dirstate: use more than a bool to control append behavior

When writing dirstate-v2, we might either append to the existing file, or
create a new file.

We are about to introduce some configuration to control this behavior.
As a prelude, we change the current way the behavior was automatically
controlled to make the change smaller/clearer.

mercurial/dirstatemap.py

@@ -1,688 +1,693 b''
 # dirstatemap.py
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 
 from .i18n import _
 
 from . import (
     error,
     pathutil,
     policy,
     txnutil,
     util,
 )
 
 from .dirstateutils import (
     docket as docketmod,
     v2,
 )
 
 parsers = policy.importmod('parsers')
 rustmod = policy.importrust('dirstate')
 
 propertycache = util.propertycache
 
 if rustmod is None:
     DirstateItem = parsers.DirstateItem
 else:
     DirstateItem = rustmod.DirstateItem
 
 rangemask = 0x7FFFFFFF
 
+WRITE_MODE_AUTO = 0
+WRITE_MODE_FORCE_NEW = 1
+
 
 class _dirstatemapcommon:
     """
     Methods that are identical for both implementations of the dirstatemap
     class, with and without Rust extensions enabled.
     """
 
     # please pytype
 
     _map = None
     copymap = None
 
     def __init__(self, ui, opener, root, nodeconstants, use_dirstate_v2):
         self._use_dirstate_v2 = use_dirstate_v2
         self._nodeconstants = nodeconstants
         self._ui = ui
         self._opener = opener
         self._root = root
         self._filename = b'dirstate'
         self._nodelen = 20  # Also update Rust code when changing this!
         self._parents = None
         self._dirtyparents = False
         self._docket = None
 
         # for consistent view between _pl() and _read() invocations
         self._pendingmode = None
 
     def preload(self):
         """Loads the underlying data, if it's not already loaded"""
         self._map
 
     def get(self, key, default=None):
         return self._map.get(key, default)
 
     def __len__(self):
         return len(self._map)
 
     def __iter__(self):
         return iter(self._map)
 
     def __contains__(self, key):
         return key in self._map
 
     def __getitem__(self, item):
         return self._map[item]
 
     ### disk interaction
 
     def _opendirstatefile(self):
         fp, mode = txnutil.trypending(self._root, self._opener, self._filename)
         if self._pendingmode is not None and self._pendingmode != mode:
             fp.close()
             raise error.Abort(
                 _(b'working directory state may be changed parallelly')
             )
         self._pendingmode = mode
         return fp
 
     def _readdirstatefile(self, size=-1):
         try:
             with self._opendirstatefile() as fp:
                 return fp.read(size)
         except FileNotFoundError:
             # File doesn't exist, so the current state is empty
             return b''
 
     @property
     def docket(self):
         if not self._docket:
             if not self._use_dirstate_v2:
                 raise error.ProgrammingError(
                     b'dirstate only has a docket in v2 format'
                 )
             self._docket = docketmod.DirstateDocket.parse(
                 self._readdirstatefile(), self._nodeconstants
             )
         return self._docket
 
     def write_v2_no_append(self, tr, st, meta, packed):
         old_docket = self.docket
         new_docket = docketmod.DirstateDocket.with_new_uuid(
             self.parents(), len(packed), meta
         )
         if old_docket.uuid == new_docket.uuid:
             raise error.ProgrammingError(b'dirstate docket name collision')
         data_filename = new_docket.data_filename()
         self._opener.write(data_filename, packed)
         # Write the new docket after the new data file has been
         # written. Because `st` was opened with `atomictemp=True`,
         # the actual `.hg/dirstate` file is only affected on close.
         st.write(new_docket.serialize())
         st.close()
         # Remove the old data file after the new docket pointing to
         # the new data file was written.
         if old_docket.uuid:
             data_filename = old_docket.data_filename()
             unlink = lambda _tr=None: self._opener.unlink(data_filename)
             if tr:
                 category = b"dirstate-v2-clean-" + old_docket.uuid
                 tr.addpostclose(category, unlink)
             else:
                 unlink()
         self._docket = new_docket
 
     ### reading/setting parents
 
     def parents(self):
         if not self._parents:
             if self._use_dirstate_v2:
                 self._parents = self.docket.parents
             else:
                 read_len = self._nodelen * 2
                 st = self._readdirstatefile(read_len)
                 l = len(st)
                 if l == read_len:
                     self._parents = (
                         st[: self._nodelen],
                         st[self._nodelen : 2 * self._nodelen],
                     )
                 elif l == 0:
                     self._parents = (
                         self._nodeconstants.nullid,
                         self._nodeconstants.nullid,
                     )
                 else:
                     raise error.Abort(
                         _(b'working directory state appears damaged!')
                     )
 
         return self._parents
 
 
 class dirstatemap(_dirstatemapcommon):
     """Map encapsulating the dirstate's contents.
 
     The dirstate contains the following state:
 
     - `identity` is the identity of the dirstate file, which can be used to
       detect when changes have occurred to the dirstate file.
 
     - `parents` is a pair containing the parents of the working copy. The
       parents are updated by calling `setparents`.
 
     - the state map maps filenames to tuples of (state, mode, size, mtime),
       where state is a single character representing 'normal', 'added',
       'removed', or 'merged'. It is read by treating the dirstate as a
       dict.  File state is updated by calling various methods (see each
       documentation for details):
 
       - `reset_state`,
       - `set_tracked`
       - `set_untracked`
       - `set_clean`
       - `set_possibly_dirty`
 
     - `copymap` maps destination filenames to their source filename.
 
     The dirstate also provides the following views onto the state:
 
     - `filefoldmap` is a dict mapping normalized filenames to the denormalized
       form that they appear as in the dirstate.
 
     - `dirfoldmap` is a dict mapping normalized directory names to the
       denormalized form that they appear as in the dirstate.
     """
 
     ### Core data storage and access
 
     @propertycache
     def _map(self):
         self._map = {}
         self.read()
         return self._map
 
     @propertycache
     def copymap(self):
         self.copymap = {}
         self._map
         return self.copymap
 
     def clear(self):
         self._map.clear()
         self.copymap.clear()
         self.setparents(self._nodeconstants.nullid, self._nodeconstants.nullid)
         util.clearcachedproperty(self, b"_dirs")
         util.clearcachedproperty(self, b"_alldirs")
         util.clearcachedproperty(self, b"filefoldmap")
         util.clearcachedproperty(self, b"dirfoldmap")
 
     def items(self):
         return self._map.items()
 
     # forward for python2,3 compat
     iteritems = items
 
     def debug_iter(self, all):
         """
         Return an iterator of (filename, state, mode, size, mtime) tuples
 
         `all` is unused when Rust is not enabled
         """
         for (filename, item) in self.items():
             yield (filename, item.state, item.mode, item.size, item.mtime)
 
     def keys(self):
         return self._map.keys()
 
     ### reading/setting parents
 
     def setparents(self, p1, p2, fold_p2=False):
         self._parents = (p1, p2)
         self._dirtyparents = True
         copies = {}
         if fold_p2:
             for f, s in self._map.items():
                 # Discard "merged" markers when moving away from a merge state
                 if s.p2_info:
                     source = self.copymap.pop(f, None)
                     if source:
                         copies[f] = source
                     s.drop_merge_data()
         return copies
 
     ### disk interaction
 
     def read(self):
         # ignore HG_PENDING because identity is used only for writing
         self.identity = util.filestat.frompath(
             self._opener.join(self._filename)
         )
 
         if self._use_dirstate_v2:
             if not self.docket.uuid:
                 return
             st = self._opener.read(self.docket.data_filename())
         else:
             st = self._readdirstatefile()
 
         if not st:
             return
 
         # TODO: adjust this estimate for dirstate-v2
         if util.safehasattr(parsers, b'dict_new_presized'):
             # Make an estimate of the number of files in the dirstate based on
             # its size. This trades wasting some memory for avoiding costly
             # resizes. Each entry have a prefix of 17 bytes followed by one or
             # two path names. Studies on various large-scale real-world repositories
             # found 54 bytes a reasonable upper limit for the average path names.
             # Copy entries are ignored for the sake of this estimate.
             self._map = parsers.dict_new_presized(len(st) // 71)
 
         # Python's garbage collector triggers a GC each time a certain number
         # of container objects (the number being defined by
         # gc.get_threshold()) are allocated. parse_dirstate creates a tuple
         # for each file in the dirstate. The C version then immediately marks
         # them as not to be tracked by the collector. However, this has no
         # effect on when GCs are triggered, only on what objects the GC looks
         # into. This means that O(number of files) GCs are unavoidable.
         # Depending on when in the process's lifetime the dirstate is parsed,
         # this can get very expensive. As a workaround, disable GC while
         # parsing the dirstate.
         #
         # (we cannot decorate the function directly since it is in a C module)
         if self._use_dirstate_v2:
             p = self.docket.parents
             meta = self.docket.tree_metadata
             parse_dirstate = util.nogc(v2.parse_dirstate)
             parse_dirstate(self._map, self.copymap, st, meta)
         else:
             parse_dirstate = util.nogc(parsers.parse_dirstate)
             p = parse_dirstate(self._map, self.copymap, st)
         if not self._dirtyparents:
             self.setparents(*p)
 
         # Avoid excess attribute lookups by fast pathing certain checks
         self.__contains__ = self._map.__contains__
         self.__getitem__ = self._map.__getitem__
         self.get = self._map.get
 
     def write(self, tr, st):
         if self._use_dirstate_v2:
             packed, meta = v2.pack_dirstate(self._map, self.copymap)
             self.write_v2_no_append(tr, st, meta, packed)
         else:
             packed = parsers.pack_dirstate(
                 self._map, self.copymap, self.parents()
             )
             st.write(packed)
             st.close()
         self._dirtyparents = False
 
     @propertycache
     def identity(self):
         self._map
         return self.identity
 
     ### code related to maintaining and accessing "extra" property
     # (e.g. "has_dir")
 
     def _dirs_incr(self, filename, old_entry=None):
         """increment the dirstate counter if applicable"""
         if (
             old_entry is None or old_entry.removed
         ) and "_dirs" in self.__dict__:
             self._dirs.addpath(filename)
         if old_entry is None and "_alldirs" in self.__dict__:
             self._alldirs.addpath(filename)
 
     def _dirs_decr(self, filename, old_entry=None, remove_variant=False):
         """decrement the dirstate counter if applicable"""
         if old_entry is not None:
             if "_dirs" in self.__dict__ and not old_entry.removed:
                 self._dirs.delpath(filename)
             if "_alldirs" in self.__dict__ and not remove_variant:
                 self._alldirs.delpath(filename)
         elif remove_variant and "_alldirs" in self.__dict__:
             self._alldirs.addpath(filename)
         if "filefoldmap" in self.__dict__:
             normed = util.normcase(filename)
             self.filefoldmap.pop(normed, None)
 
     @propertycache
     def filefoldmap(self):
         """Returns a dictionary mapping normalized case paths to their
         non-normalized versions.
         """
         try:
             makefilefoldmap = parsers.make_file_foldmap
         except AttributeError:
             pass
         else:
             return makefilefoldmap(
                 self._map, util.normcasespec, util.normcasefallback
             )
 
         f = {}
         normcase = util.normcase
         for name, s in self._map.items():
             if not s.removed:
                 f[normcase(name)] = name
         f[b'.'] = b'.'  # prevents useless util.fspath() invocation
         return f
 
     @propertycache
     def dirfoldmap(self):
         f = {}
         normcase = util.normcase
         for name in self._dirs:
             f[normcase(name)] = name
         return f
 
     def hastrackeddir(self, d):
         """
         Returns True if the dirstate contains a tracked (not removed) file
         in this directory.
         """
         return d in self._dirs
 
     def hasdir(self, d):
         """
         Returns True if the dirstate contains a file (tracked or removed)
         in this directory.
         """
         return d in self._alldirs
 
     @propertycache
     def _dirs(self):
         return pathutil.dirs(self._map, only_tracked=True)
 
     @propertycache
     def _alldirs(self):
         return pathutil.dirs(self._map)
 
     ### code related to manipulation of entries and copy-sources
 
     def reset_state(
         self,
         filename,
         wc_tracked=False,
         p1_tracked=False,
         p2_info=False,
         has_meaningful_mtime=True,
         parentfiledata=None,
     ):
         """Set a entry to a given state, diregarding all previous state
 
         This is to be used by the part of the dirstate API dedicated to
         adjusting the dirstate after a update/merge.
 
         note: calling this might result to no entry existing at all if the
         dirstate map does not see any point at having one for this file
         anymore.
         """
         # copy information are now outdated
         # (maybe new information should be in directly passed to this function)
         self.copymap.pop(filename, None)
 
         if not (p1_tracked or p2_info or wc_tracked):
             old_entry = self._map.get(filename)
             self._drop_entry(filename)
             self._dirs_decr(filename, old_entry=old_entry)
             return
 
         old_entry = self._map.get(filename)
         self._dirs_incr(filename, old_entry)
         entry = DirstateItem(
             wc_tracked=wc_tracked,
             p1_tracked=p1_tracked,
             p2_info=p2_info,
             has_meaningful_mtime=has_meaningful_mtime,
             parentfiledata=parentfiledata,
         )
         self._map[filename] = entry
 
     def set_tracked(self, filename):
         new = False
         entry = self.get(filename)
         if entry is None:
             self._dirs_incr(filename)
             entry = DirstateItem(
                 wc_tracked=True,
             )
 
             self._map[filename] = entry
             new = True
         elif not entry.tracked:
             self._dirs_incr(filename, entry)
             entry.set_tracked()
             self._refresh_entry(filename, entry)
             new = True
         else:
             # XXX This is probably overkill for more case, but we need this to
             # fully replace the `normallookup` call with `set_tracked` one.
             # Consider smoothing this in the future.
             entry.set_possibly_dirty()
             self._refresh_entry(filename, entry)
         return new
 
     def set_untracked(self, f):
         """Mark a file as no longer tracked in the dirstate map"""
         entry = self.get(f)
         if entry is None:
             return False
         else:
             self._dirs_decr(f, old_entry=entry, remove_variant=not entry.added)
             if not entry.p2_info:
                 self.copymap.pop(f, None)
             entry.set_untracked()
             self._refresh_entry(f, entry)
             return True
 
     def set_clean(self, filename, mode, size, mtime):
         """mark a file as back to a clean state"""
         entry = self[filename]
         size = size & rangemask
         entry.set_clean(mode, size, mtime)
         self._refresh_entry(filename, entry)
         self.copymap.pop(filename, None)
 
     def set_possibly_dirty(self, filename):
         """record that the current state of the file on disk is unknown"""
         entry = self[filename]
         entry.set_possibly_dirty()
         self._refresh_entry(filename, entry)
 
     def _refresh_entry(self, f, entry):
         """record updated state of an entry"""
         if not entry.any_tracked:
             self._map.pop(f, None)
 
     def _drop_entry(self, f):
         """remove any entry for file f
 
         This should also drop associated copy information
 
         The fact we actually need to drop it is the responsability of the caller"""
         self._map.pop(f, None)
         self.copymap.pop(f, None)
 
 
 if rustmod is not None:
 
     class dirstatemap(_dirstatemapcommon):
 
         ### Core data storage and access
 
         @propertycache
         def _map(self):
             """
             Fills the Dirstatemap when called.
             """
             # ignore HG_PENDING because identity is used only for writing
             self.identity = util.filestat.frompath(
                 self._opener.join(self._filename)
             )
 
             if self._use_dirstate_v2:
                 if self.docket.uuid:
                     # TODO: use mmap when possible
                     data = self._opener.read(self.docket.data_filename())
                 else:
                     data = b''
                 self._map = rustmod.DirstateMap.new_v2(
                     data, self.docket.data_size, self.docket.tree_metadata
                 )
                 parents = self.docket.parents
             else:
                 self._map, parents = rustmod.DirstateMap.new_v1(
                     self._readdirstatefile()
                 )
 
             if parents and not self._dirtyparents:
                 self.setparents(*parents)
 
             self.__contains__ = self._map.__contains__
             self.__getitem__ = self._map.__getitem__
             self.get = self._map.get
             return self._map
 
         @property
         def copymap(self):
             return self._map.copymap()
 
         def debug_iter(self, all):
             """
             Return an iterator of (filename, state, mode, size, mtime) tuples
 
             `all`: also include with `state == b' '` dirstate tree nodes that
             don't have an associated `DirstateItem`.
 
             """
             return self._map.debug_iter(all)
 
         def clear(self):
             self._map.clear()
             self.setparents(
                 self._nodeconstants.nullid, self._nodeconstants.nullid
             )
             util.clearcachedproperty(self, b"_dirs")
             util.clearcachedproperty(self, b"_alldirs")
             util.clearcachedproperty(self, b"dirfoldmap")
 
         def items(self):
             return self._map.items()
 
         # forward for python2,3 compat
         iteritems = items
 
         def keys(self):
             return iter(self._map)
 
         ### reading/setting parents
 
         def setparents(self, p1, p2, fold_p2=False):
             self._parents = (p1, p2)
             self._dirtyparents = True
             copies = {}
             if fold_p2:
                 copies = self._map.setparents_fixup()
             return copies
 
         ### disk interaction
 
         @propertycache
         def identity(self):
             self._map
             return self.identity
 
         def write(self, tr, st):
             if not self._use_dirstate_v2:
                 p1, p2 = self.parents()
                 packed = self._map.write_v1(p1, p2)
                 st.write(packed)
                 st.close()
                 self._dirtyparents = False
                 return
 
             # We can only append to an existing data file if there is one
-            can_append = self.docket.uuid is not None
-            packed, meta, append = self._map.write_v2(can_append)
+            write_mode = WRITE_MODE_AUTO
+            if self.docket.uuid is None:
+                write_mode = WRITE_MODE_FORCE_NEW
+            packed, meta, append = self._map.write_v2(write_mode)
             if append:
                 docket = self.docket
                 data_filename = docket.data_filename()
                 with self._opener(data_filename, b'r+b') as fp:
                     fp.seek(docket.data_size)
                     assert fp.tell() == docket.data_size
                     written = fp.write(packed)
                     if written is not None:  # py2 may return None
                         assert written == len(packed), (written, len(packed))
                 docket.data_size += len(packed)
                 docket.parents = self.parents()
                 docket.tree_metadata = meta
                 st.write(docket.serialize())
                 st.close()
             else:
                 self.write_v2_no_append(tr, st, meta, packed)
             # Reload from the newly-written file
             util.clearcachedproperty(self, b"_map")
             self._dirtyparents = False
 
         ### code related to maintaining and accessing "extra" property
         # (e.g. "has_dir")
 
         @propertycache
         def filefoldmap(self):
             """Returns a dictionary mapping normalized case paths to their
             non-normalized versions.
             """
             return self._map.filefoldmapasdict()
 
         def hastrackeddir(self, d):
             return self._map.hastrackeddir(d)
 
         def hasdir(self, d):
             return self._map.hasdir(d)
 
         @propertycache
         def dirfoldmap(self):
             f = {}
             normcase = util.normcase
             for name in self._map.tracked_dirs():
                 f[normcase(name)] = name
             return f
 
         ### code related to manipulation of entries and copy-sources
 
         def set_tracked(self, f):
             return self._map.set_tracked(f)
 
         def set_untracked(self, f):
             return self._map.set_untracked(f)
 
         def set_clean(self, filename, mode, size, mtime):
             self._map.set_clean(filename, mode, size, mtime)
 
         def set_possibly_dirty(self, f):
             self._map.set_possibly_dirty(f)
 
         def reset_state(
             self,
             filename,
             wc_tracked=False,
             p1_tracked=False,
             p2_info=False,
             has_meaningful_mtime=True,
             parentfiledata=None,
         ):
             return self._map.reset_state(
                 filename,
                 wc_tracked,
                 p1_tracked,
                 p2_info,
                 has_meaningful_mtime,
                 parentfiledata,
             )

rust/hg-core/src/dirstate_tree/dirstate_map.rs

@@ -1,1907 +1,1913 b''
 use bytes_cast::BytesCast;
 use micro_timer::timed;
 use std::borrow::Cow;
 use std::path::PathBuf;
 
 use super::on_disk;
 use super::on_disk::DirstateV2ParseError;
 use super::owning::OwningDirstateMap;
 use super::path_with_basename::WithBasename;
 use crate::dirstate::parsers::pack_entry;
 use crate::dirstate::parsers::packed_entry_size;
 use crate::dirstate::parsers::parse_dirstate_entries;
 use crate::dirstate::CopyMapIter;
 use crate::dirstate::DirstateV2Data;
 use crate::dirstate::ParentFileData;
 use crate::dirstate::StateMapIter;
 use crate::dirstate::TruncatedTimestamp;
 use crate::matchers::Matcher;
 use crate::utils::hg_path::{HgPath, HgPathBuf};
 use crate::DirstateEntry;
 use crate::DirstateError;
 use crate::DirstateMapError;
 use crate::DirstateParents;
 use crate::DirstateStatus;
 use crate::FastHashbrownMap as FastHashMap;
 use crate::PatternFileWarning;
 use crate::StatusError;
 use crate::StatusOptions;
 
 /// Append to an existing data file if the amount of unreachable data (not used
 /// anymore) is less than this fraction of the total amount of existing data.
 const ACCEPTABLE_UNREACHABLE_BYTES_RATIO: f32 = 0.5;
 
 #[derive(Debug, PartialEq, Eq)]
 /// Version of the on-disk format
 pub enum DirstateVersion {
     V1,
     V2,
 }
 
+#[derive(Debug, PartialEq, Eq)]
+pub enum DirstateMapWriteMode {
+    Auto,
+    ForceNewDataFile,
+}
+
 #[derive(Debug)]
 pub struct DirstateMap<'on_disk> {
     /// Contents of the `.hg/dirstate` file
     pub(super) on_disk: &'on_disk [u8],
 
     pub(super) root: ChildNodes<'on_disk>,
 
     /// Number of nodes anywhere in the tree that have `.entry.is_some()`.
     pub(super) nodes_with_entry_count: u32,
 
     /// Number of nodes anywhere in the tree that have
     /// `.copy_source.is_some()`.
     pub(super) nodes_with_copy_source_count: u32,
 
     /// See on_disk::Header
     pub(super) ignore_patterns_hash: on_disk::IgnorePatternsHash,
 
     /// How many bytes of `on_disk` are not used anymore
     pub(super) unreachable_bytes: u32,
 
     /// Size of the data used to first load this `DirstateMap`. Used in case
     /// we need to write some new metadata, but no new data on disk.
     pub(super) old_data_size: usize,
 
     pub(super) dirstate_version: DirstateVersion,
 }
 
 /// Using a plain `HgPathBuf` of the full path from the repository root as a
 /// map key would also work: all paths in a given map have the same parent
 /// path, so comparing full paths gives the same result as comparing base
 /// names. However `HashMap` would waste time always re-hashing the same
 /// string prefix.
 pub(super) type NodeKey<'on_disk> = WithBasename<Cow<'on_disk, HgPath>>;
 
 /// Similar to `&'tree Cow<'on_disk, HgPath>`, but can also be returned
 /// for on-disk nodes that don’t actually have a `Cow` to borrow.
 #[derive(Debug)]
 pub(super) enum BorrowedPath<'tree, 'on_disk> {
     InMemory(&'tree HgPathBuf),
     OnDisk(&'on_disk HgPath),
 }
 
 #[derive(Debug)]
 pub(super) enum ChildNodes<'on_disk> {
     InMemory(FastHashMap<NodeKey<'on_disk>, Node<'on_disk>>),
     OnDisk(&'on_disk [on_disk::Node]),
 }
 
 #[derive(Debug)]
 pub(super) enum ChildNodesRef<'tree, 'on_disk> {
     InMemory(&'tree FastHashMap<NodeKey<'on_disk>, Node<'on_disk>>),
     OnDisk(&'on_disk [on_disk::Node]),
 }
 
 #[derive(Debug)]
 pub(super) enum NodeRef<'tree, 'on_disk> {
     InMemory(&'tree NodeKey<'on_disk>, &'tree Node<'on_disk>),
     OnDisk(&'on_disk on_disk::Node),
 }
 
 impl<'tree, 'on_disk> BorrowedPath<'tree, 'on_disk> {
     pub fn detach_from_tree(&self) -> Cow<'on_disk, HgPath> {
         match *self {
             BorrowedPath::InMemory(in_memory) => Cow::Owned(in_memory.clone()),
             BorrowedPath::OnDisk(on_disk) => Cow::Borrowed(on_disk),
         }
     }
 }
 
 impl<'tree, 'on_disk> std::ops::Deref for BorrowedPath<'tree, 'on_disk> {
     type Target = HgPath;
 
     fn deref(&self) -> &HgPath {
         match *self {
             BorrowedPath::InMemory(in_memory) => in_memory,
             BorrowedPath::OnDisk(on_disk) => on_disk,
         }
     }
 }
 
 impl Default for ChildNodes<'_> {
     fn default() -> Self {
         ChildNodes::InMemory(Default::default())
     }
 }
 
 impl<'on_disk> ChildNodes<'on_disk> {
     pub(super) fn as_ref<'tree>(
         &'tree self,
     ) -> ChildNodesRef<'tree, 'on_disk> {
         match self {
             ChildNodes::InMemory(nodes) => ChildNodesRef::InMemory(nodes),
             ChildNodes::OnDisk(nodes) => ChildNodesRef::OnDisk(nodes),
         }
     }
 
     pub(super) fn is_empty(&self) -> bool {
         match self {
             ChildNodes::InMemory(nodes) => nodes.is_empty(),
             ChildNodes::OnDisk(nodes) => nodes.is_empty(),
         }
     }
 
     fn make_mut(
         &mut self,
         on_disk: &'on_disk [u8],
         unreachable_bytes: &mut u32,
     ) -> Result<
         &mut FastHashMap<NodeKey<'on_disk>, Node<'on_disk>>,
         DirstateV2ParseError,
     > {
         match self {
             ChildNodes::InMemory(nodes) => Ok(nodes),
             ChildNodes::OnDisk(nodes) => {
                 *unreachable_bytes +=
                     std::mem::size_of_val::<[on_disk::Node]>(nodes) as u32;
                 let nodes = nodes
                     .iter()
                     .map(|node| {
                         Ok((
                             node.path(on_disk)?,
                             node.to_in_memory_node(on_disk)?,
                         ))
                     })
                     .collect::<Result<_, _>>()?;
                 *self = ChildNodes::InMemory(nodes);
                 match self {
                     ChildNodes::InMemory(nodes) => Ok(nodes),
                     ChildNodes::OnDisk(_) => unreachable!(),
                 }
             }
         }
     }
 }
 
 impl<'tree, 'on_disk> ChildNodesRef<'tree, 'on_disk> {
     pub(super) fn get(
         &self,
         base_name: &HgPath,
         on_disk: &'on_disk [u8],
     ) -> Result<Option<NodeRef<'tree, 'on_disk>>, DirstateV2ParseError> {
         match self {
             ChildNodesRef::InMemory(nodes) => Ok(nodes
                 .get_key_value(base_name)
                 .map(|(k, v)| NodeRef::InMemory(k, v))),
             ChildNodesRef::OnDisk(nodes) => {
                 let mut parse_result = Ok(());
                 let search_result = nodes.binary_search_by(|node| {
                     match node.base_name(on_disk) {
                         Ok(node_base_name) => node_base_name.cmp(base_name),
                         Err(e) => {
                             parse_result = Err(e);
                             // Dummy comparison result, `search_result` won’t
                             // be used since `parse_result` is an error
                             std::cmp::Ordering::Equal
                         }
                     }
                 });
                 parse_result.map(|()| {
                     search_result.ok().map(|i| NodeRef::OnDisk(&nodes[i]))
                 })
             }
         }
     }
 
     /// Iterate in undefined order
     pub(super) fn iter(
         &self,
     ) -> impl Iterator<Item = NodeRef<'tree, 'on_disk>> {
         match self {
             ChildNodesRef::InMemory(nodes) => itertools::Either::Left(
                 nodes.iter().map(|(k, v)| NodeRef::InMemory(k, v)),
             ),
             ChildNodesRef::OnDisk(nodes) => {
                 itertools::Either::Right(nodes.iter().map(NodeRef::OnDisk))
             }
         }
     }
 
     /// Iterate in parallel in undefined order
     pub(super) fn par_iter(
         &self,
     ) -> impl rayon::iter::ParallelIterator<Item = NodeRef<'tree, 'on_disk>>
     {
         use rayon::prelude::*;
         match self {
             ChildNodesRef::InMemory(nodes) => rayon::iter::Either::Left(
                 nodes.par_iter().map(|(k, v)| NodeRef::InMemory(k, v)),
             ),
             ChildNodesRef::OnDisk(nodes) => rayon::iter::Either::Right(
                 nodes.par_iter().map(NodeRef::OnDisk),
             ),
         }
     }
 
     pub(super) fn sorted(&self) -> Vec<NodeRef<'tree, 'on_disk>> {
         match self {
             ChildNodesRef::InMemory(nodes) => {
                 let mut vec: Vec<_> = nodes
                     .iter()
                     .map(|(k, v)| NodeRef::InMemory(k, v))
                     .collect();
                 fn sort_key<'a>(node: &'a NodeRef) -> &'a HgPath {
                     match node {
                         NodeRef::InMemory(path, _node) => path.base_name(),
                         NodeRef::OnDisk(_) => unreachable!(),
                     }
                 }
                 // `sort_unstable_by_key` doesn’t allow keys borrowing from the
                 // value: https://github.com/rust-lang/rust/issues/34162
                 vec.sort_unstable_by(|a, b| sort_key(a).cmp(sort_key(b)));
                 vec
             }
             ChildNodesRef::OnDisk(nodes) => {
                 // Nodes on disk are already sorted
                 nodes.iter().map(NodeRef::OnDisk).collect()
             }
         }
     }
 }
 
 impl<'tree, 'on_disk> NodeRef<'tree, 'on_disk> {
     pub(super) fn full_path(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<&'tree HgPath, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(path, _node) => Ok(path.full_path()),
             NodeRef::OnDisk(node) => node.full_path(on_disk),
         }
     }
 
     /// Returns a `BorrowedPath`, which can be turned into a `Cow<'on_disk,
     /// HgPath>` detached from `'tree`
     pub(super) fn full_path_borrowed(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<BorrowedPath<'tree, 'on_disk>, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(path, _node) => match path.full_path() {
                 Cow::Borrowed(on_disk) => Ok(BorrowedPath::OnDisk(on_disk)),
                 Cow::Owned(in_memory) => Ok(BorrowedPath::InMemory(in_memory)),
             },
             NodeRef::OnDisk(node) => {
                 Ok(BorrowedPath::OnDisk(node.full_path(on_disk)?))
             }
         }
     }
 
     pub(super) fn base_name(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<&'tree HgPath, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(path, _node) => Ok(path.base_name()),
             NodeRef::OnDisk(node) => node.base_name(on_disk),
         }
     }
 
     pub(super) fn children(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<ChildNodesRef<'tree, 'on_disk>, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(_path, node) => Ok(node.children.as_ref()),
             NodeRef::OnDisk(node) => {
                 Ok(ChildNodesRef::OnDisk(node.children(on_disk)?))
             }
         }
     }
 
     pub(super) fn has_copy_source(&self) -> bool {
         match self {
             NodeRef::InMemory(_path, node) => node.copy_source.is_some(),
             NodeRef::OnDisk(node) => node.has_copy_source(),
         }
     }
 
     pub(super) fn copy_source(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<Option<&'tree HgPath>, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(_path, node) => {
                 Ok(node.copy_source.as_ref().map(|s| &**s))
             }
             NodeRef::OnDisk(node) => node.copy_source(on_disk),
         }
     }
     /// Returns a `BorrowedPath`, which can be turned into a `Cow<'on_disk,
     /// HgPath>` detached from `'tree`
     pub(super) fn copy_source_borrowed(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<Option<BorrowedPath<'tree, 'on_disk>>, DirstateV2ParseError>
     {
         Ok(match self {
             NodeRef::InMemory(_path, node) => {
                 node.copy_source.as_ref().map(|source| match source {
                     Cow::Borrowed(on_disk) => BorrowedPath::OnDisk(on_disk),
                     Cow::Owned(in_memory) => BorrowedPath::InMemory(in_memory),
                 })
             }
             NodeRef::OnDisk(node) => node
                 .copy_source(on_disk)?
                 .map(|source| BorrowedPath::OnDisk(source)),
         })
     }
 
     pub(super) fn entry(
         &self,
     ) -> Result<Option<DirstateEntry>, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(_path, node) => {
                 Ok(node.data.as_entry().copied())
             }
             NodeRef::OnDisk(node) => node.entry(),
         }
     }
 
     pub(super) fn cached_directory_mtime(
         &self,
     ) -> Result<Option<TruncatedTimestamp>, DirstateV2ParseError> {
         match self {
             NodeRef::InMemory(_path, node) => Ok(match node.data {
                 NodeData::CachedDirectory { mtime } => Some(mtime),
                 _ => None,
             }),
             NodeRef::OnDisk(node) => node.cached_directory_mtime(),
         }
     }
 
     pub(super) fn descendants_with_entry_count(&self) -> u32 {
         match self {
             NodeRef::InMemory(_path, node) => {
                 node.descendants_with_entry_count
             }
             NodeRef::OnDisk(node) => node.descendants_with_entry_count.get(),
         }
     }
 
     pub(super) fn tracked_descendants_count(&self) -> u32 {
         match self {
             NodeRef::InMemory(_path, node) => node.tracked_descendants_count,
             NodeRef::OnDisk(node) => node.tracked_descendants_count.get(),
         }
     }
 }
 
 /// Represents a file or a directory
 #[derive(Default, Debug)]
 pub(super) struct Node<'on_disk> {
     pub(super) data: NodeData,
 
     pub(super) copy_source: Option<Cow<'on_disk, HgPath>>,
 
     pub(super) children: ChildNodes<'on_disk>,
 
     /// How many (non-inclusive) descendants of this node have an entry.
     pub(super) descendants_with_entry_count: u32,
 
     /// How many (non-inclusive) descendants of this node have an entry whose
     /// state is "tracked".
     pub(super) tracked_descendants_count: u32,
 }
 
 #[derive(Debug)]
 pub(super) enum NodeData {
     Entry(DirstateEntry),
     CachedDirectory { mtime: TruncatedTimestamp },
     None,
 }
 
 impl Default for NodeData {
     fn default() -> Self {
         NodeData::None
     }
 }
 
 impl NodeData {
     fn has_entry(&self) -> bool {
         match self {
             NodeData::Entry(_) => true,
             _ => false,
         }
     }
 
     fn as_entry(&self) -> Option<&DirstateEntry> {
         match self {
             NodeData::Entry(entry) => Some(entry),
             _ => None,
         }
     }
 
     fn as_entry_mut(&mut self) -> Option<&mut DirstateEntry> {
         match self {
             NodeData::Entry(entry) => Some(entry),
             _ => None,
         }
     }
 }
 
 impl<'on_disk> DirstateMap<'on_disk> {
     pub(super) fn empty(on_disk: &'on_disk [u8]) -> Self {
         Self {
             on_disk,
             root: ChildNodes::default(),
             nodes_with_entry_count: 0,
             nodes_with_copy_source_count: 0,
             ignore_patterns_hash: [0; on_disk::IGNORE_PATTERNS_HASH_LEN],
             unreachable_bytes: 0,
             old_data_size: 0,
             dirstate_version: DirstateVersion::V1,
         }
     }
 
     #[timed]
     pub fn new_v2(
         on_disk: &'on_disk [u8],
         data_size: usize,
         metadata: &[u8],
     ) -> Result<Self, DirstateError> {
         if let Some(data) = on_disk.get(..data_size) {
             Ok(on_disk::read(data, metadata)?)
         } else {
             Err(DirstateV2ParseError::new("not enough bytes on disk").into())
         }
     }
 
     #[timed]
     pub fn new_v1(
         on_disk: &'on_disk [u8],
     ) -> Result<(Self, Option<DirstateParents>), DirstateError> {
         let mut map = Self::empty(on_disk);
         if map.on_disk.is_empty() {
             return Ok((map, None));
         }
 
         let parents = parse_dirstate_entries(
             map.on_disk,
             |path, entry, copy_source| {
                 let tracked = entry.tracked();
                 let node = Self::get_or_insert_node_inner(
                     map.on_disk,
                     &mut map.unreachable_bytes,
                     &mut map.root,
                     path,
                     WithBasename::to_cow_borrowed,
                     |ancestor| {
                         if tracked {
                             ancestor.tracked_descendants_count += 1
                         }
                         ancestor.descendants_with_entry_count += 1
                     },
                 )?;
                 assert!(
                     !node.data.has_entry(),
                     "duplicate dirstate entry in read"
                 );
                 assert!(
                     node.copy_source.is_none(),
                     "duplicate dirstate entry in read"
                 );
                 node.data = NodeData::Entry(*entry);
                 node.copy_source = copy_source.map(Cow::Borrowed);
                 map.nodes_with_entry_count += 1;
                 if copy_source.is_some() {
                     map.nodes_with_copy_source_count += 1
                 }
                 Ok(())
             },
         )?;
         let parents = Some(parents.clone());
 
         Ok((map, parents))
     }
 
     /// Assuming dirstate-v2 format, returns whether the next write should
     /// append to the existing data file that contains `self.on_disk` (true),
     /// or create a new data file from scratch (false).
     pub(super) fn write_should_append(&self) -> bool {
         let ratio = self.unreachable_bytes as f32 / self.on_disk.len() as f32;
         ratio < ACCEPTABLE_UNREACHABLE_BYTES_RATIO
     }
 
     fn get_node<'tree>(
         &'tree self,
         path: &HgPath,
     ) -> Result<Option<NodeRef<'tree, 'on_disk>>, DirstateV2ParseError> {
         let mut children = self.root.as_ref();
         let mut components = path.components();
         let mut component =
             components.next().expect("expected at least one components");
         loop {
             if let Some(child) = children.get(component, self.on_disk)? {
                 if let Some(next_component) = components.next() {
                     component = next_component;
                     children = child.children(self.on_disk)?;
                 } else {
                     return Ok(Some(child));
                 }
             } else {
                 return Ok(None);
             }
         }
     }
 
     /// Returns a mutable reference to the node at `path` if it exists
     ///
     /// `each_ancestor` is a callback that is called for each ancestor node
     /// when descending the tree. It is used to keep the different counters
     /// of the `DirstateMap` up-to-date.
     fn get_node_mut<'tree>(
         &'tree mut self,
         path: &HgPath,
         each_ancestor: impl FnMut(&mut Node),
     ) -> Result<Option<&'tree mut Node<'on_disk>>, DirstateV2ParseError> {
         Self::get_node_mut_inner(
             self.on_disk,
             &mut self.unreachable_bytes,
             &mut self.root,
             path,
             each_ancestor,
         )
     }
 
     /// Lower-level version of `get_node_mut`.
     ///
     /// This takes `root` instead of `&mut self` so that callers can mutate
     /// other fields while the returned borrow is still valid.
     ///
     /// `each_ancestor` is a callback that is called for each ancestor node
     /// when descending the tree. It is used to keep the different counters
     /// of the `DirstateMap` up-to-date.
     fn get_node_mut_inner<'tree>(
         on_disk: &'on_disk [u8],
         unreachable_bytes: &mut u32,
         root: &'tree mut ChildNodes<'on_disk>,
         path: &HgPath,
         mut each_ancestor: impl FnMut(&mut Node),
     ) -> Result<Option<&'tree mut Node<'on_disk>>, DirstateV2ParseError> {
         let mut children = root;
         let mut components = path.components();
         let mut component =
             components.next().expect("expected at least one components");
         loop {
             if let Some(child) = children
                 .make_mut(on_disk, unreachable_bytes)?
                 .get_mut(component)
             {
                 if let Some(next_component) = components.next() {
                     each_ancestor(child);
                     component = next_component;
                     children = &mut child.children;
                 } else {
                     return Ok(Some(child));
                 }
             } else {
                 return Ok(None);
             }
         }
     }
 
     /// Get a mutable reference to the node at `path`, creating it if it does
     /// not exist.
     ///
     /// `each_ancestor` is a callback that is called for each ancestor node
     /// when descending the tree. It is used to keep the different counters
     /// of the `DirstateMap` up-to-date.
     fn get_or_insert_node<'tree, 'path>(
         &'tree mut self,
         path: &'path HgPath,
         each_ancestor: impl FnMut(&mut Node),
     ) -> Result<&'tree mut Node<'on_disk>, DirstateV2ParseError> {
         Self::get_or_insert_node_inner(
             self.on_disk,
             &mut self.unreachable_bytes,
             &mut self.root,
             path,
             WithBasename::to_cow_owned,
             each_ancestor,
         )
     }
 
     /// Lower-level version of `get_or_insert_node_inner`, which is used when
     /// parsing disk data to remove allocations for new nodes.
     fn get_or_insert_node_inner<'tree, 'path>(
         on_disk: &'on_disk [u8],
         unreachable_bytes: &mut u32,
         root: &'tree mut ChildNodes<'on_disk>,
         path: &'path HgPath,
         to_cow: impl Fn(
             WithBasename<&'path HgPath>,
         ) -> WithBasename<Cow<'on_disk, HgPath>>,
         mut each_ancestor: impl FnMut(&mut Node),
     ) -> Result<&'tree mut Node<'on_disk>, DirstateV2ParseError> {
         let mut child_nodes = root;
         let mut inclusive_ancestor_paths =
             WithBasename::inclusive_ancestors_of(path);
         let mut ancestor_path = inclusive_ancestor_paths
             .next()
             .expect("expected at least one inclusive ancestor");
         loop {
             let (_, child_node) = child_nodes
                 .make_mut(on_disk, unreachable_bytes)?
                 .raw_entry_mut()
                 .from_key(ancestor_path.base_name())
                 .or_insert_with(|| (to_cow(ancestor_path), Node::default()));
             if let Some(next) = inclusive_ancestor_paths.next() {
                 each_ancestor(child_node);
                 ancestor_path = next;
                 child_nodes = &mut child_node.children;
             } else {
                 return Ok(child_node);
             }
         }
     }
 
     fn reset_state(
         &mut self,
         filename: &HgPath,
         old_entry_opt: Option<DirstateEntry>,
         wc_tracked: bool,
         p1_tracked: bool,
         p2_info: bool,
         has_meaningful_mtime: bool,
         parent_file_data_opt: Option<ParentFileData>,
     ) -> Result<(), DirstateError> {
         let (had_entry, was_tracked) = match old_entry_opt {
             Some(old_entry) => (true, old_entry.tracked()),
             None => (false, false),
         };
         let node = self.get_or_insert_node(filename, |ancestor| {
             if !had_entry {
                 ancestor.descendants_with_entry_count += 1;
             }
             if was_tracked {
                 if !wc_tracked {
                     ancestor.tracked_descendants_count = ancestor
                         .tracked_descendants_count
                         .checked_sub(1)
                         .expect("tracked count to be >= 0");
                 }
             } else {
                 if wc_tracked {
                     ancestor.tracked_descendants_count += 1;
                 }
             }
         })?;
 
         let v2_data = if let Some(parent_file_data) = parent_file_data_opt {
             DirstateV2Data {
                 wc_tracked,
                 p1_tracked,
                 p2_info,
                 mode_size: parent_file_data.mode_size,
                 mtime: if has_meaningful_mtime {
                     parent_file_data.mtime
                 } else {
                     None
                 },
                 ..Default::default()
             }
         } else {
             DirstateV2Data {
                 wc_tracked,
                 p1_tracked,
                 p2_info,
                 ..Default::default()
             }
         };
         node.data = NodeData::Entry(DirstateEntry::from_v2_data(v2_data));
         if !had_entry {
             self.nodes_with_entry_count += 1;
         }
         Ok(())
     }
 
     fn set_tracked(
         &mut self,
         filename: &HgPath,
         old_entry_opt: Option<DirstateEntry>,
     ) -> Result<bool, DirstateV2ParseError> {
         let was_tracked = old_entry_opt.map_or(false, |e| e.tracked());
         let had_entry = old_entry_opt.is_some();
         let tracked_count_increment = if was_tracked { 0 } else { 1 };
         let mut new = false;
 
         let node = self.get_or_insert_node(filename, |ancestor| {
             if !had_entry {
                 ancestor.descendants_with_entry_count += 1;
             }
 
             ancestor.tracked_descendants_count += tracked_count_increment;
         })?;
         if let Some(old_entry) = old_entry_opt {
             let mut e = old_entry.clone();
             if e.tracked() {
                 // XXX
                 // This is probably overkill for more case, but we need this to
                 // fully replace the `normallookup` call with `set_tracked`
                 // one. Consider smoothing this in the future.
                 e.set_possibly_dirty();
             } else {
                 new = true;
                 e.set_tracked();
             }
             node.data = NodeData::Entry(e)
         } else {
             node.data = NodeData::Entry(DirstateEntry::new_tracked());
             self.nodes_with_entry_count += 1;
             new = true;
         };
         Ok(new)
     }
 
     /// Set a node as untracked in the dirstate.
     ///
     /// It is the responsibility of the caller to remove the copy source and/or
     /// the entry itself if appropriate.
     ///
     /// # Panics
     ///
     /// Panics if the node does not exist.
     fn set_untracked(
         &mut self,
         filename: &HgPath,
         old_entry: DirstateEntry,
     ) -> Result<(), DirstateV2ParseError> {
         let node = self
             .get_node_mut(filename, |ancestor| {
                 ancestor.tracked_descendants_count = ancestor
                     .tracked_descendants_count
                     .checked_sub(1)
                     .expect("tracked_descendants_count should be >= 0");
             })?
             .expect("node should exist");
         let mut new_entry = old_entry.clone();
         new_entry.set_untracked();
         node.data = NodeData::Entry(new_entry);
         Ok(())
     }
 
     /// Set a node as clean in the dirstate.
     ///
     /// It is the responsibility of the caller to remove the copy source.
     ///
     /// # Panics
     ///
     /// Panics if the node does not exist.
     fn set_clean(
         &mut self,
         filename: &HgPath,
         old_entry: DirstateEntry,
         mode: u32,
         size: u32,
         mtime: TruncatedTimestamp,
     ) -> Result<(), DirstateError> {
         let node = self
             .get_node_mut(filename, |ancestor| {
                 if !old_entry.tracked() {
                     ancestor.tracked_descendants_count += 1;
                 }
             })?
             .expect("node should exist");
         let mut new_entry = old_entry.clone();
         new_entry.set_clean(mode, size, mtime);
         node.data = NodeData::Entry(new_entry);
         Ok(())
     }
 
     /// Set a node as possibly dirty in the dirstate.
     ///
     /// # Panics
     ///
     /// Panics if the node does not exist.
     fn set_possibly_dirty(
         &mut self,
         filename: &HgPath,
     ) -> Result<(), DirstateError> {
         let node = self
             .get_node_mut(filename, |_ancestor| {})?
             .expect("node should exist");
         let entry = node.data.as_entry_mut().expect("entry should exist");
         entry.set_possibly_dirty();
         node.data = NodeData::Entry(*entry);
         Ok(())
     }
 
     /// Clears the cached mtime for the (potential) folder at `path`.
     pub(super) fn clear_cached_mtime(
         &mut self,
         path: &HgPath,
     ) -> Result<(), DirstateV2ParseError> {
         let node = match self.get_node_mut(path, |_ancestor| {})? {
             Some(node) => node,
             None => return Ok(()),
         };
         if let NodeData::CachedDirectory { .. } = &node.data {
             node.data = NodeData::None
         }
         Ok(())
     }
 
     /// Sets the cached mtime for the (potential) folder at `path`.
     pub(super) fn set_cached_mtime(
         &mut self,
         path: &HgPath,
         mtime: TruncatedTimestamp,
     ) -> Result<(), DirstateV2ParseError> {
         let node = match self.get_node_mut(path, |_ancestor| {})? {
             Some(node) => node,
             None => return Ok(()),
         };
         match &node.data {
             NodeData::Entry(_) => {} // Don’t overwrite an entry
             NodeData::CachedDirectory { .. } | NodeData::None => {
                 node.data = NodeData::CachedDirectory { mtime }
             }
         }
         Ok(())
     }
 
     fn iter_nodes<'tree>(
         &'tree self,
     ) -> impl Iterator<
         Item = Result<NodeRef<'tree, 'on_disk>, DirstateV2ParseError>,
     > + 'tree {
         // Depth first tree traversal.
         //
         // If we could afford internal iteration and recursion,
         // this would look like:
         //
         // ```
         // fn traverse_children(
         //     children: &ChildNodes,
         //     each: &mut impl FnMut(&Node),
         // ) {
         //     for child in children.values() {
         //         traverse_children(&child.children, each);
         //         each(child);
         //     }
         // }
         // ```
         //
         // However we want an external iterator and therefore can’t use the
         // call stack. Use an explicit stack instead:
         let mut stack = Vec::new();
         let mut iter = self.root.as_ref().iter();
         std::iter::from_fn(move || {
             while let Some(child_node) = iter.next() {
                 let children = match child_node.children(self.on_disk) {
                     Ok(children) => children,
                     Err(error) => return Some(Err(error)),
                 };
                 // Pseudo-recursion
                 let new_iter = children.iter();
                 let old_iter = std::mem::replace(&mut iter, new_iter);
                 stack.push((child_node, old_iter));
             }
             // Found the end of a `children.iter()` iterator.
             if let Some((child_node, next_iter)) = stack.pop() {
                 // "Return" from pseudo-recursion by restoring state from the
                 // explicit stack
                 iter = next_iter;
 
                 Some(Ok(child_node))
             } else {
                 // Reached the bottom of the stack, we’re done
                 None
             }
         })
     }
 
     fn count_dropped_path(unreachable_bytes: &mut u32, path: &Cow<HgPath>) {
         if let Cow::Borrowed(path) = path {
             *unreachable_bytes += path.len() as u32
         }
     }
 }
 
 /// Like `Iterator::filter_map`, but over a fallible iterator of `Result`s.
 ///
 /// The callback is only called for incoming `Ok` values. Errors are passed
 /// through as-is. In order to let it use the `?` operator the callback is
 /// expected to return a `Result` of `Option`, instead of an `Option` of
 /// `Result`.
 fn filter_map_results<'a, I, F, A, B, E>(
     iter: I,
     f: F,
 ) -> impl Iterator<Item = Result<B, E>> + 'a
 where
     I: Iterator<Item = Result<A, E>> + 'a,
     F: Fn(A) -> Result<Option<B>, E> + 'a,
 {
     iter.filter_map(move |result| match result {
         Ok(node) => f(node).transpose(),
         Err(e) => Some(Err(e)),
     })
 }
 
 impl OwningDirstateMap {
     pub fn clear(&mut self) {
         self.with_dmap_mut(|map| {
             map.root = Default::default();
             map.nodes_with_entry_count = 0;
             map.nodes_with_copy_source_count = 0;
         });
     }
 
     pub fn set_tracked(
         &mut self,
         filename: &HgPath,
     ) -> Result<bool, DirstateV2ParseError> {
         let old_entry_opt = self.get(filename)?;
         self.with_dmap_mut(|map| map.set_tracked(filename, old_entry_opt))
     }
 
     pub fn set_untracked(
         &mut self,
         filename: &HgPath,
     ) -> Result<bool, DirstateError> {
         let old_entry_opt = self.get(filename)?;
         match old_entry_opt {
             None => Ok(false),
             Some(old_entry) => {
                 if !old_entry.tracked() {
                     // `DirstateMap::set_untracked` is not a noop if
                     // already not tracked as it will decrement the
                     // tracked counters while going down.
                     return Ok(true);
                 }
                 if old_entry.added() {
                     // Untracking an "added" entry will just result in a
                     // worthless entry (and other parts of the code will
                     // complain about it), just drop it entirely.
                     self.drop_entry_and_copy_source(filename)?;
                     return Ok(true);
                 }
                 if !old_entry.p2_info() {
                     self.copy_map_remove(filename)?;
                 }
 
                 self.with_dmap_mut(|map| {
                     map.set_untracked(filename, old_entry)?;
                     Ok(true)
                 })
             }
         }
     }
 
     pub fn set_clean(
         &mut self,
         filename: &HgPath,
         mode: u32,
         size: u32,
         mtime: TruncatedTimestamp,
     ) -> Result<(), DirstateError> {
         let old_entry = match self.get(filename)? {
             None => {
                 return Err(
                     DirstateMapError::PathNotFound(filename.into()).into()
                 )
             }
             Some(e) => e,
         };
         self.copy_map_remove(filename)?;
         self.with_dmap_mut(|map| {
             map.set_clean(filename, old_entry, mode, size, mtime)
         })
     }
 
     pub fn set_possibly_dirty(
         &mut self,
         filename: &HgPath,
     ) -> Result<(), DirstateError> {
         if self.get(filename)?.is_none() {
             return Err(DirstateMapError::PathNotFound(filename.into()).into());
         }
         self.with_dmap_mut(|map| map.set_possibly_dirty(filename))
     }
 
     pub fn reset_state(
         &mut self,
         filename: &HgPath,
         wc_tracked: bool,
         p1_tracked: bool,
         p2_info: bool,
         has_meaningful_mtime: bool,
         parent_file_data_opt: Option<ParentFileData>,
     ) -> Result<(), DirstateError> {
         if !(p1_tracked || p2_info || wc_tracked) {
             self.drop_entry_and_copy_source(filename)?;
             return Ok(());
         }
         self.copy_map_remove(filename)?;
         let old_entry_opt = self.get(filename)?;
         self.with_dmap_mut(|map| {
             map.reset_state(
                 filename,
                 old_entry_opt,
                 wc_tracked,
                 p1_tracked,
                 p2_info,
                 has_meaningful_mtime,
                 parent_file_data_opt,
             )
         })
     }
 
     pub fn drop_entry_and_copy_source(
         &mut self,
         filename: &HgPath,
     ) -> Result<(), DirstateError> {
         let was_tracked = self.get(filename)?.map_or(false, |e| e.tracked());
         struct Dropped {
             was_tracked: bool,
             had_entry: bool,
             had_copy_source: bool,
         }
 
         /// If this returns `Ok(Some((dropped, removed)))`, then
         ///
         /// * `dropped` is about the leaf node that was at `filename`
         /// * `removed` is whether this particular level of recursion just
         ///   removed a node in `nodes`.
         fn recur<'on_disk>(
             on_disk: &'on_disk [u8],
             unreachable_bytes: &mut u32,
             nodes: &mut ChildNodes<'on_disk>,
             path: &HgPath,
         ) -> Result<Option<(Dropped, bool)>, DirstateV2ParseError> {
             let (first_path_component, rest_of_path) =
                 path.split_first_component();
             let nodes = nodes.make_mut(on_disk, unreachable_bytes)?;
             let node = if let Some(node) = nodes.get_mut(first_path_component)
             {
                 node
             } else {
                 return Ok(None);
             };
             let dropped;
             if let Some(rest) = rest_of_path {
                 if let Some((d, removed)) = recur(
                     on_disk,
                     unreachable_bytes,
                     &mut node.children,
                     rest,
                 )? {
                     dropped = d;
                     if dropped.had_entry {
                         node.descendants_with_entry_count = node
                             .descendants_with_entry_count
                             .checked_sub(1)
                             .expect(
                                 "descendants_with_entry_count should be >= 0",
                             );
                     }
                     if dropped.was_tracked {
                         node.tracked_descendants_count = node
                             .tracked_descendants_count
                             .checked_sub(1)
                             .expect(
                                 "tracked_descendants_count should be >= 0",
                             );
                     }
 
                     // Directory caches must be invalidated when removing a
                     // child node
                     if removed {
                         if let NodeData::CachedDirectory { .. } = &node.data {
                             node.data = NodeData::None
                         }
                     }
                 } else {
                     return Ok(None);
                 }
             } else {
                 let entry = node.data.as_entry();
                 let was_tracked = entry.map_or(false, |entry| entry.tracked());
                 let had_entry = entry.is_some();
                 if had_entry {
                     node.data = NodeData::None
                 }
                 let mut had_copy_source = false;
                 if let Some(source) = &node.copy_source {
                     DirstateMap::count_dropped_path(unreachable_bytes, source);
                     had_copy_source = true;
                     node.copy_source = None
                 }
                 dropped = Dropped {
                     was_tracked,
                     had_entry,
                     had_copy_source,
                 };
             }
             // After recursion, for both leaf (rest_of_path is None) nodes and
             // parent nodes, remove a node if it just became empty.
             let remove = !node.data.has_entry()
                 && node.copy_source.is_none()
                 && node.children.is_empty();
             if remove {
                 let (key, _) =
                     nodes.remove_entry(first_path_component).unwrap();
                 DirstateMap::count_dropped_path(
                     unreachable_bytes,
                     key.full_path(),
                 )
             }
             Ok(Some((dropped, remove)))
         }
 
         self.with_dmap_mut(|map| {
             if let Some((dropped, _removed)) = recur(
                 map.on_disk,
                 &mut map.unreachable_bytes,
                 &mut map.root,
                 filename,
             )? {
                 if dropped.had_entry {
                     map.nodes_with_entry_count = map
                         .nodes_with_entry_count
                         .checked_sub(1)
                         .expect("nodes_with_entry_count should be >= 0");
                 }
                 if dropped.had_copy_source {
                     map.nodes_with_copy_source_count = map
                         .nodes_with_copy_source_count
                         .checked_sub(1)
                         .expect("nodes_with_copy_source_count should be >= 0");
                 }
             } else {
                 debug_assert!(!was_tracked);
             }
             Ok(())
         })
     }
 
     pub fn has_tracked_dir(
         &mut self,
         directory: &HgPath,
     ) -> Result<bool, DirstateError> {
         self.with_dmap_mut(|map| {
             if let Some(node) = map.get_node(directory)? {
                 // A node without a `DirstateEntry` was created to hold child
                 // nodes, and is therefore a directory.
                 let is_dir = node.entry()?.is_none();
                 Ok(is_dir && node.tracked_descendants_count() > 0)
             } else {
                 Ok(false)
             }
         })
     }
 
     pub fn has_dir(
         &mut self,
         directory: &HgPath,
     ) -> Result<bool, DirstateError> {
         self.with_dmap_mut(|map| {
             if let Some(node) = map.get_node(directory)? {
                 // A node without a `DirstateEntry` was created to hold child
                 // nodes, and is therefore a directory.
                 let is_dir = node.entry()?.is_none();
                 Ok(is_dir && node.descendants_with_entry_count() > 0)
             } else {
                 Ok(false)
             }
         })
     }
 
     #[timed]
     pub fn pack_v1(
         &self,
         parents: DirstateParents,
     ) -> Result<Vec<u8>, DirstateError> {
         let map = self.get_map();
         // Optizimation (to be measured?): pre-compute size to avoid `Vec`
         // reallocations
         let mut size = parents.as_bytes().len();
         for node in map.iter_nodes() {
             let node = node?;
             if node.entry()?.is_some() {
                 size += packed_entry_size(
                     node.full_path(map.on_disk)?,
                     node.copy_source(map.on_disk)?,
                 );
             }
         }
 
         let mut packed = Vec::with_capacity(size);
         packed.extend(parents.as_bytes());
 
         for node in map.iter_nodes() {
             let node = node?;
             if let Some(entry) = node.entry()? {
                 pack_entry(
                     node.full_path(map.on_disk)?,
                     &entry,
                     node.copy_source(map.on_disk)?,
                     &mut packed,
                 );
             }
         }
         Ok(packed)
     }
 
     /// Returns new data and metadata together with whether that data should be
     /// appended to the existing data file whose content is at
     /// `map.on_disk` (true), instead of written to a new data file
     /// (false), and the previous size of data on disk.
     #[timed]
     pub fn pack_v2(
         &self,
-        can_append: bool,
+        write_mode: DirstateMapWriteMode,
     ) -> Result<(Vec<u8>, on_disk::TreeMetadata, bool, usize), DirstateError>
     {
         let map = self.get_map();
-        on_disk::write(map, can_append)
+        on_disk::write(map, write_mode)
     }
 
     /// `callback` allows the caller to process and do something with the
     /// results of the status. This is needed to do so efficiently (i.e.
     /// without cloning the `DirstateStatus` object with its paths) because
     /// we need to borrow from `Self`.
     pub fn with_status<R>(
         &mut self,
         matcher: &(dyn Matcher + Sync),
         root_dir: PathBuf,
         ignore_files: Vec<PathBuf>,
         options: StatusOptions,
         callback: impl for<'r> FnOnce(
             Result<(DirstateStatus<'r>, Vec<PatternFileWarning>), StatusError>,
         ) -> R,
     ) -> R {
         self.with_dmap_mut(|map| {
             callback(super::status::status(
                 map,
                 matcher,
                 root_dir,
                 ignore_files,
                 options,
             ))
         })
     }
 
     pub fn copy_map_len(&self) -> usize {
         let map = self.get_map();
         map.nodes_with_copy_source_count as usize
     }
 
     pub fn copy_map_iter(&self) -> CopyMapIter<'_> {
         let map = self.get_map();
         Box::new(filter_map_results(map.iter_nodes(), move |node| {
             Ok(if let Some(source) = node.copy_source(map.on_disk)? {
                 Some((node.full_path(map.on_disk)?, source))
             } else {
                 None
             })
         }))
     }
 
     pub fn copy_map_contains_key(
         &self,
         key: &HgPath,
     ) -> Result<bool, DirstateV2ParseError> {
         let map = self.get_map();
         Ok(if let Some(node) = map.get_node(key)? {
             node.has_copy_source()
         } else {
             false
         })
     }
 
     pub fn copy_map_get(
         &self,
         key: &HgPath,
     ) -> Result<Option<&HgPath>, DirstateV2ParseError> {
         let map = self.get_map();
         if let Some(node) = map.get_node(key)? {
             if let Some(source) = node.copy_source(map.on_disk)? {
                 return Ok(Some(source));
             }
         }
         Ok(None)
     }
 
     pub fn copy_map_remove(
         &mut self,
         key: &HgPath,
     ) -> Result<Option<HgPathBuf>, DirstateV2ParseError> {
         self.with_dmap_mut(|map| {
             let count = &mut map.nodes_with_copy_source_count;
             let unreachable_bytes = &mut map.unreachable_bytes;
             Ok(DirstateMap::get_node_mut_inner(
                 map.on_disk,
                 unreachable_bytes,
                 &mut map.root,
                 key,
                 |_ancestor| {},
             )?
             .and_then(|node| {
                 if let Some(source) = &node.copy_source {
                     *count = count
                         .checked_sub(1)
                         .expect("nodes_with_copy_source_count should be >= 0");
                     DirstateMap::count_dropped_path(unreachable_bytes, source);
                 }
                 node.copy_source.take().map(Cow::into_owned)
             }))
         })
     }
 
     pub fn copy_map_insert(
         &mut self,
         key: &HgPath,
         value: &HgPath,
     ) -> Result<Option<HgPathBuf>, DirstateV2ParseError> {
         self.with_dmap_mut(|map| {
             let node = map.get_or_insert_node(&key, |_ancestor| {})?;
             let had_copy_source = node.copy_source.is_none();
             let old = node
                 .copy_source
                 .replace(value.to_owned().into())
                 .map(Cow::into_owned);
             if had_copy_source {
                 map.nodes_with_copy_source_count += 1
             }
             Ok(old)
         })
     }
 
     pub fn len(&self) -> usize {
         let map = self.get_map();
         map.nodes_with_entry_count as usize
     }
 
     pub fn contains_key(
         &self,
         key: &HgPath,
     ) -> Result<bool, DirstateV2ParseError> {
         Ok(self.get(key)?.is_some())
     }
 
     pub fn get(
         &self,
         key: &HgPath,
     ) -> Result<Option<DirstateEntry>, DirstateV2ParseError> {
         let map = self.get_map();
         Ok(if let Some(node) = map.get_node(key)? {
             node.entry()?
         } else {
             None
         })
     }
 
     pub fn iter(&self) -> StateMapIter<'_> {
         let map = self.get_map();
         Box::new(filter_map_results(map.iter_nodes(), move |node| {
             Ok(if let Some(entry) = node.entry()? {
                 Some((node.full_path(map.on_disk)?, entry))
             } else {
                 None
             })
         }))
     }
 
     pub fn iter_tracked_dirs(
         &mut self,
     ) -> Result<
         Box<
             dyn Iterator<Item = Result<&HgPath, DirstateV2ParseError>>
                 + Send
                 + '_,
         >,
         DirstateError,
     > {
         let map = self.get_map();
         let on_disk = map.on_disk;
         Ok(Box::new(filter_map_results(
             map.iter_nodes(),
             move |node| {
                 Ok(if node.tracked_descendants_count() > 0 {
                     Some(node.full_path(on_disk)?)
                 } else {
                     None
                 })
             },
         )))
     }
 
     /// Only public because it needs to be exposed to the Python layer.
     /// It is not the full `setparents` logic, only the parts that mutate the
     /// entries.
     pub fn setparents_fixup(
         &mut self,
     ) -> Result<Vec<(HgPathBuf, HgPathBuf)>, DirstateV2ParseError> {
         // XXX
         // All the copying and re-querying is quite inefficient, but this is
         // still a lot better than doing it from Python.
         //
         // The better solution is to develop a mechanism for `iter_mut`,
         // which will be a lot more involved: we're dealing with a lazy,
         // append-mostly, tree-like data structure. This will do for now.
         let mut copies = vec![];
         let mut files_with_p2_info = vec![];
         for res in self.iter() {
             let (path, entry) = res?;
             if entry.p2_info() {
                 files_with_p2_info.push(path.to_owned())
             }
         }
         self.with_dmap_mut(|map| {
             for path in files_with_p2_info.iter() {
                 let node = map.get_or_insert_node(path, |_| {})?;
                 let entry =
                     node.data.as_entry_mut().expect("entry should exist");
                 entry.drop_merge_data();
                 if let Some(source) = node.copy_source.take().as_deref() {
                     copies.push((path.to_owned(), source.to_owned()));
                 }
             }
             Ok(copies)
         })
     }
 
     pub fn debug_iter(
         &self,
         all: bool,
     ) -> Box<
         dyn Iterator<
                 Item = Result<
                     (&HgPath, (u8, i32, i32, i32)),
                     DirstateV2ParseError,
                 >,
             > + Send
             + '_,
     > {
         let map = self.get_map();
         Box::new(filter_map_results(map.iter_nodes(), move |node| {
             let debug_tuple = if let Some(entry) = node.entry()? {
                 entry.debug_tuple()
             } else if !all {
                 return Ok(None);
             } else if let Some(mtime) = node.cached_directory_mtime()? {
                 (b' ', 0, -1, mtime.truncated_seconds() as i32)
             } else {
                 (b' ', 0, -1, -1)
             };
             Ok(Some((node.full_path(map.on_disk)?, debug_tuple)))
         }))
     }
 }
 #[cfg(test)]
 mod tests {
     use super::*;
 
     /// Shortcut to return tracked descendants of a path.
     /// Panics if the path does not exist.
     fn tracked_descendants(map: &OwningDirstateMap, path: &[u8]) -> u32 {
         let path = dbg!(HgPath::new(path));
         let node = map.get_map().get_node(path);
         node.unwrap().unwrap().tracked_descendants_count()
     }
 
     /// Shortcut to return descendants with an entry.
     /// Panics if the path does not exist.
     fn descendants_with_an_entry(map: &OwningDirstateMap, path: &[u8]) -> u32 {
         let path = dbg!(HgPath::new(path));
         let node = map.get_map().get_node(path);
         node.unwrap().unwrap().descendants_with_entry_count()
     }
 
     fn assert_does_not_exist(map: &OwningDirstateMap, path: &[u8]) {
         let path = dbg!(HgPath::new(path));
         let node = map.get_map().get_node(path);
         assert!(node.unwrap().is_none());
     }
 
     /// Shortcut for path creation in tests
     fn p(b: &[u8]) -> &HgPath {
         HgPath::new(b)
     }
 
     /// Test the very simple case a single tracked file
     #[test]
     fn test_tracked_descendants_simple() -> Result<(), DirstateError> {
         let mut map = OwningDirstateMap::new_empty(vec![]);
         assert_eq!(map.len(), 0);
 
         map.set_tracked(p(b"some/nested/path"))?;
 
         assert_eq!(map.len(), 1);
         assert_eq!(tracked_descendants(&map, b"some"), 1);
         assert_eq!(tracked_descendants(&map, b"some/nested"), 1);
         assert_eq!(tracked_descendants(&map, b"some/nested/path"), 0);
 
         map.set_untracked(p(b"some/nested/path"))?;
         assert_eq!(map.len(), 0);
         assert!(map.get_map().get_node(p(b"some"))?.is_none());
 
         Ok(())
     }
 
     /// Test the simple case of all tracked, but multiple files
     #[test]
     fn test_tracked_descendants_multiple() -> Result<(), DirstateError> {
         let mut map = OwningDirstateMap::new_empty(vec![]);
 
         map.set_tracked(p(b"some/nested/path"))?;
         map.set_tracked(p(b"some/nested/file"))?;
         // one layer without any files to test deletion cascade
         map.set_tracked(p(b"some/other/nested/path"))?;
         map.set_tracked(p(b"root_file"))?;
         map.set_tracked(p(b"some/file"))?;
         map.set_tracked(p(b"some/file2"))?;
         map.set_tracked(p(b"some/file3"))?;
 
         assert_eq!(map.len(), 7);
         assert_eq!(tracked_descendants(&map, b"some"), 6);
         assert_eq!(tracked_descendants(&map, b"some/nested"), 2);
         assert_eq!(tracked_descendants(&map, b"some/other"), 1);
         assert_eq!(tracked_descendants(&map, b"some/other/nested"), 1);
         assert_eq!(tracked_descendants(&map, b"some/nested/path"), 0);
 
         map.set_untracked(p(b"some/nested/path"))?;
         assert_eq!(map.len(), 6);
         assert_eq!(tracked_descendants(&map, b"some"), 5);
         assert_eq!(tracked_descendants(&map, b"some/nested"), 1);
         assert_eq!(tracked_descendants(&map, b"some/other"), 1);
         assert_eq!(tracked_descendants(&map, b"some/other/nested"), 1);
 
         map.set_untracked(p(b"some/nested/file"))?;
         assert_eq!(map.len(), 5);
         assert_eq!(tracked_descendants(&map, b"some"), 4);
         assert_eq!(tracked_descendants(&map, b"some/other"), 1);
         assert_eq!(tracked_descendants(&map, b"some/other/nested"), 1);
         assert_does_not_exist(&map, b"some_nested");
 
         map.set_untracked(p(b"some/other/nested/path"))?;
         assert_eq!(map.len(), 4);
         assert_eq!(tracked_descendants(&map, b"some"), 3);
         assert_does_not_exist(&map, b"some/other");
 
         map.set_untracked(p(b"root_file"))?;
         assert_eq!(map.len(), 3);
         assert_eq!(tracked_descendants(&map, b"some"), 3);
         assert_does_not_exist(&map, b"root_file");
 
         map.set_untracked(p(b"some/file"))?;
         assert_eq!(map.len(), 2);
         assert_eq!(tracked_descendants(&map, b"some"), 2);
         assert_does_not_exist(&map, b"some/file");
 
         map.set_untracked(p(b"some/file2"))?;
         assert_eq!(map.len(), 1);
         assert_eq!(tracked_descendants(&map, b"some"), 1);
         assert_does_not_exist(&map, b"some/file2");
 
         map.set_untracked(p(b"some/file3"))?;
         assert_eq!(map.len(), 0);
         assert_does_not_exist(&map, b"some/file3");
 
         Ok(())
     }
 
     /// Check with a mix of tracked and non-tracked items
     #[test]
     fn test_tracked_descendants_different() -> Result<(), DirstateError> {
         let mut map = OwningDirstateMap::new_empty(vec![]);
 
         // A file that was just added
         map.set_tracked(p(b"some/nested/path"))?;
         // This has no information, the dirstate should ignore it
         map.reset_state(p(b"some/file"), false, false, false, false, None)?;
         assert_does_not_exist(&map, b"some/file");
 
         // A file that was removed
         map.reset_state(
             p(b"some/nested/file"),
             false,
             true,
             false,
             false,
             None,
         )?;
         assert!(!map.get(p(b"some/nested/file"))?.unwrap().tracked());
         // Only present in p2
         map.reset_state(p(b"some/file3"), false, false, true, false, None)?;
         assert!(!map.get(p(b"some/file3"))?.unwrap().tracked());
         // A file that was merged
         map.reset_state(p(b"root_file"), true, true, true, false, None)?;
         assert!(map.get(p(b"root_file"))?.unwrap().tracked());
         // A file that is added, with info from p2
         // XXX is that actually possible?
         map.reset_state(p(b"some/file2"), true, false, true, false, None)?;
         assert!(map.get(p(b"some/file2"))?.unwrap().tracked());
         // A clean file
         // One layer without any files to test deletion cascade
         map.reset_state(
             p(b"some/other/nested/path"),
             true,
             true,
             false,
             false,
             None,
         )?;
         assert!(map.get(p(b"some/other/nested/path"))?.unwrap().tracked());
 
         assert_eq!(map.len(), 6);
         assert_eq!(tracked_descendants(&map, b"some"), 3);
         assert_eq!(descendants_with_an_entry(&map, b"some"), 5);
         assert_eq!(tracked_descendants(&map, b"some/other/nested"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/other/nested"), 1);
         assert_eq!(tracked_descendants(&map, b"some/other/nested/path"), 0);
         assert_eq!(
             descendants_with_an_entry(&map, b"some/other/nested/path"),
             0
         );
         assert_eq!(tracked_descendants(&map, b"some/nested"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/nested"), 2);
 
         // might as well check this
         map.set_untracked(p(b"path/does/not/exist"))?;
         assert_eq!(map.len(), 6);
 
         map.set_untracked(p(b"some/other/nested/path"))?;
         // It is set untracked but not deleted since it held other information
         assert_eq!(map.len(), 6);
         assert_eq!(tracked_descendants(&map, b"some"), 2);
         assert_eq!(descendants_with_an_entry(&map, b"some"), 5);
         assert_eq!(descendants_with_an_entry(&map, b"some/other"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/other/nested"), 1);
         assert_eq!(tracked_descendants(&map, b"some/nested"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/nested"), 2);
 
         map.set_untracked(p(b"some/nested/path"))?;
         // It is set untracked *and* deleted since it was only added
         assert_eq!(map.len(), 5);
         assert_eq!(tracked_descendants(&map, b"some"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some"), 4);
         assert_eq!(tracked_descendants(&map, b"some/nested"), 0);
         assert_eq!(descendants_with_an_entry(&map, b"some/nested"), 1);
         assert_does_not_exist(&map, b"some/nested/path");
 
         map.set_untracked(p(b"root_file"))?;
         // Untracked but not deleted
         assert_eq!(map.len(), 5);
         assert!(map.get(p(b"root_file"))?.is_some());
 
         map.set_untracked(p(b"some/file2"))?;
         assert_eq!(map.len(), 5);
         assert_eq!(tracked_descendants(&map, b"some"), 0);
         assert!(map.get(p(b"some/file2"))?.is_some());
 
         map.set_untracked(p(b"some/file3"))?;
         assert_eq!(map.len(), 5);
         assert_eq!(tracked_descendants(&map, b"some"), 0);
         assert!(map.get(p(b"some/file3"))?.is_some());
 
         Ok(())
     }
 
     /// Check that copies counter is correctly updated
     #[test]
     fn test_copy_source() -> Result<(), DirstateError> {
         let mut map = OwningDirstateMap::new_empty(vec![]);
 
         // Clean file
         map.reset_state(p(b"files/clean"), true, true, false, false, None)?;
         // Merged file
         map.reset_state(p(b"files/from_p2"), true, true, true, false, None)?;
         // Removed file
         map.reset_state(p(b"removed"), false, true, false, false, None)?;
         // Added file
         map.reset_state(p(b"files/added"), true, false, false, false, None)?;
         // Add copy
         map.copy_map_insert(p(b"files/clean"), p(b"clean_copy_source"))?;
         assert_eq!(map.copy_map_len(), 1);
 
         // Copy override
         map.copy_map_insert(p(b"files/clean"), p(b"other_clean_copy_source"))?;
         assert_eq!(map.copy_map_len(), 1);
 
         // Multiple copies
         map.copy_map_insert(p(b"removed"), p(b"removed_copy_source"))?;
         assert_eq!(map.copy_map_len(), 2);
 
         map.copy_map_insert(p(b"files/added"), p(b"added_copy_source"))?;
         assert_eq!(map.copy_map_len(), 3);
 
         // Added, so the entry is completely removed
         map.set_untracked(p(b"files/added"))?;
         assert_does_not_exist(&map, b"files/added");
         assert_eq!(map.copy_map_len(), 2);
 
         // Removed, so the entry is kept around, so is its copy
         map.set_untracked(p(b"removed"))?;
         assert!(map.get(p(b"removed"))?.is_some());
         assert_eq!(map.copy_map_len(), 2);
 
         // Clean, so the entry is kept around, but not its copy
         map.set_untracked(p(b"files/clean"))?;
         assert!(map.get(p(b"files/clean"))?.is_some());
         assert_eq!(map.copy_map_len(), 1);
 
         map.copy_map_insert(p(b"files/from_p2"), p(b"from_p2_copy_source"))?;
         assert_eq!(map.copy_map_len(), 2);
 
         // Info from p2, so its copy source info is kept around
         map.set_untracked(p(b"files/from_p2"))?;
         assert!(map.get(p(b"files/from_p2"))?.is_some());
         assert_eq!(map.copy_map_len(), 2);
 
         Ok(())
     }
 
     /// Test with "on disk" data. For the sake of this test, the "on disk" data
     /// does not actually come from the disk, but it's opaque to the code being
     /// tested.
     #[test]
     fn test_on_disk() -> Result<(), DirstateError> {
         // First let's create some data to put "on disk"
         let mut map = OwningDirstateMap::new_empty(vec![]);
 
         // A file that was just added
         map.set_tracked(p(b"some/nested/added"))?;
         map.copy_map_insert(p(b"some/nested/added"), p(b"added_copy_source"))?;
 
         // A file that was removed
         map.reset_state(
             p(b"some/nested/removed"),
             false,
             true,
             false,
             false,
             None,
         )?;
         // Only present in p2
         map.reset_state(
             p(b"other/p2_info_only"),
             false,
             false,
             true,
             false,
             None,
         )?;
         map.copy_map_insert(
             p(b"other/p2_info_only"),
             p(b"other/p2_info_copy_source"),
         )?;
         // A file that was merged
         map.reset_state(p(b"merged"), true, true, true, false, None)?;
         // A file that is added, with info from p2
         // XXX is that actually possible?
         map.reset_state(
             p(b"other/added_with_p2"),
             true,
             false,
             true,
             false,
             None,
         )?;
         // One layer without any files to test deletion cascade
         // A clean file
         map.reset_state(
             p(b"some/other/nested/clean"),
             true,
             true,
             false,
             false,
             None,
         )?;
 
         let (packed, metadata, _should_append, _old_data_size) =
-            map.pack_v2(false)?;
+            map.pack_v2(DirstateMapWriteMode::ForceNewDataFile)?;
         let packed_len = packed.len();
         assert!(packed_len > 0);
 
         // Recreate "from disk"
         let mut map = OwningDirstateMap::new_v2(
             packed,
             packed_len,
             metadata.as_bytes(),
         )?;
 
         // Check that everything is accounted for
         assert!(map.contains_key(p(b"some/nested/added"))?);
         assert!(map.contains_key(p(b"some/nested/removed"))?);
         assert!(map.contains_key(p(b"merged"))?);
         assert!(map.contains_key(p(b"other/p2_info_only"))?);
         assert!(map.contains_key(p(b"other/added_with_p2"))?);
         assert!(map.contains_key(p(b"some/other/nested/clean"))?);
         assert_eq!(
             map.copy_map_get(p(b"some/nested/added"))?,
             Some(p(b"added_copy_source"))
         );
         assert_eq!(
             map.copy_map_get(p(b"other/p2_info_only"))?,
             Some(p(b"other/p2_info_copy_source"))
         );
         assert_eq!(tracked_descendants(&map, b"some"), 2);
         assert_eq!(descendants_with_an_entry(&map, b"some"), 3);
         assert_eq!(tracked_descendants(&map, b"other"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"other"), 2);
         assert_eq!(tracked_descendants(&map, b"some/other"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/other"), 1);
         assert_eq!(tracked_descendants(&map, b"some/other/nested"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/other/nested"), 1);
         assert_eq!(tracked_descendants(&map, b"some/nested"), 1);
         assert_eq!(descendants_with_an_entry(&map, b"some/nested"), 2);
         assert_eq!(map.len(), 6);
         assert_eq!(map.get_map().unreachable_bytes, 0);
         assert_eq!(map.copy_map_len(), 2);
 
         // Shouldn't change anything since it's already not tracked
         map.set_untracked(p(b"some/nested/removed"))?;
         assert_eq!(map.get_map().unreachable_bytes, 0);
 
         match map.get_map().root {
             ChildNodes::InMemory(_) => {
                 panic!("root should not have been mutated")
             }
             _ => (),
         }
         // We haven't mutated enough (nothing, actually), we should still be in
         // the append strategy
         assert!(map.get_map().write_should_append());
 
         // But this mutates the structure, so there should be unreachable_bytes
         assert!(map.set_untracked(p(b"some/nested/added"))?);
         let unreachable_bytes = map.get_map().unreachable_bytes;
         assert!(unreachable_bytes > 0);
 
         match map.get_map().root {
             ChildNodes::OnDisk(_) => panic!("root should have been mutated"),
             _ => (),
         }
 
         // This should not mutate the structure either, since `root` has
         // already been mutated along with its direct children.
         map.set_untracked(p(b"merged"))?;
         assert_eq!(map.get_map().unreachable_bytes, unreachable_bytes);
 
         match map.get_map().get_node(p(b"other/added_with_p2"))?.unwrap() {
             NodeRef::InMemory(_, _) => {
                 panic!("'other/added_with_p2' should not have been mutated")
             }
             _ => (),
         }
         // But this should, since it's in a different path
         // than `<root>some/nested/add`
         map.set_untracked(p(b"other/added_with_p2"))?;
         assert!(map.get_map().unreachable_bytes > unreachable_bytes);
 
         match map.get_map().get_node(p(b"other/added_with_p2"))?.unwrap() {
             NodeRef::OnDisk(_) => {
                 panic!("'other/added_with_p2' should have been mutated")
             }
             _ => (),
         }
 
         // We have rewritten most of the tree, we should create a new file
         assert!(!map.get_map().write_should_append());
 
         Ok(())
     }
 }

rust/hg-core/src/dirstate_tree/on_disk.rs

@@ -1,883 +1,888 b''
 //! The "version 2" disk representation of the dirstate
 //!
 //! See `mercurial/helptext/internals/dirstate-v2.txt`
 
 use crate::dirstate::{DirstateV2Data, TruncatedTimestamp};
 use crate::dirstate_tree::dirstate_map::DirstateVersion;
-use crate::dirstate_tree::dirstate_map::{self, DirstateMap, NodeRef};
+use crate::dirstate_tree::dirstate_map::{
+    self, DirstateMap, DirstateMapWriteMode, NodeRef,
+};
 use crate::dirstate_tree::path_with_basename::WithBasename;
 use crate::errors::HgError;
 use crate::utils::hg_path::HgPath;
 use crate::DirstateEntry;
 use crate::DirstateError;
 use crate::DirstateParents;
 use bitflags::bitflags;
 use bytes_cast::unaligned::{U16Be, U32Be};
 use bytes_cast::BytesCast;
 use format_bytes::format_bytes;
 use rand::Rng;
 use std::borrow::Cow;
 use std::convert::{TryFrom, TryInto};
 use std::fmt::Write;
 
 /// Added at the start of `.hg/dirstate` when the "v2" format is used.
 /// This a redundant sanity check more than an actual "magic number" since
 /// `.hg/requires` already governs which format should be used.
 pub const V2_FORMAT_MARKER: &[u8; 12] = b"dirstate-v2\n";
 
 /// Keep space for 256-bit hashes
 const STORED_NODE_ID_BYTES: usize = 32;
 
 /// … even though only 160 bits are used for now, with SHA-1
 const USED_NODE_ID_BYTES: usize = 20;
 
 pub(super) const IGNORE_PATTERNS_HASH_LEN: usize = 20;
 pub(super) type IgnorePatternsHash = [u8; IGNORE_PATTERNS_HASH_LEN];
 
 /// Must match constants of the same names in `mercurial/dirstateutils/v2.py`
 const TREE_METADATA_SIZE: usize = 44;
 const NODE_SIZE: usize = 44;
 
 /// Make sure that size-affecting changes are made knowingly
 #[allow(unused)]
 fn static_assert_size_of() {
     let _ = std::mem::transmute::<TreeMetadata, [u8; TREE_METADATA_SIZE]>;
     let _ = std::mem::transmute::<DocketHeader, [u8; TREE_METADATA_SIZE + 81]>;
     let _ = std::mem::transmute::<Node, [u8; NODE_SIZE]>;
 }
 
 // Must match `HEADER` in `mercurial/dirstateutils/docket.py`
 #[derive(BytesCast)]
 #[repr(C)]
 struct DocketHeader {
     marker: [u8; V2_FORMAT_MARKER.len()],
     parent_1: [u8; STORED_NODE_ID_BYTES],
     parent_2: [u8; STORED_NODE_ID_BYTES],
 
     metadata: TreeMetadata,
 
     /// Counted in bytes
     data_size: Size,
 
     uuid_size: u8,
 }
 
 pub struct Docket<'on_disk> {
     header: &'on_disk DocketHeader,
     pub uuid: &'on_disk [u8],
 }
 
 /// Fields are documented in the *Tree metadata in the docket file*
 /// section of `mercurial/helptext/internals/dirstate-v2.txt`
 #[derive(BytesCast)]
 #[repr(C)]
 pub struct TreeMetadata {
     root_nodes: ChildNodes,
     nodes_with_entry_count: Size,
     nodes_with_copy_source_count: Size,
     unreachable_bytes: Size,
     unused: [u8; 4],
 
     /// See *Optional hash of ignore patterns* section of
     /// `mercurial/helptext/internals/dirstate-v2.txt`
     ignore_patterns_hash: IgnorePatternsHash,
 }
 
 /// Fields are documented in the *The data file format*
 /// section of `mercurial/helptext/internals/dirstate-v2.txt`
 #[derive(BytesCast, Debug)]
 #[repr(C)]
 pub(super) struct Node {
     full_path: PathSlice,
 
     /// In bytes from `self.full_path.start`
     base_name_start: PathSize,
 
     copy_source: OptPathSlice,
     children: ChildNodes,
     pub(super) descendants_with_entry_count: Size,
     pub(super) tracked_descendants_count: Size,
     flags: U16Be,
     size: U32Be,
     mtime: PackedTruncatedTimestamp,
 }
 
 bitflags! {
     #[repr(C)]
     struct Flags: u16 {
         const WDIR_TRACKED = 1 << 0;
         const P1_TRACKED = 1 << 1;
         const P2_INFO = 1 << 2;
         const MODE_EXEC_PERM = 1 << 3;
         const MODE_IS_SYMLINK = 1 << 4;
         const HAS_FALLBACK_EXEC = 1 <<  5;
         const FALLBACK_EXEC = 1 <<  6;
         const HAS_FALLBACK_SYMLINK = 1 <<  7;
         const FALLBACK_SYMLINK = 1 <<  8;
         const EXPECTED_STATE_IS_MODIFIED = 1 << 9;
         const HAS_MODE_AND_SIZE = 1 <<10;
         const HAS_MTIME = 1 <<11;
         const MTIME_SECOND_AMBIGUOUS = 1 << 12;
         const DIRECTORY = 1 <<13;
         const ALL_UNKNOWN_RECORDED = 1 <<14;
         const ALL_IGNORED_RECORDED = 1 <<15;
     }
 }
 
 /// Duration since the Unix epoch
 #[derive(BytesCast, Copy, Clone, Debug)]
 #[repr(C)]
 struct PackedTruncatedTimestamp {
     truncated_seconds: U32Be,
     nanoseconds: U32Be,
 }
 
 /// Counted in bytes from the start of the file
 ///
 /// NOTE: not supporting `.hg/dirstate` files larger than 4 GiB.
 type Offset = U32Be;
 
 /// Counted in number of items
 ///
 /// NOTE: we choose not to support counting more than 4 billion nodes anywhere.
 type Size = U32Be;
 
 /// Counted in bytes
 ///
 /// NOTE: we choose not to support file names/paths longer than 64 KiB.
 type PathSize = U16Be;
 
 /// A contiguous sequence of `len` times `Node`, representing the child nodes
 /// of either some other node or of the repository root.
 ///
 /// Always sorted by ascending `full_path`, to allow binary search.
 /// Since nodes with the same parent nodes also have the same parent path,
 /// only the `base_name`s need to be compared during binary search.
 #[derive(BytesCast, Copy, Clone, Debug)]
 #[repr(C)]
 struct ChildNodes {
     start: Offset,
     len: Size,
 }
 
 /// A `HgPath` of `len` bytes
 #[derive(BytesCast, Copy, Clone, Debug)]
 #[repr(C)]
 struct PathSlice {
     start: Offset,
     len: PathSize,
 }
 
 /// Either nothing if `start == 0`, or a `HgPath` of `len` bytes
 type OptPathSlice = PathSlice;
 
 /// Unexpected file format found in `.hg/dirstate` with the "v2" format.
 ///
 /// This should only happen if Mercurial is buggy or a repository is corrupted.
 #[derive(Debug)]
 pub struct DirstateV2ParseError {
     message: String,
 }
 
 impl DirstateV2ParseError {
     pub fn new<S: Into<String>>(message: S) -> Self {
         Self {
             message: message.into(),
         }
     }
 }
 
 impl From<DirstateV2ParseError> for HgError {
     fn from(e: DirstateV2ParseError) -> Self {
         HgError::corrupted(format!("dirstate-v2 parse error: {}", e.message))
     }
 }
 
 impl From<DirstateV2ParseError> for crate::DirstateError {
     fn from(error: DirstateV2ParseError) -> Self {
         HgError::from(error).into()
     }
 }
 
 impl TreeMetadata {
     pub fn as_bytes(&self) -> &[u8] {
         BytesCast::as_bytes(self)
     }
 }
 
 impl<'on_disk> Docket<'on_disk> {
     /// Generate the identifier for a new data file
     ///
     /// TODO: support the `HGTEST_UUIDFILE` environment variable.
     /// See `mercurial/revlogutils/docket.py`
     pub fn new_uid() -> String {
         const ID_LENGTH: usize = 8;
         let mut id = String::with_capacity(ID_LENGTH);
         let mut rng = rand::thread_rng();
         for _ in 0..ID_LENGTH {
             // One random hexadecimal digit.
             // `unwrap` never panics because `impl Write for String`
             // never returns an error.
             write!(&mut id, "{:x}", rng.gen_range(0..16)).unwrap();
         }
         id
     }
 
     pub fn serialize(
         parents: DirstateParents,
         tree_metadata: TreeMetadata,
         data_size: u64,
         uuid: &[u8],
     ) -> Result<Vec<u8>, std::num::TryFromIntError> {
         let header = DocketHeader {
             marker: *V2_FORMAT_MARKER,
             parent_1: parents.p1.pad_to_256_bits(),
             parent_2: parents.p2.pad_to_256_bits(),
             metadata: tree_metadata,
             data_size: u32::try_from(data_size)?.into(),
             uuid_size: uuid.len().try_into()?,
         };
         let header = header.as_bytes();
         let mut docket = Vec::with_capacity(header.len() + uuid.len());
         docket.extend_from_slice(header);
         docket.extend_from_slice(uuid);
         Ok(docket)
     }
 
     pub fn parents(&self) -> DirstateParents {
         use crate::Node;
         let p1 = Node::try_from(&self.header.parent_1[..USED_NODE_ID_BYTES])
             .unwrap()
             .clone();
         let p2 = Node::try_from(&self.header.parent_2[..USED_NODE_ID_BYTES])
             .unwrap()
             .clone();
         DirstateParents { p1, p2 }
     }
 
     pub fn tree_metadata(&self) -> &[u8] {
         self.header.metadata.as_bytes()
     }
 
     pub fn data_size(&self) -> usize {
         // This `unwrap` could only panic on a 16-bit CPU
         self.header.data_size.get().try_into().unwrap()
     }
 
     pub fn data_filename(&self) -> String {
         String::from_utf8(format_bytes!(b"dirstate.{}", self.uuid)).unwrap()
     }
 }
 
 pub fn read_docket(
     on_disk: &[u8],
 ) -> Result<Docket<'_>, DirstateV2ParseError> {
     let (header, uuid) = DocketHeader::from_bytes(on_disk).map_err(|e| {
         DirstateV2ParseError::new(format!("when reading docket, {}", e))
     })?;
     let uuid_size = header.uuid_size as usize;
     if header.marker == *V2_FORMAT_MARKER && uuid.len() == uuid_size {
         Ok(Docket { header, uuid })
     } else {
         Err(DirstateV2ParseError::new(
             "invalid format marker or uuid size",
         ))
     }
 }
 
 pub(super) fn read<'on_disk>(
     on_disk: &'on_disk [u8],
     metadata: &[u8],
 ) -> Result<DirstateMap<'on_disk>, DirstateV2ParseError> {
     if on_disk.is_empty() {
         let mut map = DirstateMap::empty(on_disk);
         map.dirstate_version = DirstateVersion::V2;
         return Ok(map);
     }
     let (meta, _) = TreeMetadata::from_bytes(metadata).map_err(|e| {
         DirstateV2ParseError::new(format!("when parsing tree metadata, {}", e))
     })?;
     let dirstate_map = DirstateMap {
         on_disk,
         root: dirstate_map::ChildNodes::OnDisk(
             read_nodes(on_disk, meta.root_nodes).map_err(|mut e| {
                 e.message = format!("{}, when reading root notes", e.message);
                 e
             })?,
         ),
         nodes_with_entry_count: meta.nodes_with_entry_count.get(),
         nodes_with_copy_source_count: meta.nodes_with_copy_source_count.get(),
         ignore_patterns_hash: meta.ignore_patterns_hash,
         unreachable_bytes: meta.unreachable_bytes.get(),
         old_data_size: on_disk.len(),
         dirstate_version: DirstateVersion::V2,
     };
     Ok(dirstate_map)
 }
 
 impl Node {
     pub(super) fn full_path<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<&'on_disk HgPath, DirstateV2ParseError> {
         read_hg_path(on_disk, self.full_path)
     }
 
     pub(super) fn base_name_start<'on_disk>(
         &self,
     ) -> Result<usize, DirstateV2ParseError> {
         let start = self.base_name_start.get();
         if start < self.full_path.len.get() {
             let start = usize::try_from(start)
                 // u32 -> usize, could only panic on a 16-bit CPU
                 .expect("dirstate-v2 base_name_start out of bounds");
             Ok(start)
         } else {
             Err(DirstateV2ParseError::new("not enough bytes for base name"))
         }
     }
 
     pub(super) fn base_name<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<&'on_disk HgPath, DirstateV2ParseError> {
         let full_path = self.full_path(on_disk)?;
         let base_name_start = self.base_name_start()?;
         Ok(HgPath::new(&full_path.as_bytes()[base_name_start..]))
     }
 
     pub(super) fn path<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<dirstate_map::NodeKey<'on_disk>, DirstateV2ParseError> {
         Ok(WithBasename::from_raw_parts(
             Cow::Borrowed(self.full_path(on_disk)?),
             self.base_name_start()?,
         ))
     }
 
     pub(super) fn has_copy_source<'on_disk>(&self) -> bool {
         self.copy_source.start.get() != 0
     }
 
     pub(super) fn copy_source<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<Option<&'on_disk HgPath>, DirstateV2ParseError> {
         Ok(if self.has_copy_source() {
             Some(read_hg_path(on_disk, self.copy_source)?)
         } else {
             None
         })
     }
 
     fn flags(&self) -> Flags {
         Flags::from_bits_truncate(self.flags.get())
     }
 
     fn has_entry(&self) -> bool {
         self.flags().intersects(
             Flags::WDIR_TRACKED | Flags::P1_TRACKED | Flags::P2_INFO,
         )
     }
 
     pub(super) fn node_data(
         &self,
     ) -> Result<dirstate_map::NodeData, DirstateV2ParseError> {
         if self.has_entry() {
             Ok(dirstate_map::NodeData::Entry(self.assume_entry()?))
         } else if let Some(mtime) = self.cached_directory_mtime()? {
             Ok(dirstate_map::NodeData::CachedDirectory { mtime })
         } else {
             Ok(dirstate_map::NodeData::None)
         }
     }
 
     pub(super) fn cached_directory_mtime(
         &self,
     ) -> Result<Option<TruncatedTimestamp>, DirstateV2ParseError> {
         // For now we do not have code to handle the absence of
         // ALL_UNKNOWN_RECORDED, so we ignore the mtime if the flag is
         // unset.
         if self.flags().contains(Flags::DIRECTORY)
             && self.flags().contains(Flags::HAS_MTIME)
             && self.flags().contains(Flags::ALL_UNKNOWN_RECORDED)
         {
             Ok(Some(self.mtime()?))
         } else {
             Ok(None)
         }
     }
 
     fn synthesize_unix_mode(&self) -> u32 {
         let file_type = if self.flags().contains(Flags::MODE_IS_SYMLINK) {
             libc::S_IFLNK
         } else {
             libc::S_IFREG
         };
         let permisions = if self.flags().contains(Flags::MODE_EXEC_PERM) {
             0o755
         } else {
             0o644
         };
         (file_type | permisions).into()
     }
 
     fn mtime(&self) -> Result<TruncatedTimestamp, DirstateV2ParseError> {
         let mut m: TruncatedTimestamp = self.mtime.try_into()?;
         if self.flags().contains(Flags::MTIME_SECOND_AMBIGUOUS) {
             m.second_ambiguous = true;
         }
         Ok(m)
     }
 
     fn assume_entry(&self) -> Result<DirstateEntry, DirstateV2ParseError> {
         // TODO: convert through raw bits instead?
         let wc_tracked = self.flags().contains(Flags::WDIR_TRACKED);
         let p1_tracked = self.flags().contains(Flags::P1_TRACKED);
         let p2_info = self.flags().contains(Flags::P2_INFO);
         let mode_size = if self.flags().contains(Flags::HAS_MODE_AND_SIZE)
             && !self.flags().contains(Flags::EXPECTED_STATE_IS_MODIFIED)
         {
             Some((self.synthesize_unix_mode(), self.size.into()))
         } else {
             None
         };
         let mtime = if self.flags().contains(Flags::HAS_MTIME)
             && !self.flags().contains(Flags::DIRECTORY)
             && !self.flags().contains(Flags::EXPECTED_STATE_IS_MODIFIED)
         {
             Some(self.mtime()?)
         } else {
             None
         };
         let fallback_exec = if self.flags().contains(Flags::HAS_FALLBACK_EXEC)
         {
             Some(self.flags().contains(Flags::FALLBACK_EXEC))
         } else {
             None
         };
         let fallback_symlink =
             if self.flags().contains(Flags::HAS_FALLBACK_SYMLINK) {
                 Some(self.flags().contains(Flags::FALLBACK_SYMLINK))
             } else {
                 None
             };
         Ok(DirstateEntry::from_v2_data(DirstateV2Data {
             wc_tracked,
             p1_tracked,
             p2_info,
             mode_size,
             mtime,
             fallback_exec,
             fallback_symlink,
         }))
     }
 
     pub(super) fn entry(
         &self,
     ) -> Result<Option<DirstateEntry>, DirstateV2ParseError> {
         if self.has_entry() {
             Ok(Some(self.assume_entry()?))
         } else {
             Ok(None)
         }
     }
 
     pub(super) fn children<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<&'on_disk [Node], DirstateV2ParseError> {
         read_nodes(on_disk, self.children)
     }
 
     pub(super) fn to_in_memory_node<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<dirstate_map::Node<'on_disk>, DirstateV2ParseError> {
         Ok(dirstate_map::Node {
             children: dirstate_map::ChildNodes::OnDisk(
                 self.children(on_disk)?,
             ),
             copy_source: self.copy_source(on_disk)?.map(Cow::Borrowed),
             data: self.node_data()?,
             descendants_with_entry_count: self
                 .descendants_with_entry_count
                 .get(),
             tracked_descendants_count: self.tracked_descendants_count.get(),
         })
     }
 
     fn from_dirstate_entry(
         entry: &DirstateEntry,
     ) -> (Flags, U32Be, PackedTruncatedTimestamp) {
         let DirstateV2Data {
             wc_tracked,
             p1_tracked,
             p2_info,
             mode_size: mode_size_opt,
             mtime: mtime_opt,
             fallback_exec,
             fallback_symlink,
         } = entry.v2_data();
         // TODO: convert through raw flag bits instead?
         let mut flags = Flags::empty();
         flags.set(Flags::WDIR_TRACKED, wc_tracked);
         flags.set(Flags::P1_TRACKED, p1_tracked);
         flags.set(Flags::P2_INFO, p2_info);
         let size = if let Some((m, s)) = mode_size_opt {
             let exec_perm = m & (libc::S_IXUSR as u32) != 0;
             let is_symlink = m & (libc::S_IFMT as u32) == libc::S_IFLNK as u32;
             flags.set(Flags::MODE_EXEC_PERM, exec_perm);
             flags.set(Flags::MODE_IS_SYMLINK, is_symlink);
             flags.insert(Flags::HAS_MODE_AND_SIZE);
             s.into()
         } else {
             0.into()
         };
         let mtime = if let Some(m) = mtime_opt {
             flags.insert(Flags::HAS_MTIME);
             if m.second_ambiguous {
                 flags.insert(Flags::MTIME_SECOND_AMBIGUOUS);
             };
             m.into()
         } else {
             PackedTruncatedTimestamp::null()
         };
         if let Some(f_exec) = fallback_exec {
             flags.insert(Flags::HAS_FALLBACK_EXEC);
             if f_exec {
                 flags.insert(Flags::FALLBACK_EXEC);
             }
         }
         if let Some(f_symlink) = fallback_symlink {
             flags.insert(Flags::HAS_FALLBACK_SYMLINK);
             if f_symlink {
                 flags.insert(Flags::FALLBACK_SYMLINK);
             }
         }
         (flags, size, mtime)
     }
 }
 
 fn read_hg_path(
     on_disk: &[u8],
     slice: PathSlice,
 ) -> Result<&HgPath, DirstateV2ParseError> {
     read_slice(on_disk, slice.start, slice.len.get()).map(HgPath::new)
 }
 
 fn read_nodes(
     on_disk: &[u8],
     slice: ChildNodes,
 ) -> Result<&[Node], DirstateV2ParseError> {
     read_slice(on_disk, slice.start, slice.len.get())
 }
 
 fn read_slice<T, Len>(
     on_disk: &[u8],
     start: Offset,
     len: Len,
 ) -> Result<&[T], DirstateV2ParseError>
 where
     T: BytesCast,
     Len: TryInto<usize>,
 {
     // Either `usize::MAX` would result in "out of bounds" error since a single
     // `&[u8]` cannot occupy the entire addess space.
     let start = start.get().try_into().unwrap_or(std::usize::MAX);
     let len = len.try_into().unwrap_or(std::usize::MAX);
     let bytes = match on_disk.get(start..) {
         Some(bytes) => bytes,
         None => {
             return Err(DirstateV2ParseError::new(
                 "not enough bytes from disk",
             ))
         }
     };
     T::slice_from_bytes(bytes, len)
         .map_err(|e| {
             DirstateV2ParseError::new(format!("when reading a slice, {}", e))
         })
         .map(|(slice, _rest)| slice)
 }
 
 pub(crate) fn for_each_tracked_path<'on_disk>(
     on_disk: &'on_disk [u8],
     metadata: &[u8],
     mut f: impl FnMut(&'on_disk HgPath),
 ) -> Result<(), DirstateV2ParseError> {
     let (meta, _) = TreeMetadata::from_bytes(metadata).map_err(|e| {
         DirstateV2ParseError::new(format!("when parsing tree metadata, {}", e))
     })?;
     fn recur<'on_disk>(
         on_disk: &'on_disk [u8],
         nodes: ChildNodes,
         f: &mut impl FnMut(&'on_disk HgPath),
     ) -> Result<(), DirstateV2ParseError> {
         for node in read_nodes(on_disk, nodes)? {
             if let Some(entry) = node.entry()? {
                 if entry.tracked() {
                     f(node.full_path(on_disk)?)
                 }
             }
             recur(on_disk, node.children, f)?
         }
         Ok(())
     }
     recur(on_disk, meta.root_nodes, &mut f)
 }
 
 /// Returns new data and metadata, together with whether that data should be
 /// appended to the existing data file whose content is at
 /// `dirstate_map.on_disk` (true), instead of written to a new data file
 /// (false), and the previous size of data on disk.
 pub(super) fn write(
     dirstate_map: &DirstateMap,
-    can_append: bool,
+    write_mode: DirstateMapWriteMode,
 ) -> Result<(Vec<u8>, TreeMetadata, bool, usize), DirstateError> {
-    let append = can_append && dirstate_map.write_should_append();
+    let append = match write_mode {
+        DirstateMapWriteMode::Auto => dirstate_map.write_should_append(),
+        DirstateMapWriteMode::ForceNewDataFile => false,
+    };
     if append {
         log::trace!("appending to the dirstate data file");
     } else {
         log::trace!("creating new dirstate data file");
     }
 
     // This ignores the space for paths, and for nodes without an entry.
     // TODO: better estimate? Skip the `Vec` and write to a file directly?
     let size_guess = std::mem::size_of::<Node>()
         * dirstate_map.nodes_with_entry_count as usize;
 
     let mut writer = Writer {
         dirstate_map,
         append,
         out: Vec::with_capacity(size_guess),
     };
 
     let root_nodes = writer.write_nodes(dirstate_map.root.as_ref())?;
 
     let unreachable_bytes = if append {
         dirstate_map.unreachable_bytes
     } else {
         0
     };
     let meta = TreeMetadata {
         root_nodes,
         nodes_with_entry_count: dirstate_map.nodes_with_entry_count.into(),
         nodes_with_copy_source_count: dirstate_map
             .nodes_with_copy_source_count
             .into(),
         unreachable_bytes: unreachable_bytes.into(),
         unused: [0; 4],
         ignore_patterns_hash: dirstate_map.ignore_patterns_hash,
     };
     Ok((writer.out, meta, append, dirstate_map.old_data_size))
 }
 
 struct Writer<'dmap, 'on_disk> {
     dirstate_map: &'dmap DirstateMap<'on_disk>,
     append: bool,
     out: Vec<u8>,
 }
 
 impl Writer<'_, '_> {
     fn write_nodes(
         &mut self,
         nodes: dirstate_map::ChildNodesRef,
     ) -> Result<ChildNodes, DirstateError> {
         // Reuse already-written nodes if possible
         if self.append {
             if let dirstate_map::ChildNodesRef::OnDisk(nodes_slice) = nodes {
                 let start = self.on_disk_offset_of(nodes_slice).expect(
                     "dirstate-v2 OnDisk nodes not found within on_disk",
                 );
                 let len = child_nodes_len_from_usize(nodes_slice.len());
                 return Ok(ChildNodes { start, len });
             }
         }
 
         // `dirstate_map::ChildNodes::InMemory` contains a `HashMap` which has
         // undefined iteration order. Sort to enable binary search in the
         // written file.
         let nodes = nodes.sorted();
         let nodes_len = nodes.len();
 
         // First accumulate serialized nodes in a `Vec`
         let mut on_disk_nodes = Vec::with_capacity(nodes_len);
         for node in nodes {
             let children =
                 self.write_nodes(node.children(self.dirstate_map.on_disk)?)?;
             let full_path = node.full_path(self.dirstate_map.on_disk)?;
             let full_path = self.write_path(full_path.as_bytes());
             let copy_source = if let Some(source) =
                 node.copy_source(self.dirstate_map.on_disk)?
             {
                 self.write_path(source.as_bytes())
             } else {
                 PathSlice {
                     start: 0.into(),
                     len: 0.into(),
                 }
             };
             on_disk_nodes.push(match node {
                 NodeRef::InMemory(path, node) => {
                     let (flags, size, mtime) = match &node.data {
                         dirstate_map::NodeData::Entry(entry) => {
                             Node::from_dirstate_entry(entry)
                         }
                         dirstate_map::NodeData::CachedDirectory { mtime } => {
                             // we currently never set a mtime if unknown file
                             // are present.
                             // So if we have a mtime for a directory, we know
                             // they are no unknown
                             // files and we
                             // blindly set ALL_UNKNOWN_RECORDED.
                             //
                             // We never set ALL_IGNORED_RECORDED since we
                             // don't track that case
                             // currently.
                             let mut flags = Flags::DIRECTORY
                                 | Flags::HAS_MTIME
                                 | Flags::ALL_UNKNOWN_RECORDED;
                             if mtime.second_ambiguous {
                                 flags.insert(Flags::MTIME_SECOND_AMBIGUOUS)
                             }
                             (flags, 0.into(), (*mtime).into())
                         }
                         dirstate_map::NodeData::None => (
                             Flags::DIRECTORY,
                             0.into(),
                             PackedTruncatedTimestamp::null(),
                         ),
                     };
                     Node {
                         children,
                         copy_source,
                         full_path,
                         base_name_start: u16::try_from(path.base_name_start())
                             // Could only panic for paths over 64 KiB
                             .expect("dirstate-v2 path length overflow")
                             .into(),
                         descendants_with_entry_count: node
                             .descendants_with_entry_count
                             .into(),
                         tracked_descendants_count: node
                             .tracked_descendants_count
                             .into(),
                         flags: flags.bits().into(),
                         size,
                         mtime,
                     }
                 }
                 NodeRef::OnDisk(node) => Node {
                     children,
                     copy_source,
                     full_path,
                     ..*node
                 },
             })
         }
         // … so we can write them contiguously, after writing everything else
         // they refer to.
         let start = self.current_offset();
         let len = child_nodes_len_from_usize(nodes_len);
         self.out.extend(on_disk_nodes.as_bytes());
         Ok(ChildNodes { start, len })
     }
 
     /// If the given slice of items is within `on_disk`, returns its offset
     /// from the start of `on_disk`.
     fn on_disk_offset_of<T>(&self, slice: &[T]) -> Option<Offset>
     where
         T: BytesCast,
     {
         fn address_range(slice: &[u8]) -> std::ops::RangeInclusive<usize> {
             let start = slice.as_ptr() as usize;
             let end = start + slice.len();
             start..=end
         }
         let slice_addresses = address_range(slice.as_bytes());
         let on_disk_addresses = address_range(self.dirstate_map.on_disk);
         if on_disk_addresses.contains(slice_addresses.start())
             && on_disk_addresses.contains(slice_addresses.end())
         {
             let offset = slice_addresses.start() - on_disk_addresses.start();
             Some(offset_from_usize(offset))
         } else {
             None
         }
     }
 
     fn current_offset(&mut self) -> Offset {
         let mut offset = self.out.len();
         if self.append {
             offset += self.dirstate_map.on_disk.len()
         }
         offset_from_usize(offset)
     }
 
     fn write_path(&mut self, slice: &[u8]) -> PathSlice {
         let len = path_len_from_usize(slice.len());
         // Reuse an already-written path if possible
         if self.append {
             if let Some(start) = self.on_disk_offset_of(slice) {
                 return PathSlice { start, len };
             }
         }
         let start = self.current_offset();
         self.out.extend(slice.as_bytes());
         PathSlice { start, len }
     }
 }
 
 fn offset_from_usize(x: usize) -> Offset {
     u32::try_from(x)
         // Could only panic for a dirstate file larger than 4 GiB
         .expect("dirstate-v2 offset overflow")
         .into()
 }
 
 fn child_nodes_len_from_usize(x: usize) -> Size {
     u32::try_from(x)
         // Could only panic with over 4 billion nodes
         .expect("dirstate-v2 slice length overflow")
         .into()
 }
 
 fn path_len_from_usize(x: usize) -> PathSize {
     u16::try_from(x)
         // Could only panic for paths over 64 KiB
         .expect("dirstate-v2 path length overflow")
         .into()
 }
 
 impl From<TruncatedTimestamp> for PackedTruncatedTimestamp {
     fn from(timestamp: TruncatedTimestamp) -> Self {
         Self {
             truncated_seconds: timestamp.truncated_seconds().into(),
             nanoseconds: timestamp.nanoseconds().into(),
         }
     }
 }
 
 impl TryFrom<PackedTruncatedTimestamp> for TruncatedTimestamp {
     type Error = DirstateV2ParseError;
 
     fn try_from(
         timestamp: PackedTruncatedTimestamp,
     ) -> Result<Self, Self::Error> {
         Self::from_already_truncated(
             timestamp.truncated_seconds.get(),
             timestamp.nanoseconds.get(),
             false,
         )
     }
 }
 impl PackedTruncatedTimestamp {
     fn null() -> Self {
         Self {
             truncated_seconds: 0.into(),
             nanoseconds: 0.into(),
         }
     }
 }

rust/hg-core/src/repo.rs

@@ -1,577 +1,582 b''
 use crate::changelog::Changelog;
 use crate::config::{Config, ConfigError, ConfigParseError};
 use crate::dirstate::DirstateParents;
+use crate::dirstate_tree::dirstate_map::DirstateMapWriteMode;
 use crate::dirstate_tree::on_disk::Docket as DirstateDocket;
 use crate::dirstate_tree::owning::OwningDirstateMap;
 use crate::errors::HgResultExt;
 use crate::errors::{HgError, IoResultExt};
 use crate::lock::{try_with_lock_no_wait, LockError};
 use crate::manifest::{Manifest, Manifestlog};
 use crate::revlog::filelog::Filelog;
 use crate::revlog::revlog::RevlogError;
 use crate::utils::files::get_path_from_bytes;
 use crate::utils::hg_path::HgPath;
 use crate::utils::SliceExt;
 use crate::vfs::{is_dir, is_file, Vfs};
 use crate::{requirements, NodePrefix};
 use crate::{DirstateError, Revision};
 use std::cell::{Ref, RefCell, RefMut};
 use std::collections::HashSet;
 use std::io::Seek;
 use std::io::SeekFrom;
 use std::io::Write as IoWrite;
 use std::path::{Path, PathBuf};
 
 /// A repository on disk
 pub struct Repo {
     working_directory: PathBuf,
     dot_hg: PathBuf,
     store: PathBuf,
     requirements: HashSet<String>,
     config: Config,
     dirstate_parents: LazyCell<DirstateParents>,
     dirstate_data_file_uuid: LazyCell<Option<Vec<u8>>>,
     dirstate_map: LazyCell<OwningDirstateMap>,
     changelog: LazyCell<Changelog>,
     manifestlog: LazyCell<Manifestlog>,
 }
 
 #[derive(Debug, derive_more::From)]
 pub enum RepoError {
     NotFound {
         at: PathBuf,
     },
     #[from]
     ConfigParseError(ConfigParseError),
     #[from]
     Other(HgError),
 }
 
 impl From<ConfigError> for RepoError {
     fn from(error: ConfigError) -> Self {
         match error {
             ConfigError::Parse(error) => error.into(),
             ConfigError::Other(error) => error.into(),
         }
     }
 }
 
 impl Repo {
     /// tries to find nearest repository root in current working directory or
     /// its ancestors
     pub fn find_repo_root() -> Result<PathBuf, RepoError> {
         let current_directory = crate::utils::current_dir()?;
         // ancestors() is inclusive: it first yields `current_directory`
         // as-is.
         for ancestor in current_directory.ancestors() {
             if is_dir(ancestor.join(".hg"))? {
                 return Ok(ancestor.to_path_buf());
             }
         }
         return Err(RepoError::NotFound {
             at: current_directory,
         });
     }
 
     /// Find a repository, either at the given path (which must contain a `.hg`
     /// sub-directory) or by searching the current directory and its
     /// ancestors.
     ///
     /// A method with two very different "modes" like this usually a code smell
     /// to make two methods instead, but in this case an `Option` is what rhg
     /// sub-commands get from Clap for the `-R` / `--repository` CLI argument.
     /// Having two methods would just move that `if` to almost all callers.
     pub fn find(
         config: &Config,
         explicit_path: Option<PathBuf>,
     ) -> Result<Self, RepoError> {
         if let Some(root) = explicit_path {
             if is_dir(root.join(".hg"))? {
                 Self::new_at_path(root.to_owned(), config)
             } else if is_file(&root)? {
                 Err(HgError::unsupported("bundle repository").into())
             } else {
                 Err(RepoError::NotFound {
                     at: root.to_owned(),
                 })
             }
         } else {
             let root = Self::find_repo_root()?;
             Self::new_at_path(root, config)
         }
     }
 
     /// To be called after checking that `.hg` is a sub-directory
     fn new_at_path(
         working_directory: PathBuf,
         config: &Config,
     ) -> Result<Self, RepoError> {
         let dot_hg = working_directory.join(".hg");
 
         let mut repo_config_files = Vec::new();
         repo_config_files.push(dot_hg.join("hgrc"));
         repo_config_files.push(dot_hg.join("hgrc-not-shared"));
 
         let hg_vfs = Vfs { base: &dot_hg };
         let mut reqs = requirements::load_if_exists(hg_vfs)?;
         let relative =
             reqs.contains(requirements::RELATIVE_SHARED_REQUIREMENT);
         let shared =
             reqs.contains(requirements::SHARED_REQUIREMENT) || relative;
 
         // From `mercurial/localrepo.py`:
         //
         // 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
         let share_safe = reqs.contains(requirements::SHARESAFE_REQUIREMENT);
 
         let store_path;
         if !shared {
             store_path = dot_hg.join("store");
         } else {
             let bytes = hg_vfs.read("sharedpath")?;
             let mut shared_path =
                 get_path_from_bytes(bytes.trim_end_matches(|b| b == b'\n'))
                     .to_owned();
             if relative {
                 shared_path = dot_hg.join(shared_path)
             }
             if !is_dir(&shared_path)? {
                 return Err(HgError::corrupted(format!(
                     ".hg/sharedpath points to nonexistent directory {}",
                     shared_path.display()
                 ))
                 .into());
             }
 
             store_path = shared_path.join("store");
 
             let source_is_share_safe =
                 requirements::load(Vfs { base: &shared_path })?
                     .contains(requirements::SHARESAFE_REQUIREMENT);
 
             if share_safe != source_is_share_safe {
                 return Err(HgError::unsupported("share-safe mismatch").into());
             }
 
             if share_safe {
                 repo_config_files.insert(0, shared_path.join("hgrc"))
             }
         }
         if share_safe {
             reqs.extend(requirements::load(Vfs { base: &store_path })?);
         }
 
         let repo_config = if std::env::var_os("HGRCSKIPREPO").is_none() {
             config.combine_with_repo(&repo_config_files)?
         } else {
             config.clone()
         };
 
         let repo = Self {
             requirements: reqs,
             working_directory,
             store: store_path,
             dot_hg,
             config: repo_config,
             dirstate_parents: LazyCell::new(),
             dirstate_data_file_uuid: LazyCell::new(),
             dirstate_map: LazyCell::new(),
             changelog: LazyCell::new(),
             manifestlog: LazyCell::new(),
         };
 
         requirements::check(&repo)?;
 
         Ok(repo)
     }
 
     pub fn working_directory_path(&self) -> &Path {
         &self.working_directory
     }
 
     pub fn requirements(&self) -> &HashSet<String> {
         &self.requirements
     }
 
     pub fn config(&self) -> &Config {
         &self.config
     }
 
     /// For accessing repository files (in `.hg`), except for the store
     /// (`.hg/store`).
     pub fn hg_vfs(&self) -> Vfs<'_> {
         Vfs { base: &self.dot_hg }
     }
 
     /// For accessing repository store files (in `.hg/store`)
     pub fn store_vfs(&self) -> Vfs<'_> {
         Vfs { base: &self.store }
     }
 
     /// For accessing the working copy
     pub fn working_directory_vfs(&self) -> Vfs<'_> {
         Vfs {
             base: &self.working_directory,
         }
     }
 
     pub fn try_with_wlock_no_wait<R>(
         &self,
         f: impl FnOnce() -> R,
     ) -> Result<R, LockError> {
         try_with_lock_no_wait(self.hg_vfs(), "wlock", f)
     }
 
     pub fn has_dirstate_v2(&self) -> bool {
         self.requirements
             .contains(requirements::DIRSTATE_V2_REQUIREMENT)
     }
 
     pub fn has_sparse(&self) -> bool {
         self.requirements.contains(requirements::SPARSE_REQUIREMENT)
     }
 
     pub fn has_narrow(&self) -> bool {
         self.requirements.contains(requirements::NARROW_REQUIREMENT)
     }
 
     pub fn has_nodemap(&self) -> bool {
         self.requirements
             .contains(requirements::NODEMAP_REQUIREMENT)
     }
 
     fn dirstate_file_contents(&self) -> Result<Vec<u8>, HgError> {
         Ok(self
             .hg_vfs()
             .read("dirstate")
             .io_not_found_as_none()?
             .unwrap_or(Vec::new()))
     }
 
     pub fn dirstate_parents(&self) -> Result<DirstateParents, HgError> {
         Ok(*self
             .dirstate_parents
             .get_or_init(|| self.read_dirstate_parents())?)
     }
 
     fn read_dirstate_parents(&self) -> Result<DirstateParents, HgError> {
         let dirstate = self.dirstate_file_contents()?;
         let parents = if dirstate.is_empty() {
             if self.has_dirstate_v2() {
                 self.dirstate_data_file_uuid.set(None);
             }
             DirstateParents::NULL
         } else if self.has_dirstate_v2() {
             let docket =
                 crate::dirstate_tree::on_disk::read_docket(&dirstate)?;
             self.dirstate_data_file_uuid
                 .set(Some(docket.uuid.to_owned()));
             docket.parents()
         } else {
             crate::dirstate::parsers::parse_dirstate_parents(&dirstate)?
                 .clone()
         };
         self.dirstate_parents.set(parents);
         Ok(parents)
     }
 
     fn read_dirstate_data_file_uuid(
         &self,
     ) -> Result<Option<Vec<u8>>, HgError> {
         assert!(
             self.has_dirstate_v2(),
             "accessing dirstate data file ID without dirstate-v2"
         );
         let dirstate = self.dirstate_file_contents()?;
         if dirstate.is_empty() {
             self.dirstate_parents.set(DirstateParents::NULL);
             Ok(None)
         } else {
             let docket =
                 crate::dirstate_tree::on_disk::read_docket(&dirstate)?;
             self.dirstate_parents.set(docket.parents());
             Ok(Some(docket.uuid.to_owned()))
         }
     }
 
     fn new_dirstate_map(&self) -> Result<OwningDirstateMap, DirstateError> {
         let dirstate_file_contents = self.dirstate_file_contents()?;
         if dirstate_file_contents.is_empty() {
             self.dirstate_parents.set(DirstateParents::NULL);
             if self.has_dirstate_v2() {
                 self.dirstate_data_file_uuid.set(None);
             }
             Ok(OwningDirstateMap::new_empty(Vec::new()))
         } else if self.has_dirstate_v2() {
             let docket = crate::dirstate_tree::on_disk::read_docket(
                 &dirstate_file_contents,
             )?;
             self.dirstate_parents.set(docket.parents());
             self.dirstate_data_file_uuid
                 .set(Some(docket.uuid.to_owned()));
             let data_size = docket.data_size();
             let metadata = docket.tree_metadata();
             if crate::vfs::is_on_nfs_mount(docket.data_filename()) {
                 // Don't mmap on NFS to prevent `SIGBUS` error on deletion
                 OwningDirstateMap::new_v2(
                     self.hg_vfs().read(docket.data_filename())?,
                     data_size,
                     metadata,
                 )
             } else if let Some(data_mmap) = self
                 .hg_vfs()
                 .mmap_open(docket.data_filename())
                 .io_not_found_as_none()?
             {
                 OwningDirstateMap::new_v2(data_mmap, data_size, metadata)
             } else {
                 OwningDirstateMap::new_v2(Vec::new(), data_size, metadata)
             }
         } else {
             let (map, parents) =
                 OwningDirstateMap::new_v1(dirstate_file_contents)?;
             self.dirstate_parents.set(parents);
             Ok(map)
         }
     }
 
     pub fn dirstate_map(
         &self,
     ) -> Result<Ref<OwningDirstateMap>, DirstateError> {
         self.dirstate_map.get_or_init(|| self.new_dirstate_map())
     }
 
     pub fn dirstate_map_mut(
         &self,
     ) -> Result<RefMut<OwningDirstateMap>, DirstateError> {
         self.dirstate_map
             .get_mut_or_init(|| self.new_dirstate_map())
     }
 
     fn new_changelog(&self) -> Result<Changelog, HgError> {
         Changelog::open(&self.store_vfs(), self.has_nodemap())
     }
 
     pub fn changelog(&self) -> Result<Ref<Changelog>, HgError> {
         self.changelog.get_or_init(|| self.new_changelog())
     }
 
     pub fn changelog_mut(&self) -> Result<RefMut<Changelog>, HgError> {
         self.changelog.get_mut_or_init(|| self.new_changelog())
     }
 
     fn new_manifestlog(&self) -> Result<Manifestlog, HgError> {
         Manifestlog::open(&self.store_vfs(), self.has_nodemap())
     }
 
     pub fn manifestlog(&self) -> Result<Ref<Manifestlog>, HgError> {
         self.manifestlog.get_or_init(|| self.new_manifestlog())
     }
 
     pub fn manifestlog_mut(&self) -> Result<RefMut<Manifestlog>, HgError> {
         self.manifestlog.get_mut_or_init(|| self.new_manifestlog())
     }
 
     /// Returns the manifest of the *changeset* with the given node ID
     pub fn manifest_for_node(
         &self,
         node: impl Into<NodePrefix>,
     ) -> Result<Manifest, RevlogError> {
         self.manifestlog()?.data_for_node(
             self.changelog()?
                 .data_for_node(node.into())?
                 .manifest_node()?
                 .into(),
         )
     }
 
     /// Returns the manifest of the *changeset* with the given revision number
     pub fn manifest_for_rev(
         &self,
         revision: Revision,
     ) -> Result<Manifest, RevlogError> {
         self.manifestlog()?.data_for_node(
             self.changelog()?
                 .data_for_rev(revision)?
                 .manifest_node()?
                 .into(),
         )
     }
 
     pub fn has_subrepos(&self) -> Result<bool, DirstateError> {
         if let Some(entry) = self.dirstate_map()?.get(HgPath::new(".hgsub"))? {
             Ok(entry.tracked())
         } else {
             Ok(false)
         }
     }
 
     pub fn filelog(&self, path: &HgPath) -> Result<Filelog, HgError> {
         Filelog::open(self, path)
     }
 
     /// Write to disk any updates that were made through `dirstate_map_mut`.
     ///
     /// The "wlock" must be held while calling this.
     /// See for example `try_with_wlock_no_wait`.
     ///
     /// TODO: have a `WritableRepo` type only accessible while holding the
     /// lock?
     pub fn write_dirstate(&self) -> Result<(), DirstateError> {
         let map = self.dirstate_map()?;
         // TODO: Maintain a `DirstateMap::dirty` flag, and return early here if
         // it’s unset
         let parents = self.dirstate_parents()?;
         let (packed_dirstate, old_uuid_to_remove) = if self.has_dirstate_v2() {
             let uuid_opt = self
                 .dirstate_data_file_uuid
                 .get_or_init(|| self.read_dirstate_data_file_uuid())?;
             let uuid_opt = uuid_opt.as_ref();
-            let can_append = uuid_opt.is_some();
+            let write_mode = if uuid_opt.is_some() {
+                DirstateMapWriteMode::Auto
+            } else {
+                DirstateMapWriteMode::ForceNewDataFile
+            };
             let (data, tree_metadata, append, old_data_size) =
-                map.pack_v2(can_append)?;
+                map.pack_v2(write_mode)?;
 
             // Reuse the uuid, or generate a new one, keeping the old for
             // deletion.
             let (uuid, old_uuid) = match uuid_opt {
                 Some(uuid) => {
                     let as_str = std::str::from_utf8(uuid)
                         .map_err(|_| {
                             HgError::corrupted(
                                 "non-UTF-8 dirstate data file ID",
                             )
                         })?
                         .to_owned();
                     if append {
                         (as_str, None)
                     } else {
                         (DirstateDocket::new_uid(), Some(as_str))
                     }
                 }
                 None => (DirstateDocket::new_uid(), None),
             };
 
             let data_filename = format!("dirstate.{}", uuid);
             let data_filename = self.hg_vfs().join(data_filename);
             let mut options = std::fs::OpenOptions::new();
             options.write(true);
 
             // Why are we not using the O_APPEND flag when appending?
             //
             // - O_APPEND makes it trickier to deal with garbage at the end of
             //   the file, left by a previous uncommitted transaction. By
             //   starting the write at [old_data_size] we make sure we erase
             //   all such garbage.
             //
             // - O_APPEND requires to special-case 0-byte writes, whereas we
             //   don't need that.
             //
             // - Some OSes have bugs in implementation O_APPEND:
             //   revlog.py talks about a Solaris bug, but we also saw some ZFS
             //   bug: https://github.com/openzfs/zfs/pull/3124,
             //   https://github.com/openzfs/zfs/issues/13370
             //
             if !append {
                 log::trace!("creating a new dirstate data file");
                 options.create_new(true);
             } else {
                 log::trace!("appending to the dirstate data file");
             }
 
             let data_size = (|| {
                 // TODO: loop and try another random ID if !append and this
                 // returns `ErrorKind::AlreadyExists`? Collision chance of two
                 // random IDs is one in 2**32
                 let mut file = options.open(&data_filename)?;
                 if append {
                     file.seek(SeekFrom::Start(old_data_size as u64))?;
                 }
                 file.write_all(&data)?;
                 file.flush()?;
                 file.seek(SeekFrom::Current(0))
             })()
             .when_writing_file(&data_filename)?;
 
             let packed_dirstate = DirstateDocket::serialize(
                 parents,
                 tree_metadata,
                 data_size,
                 uuid.as_bytes(),
             )
             .map_err(|_: std::num::TryFromIntError| {
                 HgError::corrupted("overflow in dirstate docket serialization")
             })?;
 
             (packed_dirstate, old_uuid)
         } else {
             (map.pack_v1(parents)?, None)
         };
 
         let vfs = self.hg_vfs();
         vfs.atomic_write("dirstate", &packed_dirstate)?;
         if let Some(uuid) = old_uuid_to_remove {
             // Remove the old data file after the new docket pointing to the
             // new data file was written.
             vfs.remove_file(format!("dirstate.{}", uuid))?;
         }
         Ok(())
     }
 }
 
 /// Lazily-initialized component of `Repo` with interior mutability
 ///
 /// This differs from `OnceCell` in that the value can still be "deinitialized"
 /// later by setting its inner `Option` to `None`. It also takes the
 /// initialization function as an argument when the value is requested, not
 /// when the instance is created.
 struct LazyCell<T> {
     value: RefCell<Option<T>>,
 }
 
 impl<T> LazyCell<T> {
     fn new() -> Self {
         Self {
             value: RefCell::new(None),
         }
     }
 
     fn set(&self, value: T) {
         *self.value.borrow_mut() = Some(value)
     }
 
     fn get_or_init<E>(
         &self,
         init: impl Fn() -> Result<T, E>,
     ) -> Result<Ref<T>, E> {
         let mut borrowed = self.value.borrow();
         if borrowed.is_none() {
             drop(borrowed);
             // Only use `borrow_mut` if it is really needed to avoid panic in
             // case there is another outstanding borrow but mutation is not
             // needed.
             *self.value.borrow_mut() = Some(init()?);
             borrowed = self.value.borrow()
         }
         Ok(Ref::map(borrowed, |option| option.as_ref().unwrap()))
     }
 
     fn get_mut_or_init<E>(
         &self,
         init: impl Fn() -> Result<T, E>,
     ) -> Result<RefMut<T>, E> {
         let mut borrowed = self.value.borrow_mut();
         if borrowed.is_none() {
             *borrowed = Some(init()?);
         }
         Ok(RefMut::map(borrowed, |option| option.as_mut().unwrap()))
     }
 }

rust/hg-cpython/src/dirstate/dirstate_map.rs

@@ -1,544 +1,550 b''
 // dirstate_map.rs
 //
 // Copyright 2019 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.
 
 //! Bindings for the `hg::dirstate::dirstate_map` file provided by the
 //! `hg-core` package.
 
 use std::cell::{RefCell, RefMut};
 use std::convert::TryInto;
 
 use cpython::{
     exc, PyBool, PyBytes, PyClone, PyDict, PyErr, PyList, PyNone, PyObject,
     PyResult, Python, PythonObject, ToPyObject, UnsafePyLeaked,
 };
 use hg::dirstate::{ParentFileData, TruncatedTimestamp};
 
 use crate::{
     dirstate::copymap::{CopyMap, CopyMapItemsIterator, CopyMapKeysIterator},
     dirstate::item::DirstateItem,
     pybytes_deref::PyBytesDeref,
 };
 use hg::{
-    dirstate::StateMapIter, dirstate_tree::on_disk::DirstateV2ParseError,
+    dirstate::StateMapIter, dirstate_tree::dirstate_map::DirstateMapWriteMode,
+    dirstate_tree::on_disk::DirstateV2ParseError,
     dirstate_tree::owning::OwningDirstateMap, revlog::Node,
     utils::files::normalize_case, utils::hg_path::HgPath, DirstateEntry,
     DirstateError, DirstateParents,
 };
 
 // TODO
 //     This object needs to share references to multiple members of its Rust
 //     inner struct, namely `copy_map`, `dirs` and `all_dirs`.
 //     Right now `CopyMap` is done, but it needs to have an explicit reference
 //     to `RustDirstateMap` which itself needs to have an encapsulation for
 //     every method in `CopyMap` (copymapcopy, etc.).
 //     This is ugly and hard to maintain.
 //     The same logic applies to `dirs` and `all_dirs`, however the `Dirs`
 //     `py_class!` is already implemented and does not mention
 //     `RustDirstateMap`, rightfully so.
 //     All attributes also have to have a separate refcount data attribute for
 //     leaks, with all methods that go along for reference sharing.
 py_class!(pub class DirstateMap |py| {
     @shared data inner: OwningDirstateMap;
 
     /// Returns a `(dirstate_map, parents)` tuple
     @staticmethod
     def new_v1(
         on_disk: PyBytes,
     ) -> PyResult<PyObject> {
         let on_disk = PyBytesDeref::new(py, on_disk);
         let (map, parents) = OwningDirstateMap::new_v1(on_disk)
             .map_err(|e| dirstate_error(py, e))?;
         let map = Self::create_instance(py, map)?;
         let p1 = PyBytes::new(py, parents.p1.as_bytes());
         let p2 = PyBytes::new(py, parents.p2.as_bytes());
         let parents = (p1, p2);
         Ok((map, parents).to_py_object(py).into_object())
     }
 
     /// Returns a DirstateMap
     @staticmethod
     def new_v2(
         on_disk: PyBytes,
         data_size: usize,
         tree_metadata: PyBytes,
     ) -> PyResult<PyObject> {
         let dirstate_error = |e: DirstateError| {
             PyErr::new::<exc::OSError, _>(py, format!("Dirstate error: {:?}", e))
         };
         let on_disk = PyBytesDeref::new(py, on_disk);
         let map = OwningDirstateMap::new_v2(
             on_disk, data_size, tree_metadata.data(py),
         ).map_err(dirstate_error)?;
         let map = Self::create_instance(py, map)?;
         Ok(map.into_object())
     }
 
     def clear(&self) -> PyResult<PyObject> {
         self.inner(py).borrow_mut().clear();
         Ok(py.None())
     }
 
     def get(
         &self,
         key: PyObject,
         default: Option<PyObject> = None
     ) -> PyResult<Option<PyObject>> {
         let key = key.extract::<PyBytes>(py)?;
         match self
             .inner(py)
             .borrow()
             .get(HgPath::new(key.data(py)))
             .map_err(|e| v2_error(py, e))?
         {
             Some(entry) => {
                 Ok(Some(DirstateItem::new_as_pyobject(py, entry)?))
             },
             None => Ok(default)
         }
     }
 
     def set_tracked(&self, f: PyObject) -> PyResult<PyBool> {
         let bytes = f.extract::<PyBytes>(py)?;
         let path = HgPath::new(bytes.data(py));
         let res = self.inner(py).borrow_mut().set_tracked(path);
         let was_tracked = res.or_else(|_| {
             Err(PyErr::new::<exc::OSError, _>(py, "Dirstate error".to_string()))
         })?;
         Ok(was_tracked.to_py_object(py))
     }
 
     def set_untracked(&self, f: PyObject) -> PyResult<PyBool> {
         let bytes = f.extract::<PyBytes>(py)?;
         let path = HgPath::new(bytes.data(py));
         let res = self.inner(py).borrow_mut().set_untracked(path);
         let was_tracked = res.or_else(|_| {
             Err(PyErr::new::<exc::OSError, _>(py, "Dirstate error".to_string()))
         })?;
         Ok(was_tracked.to_py_object(py))
     }
 
     def set_clean(
         &self,
         f: PyObject,
         mode: u32,
         size: u32,
         mtime: (i64, u32, bool)
     ) -> PyResult<PyNone> {
         let (mtime_s, mtime_ns, second_ambiguous) = mtime;
         let timestamp = TruncatedTimestamp::new_truncate(
             mtime_s, mtime_ns, second_ambiguous
         );
         let bytes = f.extract::<PyBytes>(py)?;
         let path = HgPath::new(bytes.data(py));
         let res = self.inner(py).borrow_mut().set_clean(
             path, mode, size, timestamp,
         );
         res.or_else(|_| {
             Err(PyErr::new::<exc::OSError, _>(py, "Dirstate error".to_string()))
         })?;
         Ok(PyNone)
     }
 
     def set_possibly_dirty(&self, f: PyObject) -> PyResult<PyNone> {
         let bytes = f.extract::<PyBytes>(py)?;
         let path = HgPath::new(bytes.data(py));
         let res = self.inner(py).borrow_mut().set_possibly_dirty(path);
         res.or_else(|_| {
             Err(PyErr::new::<exc::OSError, _>(py, "Dirstate error".to_string()))
         })?;
         Ok(PyNone)
     }
 
     def reset_state(
         &self,
         f: PyObject,
         wc_tracked: bool,
         p1_tracked: bool,
         p2_info: bool,
         has_meaningful_mtime: bool,
         parentfiledata: Option<(u32, u32, Option<(i64, u32, bool)>)>,
     ) -> PyResult<PyNone> {
         let mut has_meaningful_mtime = has_meaningful_mtime;
         let parent_file_data = match parentfiledata {
             None => {
                 has_meaningful_mtime = false;
                 None
             },
             Some(data) => {
                 let (mode, size, mtime_info) = data;
                 let mtime = if let Some(mtime_info) = mtime_info {
                     let (mtime_s, mtime_ns, second_ambiguous) = mtime_info;
                     let timestamp = TruncatedTimestamp::new_truncate(
                         mtime_s, mtime_ns, second_ambiguous
                     );
                     Some(timestamp)
                 } else {
                     has_meaningful_mtime = false;
                     None
                 };
                 Some(ParentFileData {
                     mode_size: Some((mode, size)),
                     mtime,
                 })
             }
         };
         let bytes = f.extract::<PyBytes>(py)?;
         let path = HgPath::new(bytes.data(py));
         let res = self.inner(py).borrow_mut().reset_state(
             path,
             wc_tracked,
             p1_tracked,
             p2_info,
             has_meaningful_mtime,
             parent_file_data,
         );
         res.or_else(|_| {
             Err(PyErr::new::<exc::OSError, _>(py, "Dirstate error".to_string()))
         })?;
         Ok(PyNone)
     }
 
     def hastrackeddir(&self, d: PyObject) -> PyResult<PyBool> {
         let d = d.extract::<PyBytes>(py)?;
         Ok(self.inner(py).borrow_mut()
             .has_tracked_dir(HgPath::new(d.data(py)))
             .map_err(|e| {
                 PyErr::new::<exc::ValueError, _>(py, e.to_string())
             })?
             .to_py_object(py))
     }
 
     def hasdir(&self, d: PyObject) -> PyResult<PyBool> {
         let d = d.extract::<PyBytes>(py)?;
         Ok(self.inner(py).borrow_mut()
             .has_dir(HgPath::new(d.data(py)))
             .map_err(|e| {
                 PyErr::new::<exc::ValueError, _>(py, e.to_string())
             })?
             .to_py_object(py))
     }
 
     def write_v1(
         &self,
         p1: PyObject,
         p2: PyObject,
     ) -> PyResult<PyBytes> {
         let inner = self.inner(py).borrow();
         let parents = DirstateParents {
             p1: extract_node_id(py, &p1)?,
             p2: extract_node_id(py, &p2)?,
         };
         let result = inner.pack_v1(parents);
         match result {
             Ok(packed) => Ok(PyBytes::new(py, &packed)),
             Err(_) => Err(PyErr::new::<exc::OSError, _>(
                 py,
                 "Dirstate error".to_string(),
             )),
         }
     }
 
     /// Returns new data together with whether that data should be appended to
     /// the existing data file whose content is at `self.on_disk` (True),
     /// instead of written to a new data file (False).
     def write_v2(
         &self,
-        can_append: bool,
+        write_mode: usize,
     ) -> PyResult<PyObject> {
         let inner = self.inner(py).borrow();
-        let result = inner.pack_v2(can_append);
+        let rust_write_mode = match write_mode {
+            0 => DirstateMapWriteMode::Auto,
+            1 => DirstateMapWriteMode::ForceNewDataFile,
+            _ => DirstateMapWriteMode::Auto, // XXX should we error out?
+        };
+        let result = inner.pack_v2(rust_write_mode);
         match result {
             Ok((packed, tree_metadata, append, _old_data_size)) => {
                 let packed = PyBytes::new(py, &packed);
                 let tree_metadata = PyBytes::new(py, tree_metadata.as_bytes());
                 let tuple = (packed, tree_metadata, append);
                 Ok(tuple.to_py_object(py).into_object())
             },
             Err(_) => Err(PyErr::new::<exc::OSError, _>(
                 py,
                 "Dirstate error".to_string(),
             )),
         }
     }
 
     def filefoldmapasdict(&self) -> PyResult<PyDict> {
         let dict = PyDict::new(py);
         for item in self.inner(py).borrow_mut().iter() {
             let (path, entry) = item.map_err(|e| v2_error(py, e))?;
             if !entry.removed() {
                 let key = normalize_case(path);
                 let value = path;
                 dict.set_item(
                     py,
                     PyBytes::new(py, key.as_bytes()).into_object(),
                     PyBytes::new(py, value.as_bytes()).into_object(),
                 )?;
             }
         }
         Ok(dict)
     }
 
     def __len__(&self) -> PyResult<usize> {
         Ok(self.inner(py).borrow().len())
     }
 
     def __contains__(&self, key: PyObject) -> PyResult<bool> {
         let key = key.extract::<PyBytes>(py)?;
         self.inner(py)
             .borrow()
             .contains_key(HgPath::new(key.data(py)))
             .map_err(|e| v2_error(py, e))
     }
 
     def __getitem__(&self, key: PyObject) -> PyResult<PyObject> {
         let key = key.extract::<PyBytes>(py)?;
         let key = HgPath::new(key.data(py));
         match self
             .inner(py)
             .borrow()
             .get(key)
             .map_err(|e| v2_error(py, e))?
         {
             Some(entry) => {
                 Ok(DirstateItem::new_as_pyobject(py, entry)?)
             },
             None => Err(PyErr::new::<exc::KeyError, _>(
                 py,
                 String::from_utf8_lossy(key.as_bytes()),
             )),
         }
     }
 
     def keys(&self) -> PyResult<DirstateMapKeysIterator> {
         let leaked_ref = self.inner(py).leak_immutable();
         DirstateMapKeysIterator::from_inner(
             py,
             unsafe { leaked_ref.map(py, |o| o.iter()) },
         )
     }
 
     def items(&self) -> PyResult<DirstateMapItemsIterator> {
         let leaked_ref = self.inner(py).leak_immutable();
         DirstateMapItemsIterator::from_inner(
             py,
             unsafe { leaked_ref.map(py, |o| o.iter()) },
         )
     }
 
     def __iter__(&self) -> PyResult<DirstateMapKeysIterator> {
         let leaked_ref = self.inner(py).leak_immutable();
         DirstateMapKeysIterator::from_inner(
             py,
             unsafe { leaked_ref.map(py, |o| o.iter()) },
         )
     }
 
     // TODO all copymap* methods, see docstring above
     def copymapcopy(&self) -> PyResult<PyDict> {
         let dict = PyDict::new(py);
         for item in self.inner(py).borrow().copy_map_iter() {
             let (key, value) = item.map_err(|e| v2_error(py, e))?;
             dict.set_item(
                 py,
                 PyBytes::new(py, key.as_bytes()),
                 PyBytes::new(py, value.as_bytes()),
             )?;
         }
         Ok(dict)
     }
 
     def copymapgetitem(&self, key: PyObject) -> PyResult<PyBytes> {
         let key = key.extract::<PyBytes>(py)?;
         match self
             .inner(py)
             .borrow()
             .copy_map_get(HgPath::new(key.data(py)))
             .map_err(|e| v2_error(py, e))?
         {
             Some(copy) => Ok(PyBytes::new(py, copy.as_bytes())),
             None => Err(PyErr::new::<exc::KeyError, _>(
                 py,
                 String::from_utf8_lossy(key.data(py)),
             )),
         }
     }
     def copymap(&self) -> PyResult<CopyMap> {
         CopyMap::from_inner(py, self.clone_ref(py))
     }
 
     def copymaplen(&self) -> PyResult<usize> {
         Ok(self.inner(py).borrow().copy_map_len())
     }
     def copymapcontains(&self, key: PyObject) -> PyResult<bool> {
         let key = key.extract::<PyBytes>(py)?;
         self.inner(py)
             .borrow()
             .copy_map_contains_key(HgPath::new(key.data(py)))
             .map_err(|e| v2_error(py, e))
     }
     def copymapget(
         &self,
         key: PyObject,
         default: Option<PyObject>
     ) -> PyResult<Option<PyObject>> {
         let key = key.extract::<PyBytes>(py)?;
         match self
             .inner(py)
             .borrow()
             .copy_map_get(HgPath::new(key.data(py)))
             .map_err(|e| v2_error(py, e))?
         {
             Some(copy) => Ok(Some(
                 PyBytes::new(py, copy.as_bytes()).into_object(),
             )),
             None => Ok(default),
         }
     }
     def copymapsetitem(
         &self,
         key: PyObject,
         value: PyObject
     ) -> PyResult<PyObject> {
         let key = key.extract::<PyBytes>(py)?;
         let value = value.extract::<PyBytes>(py)?;
         self.inner(py)
             .borrow_mut()
             .copy_map_insert(
                 HgPath::new(key.data(py)),
                 HgPath::new(value.data(py)),
             )
             .map_err(|e| v2_error(py, e))?;
         Ok(py.None())
     }
     def copymappop(
         &self,
         key: PyObject,
         default: Option<PyObject>
     ) -> PyResult<Option<PyObject>> {
         let key = key.extract::<PyBytes>(py)?;
         match self
             .inner(py)
             .borrow_mut()
             .copy_map_remove(HgPath::new(key.data(py)))
             .map_err(|e| v2_error(py, e))?
         {
             Some(copy) => Ok(Some(
                 PyBytes::new(py, copy.as_bytes()).into_object(),
             )),
             None => Ok(default),
         }
     }
 
     def copymapiter(&self) -> PyResult<CopyMapKeysIterator> {
         let leaked_ref = self.inner(py).leak_immutable();
         CopyMapKeysIterator::from_inner(
             py,
             unsafe { leaked_ref.map(py, |o| o.copy_map_iter()) },
         )
     }
 
     def copymapitemsiter(&self) -> PyResult<CopyMapItemsIterator> {
         let leaked_ref = self.inner(py).leak_immutable();
         CopyMapItemsIterator::from_inner(
             py,
             unsafe { leaked_ref.map(py, |o| o.copy_map_iter()) },
         )
     }
 
     def tracked_dirs(&self) -> PyResult<PyList> {
         let dirs = PyList::new(py, &[]);
         for path in self.inner(py).borrow_mut().iter_tracked_dirs()
             .map_err(|e |dirstate_error(py, e))?
         {
             let path = path.map_err(|e| v2_error(py, e))?;
             let path = PyBytes::new(py, path.as_bytes());
             dirs.append(py, path.into_object())
         }
         Ok(dirs)
     }
 
     def setparents_fixup(&self) -> PyResult<PyDict> {
         let dict = PyDict::new(py);
         let copies = self.inner(py).borrow_mut().setparents_fixup();
         for (key, value) in copies.map_err(|e| v2_error(py, e))? {
             dict.set_item(
                 py,
                 PyBytes::new(py, key.as_bytes()),
                 PyBytes::new(py, value.as_bytes()),
             )?;
         }
         Ok(dict)
     }
 
     def debug_iter(&self, all: bool) -> PyResult<PyList> {
         let dirs = PyList::new(py, &[]);
         for item in self.inner(py).borrow().debug_iter(all) {
             let (path, (state, mode, size, mtime)) =
                 item.map_err(|e| v2_error(py, e))?;
             let path = PyBytes::new(py, path.as_bytes());
             let item = (path, state, mode, size, mtime);
             dirs.append(py, item.to_py_object(py).into_object())
         }
         Ok(dirs)
     }
 });
 
 impl DirstateMap {
     pub fn get_inner_mut<'a>(
         &'a self,
         py: Python<'a>,
     ) -> RefMut<'a, OwningDirstateMap> {
         self.inner(py).borrow_mut()
     }
     fn translate_key(
         py: Python,
         res: Result<(&HgPath, DirstateEntry), DirstateV2ParseError>,
     ) -> PyResult<Option<PyBytes>> {
         let (f, _entry) = res.map_err(|e| v2_error(py, e))?;
         Ok(Some(PyBytes::new(py, f.as_bytes())))
     }
     fn translate_key_value(
         py: Python,
         res: Result<(&HgPath, DirstateEntry), DirstateV2ParseError>,
     ) -> PyResult<Option<(PyBytes, PyObject)>> {
         let (f, entry) = res.map_err(|e| v2_error(py, e))?;
         Ok(Some((
             PyBytes::new(py, f.as_bytes()),
             DirstateItem::new_as_pyobject(py, entry)?,
         )))
     }
 }
 
 py_shared_iterator!(
     DirstateMapKeysIterator,
     UnsafePyLeaked<StateMapIter<'static>>,
     DirstateMap::translate_key,
     Option<PyBytes>
 );
 
 py_shared_iterator!(
     DirstateMapItemsIterator,
     UnsafePyLeaked<StateMapIter<'static>>,
     DirstateMap::translate_key_value,
     Option<(PyBytes, PyObject)>
 );
 
 fn extract_node_id(py: Python, obj: &PyObject) -> PyResult<Node> {
     let bytes = obj.extract::<PyBytes>(py)?;
     match bytes.data(py).try_into() {
         Ok(s) => Ok(s),
         Err(e) => Err(PyErr::new::<exc::ValueError, _>(py, e.to_string())),
     }
 }
 
 pub(super) fn v2_error(py: Python<'_>, _: DirstateV2ParseError) -> PyErr {
     PyErr::new::<exc::ValueError, _>(py, "corrupted dirstate-v2")
 }
 
 fn dirstate_error(py: Python<'_>, e: DirstateError) -> PyErr {
     PyErr::new::<exc::OSError, _>(py, format!("Dirstate error: {:?}", e))
 }