
             self.dirstate.restorebackup(None, b'undo.dirstate')
             try:
                 branch = self.vfs.read(b'undo.branch')
                 self.dirstate.setbranch(encoding.tolocal(branch))
             except IOError:
                 ui.warn(
                     _(
                         b'named branch could not be reset: '
                         b'current branch is still \'%s\'\n'
                     )
                     % self.dirstate.branch()
                 )
 
             parents = tuple([p.rev() for p in self[None].parents()])
             if len(parents) > 1:
                 ui.status(
                     _(
                         b'working directory now based on '
                         b'revisions %d and %d\n'
                     )
                     % parents
                 )
             else:
                 ui.status(
                     _(b'working directory now based on revision %d\n') % parents
                 )
             mergestatemod.mergestate.clean(self)
 
         # TODO: if we know which new heads may result from this rollback, pass
         # them to destroy(), which will prevent the branchhead cache from being
         # invalidated.
         self.destroyed()
         return 0
 
     def _buildcacheupdater(self, newtransaction):
         """called during transaction to build the callback updating cache
 
         Lives on the repository to help extension who might want to augment
         this logic. For this purpose, the created transaction is passed to the
         method.
         """
         # we must avoid cyclic reference between repo and transaction.
         reporef = weakref.ref(self)
 
         def updater(tr):
             repo = reporef()
             assert repo is not None  # help pytype
             repo.updatecaches(tr)
 
         return updater
 
     @unfilteredmethod
     def updatecaches(self, tr=None, full=False):
         """warm appropriate caches
 
         If this function is called after a transaction closed. The transaction
         will be available in the 'tr' argument. This can be used to selectively
         update caches relevant to the changes in that transaction.
 
         If 'full' is set, make sure all caches the function knows about have
         up-to-date data. Even the ones usually loaded more lazily.
         """
         if tr is not None and tr.hookargs.get(b'source') == b'strip':
             # During strip, many caches are invalid but
             # later call to `destroyed` will refresh them.
             return
 
         if tr is None or tr.changes[b'origrepolen'] < len(self):
             # accessing the 'served' branchmap should refresh all the others,
             self.ui.debug(b'updating the branch cache\n')
             self.filtered(b'served').branchmap()
             self.filtered(b'served.hidden').branchmap()
 
         if full:
             unfi = self.unfiltered()
 
             self.changelog.update_caches(transaction=tr)
             self.manifestlog.update_caches(transaction=tr)
 
             rbc = unfi.revbranchcache()
             for r in unfi.changelog:
                 rbc.branchinfo(r)
             rbc.write()
 
             # ensure the working copy parents are in the manifestfulltextcache
             for ctx in self[b'.'].parents():
                 ctx.manifest()  # accessing the manifest is enough
 
             # accessing fnode cache warms the cache
             tagsmod.fnoderevs(self.ui, unfi, unfi.changelog.revs())
             # accessing tags warm the cache
             self.tags()
             self.filtered(b'served').tags()
 
             # The `full` arg is documented as updating even the lazily-loaded
             # caches immediately, so we're forcing a write to cause these caches
             # to be warmed up even if they haven't explicitly been requested
             # yet (if they've never been used by hg, they won't ever have been
             # written, even if they're a subset of another kind of cache that
             # *has* been used).
             for filt in repoview.filtertable.keys():
                 filtered = self.filtered(filt)
                 filtered.branchmap().write(filtered)
 
     def invalidatecaches(self):
 
         if '_tagscache' in vars(self):
             # can't use delattr on proxy
             del self.__dict__['_tagscache']
 
         self._branchcaches.clear()
         self.invalidatevolatilesets()
         self._sparsesignaturecache.clear()
 
     def invalidatevolatilesets(self):
         self.filteredrevcache.clear()
         obsolete.clearobscaches(self)
         self._quick_access_changeid_invalidate()
 
     def invalidatedirstate(self):
         """Invalidates the dirstate, causing the next call to dirstate
         to check if it was modified since the last time it was read,
         rereading it if it has.
 
         This is different to dirstate.invalidate() that it doesn't always
         rereads the dirstate. Use dirstate.invalidate() if you want to
         explicitly read the dirstate again (i.e. restoring it to a previous
         known good state)."""
         if hasunfilteredcache(self, 'dirstate'):
             for k in self.dirstate._filecache:
                 try:
                     delattr(self.dirstate, k)
                 except AttributeError:
                     pass
             delattr(self.unfiltered(), 'dirstate')
 
     def invalidate(self, clearfilecache=False):
         """Invalidates both store and non-store parts other than dirstate
 
         If a transaction is running, invalidation of store is omitted,
         because discarding in-memory changes might cause inconsistency
         (e.g. incomplete fncache causes unintentional failure, but
         redundant one doesn't).
         """
         unfiltered = self.unfiltered()  # all file caches are stored unfiltered
         for k in list(self._filecache.keys()):
             # dirstate is invalidated separately in invalidatedirstate()
             if k == b'dirstate':
                 continue
             if (
                 k == b'changelog'
                 and self.currenttransaction()
                 and self.changelog._delayed
             ):
                 # The changelog object may store unwritten revisions. We don't
                 # want to lose them.
                 # TODO: Solve the problem instead of working around it.
                 continue
 
             if clearfilecache:
                 del self._filecache[k]
             try:
                 delattr(unfiltered, k)
             except AttributeError:
                 pass
         self.invalidatecaches()
         if not self.currenttransaction():
             # TODO: Changing contents of store outside transaction
             # causes inconsistency. We should make in-memory store
             # changes detectable, and abort if changed.
             self.store.invalidatecaches()
 
     def invalidateall(self):
         """Fully invalidates both store and non-store parts, causing the
         subsequent operation to reread any outside changes."""
         # extension should hook this to invalidate its caches
         self.invalidate()
         self.invalidatedirstate()
 
     @unfilteredmethod
     def _refreshfilecachestats(self, tr):
         """Reload stats of cached files so that they are flagged as valid"""
         for k, ce in self._filecache.items():
             k = pycompat.sysstr(k)
             if k == 'dirstate' or k not in self.__dict__:
                 continue
             ce.refresh()
 
     def _lock(
         self,
         vfs,
         lockname,
         wait,
         releasefn,
         acquirefn,
         desc,
     ):
         timeout = 0
         warntimeout = 0
         if wait:
             timeout = self.ui.configint(b"ui", b"timeout")
             warntimeout = self.ui.configint(b"ui", b"timeout.warn")
         # internal config: ui.signal-safe-lock
         signalsafe = self.ui.configbool(b'ui', b'signal-safe-lock')
 
         l = lockmod.trylock(
             self.ui,
             vfs,
             lockname,
             timeout,
             warntimeout,
             releasefn=releasefn,
             acquirefn=acquirefn,
             desc=desc,
             signalsafe=signalsafe,
         )
         return l
 
     def _afterlock(self, callback):
         """add a callback to be run when the repository is fully unlocked
 
         The callback will be executed when the outermost lock is released
         (with wlock being higher level than 'lock')."""
         for ref in (self._wlockref, self._lockref):
             l = ref and ref()
             if l and l.held:
                 l.postrelease.append(callback)
                 break
         else:  # no lock have been found.
             callback(True)
 
     def lock(self, wait=True):
         """Lock the repository store (.hg/store) and return a weak reference
         to the lock. Use this before modifying the store (e.g. committing or
         stripping). If you are opening a transaction, get a lock as well.)
 
         If both 'lock' and 'wlock' must be acquired, ensure you always acquires
         'wlock' first to avoid a dead-lock hazard."""
         l = self._currentlock(self._lockref)
         if l is not None:
             l.lock()
             return l
 
         l = self._lock(
             vfs=self.svfs,
             lockname=b"lock",
             wait=wait,
             releasefn=None,
             acquirefn=self.invalidate,
             desc=_(b'repository %s') % self.origroot,
         )
         self._lockref = weakref.ref(l)
         return l
 
     def wlock(self, wait=True):
         """Lock the non-store parts of the repository (everything under
         .hg except .hg/store) and return a weak reference to the lock.
 
         Use this before modifying files in .hg.
 
         If both 'lock' and 'wlock' must be acquired, ensure you always acquires
         'wlock' first to avoid a dead-lock hazard."""
         l = self._wlockref() if self._wlockref else None
         if l is not None and l.held:
             l.lock()
             return l
 
         # We do not need to check for non-waiting lock acquisition.  Such
         # acquisition would not cause dead-lock as they would just fail.
         if wait and (
             self.ui.configbool(b'devel', b'all-warnings')
             or self.ui.configbool(b'devel', b'check-locks')
         ):
             if self._currentlock(self._lockref) is not None:
                 self.ui.develwarn(b'"wlock" acquired after "lock"')
 
         def unlock():
             if self.dirstate.pendingparentchange():
                 self.dirstate.invalidate()
             else:
                 self.dirstate.write(None)
 
             self._filecache[b'dirstate'].refresh()
 
         l = self._lock(
             self.vfs,
             b"wlock",
             wait,
             unlock,
             self.invalidatedirstate,
             _(b'working directory of %s') % self.origroot,
         )
         self._wlockref = weakref.ref(l)
         return l
 
     def _currentlock(self, lockref):
         """Returns the lock if it's held, or None if it's not."""
         if lockref is None:
             return None
         l = lockref()
         if l is None or not l.held:
             return None
         return l
 
     def currentwlock(self):
         """Returns the wlock if it's held, or None if it's not."""
         return self._currentlock(self._wlockref)
 
     def checkcommitpatterns(self, wctx, match, status, fail):
         """check for commit arguments that aren't committable"""
         if match.isexact() or match.prefix():
             matched = set(status.modified + status.added + status.removed)
 
             for f in match.files():
                 f = self.dirstate.normalize(f)
                 if f == b'.' or f in matched or f in wctx.substate:
                     continue
                 if f in status.deleted:
                     fail(f, _(b'file not found!'))
                 # Is it a directory that exists or used to exist?
                 if self.wvfs.isdir(f) or wctx.p1().hasdir(f):
                     d = f + b'/'
                     for mf in matched:
                         if mf.startswith(d):
                             break
                     else:
                         fail(f, _(b"no match under directory!"))
                 elif f not in self.dirstate:
                     fail(f, _(b"file not tracked!"))
 
     @unfilteredmethod
     def commit(
         self,
         text=b"",
         user=None,
         date=None,
         match=None,
         force=False,
         editor=None,
         extra=None,
     ):
         """Add a new revision to current repository.
 
         Revision information is gathered from the working directory,
         match can be used to filter the committed files. If editor is
         supplied, it is called to get a commit message.
         """
         if extra is None:
             extra = {}
 
         def fail(f, msg):
             raise error.InputError(b'%s: %s' % (f, msg))
 
         if not match:
             match = matchmod.always()
 
         if not force:
             match.bad = fail
 
         # lock() for recent changelog (see issue4368)
         with self.wlock(), self.lock():
             wctx = self[None]
             merge = len(wctx.parents()) > 1
 
             if not force and merge and not match.always():
                 raise error.Abort(
                     _(
                         b'cannot partially commit a merge '
                         b'(do not specify files or patterns)'
                     )
                 )
 
             status = self.status(match=match, clean=force)
             if force:
                 status.modified.extend(
                     status.clean
                 )  # mq may commit clean files
 
             # check subrepos
             subs, commitsubs, newstate = subrepoutil.precommit(
                 self.ui, wctx, status, match, force=force
             )
 
             # make sure all explicit patterns are matched
             if not force:
                 self.checkcommitpatterns(wctx, match, status, fail)
 
             cctx = context.workingcommitctx(
                 self, status, text, user, date, extra
             )
 
             ms = mergestatemod.mergestate.read(self)
             mergeutil.checkunresolved(ms)
 
             # internal config: ui.allowemptycommit
             if cctx.isempty() and not self.ui.configbool(
                 b'ui', b'allowemptycommit'
             ):
                 self.ui.debug(b'nothing to commit, clearing merge state\n')
                 ms.reset()
                 return None
 
             if merge and cctx.deleted():
                 raise error.Abort(_(b"cannot commit merge with missing files"))
 
             if editor:
                 cctx._text = editor(self, cctx, subs)
             edited = text != cctx._text
 
             # Save commit message in case this transaction gets rolled back
             # (e.g. by a pretxncommit hook).  Leave the content alone on
             # the assumption that the user will use the same editor again.
             msgfn = self.savecommitmessage(cctx._text)
 
             # commit subs and write new state
             if subs:
                 uipathfn = scmutil.getuipathfn(self)
                 for s in sorted(commitsubs):
                     sub = wctx.sub(s)
                     self.ui.status(
                         _(b'committing subrepository %s\n')
                         % uipathfn(subrepoutil.subrelpath(sub))
                     )
                     sr = sub.commit(cctx._text, user, date)
                     newstate[s] = (newstate[s][0], sr)
                 subrepoutil.writestate(self, newstate)
 
             p1, p2 = self.dirstate.parents()
             hookp1, hookp2 = hex(p1), (p2 != self.nullid and hex(p2) or b'')
             try:
                 self.hook(
                     b"precommit", throw=True, parent1=hookp1, parent2=hookp2
                 )
                 with self.transaction(b'commit'):
                     ret = self.commitctx(cctx, True)
                     # update bookmarks, dirstate and mergestate
                     bookmarks.update(self, [p1, p2], ret)
                     cctx.markcommitted(ret)
                     ms.reset()
             except:  # re-raises
                 if edited:
                     self.ui.write(
                         _(b'note: commit message saved in %s\n') % msgfn
                     )
                     self.ui.write(
                         _(
                             b"note: use 'hg commit --logfile "
                             b".hg/last-message.txt --edit' to reuse it\n"
                         )
                     )
                 raise
 
         def commithook(unused_success):
             # hack for command that use a temporary commit (eg: histedit)
             # temporary commit got stripped before hook release
             if self.changelog.hasnode(ret):
                 self.hook(
                     b"commit", node=hex(ret), parent1=hookp1, parent2=hookp2
                 )
 
         self._afterlock(commithook)
         return ret
 
     @unfilteredmethod
     def commitctx(self, ctx, error=False, origctx=None):
         return commit.commitctx(self, ctx, error=error, origctx=origctx)
 
     @unfilteredmethod
     def destroying(self):
         """Inform the repository that nodes are about to be destroyed.
         Intended for use by strip and rollback, so there's a common
         place for anything that has to be done before destroying history.
 
         This is mostly useful for saving state that is in memory and waiting
         to be flushed when the current lock is released. Because a call to
         destroyed is imminent, the repo will be invalidated causing those
         changes to stay in memory (waiting for the next unlock), or vanish
         completely.
         """
         # When using the same lock to commit and strip, the phasecache is left
         # dirty after committing. Then when we strip, the repo is invalidated,
         # causing those changes to disappear.
         if '_phasecache' in vars(self):
             self._phasecache.write()
 
     @unfilteredmethod
     def destroyed(self):
         """Inform the repository that nodes have been destroyed.
         Intended for use by strip and rollback, so there's a common
         place for anything that has to be done after destroying history.
         """
         # When one tries to:
         # 1) destroy nodes thus calling this method (e.g. strip)
         # 2) use phasecache somewhere (e.g. commit)
         #
         # then 2) will fail because the phasecache contains nodes that were
         # removed. We can either remove phasecache from the filecache,
         # causing it to reload next time it is accessed, or simply filter
         # the removed nodes now and write the updated cache.
         self._phasecache.filterunknown(self)
         self._phasecache.write()
 
         # refresh all repository caches
         self.updatecaches()
 
         # Ensure the persistent tag cache is updated.  Doing it now
         # means that the tag cache only has to worry about destroyed
         # heads immediately after a strip/rollback.  That in turn
         # guarantees that "cachetip == currenttip" (comparing both rev
         # and node) always means no nodes have been added or destroyed.
 
         # XXX this is suboptimal when qrefresh'ing: we strip the current
         # head, refresh the tag cache, then immediately add a new head.
         # But I think doing it this way is necessary for the "instant
         # tag cache retrieval" case to work.
         self.invalidate()
 
     def status(
         self,
         node1=b'.',
         node2=None,
         match=None,
         ignored=False,
         clean=False,
         unknown=False,
         listsubrepos=False,
     ):
         '''a convenience method that calls node1.status(node2)'''
         return self[node1].status(
             node2, match, ignored, clean, unknown, listsubrepos
         )
 
     def addpostdsstatus(self, ps):
         """Add a callback to run within the wlock, at the point at which status
         fixups happen.
 
         On status completion, callback(wctx, status) will be called with the
         wlock held, unless the dirstate has changed from underneath or the wlock
         couldn't be grabbed.
 
         Callbacks should not capture and use a cached copy of the dirstate --
         it might change in the meanwhile. Instead, they should access the
         dirstate via wctx.repo().dirstate.
 
         This list is emptied out after each status run -- extensions should
         make sure it adds to this list each time dirstate.status is called.
         Extensions should also make sure they don't call this for statuses
         that don't involve the dirstate.
         """
 
         # The list is located here for uniqueness reasons -- it is actually
         # managed by the workingctx, but that isn't unique per-repo.
         self._postdsstatus.append(ps)
 
     def postdsstatus(self):
         """Used by workingctx to get the list of post-dirstate-status hooks."""
         return self._postdsstatus
 
     def clearpostdsstatus(self):
         """Used by workingctx to clear post-dirstate-status hooks."""
         del self._postdsstatus[:]
 
     def heads(self, start=None):
         if start is None:
             cl = self.changelog
             headrevs = reversed(cl.headrevs())
             return [cl.node(rev) for rev in headrevs]
 
         heads = self.changelog.heads(start)
         # sort the output in rev descending order
         return sorted(heads, key=self.changelog.rev, reverse=True)
 
     def branchheads(self, branch=None, start=None, closed=False):
         """return a (possibly filtered) list of heads for the given branch
 
         Heads are returned in topological order, from newest to oldest.
         If branch is None, use the dirstate branch.
         If start is not None, return only heads reachable from start.
         If closed is True, return heads that are marked as closed as well.
         """
         if branch is None:
             branch = self[None].branch()
         branches = self.branchmap()
         if not branches.hasbranch(branch):
             return []
         # the cache returns heads ordered lowest to highest
         bheads = list(reversed(branches.branchheads(branch, closed=closed)))
         if start is not None:
             # filter out the heads that cannot be reached from startrev
             fbheads = set(self.changelog.nodesbetween([start], bheads)[2])
             bheads = [h for h in bheads if h in fbheads]
         return bheads
 
     def branches(self, nodes):
         if not nodes:
             nodes = [self.changelog.tip()]
         b = []
         for n in nodes:
             t = n
             while True:
                 p = self.changelog.parents(n)
                 if p[1] != self.nullid or p[0] == self.nullid:
                     b.append((t, n, p[0], p[1]))
                     break
                 n = p[0]
         return b
 
     def between(self, pairs):
         r = []
 
         for top, bottom in pairs:
             n, l, i = top, [], 0
             f = 1
 
             while n != bottom and n != self.nullid:
                 p = self.changelog.parents(n)[0]
                 if i == f:
                     l.append(n)
                     f = f * 2
                 n = p
                 i += 1
 
             r.append(l)
 
         return r
 
     def checkpush(self, pushop):
         """Extensions can override this function if additional checks have
         to be performed before pushing, or call it if they override push
         command.
         """
 
     @unfilteredpropertycache
     def prepushoutgoinghooks(self):
         """Return util.hooks consists of a pushop with repo, remote, outgoing
         methods, which are called before pushing changesets.
         """
         return util.hooks()
 
     def pushkey(self, namespace, key, old, new):
         try:
             tr = self.currenttransaction()
             hookargs = {}
             if tr is not None:
                 hookargs.update(tr.hookargs)
             hookargs = pycompat.strkwargs(hookargs)
             hookargs['namespace'] = namespace
             hookargs['key'] = key
             hookargs['old'] = old
             hookargs['new'] = new
             self.hook(b'prepushkey', throw=True, **hookargs)
         except error.HookAbort as exc:
             self.ui.write_err(_(b"pushkey-abort: %s\n") % exc)
             if exc.hint:
                 self.ui.write_err(_(b"(%s)\n") % exc.hint)
             return False
         self.ui.debug(b'pushing key for "%s:%s"\n' % (namespace, key))
         ret = pushkey.push(self, namespace, key, old, new)
 
         def runhook(unused_success):
             self.hook(
                 b'pushkey',
                 namespace=namespace,
                 key=key,
                 old=old,
                 new=new,
                 ret=ret,
             )
 
         self._afterlock(runhook)
         return ret
 
     def listkeys(self, namespace):
         self.hook(b'prelistkeys', throw=True, namespace=namespace)
         self.ui.debug(b'listing keys for "%s"\n' % namespace)
         values = pushkey.list(self, namespace)
         self.hook(b'listkeys', namespace=namespace, values=values)
         return values
 
     def debugwireargs(self, one, two, three=None, four=None, five=None):
         '''used to test argument passing over the wire'''
         return b"%s %s %s %s %s" % (
             one,
             two,
             pycompat.bytestr(three),
             pycompat.bytestr(four),
             pycompat.bytestr(five),
         )
 
     def savecommitmessage(self, text):
         fp = self.vfs(b'last-message.txt', b'wb')
         try:
             fp.write(text)
         finally:
             fp.close()
         return self.pathto(fp.name[len(self.root) + 1 :])
 
     def register_wanted_sidedata(self, category):
         if requirementsmod.REVLOGV2_REQUIREMENT not in self.requirements:
             # Only revlogv2 repos can want sidedata.
             return
         self._wanted_sidedata.add(pycompat.bytestr(category))
 
-    def register_sidedata_computer(self, kind, category, keys, computer):
+    def register_sidedata_computer(self, kind, category, keys, computer, flags):
         if kind not in revlogconst.ALL_KINDS:
-            msg = _(b"unexpected revlog kind %r.")
+            msg = _(b"unexpected revlog kind '%s'.")
             raise error.ProgrammingError(msg % kind)
         category = pycompat.bytestr(category)
         if category in self._sidedata_computers.get(kind, []):
             msg = _(
                 b"cannot register a sidedata computer twice for category '%s'."
             )
             raise error.ProgrammingError(msg % category)
         self._sidedata_computers.setdefault(kind, {})
-        self._sidedata_computers[kind][category] = (keys, computer)
+        self._sidedata_computers[kind][category] = (keys, computer, flags)
 
 
 # used to avoid circular references so destructors work
 def aftertrans(files):
     renamefiles = [tuple(t) for t in files]
 
     def a():
         for vfs, src, dest in renamefiles:
             # if src and dest refer to a same file, vfs.rename is a no-op,
             # leaving both src and dest on disk. delete dest to make sure
             # the rename couldn't be such a no-op.
             vfs.tryunlink(dest)
             try:
                 vfs.rename(src, dest)
             except OSError:  # journal file does not yet exist
                 pass
 
     return a
 
 
 def undoname(fn):
     base, name = os.path.split(fn)
     assert name.startswith(b'journal')
     return os.path.join(base, name.replace(b'journal', b'undo', 1))
 
 
 def instance(ui, path, create, intents=None, createopts=None):
     localpath = urlutil.urllocalpath(path)
     if create:
         createrepository(ui, localpath, createopts=createopts)
 
     return makelocalrepository(ui, localpath, intents=intents)
 
 
 def islocal(path):
     return True
 
 
 def defaultcreateopts(ui, createopts=None):
     """Populate the default creation options for a repository.
 
     A dictionary of explicitly requested creation options can be passed
     in. Missing keys will be populated.
     """
     createopts = dict(createopts or {})
 
     if b'backend' not in createopts:
         # experimental config: storage.new-repo-backend
         createopts[b'backend'] = ui.config(b'storage', b'new-repo-backend')
 
     return createopts
 
 
 def newreporequirements(ui, createopts):
     """Determine the set of requirements for a new local repository.
 
     Extensions can wrap this function to specify custom requirements for
     new repositories.
     """
     # If the repo is being created from a shared repository, we copy
     # its requirements.
     if b'sharedrepo' in createopts:
         requirements = set(createopts[b'sharedrepo'].requirements)
         if createopts.get(b'sharedrelative'):
             requirements.add(requirementsmod.RELATIVE_SHARED_REQUIREMENT)
         else:
             requirements.add(requirementsmod.SHARED_REQUIREMENT)
 
         return requirements
 
     if b'backend' not in createopts:
         raise error.ProgrammingError(
             b'backend key not present in createopts; '
             b'was defaultcreateopts() called?'
         )
 
     if createopts[b'backend'] != b'revlogv1':
         raise error.Abort(
             _(
                 b'unable to determine repository requirements for '
                 b'storage backend: %s'
             )
             % createopts[b'backend']
         )
 
     requirements = {requirementsmod.REVLOGV1_REQUIREMENT}
     if ui.configbool(b'format', b'usestore'):
         requirements.add(requirementsmod.STORE_REQUIREMENT)
         if ui.configbool(b'format', b'usefncache'):
             requirements.add(requirementsmod.FNCACHE_REQUIREMENT)
             if ui.configbool(b'format', b'dotencode'):
                 requirements.add(requirementsmod.DOTENCODE_REQUIREMENT)
 
     compengines = ui.configlist(b'format', b'revlog-compression')
     for compengine in compengines:
         if compengine in util.compengines:
             engine = util.compengines[compengine]
             if engine.available() and engine.revlogheader():
                 break
     else:
         raise error.Abort(
             _(
                 b'compression engines %s defined by '
                 b'format.revlog-compression not available'
             )
             % b', '.join(b'"%s"' % e for e in compengines),
             hint=_(
                 b'run "hg debuginstall" to list available '
                 b'compression engines'
             ),
         )
 
     # zlib is the historical default and doesn't need an explicit requirement.
     if compengine == b'zstd':
         requirements.add(b'revlog-compression-zstd')
     elif compengine != b'zlib':
         requirements.add(b'exp-compression-%s' % compengine)
 
     if scmutil.gdinitconfig(ui):
         requirements.add(requirementsmod.GENERALDELTA_REQUIREMENT)
         if ui.configbool(b'format', b'sparse-revlog'):
             requirements.add(requirementsmod.SPARSEREVLOG_REQUIREMENT)
 
     # experimental config: format.exp-use-side-data
     if ui.configbool(b'format', b'exp-use-side-data'):
         requirements.discard(requirementsmod.REVLOGV1_REQUIREMENT)
         requirements.add(requirementsmod.REVLOGV2_REQUIREMENT)
         requirements.add(requirementsmod.SIDEDATA_REQUIREMENT)
     # experimental config: format.exp-use-copies-side-data-changeset
     if ui.configbool(b'format', b'exp-use-copies-side-data-changeset'):
         requirements.discard(requirementsmod.REVLOGV1_REQUIREMENT)
         requirements.add(requirementsmod.REVLOGV2_REQUIREMENT)
         requirements.add(requirementsmod.SIDEDATA_REQUIREMENT)
         requirements.add(requirementsmod.COPIESSDC_REQUIREMENT)
     if ui.configbool(b'experimental', b'treemanifest'):
         requirements.add(requirementsmod.TREEMANIFEST_REQUIREMENT)
 
     revlogv2 = ui.config(b'experimental', b'revlogv2')
     if revlogv2 == b'enable-unstable-format-and-corrupt-my-data':
         requirements.discard(requirementsmod.REVLOGV1_REQUIREMENT)
         # generaldelta is implied by revlogv2.
         requirements.discard(requirementsmod.GENERALDELTA_REQUIREMENT)
         requirements.add(requirementsmod.REVLOGV2_REQUIREMENT)
     # experimental config: format.internal-phase
     if ui.configbool(b'format', b'internal-phase'):
         requirements.add(requirementsmod.INTERNAL_PHASE_REQUIREMENT)
 
     if createopts.get(b'narrowfiles'):
         requirements.add(requirementsmod.NARROW_REQUIREMENT)
 
     if createopts.get(b'lfs'):
         requirements.add(b'lfs')
 
     if ui.configbool(b'format', b'bookmarks-in-store'):
         requirements.add(bookmarks.BOOKMARKS_IN_STORE_REQUIREMENT)
 
     if ui.configbool(b'format', b'use-persistent-nodemap'):
         requirements.add(requirementsmod.NODEMAP_REQUIREMENT)
 
     # if share-safe is enabled, let's create the new repository with the new
     # requirement
     if ui.configbool(b'format', b'use-share-safe'):
         requirements.add(requirementsmod.SHARESAFE_REQUIREMENT)
 
     return requirements
 
 
 def checkrequirementscompat(ui, requirements):
     """Checks compatibility of repository requirements enabled and disabled.
 
     Returns a set of requirements which needs to be dropped because dependend
     requirements are not enabled. Also warns users about it"""
 
     dropped = set()
 
     if requirementsmod.STORE_REQUIREMENT not in requirements:
         if bookmarks.BOOKMARKS_IN_STORE_REQUIREMENT in requirements:
             ui.warn(
                 _(
                     b'ignoring enabled \'format.bookmarks-in-store\' config '
                     b'beacuse it is incompatible with disabled '
                     b'\'format.usestore\' config\n'
                 )
             )
             dropped.add(bookmarks.BOOKMARKS_IN_STORE_REQUIREMENT)
 
         if (
             requirementsmod.SHARED_REQUIREMENT in requirements
             or requirementsmod.RELATIVE_SHARED_REQUIREMENT in requirements
         ):
             raise error.Abort(
                 _(
                     b"cannot create shared repository as source was created"
                     b" with 'format.usestore' config disabled"
                 )
             )
 
         if requirementsmod.SHARESAFE_REQUIREMENT in requirements:
             ui.warn(
                 _(
                     b"ignoring enabled 'format.use-share-safe' config because "
                     b"it is incompatible with disabled 'format.usestore'"
                     b" config\n"
                 )
             )
             dropped.add(requirementsmod.SHARESAFE_REQUIREMENT)
 
     return dropped
 
 
 def filterknowncreateopts(ui, createopts):
     """Filters a dict of repo creation options against options that are known.
 
     Receives a dict of repo creation options and returns a dict of those
     options that we don't know how to handle.
 
     This function is called as part of repository creation. If the
     returned dict contains any items, repository creation will not
     be allowed, as it means there was a request to create a repository
     with options not recognized by loaded code.
 
     Extensions can wrap this function to filter out creation options
     they know how to handle.
     """
     known = {
         b'backend',
         b'lfs',
         b'narrowfiles',
         b'sharedrepo',
         b'sharedrelative',
         b'shareditems',
         b'shallowfilestore',
     }
 
     return {k: v for k, v in createopts.items() if k not in known}
 
 
 def createrepository(ui, path, createopts=None):
     """Create a new repository in a vfs.
 
     ``path`` path to the new repo's working directory.
     ``createopts`` options for the new repository.
 
     The following keys for ``createopts`` are recognized:
 
     backend
        The storage backend to use.
     lfs
        Repository will be created with ``lfs`` requirement. The lfs extension
        will automatically be loaded when the repository is accessed.
     narrowfiles
        Set up repository to support narrow file storage.
     sharedrepo
        Repository object from which storage should be shared.
     sharedrelative
        Boolean indicating if the path to the shared repo should be
        stored as relative. By default, the pointer to the "parent" repo
        is stored as an absolute path.
     shareditems
        Set of items to share to the new repository (in addition to storage).
     shallowfilestore
        Indicates that storage for files should be shallow (not all ancestor
        revisions are known).
     """
     createopts = defaultcreateopts(ui, createopts=createopts)
 
     unknownopts = filterknowncreateopts(ui, createopts)
 
     if not isinstance(unknownopts, dict):
         raise error.ProgrammingError(
             b'filterknowncreateopts() did not return a dict'
         )
 
     if unknownopts:
         raise error.Abort(
             _(
                 b'unable to create repository because of unknown '
                 b'creation option: %s'
             )
             % b', '.join(sorted(unknownopts)),
             hint=_(b'is a required extension not loaded?'),
         )
 
     requirements = newreporequirements(ui, createopts=createopts)
     requirements -= checkrequirementscompat(ui, requirements)
 
     wdirvfs = vfsmod.vfs(path, expandpath=True, realpath=True)
 
     hgvfs = vfsmod.vfs(wdirvfs.join(b'.hg'))
     if hgvfs.exists():
         raise error.RepoError(_(b'repository %s already exists') % path)
 
     if b'sharedrepo' in createopts:
         sharedpath = createopts[b'sharedrepo'].sharedpath
 
         if createopts.get(b'sharedrelative'):
             try:
                 sharedpath = os.path.relpath(sharedpath, hgvfs.base)
                 sharedpath = util.pconvert(sharedpath)
             except (IOError, ValueError) as e:
                 # ValueError is raised on Windows if the drive letters differ
                 # on each path.
                 raise error.Abort(
                     _(b'cannot calculate relative path'),
                     hint=stringutil.forcebytestr(e),
                 )
 
     if not wdirvfs.exists():
         wdirvfs.makedirs()
 
     hgvfs.makedir(notindexed=True)
     if b'sharedrepo' not in createopts:
         hgvfs.mkdir(b'cache')
     hgvfs.mkdir(b'wcache')
 
     has_store = requirementsmod.STORE_REQUIREMENT in requirements
     if has_store and b'sharedrepo' not in createopts:
         hgvfs.mkdir(b'store')
 
         # We create an invalid changelog outside the store so very old
         # Mercurial versions (which didn't know about the requirements
         # file) encounter an error on reading the changelog. This
         # effectively locks out old clients and prevents them from
         # mucking with a repo in an unknown format.
         #
         # The revlog header has version 65535, which won't be recognized by
         # such old clients.
         hgvfs.append(
             b'00changelog.i',
             b'\0\0\xFF\xFF dummy changelog to prevent using the old repo '
             b'layout',
         )
 
     # Filter the requirements into working copy and store ones
     wcreq, storereq = scmutil.filterrequirements(requirements)
     # write working copy ones
     scmutil.writerequires(hgvfs, wcreq)
     # If there are store requirements and the current repository
     # is not a shared one, write stored requirements
     # For new shared repository, we don't need to write the store
     # requirements as they are already present in store requires
     if storereq and b'sharedrepo' not in createopts:
         storevfs = vfsmod.vfs(hgvfs.join(b'store'), cacheaudited=True)
         scmutil.writerequires(storevfs, storereq)
 
     # Write out file telling readers where to find the shared store.
     if b'sharedrepo' in createopts:
         hgvfs.write(b'sharedpath', sharedpath)
 
     if createopts.get(b'shareditems'):
         shared = b'\n'.join(sorted(createopts[b'shareditems'])) + b'\n'
         hgvfs.write(b'shared', shared)
 
 
 def poisonrepository(repo):
     """Poison a repository instance so it can no longer be used."""
     # Perform any cleanup on the instance.
     repo.close()
 
     # Our strategy is to replace the type of the object with one that
     # has all attribute lookups result in error.
     #
     # But we have to allow the close() method because some constructors
     # of repos call close() on repo references.
     class poisonedrepository(object):
         def __getattribute__(self, item):
             if item == 'close':
                 return object.__getattribute__(self, item)
 
             raise error.ProgrammingError(
                 b'repo instances should not be used after unshare'
             )
 
         def close(self):
             pass
 
     # We may have a repoview, which intercepts __setattr__. So be sure
     # we operate at the lowest level possible.
     object.__setattr__(repo, '__class__', poisonedrepository)

mercurial/metadata.py

@@ -1,964 +1,967 b''
 # coding: utf-8
 # metadata.py -- code related to various metadata computation and access.
 #
 # Copyright 2019 Google, Inc <[e-mail]>
 # Copyright 2020 Pierre-Yves David <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 from __future__ import absolute_import, print_function
 
 import multiprocessing
 import struct
 
 from .node import nullrev
 from . import (
     error,
     pycompat,
     requirements as requirementsmod,
     util,
 )
 
 from .revlogutils import (
     constants as revlogconst,
     flagutil as sidedataflag,
     sidedata as sidedatamod,
 )
 
 
 class ChangingFiles(object):
     """A class recording the changes made to files by a changeset
 
     Actions performed on files are gathered into 3 sets:
 
     - added:   files actively added in the changeset.
     - merged:  files whose history got merged
     - removed: files removed in the revision
     - salvaged: files that might have been deleted by a merge but were not
     - touched: files affected by the merge
 
     and copies information is held by 2 mappings
 
     - copied_from_p1: {"<new-name>": "<source-name-in-p1>"} mapping for copies
     - copied_from_p2: {"<new-name>": "<source-name-in-p2>"} mapping for copies
 
     See their inline help for details.
     """
 
     def __init__(
         self,
         touched=None,
         added=None,
         removed=None,
         merged=None,
         salvaged=None,
         p1_copies=None,
         p2_copies=None,
     ):
         self._added = set(() if added is None else added)
         self._merged = set(() if merged is None else merged)
         self._removed = set(() if removed is None else removed)
         self._touched = set(() if touched is None else touched)
         self._salvaged = set(() if salvaged is None else salvaged)
         self._touched.update(self._added)
         self._touched.update(self._merged)
         self._touched.update(self._removed)
         self._p1_copies = dict(() if p1_copies is None else p1_copies)
         self._p2_copies = dict(() if p2_copies is None else p2_copies)
 
     def __eq__(self, other):
         return (
             self.added == other.added
             and self.merged == other.merged
             and self.removed == other.removed
             and self.salvaged == other.salvaged
             and self.touched == other.touched
             and self.copied_from_p1 == other.copied_from_p1
             and self.copied_from_p2 == other.copied_from_p2
         )
 
     @property
     def has_copies_info(self):
         return bool(
             self.removed
             or self.merged
             or self.salvaged
             or self.copied_from_p1
             or self.copied_from_p2
         )
 
     @util.propertycache
     def added(self):
         """files actively added in the changeset
 
         Any file present in that revision that was absent in all the changeset's
         parents.
 
         In case of merge, this means a file absent in one of the parents but
         existing in the other will *not* be contained in this set. (They were
         added by an ancestor)
         """
         return frozenset(self._added)
 
     def mark_added(self, filename):
         if 'added' in vars(self):
             del self.added
         self._added.add(filename)
         self.mark_touched(filename)
 
     def update_added(self, filenames):
         for f in filenames:
             self.mark_added(f)
 
     @util.propertycache
     def merged(self):
         """files actively merged during a merge
 
         Any modified files which had modification on both size that needed merging.
 
         In this case a new filenode was created and it has two parents.
         """
         return frozenset(self._merged)
 
     def mark_merged(self, filename):
         if 'merged' in vars(self):
             del self.merged
         self._merged.add(filename)
         self.mark_touched(filename)
 
     def update_merged(self, filenames):
         for f in filenames:
             self.mark_merged(f)
 
     @util.propertycache
     def removed(self):
         """files actively removed by the changeset
 
         In case of merge this will only contain the set of files removing "new"
         content. For any file absent in the current changeset:
 
         a) If the file exists in both parents, it is clearly "actively" removed
         by this changeset.
 
         b) If a file exists in only one parent and in none of the common
         ancestors, then the file was newly added in one of the merged branches
         and then got "actively" removed.
 
         c) If a file exists in only one parent and at least one of the common
         ancestors using the same filenode, then the file was unchanged on one
         side and deleted on the other side. The merge "passively" propagated
         that deletion, but didn't "actively" remove the file. In this case the
         file is *not* included in the `removed` set.
 
         d) If a file exists in only one parent and at least one of the common
         ancestors using a different filenode, then the file was changed on one
         side and removed on the other side. The merge process "actively"
         decided to drop the new change and delete the file. Unlike in the
         previous case, (c), the file included in the `removed` set.
 
         Summary table for merge:
 
         case | exists in parents | exists in gca || removed
          (a) |       both        |     *         ||   yes
          (b) |       one         |     none      ||   yes
          (c) |       one         | same filenode ||   no
          (d) |       one         |  new filenode ||   yes
         """
         return frozenset(self._removed)
 
     def mark_removed(self, filename):
         if 'removed' in vars(self):
             del self.removed
         self._removed.add(filename)
         self.mark_touched(filename)
 
     def update_removed(self, filenames):
         for f in filenames:
             self.mark_removed(f)
 
     @util.propertycache
     def salvaged(self):
         """files that might have been deleted by a merge, but still exists.
 
         During a merge, the manifest merging might select some files for
         removal, or for a removed/changed conflict. If at commit time the file
         still exists, its removal was "reverted" and the file is "salvaged"
         """
         return frozenset(self._salvaged)
 
     def mark_salvaged(self, filename):
         if "salvaged" in vars(self):
             del self.salvaged
         self._salvaged.add(filename)
         self.mark_touched(filename)
 
     def update_salvaged(self, filenames):
         for f in filenames:
             self.mark_salvaged(f)
 
     @util.propertycache
     def touched(self):
         """files either actively modified, added or removed"""
         return frozenset(self._touched)
 
     def mark_touched(self, filename):
         if 'touched' in vars(self):
             del self.touched
         self._touched.add(filename)
 
     def update_touched(self, filenames):
         for f in filenames:
             self.mark_touched(f)
 
     @util.propertycache
     def copied_from_p1(self):
         return self._p1_copies.copy()
 
     def mark_copied_from_p1(self, source, dest):
         if 'copied_from_p1' in vars(self):
             del self.copied_from_p1
         self._p1_copies[dest] = source
 
     def update_copies_from_p1(self, copies):
         for dest, source in copies.items():
             self.mark_copied_from_p1(source, dest)
 
     @util.propertycache
     def copied_from_p2(self):
         return self._p2_copies.copy()
 
     def mark_copied_from_p2(self, source, dest):
         if 'copied_from_p2' in vars(self):
             del self.copied_from_p2
         self._p2_copies[dest] = source
 
     def update_copies_from_p2(self, copies):
         for dest, source in copies.items():
             self.mark_copied_from_p2(source, dest)
 
 
 def compute_all_files_changes(ctx):
     """compute the files changed by a revision"""
     p1 = ctx.p1()
     p2 = ctx.p2()
     if p1.rev() == nullrev and p2.rev() == nullrev:
         return _process_root(ctx)
     elif p1.rev() != nullrev and p2.rev() == nullrev:
         return _process_linear(p1, ctx)
     elif p1.rev() == nullrev and p2.rev() != nullrev:
         # In the wild, one can encounter changeset where p1 is null but p2 is not
         return _process_linear(p1, ctx, parent=2)
     elif p1.rev() == p2.rev():
         # In the wild, one can encounter such "non-merge"
         return _process_linear(p1, ctx)
     else:
         return _process_merge(p1, p2, ctx)
 
 
 def _process_root(ctx):
     """compute the appropriate changed files for a changeset with no parents"""
     # Simple, there was nothing before it, so everything is added.
     md = ChangingFiles()
     manifest = ctx.manifest()
     for filename in manifest:
         md.mark_added(filename)
     return md
 
 
 def _process_linear(parent_ctx, children_ctx, parent=1):
     """compute the appropriate changed files for a changeset with a single parent"""
     md = ChangingFiles()
     parent_manifest = parent_ctx.manifest()
     children_manifest = children_ctx.manifest()
 
     copies_candidate = []
 
     for filename, d in parent_manifest.diff(children_manifest).items():
         if d[1][0] is None:
             # no filenode for the "new" value, file is absent
             md.mark_removed(filename)
         else:
             copies_candidate.append(filename)
             if d[0][0] is None:
                 # not filenode for the "old" value file was absent
                 md.mark_added(filename)
             else:
                 # filenode for both "old" and "new"
                 md.mark_touched(filename)
 
     if parent == 1:
         copied = md.mark_copied_from_p1
     elif parent == 2:
         copied = md.mark_copied_from_p2
     else:
         assert False, "bad parent value %d" % parent
 
     for filename in copies_candidate:
         copy_info = children_ctx[filename].renamed()
         if copy_info:
             source, srcnode = copy_info
             copied(source, filename)
 
     return md
 
 
 def _process_merge(p1_ctx, p2_ctx, ctx):
     """compute the appropriate changed files for a changeset with two parents
 
     This is a more advance case. The information we need to record is summarise
     in the following table:
 
     ┌──────────────┬──────────────┬──────────────┬──────────────┬──────────────┐
     │ diff ╲  diff │       ø      │ (Some, None) │ (None, Some) │ (Some, Some) │
     │  p2   ╲  p1  │              │              │              │              │
     ├──────────────┼──────────────┼──────────────┼──────────────┼──────────────┤
     │              │              │🄱  No Changes │🄳  No Changes │              │
     │  ø           │🄰  No Changes │      OR      │     OR       │🄵  No Changes │
     │              │              │🄲  Deleted[1] │🄴  Salvaged[2]│     [3]      │
     ├──────────────┼──────────────┼──────────────┼──────────────┼──────────────┤
     │              │🄶  No Changes │              │              │              │
     │ (Some, None) │      OR      │🄻  Deleted    │       ø      │      ø       │
     │              │🄷  Deleted[1] │              │              │              │
     ├──────────────┼──────────────┼──────────────┼──────────────┼──────────────┤
     │              │🄸  No Changes │              │              │   🄽 Touched  │
     │ (None, Some) │     OR       │      ø       │🄼   Added     │OR 🅀 Salvaged │
     │              │🄹  Salvaged[2]│              │   (copied?)  │   (copied?)  │
     ├──────────────┼──────────────┼──────────────┼──────────────┼──────────────┤
     │              │              │              │   🄾 Touched  │   🄿 Merged   │
     │ (Some, Some) │🄺  No Changes │      ø       │OR 🅁 Salvaged │OR 🅂 Touched  │
     │              │     [3]      │              │   (copied?)  │   (copied?)  │
     └──────────────┴──────────────┴──────────────┴──────────────┴──────────────┘
 
     Special case [1]:
 
       The situation is:
         - parent-A:     file exists,
         - parent-B:     no file,
         - working-copy: no file.
 
       Detecting a "deletion" will depend on the presence of actual change on
       the "parent-A" branch:
 
       Subcase 🄱 or 🄶 : if the state of the file in "parent-A" is unchanged
       compared to the merge ancestors, then parent-A branch left the file
       untouched while parent-B deleted it. We simply apply the change from
       "parent-B" branch the file was automatically dropped.
       The result is:
           - file is not recorded as touched by the merge.
 
       Subcase 🄲 or 🄷 : otherwise, the change from parent-A branch were explicitly dropped and
       the file was "deleted again". From a user perspective, the message
       about "locally changed" while "remotely deleted" (or the other way
       around) was issued and the user chose to deleted the file.
       The result:
           - file is recorded as touched by the merge.
 
 
     Special case [2]:
 
       The situation is:
         - parent-A:     no file,
         - parent-B:     file,
         - working-copy: file (same content as parent-B).
 
       There are three subcases depending on the ancestors contents:
 
       - A) the file is missing in all ancestors,
       - B) at least one ancestor has the file with filenode ≠ from parent-B,
       - C) all ancestors use the same filenode as parent-B,
 
       Subcase (A) is the simpler, nothing happend on parent-A side while
       parent-B added it.
 
         The result:
             - the file is not marked as touched by the merge.
 
       Subcase (B) is the counter part of "Special case [1]", the file was
         modified on parent-B side, while parent-A side deleted it. However this
         time, the conflict was solved by keeping the file (and its
         modification). We consider the file as "salvaged".
 
         The result:
             - the file is marked as "salvaged" by the merge.
 
       Subcase (C) is subtle variation of the case above. In this case, the
         file in unchanged on the parent-B side and actively removed on the
         parent-A side. So the merge machinery correctly decide it should be
         removed. However, the file was explicitly restored to its parent-B
         content before the merge was commited. The file is be marked
         as salvaged too. From the merge result perspective, this is similar to
         Subcase (B), however from the merge resolution perspective they differ
         since in (C), there was some conflict not obvious solution to the
         merge (That got reversed)
 
     Special case [3]:
 
       The situation is:
         - parent-A:     file,
         - parent-B:     file (different filenode as parent-A),
         - working-copy: file (same filenode as parent-B).
 
       This case is in theory much simple, for this to happens, this mean the
       filenode in parent-A is purely replacing the one in parent-B (either a
       descendant, or a full new file history, see changeset). So the merge
       introduce no changes, and the file is not affected by the merge...
 
       However, in the wild it is possible to find commit with the above is not
       True. For example repository have some commit where the *new* node is an
       ancestor of the node in parent-A, or where parent-A and parent-B are two
       branches of the same file history, yet not merge-filenode were created
       (while the "merge" should have led to a "modification").
 
       Detecting such cases (and not recording the file as modified) would be a
       nice bonus. However do not any of this yet.
     """
 
     repo = ctx.repo()
     md = ChangingFiles()
 
     m = ctx.manifest()
     p1m = p1_ctx.manifest()
     p2m = p2_ctx.manifest()
     diff_p1 = p1m.diff(m)
     diff_p2 = p2m.diff(m)
 
     cahs = ctx.repo().changelog.commonancestorsheads(
         p1_ctx.node(), p2_ctx.node()
     )
     if not cahs:
         cahs = [nullrev]
     mas = [ctx.repo()[r].manifest() for r in cahs]
 
     copy_candidates = []
 
     # Dealing with case 🄰 happens automatically.  Since there are no entry in
     # d1 nor d2, we won't iterate on it ever.
 
     # Iteration over d1 content will deal with all cases, but the one in the
     # first column of the table.
     for filename, d1 in diff_p1.items():
 
         d2 = diff_p2.pop(filename, None)
 
         if d2 is None:
             # this deal with the first line of the table.
             _process_other_unchanged(md, mas, filename, d1)
         else:
 
             if d1[0][0] is None and d2[0][0] is None:
                 # case 🄼 — both deleted the file.
                 md.mark_added(filename)
                 copy_candidates.append(filename)
             elif d1[1][0] is None and d2[1][0] is None:
                 # case 🄻 — both deleted the file.
                 md.mark_removed(filename)
             elif d1[1][0] is not None and d2[1][0] is not None:
                 if d1[0][0] is None or d2[0][0] is None:
                     if any(_find(ma, filename) is not None for ma in mas):
                         # case 🅀 or 🅁
                         md.mark_salvaged(filename)
                     else:
                         # case 🄽 🄾 : touched
                         md.mark_touched(filename)
                 else:
                     fctx = repo.filectx(filename, fileid=d1[1][0])
                     if fctx.p2().rev() == nullrev:
                         # case 🅂
                         # lets assume we can trust the file history. If the
                         # filenode is not a merge, the file was not merged.
                         md.mark_touched(filename)
                     else:
                         # case 🄿
                         md.mark_merged(filename)
                 copy_candidates.append(filename)
             else:
                 # Impossible case, the post-merge file status cannot be None on
                 # one side and Something on the other side.
                 assert False, "unreachable"
 
     # Iteration over remaining d2 content deal with the first column of the
     # table.
     for filename, d2 in diff_p2.items():
         _process_other_unchanged(md, mas, filename, d2)
 
     for filename in copy_candidates:
         copy_info = ctx[filename].renamed()
         if copy_info:
             source, srcnode = copy_info
             if source in p1_ctx and p1_ctx[source].filenode() == srcnode:
                 md.mark_copied_from_p1(source, filename)
             elif source in p2_ctx and p2_ctx[source].filenode() == srcnode:
                 md.mark_copied_from_p2(source, filename)
     return md
 
 
 def _find(manifest, filename):
     """return the associate filenode or None"""
     if filename not in manifest:
         return None
     return manifest.find(filename)[0]
 
 
 def _process_other_unchanged(md, mas, filename, diff):
     source_node = diff[0][0]
     target_node = diff[1][0]
 
     if source_node is not None and target_node is None:
         if any(not _find(ma, filename) == source_node for ma in mas):
             # case 🄲 of 🄷
             md.mark_removed(filename)
         # else, we have case 🄱 or 🄶 : no change need to be recorded
     elif source_node is None and target_node is not None:
         if any(_find(ma, filename) is not None for ma in mas):
             # case 🄴 or 🄹
             md.mark_salvaged(filename)
         # else, we have case 🄳 or 🄸 : simple merge without intervention
     elif source_node is not None and target_node is not None:
         # case 🄵  or 🄺 : simple merge without intervention
         #
         # In buggy case where source_node is not an ancestors of target_node.
         # There should have a been a new filenode created, recording this as
         # "modified". We do not deal with them yet.
         pass
     else:
         # An impossible case, the diff algorithm should not return entry if the
         # file is missing on both side.
         assert False, "unreachable"
 
 
 def _missing_from_all_ancestors(mas, filename):
     return all(_find(ma, filename) is None for ma in mas)
 
 
 def computechangesetfilesadded(ctx):
     """return the list of files added in a changeset"""
     added = []
     for f in ctx.files():
         if not any(f in p for p in ctx.parents()):
             added.append(f)
     return added
 
 
 def get_removal_filter(ctx, x=None):
     """return a function to detect files "wrongly" detected as `removed`
 
     When a file is removed relative to p1 in a merge, this
     function determines whether the absence is due to a
     deletion from a parent, or whether the merge commit
     itself deletes the file. We decide this by doing a
     simplified three way merge of the manifest entry for
     the file. There are two ways we decide the merge
     itself didn't delete a file:
     - neither parent (nor the merge) contain the file
     - exactly one parent contains the file, and that
       parent has the same filelog entry as the merge
       ancestor (or all of them if there two). In other
       words, that parent left the file unchanged while the
       other one deleted it.
     One way to think about this is that deleting a file is
     similar to emptying it, so the list of changed files
     should be similar either way. The computation
     described above is not done directly in _filecommit
     when creating the list of changed files, however
     it does something very similar by comparing filelog
     nodes.
     """
 
     if x is not None:
         p1, p2, m1, m2 = x
     else:
         p1 = ctx.p1()
         p2 = ctx.p2()
         m1 = p1.manifest()
         m2 = p2.manifest()
 
     @util.cachefunc
     def mas():
         p1n = p1.node()
         p2n = p2.node()
         cahs = ctx.repo().changelog.commonancestorsheads(p1n, p2n)
         if not cahs:
             cahs = [nullrev]
         return [ctx.repo()[r].manifest() for r in cahs]
 
     def deletionfromparent(f):
         if f in m1:
             return f not in m2 and all(
                 f in ma and ma.find(f) == m1.find(f) for ma in mas()
             )
         elif f in m2:
             return all(f in ma and ma.find(f) == m2.find(f) for ma in mas())
         else:
             return True
 
     return deletionfromparent
 
 
 def computechangesetfilesremoved(ctx):
     """return the list of files removed in a changeset"""
     removed = []
     for f in ctx.files():
         if f not in ctx:
             removed.append(f)
     if removed:
         rf = get_removal_filter(ctx)
         removed = [r for r in removed if not rf(r)]
     return removed
 
 
 def computechangesetfilesmerged(ctx):
     """return the list of files merged in a changeset"""
     merged = []
     if len(ctx.parents()) < 2:
         return merged
     for f in ctx.files():
         if f in ctx:
             fctx = ctx[f]
             parents = fctx._filelog.parents(fctx._filenode)
             if parents[1] != ctx.repo().nullid:
                 merged.append(f)
     return merged
 
 
 def computechangesetcopies(ctx):
     """return the copies data for a changeset
 
     The copies data are returned as a pair of dictionnary (p1copies, p2copies).
 
     Each dictionnary are in the form: `{newname: oldname}`
     """
     p1copies = {}
     p2copies = {}
     p1 = ctx.p1()
     p2 = ctx.p2()
     narrowmatch = ctx._repo.narrowmatch()
     for dst in ctx.files():
         if not narrowmatch(dst) or dst not in ctx:
             continue
         copied = ctx[dst].renamed()
         if not copied:
             continue
         src, srcnode = copied
         if src in p1 and p1[src].filenode() == srcnode:
             p1copies[dst] = src
         elif src in p2 and p2[src].filenode() == srcnode:
             p2copies[dst] = src
     return p1copies, p2copies
 
 
 def encodecopies(files, copies):
     items = []
     for i, dst in enumerate(files):
         if dst in copies:
             items.append(b'%d\0%s' % (i, copies[dst]))
     if len(items) != len(copies):
         raise error.ProgrammingError(
             b'some copy targets missing from file list'
         )
     return b"\n".join(items)
 
 
 def decodecopies(files, data):
     try:
         copies = {}
         if not data:
             return copies
         for l in data.split(b'\n'):
             strindex, src = l.split(b'\0')
             i = int(strindex)
             dst = files[i]
             copies[dst] = src
         return copies
     except (ValueError, IndexError):
         # Perhaps someone had chosen the same key name (e.g. "p1copies") and
         # used different syntax for the value.
         return None
 
 
 def encodefileindices(files, subset):
     subset = set(subset)
     indices = []
     for i, f in enumerate(files):
         if f in subset:
             indices.append(b'%d' % i)
     return b'\n'.join(indices)
 
 
 def decodefileindices(files, data):
     try:
         subset = []
         if not data:
             return subset
         for strindex in data.split(b'\n'):
             i = int(strindex)
             if i < 0 or i >= len(files):
                 return None
             subset.append(files[i])
         return subset
     except (ValueError, IndexError):
         # Perhaps someone had chosen the same key name (e.g. "added") and
         # used different syntax for the value.
         return None
 
 
 # see mercurial/helptext/internals/revlogs.txt for details about the format
 
 ACTION_MASK = int("111" "00", 2)
 # note: untouched file used as copy source will as `000` for this mask.
 ADDED_FLAG = int("001" "00", 2)
 MERGED_FLAG = int("010" "00", 2)
 REMOVED_FLAG = int("011" "00", 2)
 SALVAGED_FLAG = int("100" "00", 2)
 TOUCHED_FLAG = int("101" "00", 2)
 
 COPIED_MASK = int("11", 2)
 COPIED_FROM_P1_FLAG = int("10", 2)
 COPIED_FROM_P2_FLAG = int("11", 2)
 
 # structure is <flag><filename-end><copy-source>
 INDEX_HEADER = struct.Struct(">L")
 INDEX_ENTRY = struct.Struct(">bLL")
 
 
 def encode_files_sidedata(files):
     all_files = set(files.touched)
     all_files.update(files.copied_from_p1.values())
     all_files.update(files.copied_from_p2.values())
     all_files = sorted(all_files)
     file_idx = {f: i for (i, f) in enumerate(all_files)}
     file_idx[None] = 0
 
     chunks = [INDEX_HEADER.pack(len(all_files))]
 
     filename_length = 0
     for f in all_files:
         filename_size = len(f)
         filename_length += filename_size
         flag = 0
         if f in files.added:
             flag |= ADDED_FLAG
         elif f in files.merged:
             flag |= MERGED_FLAG
         elif f in files.removed:
             flag |= REMOVED_FLAG
         elif f in files.salvaged:
             flag |= SALVAGED_FLAG
         elif f in files.touched:
             flag |= TOUCHED_FLAG
 
         copy = None
         if f in files.copied_from_p1:
             flag |= COPIED_FROM_P1_FLAG
             copy = files.copied_from_p1.get(f)
         elif f in files.copied_from_p2:
             copy = files.copied_from_p2.get(f)
             flag |= COPIED_FROM_P2_FLAG
         copy_idx = file_idx[copy]
         chunks.append(INDEX_ENTRY.pack(flag, filename_length, copy_idx))
     chunks.extend(all_files)
     return {sidedatamod.SD_FILES: b''.join(chunks)}
 
 
 def decode_files_sidedata(sidedata):
     md = ChangingFiles()
     raw = sidedata.get(sidedatamod.SD_FILES)
 
     if raw is None:
         return md
 
     copies = []
     all_files = []
 
     assert len(raw) >= INDEX_HEADER.size
     total_files = INDEX_HEADER.unpack_from(raw, 0)[0]
 
     offset = INDEX_HEADER.size
     file_offset_base = offset + (INDEX_ENTRY.size * total_files)
     file_offset_last = file_offset_base
 
     assert len(raw) >= file_offset_base
 
     for idx in range(total_files):
         flag, file_end, copy_idx = INDEX_ENTRY.unpack_from(raw, offset)
         file_end += file_offset_base
         filename = raw[file_offset_last:file_end]
         filesize = file_end - file_offset_last
         assert len(filename) == filesize
         offset += INDEX_ENTRY.size
         file_offset_last = file_end
         all_files.append(filename)
         if flag & ACTION_MASK == ADDED_FLAG:
             md.mark_added(filename)
         elif flag & ACTION_MASK == MERGED_FLAG:
             md.mark_merged(filename)
         elif flag & ACTION_MASK == REMOVED_FLAG:
             md.mark_removed(filename)
         elif flag & ACTION_MASK == SALVAGED_FLAG:
             md.mark_salvaged(filename)
         elif flag & ACTION_MASK == TOUCHED_FLAG:
             md.mark_touched(filename)
 
         copied = None
         if flag & COPIED_MASK == COPIED_FROM_P1_FLAG:
             copied = md.mark_copied_from_p1
         elif flag & COPIED_MASK == COPIED_FROM_P2_FLAG:
             copied = md.mark_copied_from_p2
 
         if copied is not None:
             copies.append((copied, filename, copy_idx))
 
     for copied, filename, copy_idx in copies:
         copied(all_files[copy_idx], filename)
 
     return md
 
 
 def _getsidedata(srcrepo, rev):
     ctx = srcrepo[rev]
     files = compute_all_files_changes(ctx)
     return encode_files_sidedata(files), files.has_copies_info
 
 
 def copies_sidedata_computer(repo, revlog, rev, existing_sidedata):
-    return _getsidedata(repo, rev)[0]
+    sidedata, has_copies_info = _getsidedata(repo, rev)
+    flags_to_add = sidedataflag.REVIDX_HASCOPIESINFO if has_copies_info else 0
+    return sidedata, (flags_to_add, 0)
 
 
 def set_sidedata_spec_for_repo(repo):
     if requirementsmod.COPIESSDC_REQUIREMENT in repo.requirements:
         repo.register_wanted_sidedata(sidedatamod.SD_FILES)
     repo.register_sidedata_computer(
         revlogconst.KIND_CHANGELOG,
         sidedatamod.SD_FILES,
         (sidedatamod.SD_FILES,),
         copies_sidedata_computer,
+        sidedataflag.REVIDX_HASCOPIESINFO,
     )
 
 
 def getsidedataadder(srcrepo, destrepo):
     use_w = srcrepo.ui.configbool(b'experimental', b'worker.repository-upgrade')
     if pycompat.iswindows or not use_w:
         return _get_simple_sidedata_adder(srcrepo, destrepo)
     else:
         return _get_worker_sidedata_adder(srcrepo, destrepo)
 
 
 def _sidedata_worker(srcrepo, revs_queue, sidedata_queue, tokens):
     """The function used by worker precomputing sidedata
 
     It read an input queue containing revision numbers
     It write in an output queue containing (rev, <sidedata-map>)
 
     The `None` input value is used as a stop signal.
 
     The `tokens` semaphore is user to avoid having too many unprocessed
     entries. The workers needs to acquire one token before fetching a task.
     They will be released by the consumer of the produced data.
     """
     tokens.acquire()
     rev = revs_queue.get()
     while rev is not None:
         data = _getsidedata(srcrepo, rev)
         sidedata_queue.put((rev, data))
         tokens.acquire()
         rev = revs_queue.get()
     # processing of `None` is completed, release the token.
     tokens.release()
 
 
 BUFF_PER_WORKER = 50
 
 
 def _get_worker_sidedata_adder(srcrepo, destrepo):
     """The parallel version of the sidedata computation
 
     This code spawn a pool of worker that precompute a buffer of sidedata
     before we actually need them"""
     # avoid circular import copies -> scmutil -> worker -> copies
     from . import worker
 
     nbworkers = worker._numworkers(srcrepo.ui)
 
     tokens = multiprocessing.BoundedSemaphore(nbworkers * BUFF_PER_WORKER)
     revsq = multiprocessing.Queue()
     sidedataq = multiprocessing.Queue()
 
     assert srcrepo.filtername is None
     # queue all tasks beforehand, revision numbers are small and it make
     # synchronisation simpler
     #
     # Since the computation for each node can be quite expensive, the overhead
     # of using a single queue is not revelant. In practice, most computation
     # are fast but some are very expensive and dominate all the other smaller
     # cost.
     for r in srcrepo.changelog.revs():
         revsq.put(r)
     # queue the "no more tasks" markers
     for i in range(nbworkers):
         revsq.put(None)
 
     allworkers = []
     for i in range(nbworkers):
         args = (srcrepo, revsq, sidedataq, tokens)
         w = multiprocessing.Process(target=_sidedata_worker, args=args)
         allworkers.append(w)
         w.start()
 
     # dictionnary to store results for revision higher than we one we are
     # looking for. For example, if we need the sidedatamap for 42, and 43 is
     # received, when shelve 43 for later use.
     staging = {}
 
     def sidedata_companion(revlog, rev):
         data = {}, False
         if util.safehasattr(revlog, b'filteredrevs'):  # this is a changelog
             # Is the data previously shelved ?
             data = staging.pop(rev, None)
             if data is None:
                 # look at the queued result until we find the one we are lookig
                 # for (shelve the other ones)
                 r, data = sidedataq.get()
                 while r != rev:
                     staging[r] = data
                     r, data = sidedataq.get()
             tokens.release()
         sidedata, has_copies_info = data
         new_flag = 0
         if has_copies_info:
             new_flag = sidedataflag.REVIDX_HASCOPIESINFO
         return False, (), sidedata, new_flag, 0
 
     return sidedata_companion
 
 
 def _get_simple_sidedata_adder(srcrepo, destrepo):
     """The simple version of the sidedata computation
 
     It just compute it in the same thread on request"""
 
     def sidedatacompanion(revlog, rev):
         sidedata, has_copies_info = {}, False
         if util.safehasattr(revlog, 'filteredrevs'):  # this is a changelog
             sidedata, has_copies_info = _getsidedata(srcrepo, rev)
         new_flag = 0
         if has_copies_info:
             new_flag = sidedataflag.REVIDX_HASCOPIESINFO
 
         return False, (), sidedata, new_flag, 0
 
     return sidedatacompanion
 
 
 def getsidedataremover(srcrepo, destrepo):
     def sidedatacompanion(revlog, rev):
         f = ()
         if util.safehasattr(revlog, 'filteredrevs'):  # this is a changelog
             if revlog.flags(rev) & sidedataflag.REVIDX_SIDEDATA:
                 f = (
                     sidedatamod.SD_P1COPIES,
                     sidedatamod.SD_P2COPIES,
                     sidedatamod.SD_FILESADDED,
                     sidedatamod.SD_FILESREMOVED,
                 )
         return False, f, {}, 0, sidedataflag.REVIDX_HASCOPIESINFO
 
     return sidedatacompanion

mercurial/pure/parsers.py

@@ -1,381 +1,384 b''
 # parsers.py - Python implementation of parsers.c
 #
 # Copyright 2009 Olivia Mackall <[e-mail]> and others
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 import struct
 import zlib
 
 from ..node import (
     nullrev,
     sha1nodeconstants,
 )
 from .. import (
     pycompat,
     util,
 )
 
 from ..revlogutils import nodemap as nodemaputil
 from ..revlogutils import constants as revlog_constants
 
 stringio = pycompat.bytesio
 
 
 _pack = struct.pack
 _unpack = struct.unpack
 _compress = zlib.compress
 _decompress = zlib.decompress
 
 # Some code below makes tuples directly because it's more convenient. However,
 # code outside this module should always use dirstatetuple.
 def dirstatetuple(*x):
     # x is a tuple
     return x
 
 
 def gettype(q):
     return int(q & 0xFFFF)
 
 
 def offset_type(offset, type):
     return int(int(offset) << 16 | type)
 
 
 class BaseIndexObject(object):
     # Format of an index entry according to Python's `struct` language
     index_format = revlog_constants.INDEX_ENTRY_V1
     # Size of a C unsigned long long int, platform independent
     big_int_size = struct.calcsize(b'>Q')
     # Size of a C long int, platform independent
     int_size = struct.calcsize(b'>i')
     # An empty index entry, used as a default value to be overridden, or nullrev
     null_item = (0, 0, 0, -1, -1, -1, -1, sha1nodeconstants.nullid)
 
     @util.propertycache
     def entry_size(self):
         return self.index_format.size
 
     @property
     def nodemap(self):
         msg = b"index.nodemap is deprecated, use index.[has_node|rev|get_rev]"
         util.nouideprecwarn(msg, b'5.3', stacklevel=2)
         return self._nodemap
 
     @util.propertycache
     def _nodemap(self):
         nodemap = nodemaputil.NodeMap({sha1nodeconstants.nullid: nullrev})
         for r in range(0, len(self)):
             n = self[r][7]
             nodemap[n] = r
         return nodemap
 
     def has_node(self, node):
         """return True if the node exist in the index"""
         return node in self._nodemap
 
     def rev(self, node):
         """return a revision for a node
 
         If the node is unknown, raise a RevlogError"""
         return self._nodemap[node]
 
     def get_rev(self, node):
         """return a revision for a node
 
         If the node is unknown, return None"""
         return self._nodemap.get(node)
 
     def _stripnodes(self, start):
         if '_nodemap' in vars(self):
             for r in range(start, len(self)):
                 n = self[r][7]
                 del self._nodemap[n]
 
     def clearcaches(self):
         self.__dict__.pop('_nodemap', None)
 
     def __len__(self):
         return self._lgt + len(self._extra)
 
     def append(self, tup):
         if '_nodemap' in vars(self):
             self._nodemap[tup[7]] = len(self)
         data = self.index_format.pack(*tup)
         self._extra.append(data)
 
     def _check_index(self, i):
         if not isinstance(i, int):
             raise TypeError(b"expecting int indexes")
         if i < 0 or i >= len(self):
             raise IndexError
 
     def __getitem__(self, i):
         if i == -1:
             return self.null_item
         self._check_index(i)
         if i >= self._lgt:
             data = self._extra[i - self._lgt]
         else:
             index = self._calculate_index(i)
             data = self._data[index : index + self.entry_size]
         r = self.index_format.unpack(data)
         if self._lgt and i == 0:
             r = (offset_type(0, gettype(r[0])),) + r[1:]
         return r
 
     def pack_header(self, header):
         """pack header information as binary"""
         v_fmt = revlog_constants.INDEX_HEADER
         return v_fmt.pack(header)
 
     def entry_binary(self, rev):
         """return the raw binary string representing a revision"""
         entry = self[rev]
         p = revlog_constants.INDEX_ENTRY_V1.pack(*entry)
         if rev == 0:
             p = p[revlog_constants.INDEX_HEADER.size :]
         return p
 
 
 class IndexObject(BaseIndexObject):
     def __init__(self, data):
         assert len(data) % self.entry_size == 0, (
             len(data),
             self.entry_size,
             len(data) % self.entry_size,
         )
         self._data = data
         self._lgt = len(data) // self.entry_size
         self._extra = []
 
     def _calculate_index(self, i):
         return i * self.entry_size
 
     def __delitem__(self, i):
         if not isinstance(i, slice) or not i.stop == -1 or i.step is not None:
             raise ValueError(b"deleting slices only supports a:-1 with step 1")
         i = i.start
         self._check_index(i)
         self._stripnodes(i)
         if i < self._lgt:
             self._data = self._data[: i * self.entry_size]
             self._lgt = i
             self._extra = []
         else:
             self._extra = self._extra[: i - self._lgt]
 
 
 class PersistentNodeMapIndexObject(IndexObject):
     """a Debug oriented class to test persistent nodemap
 
     We need a simple python object to test API and higher level behavior. See
     the Rust implementation for  more serious usage. This should be used only
     through the dedicated `devel.persistent-nodemap` config.
     """
 
     def nodemap_data_all(self):
         """Return bytes containing a full serialization of a nodemap
 
         The nodemap should be valid for the full set of revisions in the
         index."""
         return nodemaputil.persistent_data(self)
 
     def nodemap_data_incremental(self):
         """Return bytes containing a incremental update to persistent nodemap
 
         This containst the data for an append-only update of the data provided
         in the last call to `update_nodemap_data`.
         """
         if self._nm_root is None:
             return None
         docket = self._nm_docket
         changed, data = nodemaputil.update_persistent_data(
             self, self._nm_root, self._nm_max_idx, self._nm_docket.tip_rev
         )
 
         self._nm_root = self._nm_max_idx = self._nm_docket = None
         return docket, changed, data
 
     def update_nodemap_data(self, docket, nm_data):
         """provide full block of persisted binary data for a nodemap
 
         The data are expected to come from disk. See `nodemap_data_all` for a
         produceur of such data."""
         if nm_data is not None:
             self._nm_root, self._nm_max_idx = nodemaputil.parse_data(nm_data)
             if self._nm_root:
                 self._nm_docket = docket
             else:
                 self._nm_root = self._nm_max_idx = self._nm_docket = None
 
 
 class InlinedIndexObject(BaseIndexObject):
     def __init__(self, data, inline=0):
         self._data = data
         self._lgt = self._inline_scan(None)
         self._inline_scan(self._lgt)
         self._extra = []
 
     def _inline_scan(self, lgt):
         off = 0
         if lgt is not None:
             self._offsets = [0] * lgt
         count = 0
         while off <= len(self._data) - self.entry_size:
             start = off + self.big_int_size
             (s,) = struct.unpack(
                 b'>i',
                 self._data[start : start + self.int_size],
             )
             if lgt is not None:
                 self._offsets[count] = off
             count += 1
             off += self.entry_size + s
         if off != len(self._data):
             raise ValueError(b"corrupted data")
         return count
 
     def __delitem__(self, i):
         if not isinstance(i, slice) or not i.stop == -1 or i.step is not None:
             raise ValueError(b"deleting slices only supports a:-1 with step 1")
         i = i.start
         self._check_index(i)
         self._stripnodes(i)
         if i < self._lgt:
             self._offsets = self._offsets[:i]
             self._lgt = i
             self._extra = []
         else:
             self._extra = self._extra[: i - self._lgt]
 
     def _calculate_index(self, i):
         return self._offsets[i]
 
 
 def parse_index2(data, inline, revlogv2=False):
     if not inline:
         cls = IndexObject2 if revlogv2 else IndexObject
         return cls(data), None
     cls = InlinedIndexObject2 if revlogv2 else InlinedIndexObject
     return cls(data, inline), (0, data)
 
 
 class Index2Mixin(object):
     index_format = revlog_constants.INDEX_ENTRY_V2
     null_item = (0, 0, 0, -1, -1, -1, -1, sha1nodeconstants.nullid, 0, 0)
 
-    def replace_sidedata_info(self, i, sidedata_offset, sidedata_length):
+    def replace_sidedata_info(
+        self, i, sidedata_offset, sidedata_length, offset_flags
+    ):
         """
         Replace an existing index entry's sidedata offset and length with new
         ones.
         This cannot be used outside of the context of sidedata rewriting,
         inside the transaction that creates the revision `i`.
         """
         if i < 0:
             raise KeyError
         self._check_index(i)
         sidedata_format = b">Qi"
         packed_size = struct.calcsize(sidedata_format)
         if i >= self._lgt:
             packed = _pack(sidedata_format, sidedata_offset, sidedata_length)
             old = self._extra[i - self._lgt]
-            new = old[:64] + packed + old[64 + packed_size :]
+            offset_flags = struct.pack(b">Q", offset_flags)
+            new = offset_flags + old[8:64] + packed + old[64 + packed_size :]
             self._extra[i - self._lgt] = new
         else:
             msg = b"cannot rewrite entries outside of this transaction"
             raise KeyError(msg)
 
     def entry_binary(self, rev):
         """return the raw binary string representing a revision"""
         entry = self[rev]
         p = revlog_constants.INDEX_ENTRY_V2.pack(*entry)
         if rev == 0:
             p = p[revlog_constants.INDEX_HEADER.size :]
         return p
 
 
 class IndexObject2(Index2Mixin, IndexObject):
     pass
 
 
 class InlinedIndexObject2(Index2Mixin, InlinedIndexObject):
     def _inline_scan(self, lgt):
         sidedata_length_pos = 72
         off = 0
         if lgt is not None:
             self._offsets = [0] * lgt
         count = 0
         while off <= len(self._data) - self.entry_size:
             start = off + self.big_int_size
             (data_size,) = struct.unpack(
                 b'>i',
                 self._data[start : start + self.int_size],
             )
             start = off + sidedata_length_pos
             (side_data_size,) = struct.unpack(
                 b'>i', self._data[start : start + self.int_size]
             )
             if lgt is not None:
                 self._offsets[count] = off
             count += 1
             off += self.entry_size + data_size + side_data_size
         if off != len(self._data):
             raise ValueError(b"corrupted data")
         return count
 
 
 def parse_index_devel_nodemap(data, inline):
     """like parse_index2, but alway return a PersistentNodeMapIndexObject"""
     return PersistentNodeMapIndexObject(data), None
 
 
 def parse_dirstate(dmap, copymap, st):
     parents = [st[:20], st[20:40]]
     # dereference fields so they will be local in loop
     format = b">cllll"
     e_size = struct.calcsize(format)
     pos1 = 40
     l = len(st)
 
     # the inner loop
     while pos1 < l:
         pos2 = pos1 + e_size
         e = _unpack(b">cllll", st[pos1:pos2])  # a literal here is faster
         pos1 = pos2 + e[4]
         f = st[pos2:pos1]
         if b'\0' in f:
             f, c = f.split(b'\0')
             copymap[f] = c
         dmap[f] = e[:4]
     return parents
 
 
 def pack_dirstate(dmap, copymap, pl, now):
     now = int(now)
     cs = stringio()
     write = cs.write
     write(b"".join(pl))
     for f, e in pycompat.iteritems(dmap):
         if e[0] == b'n' and e[3] == now:
             # The file was last modified "simultaneously" with the current
             # write to dirstate (i.e. within the same second for file-
             # systems with a granularity of 1 sec). This commonly happens
             # for at least a couple of files on 'update'.
             # The user could change the file without changing its size
             # within the same second. Invalidate the file's mtime in
             # dirstate, forcing future 'status' calls to compare the
             # contents of the file if the size is the same. This prevents
             # mistakenly treating such files as clean.
             e = dirstatetuple(e[0], e[1], e[2], -1)
             dmap[f] = e
 
         if f in copymap:
             f = b"%s\0%s" % (f, copymap[f])
         e = _pack(b">cllll", e[0], e[1], e[2], e[3], len(f))
         write(e)
         write(f)
     return cs.getvalue()

mercurial/revlog.py

@@ -1,3141 +1,3145 b''
 # revlog.py - storage back-end for mercurial
 #
 # Copyright 2005-2007 Olivia Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 """Storage back-end for Mercurial.
 
 This provides efficient delta storage with O(1) retrieve and append
 and O(changes) merge between branches.
 """
 
 from __future__ import absolute_import
 
 import binascii
 import collections
 import contextlib
 import errno
 import io
 import os
 import struct
 import zlib
 
 # import stuff from node for others to import from revlog
 from .node import (
     bin,
     hex,
     nullrev,
     sha1nodeconstants,
     short,
     wdirrev,
 )
 from .i18n import _
 from .pycompat import getattr
 from .revlogutils.constants import (
     ALL_KINDS,
     FLAG_GENERALDELTA,
     FLAG_INLINE_DATA,
     INDEX_HEADER,
     REVLOGV0,
     REVLOGV1,
     REVLOGV1_FLAGS,
     REVLOGV2,
     REVLOGV2_FLAGS,
     REVLOG_DEFAULT_FLAGS,
     REVLOG_DEFAULT_FORMAT,
     REVLOG_DEFAULT_VERSION,
 )
 from .revlogutils.flagutil import (
     REVIDX_DEFAULT_FLAGS,
     REVIDX_ELLIPSIS,
     REVIDX_EXTSTORED,
     REVIDX_FLAGS_ORDER,
     REVIDX_HASCOPIESINFO,
     REVIDX_ISCENSORED,
     REVIDX_RAWTEXT_CHANGING_FLAGS,
 )
 from .thirdparty import attr
 from . import (
     ancestor,
     dagop,
     error,
     mdiff,
     policy,
     pycompat,
     templatefilters,
     util,
 )
 from .interfaces import (
     repository,
     util as interfaceutil,
 )
 from .revlogutils import (
     deltas as deltautil,
     flagutil,
     nodemap as nodemaputil,
     revlogv0,
     sidedata as sidedatautil,
 )
 from .utils import (
     storageutil,
     stringutil,
 )
 
 # blanked usage of all the name to prevent pyflakes constraints
 # We need these name available in the module for extensions.
 REVLOGV0
 REVLOGV1
 REVLOGV2
 FLAG_INLINE_DATA
 FLAG_GENERALDELTA
 REVLOG_DEFAULT_FLAGS
 REVLOG_DEFAULT_FORMAT
 REVLOG_DEFAULT_VERSION
 REVLOGV1_FLAGS
 REVLOGV2_FLAGS
 REVIDX_ISCENSORED
 REVIDX_ELLIPSIS
 REVIDX_HASCOPIESINFO
 REVIDX_EXTSTORED
 REVIDX_DEFAULT_FLAGS
 REVIDX_FLAGS_ORDER
 REVIDX_RAWTEXT_CHANGING_FLAGS
 
 parsers = policy.importmod('parsers')
 rustancestor = policy.importrust('ancestor')
 rustdagop = policy.importrust('dagop')
 rustrevlog = policy.importrust('revlog')
 
 # Aliased for performance.
 _zlibdecompress = zlib.decompress
 
 # max size of revlog with inline data
 _maxinline = 131072
 _chunksize = 1048576
 
 # Flag processors for REVIDX_ELLIPSIS.
 def ellipsisreadprocessor(rl, text):
     return text, False
 
 
 def ellipsiswriteprocessor(rl, text):
     return text, False
 
 
 def ellipsisrawprocessor(rl, text):
     return False
 
 
 ellipsisprocessor = (
     ellipsisreadprocessor,
     ellipsiswriteprocessor,
     ellipsisrawprocessor,
 )
 
 
 def offset_type(offset, type):
     if (type & ~flagutil.REVIDX_KNOWN_FLAGS) != 0:
         raise ValueError(b'unknown revlog index flags')
     return int(int(offset) << 16 | type)
 
 
 def _verify_revision(rl, skipflags, state, node):
     """Verify the integrity of the given revlog ``node`` while providing a hook
     point for extensions to influence the operation."""
     if skipflags:
         state[b'skipread'].add(node)
     else:
         # Side-effect: read content and verify hash.
         rl.revision(node)
 
 
 # True if a fast implementation for persistent-nodemap is available
 #
 # We also consider we have a "fast" implementation in "pure" python because
 # people using pure don't really have performance consideration (and a
 # wheelbarrow of other slowness source)
 HAS_FAST_PERSISTENT_NODEMAP = rustrevlog is not None or util.safehasattr(
     parsers, 'BaseIndexObject'
 )
 
 
 @attr.s(slots=True, frozen=True)
 class _revisioninfo(object):
     """Information about a revision that allows building its fulltext
     node:       expected hash of the revision
     p1, p2:     parent revs of the revision
     btext:      built text cache consisting of a one-element list
     cachedelta: (baserev, uncompressed_delta) or None
     flags:      flags associated to the revision storage
 
     One of btext[0] or cachedelta must be set.
     """
 
     node = attr.ib()
     p1 = attr.ib()
     p2 = attr.ib()
     btext = attr.ib()
     textlen = attr.ib()
     cachedelta = attr.ib()
     flags = attr.ib()
 
 
 @interfaceutil.implementer(repository.irevisiondelta)
 @attr.s(slots=True)
 class revlogrevisiondelta(object):
     node = attr.ib()
     p1node = attr.ib()
     p2node = attr.ib()
     basenode = attr.ib()
     flags = attr.ib()
     baserevisionsize = attr.ib()
     revision = attr.ib()
     delta = attr.ib()
     sidedata = attr.ib()
     protocol_flags = attr.ib()
     linknode = attr.ib(default=None)
 
 
 @interfaceutil.implementer(repository.iverifyproblem)
 @attr.s(frozen=True)
 class revlogproblem(object):
     warning = attr.ib(default=None)
     error = attr.ib(default=None)
     node = attr.ib(default=None)
 
 
 def parse_index_v1(data, inline):
     # call the C implementation to parse the index data
     index, cache = parsers.parse_index2(data, inline)
     return index, cache
 
 
 def parse_index_v2(data, inline):
     # call the C implementation to parse the index data
     index, cache = parsers.parse_index2(data, inline, revlogv2=True)
     return index, cache
 
 
 if util.safehasattr(parsers, 'parse_index_devel_nodemap'):
 
     def parse_index_v1_nodemap(data, inline):
         index, cache = parsers.parse_index_devel_nodemap(data, inline)
         return index, cache
 
 
 else:
     parse_index_v1_nodemap = None
 
 
 def parse_index_v1_mixed(data, inline):
     index, cache = parse_index_v1(data, inline)
     return rustrevlog.MixedIndex(index), cache
 
 
 # corresponds to uncompressed length of indexformatng (2 gigs, 4-byte
 # signed integer)
 _maxentrysize = 0x7FFFFFFF
 
 
 class revlog(object):
     """
     the underlying revision storage object
 
     A revlog consists of two parts, an index and the revision data.
 
     The index is a file with a fixed record size containing
     information on each revision, including its nodeid (hash), the
     nodeids of its parents, the position and offset of its data within
     the data file, and the revision it's based on. Finally, each entry
     contains a linkrev entry that can serve as a pointer to external
     data.
 
     The revision data itself is a linear collection of data chunks.
     Each chunk represents a revision and is usually represented as a
     delta against the previous chunk. To bound lookup time, runs of
     deltas are limited to about 2 times the length of the original
     version data. This makes retrieval of a version proportional to
     its size, or O(1) relative to the number of revisions.
 
     Both pieces of the revlog are written to in an append-only
     fashion, which means we never need to rewrite a file to insert or
     remove data, and can use some simple techniques to avoid the need
     for locking while reading.
 
     If checkambig, indexfile is opened with checkambig=True at
     writing, to avoid file stat ambiguity.
 
     If mmaplargeindex is True, and an mmapindexthreshold is set, the
     index will be mmapped rather than read if it is larger than the
     configured threshold.
 
     If censorable is True, the revlog can have censored revisions.
 
     If `upperboundcomp` is not None, this is the expected maximal gain from
     compression for the data content.
 
     `concurrencychecker` is an optional function that receives 3 arguments: a
     file handle, a filename, and an expected position. It should check whether
     the current position in the file handle is valid, and log/warn/fail (by
     raising).
     """
 
     _flagserrorclass = error.RevlogError
 
     def __init__(
         self,
         opener,
         target,
         indexfile=None,
         datafile=None,
         checkambig=False,
         mmaplargeindex=False,
         censorable=False,
         upperboundcomp=None,
         persistentnodemap=False,
         concurrencychecker=None,
     ):
         """
         create a revlog object
 
         opener is a function that abstracts the file opening operation
         and can be used to implement COW semantics or the like.
 
         `target`: a (KIND, ID) tuple that identify the content stored in
         this revlog. It help the rest of the code to understand what the revlog
         is about without having to resort to heuristic and index filename
         analysis. Note: that this must be reliably be set by normal code, but
         that test, debug, or performance measurement code might not set this to
         accurate value.
         """
         self.upperboundcomp = upperboundcomp
         self.indexfile = indexfile
         self.datafile = datafile or (indexfile[:-2] + b".d")
         self.nodemap_file = None
         if persistentnodemap:
             self.nodemap_file = nodemaputil.get_nodemap_file(
                 opener, self.indexfile
             )
 
         self.opener = opener
         assert target[0] in ALL_KINDS
         assert len(target) == 2
         self.target = target
         #  When True, indexfile is opened with checkambig=True at writing, to
         #  avoid file stat ambiguity.
         self._checkambig = checkambig
         self._mmaplargeindex = mmaplargeindex
         self._censorable = censorable
         # 3-tuple of (node, rev, text) for a raw revision.
         self._revisioncache = None
         # Maps rev to chain base rev.
         self._chainbasecache = util.lrucachedict(100)
         # 2-tuple of (offset, data) of raw data from the revlog at an offset.
         self._chunkcache = (0, b'')
         # How much data to read and cache into the raw revlog data cache.
         self._chunkcachesize = 65536
         self._maxchainlen = None
         self._deltabothparents = True
         self.index = None
         self._nodemap_docket = None
         # Mapping of partial identifiers to full nodes.
         self._pcache = {}
         # Mapping of revision integer to full node.
         self._compengine = b'zlib'
         self._compengineopts = {}
         self._maxdeltachainspan = -1
         self._withsparseread = False
         self._sparserevlog = False
         self._srdensitythreshold = 0.50
         self._srmingapsize = 262144
 
         # Make copy of flag processors so each revlog instance can support
         # custom flags.
         self._flagprocessors = dict(flagutil.flagprocessors)
 
         # 2-tuple of file handles being used for active writing.
         self._writinghandles = None
 
         self._loadindex()
 
         self._concurrencychecker = concurrencychecker
 
     def _loadindex(self):
         mmapindexthreshold = None
         opts = self.opener.options
 
         if b'revlogv2' in opts:
             newversionflags = REVLOGV2 | FLAG_INLINE_DATA
         elif b'revlogv1' in opts:
             newversionflags = REVLOGV1 | FLAG_INLINE_DATA
             if b'generaldelta' in opts:
                 newversionflags |= FLAG_GENERALDELTA
         elif b'revlogv0' in self.opener.options:
             newversionflags = REVLOGV0
         else:
             newversionflags = REVLOG_DEFAULT_VERSION
 
         if b'chunkcachesize' in opts:
             self._chunkcachesize = opts[b'chunkcachesize']
         if b'maxchainlen' in opts:
             self._maxchainlen = opts[b'maxchainlen']
         if b'deltabothparents' in opts:
             self._deltabothparents = opts[b'deltabothparents']
         self._lazydelta = bool(opts.get(b'lazydelta', True))
         self._lazydeltabase = False
         if self._lazydelta:
             self._lazydeltabase = bool(opts.get(b'lazydeltabase', False))
         if b'compengine' in opts:
             self._compengine = opts[b'compengine']
         if b'zlib.level' in opts:
             self._compengineopts[b'zlib.level'] = opts[b'zlib.level']
         if b'zstd.level' in opts:
             self._compengineopts[b'zstd.level'] = opts[b'zstd.level']
         if b'maxdeltachainspan' in opts:
             self._maxdeltachainspan = opts[b'maxdeltachainspan']
         if self._mmaplargeindex and b'mmapindexthreshold' in opts:
             mmapindexthreshold = opts[b'mmapindexthreshold']
         self.hassidedata = bool(opts.get(b'side-data', False))
         self._sparserevlog = bool(opts.get(b'sparse-revlog', False))
         withsparseread = bool(opts.get(b'with-sparse-read', False))
         # sparse-revlog forces sparse-read
         self._withsparseread = self._sparserevlog or withsparseread
         if b'sparse-read-density-threshold' in opts:
             self._srdensitythreshold = opts[b'sparse-read-density-threshold']
         if b'sparse-read-min-gap-size' in opts:
             self._srmingapsize = opts[b'sparse-read-min-gap-size']
         if opts.get(b'enableellipsis'):
             self._flagprocessors[REVIDX_ELLIPSIS] = ellipsisprocessor
 
         # revlog v0 doesn't have flag processors
         for flag, processor in pycompat.iteritems(
             opts.get(b'flagprocessors', {})
         ):
             flagutil.insertflagprocessor(flag, processor, self._flagprocessors)
 
         if self._chunkcachesize <= 0:
             raise error.RevlogError(
                 _(b'revlog chunk cache size %r is not greater than 0')
                 % self._chunkcachesize
             )
         elif self._chunkcachesize & (self._chunkcachesize - 1):
             raise error.RevlogError(
                 _(b'revlog chunk cache size %r is not a power of 2')
                 % self._chunkcachesize
             )
 
         indexdata = b''
         self._initempty = True
         try:
             with self._indexfp() as f:
                 if (
                     mmapindexthreshold is not None
                     and self.opener.fstat(f).st_size >= mmapindexthreshold
                 ):
                     # TODO: should .close() to release resources without
                     # relying on Python GC
                     indexdata = util.buffer(util.mmapread(f))
                 else:
                     indexdata = f.read()
             if len(indexdata) > 0:
                 versionflags = INDEX_HEADER.unpack(indexdata[:4])[0]
                 self._initempty = False
             else:
                 versionflags = newversionflags
         except IOError as inst:
             if inst.errno != errno.ENOENT:
                 raise
 
             versionflags = newversionflags
 
         self.version = versionflags
 
         flags = versionflags & ~0xFFFF
         fmt = versionflags & 0xFFFF
 
         if fmt == REVLOGV0:
             if flags:
                 raise error.RevlogError(
                     _(b'unknown flags (%#04x) in version %d revlog %s')
                     % (flags >> 16, fmt, self.indexfile)
                 )
 
             self._inline = False
             self._generaldelta = False
 
         elif fmt == REVLOGV1:
             if flags & ~REVLOGV1_FLAGS:
                 raise error.RevlogError(
                     _(b'unknown flags (%#04x) in version %d revlog %s')
                     % (flags >> 16, fmt, self.indexfile)
                 )
 
             self._inline = versionflags & FLAG_INLINE_DATA
             self._generaldelta = versionflags & FLAG_GENERALDELTA
 
         elif fmt == REVLOGV2:
             if flags & ~REVLOGV2_FLAGS:
                 raise error.RevlogError(
                     _(b'unknown flags (%#04x) in version %d revlog %s')
                     % (flags >> 16, fmt, self.indexfile)
                 )
 
             # There is a bug in the transaction handling when going from an
             # inline revlog to a separate index and data file. Turn it off until
             # it's fixed, since v2 revlogs sometimes get rewritten on exchange.
             # See issue6485
             self._inline = False
             # generaldelta implied by version 2 revlogs.
             self._generaldelta = True
 
         else:
             raise error.RevlogError(
                 _(b'unknown version (%d) in revlog %s') % (fmt, self.indexfile)
             )
 
         self.nodeconstants = sha1nodeconstants
         self.nullid = self.nodeconstants.nullid
 
         # sparse-revlog can't be on without general-delta (issue6056)
         if not self._generaldelta:
             self._sparserevlog = False
 
         self._storedeltachains = True
 
         devel_nodemap = (
             self.nodemap_file
             and opts.get(b'devel-force-nodemap', False)
             and parse_index_v1_nodemap is not None
         )
 
         use_rust_index = False
         if rustrevlog is not None:
             if self.nodemap_file is not None:
                 use_rust_index = True
             else:
                 use_rust_index = self.opener.options.get(b'rust.index')
 
         self._parse_index = parse_index_v1
         if self.version == REVLOGV0:
             self._parse_index = revlogv0.parse_index_v0
         elif fmt == REVLOGV2:
             self._parse_index = parse_index_v2
         elif devel_nodemap:
             self._parse_index = parse_index_v1_nodemap
         elif use_rust_index:
             self._parse_index = parse_index_v1_mixed
         try:
             d = self._parse_index(indexdata, self._inline)
             index, _chunkcache = d
             use_nodemap = (
                 not self._inline
                 and self.nodemap_file is not None
                 and util.safehasattr(index, 'update_nodemap_data')
             )
             if use_nodemap:
                 nodemap_data = nodemaputil.persisted_data(self)
                 if nodemap_data is not None:
                     docket = nodemap_data[0]
                     if (
                         len(d[0]) > docket.tip_rev
                         and d[0][docket.tip_rev][7] == docket.tip_node
                     ):
                         # no changelog tampering
                         self._nodemap_docket = docket
                         index.update_nodemap_data(*nodemap_data)
         except (ValueError, IndexError):
             raise error.RevlogError(
                 _(b"index %s is corrupted") % self.indexfile
             )
         self.index, self._chunkcache = d
         if not self._chunkcache:
             self._chunkclear()
         # revnum -> (chain-length, sum-delta-length)
         self._chaininfocache = util.lrucachedict(500)
         # revlog header -> revlog compressor
         self._decompressors = {}
 
     @util.propertycache
     def revlog_kind(self):
         return self.target[0]
 
     @util.propertycache
     def _compressor(self):
         engine = util.compengines[self._compengine]
         return engine.revlogcompressor(self._compengineopts)
 
     def _indexfp(self, mode=b'r'):
         """file object for the revlog's index file"""
         args = {'mode': mode}
         if mode != b'r':
             args['checkambig'] = self._checkambig
         if mode == b'w':
             args['atomictemp'] = True
         return self.opener(self.indexfile, **args)
 
     def _datafp(self, mode=b'r'):
         """file object for the revlog's data file"""
         return self.opener(self.datafile, mode=mode)
 
     @contextlib.contextmanager
     def _datareadfp(self, existingfp=None):
         """file object suitable to read data"""
         # Use explicit file handle, if given.
         if existingfp is not None:
             yield existingfp
 
         # Use a file handle being actively used for writes, if available.
         # There is some danger to doing this because reads will seek the
         # file. However, _writeentry() performs a SEEK_END before all writes,
         # so we should be safe.
         elif self._writinghandles:
             if self._inline:
                 yield self._writinghandles[0]
             else:
                 yield self._writinghandles[1]
 
         # Otherwise open a new file handle.
         else:
             if self._inline:
                 func = self._indexfp
             else:
                 func = self._datafp
             with func() as fp:
                 yield fp
 
     def tiprev(self):
         return len(self.index) - 1
 
     def tip(self):
         return self.node(self.tiprev())
 
     def __contains__(self, rev):
         return 0 <= rev < len(self)
 
     def __len__(self):
         return len(self.index)
 
     def __iter__(self):
         return iter(pycompat.xrange(len(self)))
 
     def revs(self, start=0, stop=None):
         """iterate over all rev in this revlog (from start to stop)"""
         return storageutil.iterrevs(len(self), start=start, stop=stop)
 
     @property
     def nodemap(self):
         msg = (
             b"revlog.nodemap is deprecated, "
             b"use revlog.index.[has_node|rev|get_rev]"
         )
         util.nouideprecwarn(msg, b'5.3', stacklevel=2)
         return self.index.nodemap
 
     @property
     def _nodecache(self):
         msg = b"revlog._nodecache is deprecated, use revlog.index.nodemap"
         util.nouideprecwarn(msg, b'5.3', stacklevel=2)
         return self.index.nodemap
 
     def hasnode(self, node):
         try:
             self.rev(node)
             return True
         except KeyError:
             return False
 
     def candelta(self, baserev, rev):
         """whether two revisions (baserev, rev) can be delta-ed or not"""
         # Disable delta if either rev requires a content-changing flag
         # processor (ex. LFS). This is because such flag processor can alter
         # the rawtext content that the delta will be based on, and two clients
         # could have a same revlog node with different flags (i.e. different
         # rawtext contents) and the delta could be incompatible.
         if (self.flags(baserev) & REVIDX_RAWTEXT_CHANGING_FLAGS) or (
             self.flags(rev) & REVIDX_RAWTEXT_CHANGING_FLAGS
         ):
             return False
         return True
 
     def update_caches(self, transaction):
         if self.nodemap_file is not None:
             if transaction is None:
                 nodemaputil.update_persistent_nodemap(self)
             else:
                 nodemaputil.setup_persistent_nodemap(transaction, self)
 
     def clearcaches(self):
         self._revisioncache = None
         self._chainbasecache.clear()
         self._chunkcache = (0, b'')
         self._pcache = {}
         self._nodemap_docket = None
         self.index.clearcaches()
         # The python code is the one responsible for validating the docket, we
         # end up having to refresh it here.
         use_nodemap = (
             not self._inline
             and self.nodemap_file is not None
             and util.safehasattr(self.index, 'update_nodemap_data')
         )
         if use_nodemap:
             nodemap_data = nodemaputil.persisted_data(self)
             if nodemap_data is not None:
                 self._nodemap_docket = nodemap_data[0]
                 self.index.update_nodemap_data(*nodemap_data)
 
     def rev(self, node):
         try:
             return self.index.rev(node)
         except TypeError:
             raise
         except error.RevlogError:
             # parsers.c radix tree lookup failed
             if (
                 node == self.nodeconstants.wdirid
                 or node in self.nodeconstants.wdirfilenodeids
             ):
                 raise error.WdirUnsupported
             raise error.LookupError(node, self.indexfile, _(b'no node'))
 
     # Accessors for index entries.
 
     # First tuple entry is 8 bytes. First 6 bytes are offset. Last 2 bytes
     # are flags.
     def start(self, rev):
         return int(self.index[rev][0] >> 16)
 
     def flags(self, rev):
         return self.index[rev][0] & 0xFFFF
 
     def length(self, rev):
         return self.index[rev][1]
 
     def sidedata_length(self, rev):
         if self.version & 0xFFFF != REVLOGV2:
             return 0
         return self.index[rev][9]
 
     def rawsize(self, rev):
         """return the length of the uncompressed text for a given revision"""
         l = self.index[rev][2]
         if l >= 0:
             return l
 
         t = self.rawdata(rev)
         return len(t)
 
     def size(self, rev):
         """length of non-raw text (processed by a "read" flag processor)"""
         # fast path: if no "read" flag processor could change the content,
         # size is rawsize. note: ELLIPSIS is known to not change the content.
         flags = self.flags(rev)
         if flags & (flagutil.REVIDX_KNOWN_FLAGS ^ REVIDX_ELLIPSIS) == 0:
             return self.rawsize(rev)
 
         return len(self.revision(rev, raw=False))
 
     def chainbase(self, rev):
         base = self._chainbasecache.get(rev)
         if base is not None:
             return base
 
         index = self.index
         iterrev = rev
         base = index[iterrev][3]
         while base != iterrev:
             iterrev = base
             base = index[iterrev][3]
 
         self._chainbasecache[rev] = base
         return base
 
     def linkrev(self, rev):
         return self.index[rev][4]
 
     def parentrevs(self, rev):
         try:
             entry = self.index[rev]
         except IndexError:
             if rev == wdirrev:
                 raise error.WdirUnsupported
             raise
         if entry[5] == nullrev:
             return entry[6], entry[5]
         else:
             return entry[5], entry[6]
 
     # fast parentrevs(rev) where rev isn't filtered
     _uncheckedparentrevs = parentrevs
 
     def node(self, rev):
         try:
             return self.index[rev][7]
         except IndexError:
             if rev == wdirrev:
                 raise error.WdirUnsupported
             raise
 
     # Derived from index values.
 
     def end(self, rev):
         return self.start(rev) + self.length(rev)
 
     def parents(self, node):
         i = self.index
         d = i[self.rev(node)]
         # inline node() to avoid function call overhead
         if d[5] == self.nullid:
             return i[d[6]][7], i[d[5]][7]
         else:
             return i[d[5]][7], i[d[6]][7]
 
     def chainlen(self, rev):
         return self._chaininfo(rev)[0]
 
     def _chaininfo(self, rev):
         chaininfocache = self._chaininfocache
         if rev in chaininfocache:
             return chaininfocache[rev]
         index = self.index
         generaldelta = self._generaldelta
         iterrev = rev
         e = index[iterrev]
         clen = 0
         compresseddeltalen = 0
         while iterrev != e[3]:
             clen += 1
             compresseddeltalen += e[1]
             if generaldelta:
                 iterrev = e[3]
             else:
                 iterrev -= 1
             if iterrev in chaininfocache:
                 t = chaininfocache[iterrev]
                 clen += t[0]
                 compresseddeltalen += t[1]
                 break
             e = index[iterrev]
         else:
             # Add text length of base since decompressing that also takes
             # work. For cache hits the length is already included.
             compresseddeltalen += e[1]
         r = (clen, compresseddeltalen)
         chaininfocache[rev] = r
         return r
 
     def _deltachain(self, rev, stoprev=None):
         """Obtain the delta chain for a revision.
 
         ``stoprev`` specifies a revision to stop at. If not specified, we
         stop at the base of the chain.
 
         Returns a 2-tuple of (chain, stopped) where ``chain`` is a list of
         revs in ascending order and ``stopped`` is a bool indicating whether
         ``stoprev`` was hit.
         """
         # Try C implementation.
         try:
             return self.index.deltachain(rev, stoprev, self._generaldelta)
         except AttributeError:
             pass
 
         chain = []
 
         # Alias to prevent attribute lookup in tight loop.
         index = self.index
         generaldelta = self._generaldelta
 
         iterrev = rev
         e = index[iterrev]
         while iterrev != e[3] and iterrev != stoprev:
             chain.append(iterrev)
             if generaldelta:
                 iterrev = e[3]
             else:
                 iterrev -= 1
             e = index[iterrev]
 
         if iterrev == stoprev:
             stopped = True
         else:
             chain.append(iterrev)
             stopped = False
 
         chain.reverse()
         return chain, stopped
 
     def ancestors(self, revs, stoprev=0, inclusive=False):
         """Generate the ancestors of 'revs' in reverse revision order.
         Does not generate revs lower than stoprev.
 
         See the documentation for ancestor.lazyancestors for more details."""
 
         # first, make sure start revisions aren't filtered
         revs = list(revs)
         checkrev = self.node
         for r in revs:
             checkrev(r)
         # and we're sure ancestors aren't filtered as well
 
         if rustancestor is not None:
             lazyancestors = rustancestor.LazyAncestors
             arg = self.index
         else:
             lazyancestors = ancestor.lazyancestors
             arg = self._uncheckedparentrevs
         return lazyancestors(arg, revs, stoprev=stoprev, inclusive=inclusive)
 
     def descendants(self, revs):
         return dagop.descendantrevs(revs, self.revs, self.parentrevs)
 
     def findcommonmissing(self, common=None, heads=None):
         """Return a tuple of the ancestors of common and the ancestors of heads
         that are not ancestors of common. In revset terminology, we return the
         tuple:
 
           ::common, (::heads) - (::common)
 
         The list is sorted by revision number, meaning it is
         topologically sorted.
 
         'heads' and 'common' are both lists of node IDs.  If heads is
         not supplied, uses all of the revlog's heads.  If common is not
         supplied, uses nullid."""
         if common is None:
             common = [self.nullid]
         if heads is None:
             heads = self.heads()
 
         common = [self.rev(n) for n in common]
         heads = [self.rev(n) for n in heads]
 
         # we want the ancestors, but inclusive
         class lazyset(object):
             def __init__(self, lazyvalues):
                 self.addedvalues = set()
                 self.lazyvalues = lazyvalues
 
             def __contains__(self, value):
                 return value in self.addedvalues or value in self.lazyvalues
 
             def __iter__(self):
                 added = self.addedvalues
                 for r in added:
                     yield r
                 for r in self.lazyvalues:
                     if not r in added:
                         yield r
 
             def add(self, value):
                 self.addedvalues.add(value)
 
             def update(self, values):
                 self.addedvalues.update(values)
 
         has = lazyset(self.ancestors(common))
         has.add(nullrev)
         has.update(common)
 
         # take all ancestors from heads that aren't in has
         missing = set()
         visit = collections.deque(r for r in heads if r not in has)
         while visit:
             r = visit.popleft()
             if r in missing:
                 continue
             else:
                 missing.add(r)
                 for p in self.parentrevs(r):
                     if p not in has:
                         visit.append(p)
         missing = list(missing)
         missing.sort()
         return has, [self.node(miss) for miss in missing]
 
     def incrementalmissingrevs(self, common=None):
         """Return an object that can be used to incrementally compute the
         revision numbers of the ancestors of arbitrary sets that are not
         ancestors of common. This is an ancestor.incrementalmissingancestors
         object.
 
         'common' is a list of revision numbers. If common is not supplied, uses
         nullrev.
         """
         if common is None:
             common = [nullrev]
 
         if rustancestor is not None:
             return rustancestor.MissingAncestors(self.index, common)
         return ancestor.incrementalmissingancestors(self.parentrevs, common)
 
     def findmissingrevs(self, common=None, heads=None):
         """Return the revision numbers of the ancestors of heads that
         are not ancestors of common.
 
         More specifically, return a list of revision numbers corresponding to
         nodes N such that every N satisfies the following constraints:
 
           1. N is an ancestor of some node in 'heads'
           2. N is not an ancestor of any node in 'common'
 
         The list is sorted by revision number, meaning it is
         topologically sorted.
 
         'heads' and 'common' are both lists of revision numbers.  If heads is
         not supplied, uses all of the revlog's heads.  If common is not
         supplied, uses nullid."""
         if common is None:
             common = [nullrev]
         if heads is None:
             heads = self.headrevs()
 
         inc = self.incrementalmissingrevs(common=common)
         return inc.missingancestors(heads)
 
     def findmissing(self, common=None, heads=None):
         """Return the ancestors of heads that are not ancestors of common.
 
         More specifically, return a list of nodes N such that every N
         satisfies the following constraints:
 
           1. N is an ancestor of some node in 'heads'
           2. N is not an ancestor of any node in 'common'
 
         The list is sorted by revision number, meaning it is
         topologically sorted.
 
         'heads' and 'common' are both lists of node IDs.  If heads is
         not supplied, uses all of the revlog's heads.  If common is not
         supplied, uses nullid."""
         if common is None:
             common = [self.nullid]
         if heads is None:
             heads = self.heads()
 
         common = [self.rev(n) for n in common]
         heads = [self.rev(n) for n in heads]
 
         inc = self.incrementalmissingrevs(common=common)
         return [self.node(r) for r in inc.missingancestors(heads)]
 
     def nodesbetween(self, roots=None, heads=None):
         """Return a topological path from 'roots' to 'heads'.
 
         Return a tuple (nodes, outroots, outheads) where 'nodes' is a
         topologically sorted list of all nodes N that satisfy both of
         these constraints:
 
           1. N is a descendant of some node in 'roots'
           2. N is an ancestor of some node in 'heads'
 
         Every node is considered to be both a descendant and an ancestor
         of itself, so every reachable node in 'roots' and 'heads' will be
         included in 'nodes'.
 
         'outroots' is the list of reachable nodes in 'roots', i.e., the
         subset of 'roots' that is returned in 'nodes'.  Likewise,
         'outheads' is the subset of 'heads' that is also in 'nodes'.
 
         'roots' and 'heads' are both lists of node IDs.  If 'roots' is
         unspecified, uses nullid as the only root.  If 'heads' is
         unspecified, uses list of all of the revlog's heads."""
         nonodes = ([], [], [])
         if roots is not None:
             roots = list(roots)
             if not roots:
                 return nonodes
             lowestrev = min([self.rev(n) for n in roots])
         else:
             roots = [self.nullid]  # Everybody's a descendant of nullid
             lowestrev = nullrev
         if (lowestrev == nullrev) and (heads is None):
             # We want _all_ the nodes!
             return (
                 [self.node(r) for r in self],
                 [self.nullid],
                 list(self.heads()),
             )
         if heads is None:
             # All nodes are ancestors, so the latest ancestor is the last
             # node.
             highestrev = len(self) - 1
             # Set ancestors to None to signal that every node is an ancestor.
             ancestors = None
             # Set heads to an empty dictionary for later discovery of heads
             heads = {}
         else:
             heads = list(heads)
             if not heads:
                 return nonodes
             ancestors = set()
             # Turn heads into a dictionary so we can remove 'fake' heads.
             # Also, later we will be using it to filter out the heads we can't
             # find from roots.
             heads = dict.fromkeys(heads, False)
             # Start at the top and keep marking parents until we're done.
             nodestotag = set(heads)
             # Remember where the top was so we can use it as a limit later.
             highestrev = max([self.rev(n) for n in nodestotag])
             while nodestotag:
                 # grab a node to tag
                 n = nodestotag.pop()
                 # Never tag nullid
                 if n == self.nullid:
                     continue
                 # A node's revision number represents its place in a
                 # topologically sorted list of nodes.
                 r = self.rev(n)
                 if r >= lowestrev:
                     if n not in ancestors:
                         # If we are possibly a descendant of one of the roots
                         # and we haven't already been marked as an ancestor
                         ancestors.add(n)  # Mark as ancestor
                         # Add non-nullid parents to list of nodes to tag.
                         nodestotag.update(
                             [p for p in self.parents(n) if p != self.nullid]
                         )
                     elif n in heads:  # We've seen it before, is it a fake head?
                         # So it is, real heads should not be the ancestors of
                         # any other heads.
                         heads.pop(n)
             if not ancestors:
                 return nonodes
             # Now that we have our set of ancestors, we want to remove any
             # roots that are not ancestors.
 
             # If one of the roots was nullid, everything is included anyway.
             if lowestrev > nullrev:
                 # But, since we weren't, let's recompute the lowest rev to not
                 # include roots that aren't ancestors.
 
                 # Filter out roots that aren't ancestors of heads
                 roots = [root for root in roots if root in ancestors]
                 # Recompute the lowest revision
                 if roots:
                     lowestrev = min([self.rev(root) for root in roots])
                 else:
                     # No more roots?  Return empty list
                     return nonodes
             else:
                 # We are descending from nullid, and don't need to care about
                 # any other roots.
                 lowestrev = nullrev
                 roots = [self.nullid]
         # Transform our roots list into a set.
         descendants = set(roots)
         # Also, keep the original roots so we can filter out roots that aren't
         # 'real' roots (i.e. are descended from other roots).
         roots = descendants.copy()
         # Our topologically sorted list of output nodes.
         orderedout = []
         # Don't start at nullid since we don't want nullid in our output list,
         # and if nullid shows up in descendants, empty parents will look like
         # they're descendants.
         for r in self.revs(start=max(lowestrev, 0), stop=highestrev + 1):
             n = self.node(r)
             isdescendant = False
             if lowestrev == nullrev:  # Everybody is a descendant of nullid
                 isdescendant = True
             elif n in descendants:
                 # n is already a descendant
                 isdescendant = True
                 # This check only needs to be done here because all the roots
                 # will start being marked is descendants before the loop.
                 if n in roots:
                     # If n was a root, check if it's a 'real' root.
                     p = tuple(self.parents(n))
                     # If any of its parents are descendants, it's not a root.
                     if (p[0] in descendants) or (p[1] in descendants):
                         roots.remove(n)
             else:
                 p = tuple(self.parents(n))
                 # A node is a descendant if either of its parents are
                 # descendants.  (We seeded the dependents list with the roots
                 # up there, remember?)
                 if (p[0] in descendants) or (p[1] in descendants):
                     descendants.add(n)
                     isdescendant = True
             if isdescendant and ((ancestors is None) or (n in ancestors)):
                 # Only include nodes that are both descendants and ancestors.
                 orderedout.append(n)
                 if (ancestors is not None) and (n in heads):
                     # We're trying to figure out which heads are reachable
                     # from roots.
                     # Mark this head as having been reached
                     heads[n] = True
                 elif ancestors is None:
                     # Otherwise, we're trying to discover the heads.
                     # Assume this is a head because if it isn't, the next step
                     # will eventually remove it.
                     heads[n] = True
                     # But, obviously its parents aren't.
                     for p in self.parents(n):
                         heads.pop(p, None)
         heads = [head for head, flag in pycompat.iteritems(heads) if flag]
         roots = list(roots)
         assert orderedout
         assert roots
         assert heads
         return (orderedout, roots, heads)
 
     def headrevs(self, revs=None):
         if revs is None:
             try:
                 return self.index.headrevs()
             except AttributeError:
                 return self._headrevs()
         if rustdagop is not None:
             return rustdagop.headrevs(self.index, revs)
         return dagop.headrevs(revs, self._uncheckedparentrevs)
 
     def computephases(self, roots):
         return self.index.computephasesmapsets(roots)
 
     def _headrevs(self):
         count = len(self)
         if not count:
             return [nullrev]
         # we won't iter over filtered rev so nobody is a head at start
         ishead = [0] * (count + 1)
         index = self.index
         for r in self:
             ishead[r] = 1  # I may be an head
             e = index[r]
             ishead[e[5]] = ishead[e[6]] = 0  # my parent are not
         return [r for r, val in enumerate(ishead) if val]
 
     def heads(self, start=None, stop=None):
         """return the list of all nodes that have no children
 
         if start is specified, only heads that are descendants of
         start will be returned
         if stop is specified, it will consider all the revs from stop
         as if they had no children
         """
         if start is None and stop is None:
             if not len(self):
                 return [self.nullid]
             return [self.node(r) for r in self.headrevs()]
 
         if start is None:
             start = nullrev
         else:
             start = self.rev(start)
 
         stoprevs = {self.rev(n) for n in stop or []}
 
         revs = dagop.headrevssubset(
             self.revs, self.parentrevs, startrev=start, stoprevs=stoprevs
         )
 
         return [self.node(rev) for rev in revs]
 
     def children(self, node):
         """find the children of a given node"""
         c = []
         p = self.rev(node)
         for r in self.revs(start=p + 1):
             prevs = [pr for pr in self.parentrevs(r) if pr != nullrev]
             if prevs:
                 for pr in prevs:
                     if pr == p:
                         c.append(self.node(r))
             elif p == nullrev:
                 c.append(self.node(r))
         return c
 
     def commonancestorsheads(self, a, b):
         """calculate all the heads of the common ancestors of nodes a and b"""
         a, b = self.rev(a), self.rev(b)
         ancs = self._commonancestorsheads(a, b)
         return pycompat.maplist(self.node, ancs)
 
     def _commonancestorsheads(self, *revs):
         """calculate all the heads of the common ancestors of revs"""
         try:
             ancs = self.index.commonancestorsheads(*revs)
         except (AttributeError, OverflowError):  # C implementation failed
             ancs = ancestor.commonancestorsheads(self.parentrevs, *revs)
         return ancs
 
     def isancestor(self, a, b):
         """return True if node a is an ancestor of node b
 
         A revision is considered an ancestor of itself."""
         a, b = self.rev(a), self.rev(b)
         return self.isancestorrev(a, b)
 
     def isancestorrev(self, a, b):
         """return True if revision a is an ancestor of revision b
 
         A revision is considered an ancestor of itself.
 
         The implementation of this is trivial but the use of
         reachableroots is not."""
         if a == nullrev:
             return True
         elif a == b:
             return True
         elif a > b:
             return False
         return bool(self.reachableroots(a, [b], [a], includepath=False))
 
     def reachableroots(self, minroot, heads, roots, includepath=False):
         """return (heads(::(<roots> and <roots>::<heads>)))
 
         If includepath is True, return (<roots>::<heads>)."""
         try:
             return self.index.reachableroots2(
                 minroot, heads, roots, includepath
             )
         except AttributeError:
             return dagop._reachablerootspure(
                 self.parentrevs, minroot, roots, heads, includepath
             )
 
     def ancestor(self, a, b):
         """calculate the "best" common ancestor of nodes a and b"""
 
         a, b = self.rev(a), self.rev(b)
         try:
             ancs = self.index.ancestors(a, b)
         except (AttributeError, OverflowError):
             ancs = ancestor.ancestors(self.parentrevs, a, b)
         if ancs:
             # choose a consistent winner when there's a tie
             return min(map(self.node, ancs))
         return self.nullid
 
     def _match(self, id):
         if isinstance(id, int):
             # rev
             return self.node(id)
         if len(id) == self.nodeconstants.nodelen:
             # possibly a binary node
             # odds of a binary node being all hex in ASCII are 1 in 10**25
             try:
                 node = id
                 self.rev(node)  # quick search the index
                 return node
             except error.LookupError:
                 pass  # may be partial hex id
         try:
             # str(rev)
             rev = int(id)
             if b"%d" % rev != id:
                 raise ValueError
             if rev < 0:
                 rev = len(self) + rev
             if rev < 0 or rev >= len(self):
                 raise ValueError
             return self.node(rev)
         except (ValueError, OverflowError):
             pass
         if len(id) == 2 * self.nodeconstants.nodelen:
             try:
                 # a full hex nodeid?
                 node = bin(id)
                 self.rev(node)
                 return node
             except (TypeError, error.LookupError):
                 pass
 
     def _partialmatch(self, id):
         # we don't care wdirfilenodeids as they should be always full hash
         maybewdir = self.nodeconstants.wdirhex.startswith(id)
         try:
             partial = self.index.partialmatch(id)
             if partial and self.hasnode(partial):
                 if maybewdir:
                     # single 'ff...' match in radix tree, ambiguous with wdir
                     raise error.RevlogError
                 return partial
             if maybewdir:
                 # no 'ff...' match in radix tree, wdir identified
                 raise error.WdirUnsupported
             return None
         except error.RevlogError:
             # parsers.c radix tree lookup gave multiple matches
             # fast path: for unfiltered changelog, radix tree is accurate
             if not getattr(self, 'filteredrevs', None):
                 raise error.AmbiguousPrefixLookupError(
                     id, self.indexfile, _(b'ambiguous identifier')
                 )
             # fall through to slow path that filters hidden revisions
         except (AttributeError, ValueError):
             # we are pure python, or key was too short to search radix tree
             pass
 
         if id in self._pcache:
             return self._pcache[id]
 
         if len(id) <= 40:
             try:
                 # hex(node)[:...]
                 l = len(id) // 2  # grab an even number of digits
                 prefix = bin(id[: l * 2])
                 nl = [e[7] for e in self.index if e[7].startswith(prefix)]
                 nl = [
                     n for n in nl if hex(n).startswith(id) and self.hasnode(n)
                 ]
                 if self.nodeconstants.nullhex.startswith(id):
                     nl.append(self.nullid)
                 if len(nl) > 0:
                     if len(nl) == 1 and not maybewdir:
                         self._pcache[id] = nl[0]
                         return nl[0]
                     raise error.AmbiguousPrefixLookupError(
                         id, self.indexfile, _(b'ambiguous identifier')
                     )
                 if maybewdir:
                     raise error.WdirUnsupported
                 return None
             except TypeError:
                 pass
 
     def lookup(self, id):
         """locate a node based on:
         - revision number or str(revision number)
         - nodeid or subset of hex nodeid
         """
         n = self._match(id)
         if n is not None:
             return n
         n = self._partialmatch(id)
         if n:
             return n
 
         raise error.LookupError(id, self.indexfile, _(b'no match found'))
 
     def shortest(self, node, minlength=1):
         """Find the shortest unambiguous prefix that matches node."""
 
         def isvalid(prefix):
             try:
                 matchednode = self._partialmatch(prefix)
             except error.AmbiguousPrefixLookupError:
                 return False
             except error.WdirUnsupported:
                 # single 'ff...' match
                 return True
             if matchednode is None:
                 raise error.LookupError(node, self.indexfile, _(b'no node'))
             return True
 
         def maybewdir(prefix):
             return all(c == b'f' for c in pycompat.iterbytestr(prefix))
 
         hexnode = hex(node)
 
         def disambiguate(hexnode, minlength):
             """Disambiguate against wdirid."""
             for length in range(minlength, len(hexnode) + 1):
                 prefix = hexnode[:length]
                 if not maybewdir(prefix):
                     return prefix
 
         if not getattr(self, 'filteredrevs', None):
             try:
                 length = max(self.index.shortest(node), minlength)
                 return disambiguate(hexnode, length)
             except error.RevlogError:
                 if node != self.nodeconstants.wdirid:
                     raise error.LookupError(node, self.indexfile, _(b'no node'))
             except AttributeError:
                 # Fall through to pure code
                 pass
 
         if node == self.nodeconstants.wdirid:
             for length in range(minlength, len(hexnode) + 1):
                 prefix = hexnode[:length]
                 if isvalid(prefix):
                     return prefix
 
         for length in range(minlength, len(hexnode) + 1):
             prefix = hexnode[:length]
             if isvalid(prefix):
                 return disambiguate(hexnode, length)
 
     def cmp(self, node, text):
         """compare text with a given file revision
 
         returns True if text is different than what is stored.
         """
         p1, p2 = self.parents(node)
         return storageutil.hashrevisionsha1(text, p1, p2) != node
 
     def _cachesegment(self, offset, data):
         """Add a segment to the revlog cache.
 
         Accepts an absolute offset and the data that is at that location.
         """
         o, d = self._chunkcache
         # try to add to existing cache
         if o + len(d) == offset and len(d) + len(data) < _chunksize:
             self._chunkcache = o, d + data
         else:
             self._chunkcache = offset, data
 
     def _readsegment(self, offset, length, df=None):
         """Load a segment of raw data from the revlog.
 
         Accepts an absolute offset, length to read, and an optional existing
         file handle to read from.
 
         If an existing file handle is passed, it will be seeked and the
         original seek position will NOT be restored.
 
         Returns a str or buffer of raw byte data.
 
         Raises if the requested number of bytes could not be read.
         """
         # Cache data both forward and backward around the requested
         # data, in a fixed size window. This helps speed up operations
         # involving reading the revlog backwards.
         cachesize = self._chunkcachesize
         realoffset = offset & ~(cachesize - 1)
         reallength = (
             (offset + length + cachesize) & ~(cachesize - 1)
         ) - realoffset
         with self._datareadfp(df) as df:
             df.seek(realoffset)
             d = df.read(reallength)
 
         self._cachesegment(realoffset, d)
         if offset != realoffset or reallength != length:
             startoffset = offset - realoffset
             if len(d) - startoffset < length:
                 raise error.RevlogError(
                     _(
                         b'partial read of revlog %s; expected %d bytes from '
                         b'offset %d, got %d'
                     )
                     % (
                         self.indexfile if self._inline else self.datafile,
                         length,
                         realoffset,
                         len(d) - startoffset,
                     )
                 )
 
             return util.buffer(d, startoffset, length)
 
         if len(d) < length:
             raise error.RevlogError(
                 _(
                     b'partial read of revlog %s; expected %d bytes from offset '
                     b'%d, got %d'
                 )
                 % (
                     self.indexfile if self._inline else self.datafile,
                     length,
                     offset,
                     len(d),
                 )
             )
 
         return d
 
     def _getsegment(self, offset, length, df=None):
         """Obtain a segment of raw data from the revlog.
 
         Accepts an absolute offset, length of bytes to obtain, and an
         optional file handle to the already-opened revlog. If the file
         handle is used, it's original seek position will not be preserved.
 
         Requests for data may be returned from a cache.
 
         Returns a str or a buffer instance of raw byte data.
         """
         o, d = self._chunkcache
         l = len(d)
 
         # is it in the cache?
         cachestart = offset - o
         cacheend = cachestart + length
         if cachestart >= 0 and cacheend <= l:
             if cachestart == 0 and cacheend == l:
                 return d  # avoid a copy
             return util.buffer(d, cachestart, cacheend - cachestart)
 
         return self._readsegment(offset, length, df=df)
 
     def _getsegmentforrevs(self, startrev, endrev, df=None):
         """Obtain a segment of raw data corresponding to a range of revisions.
 
         Accepts the start and end revisions and an optional already-open
         file handle to be used for reading. If the file handle is read, its
         seek position will not be preserved.
 
         Requests for data may be satisfied by a cache.
 
         Returns a 2-tuple of (offset, data) for the requested range of
         revisions. Offset is the integer offset from the beginning of the
         revlog and data is a str or buffer of the raw byte data.
 
         Callers will need to call ``self.start(rev)`` and ``self.length(rev)``
         to determine where each revision's data begins and ends.
         """
         # Inlined self.start(startrev) & self.end(endrev) for perf reasons
         # (functions are expensive).
         index = self.index
         istart = index[startrev]
         start = int(istart[0] >> 16)
         if startrev == endrev:
             end = start + istart[1]
         else:
             iend = index[endrev]
             end = int(iend[0] >> 16) + iend[1]
 
         if self._inline:
             start += (startrev + 1) * self.index.entry_size
             end += (endrev + 1) * self.index.entry_size
         length = end - start
 
         return start, self._getsegment(start, length, df=df)
 
     def _chunk(self, rev, df=None):
         """Obtain a single decompressed chunk for a revision.
 
         Accepts an integer revision and an optional already-open file handle
         to be used for reading. If used, the seek position of the file will not
         be preserved.
 
         Returns a str holding uncompressed data for the requested revision.
         """
         return self.decompress(self._getsegmentforrevs(rev, rev, df=df)[1])
 
     def _chunks(self, revs, df=None, targetsize=None):
         """Obtain decompressed chunks for the specified revisions.
 
         Accepts an iterable of numeric revisions that are assumed to be in
         ascending order. Also accepts an optional already-open file handle
         to be used for reading. If used, the seek position of the file will
         not be preserved.
 
         This function is similar to calling ``self._chunk()`` multiple times,
         but is faster.
 
         Returns a list with decompressed data for each requested revision.
         """
         if not revs:
             return []
         start = self.start
         length = self.length
         inline = self._inline
         iosize = self.index.entry_size
         buffer = util.buffer
 
         l = []
         ladd = l.append
 
         if not self._withsparseread:
             slicedchunks = (revs,)
         else:
             slicedchunks = deltautil.slicechunk(
                 self, revs, targetsize=targetsize
             )
 
         for revschunk in slicedchunks:
             firstrev = revschunk[0]
             # Skip trailing revisions with empty diff
             for lastrev in revschunk[::-1]:
                 if length(lastrev) != 0:
                     break
 
             try:
                 offset, data = self._getsegmentforrevs(firstrev, lastrev, df=df)
             except OverflowError:
                 # issue4215 - we can't cache a run of chunks greater than
                 # 2G on Windows
                 return [self._chunk(rev, df=df) for rev in revschunk]
 
             decomp = self.decompress
             for rev in revschunk:
                 chunkstart = start(rev)
                 if inline:
                     chunkstart += (rev + 1) * iosize
                 chunklength = length(rev)
                 ladd(decomp(buffer(data, chunkstart - offset, chunklength)))
 
         return l
 
     def _chunkclear(self):
         """Clear the raw chunk cache."""
         self._chunkcache = (0, b'')
 
     def deltaparent(self, rev):
         """return deltaparent of the given revision"""
         base = self.index[rev][3]
         if base == rev:
             return nullrev
         elif self._generaldelta:
             return base
         else:
             return rev - 1
 
     def issnapshot(self, rev):
         """tells whether rev is a snapshot"""
         if not self._sparserevlog:
             return self.deltaparent(rev) == nullrev
         elif util.safehasattr(self.index, b'issnapshot'):
             # directly assign the method to cache the testing and access
             self.issnapshot = self.index.issnapshot
             return self.issnapshot(rev)
         if rev == nullrev:
             return True
         entry = self.index[rev]
         base = entry[3]
         if base == rev:
             return True
         if base == nullrev:
             return True
         p1 = entry[5]
         p2 = entry[6]
         if base == p1 or base == p2:
             return False
         return self.issnapshot(base)
 
     def snapshotdepth(self, rev):
         """number of snapshot in the chain before this one"""
         if not self.issnapshot(rev):
             raise error.ProgrammingError(b'revision %d not a snapshot')
         return len(self._deltachain(rev)[0]) - 1
 
     def revdiff(self, rev1, rev2):
         """return or calculate a delta between two revisions
 
         The delta calculated is in binary form and is intended to be written to
         revlog data directly. So this function needs raw revision data.
         """
         if rev1 != nullrev and self.deltaparent(rev2) == rev1:
             return bytes(self._chunk(rev2))
 
         return mdiff.textdiff(self.rawdata(rev1), self.rawdata(rev2))
 
     def _processflags(self, text, flags, operation, raw=False):
         """deprecated entry point to access flag processors"""
         msg = b'_processflag(...) use the specialized variant'
         util.nouideprecwarn(msg, b'5.2', stacklevel=2)
         if raw:
             return text, flagutil.processflagsraw(self, text, flags)
         elif operation == b'read':
             return flagutil.processflagsread(self, text, flags)
         else:  # write operation
             return flagutil.processflagswrite(self, text, flags)
 
     def revision(self, nodeorrev, _df=None, raw=False):
         """return an uncompressed revision of a given node or revision
         number.
 
         _df - an existing file handle to read from. (internal-only)
         raw - an optional argument specifying if the revision data is to be
         treated as raw data when applying flag transforms. 'raw' should be set
         to True when generating changegroups or in debug commands.
         """
         if raw:
             msg = (
                 b'revlog.revision(..., raw=True) is deprecated, '
                 b'use revlog.rawdata(...)'
             )
             util.nouideprecwarn(msg, b'5.2', stacklevel=2)
         return self._revisiondata(nodeorrev, _df, raw=raw)[0]
 
     def sidedata(self, nodeorrev, _df=None):
         """a map of extra data related to the changeset but not part of the hash
 
         This function currently return a dictionary. However, more advanced
         mapping object will likely be used in the future for a more
         efficient/lazy code.
         """
         return self._revisiondata(nodeorrev, _df)[1]
 
     def _revisiondata(self, nodeorrev, _df=None, raw=False):
         # deal with <nodeorrev> argument type
         if isinstance(nodeorrev, int):
             rev = nodeorrev
             node = self.node(rev)
         else:
             node = nodeorrev
             rev = None
 
         # fast path the special `nullid` rev
         if node == self.nullid:
             return b"", {}
 
         # ``rawtext`` is the text as stored inside the revlog. Might be the
         # revision or might need to be processed to retrieve the revision.
         rev, rawtext, validated = self._rawtext(node, rev, _df=_df)
 
         if self.version & 0xFFFF == REVLOGV2:
             if rev is None:
                 rev = self.rev(node)
             sidedata = self._sidedata(rev)
         else:
             sidedata = {}
 
         if raw and validated:
             # if we don't want to process the raw text and that raw
             # text is cached, we can exit early.
             return rawtext, sidedata
         if rev is None:
             rev = self.rev(node)
         # the revlog's flag for this revision
         # (usually alter its state or content)
         flags = self.flags(rev)
 
         if validated and flags == REVIDX_DEFAULT_FLAGS:
             # no extra flags set, no flag processor runs, text = rawtext
             return rawtext, sidedata
 
         if raw:
             validatehash = flagutil.processflagsraw(self, rawtext, flags)
             text = rawtext
         else:
             r = flagutil.processflagsread(self, rawtext, flags)
             text, validatehash = r
         if validatehash:
             self.checkhash(text, node, rev=rev)
         if not validated:
             self._revisioncache = (node, rev, rawtext)
 
         return text, sidedata
 
     def _rawtext(self, node, rev, _df=None):
         """return the possibly unvalidated rawtext for a revision
 
         returns (rev, rawtext, validated)
         """
 
         # revision in the cache (could be useful to apply delta)
         cachedrev = None
         # An intermediate text to apply deltas to
         basetext = None
 
         # Check if we have the entry in cache
         # The cache entry looks like (node, rev, rawtext)
         if self._revisioncache:
             if self._revisioncache[0] == node:
                 return (rev, self._revisioncache[2], True)
             cachedrev = self._revisioncache[1]
 
         if rev is None:
             rev = self.rev(node)
 
         chain, stopped = self._deltachain(rev, stoprev=cachedrev)
         if stopped:
             basetext = self._revisioncache[2]
 
         # drop cache to save memory, the caller is expected to
         # update self._revisioncache after validating the text
         self._revisioncache = None
 
         targetsize = None
         rawsize = self.index[rev][2]
         if 0 <= rawsize:
             targetsize = 4 * rawsize
 
         bins = self._chunks(chain, df=_df, targetsize=targetsize)
         if basetext is None:
             basetext = bytes(bins[0])
             bins = bins[1:]
 
         rawtext = mdiff.patches(basetext, bins)
         del basetext  # let us have a chance to free memory early
         return (rev, rawtext, False)
 
     def _sidedata(self, rev):
         """Return the sidedata for a given revision number."""
         index_entry = self.index[rev]
         sidedata_offset = index_entry[8]
         sidedata_size = index_entry[9]
 
         if self._inline:
             sidedata_offset += self.index.entry_size * (1 + rev)
         if sidedata_size == 0:
             return {}
 
         segment = self._getsegment(sidedata_offset, sidedata_size)
         sidedata = sidedatautil.deserialize_sidedata(segment)
         return sidedata
 
     def rawdata(self, nodeorrev, _df=None):
         """return an uncompressed raw data of a given node or revision number.
 
         _df - an existing file handle to read from. (internal-only)
         """
         return self._revisiondata(nodeorrev, _df, raw=True)[0]
 
     def hash(self, text, p1, p2):
         """Compute a node hash.
 
         Available as a function so that subclasses can replace the hash
         as needed.
         """
         return storageutil.hashrevisionsha1(text, p1, p2)
 
     def checkhash(self, text, node, p1=None, p2=None, rev=None):
         """Check node hash integrity.
 
         Available as a function so that subclasses can extend hash mismatch
         behaviors as needed.
         """
         try:
             if p1 is None and p2 is None:
                 p1, p2 = self.parents(node)
             if node != self.hash(text, p1, p2):
                 # Clear the revision cache on hash failure. The revision cache
                 # only stores the raw revision and clearing the cache does have
                 # the side-effect that we won't have a cache hit when the raw
                 # revision data is accessed. But this case should be rare and
                 # it is extra work to teach the cache about the hash
                 # verification state.
                 if self._revisioncache and self._revisioncache[0] == node:
                     self._revisioncache = None
 
                 revornode = rev
                 if revornode is None:
                     revornode = templatefilters.short(hex(node))
                 raise error.RevlogError(
                     _(b"integrity check failed on %s:%s")
                     % (self.indexfile, pycompat.bytestr(revornode))
                 )
         except error.RevlogError:
             if self._censorable and storageutil.iscensoredtext(text):
                 raise error.CensoredNodeError(self.indexfile, node, text)
             raise
 
     def _enforceinlinesize(self, tr, fp=None):
         """Check if the revlog is too big for inline and convert if so.
 
         This should be called after revisions are added to the revlog. If the
         revlog has grown too large to be an inline revlog, it will convert it
         to use multiple index and data files.
         """
         tiprev = len(self) - 1
         if (
             not self._inline
             or (self.start(tiprev) + self.length(tiprev)) < _maxinline
         ):
             return
 
         troffset = tr.findoffset(self.indexfile)
         if troffset is None:
             raise error.RevlogError(
                 _(b"%s not found in the transaction") % self.indexfile
             )
         trindex = 0
         tr.add(self.datafile, 0)
 
         if fp:
             fp.flush()
             fp.close()
             # We can't use the cached file handle after close(). So prevent
             # its usage.
             self._writinghandles = None
 
         with self._indexfp(b'r') as ifh, self._datafp(b'w') as dfh:
             for r in self:
                 dfh.write(self._getsegmentforrevs(r, r, df=ifh)[1])
                 if troffset <= self.start(r):
                     trindex = r
 
         with self._indexfp(b'w') as fp:
             self.version &= ~FLAG_INLINE_DATA
             self._inline = False
             for i in self:
                 e = self.index.entry_binary(i)
                 if i == 0:
                     header = self.index.pack_header(self.version)
                     e = header + e
                 fp.write(e)
 
             # the temp file replace the real index when we exit the context
             # manager
 
         tr.replace(self.indexfile, trindex * self.index.entry_size)
         nodemaputil.setup_persistent_nodemap(tr, self)
         self._chunkclear()
 
     def _nodeduplicatecallback(self, transaction, node):
         """called when trying to add a node already stored."""
 
     def addrevision(
         self,
         text,
         transaction,
         link,
         p1,
         p2,
         cachedelta=None,
         node=None,
         flags=REVIDX_DEFAULT_FLAGS,
         deltacomputer=None,
         sidedata=None,
     ):
         """add a revision to the log
 
         text - the revision data to add
         transaction - the transaction object used for rollback
         link - the linkrev data to add
         p1, p2 - the parent nodeids of the revision
         cachedelta - an optional precomputed delta
         node - nodeid of revision; typically node is not specified, and it is
             computed by default as hash(text, p1, p2), however subclasses might
             use different hashing method (and override checkhash() in such case)
         flags - the known flags to set on the revision
         deltacomputer - an optional deltacomputer instance shared between
             multiple calls
         """
         if link == nullrev:
             raise error.RevlogError(
                 _(b"attempted to add linkrev -1 to %s") % self.indexfile
             )
 
         if sidedata is None:
             sidedata = {}
         elif not self.hassidedata:
             raise error.ProgrammingError(
                 _(b"trying to add sidedata to a revlog who don't support them")
             )
 
         if flags:
             node = node or self.hash(text, p1, p2)
 
         rawtext, validatehash = flagutil.processflagswrite(self, text, flags)
 
         # If the flag processor modifies the revision data, ignore any provided
         # cachedelta.
         if rawtext != text:
             cachedelta = None
 
         if len(rawtext) > _maxentrysize:
             raise error.RevlogError(
                 _(
                     b"%s: size of %d bytes exceeds maximum revlog storage of 2GiB"
                 )
                 % (self.indexfile, len(rawtext))
             )
 
         node = node or self.hash(rawtext, p1, p2)
         rev = self.index.get_rev(node)
         if rev is not None:
             return rev
 
         if validatehash:
             self.checkhash(rawtext, node, p1=p1, p2=p2)
 
         return self.addrawrevision(
             rawtext,
             transaction,
             link,
             p1,
             p2,
             node,
             flags,
             cachedelta=cachedelta,
             deltacomputer=deltacomputer,
             sidedata=sidedata,
         )
 
     def addrawrevision(
         self,
         rawtext,
         transaction,
         link,
         p1,
         p2,
         node,
         flags,
         cachedelta=None,
         deltacomputer=None,
         sidedata=None,
     ):
         """add a raw revision with known flags, node and parents
         useful when reusing a revision not stored in this revlog (ex: received
         over wire, or read from an external bundle).
         """
         dfh = None
         if not self._inline:
             dfh = self._datafp(b"a+")
         ifh = self._indexfp(b"a+")
         try:
             return self._addrevision(
                 node,
                 rawtext,
                 transaction,
                 link,
                 p1,
                 p2,
                 flags,
                 cachedelta,
                 ifh,
                 dfh,
                 deltacomputer=deltacomputer,
                 sidedata=sidedata,
             )
         finally:
             if dfh:
                 dfh.close()
             ifh.close()
 
     def compress(self, data):
         """Generate a possibly-compressed representation of data."""
         if not data:
             return b'', data
 
         compressed = self._compressor.compress(data)
 
         if compressed:
             # The revlog compressor added the header in the returned data.
             return b'', compressed
 
         if data[0:1] == b'\0':
             return b'', data
         return b'u', data
 
     def decompress(self, data):
         """Decompress a revlog chunk.
 
         The chunk is expected to begin with a header identifying the
         format type so it can be routed to an appropriate decompressor.
         """
         if not data:
             return data
 
         # Revlogs are read much more frequently than they are written and many
         # chunks only take microseconds to decompress, so performance is
         # important here.
         #
         # We can make a few assumptions about revlogs:
         #
         # 1) the majority of chunks will be compressed (as opposed to inline
         #    raw data).
         # 2) decompressing *any* data will likely by at least 10x slower than
         #    returning raw inline data.
         # 3) we want to prioritize common and officially supported compression
         #    engines
         #
         # It follows that we want to optimize for "decompress compressed data
         # when encoded with common and officially supported compression engines"
         # case over "raw data" and "data encoded by less common or non-official
         # compression engines." That is why we have the inline lookup first
         # followed by the compengines lookup.
         #
         # According to `hg perfrevlogchunks`, this is ~0.5% faster for zlib
         # compressed chunks. And this matters for changelog and manifest reads.
         t = data[0:1]
 
         if t == b'x':
             try:
                 return _zlibdecompress(data)
             except zlib.error as e:
                 raise error.RevlogError(
                     _(b'revlog decompress error: %s')
                     % stringutil.forcebytestr(e)
                 )
         # '\0' is more common than 'u' so it goes first.
         elif t == b'\0':
             return data
         elif t == b'u':
             return util.buffer(data, 1)
 
         try:
             compressor = self._decompressors[t]
         except KeyError:
             try:
                 engine = util.compengines.forrevlogheader(t)
                 compressor = engine.revlogcompressor(self._compengineopts)
                 self._decompressors[t] = compressor
             except KeyError:
                 raise error.RevlogError(
                     _(b'unknown compression type %s') % binascii.hexlify(t)
                 )
 
         return compressor.decompress(data)
 
     def _addrevision(
         self,
         node,
         rawtext,
         transaction,
         link,
         p1,
         p2,
         flags,
         cachedelta,
         ifh,
         dfh,
         alwayscache=False,
         deltacomputer=None,
         sidedata=None,
     ):
         """internal function to add revisions to the log
 
         see addrevision for argument descriptions.
 
         note: "addrevision" takes non-raw text, "_addrevision" takes raw text.
 
         if "deltacomputer" is not provided or None, a defaultdeltacomputer will
         be used.
 
         invariants:
         - rawtext is optional (can be None); if not set, cachedelta must be set.
           if both are set, they must correspond to each other.
         """
         if node == self.nullid:
             raise error.RevlogError(
                 _(b"%s: attempt to add null revision") % self.indexfile
             )
         if (
             node == self.nodeconstants.wdirid
             or node in self.nodeconstants.wdirfilenodeids
         ):
             raise error.RevlogError(
                 _(b"%s: attempt to add wdir revision") % self.indexfile
             )
 
         if self._inline:
             fh = ifh
         else:
             fh = dfh
 
         btext = [rawtext]
 
         curr = len(self)
         prev = curr - 1
 
         offset = self._get_data_offset(prev)
 
         if self._concurrencychecker:
             if self._inline:
                 # offset is "as if" it were in the .d file, so we need to add on
                 # the size of the entry metadata.
                 self._concurrencychecker(
                     ifh, self.indexfile, offset + curr * self.index.entry_size
                 )
             else:
                 # Entries in the .i are a consistent size.
                 self._concurrencychecker(
                     ifh, self.indexfile, curr * self.index.entry_size
                 )
                 self._concurrencychecker(dfh, self.datafile, offset)
 
         p1r, p2r = self.rev(p1), self.rev(p2)
 
         # full versions are inserted when the needed deltas
         # become comparable to the uncompressed text
         if rawtext is None:
             # need rawtext size, before changed by flag processors, which is
             # the non-raw size. use revlog explicitly to avoid filelog's extra
             # logic that might remove metadata size.
             textlen = mdiff.patchedsize(
                 revlog.size(self, cachedelta[0]), cachedelta[1]
             )
         else:
             textlen = len(rawtext)
 
         if deltacomputer is None:
             deltacomputer = deltautil.deltacomputer(self)
 
         revinfo = _revisioninfo(node, p1, p2, btext, textlen, cachedelta, flags)
 
         deltainfo = deltacomputer.finddeltainfo(revinfo, fh)
 
         if sidedata and self.version & 0xFFFF == REVLOGV2:
             serialized_sidedata = sidedatautil.serialize_sidedata(sidedata)
             sidedata_offset = offset + deltainfo.deltalen
         else:
             serialized_sidedata = b""
             # Don't store the offset if the sidedata is empty, that way
             # we can easily detect empty sidedata and they will be no different
             # than ones we manually add.
             sidedata_offset = 0
 
         e = (
             offset_type(offset, flags),
             deltainfo.deltalen,
             textlen,
             deltainfo.base,
             link,
             p1r,
             p2r,
             node,
             sidedata_offset,
             len(serialized_sidedata),
         )
 
         if self.version & 0xFFFF != REVLOGV2:
             e = e[:8]
 
         self.index.append(e)
         entry = self.index.entry_binary(curr)
         if curr == 0:
             header = self.index.pack_header(self.version)
             entry = header + entry
         self._writeentry(
             transaction,
             ifh,
             dfh,
             entry,
             deltainfo.data,
             link,
             offset,
             serialized_sidedata,
         )
 
         rawtext = btext[0]
 
         if alwayscache and rawtext is None:
             rawtext = deltacomputer.buildtext(revinfo, fh)
 
         if type(rawtext) == bytes:  # only accept immutable objects
             self._revisioncache = (node, curr, rawtext)
         self._chainbasecache[curr] = deltainfo.chainbase
         return curr
 
     def _get_data_offset(self, prev):
         """Returns the current offset in the (in-transaction) data file.
         Versions < 2 of the revlog can get this 0(1), revlog v2 needs a docket
         file to store that information: since sidedata can be rewritten to the
         end of the data file within a transaction, you can have cases where, for
         example, rev `n` does not have sidedata while rev `n - 1` does, leading
         to `n - 1`'s sidedata being written after `n`'s data.
 
         TODO cache this in a docket file before getting out of experimental."""
         if self.version & 0xFFFF != REVLOGV2:
             return self.end(prev)
 
         offset = 0
         for rev, entry in enumerate(self.index):
             sidedata_end = entry[8] + entry[9]
             # Sidedata for a previous rev has potentially been written after
             # this rev's end, so take the max.
             offset = max(self.end(rev), offset, sidedata_end)
         return offset
 
     def _writeentry(
         self, transaction, ifh, dfh, entry, data, link, offset, sidedata
     ):
         # Files opened in a+ mode have inconsistent behavior on various
         # platforms. Windows requires that a file positioning call be made
         # when the file handle transitions between reads and writes. See
         # 3686fa2b8eee and the mixedfilemodewrapper in windows.py. On other
         # platforms, Python or the platform itself can be buggy. Some versions
         # of Solaris have been observed to not append at the end of the file
         # if the file was seeked to before the end. See issue4943 for more.
         #
         # We work around this issue by inserting a seek() before writing.
         # Note: This is likely not necessary on Python 3. However, because
         # the file handle is reused for reads and may be seeked there, we need
         # to be careful before changing this.
         ifh.seek(0, os.SEEK_END)
         if dfh:
             dfh.seek(0, os.SEEK_END)
 
         curr = len(self) - 1
         if not self._inline:
             transaction.add(self.datafile, offset)
             transaction.add(self.indexfile, curr * len(entry))
             if data[0]:
                 dfh.write(data[0])
             dfh.write(data[1])
             if sidedata:
                 dfh.write(sidedata)
             ifh.write(entry)
         else:
             offset += curr * self.index.entry_size
             transaction.add(self.indexfile, offset)
             ifh.write(entry)
             ifh.write(data[0])
             ifh.write(data[1])
             if sidedata:
                 ifh.write(sidedata)
             self._enforceinlinesize(transaction, ifh)
         nodemaputil.setup_persistent_nodemap(transaction, self)
 
     def addgroup(
         self,
         deltas,
         linkmapper,
         transaction,
         alwayscache=False,
         addrevisioncb=None,
         duplicaterevisioncb=None,
     ):
         """
         add a delta group
 
         given a set of deltas, add them to the revision log. the
         first delta is against its parent, which should be in our
         log, the rest are against the previous delta.
 
         If ``addrevisioncb`` is defined, it will be called with arguments of
         this revlog and the node that was added.
         """
 
         if self._writinghandles:
             raise error.ProgrammingError(b'cannot nest addgroup() calls')
 
         r = len(self)
         end = 0
         if r:
             end = self.end(r - 1)
         ifh = self._indexfp(b"a+")
         isize = r * self.index.entry_size
         if self._inline:
             transaction.add(self.indexfile, end + isize)
             dfh = None
         else:
             transaction.add(self.indexfile, isize)
             transaction.add(self.datafile, end)
             dfh = self._datafp(b"a+")
 
         def flush():
             if dfh:
                 dfh.flush()
             ifh.flush()
 
         self._writinghandles = (ifh, dfh)
         empty = True
 
         try:
             deltacomputer = deltautil.deltacomputer(self)
             # loop through our set of deltas
             for data in deltas:
                 node, p1, p2, linknode, deltabase, delta, flags, sidedata = data
                 link = linkmapper(linknode)
                 flags = flags or REVIDX_DEFAULT_FLAGS
 
                 rev = self.index.get_rev(node)
                 if rev is not None:
                     # this can happen if two branches make the same change
                     self._nodeduplicatecallback(transaction, rev)
                     if duplicaterevisioncb:
                         duplicaterevisioncb(self, rev)
                     empty = False
                     continue
 
                 for p in (p1, p2):
                     if not self.index.has_node(p):
                         raise error.LookupError(
                             p, self.indexfile, _(b'unknown parent')
                         )
 
                 if not self.index.has_node(deltabase):
                     raise error.LookupError(
                         deltabase, self.indexfile, _(b'unknown delta base')
                     )
 
                 baserev = self.rev(deltabase)
 
                 if baserev != nullrev and self.iscensored(baserev):
                     # if base is censored, delta must be full replacement in a
                     # single patch operation
                     hlen = struct.calcsize(b">lll")
                     oldlen = self.rawsize(baserev)
                     newlen = len(delta) - hlen
                     if delta[:hlen] != mdiff.replacediffheader(oldlen, newlen):
                         raise error.CensoredBaseError(
                             self.indexfile, self.node(baserev)
                         )
 
                 if not flags and self._peek_iscensored(baserev, delta, flush):
                     flags |= REVIDX_ISCENSORED
 
                 # We assume consumers of addrevisioncb will want to retrieve
                 # the added revision, which will require a call to
                 # revision(). revision() will fast path if there is a cache
                 # hit. So, we tell _addrevision() to always cache in this case.
                 # We're only using addgroup() in the context of changegroup
                 # generation so the revision data can always be handled as raw
                 # by the flagprocessor.
                 rev = self._addrevision(
                     node,
                     None,
                     transaction,
                     link,
                     p1,
                     p2,
                     flags,
                     (baserev, delta),
                     ifh,
                     dfh,
                     alwayscache=alwayscache,
                     deltacomputer=deltacomputer,
                     sidedata=sidedata,
                 )
 
                 if addrevisioncb:
                     addrevisioncb(self, rev)
                 empty = False
 
                 if not dfh and not self._inline:
                     # addrevision switched from inline to conventional
                     # reopen the index
                     ifh.close()
                     dfh = self._datafp(b"a+")
                     ifh = self._indexfp(b"a+")
                     self._writinghandles = (ifh, dfh)
         finally:
             self._writinghandles = None
 
             if dfh:
                 dfh.close()
             ifh.close()
         return not empty
 
     def iscensored(self, rev):
         """Check if a file revision is censored."""
         if not self._censorable:
             return False
 
         return self.flags(rev) & REVIDX_ISCENSORED
 
     def _peek_iscensored(self, baserev, delta, flush):
         """Quickly check if a delta produces a censored revision."""
         if not self._censorable:
             return False
 
         return storageutil.deltaiscensored(delta, baserev, self.rawsize)
 
     def getstrippoint(self, minlink):
         """find the minimum rev that must be stripped to strip the linkrev
 
         Returns a tuple containing the minimum rev and a set of all revs that
         have linkrevs that will be broken by this strip.
         """
         return storageutil.resolvestripinfo(
             minlink,
             len(self) - 1,
             self.headrevs(),
             self.linkrev,
             self.parentrevs,
         )
 
     def strip(self, minlink, transaction):
         """truncate the revlog on the first revision with a linkrev >= minlink
 
         This function is called when we're stripping revision minlink and
         its descendants from the repository.
 
         We have to remove all revisions with linkrev >= minlink, because
         the equivalent changelog revisions will be renumbered after the
         strip.
 
         So we truncate the revlog on the first of these revisions, and
         trust that the caller has saved the revisions that shouldn't be
         removed and that it'll re-add them after this truncation.
         """
         if len(self) == 0:
             return
 
         rev, _ = self.getstrippoint(minlink)
         if rev == len(self):
             return
 
         # first truncate the files on disk
         end = self.start(rev)
         if not self._inline:
             transaction.add(self.datafile, end)
             end = rev * self.index.entry_size
         else:
             end += rev * self.index.entry_size
 
         transaction.add(self.indexfile, end)
 
         # then reset internal state in memory to forget those revisions
         self._revisioncache = None
         self._chaininfocache = util.lrucachedict(500)
         self._chunkclear()
 
         del self.index[rev:-1]
 
     def checksize(self):
         """Check size of index and data files
 
         return a (dd, di) tuple.
         - dd: extra bytes for the "data" file
         - di: extra bytes for the "index" file
 
         A healthy revlog will return (0, 0).
         """
         expected = 0
         if len(self):
             expected = max(0, self.end(len(self) - 1))
 
         try:
             with self._datafp() as f:
                 f.seek(0, io.SEEK_END)
                 actual = f.tell()
             dd = actual - expected
         except IOError as inst:
             if inst.errno != errno.ENOENT:
                 raise
             dd = 0
 
         try:
             f = self.opener(self.indexfile)
             f.seek(0, io.SEEK_END)
             actual = f.tell()
             f.close()
             s = self.index.entry_size
             i = max(0, actual // s)
             di = actual - (i * s)
             if self._inline:
                 databytes = 0
                 for r in self:
                     databytes += max(0, self.length(r))
                 dd = 0
                 di = actual - len(self) * s - databytes
         except IOError as inst:
             if inst.errno != errno.ENOENT:
                 raise
             di = 0
 
         return (dd, di)
 
     def files(self):
         res = [self.indexfile]
         if not self._inline:
             res.append(self.datafile)
         return res
 
     def emitrevisions(
         self,
         nodes,
         nodesorder=None,
         revisiondata=False,
         assumehaveparentrevisions=False,
         deltamode=repository.CG_DELTAMODE_STD,
         sidedata_helpers=None,
     ):
         if nodesorder not in (b'nodes', b'storage', b'linear', None):
             raise error.ProgrammingError(
                 b'unhandled value for nodesorder: %s' % nodesorder
             )
 
         if nodesorder is None and not self._generaldelta:
             nodesorder = b'storage'
 
         if (
             not self._storedeltachains
             and deltamode != repository.CG_DELTAMODE_PREV
         ):
             deltamode = repository.CG_DELTAMODE_FULL
 
         return storageutil.emitrevisions(
             self,
             nodes,
             nodesorder,
             revlogrevisiondelta,
             deltaparentfn=self.deltaparent,
             candeltafn=self.candelta,
             rawsizefn=self.rawsize,
             revdifffn=self.revdiff,
             flagsfn=self.flags,
             deltamode=deltamode,
             revisiondata=revisiondata,
             assumehaveparentrevisions=assumehaveparentrevisions,
             sidedata_helpers=sidedata_helpers,
         )
 
     DELTAREUSEALWAYS = b'always'
     DELTAREUSESAMEREVS = b'samerevs'
     DELTAREUSENEVER = b'never'
 
     DELTAREUSEFULLADD = b'fulladd'
 
     DELTAREUSEALL = {b'always', b'samerevs', b'never', b'fulladd'}
 
     def clone(
         self,
         tr,
         destrevlog,
         addrevisioncb=None,
         deltareuse=DELTAREUSESAMEREVS,
         forcedeltabothparents=None,
         sidedatacompanion=None,
     ):
         """Copy this revlog to another, possibly with format changes.
 
         The destination revlog will contain the same revisions and nodes.
         However, it may not be bit-for-bit identical due to e.g. delta encoding
         differences.
 
         The ``deltareuse`` argument control how deltas from the existing revlog
         are preserved in the destination revlog. The argument can have the
         following values:
 
         DELTAREUSEALWAYS
            Deltas will always be reused (if possible), even if the destination
            revlog would not select the same revisions for the delta. This is the
            fastest mode of operation.
         DELTAREUSESAMEREVS
            Deltas will be reused if the destination revlog would pick the same
            revisions for the delta. This mode strikes a balance between speed
            and optimization.
         DELTAREUSENEVER
            Deltas will never be reused. This is the slowest mode of execution.
            This mode can be used to recompute deltas (e.g. if the diff/delta
            algorithm changes).
         DELTAREUSEFULLADD
            Revision will be re-added as if their were new content. This is
            slower than DELTAREUSEALWAYS but allow more mechanism to kicks in.
            eg: large file detection and handling.
 
         Delta computation can be slow, so the choice of delta reuse policy can
         significantly affect run time.
 
         The default policy (``DELTAREUSESAMEREVS``) strikes a balance between
         two extremes. Deltas will be reused if they are appropriate. But if the
         delta could choose a better revision, it will do so. This means if you
         are converting a non-generaldelta revlog to a generaldelta revlog,
         deltas will be recomputed if the delta's parent isn't a parent of the
         revision.
 
         In addition to the delta policy, the ``forcedeltabothparents``
         argument controls whether to force compute deltas against both parents
         for merges. By default, the current default is used.
 
         If not None, the `sidedatacompanion` is callable that accept two
         arguments:
 
             (srcrevlog, rev)
 
         and return a quintet that control changes to sidedata content from the
         old revision to the new clone result:
 
             (dropall, filterout, update, new_flags, dropped_flags)
 
         * if `dropall` is True, all sidedata should be dropped
         * `filterout` is a set of sidedata keys that should be dropped
         * `update` is a mapping of additionnal/new key -> value
         * new_flags is a bitfields of new flags that the revision should get
         * dropped_flags is a bitfields of new flags that the revision shoudl not longer have
         """
         if deltareuse not in self.DELTAREUSEALL:
             raise ValueError(
                 _(b'value for deltareuse invalid: %s') % deltareuse
             )
 
         if len(destrevlog):
             raise ValueError(_(b'destination revlog is not empty'))
 
         if getattr(self, 'filteredrevs', None):
             raise ValueError(_(b'source revlog has filtered revisions'))
         if getattr(destrevlog, 'filteredrevs', None):
             raise ValueError(_(b'destination revlog has filtered revisions'))
 
         # lazydelta and lazydeltabase controls whether to reuse a cached delta,
         # if possible.
         oldlazydelta = destrevlog._lazydelta
         oldlazydeltabase = destrevlog._lazydeltabase
         oldamd = destrevlog._deltabothparents
 
         try:
             if deltareuse == self.DELTAREUSEALWAYS:
                 destrevlog._lazydeltabase = True
                 destrevlog._lazydelta = True
             elif deltareuse == self.DELTAREUSESAMEREVS:
                 destrevlog._lazydeltabase = False
                 destrevlog._lazydelta = True
             elif deltareuse == self.DELTAREUSENEVER:
                 destrevlog._lazydeltabase = False
                 destrevlog._lazydelta = False
 
             destrevlog._deltabothparents = forcedeltabothparents or oldamd
 
             self._clone(
                 tr,
                 destrevlog,
                 addrevisioncb,
                 deltareuse,
                 forcedeltabothparents,
                 sidedatacompanion,
             )
 
         finally:
             destrevlog._lazydelta = oldlazydelta
             destrevlog._lazydeltabase = oldlazydeltabase
             destrevlog._deltabothparents = oldamd
 
     def _clone(
         self,
         tr,
         destrevlog,
         addrevisioncb,
         deltareuse,
         forcedeltabothparents,
         sidedatacompanion,
     ):
         """perform the core duty of `revlog.clone` after parameter processing"""
         deltacomputer = deltautil.deltacomputer(destrevlog)
         index = self.index
         for rev in self:
             entry = index[rev]
 
             # Some classes override linkrev to take filtered revs into
             # account. Use raw entry from index.
             flags = entry[0] & 0xFFFF
             linkrev = entry[4]
             p1 = index[entry[5]][7]
             p2 = index[entry[6]][7]
             node = entry[7]
 
             sidedataactions = (False, [], {}, 0, 0)
             if sidedatacompanion is not None:
                 sidedataactions = sidedatacompanion(self, rev)
 
             # (Possibly) reuse the delta from the revlog if allowed and
             # the revlog chunk is a delta.
             cachedelta = None
             rawtext = None
             if any(sidedataactions) or deltareuse == self.DELTAREUSEFULLADD:
                 dropall = sidedataactions[0]
                 filterout = sidedataactions[1]
                 update = sidedataactions[2]
                 new_flags = sidedataactions[3]
                 dropped_flags = sidedataactions[4]
                 text, sidedata = self._revisiondata(rev)
                 if dropall:
                     sidedata = {}
                 for key in filterout:
                     sidedata.pop(key, None)
                 sidedata.update(update)
                 if not sidedata:
                     sidedata = None
 
                 flags |= new_flags
                 flags &= ~dropped_flags
 
                 destrevlog.addrevision(
                     text,
                     tr,
                     linkrev,
                     p1,
                     p2,
                     cachedelta=cachedelta,
                     node=node,
                     flags=flags,
                     deltacomputer=deltacomputer,
                     sidedata=sidedata,
                 )
             else:
                 if destrevlog._lazydelta:
                     dp = self.deltaparent(rev)
                     if dp != nullrev:
                         cachedelta = (dp, bytes(self._chunk(rev)))
 
                 if not cachedelta:
                     rawtext = self.rawdata(rev)
 
                 ifh = destrevlog.opener(
                     destrevlog.indexfile, b'a+', checkambig=False
                 )
                 dfh = None
                 if not destrevlog._inline:
                     dfh = destrevlog.opener(destrevlog.datafile, b'a+')
                 try:
                     destrevlog._addrevision(
                         node,
                         rawtext,
                         tr,
                         linkrev,
                         p1,
                         p2,
                         flags,
                         cachedelta,
                         ifh,
                         dfh,
                         deltacomputer=deltacomputer,
                     )
                 finally:
                     if dfh:
                         dfh.close()
                     ifh.close()
 
             if addrevisioncb:
                 addrevisioncb(self, rev, node)
 
     def censorrevision(self, tr, censornode, tombstone=b''):
         if (self.version & 0xFFFF) == REVLOGV0:
             raise error.RevlogError(
                 _(b'cannot censor with version %d revlogs') % self.version
             )
 
         censorrev = self.rev(censornode)
         tombstone = storageutil.packmeta({b'censored': tombstone}, b'')
 
         if len(tombstone) > self.rawsize(censorrev):
             raise error.Abort(
                 _(b'censor tombstone must be no longer than censored data')
             )
 
         # Rewriting the revlog in place is hard. Our strategy for censoring is
         # to create a new revlog, copy all revisions to it, then replace the
         # revlogs on transaction close.
 
         newindexfile = self.indexfile + b'.tmpcensored'
         newdatafile = self.datafile + b'.tmpcensored'
 
         # This is a bit dangerous. We could easily have a mismatch of state.
         newrl = revlog(
             self.opener,
             target=self.target,
             indexfile=newindexfile,
             datafile=newdatafile,
             censorable=True,
         )
         newrl.version = self.version
         newrl._generaldelta = self._generaldelta
         newrl._parse_index = self._parse_index
 
         for rev in self.revs():
             node = self.node(rev)
             p1, p2 = self.parents(node)
 
             if rev == censorrev:
                 newrl.addrawrevision(
                     tombstone,
                     tr,
                     self.linkrev(censorrev),
                     p1,
                     p2,
                     censornode,
                     REVIDX_ISCENSORED,
                 )
 
                 if newrl.deltaparent(rev) != nullrev:
                     raise error.Abort(
                         _(
                             b'censored revision stored as delta; '
                             b'cannot censor'
                         ),
                         hint=_(
                             b'censoring of revlogs is not '
                             b'fully implemented; please report '
                             b'this bug'
                         ),
                     )
                 continue
 
             if self.iscensored(rev):
                 if self.deltaparent(rev) != nullrev:
                     raise error.Abort(
                         _(
                             b'cannot censor due to censored '
                             b'revision having delta stored'
                         )
                     )
                 rawtext = self._chunk(rev)
             else:
                 rawtext = self.rawdata(rev)
 
             newrl.addrawrevision(
                 rawtext, tr, self.linkrev(rev), p1, p2, node, self.flags(rev)
             )
 
         tr.addbackup(self.indexfile, location=b'store')
         if not self._inline:
             tr.addbackup(self.datafile, location=b'store')
 
         self.opener.rename(newrl.indexfile, self.indexfile)
         if not self._inline:
             self.opener.rename(newrl.datafile, self.datafile)
 
         self.clearcaches()
         self._loadindex()
 
     def verifyintegrity(self, state):
         """Verifies the integrity of the revlog.
 
         Yields ``revlogproblem`` instances describing problems that are
         found.
         """
         dd, di = self.checksize()
         if dd:
             yield revlogproblem(error=_(b'data length off by %d bytes') % dd)
         if di:
             yield revlogproblem(error=_(b'index contains %d extra bytes') % di)
 
         version = self.version & 0xFFFF
 
         # The verifier tells us what version revlog we should be.
         if version != state[b'expectedversion']:
             yield revlogproblem(
                 warning=_(b"warning: '%s' uses revlog format %d; expected %d")
                 % (self.indexfile, version, state[b'expectedversion'])
             )
 
         state[b'skipread'] = set()
         state[b'safe_renamed'] = set()
 
         for rev in self:
             node = self.node(rev)
 
             # Verify contents. 4 cases to care about:
             #
             #   common: the most common case
             #   rename: with a rename
             #   meta: file content starts with b'\1\n', the metadata
             #         header defined in filelog.py, but without a rename
             #   ext: content stored externally
             #
             # More formally, their differences are shown below:
             #
             #                       | common | rename | meta  | ext
             #  -------------------------------------------------------
             #   flags()             | 0      | 0      | 0     | not 0
             #   renamed()           | False  | True   | False | ?
             #   rawtext[0:2]=='\1\n'| False  | True   | True  | ?
             #
             # "rawtext" means the raw text stored in revlog data, which
             # could be retrieved by "rawdata(rev)". "text"
             # mentioned below is "revision(rev)".
             #
             # There are 3 different lengths stored physically:
             #  1. L1: rawsize, stored in revlog index
             #  2. L2: len(rawtext), stored in revlog data
             #  3. L3: len(text), stored in revlog data if flags==0, or
             #     possibly somewhere else if flags!=0
             #
             # L1 should be equal to L2. L3 could be different from them.
             # "text" may or may not affect commit hash depending on flag
             # processors (see flagutil.addflagprocessor).
             #
             #              | common  | rename | meta  | ext
             # -------------------------------------------------
             #    rawsize() | L1      | L1     | L1    | L1
             #       size() | L1      | L2-LM  | L1(*) | L1 (?)
             # len(rawtext) | L2      | L2     | L2    | L2
             #    len(text) | L2      | L2     | L2    | L3
             #  len(read()) | L2      | L2-LM  | L2-LM | L3 (?)
             #
             # LM:  length of metadata, depending on rawtext
             # (*): not ideal, see comment in filelog.size
             # (?): could be "- len(meta)" if the resolved content has
             #      rename metadata
             #
             # Checks needed to be done:
             #  1. length check: L1 == L2, in all cases.
             #  2. hash check: depending on flag processor, we may need to
             #     use either "text" (external), or "rawtext" (in revlog).
 
             try:
                 skipflags = state.get(b'skipflags', 0)
                 if skipflags:
                     skipflags &= self.flags(rev)
 
                 _verify_revision(self, skipflags, state, node)
 
                 l1 = self.rawsize(rev)
                 l2 = len(self.rawdata(node))
 
                 if l1 != l2:
                     yield revlogproblem(
                         error=_(b'unpacked size is %d, %d expected') % (l2, l1),
                         node=node,
                     )
 
             except error.CensoredNodeError:
                 if state[b'erroroncensored']:
                     yield revlogproblem(
                         error=_(b'censored file data'), node=node
                     )
                     state[b'skipread'].add(node)
             except Exception as e:
                 yield revlogproblem(
                     error=_(b'unpacking %s: %s')
                     % (short(node), stringutil.forcebytestr(e)),
                     node=node,
                 )
                 state[b'skipread'].add(node)
 
     def storageinfo(
         self,
         exclusivefiles=False,
         sharedfiles=False,
         revisionscount=False,
         trackedsize=False,
         storedsize=False,
     ):
         d = {}
 
         if exclusivefiles:
             d[b'exclusivefiles'] = [(self.opener, self.indexfile)]
             if not self._inline:
                 d[b'exclusivefiles'].append((self.opener, self.datafile))
 
         if sharedfiles:
             d[b'sharedfiles'] = []
 
         if revisionscount:
             d[b'revisionscount'] = len(self)
 
         if trackedsize:
             d[b'trackedsize'] = sum(map(self.rawsize, iter(self)))
 
         if storedsize:
             d[b'storedsize'] = sum(
                 self.opener.stat(path).st_size for path in self.files()
             )
 
         return d
 
     def rewrite_sidedata(self, helpers, startrev, endrev):
         if self.version & 0xFFFF != REVLOGV2:
             return
         # inline are not yet supported because they suffer from an issue when
         # rewriting them (since it's not an append-only operation).
         # See issue6485.
         assert not self._inline
         if not helpers[1] and not helpers[2]:
             # Nothing to generate or remove
             return
 
         new_entries = []
         # append the new sidedata
         with self._datafp(b'a+') as fp:
             # Maybe this bug still exists, see revlog._writeentry
             fp.seek(0, os.SEEK_END)
             current_offset = fp.tell()
             for rev in range(startrev, endrev + 1):
                 entry = self.index[rev]
-                new_sidedata = storageutil.run_sidedata_helpers(
+                new_sidedata, flags = storageutil.run_sidedata_helpers(
                     store=self,
                     sidedata_helpers=helpers,
                     sidedata={},
                     rev=rev,
                 )
 
                 serialized_sidedata = sidedatautil.serialize_sidedata(
                     new_sidedata
                 )
                 if entry[8] != 0 or entry[9] != 0:
                     # rewriting entries that already have sidedata is not
                     # supported yet, because it introduces garbage data in the
                     # revlog.
                     msg = b"Rewriting existing sidedata is not supported yet"
                     raise error.Abort(msg)
-                entry = entry[:8]
+
+                # Apply (potential) flags to add and to remove after running
+                # the sidedata helpers
+                new_offset_flags = entry[0] | flags[0] & ~flags[1]
+                entry = (new_offset_flags,) + entry[1:8]
                 entry += (current_offset, len(serialized_sidedata))
 
                 fp.write(serialized_sidedata)
                 new_entries.append(entry)
                 current_offset += len(serialized_sidedata)
 
         # rewrite the new index entries
         with self._indexfp(b'w+') as fp:
             fp.seek(startrev * self.index.entry_size)
-            for i, entry in enumerate(new_entries):
+            for i, e in enumerate(new_entries):
                 rev = startrev + i
-                self.index.replace_sidedata_info(rev, entry[8], entry[9])
+                self.index.replace_sidedata_info(rev, e[8], e[9], e[0])
                 packed = self.index.entry_binary(rev)
                 if rev == 0:
                     header = self.index.pack_header(self.version)
                     packed = header + packed
                 fp.write(packed)

mercurial/utils/storageutil.py

@@ -1,572 +1,585 b''
 # storageutil.py - Storage functionality agnostic of backend implementation.
 #
 # Copyright 2018 Gregory Szorc <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 import re
 import struct
 
 from ..i18n import _
 from ..node import (
     bin,
     nullrev,
     sha1nodeconstants,
 )
 from .. import (
     dagop,
     error,
     mdiff,
     pycompat,
 )
 from ..interfaces import repository
 from ..revlogutils import sidedata as sidedatamod
 from ..utils import hashutil
 
 _nullhash = hashutil.sha1(sha1nodeconstants.nullid)
 
 # revision data contains extra metadata not part of the official digest
 # Only used in changegroup >= v4.
 CG_FLAG_SIDEDATA = 1
 
 
 def hashrevisionsha1(text, p1, p2):
     """Compute the SHA-1 for revision data and its parents.
 
     This hash combines both the current file contents and its history
     in a manner that makes it easy to distinguish nodes with the same
     content in the revision graph.
     """
     # As of now, if one of the parent node is null, p2 is null
     if p2 == sha1nodeconstants.nullid:
         # deep copy of a hash is faster than creating one
         s = _nullhash.copy()
         s.update(p1)
     else:
         # none of the parent nodes are nullid
         if p1 < p2:
             a = p1
             b = p2
         else:
             a = p2
             b = p1
         s = hashutil.sha1(a)
         s.update(b)
     s.update(text)
     return s.digest()
 
 
 METADATA_RE = re.compile(b'\x01\n')
 
 
 def parsemeta(text):
     """Parse metadata header from revision data.
 
     Returns a 2-tuple of (metadata, offset), where both can be None if there
     is no metadata.
     """
     # text can be buffer, so we can't use .startswith or .index
     if text[:2] != b'\x01\n':
         return None, None
     s = METADATA_RE.search(text, 2).start()
     mtext = text[2:s]
     meta = {}
     for l in mtext.splitlines():
         k, v = l.split(b': ', 1)
         meta[k] = v
     return meta, s + 2
 
 
 def packmeta(meta, text):
     """Add metadata to fulltext to produce revision text."""
     keys = sorted(meta)
     metatext = b''.join(b'%s: %s\n' % (k, meta[k]) for k in keys)
     return b'\x01\n%s\x01\n%s' % (metatext, text)
 
 
 def iscensoredtext(text):
     meta = parsemeta(text)[0]
     return meta and b'censored' in meta
 
 
 def filtermetadata(text):
     """Extract just the revision data from source text.
 
     Returns ``text`` unless it has a metadata header, in which case we return
     a new buffer without hte metadata.
     """
     if not text.startswith(b'\x01\n'):
         return text
 
     offset = text.index(b'\x01\n', 2)
     return text[offset + 2 :]
 
 
 def filerevisioncopied(store, node):
     """Resolve file revision copy metadata.
 
     Returns ``False`` if the file has no copy metadata. Otherwise a
     2-tuple of the source filename and node.
     """
     if store.parents(node)[0] != sha1nodeconstants.nullid:
         return False
 
     meta = parsemeta(store.revision(node))[0]
 
     # copy and copyrev occur in pairs. In rare cases due to old bugs,
     # one can occur without the other. So ensure both are present to flag
     # as a copy.
     if meta and b'copy' in meta and b'copyrev' in meta:
         return meta[b'copy'], bin(meta[b'copyrev'])
 
     return False
 
 
 def filedataequivalent(store, node, filedata):
     """Determines whether file data is equivalent to a stored node.
 
     Returns True if the passed file data would hash to the same value
     as a stored revision and False otherwise.
 
     When a stored revision is censored, filedata must be empty to have
     equivalence.
 
     When a stored revision has copy metadata, it is ignored as part
     of the compare.
     """
 
     if filedata.startswith(b'\x01\n'):
         revisiontext = b'\x01\n\x01\n' + filedata
     else:
         revisiontext = filedata
 
     p1, p2 = store.parents(node)
 
     computednode = hashrevisionsha1(revisiontext, p1, p2)
 
     if computednode == node:
         return True
 
     # Censored files compare against the empty file.
     if store.iscensored(store.rev(node)):
         return filedata == b''
 
     # Renaming a file produces a different hash, even if the data
     # remains unchanged. Check if that's the case.
     if store.renamed(node):
         return store.read(node) == filedata
 
     return False
 
 
 def iterrevs(storelen, start=0, stop=None):
     """Iterate over revision numbers in a store."""
     step = 1
 
     if stop is not None:
         if start > stop:
             step = -1
         stop += step
         if stop > storelen:
             stop = storelen
     else:
         stop = storelen
 
     return pycompat.xrange(start, stop, step)
 
 
 def fileidlookup(store, fileid, identifier):
     """Resolve the file node for a value.
 
     ``store`` is an object implementing the ``ifileindex`` interface.
 
     ``fileid`` can be:
 
     * A 20 or 32 byte binary node.
     * An integer revision number
     * A 40 or 64 byte hex node.
     * A bytes that can be parsed as an integer representing a revision number.
 
     ``identifier`` is used to populate ``error.LookupError`` with an identifier
     for the store.
 
     Raises ``error.LookupError`` on failure.
     """
     if isinstance(fileid, int):
         try:
             return store.node(fileid)
         except IndexError:
             raise error.LookupError(
                 b'%d' % fileid, identifier, _(b'no match found')
             )
 
     if len(fileid) in (20, 32):
         try:
             store.rev(fileid)
             return fileid
         except error.LookupError:
             pass
 
     if len(fileid) in (40, 64):
         try:
             rawnode = bin(fileid)
             store.rev(rawnode)
             return rawnode
         except TypeError:
             pass
 
     try:
         rev = int(fileid)
 
         if b'%d' % rev != fileid:
             raise ValueError
 
         try:
             return store.node(rev)
         except (IndexError, TypeError):
             pass
     except (ValueError, OverflowError):
         pass
 
     raise error.LookupError(fileid, identifier, _(b'no match found'))
 
 
 def resolvestripinfo(minlinkrev, tiprev, headrevs, linkrevfn, parentrevsfn):
     """Resolve information needed to strip revisions.
 
     Finds the minimum revision number that must be stripped in order to
     strip ``minlinkrev``.
 
     Returns a 2-tuple of the minimum revision number to do that and a set
     of all revision numbers that have linkrevs that would be broken
     by that strip.
 
     ``tiprev`` is the current tip-most revision. It is ``len(store) - 1``.
     ``headrevs`` is an iterable of head revisions.
     ``linkrevfn`` is a callable that receives a revision and returns a linked
     revision.
     ``parentrevsfn`` is a callable that receives a revision number and returns
     an iterable of its parent revision numbers.
     """
     brokenrevs = set()
     strippoint = tiprev + 1
 
     heads = {}
     futurelargelinkrevs = set()
     for head in headrevs:
         headlinkrev = linkrevfn(head)
         heads[head] = headlinkrev
         if headlinkrev >= minlinkrev:
             futurelargelinkrevs.add(headlinkrev)
 
     # This algorithm involves walking down the rev graph, starting at the
     # heads. Since the revs are topologically sorted according to linkrev,
     # once all head linkrevs are below the minlink, we know there are
     # no more revs that could have a linkrev greater than minlink.
     # So we can stop walking.
     while futurelargelinkrevs:
         strippoint -= 1
         linkrev = heads.pop(strippoint)
 
         if linkrev < minlinkrev:
             brokenrevs.add(strippoint)
         else:
             futurelargelinkrevs.remove(linkrev)
 
         for p in parentrevsfn(strippoint):
             if p != nullrev:
                 plinkrev = linkrevfn(p)
                 heads[p] = plinkrev
                 if plinkrev >= minlinkrev:
                     futurelargelinkrevs.add(plinkrev)
 
     return strippoint, brokenrevs
 
 
 def emitrevisions(
     store,
     nodes,
     nodesorder,
     resultcls,
     deltaparentfn=None,
     candeltafn=None,
     rawsizefn=None,
     revdifffn=None,
     flagsfn=None,
     deltamode=repository.CG_DELTAMODE_STD,
     revisiondata=False,
     assumehaveparentrevisions=False,
     sidedata_helpers=None,
 ):
     """Generic implementation of ifiledata.emitrevisions().
 
     Emitting revision data is subtly complex. This function attempts to
     encapsulate all the logic for doing so in a backend-agnostic way.
 
     ``store``
        Object conforming to ``ifilestorage`` interface.
 
     ``nodes``
        List of revision nodes whose data to emit.
 
     ``resultcls``
        A type implementing the ``irevisiondelta`` interface that will be
        constructed and returned.
 
     ``deltaparentfn`` (optional)
        Callable receiving a revision number and returning the revision number
        of a revision that the internal delta is stored against. This delta
        will be preferred over computing a new arbitrary delta.
 
        If not defined, a delta will always be computed from raw revision
        data.
 
     ``candeltafn`` (optional)
        Callable receiving a pair of revision numbers that returns a bool
        indicating whether a delta between them can be produced.
 
        If not defined, it is assumed that any two revisions can delta with
        each other.
 
     ``rawsizefn`` (optional)
        Callable receiving a revision number and returning the length of the
        ``store.rawdata(rev)``.
 
        If not defined, ``len(store.rawdata(rev))`` will be called.
 
     ``revdifffn`` (optional)
        Callable receiving a pair of revision numbers that returns a delta
        between them.
 
        If not defined, a delta will be computed by invoking mdiff code
        on ``store.revision()`` results.
 
        Defining this function allows a precomputed or stored delta to be
        used without having to compute on.
 
     ``flagsfn`` (optional)
        Callable receiving a revision number and returns the integer flags
        value for it. If not defined, flags value will be 0.
 
     ``deltamode``
        constaint on delta to be sent:
        * CG_DELTAMODE_STD  - normal mode, try to reuse storage deltas,
        * CG_DELTAMODE_PREV - only delta against "prev",
        * CG_DELTAMODE_FULL - only issue full snapshot.
 
        Whether to send fulltext revisions instead of deltas, if allowed.
 
     ``nodesorder``
     ``revisiondata``
     ``assumehaveparentrevisions``
     ``sidedata_helpers`` (optional)
         If not None, means that sidedata should be included.
         A dictionary of revlog type to tuples of `(repo, computers, removers)`:
             * `repo` is used as an argument for computers
-            * `computers` is a list of `(category, (keys, computer)` that
+            * `computers` is a list of `(category, (keys, computer, flags)` that
                compute the missing sidedata categories that were asked:
                * `category` is the sidedata category
                * `keys` are the sidedata keys to be affected
+               * `flags` is a bitmask (an integer) of flags to remove when
+                  removing the category.
                * `computer` is the function `(repo, store, rev, sidedata)` that
-                 returns a new sidedata dict.
+                 returns a tuple of
+                 `(new sidedata dict, (flags to add, flags to remove))`.
+                 For example, it will return `({}, (0, 1 << 15))` to return no
+                 sidedata, with no flags to add and one flag to remove.
             * `removers` will remove the keys corresponding to the categories
               that are present, but not needed.
         If both `computers` and `removers` are empty, sidedata are simply not
         transformed.
         Revlog types are `changelog`, `manifest` or `filelog`.
     """
 
     fnode = store.node
     frev = store.rev
 
     if nodesorder == b'nodes':
         revs = [frev(n) for n in nodes]
     elif nodesorder == b'linear':
         revs = {frev(n) for n in nodes}
         revs = dagop.linearize(revs, store.parentrevs)
     else:  # storage and default
         revs = sorted(frev(n) for n in nodes)
 
     prevrev = None
 
     if deltamode == repository.CG_DELTAMODE_PREV or assumehaveparentrevisions:
         prevrev = store.parentrevs(revs[0])[0]
 
     # Set of revs available to delta against.
     available = set()
 
     for rev in revs:
         if rev == nullrev:
             continue
 
         node = fnode(rev)
         p1rev, p2rev = store.parentrevs(rev)
 
         if deltaparentfn:
             deltaparentrev = deltaparentfn(rev)
         else:
             deltaparentrev = nullrev
 
         # Forced delta against previous mode.
         if deltamode == repository.CG_DELTAMODE_PREV:
             baserev = prevrev
 
         # We're instructed to send fulltext. Honor that.
         elif deltamode == repository.CG_DELTAMODE_FULL:
             baserev = nullrev
         # We're instructed to use p1. Honor that
         elif deltamode == repository.CG_DELTAMODE_P1:
             baserev = p1rev
 
         # There is a delta in storage. We try to use that because it
         # amounts to effectively copying data from storage and is
         # therefore the fastest.
         elif deltaparentrev != nullrev:
             # Base revision was already emitted in this group. We can
             # always safely use the delta.
             if deltaparentrev in available:
                 baserev = deltaparentrev
 
             # Base revision is a parent that hasn't been emitted already.
             # Use it if we can assume the receiver has the parent revision.
             elif assumehaveparentrevisions and deltaparentrev in (p1rev, p2rev):
                 baserev = deltaparentrev
 
             # No guarantee the receiver has the delta parent. Send delta
             # against last revision (if possible), which in the common case
             # should be similar enough to this revision that the delta is
             # reasonable.
             elif prevrev is not None:
                 baserev = prevrev
             else:
                 baserev = nullrev
 
         # Storage has a fulltext revision.
 
         # Let's use the previous revision, which is as good a guess as any.
         # There is definitely room to improve this logic.
         elif prevrev is not None:
             baserev = prevrev
         else:
             baserev = nullrev
 
         # But we can't actually use our chosen delta base for whatever
         # reason. Reset to fulltext.
         if baserev != nullrev and (candeltafn and not candeltafn(baserev, rev)):
             baserev = nullrev
 
         revision = None
         delta = None
         baserevisionsize = None
 
         if revisiondata:
             if store.iscensored(baserev) or store.iscensored(rev):
                 try:
                     revision = store.rawdata(node)
                 except error.CensoredNodeError as e:
                     revision = e.tombstone
 
                 if baserev != nullrev:
                     if rawsizefn:
                         baserevisionsize = rawsizefn(baserev)
                     else:
                         baserevisionsize = len(store.rawdata(baserev))
 
             elif (
                 baserev == nullrev and deltamode != repository.CG_DELTAMODE_PREV
             ):
                 revision = store.rawdata(node)
                 available.add(rev)
             else:
                 if revdifffn:
                     delta = revdifffn(baserev, rev)
                 else:
                     delta = mdiff.textdiff(
                         store.rawdata(baserev), store.rawdata(rev)
                     )
 
                 available.add(rev)
 
         serialized_sidedata = None
+        sidedata_flags = (0, 0)
         if sidedata_helpers:
-            sidedata = store.sidedata(rev)
-            sidedata = run_sidedata_helpers(
+            old_sidedata = store.sidedata(rev)
+            sidedata, sidedata_flags = run_sidedata_helpers(
                 store=store,
                 sidedata_helpers=sidedata_helpers,
-                sidedata=sidedata,
+                sidedata=old_sidedata,
                 rev=rev,
             )
             if sidedata:
                 serialized_sidedata = sidedatamod.serialize_sidedata(sidedata)
 
         flags = flagsfn(rev) if flagsfn else 0
         protocol_flags = 0
         if serialized_sidedata:
             # Advertise that sidedata exists to the other side
             protocol_flags |= CG_FLAG_SIDEDATA
+            # Computers and removers can return flags to add and/or remove
+            flags = flags | sidedata_flags[0] & ~sidedata_flags[1]
 
         yield resultcls(
             node=node,
             p1node=fnode(p1rev),
             p2node=fnode(p2rev),
             basenode=fnode(baserev),
             flags=flags,
             baserevisionsize=baserevisionsize,
             revision=revision,
             delta=delta,
             sidedata=serialized_sidedata,
             protocol_flags=protocol_flags,
         )
 
         prevrev = rev
 
 
 def run_sidedata_helpers(store, sidedata_helpers, sidedata, rev):
     """Returns the sidedata for the given revision after running through
     the given helpers.
     - `store`: the revlog this applies to (changelog, manifest, or filelog
       instance)
     - `sidedata_helpers`: see `storageutil.emitrevisions`
     - `sidedata`: previous sidedata at the given rev, if any
     - `rev`: affected rev of `store`
     """
     repo, sd_computers, sd_removers = sidedata_helpers
     kind = store.revlog_kind
-    for _keys, sd_computer in sd_computers.get(kind, []):
-        sidedata = sd_computer(repo, store, rev, sidedata)
-    for keys, _computer in sd_removers.get(kind, []):
+    flags_to_add = 0
+    flags_to_remove = 0
+    for _keys, sd_computer, _flags in sd_computers.get(kind, []):
+        sidedata, flags = sd_computer(repo, store, rev, sidedata)
+        flags_to_add |= flags[0]
+        flags_to_remove |= flags[1]
+    for keys, _computer, flags in sd_removers.get(kind, []):
         for key in keys:
             sidedata.pop(key, None)
-    return sidedata
+        flags_to_remove |= flags
+    return sidedata, (flags_to_add, flags_to_remove)
 
 
 def deltaiscensored(delta, baserev, baselenfn):
     """Determine if a delta represents censored revision data.
 
     ``baserev`` is the base revision this delta is encoded against.
     ``baselenfn`` is a callable receiving a revision number that resolves the
     length of the revision fulltext.
 
     Returns a bool indicating if the result of the delta represents a censored
     revision.
     """
     # Fragile heuristic: unless new file meta keys are added alphabetically
     # preceding "censored", all censored revisions are prefixed by
     # "\1\ncensored:". A delta producing such a censored revision must be a
     # full-replacement delta, so we inspect the first and only patch in the
     # delta for this prefix.
     hlen = struct.calcsize(b">lll")
     if len(delta) <= hlen:
         return False
 
     oldlen = baselenfn(baserev)
     newlen = len(delta) - hlen
     if delta[:hlen] != mdiff.replacediffheader(oldlen, newlen):
         return False
 
     add = b"\1\ncensored:"
     addlen = len(add)
     return newlen >= addlen and delta[hlen : hlen + addlen] == add

tests/testlib/ext-sidedata-2.py

@@ -1,51 +1,55 b''
 # coding: utf8
 # ext-sidedata-2.py - small extension to test (differently) the sidedata logic
 #
 # Simulates a client for a complex sidedata exchange.
 #
 # Copyright 2021 Raphaël Gomès <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 import hashlib
 import struct
 
 from mercurial.revlogutils import sidedata as sidedatamod
 from mercurial.revlogutils import constants
 
+NO_FLAGS = (0, 0)  # hoot
+
 
 def compute_sidedata_1(repo, revlog, rev, sidedata, text=None):
     sidedata = sidedata.copy()
     if text is None:
         text = revlog.revision(rev)
     sidedata[sidedatamod.SD_TEST1] = struct.pack('>I', len(text))
-    return sidedata
+    return sidedata, NO_FLAGS
 
 
 def compute_sidedata_2(repo, revlog, rev, sidedata, text=None):
     sidedata = sidedata.copy()
     if text is None:
         text = revlog.revision(rev)
     sha256 = hashlib.sha256(text).digest()
     sidedata[sidedatamod.SD_TEST2] = struct.pack('>32s', sha256)
-    return sidedata
+    return sidedata, NO_FLAGS
 
 
 def reposetup(ui, repo):
     # Sidedata keys happen to be the same as the categories, easier for testing.
     for kind in constants.ALL_KINDS:
         repo.register_sidedata_computer(
             kind,
             sidedatamod.SD_TEST1,
             (sidedatamod.SD_TEST1,),
             compute_sidedata_1,
+            0,
         )
         repo.register_sidedata_computer(
             kind,
             sidedatamod.SD_TEST2,
             (sidedatamod.SD_TEST2,),
             compute_sidedata_2,
+            0,
         )

tests/testlib/ext-sidedata-3.py

@@ -1,89 +1,94 b''
 # coding: utf8
 # ext-sidedata-3.py - small extension to test (differently still) the sidedata
 # logic
 #
 # Simulates a client for a complex sidedata exchange.
 #
 # Copyright 2021 Raphaël Gomès <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 import hashlib
 import struct
 
 from mercurial import (
     extensions,
     revlog,
 )
 
 from mercurial.revlogutils import sidedata as sidedatamod
 from mercurial.revlogutils import constants
 
+NO_FLAGS = (0, 0)
+
 
 def compute_sidedata_1(repo, revlog, rev, sidedata, text=None):
     sidedata = sidedata.copy()
     if text is None:
         text = revlog.revision(rev)
     sidedata[sidedatamod.SD_TEST1] = struct.pack('>I', len(text))
-    return sidedata
+    return sidedata, NO_FLAGS
 
 
 def compute_sidedata_2(repo, revlog, rev, sidedata, text=None):
     sidedata = sidedata.copy()
     if text is None:
         text = revlog.revision(rev)
     sha256 = hashlib.sha256(text).digest()
     sidedata[sidedatamod.SD_TEST2] = struct.pack('>32s', sha256)
-    return sidedata
+    return sidedata, NO_FLAGS
 
 
 def compute_sidedata_3(repo, revlog, rev, sidedata, text=None):
     sidedata = sidedata.copy()
     if text is None:
         text = revlog.revision(rev)
     sha384 = hashlib.sha384(text).digest()
     sidedata[sidedatamod.SD_TEST3] = struct.pack('>48s', sha384)
-    return sidedata
+    return sidedata, NO_FLAGS
 
 
 def wrapaddrevision(
     orig, self, text, transaction, link, p1, p2, *args, **kwargs
 ):
     if kwargs.get('sidedata') is None:
         kwargs['sidedata'] = {}
     sd = kwargs['sidedata']
-    sd = compute_sidedata_1(None, self, None, sd, text=text)
-    kwargs['sidedata'] = compute_sidedata_2(None, self, None, sd, text=text)
+    sd, flags = compute_sidedata_1(None, self, None, sd, text=text)
+    kwargs['sidedata'] = compute_sidedata_2(None, self, None, sd, text=text)[0]
     return orig(self, text, transaction, link, p1, p2, *args, **kwargs)
 
 
 def extsetup(ui):
     extensions.wrapfunction(revlog.revlog, 'addrevision', wrapaddrevision)
 
 
 def reposetup(ui, repo):
     # Sidedata keys happen to be the same as the categories, easier for testing.
     for kind in constants.ALL_KINDS:
         repo.register_sidedata_computer(
             kind,
             sidedatamod.SD_TEST1,
             (sidedatamod.SD_TEST1,),
             compute_sidedata_1,
+            0,
         )
         repo.register_sidedata_computer(
             kind,
             sidedatamod.SD_TEST2,
             (sidedatamod.SD_TEST2,),
             compute_sidedata_2,
+            0,
         )
         repo.register_sidedata_computer(
             kind,
             sidedatamod.SD_TEST3,
             (sidedatamod.SD_TEST3,),
             compute_sidedata_3,
+            0,
         )
     repo.register_wanted_sidedata(sidedatamod.SD_TEST1)
     repo.register_wanted_sidedata(sidedatamod.SD_TEST2)

tests/testlib/ext-sidedata-5.py

@@ -1,82 +1,86 b''
 # coding: utf8
 # ext-sidedata-5.py - small extension to test (differently still) the sidedata
 # logic
 #
 # Simulates a server for a simple sidedata exchange.
 #
 # Copyright 2021 Raphaël Gomès <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 import hashlib
 import struct
 
 from mercurial import (
     extensions,
     revlog,
 )
 
 
 from mercurial.revlogutils import sidedata as sidedatamod
 from mercurial.revlogutils import constants
 
+NO_FLAGS = (0, 0)
+
 
 def compute_sidedata_1(repo, revlog, rev, sidedata, text=None):
     sidedata = sidedata.copy()
     if text is None:
         text = revlog.revision(rev)
     sidedata[sidedatamod.SD_TEST1] = struct.pack('>I', len(text))
-    return sidedata
+    return sidedata, NO_FLAGS
 
 
 def compute_sidedata_2(repo, revlog, rev, sidedata, text=None):
     sidedata = sidedata.copy()
     if text is None:
         text = revlog.revision(rev)
     sha256 = hashlib.sha256(text).digest()
     sidedata[sidedatamod.SD_TEST2] = struct.pack('>32s', sha256)
-    return sidedata
+    return sidedata, NO_FLAGS
 
 
 def reposetup(ui, repo):
     # Sidedata keys happen to be the same as the categories, easier for testing.
     for kind in constants.ALL_KINDS:
         repo.register_sidedata_computer(
             kind,
             sidedatamod.SD_TEST1,
             (sidedatamod.SD_TEST1,),
             compute_sidedata_1,
+            0,
         )
         repo.register_sidedata_computer(
             kind,
             sidedatamod.SD_TEST2,
             (sidedatamod.SD_TEST2,),
             compute_sidedata_2,
+            0,
         )
 
     # We don't register sidedata computers because we don't care within these
     # tests
     repo.register_wanted_sidedata(sidedatamod.SD_TEST1)
     repo.register_wanted_sidedata(sidedatamod.SD_TEST2)
 
 
 def wrapaddrevision(
     orig, self, text, transaction, link, p1, p2, *args, **kwargs
 ):
     if kwargs.get('sidedata') is None:
         kwargs['sidedata'] = {}
     sd = kwargs['sidedata']
     ## let's store some arbitrary data just for testing
     # text length
     sd[sidedatamod.SD_TEST1] = struct.pack('>I', len(text))
     # and sha2 hashes
     sha256 = hashlib.sha256(text).digest()
     sd[sidedatamod.SD_TEST2] = struct.pack('>32s', sha256)
     return orig(self, text, transaction, link, p1, p2, *args, **kwargs)
 
 
 def extsetup(ui):
     extensions.wrapfunction(revlog.revlog, 'addrevision', wrapaddrevision)