parsers: use base-16 trie for faster node->rev mapping

This greatly speeds up node->rev lookups, with results that are
often user-perceptible: for instance, "hg --time log" of the node
associated with rev 1000 on a linux-2.6 repo improves from 0.3
seconds to 0.03.  I have not found any instances of slowdowns.

The new perfnodelookup command in contrib/perf.py demonstrates the
speedup more dramatically, since it performs no I/O.  For a single
lookup, the new code is about 40x faster.

These changes also prepare the ground for the possibility of further
improving the performance of prefix-based node lookups.

contrib/perf.py

@@ -1,206 +1,227 b''
 # perf.py - performance test routines
 '''helper extension to measure performance'''
 
-from mercurial import cmdutil, scmutil, match, commands
+from mercurial import cmdutil, scmutil, util, match, commands
 import time, os, sys
 
 def timer(func, title=None):
     results = []
     begin = time.time()
     count = 0
     while True:
         ostart = os.times()
         cstart = time.time()
         r = func()
         cstop = time.time()
         ostop = os.times()
         count += 1
         a, b = ostart, ostop
         results.append((cstop - cstart, b[0] - a[0], b[1]-a[1]))
         if cstop - begin > 3 and count >= 100:
             break
         if cstop - begin > 10 and count >= 3:
             break
     if title:
         sys.stderr.write("! %s\n" % title)
     if r:
         sys.stderr.write("! result: %s\n" % r)
     m = min(results)
     sys.stderr.write("! wall %f comb %f user %f sys %f (best of %d)\n"
                      % (m[0], m[1] + m[2], m[1], m[2], count))
 
 def perfwalk(ui, repo, *pats):
     try:
         m = scmutil.match(repo[None], pats, {})
         timer(lambda: len(list(repo.dirstate.walk(m, [], True, False))))
     except:
         try:
             m = scmutil.match(repo[None], pats, {})
             timer(lambda: len([b for a, b, c in repo.dirstate.statwalk([], m)]))
         except:
             timer(lambda: len(list(cmdutil.walk(repo, pats, {}))))
 
 def perfstatus(ui, repo, *pats):
     #m = match.always(repo.root, repo.getcwd())
     #timer(lambda: sum(map(len, repo.dirstate.status(m, [], False, False, False))))
     timer(lambda: sum(map(len, repo.status())))
 
 def perfheads(ui, repo):
     timer(lambda: len(repo.changelog.headrevs()))
 
 def perftags(ui, repo):
     import mercurial.changelog, mercurial.manifest
     def t():
         repo.changelog = mercurial.changelog.changelog(repo.sopener)
         repo.manifest = mercurial.manifest.manifest(repo.sopener)
         repo._tags = None
         return len(repo.tags())
     timer(t)
 
 def perfdirstate(ui, repo):
     "a" in repo.dirstate
     def d():
         repo.dirstate.invalidate()
         "a" in repo.dirstate
     timer(d)
 
 def perfdirstatedirs(ui, repo):
     "a" in repo.dirstate
     def d():
         "a" in repo.dirstate._dirs
         del repo.dirstate._dirs
     timer(d)
 
 def perfmanifest(ui, repo):
     def d():
         t = repo.manifest.tip()
         m = repo.manifest.read(t)
         repo.manifest.mapcache = None
         repo.manifest._cache = None
     timer(d)
 
 def perfchangeset(ui, repo, rev):
     n = repo[rev].node()
     def d():
         c = repo.changelog.read(n)
         #repo.changelog._cache = None
     timer(d)
 
 def perfindex(ui, repo):
     import mercurial.revlog
     mercurial.revlog._prereadsize = 2**24 # disable lazy parser in old hg
     n = repo["tip"].node()
     def d():
         cl = mercurial.revlog.revlog(repo.sopener, "00changelog.i")
         cl.rev(n)
     timer(d)
 
 def perfstartup(ui, repo):
     cmd = sys.argv[0]
     def d():
         os.system("HGRCPATH= %s version -q > /dev/null" % cmd)
     timer(d)
 
 def perfparents(ui, repo):
     nl = [repo.changelog.node(i) for i in xrange(1000)]
     def d():
         for n in nl:
             repo.changelog.parents(n)
     timer(d)
 
 def perflookup(ui, repo, rev):
     timer(lambda: len(repo.lookup(rev)))
 
 def perfnodelookup(ui, repo, rev):
     import mercurial.revlog
     mercurial.revlog._prereadsize = 2**24 # disable lazy parser in old hg
     n = repo[rev].node()
     def d():
         cl = mercurial.revlog.revlog(repo.sopener, "00changelog.i")
         cl.rev(n)
     timer(d)
 
+def perfnodelookup(ui, repo, rev):
+    import mercurial.revlog
+    mercurial.revlog._prereadsize = 2**24 # disable lazy parser in old hg
+    n = repo[rev].node()
+    cl = mercurial.revlog.revlog(repo.sopener, "00changelog.i")
+    # behave somewhat consistently across internal API changes
+    if util.safehasattr(cl, 'clearcaches'):
+        clearcaches = cl.clearcaches
+    elif util.safehasattr(cl, '_nodecache'):
+        from mercurial.node import nullid, nullrev
+        def clearcaches():
+            cl._nodecache = {nullid: nullrev}
+            cl._nodepos = None
+    else:
+        def clearcaches():
+            pass
+    def d():
+        cl.rev(n)
+        clearcaches()
+    timer(d)
+
 def perflog(ui, repo, **opts):
     ui.pushbuffer()
     timer(lambda: commands.log(ui, repo, rev=[], date='', user='',
                                copies=opts.get('rename')))
     ui.popbuffer()
 
 def perftemplating(ui, repo):
     ui.pushbuffer()
     timer(lambda: commands.log(ui, repo, rev=[], date='', user='',
                                template='{date|shortdate} [{rev}:{node|short}]'
                                ' {author|person}: {desc|firstline}\n'))
     ui.popbuffer()
 
 def perfcca(ui, repo):
     timer(lambda: scmutil.casecollisionauditor(ui, False, repo[None]))
 
 def perffncacheload(ui, repo):
     from mercurial import scmutil, store
     s = store.store(set(['store','fncache']), repo.path, scmutil.opener)
     def d():
         s.fncache._load()
     timer(d)
 
 def perffncachewrite(ui, repo):
     from mercurial import scmutil, store
     s = store.store(set(['store','fncache']), repo.path, scmutil.opener)
     s.fncache._load()
     def d():
         s.fncache._dirty = True
         s.fncache.write()
     timer(d)
 
 def perfdiffwd(ui, repo):
     """Profile diff of working directory changes"""
     options = {
         'w': 'ignore_all_space',
         'b': 'ignore_space_change',
         'B': 'ignore_blank_lines',
         }
 
     for diffopt in ('', 'w', 'b', 'B', 'wB'):
         opts = dict((options[c], '1') for c in diffopt)
         def d():
             ui.pushbuffer()
             commands.diff(ui, repo, **opts)
             ui.popbuffer()
         title = 'diffopts: %s' % (diffopt and ('-' + diffopt) or 'none')
         timer(d, title)
 
 def perfrevlog(ui, repo, file_, **opts):
     from mercurial import revlog
     dist = opts['dist']
     def d():
         r = revlog.revlog(lambda fn: open(fn, 'rb'), file_)
         for x in xrange(0, len(r), dist):
             r.revision(r.node(x))
 
     timer(d)
 
 cmdtable = {
     'perfcca': (perfcca, []),
     'perffncacheload': (perffncacheload, []),
     'perffncachewrite': (perffncachewrite, []),
     'perflookup': (perflookup, []),
     'perfnodelookup': (perfnodelookup, []),
     'perfparents': (perfparents, []),
     'perfstartup': (perfstartup, []),
     'perfstatus': (perfstatus, []),
     'perfwalk': (perfwalk, []),
     'perfmanifest': (perfmanifest, []),
     'perfchangeset': (perfchangeset, []),
     'perfindex': (perfindex, []),
     'perfheads': (perfheads, []),
     'perftags': (perftags, []),
     'perfdirstate': (perfdirstate, []),
     'perfdirstatedirs': (perfdirstate, []),
     'perflog': (perflog,
                 [('', 'rename', False, 'ask log to follow renames')]),
     'perftemplating': (perftemplating, []),
     'perfdiffwd': (perfdiffwd, []),
     'perfrevlog': (perfrevlog,
                    [('d', 'dist', 100, 'distance between the revisions')],
                    "[INDEXFILE]"),
 }

mercurial/parsers.c

@@ -1,713 +1,1200 b''
 /*
  parsers.c - efficient content parsing
 
  Copyright 2008 Matt 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.
 */
 
 #include <Python.h>
 #include <ctype.h>
 #include <string.h>
 
 #include "util.h"
 
 static int hexdigit(char c)
 {
 	if (c >= '0' && c <= '9')
 		return c - '0';
 	if (c >= 'a' && c <= 'f')
 		return c - 'a' + 10;
 	if (c >= 'A' && c <= 'F')
 		return c - 'A' + 10;
 
 	PyErr_SetString(PyExc_ValueError, "input contains non-hex character");
 	return 0;
 }
 
 /*
  * Turn a hex-encoded string into binary.
  */
 static PyObject *unhexlify(const char *str, int len)
 {
 	PyObject *ret;
 	const char *c;
 	char *d;
 
 	ret = PyBytes_FromStringAndSize(NULL, len / 2);
 
 	if (!ret)
 		return NULL;
 
 	d = PyBytes_AsString(ret);
 
 	for (c = str; c < str + len;) {
 		int hi = hexdigit(*c++);
 		int lo = hexdigit(*c++);
 		*d++ = (hi << 4) | lo;
 	}
 
 	return ret;
 }
 
 /*
  * This code assumes that a manifest is stitched together with newline
  * ('\n') characters.
  */
 static PyObject *parse_manifest(PyObject *self, PyObject *args)
 {
 	PyObject *mfdict, *fdict;
 	char *str, *cur, *start, *zero;
 	int len;
 
 	if (!PyArg_ParseTuple(args, "O!O!s#:parse_manifest",
 			      &PyDict_Type, &mfdict,
 			      &PyDict_Type, &fdict,
 			      &str, &len))
 		goto quit;
 
 	for (start = cur = str, zero = NULL; cur < str + len; cur++) {
 		PyObject *file = NULL, *node = NULL;
 		PyObject *flags = NULL;
 		int nlen;
 
 		if (!*cur) {
 			zero = cur;
 			continue;
 		}
 		else if (*cur != '\n')
 			continue;
 
 		if (!zero) {
 			PyErr_SetString(PyExc_ValueError,
 					"manifest entry has no separator");
 			goto quit;
 		}
 
 		file = PyBytes_FromStringAndSize(start, zero - start);
 
 		if (!file)
 			goto bail;
 
 		nlen = cur - zero - 1;
 
 		node = unhexlify(zero + 1, nlen > 40 ? 40 : nlen);
 		if (!node)
 			goto bail;
 
 		if (nlen > 40) {
 			flags = PyBytes_FromStringAndSize(zero + 41,
 							   nlen - 40);
 			if (!flags)
 				goto bail;
 
 			if (PyDict_SetItem(fdict, file, flags) == -1)
 				goto bail;
 		}
 
 		if (PyDict_SetItem(mfdict, file, node) == -1)
 			goto bail;
 
 		start = cur + 1;
 		zero = NULL;
 
 		Py_XDECREF(flags);
 		Py_XDECREF(node);
 		Py_XDECREF(file);
 		continue;
 	bail:
 		Py_XDECREF(flags);
 		Py_XDECREF(node);
 		Py_XDECREF(file);
 		goto quit;
 	}
 
 	if (len > 0 && *(cur - 1) != '\n') {
 		PyErr_SetString(PyExc_ValueError,
 				"manifest contains trailing garbage");
 		goto quit;
 	}
 
 	Py_INCREF(Py_None);
 	return Py_None;
 quit:
 	return NULL;
 }
 
 static PyObject *parse_dirstate(PyObject *self, PyObject *args)
 {
 	PyObject *dmap, *cmap, *parents = NULL, *ret = NULL;
 	PyObject *fname = NULL, *cname = NULL, *entry = NULL;
 	char *str, *cur, *end, *cpos;
 	int state, mode, size, mtime;
 	unsigned int flen;
 	int len;
 	uint32_t decode[4]; /* for alignment */
 
 	if (!PyArg_ParseTuple(args, "O!O!s#:parse_dirstate",
 			      &PyDict_Type, &dmap,
 			      &PyDict_Type, &cmap,
 			      &str, &len))
 		goto quit;
 
 	/* read parents */
 	if (len < 40)
 		goto quit;
 
 	parents = Py_BuildValue("s#s#", str, 20, str + 20, 20);
 	if (!parents)
 		goto quit;
 
 	/* read filenames */
 	cur = str + 40;
 	end = str + len;
 
 	while (cur < end - 17) {
 		/* unpack header */
 		state = *cur;
 		memcpy(decode, cur + 1, 16);
 		mode = ntohl(decode[0]);
 		size = ntohl(decode[1]);
 		mtime = ntohl(decode[2]);
 		flen = ntohl(decode[3]);
 		cur += 17;
 		if (cur + flen > end || cur + flen < cur) {
 			PyErr_SetString(PyExc_ValueError, "overflow in dirstate");
 			goto quit;
 		}
 
 		entry = Py_BuildValue("ciii", state, mode, size, mtime);
 		if (!entry)
 			goto quit;
 		PyObject_GC_UnTrack(entry); /* don't waste time with this */
 
 		cpos = memchr(cur, 0, flen);
 		if (cpos) {
 			fname = PyBytes_FromStringAndSize(cur, cpos - cur);
 			cname = PyBytes_FromStringAndSize(cpos + 1,
 							   flen - (cpos - cur) - 1);
 			if (!fname || !cname ||
 			    PyDict_SetItem(cmap, fname, cname) == -1 ||
 			    PyDict_SetItem(dmap, fname, entry) == -1)
 				goto quit;
 			Py_DECREF(cname);
 		} else {
 			fname = PyBytes_FromStringAndSize(cur, flen);
 			if (!fname ||
 			    PyDict_SetItem(dmap, fname, entry) == -1)
 				goto quit;
 		}
 		cur += flen;
 		Py_DECREF(fname);
 		Py_DECREF(entry);
 		fname = cname = entry = NULL;
 	}
 
 	ret = parents;
 	Py_INCREF(ret);
 quit:
 	Py_XDECREF(fname);
 	Py_XDECREF(cname);
 	Py_XDECREF(entry);
 	Py_XDECREF(parents);
 	return ret;
 }
 
 /*
- * A list-like object that decodes the contents of a RevlogNG index
- * file on demand. It has limited support for insert and delete at the
- * last element before the end.  The last entry is always a sentinel
- * nullid.
+ * 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 + 1)
+ * Zero is empty
+ */
+typedef struct {
+	int children[16];
+} nodetree;
+
+/*
+ * This class has two behaviours.
+ *
+ * When used in a list-like way (with integer keys), we decode an
+ * entry in a RevlogNG index file on demand. Our last entry is a
+ * sentinel, always a nullid.  We have limited support for
+ * integer-keyed insert and delete, only at elements right before the
+ * sentinel.
+ *
+ * With string keys, we lazily perform a reverse mapping from node to
+ * rev, using a base-16 trie.
  */
 typedef struct {
 	PyObject_HEAD
 	/* Type-specific fields go here. */
 	PyObject *data;        /* raw bytes of index */
 	PyObject **cache;      /* cached tuples */
 	const char **offsets;  /* populated on demand */
 	Py_ssize_t raw_length; /* original number of elements */
 	Py_ssize_t length;     /* current number of elements */
 	PyObject *added;       /* populated on demand */
+	nodetree *nt;          /* base-16 trie */
+	int ntlength;          /* # nodes in use */
+	int ntcapacity;        /* # nodes allocated */
+	int ntdepth;           /* maximum depth of tree */
+	int ntsplits;          /* # splits performed */
+	int ntrev;             /* last rev scanned */
+	int ntlookups;         /* # lookups */
+	int ntmisses;          /* # lookups that miss the cache */
 	int inlined;
 } indexObject;
 
-static Py_ssize_t index_length(indexObject *self)
+static Py_ssize_t index_length(const indexObject *self)
 {
 	if (self->added == NULL)
 		return self->length;
 	return self->length + PyList_GET_SIZE(self->added);
 }
 
 static PyObject *nullentry;
+static const char nullid[20];
 
 static long inline_scan(indexObject *self, const char **offsets);
 
 #if LONG_MAX == 0x7fffffffL
 static char *tuple_format = "Kiiiiiis#";
 #else
 static char *tuple_format = "kiiiiiis#";
 #endif
 
-/* RevlogNG format (all in big endian, data may be inlined):
+/*
+ * Return a pointer to the beginning of a RevlogNG record.
+ */
+static const char *index_deref(indexObject *self, Py_ssize_t pos)
+{
+	if (self->inlined && pos > 0) {
+		if (self->offsets == NULL) {
+			self->offsets = malloc(self->raw_length *
+					       sizeof(*self->offsets));
+			if (self->offsets == NULL)
+				return (const char *)PyErr_NoMemory();
+			inline_scan(self, self->offsets);
+		}
+		return self->offsets[pos];
+	}
+
+	return PyString_AS_STRING(self->data) + pos * 64;
+}
+
+/*
+ * 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)
  */
 static PyObject *index_get(indexObject *self, Py_ssize_t pos)
 {
 	uint32_t decode[8]; /* to enforce alignment with inline data */
 	uint64_t offset_flags;
 	int comp_len, uncomp_len, base_rev, link_rev, parent_1, parent_2;
 	const char *c_node_id;
 	const char *data;
 	Py_ssize_t length = index_length(self);
 	PyObject *entry;
 
-	if (pos >= length) {
+	if (pos < 0)
+		pos += length;
+
+	if (pos < 0 || pos >= length) {
 		PyErr_SetString(PyExc_IndexError, "revlog index out of range");
 		return NULL;
 	}
 
 	if (pos == length - 1) {
 		Py_INCREF(nullentry);
 		return nullentry;
 	}
 
 	if (pos >= self->length - 1) {
 		PyObject *obj;
 		obj = PyList_GET_ITEM(self->added, pos - self->length + 1);
 		Py_INCREF(obj);
 		return obj;
 	}
 
 	if (self->cache) {
 		if (self->cache[pos]) {
 			Py_INCREF(self->cache[pos]);
 			return self->cache[pos];
 		}
 	} else {
 		self->cache = calloc(self->raw_length, sizeof(PyObject *));
 		if (self->cache == NULL)
 			return PyErr_NoMemory();
 	}
 
-	if (self->inlined && pos > 0) {
-		if (self->offsets == NULL) {
-			self->offsets = malloc(self->raw_length *
-					       sizeof(*self->offsets));
-			if (self->offsets == NULL)
-				return PyErr_NoMemory();
-			inline_scan(self, self->offsets);
-		}
-		data = self->offsets[pos];
-	} else
-		data = PyString_AS_STRING(self->data) + pos * 64;
+	data = index_deref(self, pos);
+	if (data == NULL)
+		return NULL;
 
 	memcpy(decode, data, 8 * sizeof(uint32_t));
 
 	offset_flags = ntohl(decode[1]);
 	if (pos == 0) /* mask out version number for the first entry */
 		offset_flags &= 0xFFFF;
 	else {
 		uint32_t offset_high = ntohl(decode[0]);
 		offset_flags |= ((uint64_t)offset_high) << 32;
 	}
 
 	comp_len = ntohl(decode[2]);
 	uncomp_len = ntohl(decode[3]);
 	base_rev = ntohl(decode[4]);
 	link_rev = ntohl(decode[5]);
 	parent_1 = ntohl(decode[6]);
 	parent_2 = ntohl(decode[7]);
 	c_node_id = data + 32;
 
 	entry = Py_BuildValue(tuple_format, offset_flags, comp_len,
 			      uncomp_len, base_rev, link_rev,
 			      parent_1, parent_2, c_node_id, 20);
 
 	if (entry)
 		PyObject_GC_UnTrack(entry);
 
 	self->cache[pos] = entry;
 	Py_INCREF(entry);
 
 	return entry;
 }
 
+/*
+ * Return the 20-byte SHA of the 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 == length - 1)
+		return nullid;
+
+	if (pos >= length)
+		return NULL;
+
+	if (pos >= self->length - 1) {
+		PyObject *tuple, *str;
+		tuple = PyList_GET_ITEM(self->added, pos - self->length + 1);
+		str = PyTuple_GetItem(tuple, 7);
+		return str ? PyString_AS_STRING(str) : NULL;
+	}
+
+	data = index_deref(self, pos);
+	return data ? data + 32 : NULL;
+}
+
+static int nt_insert(indexObject *self, const char *node, int rev);
+
+static int node_check(PyObject *obj, char **node, Py_ssize_t *nodelen)
+{
+	if (PyString_AsStringAndSize(obj, node, nodelen) == -1)
+		return -1;
+	if (*nodelen == 20)
+		return 0;
+	PyErr_SetString(PyExc_ValueError, "20-byte hash required");
+	return -1;
+}
+
 static PyObject *index_insert(indexObject *self, PyObject *args)
 {
-	PyObject *obj, *node;
+	PyObject *obj;
+	char *node;
 	long offset;
-	Py_ssize_t len;
+	Py_ssize_t len, nodelen;
 
 	if (!PyArg_ParseTuple(args, "lO", &offset, &obj))
 		return NULL;
 
 	if (!PyTuple_Check(obj) || PyTuple_GET_SIZE(obj) != 8) {
-		PyErr_SetString(PyExc_ValueError, "8-tuple required");
+		PyErr_SetString(PyExc_TypeError, "8-tuple required");
 		return NULL;
 	}
 
-	node = PyTuple_GET_ITEM(obj, 7);
-	if (!PyString_Check(node) || PyString_GET_SIZE(node) != 20) {
-		PyErr_SetString(PyExc_ValueError,
-				"20-byte hash required as last element");
+	if (node_check(PyTuple_GET_ITEM(obj, 7), &node, &nodelen) == -1)
 		return NULL;
-	}
 
 	len = index_length(self);
 
 	if (offset < 0)
 		offset += len;
 
 	if (offset != len - 1) {
 		PyErr_SetString(PyExc_IndexError,
 				"insert only supported at index -1");
 		return NULL;
 	}
 
+	if (offset > INT_MAX) {
+		PyErr_SetString(PyExc_ValueError,
+				"currently only 2**31 revs supported");
+		return NULL;
+	}
+
 	if (self->added == NULL) {
 		self->added = PyList_New(0);
 		if (self->added == NULL)
 			return NULL;
 	}
 
 	if (PyList_Append(self->added, obj) == -1)
 		return NULL;
 
+	if (self->nt)
+		nt_insert(self, node, (int)offset);
+
 	Py_RETURN_NONE;
 }
 
 static void _index_clearcaches(indexObject *self)
 {
 	if (self->cache) {
 		Py_ssize_t i;
 
 		for (i = 0; i < self->raw_length; i++) {
 			Py_XDECREF(self->cache[i]);
 			self->cache[i] = NULL;
 		}
 		free(self->cache);
 		self->cache = NULL;
 	}
 	if (self->offsets) {
 		free(self->offsets);
 		self->offsets = NULL;
 	}
+	if (self->nt) {
+		free(self->nt);
+		self->nt = NULL;
+	}
 }
 
 static PyObject *index_clearcaches(indexObject *self)
 {
 	_index_clearcaches(self);
+	self->ntlength = self->ntcapacity = 0;
+	self->ntdepth = self->ntsplits = 0;
+	self->ntrev = -1;
+	self->ntlookups = self->ntmisses = 0;
 	Py_RETURN_NONE;
 }
 
-static int index_assign_subscript(indexObject *self, PyObject *item,
-				  PyObject *value)
+static PyObject *index_stats(indexObject *self)
+{
+	PyObject *obj = PyDict_New();
+
+	if (obj == NULL)
+		return NULL;
+
+#define istat(__n, __d) \
+	if (PyDict_SetItemString(obj, __d, PyInt_FromLong(self->__n)) == -1) \
+		goto bail;
+
+	if (self->added) {
+		Py_ssize_t len = PyList_GET_SIZE(self->added);
+		if (PyDict_SetItemString(obj, "index entries added",
+					 PyInt_FromLong(len)) == -1)
+			goto bail;
+	}
+
+	if (self->raw_length != self->length - 1)
+		istat(raw_length, "revs on disk");
+	istat(length, "revs in memory");
+	istat(ntcapacity, "node trie capacity");
+	istat(ntdepth, "node trie depth");
+	istat(ntlength, "node trie count");
+	istat(ntlookups, "node trie lookups");
+	istat(ntmisses, "node trie misses");
+	istat(ntrev, "node trie last rev scanned");
+	istat(ntsplits, "node trie splits");
+
+#undef istat
+
+	return obj;
+
+bail:
+	Py_XDECREF(obj);
+	return NULL;
+}
+
+static inline int nt_level(const char *node, int level)
+{
+	int v = node[level>>1];
+	if (!(level & 1))
+		v >>= 4;
+	return v & 0xf;
+}
+
+static int nt_find(indexObject *self, const char *node, Py_ssize_t nodelen)
+{
+	int level, off;
+
+	if (nodelen == 20 && node[0] == '\0' && memcmp(node, nullid, 20) == 0)
+		return -1;
+
+	if (self->nt == NULL)
+		return -2;
+
+	for (level = off = 0; level < nodelen; level++) {
+		int k = nt_level(node, level);
+		nodetree *n = &self->nt[off];
+		int v = n->children[k];
+
+		if (v < 0) {
+			const char *n;
+			v = -v - 1;
+			n = index_node(self, v);
+			if (n == NULL)
+				return -2;
+			return memcmp(node, n, nodelen > 20 ? 20 : nodelen)
+				? -2 : v;
+		}
+		if (v == 0)
+			return -2;
+		off = v;
+	}
+	return -2;
+}
+
+static int nt_new(indexObject *self)
+{
+	if (self->ntlength == self->ntcapacity) {
+		self->ntcapacity *= 2;
+		self->nt = realloc(self->nt,
+				   self->ntcapacity * sizeof(nodetree));
+		if (self->nt == NULL) {
+			PyErr_SetString(PyExc_MemoryError, "out of memory");
+			return -1;
+		}
+		memset(&self->nt[self->ntlength], 0,
+		       sizeof(nodetree) * (self->ntcapacity - self->ntlength));
+	}
+	return self->ntlength++;
+}
+
+static int nt_insert(indexObject *self, const char *node, int rev)
+{
+	int level = 0;
+	int off = 0;
+
+	while (level < 20) {
+		int k = nt_level(node, level);
+		nodetree *n;
+		int v;
+
+		n = &self->nt[off];
+		v = n->children[k];
+
+		if (v == 0) {
+			n->children[k] = -rev - 1;
+			return 0;
+		}
+		if (v < 0) {
+			const char *oldnode = index_node(self, -v - 1);
+			int noff;
+
+			if (!oldnode || !memcmp(oldnode, node, 20)) {
+				n->children[k] = -rev - 1;
+				return 0;
+			}
+			noff = nt_new(self);
+			if (noff == -1)
+				return -1;
+			/* self->nt may have been changed by realloc */
+			self->nt[off].children[k] = noff;
+			off = noff;
+			n = &self->nt[off];
+			n->children[nt_level(oldnode, ++level)] = v;
+			if (level > self->ntdepth)
+				self->ntdepth = level;
+			self->ntsplits += 1;
+		} else {
+			level += 1;
+			off = v;
+		}
+	}
+
+	return -1;
+}
+
+/*
+ * 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, Py_ssize_t nodelen)
+{
+	int rev;
+
+	self->ntlookups++;
+	rev = nt_find(self, node, nodelen);
+	if (rev >= -1)
+		return rev;
+
+	if (self->nt == NULL) {
+		self->ntcapacity = self->raw_length < 4
+			? 4 : self->raw_length / 2;
+		self->nt = calloc(self->ntcapacity, sizeof(nodetree));
+		if (self->nt == NULL) {
+			PyErr_SetString(PyExc_MemoryError, "out of memory");
+			return -3;
+		}
+		self->ntlength = 1;
+		self->ntrev = (int)index_length(self) - 1;
+		self->ntlookups = 1;
+		self->ntmisses = 0;
+	}
+
+	/*
+	 * 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(self, rev);
+			if (n == NULL)
+				return -2;
+			if (memcmp(node, n, nodelen > 20 ? 20 : nodelen) == 0) {
+				if (nt_insert(self, n, rev) == -1)
+					return -3;
+				break;
+			}
+		}
+	} else {
+		for (rev = self->ntrev - 1; rev >= 0; rev--) {
+			const char *n = index_node(self, rev);
+			if (n == NULL)
+				return -2;
+			if (nt_insert(self, n, rev) == -1)
+				return -3;
+			if (memcmp(node, n, nodelen > 20 ? 20 : nodelen) == 0) {
+				break;
+			}
+		}
+		self->ntrev = rev;
+	}
+
+	if (rev >= 0)
+		return rev;
+	return -2;
+}
+
+static PyObject *raise_revlog_error(void)
+{
+	static PyObject *errclass;
+	PyObject *mod = NULL, *errobj;
+
+	if (errclass == NULL) {
+		PyObject *dict;
+
+		mod = PyImport_ImportModule("mercurial.error");
+		if (mod == NULL)
+			goto classfail;
+
+		dict = PyModule_GetDict(mod);
+		if (dict == NULL)
+			goto classfail;
+
+		errclass = PyDict_GetItemString(dict, "RevlogError");
+		if (errclass == NULL) {
+			PyErr_SetString(PyExc_SystemError,
+					"could not find RevlogError");
+			goto classfail;
+		}
+		Py_INCREF(errclass);
+	}
+
+	errobj = PyObject_CallFunction(errclass, NULL);
+	if (errobj == NULL)
+		return NULL;
+	PyErr_SetObject(errclass, errobj);
+	return errobj;
+
+classfail:
+	Py_XDECREF(mod);
+	return NULL;
+}
+
+static PyObject *index_getitem(indexObject *self, PyObject *value)
+{
+	char *node;
+	Py_ssize_t nodelen;
+	int rev;
+
+	if (PyInt_Check(value))
+		return index_get(self, PyInt_AS_LONG(value));
+
+	if (PyString_AsStringAndSize(value, &node, &nodelen) == -1)
+		return NULL;
+	rev = index_find_node(self, node, nodelen);
+	if (rev >= -1)
+		return PyInt_FromLong(rev);
+	if (rev == -2)
+		raise_revlog_error();
+	return NULL;
+}
+
+static PyObject *index_m_get(indexObject *self, PyObject *args)
+{
+	char *node;
+	int nodelen, rev;
+
+	if (!PyArg_ParseTuple(args, "s#", &node, &nodelen))
+		return NULL;
+
+	rev = index_find_node(self, node, nodelen);
+	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;
+	Py_ssize_t nodelen;
+
+	if (PyInt_Check(value)) {
+		long rev = PyInt_AS_LONG(value);
+		return rev >= -1 && rev < index_length(self);
+	}
+
+	if (!PyString_Check(value))
+		return 0;
+
+	node = PyString_AS_STRING(value);
+	nodelen = PyString_GET_SIZE(value);
+
+	switch (index_find_node(self, node, nodelen)) {
+	case -3:
+		return -1;
+	case -2:
+		return 0;
+	default:
+		return 1;
+	}
+}
+
+/*
+ * Invalidate any trie entries introduced by added revs.
+ */
+static void nt_invalidate_added(indexObject *self, Py_ssize_t start)
+{
+	Py_ssize_t i, len = PyList_GET_SIZE(self->added);
+
+	for (i = start; i < len; i++) {
+		PyObject *tuple = PyList_GET_ITEM(self->added, i);
+		PyObject *node = PyTuple_GET_ITEM(tuple, 7);
+
+		nt_insert(self, PyString_AS_STRING(node), -1);
+	}
+
+	if (start == 0) {
+		Py_DECREF(self->added);
+		self->added = NULL;
+	}
+}
+
+/*
+ * Delete a numeric range of revs, which must be at the end of the
+ * range, but exclude the sentinel nullid entry.
+ */
+static int index_slice_del(indexObject *self, PyObject *item)
 {
 	Py_ssize_t start, stop, step, slicelength;
 	Py_ssize_t length = index_length(self);
 
-	if (!PySlice_Check(item) || value != NULL) {
-		PyErr_SetString(PyExc_TypeError,
-				"revlog index only supports slice deletion");
-		return -1;
-	}
-
 	if (PySlice_GetIndicesEx((PySliceObject*)item, length,
 				 &start, &stop, &step, &slicelength) < 0)
 		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 (start < self->length - 1) {
+		if (self->nt) {
+			Py_ssize_t i;
+
+			for (i = start + 1; i < self->length - 1; i++) {
+				const char *node = index_node(self, i);
+
+				if (node)
+					nt_insert(self, node, -1);
+			}
+			if (self->added)
+				nt_invalidate_added(self, 0);
+			if (self->ntrev > start)
+				self->ntrev = (int)start;
+		}
 		self->length = start + 1;
-		if (self->added) {
-			Py_DECREF(self->added);
-			self->added = NULL;
-		}
 		return 0;
 	}
 
-	return PyList_SetSlice(self->added, start - self->length + 1,
-			       PyList_GET_SIZE(self->added),
-			       NULL);
+	if (self->nt) {
+		nt_invalidate_added(self, start - self->length + 1);
+		if (self->ntrev > start)
+			self->ntrev = (int)start;
+	}
+	return self->added
+		? PyList_SetSlice(self->added, start - self->length + 1,
+				  PyList_GET_SIZE(self->added), NULL)
+		: 0;
 }
 
+/*
+ * 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;
+	Py_ssize_t nodelen;
+	long rev;
+
+	if (PySlice_Check(item) && value == NULL)
+		return index_slice_del(self, item);
+
+	if (node_check(item, &node, &nodelen) == -1)
+		return -1;
+
+	if (value == NULL)
+		return self->nt ? nt_insert(self, node, -1) : 0;
+	rev = PyInt_AsLong(value);
+	if (rev > INT_MAX || rev < 0) {
+		if (!PyErr_Occurred())
+			PyErr_SetString(PyExc_ValueError, "rev out of range");
+		return -1;
+	}
+	return nt_insert(self, node, (int)rev);
+}
+
+/*
+ * Find all RevlogNG entries in an index that has inline data. Update
+ * the optional "offsets" table with those entries.
+ */
 static long inline_scan(indexObject *self, const char **offsets)
 {
 	const char *data = PyString_AS_STRING(self->data);
 	const char *end = data + PyString_GET_SIZE(self->data);
 	const long hdrsize = 64;
 	long incr = hdrsize;
 	Py_ssize_t len = 0;
 
 	while (data + hdrsize <= end) {
 		uint32_t comp_len;
 		const char *old_data;
 		/* 3rd element of header is length of compressed inline data */
 		memcpy(&comp_len, data + 8, sizeof(uint32_t));
 		incr = hdrsize + ntohl(comp_len);
 		if (incr < hdrsize)
 			break;
 		if (offsets)
 			offsets[len] = data;
 		len++;
 		old_data = data;
 		data += incr;
 		if (data <= old_data)
 			break;
 	}
 
 	if (data != end && data + hdrsize != end) {
 		if (!PyErr_Occurred())
 			PyErr_SetString(PyExc_ValueError, "corrupt index file");
 		return -1;
 	}
 
 	return len;
 }
 
 static int index_real_init(indexObject *self, const char *data, int size,
 			   PyObject *inlined_obj, PyObject *data_obj)
 {
 	self->inlined = inlined_obj && PyObject_IsTrue(inlined_obj);
 	self->data = data_obj;
 	self->cache = NULL;
 
 	self->added = NULL;
 	self->offsets = NULL;
+	self->nt = NULL;
+	self->ntlength = self->ntcapacity = 0;
+	self->ntdepth = self->ntsplits = 0;
+	self->ntlookups = self->ntmisses = 0;
+	self->ntrev = -1;
 	Py_INCREF(self->data);
 
 	if (self->inlined) {
 		long len = inline_scan(self, NULL);
 		if (len == -1)
 			goto bail;
 		self->raw_length = len;
 		self->length = len + 1;
 	} else {
 		if (size % 64) {
 			PyErr_SetString(PyExc_ValueError, "corrupt index file");
 			goto bail;
 		}
 		self->raw_length = size / 64;
 		self->length = self->raw_length + 1;
 	}
 
 	return 0;
 bail:
 	return -1;
 }
 
 static int index_init(indexObject *self, PyObject *args, PyObject *kwds)
 {
 	const char *data;
 	int size;
 	PyObject *inlined_obj;
 
 	if (!PyArg_ParseTuple(args, "s#O", &data, &size, &inlined_obj))
 		return -1;
 
 	return index_real_init(self, data, size, inlined_obj,
 			       PyTuple_GET_ITEM(args, 0));
 }
 
+static PyObject *index_nodemap(indexObject *self)
+{
+	return (PyObject *)self;
+}
+
 static void index_dealloc(indexObject *self)
 {
 	_index_clearcaches(self);
 	Py_DECREF(self->data);
 	Py_XDECREF(self->added);
 	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 */
-	NULL,                                  /* mp_subscript */
+	(binaryfunc)index_getitem,             /* mp_subscript */
 	(objobjargproc)index_assign_subscript, /* mp_ass_subscript */
 };
 
 static PyMethodDef index_methods[] = {
 	{"clearcaches", (PyCFunction)index_clearcaches, METH_NOARGS,
 	 "clear the index caches"},
+	{"get", (PyCFunction)index_m_get, METH_VARARGS,
+	 "get an index entry"},
 	{"insert", (PyCFunction)index_insert, METH_VARARGS,
 	 "insert an index entry"},
+	{"stats", (PyCFunction)index_stats, METH_NOARGS,
+	 "stats for the index"},
+	{NULL} /* Sentinel */
+};
+
+static PyGetSetDef index_getset[] = {
+	{"nodemap", (getter)index_nodemap, NULL, "nodemap", NULL},
 	{NULL} /* Sentinel */
 };
 
 static PyTypeObject indexType = {
 	PyObject_HEAD_INIT(NULL)
 	0,                         /* ob_size */
 	"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 */
 	0,                         /* tp_members */
-	0,                         /* tp_getset */
+	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 */
 	PyType_GenericNew,         /* tp_new */
 };
 
 /*
- * returns a tuple of the form (index, None, cache) with elements as
+ * returns a tuple of the form (index, index, cache) with elements as
  * follows:
  *
- * index: an index object that lazily parses the RevlogNG records
+ * index: an index object that lazily parses RevlogNG records
  * cache: if data is inlined, a tuple (index_file_content, 0), else None
  *
  * added complications are for backwards compatibility
  */
 static PyObject *parse_index2(PyObject *self, PyObject *args)
 {
 	const char *data;
 	int size, ret;
 	PyObject *inlined_obj, *tuple = NULL, *cache = NULL;
 	indexObject *idx;
 
 	if (!PyArg_ParseTuple(args, "s#O", &data, &size, &inlined_obj))
 		return NULL;
 
 	idx = PyObject_New(indexObject, &indexType);
 
 	if (idx == NULL)
 		goto bail;
 
 	ret = index_real_init(idx, data, size, inlined_obj,
 			      PyTuple_GET_ITEM(args, 0));
 	if (ret)
 		goto bail;
 
 	if (idx->inlined) {
 		Py_INCREF(idx->data);
 		cache = Py_BuildValue("iO", 0, idx->data);
 		if (cache == NULL)
 			goto bail;
 	} else {
 		cache = Py_None;
 		Py_INCREF(cache);
 	}
 
+	Py_INCREF(idx);
+
 	tuple = Py_BuildValue("NN", idx, cache);
 	if (!tuple)
 		goto bail;
 	return tuple;
 
 bail:
 	Py_XDECREF(idx);
 	Py_XDECREF(cache);
 	Py_XDECREF(tuple);
 	return NULL;
 }
 
 static char parsers_doc[] = "Efficient content parsing.";
 
 static PyMethodDef methods[] = {
 	{"parse_manifest", parse_manifest, METH_VARARGS, "parse a manifest\n"},
 	{"parse_dirstate", parse_dirstate, METH_VARARGS, "parse a dirstate\n"},
 	{"parse_index2", parse_index2, METH_VARARGS, "parse a revlog index\n"},
 	{NULL, NULL}
 };
 
 static void module_init(PyObject *mod)
 {
-	static const char nullid[20];
-
 	if (PyType_Ready(&indexType) < 0)
 		return;
 	Py_INCREF(&indexType);
 
 	PyModule_AddObject(mod, "index", (PyObject *)&indexType);
 
 	nullentry = Py_BuildValue("iiiiiiis#", 0, 0, 0,
 				  -1, -1, -1, -1, nullid, 20);
 	if (nullentry)
 		PyObject_GC_UnTrack(nullentry);
 }
 
 #ifdef IS_PY3K
 static struct PyModuleDef parsers_module = {
 	PyModuleDef_HEAD_INIT,
 	"parsers",
 	parsers_doc,
 	-1,
 	methods
 };
 
 PyMODINIT_FUNC PyInit_parsers(void)
 {
 	PyObject *mod = PyModule_Create(&parsers_module);
 	module_init(mod);
 	return mod;
 }
 #else
 PyMODINIT_FUNC initparsers(void)
 {
 	PyObject *mod = Py_InitModule3("parsers", methods, parsers_doc);
 	module_init(mod);
 }
 #endif
-

mercurial/revlog.py

@@ -1,1295 +1,1306 b''
 # revlog.py - storage back-end for mercurial
 #
 # Copyright 2005-2007 Matt 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.
 """
 
 # import stuff from node for others to import from revlog
 from node import bin, hex, nullid, nullrev
 from i18n import _
 import ancestor, mdiff, parsers, error, util, dagutil
 import struct, zlib, errno
 
 _pack = struct.pack
 _unpack = struct.unpack
 _compress = zlib.compress
 _decompress = zlib.decompress
 _sha = util.sha1
 
 # revlog header flags
 REVLOGV0 = 0
 REVLOGNG = 1
 REVLOGNGINLINEDATA = (1 << 16)
 REVLOGGENERALDELTA = (1 << 17)
 REVLOG_DEFAULT_FLAGS = REVLOGNGINLINEDATA
 REVLOG_DEFAULT_FORMAT = REVLOGNG
 REVLOG_DEFAULT_VERSION = REVLOG_DEFAULT_FORMAT | REVLOG_DEFAULT_FLAGS
 REVLOGNG_FLAGS = REVLOGNGINLINEDATA | REVLOGGENERALDELTA
 
 # revlog index flags
 REVIDX_KNOWN_FLAGS = 0
 
 # max size of revlog with inline data
 _maxinline = 131072
 _chunksize = 1048576
 
 RevlogError = error.RevlogError
 LookupError = error.LookupError
 
 def getoffset(q):
     return int(q >> 16)
 
 def gettype(q):
     return int(q & 0xFFFF)
 
 def offset_type(offset, type):
     return long(long(offset) << 16 | type)
 
 nullhash = _sha(nullid)
 
 def hash(text, p1, p2):
     """generate a hash from the given text and its parent hashes
 
     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 == 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
         l = [p1, p2]
         l.sort()
         s = _sha(l[0])
         s.update(l[1])
     s.update(text)
     return s.digest()
 
 def compress(text):
     """ generate a possibly-compressed representation of text """
     if not text:
         return ("", text)
     l = len(text)
     bin = None
     if l < 44:
         pass
     elif l > 1000000:
         # zlib makes an internal copy, thus doubling memory usage for
         # large files, so lets do this in pieces
         z = zlib.compressobj()
         p = []
         pos = 0
         while pos < l:
             pos2 = pos + 2**20
             p.append(z.compress(text[pos:pos2]))
             pos = pos2
         p.append(z.flush())
         if sum(map(len, p)) < l:
             bin = "".join(p)
     else:
         bin = _compress(text)
     if bin is None or len(bin) > l:
         if text[0] == '\0':
             return ("", text)
         return ('u', text)
     return ("", bin)
 
 def decompress(bin):
     """ decompress the given input """
     if not bin:
         return bin
     t = bin[0]
     if t == '\0':
         return bin
     if t == 'x':
         return _decompress(bin)
     if t == 'u':
         return bin[1:]
     raise RevlogError(_("unknown compression type %r") % t)
 
 indexformatv0 = ">4l20s20s20s"
 v0shaoffset = 56
 
 class revlogoldio(object):
     def __init__(self):
         self.size = struct.calcsize(indexformatv0)
 
     def parseindex(self, data, inline):
         s = self.size
         index = []
         nodemap =  {nullid: nullrev}
         n = off = 0
         l = len(data)
         while off + s <= l:
             cur = data[off:off + s]
             off += s
             e = _unpack(indexformatv0, cur)
             # transform to revlogv1 format
             e2 = (offset_type(e[0], 0), e[1], -1, e[2], e[3],
                   nodemap.get(e[4], nullrev), nodemap.get(e[5], nullrev), e[6])
             index.append(e2)
             nodemap[e[6]] = n
             n += 1
 
         # add the magic null revision at -1
         index.append((0, 0, 0, -1, -1, -1, -1, nullid))
 
         return index, nodemap, None
 
     def packentry(self, entry, node, version, rev):
         if gettype(entry[0]):
             raise RevlogError(_("index entry flags need RevlogNG"))
         e2 = (getoffset(entry[0]), entry[1], entry[3], entry[4],
               node(entry[5]), node(entry[6]), entry[7])
         return _pack(indexformatv0, *e2)
 
 # index ng:
 #  6 bytes: offset
 #  2 bytes: flags
 #  4 bytes: compressed length
 #  4 bytes: uncompressed length
 #  4 bytes: base rev
 #  4 bytes: link rev
 #  4 bytes: parent 1 rev
 #  4 bytes: parent 2 rev
 # 32 bytes: nodeid
 indexformatng = ">Qiiiiii20s12x"
 ngshaoffset = 32
 versionformat = ">I"
 
 class revlogio(object):
     def __init__(self):
         self.size = struct.calcsize(indexformatng)
 
     def parseindex(self, data, inline):
         # call the C implementation to parse the index data
         index, cache = parsers.parse_index2(data, inline)
-        return index, None, cache
+        return index, getattr(index, 'nodemap', None), cache
 
     def packentry(self, entry, node, version, rev):
         p = _pack(indexformatng, *entry)
         if rev == 0:
             p = _pack(versionformat, version) + p[4:]
         return p
 
 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.
     """
     def __init__(self, opener, indexfile):
         """
         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.
         """
         self.indexfile = indexfile
         self.datafile = indexfile[:-2] + ".d"
         self.opener = opener
         self._cache = None
         self._basecache = (0, 0)
         self._chunkcache = (0, '')
         self.index = []
         self._pcache = {}
         self._nodecache = {nullid: nullrev}
         self._nodepos = None
 
         v = REVLOG_DEFAULT_VERSION
         opts = getattr(opener, 'options', None)
         if opts is not None:
             if 'revlogv1' in opts:
                 if 'generaldelta' in opts:
                     v |= REVLOGGENERALDELTA
             else:
                 v = 0
 
         i = ''
         self._initempty = True
         try:
             f = self.opener(self.indexfile)
             i = f.read()
             f.close()
             if len(i) > 0:
                 v = struct.unpack(versionformat, i[:4])[0]
                 self._initempty = False
         except IOError, inst:
             if inst.errno != errno.ENOENT:
                 raise
 
         self.version = v
         self._inline = v & REVLOGNGINLINEDATA
         self._generaldelta = v & REVLOGGENERALDELTA
         flags = v & ~0xFFFF
         fmt = v & 0xFFFF
         if fmt == REVLOGV0 and flags:
             raise RevlogError(_("index %s unknown flags %#04x for format v0")
                               % (self.indexfile, flags >> 16))
         elif fmt == REVLOGNG and flags & ~REVLOGNG_FLAGS:
             raise RevlogError(_("index %s unknown flags %#04x for revlogng")
                               % (self.indexfile, flags >> 16))
         elif fmt > REVLOGNG:
             raise RevlogError(_("index %s unknown format %d")
                               % (self.indexfile, fmt))
 
         self._io = revlogio()
         if self.version == REVLOGV0:
             self._io = revlogoldio()
         try:
             d = self._io.parseindex(i, self._inline)
         except (ValueError, IndexError):
             raise RevlogError(_("index %s is corrupted") % (self.indexfile))
         self.index, nodemap, self._chunkcache = d
         if nodemap is not None:
             self.nodemap = self._nodecache = nodemap
         if not self._chunkcache:
             self._chunkclear()
 
     def tip(self):
         return self.node(len(self.index) - 2)
     def __len__(self):
         return len(self.index) - 1
     def __iter__(self):
         for i in xrange(len(self)):
             yield i
 
     @util.propertycache
     def nodemap(self):
         self.rev(self.node(0))
         return self._nodecache
 
     def hasnode(self, node):
         try:
             self.rev(node)
             return True
         except KeyError:
             return False
 
+    def clearcaches(self):
+        try:
+            self._nodecache.clearcaches()
+        except AttributeError:
+            self._nodecache = {nullid: nullrev}
+            self._nodepos = None
+
     def rev(self, node):
         try:
             return self._nodecache[node]
+        except RevlogError:
+            # parsers.c radix tree lookup failed
+            raise LookupError(node, self.indexfile, _('no node'))
         except KeyError:
+            # pure python cache lookup failed
             n = self._nodecache
             i = self.index
             p = self._nodepos
             if p is None:
                 p = len(i) - 2
             for r in xrange(p, -1, -1):
                 v = i[r][7]
                 n[v] = r
                 if v == node:
                     self._nodepos = r - 1
                     return r
             raise LookupError(node, self.indexfile, _('no node'))
 
     def node(self, rev):
         return self.index[rev][7]
     def linkrev(self, rev):
         return self.index[rev][4]
     def parents(self, node):
         i = self.index
         d = i[self.rev(node)]
         return i[d[5]][7], i[d[6]][7] # map revisions to nodes inline
     def parentrevs(self, rev):
         return self.index[rev][5:7]
     def start(self, rev):
         return int(self.index[rev][0] >> 16)
     def end(self, rev):
         return self.start(rev) + self.length(rev)
     def length(self, rev):
         return self.index[rev][1]
     def chainbase(self, rev):
         index = self.index
         base = index[rev][3]
         while base != rev:
             rev = base
             base = index[rev][3]
         return base
     def flags(self, rev):
         return self.index[rev][0] & 0xFFFF
     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.revision(self.node(rev))
         return len(t)
     size = rawsize
 
     def reachable(self, node, stop=None):
         """return the set of all nodes ancestral to a given node, including
          the node itself, stopping when stop is matched"""
         reachable = set((node,))
         visit = [node]
         if stop:
             stopn = self.rev(stop)
         else:
             stopn = 0
         while visit:
             n = visit.pop(0)
             if n == stop:
                 continue
             if n == nullid:
                 continue
             for p in self.parents(n):
                 if self.rev(p) < stopn:
                     continue
                 if p not in reachable:
                     reachable.add(p)
                     visit.append(p)
         return reachable
 
     def ancestors(self, *revs):
         """Generate the ancestors of 'revs' in reverse topological order.
 
         Yield a sequence of revision numbers starting with the parents
         of each revision in revs, i.e., each revision is *not* considered
         an ancestor of itself.  Results are in breadth-first order:
         parents of each rev in revs, then parents of those, etc.  Result
         does not include the null revision."""
         visit = list(revs)
         seen = set([nullrev])
         while visit:
             for parent in self.parentrevs(visit.pop(0)):
                 if parent not in seen:
                     visit.append(parent)
                     seen.add(parent)
                     yield parent
 
     def descendants(self, *revs):
         """Generate the descendants of 'revs' in revision order.
 
         Yield a sequence of revision numbers starting with a child of
         some rev in revs, i.e., each revision is *not* considered a
         descendant of itself.  Results are ordered by revision number (a
         topological sort)."""
         first = min(revs)
         if first == nullrev:
             for i in self:
                 yield i
             return
 
         seen = set(revs)
         for i in xrange(first + 1, len(self)):
             for x in self.parentrevs(i):
                 if x != nullrev and x in seen:
                     seen.add(i)
                     yield i
                     break
 
     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 = [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
         has = set(self.ancestors(*common))
         has.add(nullrev)
         has.update(common)
 
         # take all ancestors from heads that aren't in has
         missing = set()
         visit = [r for r in heads if r not in has]
         while visit:
             r = visit.pop(0)
             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(r) for r in missing]
 
     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."""
         _common, missing = self.findcommonmissing(common, heads)
         return missing
 
     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 = [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], [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 == 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 != 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 = [n for n in roots if n in ancestors]
                 # Recompute the lowest revision
                 if roots:
                     lowestrev = min([self.rev(n) for n 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 = [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 descedents, empty parents will look like
         # they're descendants.
         for r in xrange(max(lowestrev, 0), 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 = [n for n, flag in heads.iteritems() if flag]
         roots = list(roots)
         assert orderedout
         assert roots
         assert heads
         return (orderedout, roots, heads)
 
     def headrevs(self):
         count = len(self)
         if not count:
             return [nullrev]
         ishead = [1] * (count + 1)
         index = self.index
         for r in xrange(count):
             e = index[r]
             ishead[e[5]] = ishead[e[6]] = 0
         return [r for r in xrange(count) if ishead[r]]
 
     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 [nullid]
             return [self.node(r) for r in self.headrevs()]
 
         if start is None:
             start = nullid
         if stop is None:
             stop = []
         stoprevs = set([self.rev(n) for n in stop])
         startrev = self.rev(start)
         reachable = set((startrev,))
         heads = set((startrev,))
 
         parentrevs = self.parentrevs
         for r in xrange(startrev + 1, len(self)):
             for p in parentrevs(r):
                 if p in reachable:
                     if r not in stoprevs:
                         reachable.add(r)
                     heads.add(r)
                 if p in heads and p not in stoprevs:
                     heads.remove(p)
 
         return [self.node(r) for r in heads]
 
     def children(self, node):
         """find the children of a given node"""
         c = []
         p = self.rev(node)
         for r in range(p + 1, len(self)):
             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 descendant(self, start, end):
         if start == nullrev:
             return True
         for i in self.descendants(start):
             if i == end:
                 return True
             elif i > end:
                 break
         return False
 
     def ancestor(self, a, b):
         """calculate the least common ancestor of nodes a and b"""
 
         # fast path, check if it is a descendant
         a, b = self.rev(a), self.rev(b)
         start, end = sorted((a, b))
         if self.descendant(start, end):
             return self.node(start)
 
         def parents(rev):
             return [p for p in self.parentrevs(rev) if p != nullrev]
 
         c = ancestor.ancestor(a, b, parents)
         if c is None:
             return nullid
 
         return self.node(c)
 
     def _match(self, id):
         if isinstance(id, (long, int)):
             # rev
             return self.node(id)
         if len(id) == 20:
             # 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 LookupError:
                 pass # may be partial hex id
         try:
             # str(rev)
             rev = int(id)
             if str(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) == 40:
             try:
                 # a full hex nodeid?
                 node = bin(id)
                 self.rev(node)
                 return node
             except (TypeError, LookupError):
                 pass
 
     def _partialmatch(self, id):
         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)]
                 if len(nl) > 0:
                     if len(nl) == 1:
                         self._pcache[id] = nl[0]
                         return nl[0]
                     raise LookupError(id, self.indexfile,
                                       _('ambiguous identifier'))
                 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 LookupError(id, self.indexfile, _('no match found'))
 
     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 hash(text, p1, p2) != node
 
     def _addchunk(self, offset, data):
         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 _loadchunk(self, offset, length):
         if self._inline:
             df = self.opener(self.indexfile)
         else:
             df = self.opener(self.datafile)
 
         readahead = max(65536, length)
         df.seek(offset)
         d = df.read(readahead)
         df.close()
         self._addchunk(offset, d)
         if readahead > length:
             return d[:length]
         return d
 
     def _getchunk(self, offset, length):
         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 d[cachestart:cacheend]
 
         return self._loadchunk(offset, length)
 
     def _chunkraw(self, startrev, endrev):
         start = self.start(startrev)
         length = self.end(endrev) - start
         if self._inline:
             start += (startrev + 1) * self._io.size
         return self._getchunk(start, length)
 
     def _chunk(self, rev):
         return decompress(self._chunkraw(rev, rev))
 
     def _chunkbase(self, rev):
         return self._chunk(rev)
 
     def _chunkclear(self):
         self._chunkcache = (0, '')
 
     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 revdiff(self, rev1, rev2):
         """return or calculate a delta between two revisions"""
         if rev1 != nullrev and self.deltaparent(rev2) == rev1:
             return self._chunk(rev2)
 
         return mdiff.textdiff(self.revision(self.node(rev1)),
                               self.revision(self.node(rev2)))
 
     def revision(self, nodeorrev):
         """return an uncompressed revision of a given node or"""
         if isinstance(nodeorrev, int):
             rev = nodeorrev
             node = self.node(rev)
         else:
             node = nodeorrev
             rev = None
 
         cachedrev = None
         if node == nullid:
             return ""
         if self._cache:
             if self._cache[0] == node:
                 return self._cache[2]
             cachedrev = self._cache[1]
 
         # look up what we need to read
         text = None
         if rev is None:
             rev = self.rev(node)
 
         # check rev flags
         if self.flags(rev) & ~REVIDX_KNOWN_FLAGS:
             raise RevlogError(_('incompatible revision flag %x') %
                               (self.flags(rev) & ~REVIDX_KNOWN_FLAGS))
 
         # build delta chain
         chain = []
         index = self.index # for performance
         generaldelta = self._generaldelta
         iterrev = rev
         e = index[iterrev]
         while iterrev != e[3] and iterrev != cachedrev:
             chain.append(iterrev)
             if generaldelta:
                 iterrev = e[3]
             else:
                 iterrev -= 1
             e = index[iterrev]
         chain.reverse()
         base = iterrev
 
         if iterrev == cachedrev:
             # cache hit
             text = self._cache[2]
 
         # drop cache to save memory
         self._cache = None
 
         self._chunkraw(base, rev)
         if text is None:
             text = self._chunkbase(base)
 
         bins = [self._chunk(r) for r in chain]
         text = mdiff.patches(text, bins)
 
         text = self._checkhash(text, node, rev)
 
         self._cache = (node, rev, text)
         return text
 
     def _checkhash(self, text, node, rev):
         p1, p2 = self.parents(node)
         if node != hash(text, p1, p2):
             raise RevlogError(_("integrity check failed on %s:%d")
                               % (self.indexfile, rev))
         return text
 
     def checkinlinesize(self, tr, fp=None):
         if not self._inline or (self.start(-2) + self.length(-2)) < _maxinline:
             return
 
         trinfo = tr.find(self.indexfile)
         if trinfo is None:
             raise RevlogError(_("%s not found in the transaction")
                               % self.indexfile)
 
         trindex = trinfo[2]
         dataoff = self.start(trindex)
 
         tr.add(self.datafile, dataoff)
 
         if fp:
             fp.flush()
             fp.close()
 
         df = self.opener(self.datafile, 'w')
         try:
             for r in self:
                 df.write(self._chunkraw(r, r))
         finally:
             df.close()
 
         fp = self.opener(self.indexfile, 'w', atomictemp=True)
         self.version &= ~(REVLOGNGINLINEDATA)
         self._inline = False
         for i in self:
             e = self._io.packentry(self.index[i], self.node, self.version, i)
             fp.write(e)
 
         # if we don't call close, the temp file will never replace the
         # real index
         fp.close()
 
         tr.replace(self.indexfile, trindex * self._io.size)
         self._chunkclear()
 
     def addrevision(self, text, transaction, link, p1, p2, cachedelta=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 = hash(text, p1, p2)
         if node in self.nodemap:
             return node
 
         dfh = None
         if not self._inline:
             dfh = self.opener(self.datafile, "a")
         ifh = self.opener(self.indexfile, "a+")
         try:
             return self._addrevision(node, text, transaction, link, p1, p2,
                                      cachedelta, ifh, dfh)
         finally:
             if dfh:
                 dfh.close()
             ifh.close()
 
     def _addrevision(self, node, text, transaction, link, p1, p2,
                      cachedelta, ifh, dfh):
         """internal function to add revisions to the log
 
         see addrevision for argument descriptions.
         invariants:
         - text is optional (can be None); if not set, cachedelta must be set.
           if both are set, they must correspond to eachother.
         """
         btext = [text]
         def buildtext():
             if btext[0] is not None:
                 return btext[0]
             # flush any pending writes here so we can read it in revision
             if dfh:
                 dfh.flush()
             ifh.flush()
             basetext = self.revision(self.node(cachedelta[0]))
             btext[0] = mdiff.patch(basetext, cachedelta[1])
             chk = hash(btext[0], p1, p2)
             if chk != node:
                 raise RevlogError(_("consistency error in delta"))
             return btext[0]
 
         def builddelta(rev):
             # can we use the cached delta?
             if cachedelta and cachedelta[0] == rev:
                 delta = cachedelta[1]
             else:
                 t = buildtext()
                 ptext = self.revision(self.node(rev))
                 delta = mdiff.textdiff(ptext, t)
             data = compress(delta)
             l = len(data[1]) + len(data[0])
             if basecache[0] == rev:
                 chainbase = basecache[1]
             else:
                 chainbase = self.chainbase(rev)
             dist = l + offset - self.start(chainbase)
             if self._generaldelta:
                 base = rev
             else:
                 base = chainbase
             return dist, l, data, base, chainbase
 
         curr = len(self)
         prev = curr - 1
         base = chainbase = curr
         offset = self.end(prev)
         flags = 0
         d = None
         basecache = self._basecache
         p1r, p2r = self.rev(p1), self.rev(p2)
 
         # should we try to build a delta?
         if prev != nullrev:
             if self._generaldelta:
                 if p1r >= basecache[1]:
                     d = builddelta(p1r)
                 elif p2r >= basecache[1]:
                     d = builddelta(p2r)
                 else:
                     d = builddelta(prev)
             else:
                 d = builddelta(prev)
             dist, l, data, base, chainbase = d
 
         # full versions are inserted when the needed deltas
         # become comparable to the uncompressed text
         if text is None:
             textlen = mdiff.patchedsize(self.rawsize(cachedelta[0]),
                                         cachedelta[1])
         else:
             textlen = len(text)
         if d is None or dist > textlen * 2:
             text = buildtext()
             data = compress(text)
             l = len(data[1]) + len(data[0])
             base = chainbase = curr
 
         e = (offset_type(offset, flags), l, textlen,
              base, link, p1r, p2r, node)
         self.index.insert(-1, e)
         self.nodemap[node] = curr
 
         entry = self._io.packentry(e, self.node, self.version, curr)
         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])
             dfh.flush()
             ifh.write(entry)
         else:
             offset += curr * self._io.size
             transaction.add(self.indexfile, offset, curr)
             ifh.write(entry)
             ifh.write(data[0])
             ifh.write(data[1])
             self.checkinlinesize(transaction, ifh)
 
         if type(text) == str: # only accept immutable objects
             self._cache = (node, curr, text)
         self._basecache = (curr, chainbase)
         return node
 
     def group(self, nodelist, bundler, reorder=None):
         """Calculate a delta group, yielding a sequence of changegroup chunks
         (strings).
 
         Given a list of changeset revs, return a set of deltas and
         metadata corresponding to nodes. The first delta is
         first parent(nodelist[0]) -> nodelist[0], the receiver is
         guaranteed to have this parent as it has all history before
         these changesets. In the case firstparent is nullrev the
         changegroup starts with a full revision.
         """
 
         # if we don't have any revisions touched by these changesets, bail
         if len(nodelist) == 0:
             yield bundler.close()
             return
 
         # for generaldelta revlogs, we linearize the revs; this will both be
         # much quicker and generate a much smaller bundle
         if (self._generaldelta and reorder is not False) or reorder:
             dag = dagutil.revlogdag(self)
             revs = set(self.rev(n) for n in nodelist)
             revs = dag.linearize(revs)
         else:
             revs = sorted([self.rev(n) for n in nodelist])
 
         # add the parent of the first rev
         p = self.parentrevs(revs[0])[0]
         revs.insert(0, p)
 
         # build deltas
         for r in xrange(len(revs) - 1):
             prev, curr = revs[r], revs[r + 1]
             for c in bundler.revchunk(self, curr, prev):
                 yield c
 
         yield bundler.close()
 
     def addgroup(self, bundle, linkmapper, transaction):
         """
         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.
         """
 
         # track the base of the current delta log
         content = []
         node = None
 
         r = len(self)
         end = 0
         if r:
             end = self.end(r - 1)
         ifh = self.opener(self.indexfile, "a+")
         isize = r * self._io.size
         if self._inline:
             transaction.add(self.indexfile, end + isize, r)
             dfh = None
         else:
             transaction.add(self.indexfile, isize, r)
             transaction.add(self.datafile, end)
             dfh = self.opener(self.datafile, "a")
 
         try:
             # loop through our set of deltas
             chain = None
             while True:
                 chunkdata = bundle.deltachunk(chain)
                 if not chunkdata:
                     break
                 node = chunkdata['node']
                 p1 = chunkdata['p1']
                 p2 = chunkdata['p2']
                 cs = chunkdata['cs']
                 deltabase = chunkdata['deltabase']
                 delta = chunkdata['delta']
 
                 content.append(node)
 
                 link = linkmapper(cs)
                 if node in self.nodemap:
                     # this can happen if two branches make the same change
                     chain = node
                     continue
 
                 for p in (p1, p2):
                     if not p in self.nodemap:
                         raise LookupError(p, self.indexfile,
                                           _('unknown parent'))
 
                 if deltabase not in self.nodemap:
                     raise LookupError(deltabase, self.indexfile,
                                       _('unknown delta base'))
 
                 baserev = self.rev(deltabase)
                 chain = self._addrevision(node, None, transaction, link,
                                           p1, p2, (baserev, delta), ifh, dfh)
                 if not dfh and not self._inline:
                     # addrevision switched from inline to conventional
                     # reopen the index
                     ifh.close()
                     dfh = self.opener(self.datafile, "a")
                     ifh = self.opener(self.indexfile, "a")
         finally:
             if dfh:
                 dfh.close()
             ifh.close()
 
         return content
 
     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
 
         for rev in self:
             if self.index[rev][4] >= minlink:
                 break
         else:
             return
 
         # first truncate the files on disk
         end = self.start(rev)
         if not self._inline:
             transaction.add(self.datafile, end)
             end = rev * self._io.size
         else:
             end += rev * self._io.size
 
         transaction.add(self.indexfile, end)
 
         # then reset internal state in memory to forget those revisions
         self._cache = None
         self._chunkclear()
         for x in xrange(rev, len(self)):
             del self.nodemap[self.node(x)]
 
         del self.index[rev:-1]
 
     def checksize(self):
         expected = 0
         if len(self):
             expected = max(0, self.end(len(self) - 1))
 
         try:
             f = self.opener(self.datafile)
             f.seek(0, 2)
             actual = f.tell()
             f.close()
             dd = actual - expected
         except IOError, inst:
             if inst.errno != errno.ENOENT:
                 raise
             dd = 0
 
         try:
             f = self.opener(self.indexfile)
             f.seek(0, 2)
             actual = f.tell()
             f.close()
             s = self._io.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, 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

tests/test-parseindex2.py

@@ -1,115 +1,122 b''
 from mercurial import parsers
 from mercurial.node import nullid, nullrev
 import struct
 
 # This unit test compares the return value of the original Python
 # implementation of parseindex and the new C implementation for
 # an index file with and without inlined data
 
 # original python implementation
 def gettype(q):
     return int(q & 0xFFFF)
 
 def offset_type(offset, type):
     return long(long(offset) << 16 | type)
 
 indexformatng = ">Qiiiiii20s12x"
 
 def py_parseindex(data, inline) :
     s = 64
     cache = None
     index = []
     nodemap =  {nullid: nullrev}
     n = off = 0
 
     l = len(data) - s
     append = index.append
     if inline:
         cache = (0, data)
         while off <= l:
             e = struct.unpack(indexformatng, data[off:off + s])
             nodemap[e[7]] = n
             append(e)
             n += 1
             if e[1] < 0:
                 break
             off += e[1] + s
     else:
         while off <= l:
             e = struct.unpack(indexformatng, data[off:off + s])
             nodemap[e[7]] = n
             append(e)
             n += 1
             off += s
 
     e = list(index[0])
     type = gettype(e[0])
     e[0] = offset_type(0, type)
     index[0] = tuple(e)
 
     # add the magic null revision at -1
     index.append((0, 0, 0, -1, -1, -1, -1, nullid))
 
     return index, cache
 
 data_inlined = '\x00\x01\x00\x01\x00\x00\x00\x00\x00\x00\x01\x8c' \
     '\x00\x00\x04\x07\x00\x00\x00\x00\x00\x00\x15\x15\xff\xff\xff' \
     '\xff\xff\xff\xff\xff\xebG\x97\xb7\x1fB\x04\xcf\x13V\x81\tw\x1b' \
     'w\xdduR\xda\xc6\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00' \
     'x\x9c\x9d\x93?O\xc30\x10\xc5\xf7|\x8a\xdb\x9a\xa8m\x06\xd8*\x95' \
     '\x81B\xa1\xa2\xa2R\xcb\x86Pd\x9a\x0b5$vd_\x04\xfd\xf6\x9c\xff@' \
     '\x11!\x0b\xd9\xec\xf7\xbbw\xe7gG6\xad6\x04\xdaN\xc0\x92\xa0$)' \
     '\xb1\x82\xa2\xd1%\x16\xa4\x8b7\xa9\xca\xd4-\xb2Y\x02\xfc\xc9' \
     '\xcaS\xf9\xaeX\xed\xb6\xd77Q\x02\x83\xd4\x19\xf5--Y\xea\xe1W' \
     '\xab\xed\x10\xceR\x0f_\xdf\xdf\r\xe1,\xf5\xf0\xcb\xf5 \xceR\x0f' \
     '_\xdc\x0e\x0e\xc3R\x0f_\xae\x96\x9b!\x9e\xa5\x1e\xbf\xdb,\x06' \
     '\xc7q\x9a/\x88\x82\xc3B\xea\xb5\xb4TJ\x93\xb6\x82\x0e\xe16\xe6' \
     'KQ\xdb\xaf\xecG\xa3\xd1 \x01\xd3\x0b_^\xe8\xaa\xa0\xae\xad\xd1' \
     '&\xbef\x1bz\x08\xb0|\xc9Xz\x06\xf6Z\x91\x90J\xaa\x17\x90\xaa' \
     '\xd2\xa6\x11$5C\xcf\xba#\xa0\x03\x02*2\x92-\xfc\xb1\x94\xdf\xe2' \
     '\xae\xb8\'m\x8ey0^\x85\xd3\x82\xb4\xf0`:\x9c\x00\x8a\xfd\x01' \
     '\xb0\xc6\x86\x8b\xdd\xae\x80\xf3\xa9\x9fd\x16\n\x00R%\x1a\x06' \
     '\xe9\xd8b\x98\x1d\xf4\xf3+\x9bf\x01\xd8p\x1b\xf3.\xed\x9f^g\xc3' \
     '^\xd9W81T\xdb\xd5\x04sx|\xf2\xeb\xd6`%?x\xed"\x831\xbf\xf3\xdc' \
     'b\xeb%gaY\xe1\xad\x9f\xb9f\'1w\xa9\xa5a\x83s\x82J\xb98\xbc4\x8b' \
     '\x83\x00\x9f$z\xb8#\xa5\xb1\xdf\x98\xd9\xec\x1b\x89O\xe3Ts\x9a4' \
     '\x17m\x8b\xfc\x8f\xa5\x95\x9a\xfc\xfa\xed,\xe5|\xa1\xfe\x15\xb9' \
     '\xbc\xb2\x93\x1f\xf2\x95\xff\xdf,\x1a\xc5\xe7\x17*\x93Oz:>\x0e'
 
 data_non_inlined = '\x00\x00\x00\x01\x00\x00\x00\x00\x00\x01D\x19' \
     '\x00\x07e\x12\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff\xff\xff' \
     '\xff\xff\xff\xff\xd1\xf4\xbb\xb0\xbe\xfc\x13\xbd\x8c\xd3\x9d' \
     '\x0f\xcd\xd9;\x8c\x07\x8cJ/\x00\x00\x00\x00\x00\x00\x00\x00\x00' \
     '\x00\x00\x00\x00\x00\x00\x01D\x19\x00\x00\x00\x00\x00\xdf\x00' \
     '\x00\x01q\x00\x00\x00\x01\x00\x00\x00\x01\x00\x00\x00\x00\xff' \
     '\xff\xff\xff\xc1\x12\xb9\x04\x96\xa4Z1t\x91\xdfsJ\x90\xf0\x9bh' \
     '\x07l&\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00' \
     '\x00\x01D\xf8\x00\x00\x00\x00\x01\x1b\x00\x00\x01\xb8\x00\x00' \
     '\x00\x01\x00\x00\x00\x02\x00\x00\x00\x01\xff\xff\xff\xff\x02\n' \
     '\x0e\xc6&\xa1\x92\xae6\x0b\x02i\xfe-\xe5\xbao\x05\xd1\xe7\x00' \
     '\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01F' \
     '\x13\x00\x00\x00\x00\x01\xec\x00\x00\x03\x06\x00\x00\x00\x01' \
     '\x00\x00\x00\x03\x00\x00\x00\x02\xff\xff\xff\xff\x12\xcb\xeby1' \
     '\xb6\r\x98B\xcb\x07\xbd`\x8f\x92\xd9\xc4\x84\xbdK\x00\x00\x00' \
     '\x00\x00\x00\x00\x00\x00\x00\x00\x00'
 
 def parse_index2(data, inline):
     index, chunkcache = parsers.parse_index2(data, inline)
     return list(index), chunkcache
 
 def runtest() :
     py_res_1 = py_parseindex(data_inlined, True)
     c_res_1 = parse_index2(data_inlined, True)
 
     py_res_2 = py_parseindex(data_non_inlined, False)
     c_res_2 = parse_index2(data_non_inlined, False)
 
     if py_res_1 != c_res_1:
         print "Parse index result (with inlined data) differs!"
 
     if py_res_2 != c_res_2:
         print "Parse index result (no inlined data) differs!"
 
+    ix = parsers.parse_index2(data_inlined, True)[0]
+    for i, r in enumerate(ix):
+        if r[7] == nullid:
+            i = -1
+        if ix[r[7]] != i:
+            print 'Reverse lookup inconsistent for %r' % r[7].encode('hex')
+
     print "done"
 
 runtest()