revlog: improve the robustness of the splitting process

The previous "in-place" splitting, preserving the splitting on transaction
failure had a couple of issue in case of transaction rollback:
- a race windows that could still lead to a crash and data loss
- it corrupted the `fncache`.

So instead, we use a new approach that we summarized as "we do a backup of the
inline revlog pre-split, and we restore this in case of failure".
To make readers live easier, we don't overwrite the inline index file until
transaction finalization. (once the transaction get into its finalization phase,
it is not expected to rollback, unless some crash happens).

To do so, we write the index of the split index in a temporary file that we use
until transaction finalization. We also keep a backup of the initial inline file
to be able to rollback the split if needed.

As a result, transaction rollback cancel the split and no longer corrupt
fncache. We also no longer have a small inconsistency windows where the
transaction could be unrecoverable.

mercurial/changelog.py

@@ -1,632 +1,632 b''
 # changelog.py - changelog class for mercurial
 #
 # Copyright 2005-2007 Olivia Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 
 from .i18n import _
 from .node import (
     bin,
     hex,
 )
 from .thirdparty import attr
 
 from . import (
     encoding,
     error,
     metadata,
     pycompat,
     revlog,
 )
 from .utils import (
     dateutil,
     stringutil,
 )
 from .revlogutils import (
     constants as revlog_constants,
     flagutil,
 )
 
 _defaultextra = {b'branch': b'default'}
 
 
 def _string_escape(text):
     """
     >>> from .pycompat import bytechr as chr
     >>> d = {b'nl': chr(10), b'bs': chr(92), b'cr': chr(13), b'nul': chr(0)}
     >>> s = b"ab%(nl)scd%(bs)s%(bs)sn%(nul)s12ab%(cr)scd%(bs)s%(nl)s" % d
     >>> s
     'ab\\ncd\\\\\\\\n\\x0012ab\\rcd\\\\\\n'
     >>> res = _string_escape(s)
     >>> s == _string_unescape(res)
     True
     """
     # subset of the string_escape codec
     text = (
         text.replace(b'\\', b'\\\\')
         .replace(b'\n', b'\\n')
         .replace(b'\r', b'\\r')
     )
     return text.replace(b'\0', b'\\0')
 
 
 def _string_unescape(text):
     if b'\\0' in text:
         # fix up \0 without getting into trouble with \\0
         text = text.replace(b'\\\\', b'\\\\\n')
         text = text.replace(b'\\0', b'\0')
         text = text.replace(b'\n', b'')
     return stringutil.unescapestr(text)
 
 
 def decodeextra(text):
     """
     >>> from .pycompat import bytechr as chr
     >>> sorted(decodeextra(encodeextra({b'foo': b'bar', b'baz': chr(0) + b'2'})
     ...                    ).items())
     [('baz', '\\x002'), ('branch', 'default'), ('foo', 'bar')]
     >>> sorted(decodeextra(encodeextra({b'foo': b'bar',
     ...                                 b'baz': chr(92) + chr(0) + b'2'})
     ...                    ).items())
     [('baz', '\\\\\\x002'), ('branch', 'default'), ('foo', 'bar')]
     """
     extra = _defaultextra.copy()
     for l in text.split(b'\0'):
         if l:
             k, v = _string_unescape(l).split(b':', 1)
             extra[k] = v
     return extra
 
 
 def encodeextra(d):
     # keys must be sorted to produce a deterministic changelog entry
     items = [_string_escape(b'%s:%s' % (k, d[k])) for k in sorted(d)]
     return b"\0".join(items)
 
 
 def stripdesc(desc):
     """strip trailing whitespace and leading and trailing empty lines"""
     return b'\n'.join([l.rstrip() for l in desc.splitlines()]).strip(b'\n')
 
 
 class appender:
     """the changelog index must be updated last on disk, so we use this class
     to delay writes to it"""
 
     def __init__(self, vfs, name, mode, buf):
         self.data = buf
         fp = vfs(name, mode)
         self.fp = fp
         self.offset = fp.tell()
         self.size = vfs.fstat(fp).st_size
         self._end = self.size
 
     def end(self):
         return self._end
 
     def tell(self):
         return self.offset
 
     def flush(self):
         pass
 
     @property
     def closed(self):
         return self.fp.closed
 
     def close(self):
         self.fp.close()
 
     def seek(self, offset, whence=0):
         '''virtual file offset spans real file and data'''
         if whence == 0:
             self.offset = offset
         elif whence == 1:
             self.offset += offset
         elif whence == 2:
             self.offset = self.end() + offset
         if self.offset < self.size:
             self.fp.seek(self.offset)
 
     def read(self, count=-1):
         '''only trick here is reads that span real file and data'''
         ret = b""
         if self.offset < self.size:
             s = self.fp.read(count)
             ret = s
             self.offset += len(s)
             if count > 0:
                 count -= len(s)
         if count != 0:
             doff = self.offset - self.size
             self.data.insert(0, b"".join(self.data))
             del self.data[1:]
             s = self.data[0][doff : doff + count]
             self.offset += len(s)
             ret += s
         return ret
 
     def write(self, s):
         self.data.append(bytes(s))
         self.offset += len(s)
         self._end += len(s)
 
     def __enter__(self):
         self.fp.__enter__()
         return self
 
     def __exit__(self, *args):
         return self.fp.__exit__(*args)
 
 
 class _divertopener:
     def __init__(self, opener, target):
         self._opener = opener
         self._target = target
 
     def __call__(self, name, mode=b'r', checkambig=False, **kwargs):
         if name != self._target:
             return self._opener(name, mode, **kwargs)
         return self._opener(name + b".a", mode, **kwargs)
 
     def __getattr__(self, attr):
         return getattr(self._opener, attr)
 
 
 def _delayopener(opener, target, buf):
     """build an opener that stores chunks in 'buf' instead of 'target'"""
 
     def _delay(name, mode=b'r', checkambig=False, **kwargs):
         if name != target:
             return opener(name, mode, **kwargs)
         assert not kwargs
         return appender(opener, name, mode, buf)
 
     return _delay
 
 
 @attr.s
 class _changelogrevision:
     # Extensions might modify _defaultextra, so let the constructor below pass
     # it in
     extra = attr.ib()
     manifest = attr.ib()
     user = attr.ib(default=b'')
     date = attr.ib(default=(0, 0))
     files = attr.ib(default=attr.Factory(list))
     filesadded = attr.ib(default=None)
     filesremoved = attr.ib(default=None)
     p1copies = attr.ib(default=None)
     p2copies = attr.ib(default=None)
     description = attr.ib(default=b'')
     branchinfo = attr.ib(default=(_defaultextra[b'branch'], False))
 
 
 class changelogrevision:
     """Holds results of a parsed changelog revision.
 
     Changelog revisions consist of multiple pieces of data, including
     the manifest node, user, and date. This object exposes a view into
     the parsed object.
     """
 
     __slots__ = (
         '_offsets',
         '_text',
         '_sidedata',
         '_cpsd',
         '_changes',
     )
 
     def __new__(cls, cl, text, sidedata, cpsd):
         if not text:
             return _changelogrevision(extra=_defaultextra, manifest=cl.nullid)
 
         self = super(changelogrevision, cls).__new__(cls)
         # We could return here and implement the following as an __init__.
         # But doing it here is equivalent and saves an extra function call.
 
         # format used:
         # nodeid\n        : manifest node in ascii
         # user\n          : user, no \n or \r allowed
         # time tz extra\n : date (time is int or float, timezone is int)
         #                 : extra is metadata, encoded and separated by '\0'
         #                 : older versions ignore it
         # files\n\n       : files modified by the cset, no \n or \r allowed
         # (.*)            : comment (free text, ideally utf-8)
         #
         # changelog v0 doesn't use extra
 
         nl1 = text.index(b'\n')
         nl2 = text.index(b'\n', nl1 + 1)
         nl3 = text.index(b'\n', nl2 + 1)
 
         # The list of files may be empty. Which means nl3 is the first of the
         # double newline that precedes the description.
         if text[nl3 + 1 : nl3 + 2] == b'\n':
             doublenl = nl3
         else:
             doublenl = text.index(b'\n\n', nl3 + 1)
 
         self._offsets = (nl1, nl2, nl3, doublenl)
         self._text = text
         self._sidedata = sidedata
         self._cpsd = cpsd
         self._changes = None
 
         return self
 
     @property
     def manifest(self):
         return bin(self._text[0 : self._offsets[0]])
 
     @property
     def user(self):
         off = self._offsets
         return encoding.tolocal(self._text[off[0] + 1 : off[1]])
 
     @property
     def _rawdate(self):
         off = self._offsets
         dateextra = self._text[off[1] + 1 : off[2]]
         return dateextra.split(b' ', 2)[0:2]
 
     @property
     def _rawextra(self):
         off = self._offsets
         dateextra = self._text[off[1] + 1 : off[2]]
         fields = dateextra.split(b' ', 2)
         if len(fields) != 3:
             return None
 
         return fields[2]
 
     @property
     def date(self):
         raw = self._rawdate
         time = float(raw[0])
         # Various tools did silly things with the timezone.
         try:
             timezone = int(raw[1])
         except ValueError:
             timezone = 0
 
         return time, timezone
 
     @property
     def extra(self):
         raw = self._rawextra
         if raw is None:
             return _defaultextra
 
         return decodeextra(raw)
 
     @property
     def changes(self):
         if self._changes is not None:
             return self._changes
         if self._cpsd:
             changes = metadata.decode_files_sidedata(self._sidedata)
         else:
             changes = metadata.ChangingFiles(
                 touched=self.files or (),
                 added=self.filesadded or (),
                 removed=self.filesremoved or (),
                 p1_copies=self.p1copies or {},
                 p2_copies=self.p2copies or {},
             )
         self._changes = changes
         return changes
 
     @property
     def files(self):
         if self._cpsd:
             return sorted(self.changes.touched)
         off = self._offsets
         if off[2] == off[3]:
             return []
 
         return self._text[off[2] + 1 : off[3]].split(b'\n')
 
     @property
     def filesadded(self):
         if self._cpsd:
             return self.changes.added
         else:
             rawindices = self.extra.get(b'filesadded')
         if rawindices is None:
             return None
         return metadata.decodefileindices(self.files, rawindices)
 
     @property
     def filesremoved(self):
         if self._cpsd:
             return self.changes.removed
         else:
             rawindices = self.extra.get(b'filesremoved')
         if rawindices is None:
             return None
         return metadata.decodefileindices(self.files, rawindices)
 
     @property
     def p1copies(self):
         if self._cpsd:
             return self.changes.copied_from_p1
         else:
             rawcopies = self.extra.get(b'p1copies')
         if rawcopies is None:
             return None
         return metadata.decodecopies(self.files, rawcopies)
 
     @property
     def p2copies(self):
         if self._cpsd:
             return self.changes.copied_from_p2
         else:
             rawcopies = self.extra.get(b'p2copies')
         if rawcopies is None:
             return None
         return metadata.decodecopies(self.files, rawcopies)
 
     @property
     def description(self):
         return encoding.tolocal(self._text[self._offsets[3] + 2 :])
 
     @property
     def branchinfo(self):
         extra = self.extra
         return encoding.tolocal(extra.get(b"branch")), b'close' in extra
 
 
 class changelog(revlog.revlog):
     def __init__(self, opener, trypending=False, concurrencychecker=None):
         """Load a changelog revlog using an opener.
 
         If ``trypending`` is true, we attempt to load the index from a
         ``00changelog.i.a`` file instead of the default ``00changelog.i``.
         The ``00changelog.i.a`` file contains index (and possibly inline
         revision) data for a transaction that hasn't been finalized yet.
         It exists in a separate file to facilitate readers (such as
         hooks processes) accessing data before a transaction is finalized.
 
         ``concurrencychecker`` will be passed to the revlog init function, see
         the documentation there.
         """
         revlog.revlog.__init__(
             self,
             opener,
             target=(revlog_constants.KIND_CHANGELOG, None),
             radix=b'00changelog',
             checkambig=True,
             mmaplargeindex=True,
             persistentnodemap=opener.options.get(b'persistent-nodemap', False),
             concurrencychecker=concurrencychecker,
             trypending=trypending,
         )
 
         if self._initempty and (self._format_version == revlog.REVLOGV1):
             # changelogs don't benefit from generaldelta.
 
             self._format_flags &= ~revlog.FLAG_GENERALDELTA
             self._generaldelta = False
 
         # Delta chains for changelogs tend to be very small because entries
         # tend to be small and don't delta well with each. So disable delta
         # chains.
         self._storedeltachains = False
 
         self._realopener = opener
         self._delayed = False
         self._delaybuf = None
         self._divert = False
         self._filteredrevs = frozenset()
         self._filteredrevs_hashcache = {}
         self._copiesstorage = opener.options.get(b'copies-storage')
 
     @property
     def filteredrevs(self):
         return self._filteredrevs
 
     @filteredrevs.setter
     def filteredrevs(self, val):
         # Ensure all updates go through this function
         assert isinstance(val, frozenset)
         self._filteredrevs = val
         self._filteredrevs_hashcache = {}
 
     def _write_docket(self, tr):
         if not self._delayed:
             super(changelog, self)._write_docket(tr)
 
     def delayupdate(self, tr):
         """delay visibility of index updates to other readers"""
         if self._docket is None and not self._delayed:
             if len(self) == 0:
                 self._divert = True
                 if self._realopener.exists(self._indexfile + b'.a'):
                     self._realopener.unlink(self._indexfile + b'.a')
                 self.opener = _divertopener(self._realopener, self._indexfile)
             else:
                 self._delaybuf = []
                 self.opener = _delayopener(
                     self._realopener, self._indexfile, self._delaybuf
                 )
             self._segmentfile.opener = self.opener
             self._segmentfile_sidedata.opener = self.opener
         self._delayed = True
         tr.addpending(b'cl-%i' % id(self), self._writepending)
         tr.addfinalize(b'cl-%i' % id(self), self._finalize)
 
     def _finalize(self, tr):
         """finalize index updates"""
         self._delayed = False
         self.opener = self._realopener
         self._segmentfile.opener = self.opener
         self._segmentfile_sidedata.opener = self.opener
         # move redirected index data back into place
         if self._docket is not None:
             self._write_docket(tr)
         elif self._divert:
             assert not self._delaybuf
             tmpname = self._indexfile + b".a"
             nfile = self.opener.open(tmpname)
             nfile.close()
             self.opener.rename(tmpname, self._indexfile, checkambig=True)
         elif self._delaybuf:
             fp = self.opener(self._indexfile, b'a', checkambig=True)
             fp.write(b"".join(self._delaybuf))
             fp.close()
             self._delaybuf = None
         self._divert = False
         # split when we're done
-        self._enforceinlinesize(tr)
+        self._enforceinlinesize(tr, side_write=False)
 
     def _writepending(self, tr):
         """create a file containing the unfinalized state for
         pretxnchangegroup"""
         if self._docket:
             return self._docket.write(tr, pending=True)
         if self._delaybuf:
             # make a temporary copy of the index
             fp1 = self._realopener(self._indexfile)
             pendingfilename = self._indexfile + b".a"
             # register as a temp file to ensure cleanup on failure
             tr.registertmp(pendingfilename)
             # write existing data
             fp2 = self._realopener(pendingfilename, b"w")
             fp2.write(fp1.read())
             # add pending data
             fp2.write(b"".join(self._delaybuf))
             fp2.close()
             # switch modes so finalize can simply rename
             self._delaybuf = None
             self._divert = True
             self.opener = _divertopener(self._realopener, self._indexfile)
             self._segmentfile.opener = self.opener
             self._segmentfile_sidedata.opener = self.opener
 
         if self._divert:
             return True
 
         return False
 
-    def _enforceinlinesize(self, tr):
+    def _enforceinlinesize(self, tr, side_write=True):
         if not self._delayed:
-            revlog.revlog._enforceinlinesize(self, tr)
+            revlog.revlog._enforceinlinesize(self, tr, side_write=side_write)
 
     def read(self, nodeorrev):
         """Obtain data from a parsed changelog revision.
 
         Returns a 6-tuple of:
 
            - manifest node in binary
            - author/user as a localstr
            - date as a 2-tuple of (time, timezone)
            - list of files
            - commit message as a localstr
            - dict of extra metadata
 
         Unless you need to access all fields, consider calling
         ``changelogrevision`` instead, as it is faster for partial object
         access.
         """
         d = self._revisiondata(nodeorrev)
         sidedata = self.sidedata(nodeorrev)
         copy_sd = self._copiesstorage == b'changeset-sidedata'
         c = changelogrevision(self, d, sidedata, copy_sd)
         return (c.manifest, c.user, c.date, c.files, c.description, c.extra)
 
     def changelogrevision(self, nodeorrev):
         """Obtain a ``changelogrevision`` for a node or revision."""
         text = self._revisiondata(nodeorrev)
         sidedata = self.sidedata(nodeorrev)
         return changelogrevision(
             self, text, sidedata, self._copiesstorage == b'changeset-sidedata'
         )
 
     def readfiles(self, nodeorrev):
         """
         short version of read that only returns the files modified by the cset
         """
         text = self.revision(nodeorrev)
         if not text:
             return []
         last = text.index(b"\n\n")
         l = text[:last].split(b'\n')
         return l[3:]
 
     def add(
         self,
         manifest,
         files,
         desc,
         transaction,
         p1,
         p2,
         user,
         date=None,
         extra=None,
     ):
         # Convert to UTF-8 encoded bytestrings as the very first
         # thing: calling any method on a localstr object will turn it
         # into a str object and the cached UTF-8 string is thus lost.
         user, desc = encoding.fromlocal(user), encoding.fromlocal(desc)
 
         user = user.strip()
         # An empty username or a username with a "\n" will make the
         # revision text contain two "\n\n" sequences -> corrupt
         # repository since read cannot unpack the revision.
         if not user:
             raise error.StorageError(_(b"empty username"))
         if b"\n" in user:
             raise error.StorageError(
                 _(b"username %r contains a newline") % pycompat.bytestr(user)
             )
 
         desc = stripdesc(desc)
 
         if date:
             parseddate = b"%d %d" % dateutil.parsedate(date)
         else:
             parseddate = b"%d %d" % dateutil.makedate()
         if extra:
             branch = extra.get(b"branch")
             if branch in (b"default", b""):
                 del extra[b"branch"]
             elif branch in (b".", b"null", b"tip"):
                 raise error.StorageError(
                     _(b'the name \'%s\' is reserved') % branch
                 )
         sortedfiles = sorted(files.touched)
         flags = 0
         sidedata = None
         if self._copiesstorage == b'changeset-sidedata':
             if files.has_copies_info:
                 flags |= flagutil.REVIDX_HASCOPIESINFO
             sidedata = metadata.encode_files_sidedata(files)
 
         if extra:
             extra = encodeextra(extra)
             parseddate = b"%s %s" % (parseddate, extra)
         l = [hex(manifest), user, parseddate] + sortedfiles + [b"", desc]
         text = b"\n".join(l)
         rev = self.addrevision(
             text, transaction, len(self), p1, p2, sidedata=sidedata, flags=flags
         )
         return self.node(rev)
 
     def branchinfo(self, rev):
         """return the branch name and open/close state of a revision
 
         This function exists because creating a changectx object
         just to access this is costly."""
         return self.changelogrevision(rev).branchinfo
 
     def _nodeduplicatecallback(self, transaction, rev):
         # keep track of revisions that got "re-added", eg: unbunde of know rev.
         #
         # We track them in a list to preserve their order from the source bundle
         duplicates = transaction.changes.setdefault(b'revduplicates', [])
         duplicates.append(rev)

mercurial/filelog.py

@@ -1,303 +1,304 b''
 # filelog.py - file history class for mercurial
 #
 # Copyright 2005-2007 Olivia Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 
 from .i18n import _
 from .node import nullrev
 from . import (
     error,
     revlog,
 )
 from .interfaces import (
     repository,
     util as interfaceutil,
 )
 from .utils import storageutil
 from .revlogutils import (
     constants as revlog_constants,
     rewrite,
 )
 
 
 @interfaceutil.implementer(repository.ifilestorage)
 class filelog:
-    def __init__(self, opener, path):
+    def __init__(self, opener, path, try_split=False):
         self._revlog = revlog.revlog(
             opener,
             # XXX should use the unencoded path
             target=(revlog_constants.KIND_FILELOG, path),
             radix=b'/'.join((b'data', path)),
             censorable=True,
             canonical_parent_order=False,  # see comment in revlog.py
+            try_split=try_split,
         )
         # Full name of the user visible file, relative to the repository root.
         # Used by LFS.
         self._revlog.filename = path
         self.nullid = self._revlog.nullid
         opts = opener.options
         self._fix_issue6528 = opts.get(b'issue6528.fix-incoming', True)
 
     def __len__(self):
         return len(self._revlog)
 
     def __iter__(self):
         return self._revlog.__iter__()
 
     def hasnode(self, node):
         if node in (self.nullid, nullrev):
             return False
 
         try:
             self._revlog.rev(node)
             return True
         except (TypeError, ValueError, IndexError, error.LookupError):
             return False
 
     def revs(self, start=0, stop=None):
         return self._revlog.revs(start=start, stop=stop)
 
     def parents(self, node):
         return self._revlog.parents(node)
 
     def parentrevs(self, rev):
         return self._revlog.parentrevs(rev)
 
     def rev(self, node):
         return self._revlog.rev(node)
 
     def node(self, rev):
         return self._revlog.node(rev)
 
     def lookup(self, node):
         return storageutil.fileidlookup(
             self._revlog, node, self._revlog.display_id
         )
 
     def linkrev(self, rev):
         return self._revlog.linkrev(rev)
 
     def commonancestorsheads(self, node1, node2):
         return self._revlog.commonancestorsheads(node1, node2)
 
     # Used by dagop.blockdescendants().
     def descendants(self, revs):
         return self._revlog.descendants(revs)
 
     def heads(self, start=None, stop=None):
         return self._revlog.heads(start, stop)
 
     # Used by hgweb, children extension.
     def children(self, node):
         return self._revlog.children(node)
 
     def iscensored(self, rev):
         return self._revlog.iscensored(rev)
 
     def revision(self, node, _df=None):
         return self._revlog.revision(node, _df=_df)
 
     def rawdata(self, node, _df=None):
         return self._revlog.rawdata(node, _df=_df)
 
     def emitrevisions(
         self,
         nodes,
         nodesorder=None,
         revisiondata=False,
         assumehaveparentrevisions=False,
         deltamode=repository.CG_DELTAMODE_STD,
         sidedata_helpers=None,
         debug_info=None,
     ):
         return self._revlog.emitrevisions(
             nodes,
             nodesorder=nodesorder,
             revisiondata=revisiondata,
             assumehaveparentrevisions=assumehaveparentrevisions,
             deltamode=deltamode,
             sidedata_helpers=sidedata_helpers,
             debug_info=debug_info,
         )
 
     def addrevision(
         self,
         revisiondata,
         transaction,
         linkrev,
         p1,
         p2,
         node=None,
         flags=revlog.REVIDX_DEFAULT_FLAGS,
         cachedelta=None,
     ):
         return self._revlog.addrevision(
             revisiondata,
             transaction,
             linkrev,
             p1,
             p2,
             node=node,
             flags=flags,
             cachedelta=cachedelta,
         )
 
     def addgroup(
         self,
         deltas,
         linkmapper,
         transaction,
         addrevisioncb=None,
         duplicaterevisioncb=None,
         maybemissingparents=False,
         debug_info=None,
         delta_base_reuse_policy=None,
     ):
         if maybemissingparents:
             raise error.Abort(
                 _(
                     b'revlog storage does not support missing '
                     b'parents write mode'
                 )
             )
 
         with self._revlog._writing(transaction):
 
             if self._fix_issue6528:
                 deltas = rewrite.filter_delta_issue6528(self._revlog, deltas)
 
             return self._revlog.addgroup(
                 deltas,
                 linkmapper,
                 transaction,
                 addrevisioncb=addrevisioncb,
                 duplicaterevisioncb=duplicaterevisioncb,
                 debug_info=debug_info,
                 delta_base_reuse_policy=delta_base_reuse_policy,
             )
 
     def getstrippoint(self, minlink):
         return self._revlog.getstrippoint(minlink)
 
     def strip(self, minlink, transaction):
         return self._revlog.strip(minlink, transaction)
 
     def censorrevision(self, tr, node, tombstone=b''):
         return self._revlog.censorrevision(tr, node, tombstone=tombstone)
 
     def files(self):
         return self._revlog.files()
 
     def read(self, node):
         return storageutil.filtermetadata(self.revision(node))
 
     def add(self, text, meta, transaction, link, p1=None, p2=None):
         if meta or text.startswith(b'\1\n'):
             text = storageutil.packmeta(meta, text)
         rev = self.addrevision(text, transaction, link, p1, p2)
         return self.node(rev)
 
     def renamed(self, node):
         return storageutil.filerevisioncopied(self, node)
 
     def size(self, rev):
         """return the size of a given revision"""
 
         # for revisions with renames, we have to go the slow way
         node = self.node(rev)
         if self.iscensored(rev):
             return 0
         if self.renamed(node):
             return len(self.read(node))
 
         # XXX if self.read(node).startswith("\1\n"), this returns (size+4)
         # XXX See also basefilectx.cmp.
         return self._revlog.size(rev)
 
     def cmp(self, node, text):
         """compare text with a given file revision
 
         returns True if text is different than what is stored.
         """
         return not storageutil.filedataequivalent(self, node, text)
 
     def verifyintegrity(self, state):
         return self._revlog.verifyintegrity(state)
 
     def storageinfo(
         self,
         exclusivefiles=False,
         sharedfiles=False,
         revisionscount=False,
         trackedsize=False,
         storedsize=False,
     ):
         return self._revlog.storageinfo(
             exclusivefiles=exclusivefiles,
             sharedfiles=sharedfiles,
             revisionscount=revisionscount,
             trackedsize=trackedsize,
             storedsize=storedsize,
         )
 
     # Used by repo upgrade.
     def clone(self, tr, destrevlog, **kwargs):
         if not isinstance(destrevlog, filelog):
             msg = b'expected filelog to clone(), not %r'
             msg %= destrevlog
             raise error.ProgrammingError(msg)
 
         return self._revlog.clone(tr, destrevlog._revlog, **kwargs)
 
 
 class narrowfilelog(filelog):
     """Filelog variation to be used with narrow stores."""
 
-    def __init__(self, opener, path, narrowmatch):
-        super(narrowfilelog, self).__init__(opener, path)
+    def __init__(self, opener, path, narrowmatch, try_split=False):
+        super(narrowfilelog, self).__init__(opener, path, try_split=try_split)
         self._narrowmatch = narrowmatch
 
     def renamed(self, node):
         res = super(narrowfilelog, self).renamed(node)
 
         # Renames that come from outside the narrowspec are problematic
         # because we may lack the base text for the rename. This can result
         # in code attempting to walk the ancestry or compute a diff
         # encountering a missing revision. We address this by silently
         # removing rename metadata if the source file is outside the
         # narrow spec.
         #
         # A better solution would be to see if the base revision is available,
         # rather than assuming it isn't.
         #
         # An even better solution would be to teach all consumers of rename
         # metadata that the base revision may not be available.
         #
         # TODO consider better ways of doing this.
         if res and not self._narrowmatch(res[0]):
             return None
 
         return res
 
     def size(self, rev):
         # Because we have a custom renamed() that may lie, we need to call
         # the base renamed() to report accurate results.
         node = self.node(rev)
         if super(narrowfilelog, self).renamed(node):
             return len(self.read(node))
         else:
             return super(narrowfilelog, self).size(rev)
 
     def cmp(self, node, text):
         # We don't call `super` because narrow parents can be buggy in case of a
         # ambiguous dirstate. Always take the slow path until there is a better
         # fix, see issue6150.
 
         # Censored files compare against the empty file.
         if self.iscensored(self.rev(node)):
             return text != b''
 
         return self.read(node) != text

mercurial/localrepo.py

@@ -1,3984 +1,3995 b''
 # localrepo.py - read/write repository class for mercurial
 # coding: utf-8
 #
 # Copyright 2005-2007 Olivia Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 
 import functools
 import os
 import random
 import re
 import sys
 import time
 import weakref
 
 from concurrent import futures
 from typing import (
     Optional,
 )
 
 from .i18n import _
 from .node import (
     bin,
     hex,
     nullrev,
     sha1nodeconstants,
     short,
 )
 from .pycompat import (
     delattr,
     getattr,
 )
 from . import (
     bookmarks,
     branchmap,
     bundle2,
     bundlecaches,
     changegroup,
     color,
     commit,
     context,
     dirstate,
     discovery,
     encoding,
     error,
     exchange,
     extensions,
     filelog,
     hook,
     lock as lockmod,
     match as matchmod,
     mergestate as mergestatemod,
     mergeutil,
     namespaces,
     narrowspec,
     obsolete,
     pathutil,
     phases,
     pushkey,
     pycompat,
     rcutil,
     repoview,
     requirements as requirementsmod,
     revlog,
     revset,
     revsetlang,
     scmutil,
     sparse,
     store as storemod,
     subrepoutil,
     tags as tagsmod,
     transaction,
     txnutil,
     util,
     vfs as vfsmod,
     wireprototypes,
 )
 
 from .interfaces import (
     repository,
     util as interfaceutil,
 )
 
 from .utils import (
     hashutil,
     procutil,
     stringutil,
     urlutil,
 )
 
 from .revlogutils import (
     concurrency_checker as revlogchecker,
     constants as revlogconst,
     sidedata as sidedatamod,
 )
 
 release = lockmod.release
 urlerr = util.urlerr
 urlreq = util.urlreq
 
 RE_SKIP_DIRSTATE_ROLLBACK = re.compile(
     b"^((dirstate|narrowspec.dirstate).*|branch$)"
 )
 
 # set of (path, vfs-location) tuples. vfs-location is:
 # - 'plain for vfs relative paths
 # - '' for svfs relative paths
 _cachedfiles = set()
 
 
 class _basefilecache(scmutil.filecache):
     """All filecache usage on repo are done for logic that should be unfiltered"""
 
     def __get__(self, repo, type=None):
         if repo is None:
             return self
         # proxy to unfiltered __dict__ since filtered repo has no entry
         unfi = repo.unfiltered()
         try:
             return unfi.__dict__[self.sname]
         except KeyError:
             pass
         return super(_basefilecache, self).__get__(unfi, type)
 
     def set(self, repo, value):
         return super(_basefilecache, self).set(repo.unfiltered(), value)
 
 
 class repofilecache(_basefilecache):
     """filecache for files in .hg but outside of .hg/store"""
 
     def __init__(self, *paths):
         super(repofilecache, self).__init__(*paths)
         for path in paths:
             _cachedfiles.add((path, b'plain'))
 
     def join(self, obj, fname):
         return obj.vfs.join(fname)
 
 
 class storecache(_basefilecache):
     """filecache for files in the store"""
 
     def __init__(self, *paths):
         super(storecache, self).__init__(*paths)
         for path in paths:
             _cachedfiles.add((path, b''))
 
     def join(self, obj, fname):
         return obj.sjoin(fname)
 
 
 class changelogcache(storecache):
     """filecache for the changelog"""
 
     def __init__(self):
         super(changelogcache, self).__init__()
         _cachedfiles.add((b'00changelog.i', b''))
         _cachedfiles.add((b'00changelog.n', b''))
 
     def tracked_paths(self, obj):
         paths = [self.join(obj, b'00changelog.i')]
         if obj.store.opener.options.get(b'persistent-nodemap', False):
             paths.append(self.join(obj, b'00changelog.n'))
         return paths
 
 
 class manifestlogcache(storecache):
     """filecache for the manifestlog"""
 
     def __init__(self):
         super(manifestlogcache, self).__init__()
         _cachedfiles.add((b'00manifest.i', b''))
         _cachedfiles.add((b'00manifest.n', b''))
 
     def tracked_paths(self, obj):
         paths = [self.join(obj, b'00manifest.i')]
         if obj.store.opener.options.get(b'persistent-nodemap', False):
             paths.append(self.join(obj, b'00manifest.n'))
         return paths
 
 
 class mixedrepostorecache(_basefilecache):
     """filecache for a mix files in .hg/store and outside"""
 
     def __init__(self, *pathsandlocations):
         # scmutil.filecache only uses the path for passing back into our
         # join(), so we can safely pass a list of paths and locations
         super(mixedrepostorecache, self).__init__(*pathsandlocations)
         _cachedfiles.update(pathsandlocations)
 
     def join(self, obj, fnameandlocation):
         fname, location = fnameandlocation
         if location == b'plain':
             return obj.vfs.join(fname)
         else:
             if location != b'':
                 raise error.ProgrammingError(
                     b'unexpected location: %s' % location
                 )
             return obj.sjoin(fname)
 
 
 def isfilecached(repo, name):
     """check if a repo has already cached "name" filecache-ed property
 
     This returns (cachedobj-or-None, iscached) tuple.
     """
     cacheentry = repo.unfiltered()._filecache.get(name, None)
     if not cacheentry:
         return None, False
     return cacheentry.obj, True
 
 
 class unfilteredpropertycache(util.propertycache):
     """propertycache that apply to unfiltered repo only"""
 
     def __get__(self, repo, type=None):
         unfi = repo.unfiltered()
         if unfi is repo:
             return super(unfilteredpropertycache, self).__get__(unfi)
         return getattr(unfi, self.name)
 
 
 class filteredpropertycache(util.propertycache):
     """propertycache that must take filtering in account"""
 
     def cachevalue(self, obj, value):
         object.__setattr__(obj, self.name, value)
 
 
 def hasunfilteredcache(repo, name):
     """check if a repo has an unfilteredpropertycache value for <name>"""
     return name in vars(repo.unfiltered())
 
 
 def unfilteredmethod(orig):
     """decorate method that always need to be run on unfiltered version"""
 
     @functools.wraps(orig)
     def wrapper(repo, *args, **kwargs):
         return orig(repo.unfiltered(), *args, **kwargs)
 
     return wrapper
 
 
 moderncaps = {
     b'lookup',
     b'branchmap',
     b'pushkey',
     b'known',
     b'getbundle',
     b'unbundle',
 }
 legacycaps = moderncaps.union({b'changegroupsubset'})
 
 
 @interfaceutil.implementer(repository.ipeercommandexecutor)
 class localcommandexecutor:
     def __init__(self, peer):
         self._peer = peer
         self._sent = False
         self._closed = False
 
     def __enter__(self):
         return self
 
     def __exit__(self, exctype, excvalue, exctb):
         self.close()
 
     def callcommand(self, command, args):
         if self._sent:
             raise error.ProgrammingError(
                 b'callcommand() cannot be used after sendcommands()'
             )
 
         if self._closed:
             raise error.ProgrammingError(
                 b'callcommand() cannot be used after close()'
             )
 
         # We don't need to support anything fancy. Just call the named
         # method on the peer and return a resolved future.
         fn = getattr(self._peer, pycompat.sysstr(command))
 
         f = futures.Future()
 
         try:
             result = fn(**pycompat.strkwargs(args))
         except Exception:
             pycompat.future_set_exception_info(f, sys.exc_info()[1:])
         else:
             f.set_result(result)
 
         return f
 
     def sendcommands(self):
         self._sent = True
 
     def close(self):
         self._closed = True
 
 
 @interfaceutil.implementer(repository.ipeercommands)
 class localpeer(repository.peer):
     '''peer for a local repo; reflects only the most recent API'''
 
     def __init__(self, repo, caps=None, path=None):
         super(localpeer, self).__init__(repo.ui, path=path)
 
         if caps is None:
             caps = moderncaps.copy()
         self._repo = repo.filtered(b'served')
 
         if repo._wanted_sidedata:
             formatted = bundle2.format_remote_wanted_sidedata(repo)
             caps.add(b'exp-wanted-sidedata=' + formatted)
 
         self._caps = repo._restrictcapabilities(caps)
 
     # Begin of _basepeer interface.
 
     def url(self):
         return self._repo.url()
 
     def local(self):
         return self._repo
 
     def canpush(self):
         return True
 
     def close(self):
         self._repo.close()
 
     # End of _basepeer interface.
 
     # Begin of _basewirecommands interface.
 
     def branchmap(self):
         return self._repo.branchmap()
 
     def capabilities(self):
         return self._caps
 
     def clonebundles(self):
         return self._repo.tryread(bundlecaches.CB_MANIFEST_FILE)
 
     def debugwireargs(self, one, two, three=None, four=None, five=None):
         """Used to test argument passing over the wire"""
         return b"%s %s %s %s %s" % (
             one,
             two,
             pycompat.bytestr(three),
             pycompat.bytestr(four),
             pycompat.bytestr(five),
         )
 
     def getbundle(
         self,
         source,
         heads=None,
         common=None,
         bundlecaps=None,
         remote_sidedata=None,
         **kwargs
     ):
         chunks = exchange.getbundlechunks(
             self._repo,
             source,
             heads=heads,
             common=common,
             bundlecaps=bundlecaps,
             remote_sidedata=remote_sidedata,
             **kwargs
         )[1]
         cb = util.chunkbuffer(chunks)
 
         if exchange.bundle2requested(bundlecaps):
             # When requesting a bundle2, getbundle returns a stream to make the
             # wire level function happier. We need to build a proper object
             # from it in local peer.
             return bundle2.getunbundler(self.ui, cb)
         else:
             return changegroup.getunbundler(b'01', cb, None)
 
     def heads(self):
         return self._repo.heads()
 
     def known(self, nodes):
         return self._repo.known(nodes)
 
     def listkeys(self, namespace):
         return self._repo.listkeys(namespace)
 
     def lookup(self, key):
         return self._repo.lookup(key)
 
     def pushkey(self, namespace, key, old, new):
         return self._repo.pushkey(namespace, key, old, new)
 
     def stream_out(self):
         raise error.Abort(_(b'cannot perform stream clone against local peer'))
 
     def unbundle(self, bundle, heads, url):
         """apply a bundle on a repo
 
         This function handles the repo locking itself."""
         try:
             try:
                 bundle = exchange.readbundle(self.ui, bundle, None)
                 ret = exchange.unbundle(self._repo, bundle, heads, b'push', url)
                 if util.safehasattr(ret, b'getchunks'):
                     # This is a bundle20 object, turn it into an unbundler.
                     # This little dance should be dropped eventually when the
                     # API is finally improved.
                     stream = util.chunkbuffer(ret.getchunks())
                     ret = bundle2.getunbundler(self.ui, stream)
                 return ret
             except Exception as exc:
                 # If the exception contains output salvaged from a bundle2
                 # reply, we need to make sure it is printed before continuing
                 # to fail. So we build a bundle2 with such output and consume
                 # it directly.
                 #
                 # This is not very elegant but allows a "simple" solution for
                 # issue4594
                 output = getattr(exc, '_bundle2salvagedoutput', ())
                 if output:
                     bundler = bundle2.bundle20(self._repo.ui)
                     for out in output:
                         bundler.addpart(out)
                     stream = util.chunkbuffer(bundler.getchunks())
                     b = bundle2.getunbundler(self.ui, stream)
                     bundle2.processbundle(self._repo, b)
                 raise
         except error.PushRaced as exc:
             raise error.ResponseError(
                 _(b'push failed:'), stringutil.forcebytestr(exc)
             )
 
     # End of _basewirecommands interface.
 
     # Begin of peer interface.
 
     def commandexecutor(self):
         return localcommandexecutor(self)
 
     # End of peer interface.
 
 
 @interfaceutil.implementer(repository.ipeerlegacycommands)
 class locallegacypeer(localpeer):
     """peer extension which implements legacy methods too; used for tests with
     restricted capabilities"""
 
     def __init__(self, repo, path=None):
         super(locallegacypeer, self).__init__(repo, caps=legacycaps, path=path)
 
     # Begin of baselegacywirecommands interface.
 
     def between(self, pairs):
         return self._repo.between(pairs)
 
     def branches(self, nodes):
         return self._repo.branches(nodes)
 
     def changegroup(self, nodes, source):
         outgoing = discovery.outgoing(
             self._repo, missingroots=nodes, ancestorsof=self._repo.heads()
         )
         return changegroup.makechangegroup(self._repo, outgoing, b'01', source)
 
     def changegroupsubset(self, bases, heads, source):
         outgoing = discovery.outgoing(
             self._repo, missingroots=bases, ancestorsof=heads
         )
         return changegroup.makechangegroup(self._repo, outgoing, b'01', source)
 
     # End of baselegacywirecommands interface.
 
 
 # Functions receiving (ui, features) that extensions can register to impact
 # the ability to load repositories with custom requirements. Only
 # functions defined in loaded extensions are called.
 #
 # The function receives a set of requirement strings that the repository
 # is capable of opening. Functions will typically add elements to the
 # set to reflect that the extension knows how to handle that requirements.
 featuresetupfuncs = set()
 
 
 def _getsharedvfs(hgvfs, requirements):
     """returns the vfs object pointing to root of shared source
     repo for a shared repository
 
     hgvfs is vfs pointing at .hg/ of current repo (shared one)
     requirements is a set of requirements of current repo (shared one)
     """
     # The ``shared`` or ``relshared`` requirements indicate the
     # store lives in the path contained in the ``.hg/sharedpath`` file.
     # This is an absolute path for ``shared`` and relative to
     # ``.hg/`` for ``relshared``.
     sharedpath = hgvfs.read(b'sharedpath').rstrip(b'\n')
     if requirementsmod.RELATIVE_SHARED_REQUIREMENT in requirements:
         sharedpath = util.normpath(hgvfs.join(sharedpath))
 
     sharedvfs = vfsmod.vfs(sharedpath, realpath=True)
 
     if not sharedvfs.exists():
         raise error.RepoError(
             _(b'.hg/sharedpath points to nonexistent directory %s')
             % sharedvfs.base
         )
     return sharedvfs
 
 
 def _readrequires(vfs, allowmissing):
     """reads the require file present at root of this vfs
     and return a set of requirements
 
     If allowmissing is True, we suppress FileNotFoundError if raised"""
     # requires file contains a newline-delimited list of
     # features/capabilities the opener (us) must have in order to use
     # the repository. This file was introduced in Mercurial 0.9.2,
     # which means very old repositories may not have one. We assume
     # a missing file translates to no requirements.
     read = vfs.tryread if allowmissing else vfs.read
     return set(read(b'requires').splitlines())
 
 
 def makelocalrepository(baseui, path: bytes, intents=None):
     """Create a local repository object.
 
     Given arguments needed to construct a local repository, this function
     performs various early repository loading functionality (such as
     reading the ``.hg/requires`` and ``.hg/hgrc`` files), validates that
     the repository can be opened, derives a type suitable for representing
     that repository, and returns an instance of it.
 
     The returned object conforms to the ``repository.completelocalrepository``
     interface.
 
     The repository type is derived by calling a series of factory functions
     for each aspect/interface of the final repository. These are defined by
     ``REPO_INTERFACES``.
 
     Each factory function is called to produce a type implementing a specific
     interface. The cumulative list of returned types will be combined into a
     new type and that type will be instantiated to represent the local
     repository.
 
     The factory functions each receive various state that may be consulted
     as part of deriving a type.
 
     Extensions should wrap these factory functions to customize repository type
     creation. Note that an extension's wrapped function may be called even if
     that extension is not loaded for the repo being constructed. Extensions
     should check if their ``__name__`` appears in the
     ``extensionmodulenames`` set passed to the factory function and no-op if
     not.
     """
     ui = baseui.copy()
     # Prevent copying repo configuration.
     ui.copy = baseui.copy
 
     # Working directory VFS rooted at repository root.
     wdirvfs = vfsmod.vfs(path, expandpath=True, realpath=True)
 
     # Main VFS for .hg/ directory.
     hgpath = wdirvfs.join(b'.hg')
     hgvfs = vfsmod.vfs(hgpath, cacheaudited=True)
     # Whether this repository is shared one or not
     shared = False
     # If this repository is shared, vfs pointing to shared repo
     sharedvfs = None
 
     # The .hg/ path should exist and should be a directory. All other
     # cases are errors.
     if not hgvfs.isdir():
         try:
             hgvfs.stat()
         except FileNotFoundError:
             pass
         except ValueError as e:
             # Can be raised on Python 3.8 when path is invalid.
             raise error.Abort(
                 _(b'invalid path %s: %s') % (path, stringutil.forcebytestr(e))
             )
 
         raise error.RepoError(_(b'repository %s not found') % path)
 
     requirements = _readrequires(hgvfs, True)
     shared = (
         requirementsmod.SHARED_REQUIREMENT in requirements
         or requirementsmod.RELATIVE_SHARED_REQUIREMENT in requirements
     )
     storevfs = None
     if shared:
         # This is a shared repo
         sharedvfs = _getsharedvfs(hgvfs, requirements)
         storevfs = vfsmod.vfs(sharedvfs.join(b'store'))
     else:
         storevfs = vfsmod.vfs(hgvfs.join(b'store'))
 
     # if .hg/requires contains the sharesafe requirement, it means
     # there exists a `.hg/store/requires` too and we should read it
     # NOTE: presence of SHARESAFE_REQUIREMENT imply that store requirement
     # is present. We never write SHARESAFE_REQUIREMENT for a repo if store
     # is not present, refer checkrequirementscompat() for that
     #
     # However, if SHARESAFE_REQUIREMENT is not present, it means that the
     # repository was shared the old way. We check the share source .hg/requires
     # for SHARESAFE_REQUIREMENT to detect whether the current repository needs
     # to be reshared
     hint = _(b"see `hg help config.format.use-share-safe` for more information")
     if requirementsmod.SHARESAFE_REQUIREMENT in requirements:
         if (
             shared
             and requirementsmod.SHARESAFE_REQUIREMENT
             not in _readrequires(sharedvfs, True)
         ):
             mismatch_warn = ui.configbool(
                 b'share', b'safe-mismatch.source-not-safe.warn'
             )
             mismatch_config = ui.config(
                 b'share', b'safe-mismatch.source-not-safe'
             )
             mismatch_verbose_upgrade = ui.configbool(
                 b'share', b'safe-mismatch.source-not-safe:verbose-upgrade'
             )
             if mismatch_config in (
                 b'downgrade-allow',
                 b'allow',
                 b'downgrade-abort',
             ):
                 # prevent cyclic import localrepo -> upgrade -> localrepo
                 from . import upgrade
 
                 upgrade.downgrade_share_to_non_safe(
                     ui,
                     hgvfs,
                     sharedvfs,
                     requirements,
                     mismatch_config,
                     mismatch_warn,
                     mismatch_verbose_upgrade,
                 )
             elif mismatch_config == b'abort':
                 raise error.Abort(
                     _(b"share source does not support share-safe requirement"),
                     hint=hint,
                 )
             else:
                 raise error.Abort(
                     _(
                         b"share-safe mismatch with source.\nUnrecognized"
                         b" value '%s' of `share.safe-mismatch.source-not-safe`"
                         b" set."
                     )
                     % mismatch_config,
                     hint=hint,
                 )
         else:
             requirements |= _readrequires(storevfs, False)
     elif shared:
         sourcerequires = _readrequires(sharedvfs, False)
         if requirementsmod.SHARESAFE_REQUIREMENT in sourcerequires:
             mismatch_config = ui.config(b'share', b'safe-mismatch.source-safe')
             mismatch_warn = ui.configbool(
                 b'share', b'safe-mismatch.source-safe.warn'
             )
             mismatch_verbose_upgrade = ui.configbool(
                 b'share', b'safe-mismatch.source-safe:verbose-upgrade'
             )
             if mismatch_config in (
                 b'upgrade-allow',
                 b'allow',
                 b'upgrade-abort',
             ):
                 # prevent cyclic import localrepo -> upgrade -> localrepo
                 from . import upgrade
 
                 upgrade.upgrade_share_to_safe(
                     ui,
                     hgvfs,
                     storevfs,
                     requirements,
                     mismatch_config,
                     mismatch_warn,
                     mismatch_verbose_upgrade,
                 )
             elif mismatch_config == b'abort':
                 raise error.Abort(
                     _(
                         b'version mismatch: source uses share-safe'
                         b' functionality while the current share does not'
                     ),
                     hint=hint,
                 )
             else:
                 raise error.Abort(
                     _(
                         b"share-safe mismatch with source.\nUnrecognized"
                         b" value '%s' of `share.safe-mismatch.source-safe` set."
                     )
                     % mismatch_config,
                     hint=hint,
                 )
 
     # The .hg/hgrc file may load extensions or contain config options
     # that influence repository construction. Attempt to load it and
     # process any new extensions that it may have pulled in.
     if loadhgrc(ui, wdirvfs, hgvfs, requirements, sharedvfs):
         afterhgrcload(ui, wdirvfs, hgvfs, requirements)
         extensions.loadall(ui)
         extensions.populateui(ui)
 
     # Set of module names of extensions loaded for this repository.
     extensionmodulenames = {m.__name__ for n, m in extensions.extensions(ui)}
 
     supportedrequirements = gathersupportedrequirements(ui)
 
     # We first validate the requirements are known.
     ensurerequirementsrecognized(requirements, supportedrequirements)
 
     # Then we validate that the known set is reasonable to use together.
     ensurerequirementscompatible(ui, requirements)
 
     # TODO there are unhandled edge cases related to opening repositories with
     # shared storage. If storage is shared, we should also test for requirements
     # compatibility in the pointed-to repo. This entails loading the .hg/hgrc in
     # that repo, as that repo may load extensions needed to open it. This is a
     # bit complicated because we don't want the other hgrc to overwrite settings
     # in this hgrc.
     #
     # This bug is somewhat mitigated by the fact that we copy the .hg/requires
     # file when sharing repos. But if a requirement is added after the share is
     # performed, thereby introducing a new requirement for the opener, we may
     # will not see that and could encounter a run-time error interacting with
     # that shared store since it has an unknown-to-us requirement.
 
     # At this point, we know we should be capable of opening the repository.
     # Now get on with doing that.
 
     features = set()
 
     # The "store" part of the repository holds versioned data. How it is
     # accessed is determined by various requirements. If `shared` or
     # `relshared` requirements are present, this indicates current repository
     # is a share and store exists in path mentioned in `.hg/sharedpath`
     if shared:
         storebasepath = sharedvfs.base
         cachepath = sharedvfs.join(b'cache')
         features.add(repository.REPO_FEATURE_SHARED_STORAGE)
     else:
         storebasepath = hgvfs.base
         cachepath = hgvfs.join(b'cache')
     wcachepath = hgvfs.join(b'wcache')
 
     # The store has changed over time and the exact layout is dictated by
     # requirements. The store interface abstracts differences across all
     # of them.
     store = makestore(
         requirements,
         storebasepath,
         lambda base: vfsmod.vfs(base, cacheaudited=True),
     )
     hgvfs.createmode = store.createmode
 
     storevfs = store.vfs
     storevfs.options = resolvestorevfsoptions(ui, requirements, features)
 
     if (
         requirementsmod.REVLOGV2_REQUIREMENT in requirements
         or requirementsmod.CHANGELOGV2_REQUIREMENT in requirements
     ):
         features.add(repository.REPO_FEATURE_SIDE_DATA)
         # the revlogv2 docket introduced race condition that we need to fix
         features.discard(repository.REPO_FEATURE_STREAM_CLONE)
 
     # The cache vfs is used to manage cache files.
     cachevfs = vfsmod.vfs(cachepath, cacheaudited=True)
     cachevfs.createmode = store.createmode
     # The cache vfs is used to manage cache files related to the working copy
     wcachevfs = vfsmod.vfs(wcachepath, cacheaudited=True)
     wcachevfs.createmode = store.createmode
 
     # Now resolve the type for the repository object. We do this by repeatedly
     # calling a factory function to produces types for specific aspects of the
     # repo's operation. The aggregate returned types are used as base classes
     # for a dynamically-derived type, which will represent our new repository.
 
     bases = []
     extrastate = {}
 
     for iface, fn in REPO_INTERFACES:
         # We pass all potentially useful state to give extensions tons of
         # flexibility.
         typ = fn()(
             ui=ui,
             intents=intents,
             requirements=requirements,
             features=features,
             wdirvfs=wdirvfs,
             hgvfs=hgvfs,
             store=store,
             storevfs=storevfs,
             storeoptions=storevfs.options,
             cachevfs=cachevfs,
             wcachevfs=wcachevfs,
             extensionmodulenames=extensionmodulenames,
             extrastate=extrastate,
             baseclasses=bases,
         )
 
         if not isinstance(typ, type):
             raise error.ProgrammingError(
                 b'unable to construct type for %s' % iface
             )
 
         bases.append(typ)
 
     # type() allows you to use characters in type names that wouldn't be
     # recognized as Python symbols in source code. We abuse that to add
     # rich information about our constructed repo.
     name = pycompat.sysstr(
         b'derivedrepo:%s<%s>' % (wdirvfs.base, b','.join(sorted(requirements)))
     )
 
     cls = type(name, tuple(bases), {})
 
     return cls(
         baseui=baseui,
         ui=ui,
         origroot=path,
         wdirvfs=wdirvfs,
         hgvfs=hgvfs,
         requirements=requirements,
         supportedrequirements=supportedrequirements,
         sharedpath=storebasepath,
         store=store,
         cachevfs=cachevfs,
         wcachevfs=wcachevfs,
         features=features,
         intents=intents,
     )
 
 
 def loadhgrc(
     ui,
     wdirvfs: vfsmod.vfs,
     hgvfs: vfsmod.vfs,
     requirements,
     sharedvfs: Optional[vfsmod.vfs] = None,
 ):
     """Load hgrc files/content into a ui instance.
 
     This is called during repository opening to load any additional
     config files or settings relevant to the current repository.
 
     Returns a bool indicating whether any additional configs were loaded.
 
     Extensions should monkeypatch this function to modify how per-repo
     configs are loaded. For example, an extension may wish to pull in
     configs from alternate files or sources.
 
     sharedvfs is vfs object pointing to source repo if the current one is a
     shared one
     """
     if not rcutil.use_repo_hgrc():
         return False
 
     ret = False
     # first load config from shared source if we has to
     if requirementsmod.SHARESAFE_REQUIREMENT in requirements and sharedvfs:
         try:
             ui.readconfig(sharedvfs.join(b'hgrc'), root=sharedvfs.base)
             ret = True
         except IOError:
             pass
 
     try:
         ui.readconfig(hgvfs.join(b'hgrc'), root=wdirvfs.base)
         ret = True
     except IOError:
         pass
 
     try:
         ui.readconfig(hgvfs.join(b'hgrc-not-shared'), root=wdirvfs.base)
         ret = True
     except IOError:
         pass
 
     return ret
 
 
 def afterhgrcload(ui, wdirvfs, hgvfs, requirements):
     """Perform additional actions after .hg/hgrc is loaded.
 
     This function is called during repository loading immediately after
     the .hg/hgrc file is loaded and before per-repo extensions are loaded.
 
     The function can be used to validate configs, automatically add
     options (including extensions) based on requirements, etc.
     """
 
     # Map of requirements to list of extensions to load automatically when
     # requirement is present.
     autoextensions = {
         b'git': [b'git'],
         b'largefiles': [b'largefiles'],
         b'lfs': [b'lfs'],
     }
 
     for requirement, names in sorted(autoextensions.items()):
         if requirement not in requirements:
             continue
 
         for name in names:
             if not ui.hasconfig(b'extensions', name):
                 ui.setconfig(b'extensions', name, b'', source=b'autoload')
 
 
 def gathersupportedrequirements(ui):
     """Determine the complete set of recognized requirements."""
     # Start with all requirements supported by this file.
     supported = set(localrepository._basesupported)
 
     # Execute ``featuresetupfuncs`` entries if they belong to an extension
     # relevant to this ui instance.
     modules = {m.__name__ for n, m in extensions.extensions(ui)}
 
     for fn in featuresetupfuncs:
         if fn.__module__ in modules:
             fn(ui, supported)
 
     # Add derived requirements from registered compression engines.
     for name in util.compengines:
         engine = util.compengines[name]
         if engine.available() and engine.revlogheader():
             supported.add(b'exp-compression-%s' % name)
             if engine.name() == b'zstd':
                 supported.add(requirementsmod.REVLOG_COMPRESSION_ZSTD)
 
     return supported
 
 
 def ensurerequirementsrecognized(requirements, supported):
     """Validate that a set of local requirements is recognized.
 
     Receives a set of requirements. Raises an ``error.RepoError`` if there
     exists any requirement in that set that currently loaded code doesn't
     recognize.
 
     Returns a set of supported requirements.
     """
     missing = set()
 
     for requirement in requirements:
         if requirement in supported:
             continue
 
         if not requirement or not requirement[0:1].isalnum():
             raise error.RequirementError(_(b'.hg/requires file is corrupt'))
 
         missing.add(requirement)
 
     if missing:
         raise error.RequirementError(
             _(b'repository requires features unknown to this Mercurial: %s')
             % b' '.join(sorted(missing)),
             hint=_(
                 b'see https://mercurial-scm.org/wiki/MissingRequirement '
                 b'for more information'
             ),
         )
 
 
 def ensurerequirementscompatible(ui, requirements):
     """Validates that a set of recognized requirements is mutually compatible.
 
     Some requirements may not be compatible with others or require
     config options that aren't enabled. This function is called during
     repository opening to ensure that the set of requirements needed
     to open a repository is sane and compatible with config options.
 
     Extensions can monkeypatch this function to perform additional
     checking.
 
     ``error.RepoError`` should be raised on failure.
     """
     if (
         requirementsmod.SPARSE_REQUIREMENT in requirements
         and not sparse.enabled
     ):
         raise error.RepoError(
             _(
                 b'repository is using sparse feature but '
                 b'sparse is not enabled; enable the '
                 b'"sparse" extensions to access'
             )
         )
 
 
 def makestore(requirements, path, vfstype):
     """Construct a storage object for a repository."""
     if requirementsmod.STORE_REQUIREMENT in requirements:
         if requirementsmod.FNCACHE_REQUIREMENT in requirements:
             dotencode = requirementsmod.DOTENCODE_REQUIREMENT in requirements
             return storemod.fncachestore(path, vfstype, dotencode)
 
         return storemod.encodedstore(path, vfstype)
 
     return storemod.basicstore(path, vfstype)
 
 
 def resolvestorevfsoptions(ui, requirements, features):
     """Resolve the options to pass to the store vfs opener.
 
     The returned dict is used to influence behavior of the storage layer.
     """
     options = {}
 
     if requirementsmod.TREEMANIFEST_REQUIREMENT in requirements:
         options[b'treemanifest'] = True
 
     # experimental config: format.manifestcachesize
     manifestcachesize = ui.configint(b'format', b'manifestcachesize')
     if manifestcachesize is not None:
         options[b'manifestcachesize'] = manifestcachesize
 
     # In the absence of another requirement superseding a revlog-related
     # requirement, we have to assume the repo is using revlog version 0.
     # This revlog format is super old and we don't bother trying to parse
     # opener options for it because those options wouldn't do anything
     # meaningful on such old repos.
     if (
         requirementsmod.REVLOGV1_REQUIREMENT in requirements
         or requirementsmod.REVLOGV2_REQUIREMENT in requirements
     ):
         options.update(resolverevlogstorevfsoptions(ui, requirements, features))
     else:  # explicitly mark repo as using revlogv0
         options[b'revlogv0'] = True
 
     if requirementsmod.COPIESSDC_REQUIREMENT in requirements:
         options[b'copies-storage'] = b'changeset-sidedata'
     else:
         writecopiesto = ui.config(b'experimental', b'copies.write-to')
         copiesextramode = (b'changeset-only', b'compatibility')
         if writecopiesto in copiesextramode:
             options[b'copies-storage'] = b'extra'
 
     return options
 
 
 def resolverevlogstorevfsoptions(ui, requirements, features):
     """Resolve opener options specific to revlogs."""
 
     options = {}
     options[b'flagprocessors'] = {}
 
     if requirementsmod.REVLOGV1_REQUIREMENT in requirements:
         options[b'revlogv1'] = True
     if requirementsmod.REVLOGV2_REQUIREMENT in requirements:
         options[b'revlogv2'] = True
     if requirementsmod.CHANGELOGV2_REQUIREMENT in requirements:
         options[b'changelogv2'] = True
         cmp_rank = ui.configbool(b'experimental', b'changelog-v2.compute-rank')
         options[b'changelogv2.compute-rank'] = cmp_rank
 
     if requirementsmod.GENERALDELTA_REQUIREMENT in requirements:
         options[b'generaldelta'] = True
 
     # experimental config: format.chunkcachesize
     chunkcachesize = ui.configint(b'format', b'chunkcachesize')
     if chunkcachesize is not None:
         options[b'chunkcachesize'] = chunkcachesize
 
     deltabothparents = ui.configbool(
         b'storage', b'revlog.optimize-delta-parent-choice'
     )
     options[b'deltabothparents'] = deltabothparents
     dps_cgds = ui.configint(
         b'storage',
         b'revlog.delta-parent-search.candidate-group-chunk-size',
     )
     options[b'delta-parent-search.candidate-group-chunk-size'] = dps_cgds
     options[b'debug-delta'] = ui.configbool(b'debug', b'revlog.debug-delta')
 
     issue6528 = ui.configbool(b'storage', b'revlog.issue6528.fix-incoming')
     options[b'issue6528.fix-incoming'] = issue6528
 
     lazydelta = ui.configbool(b'storage', b'revlog.reuse-external-delta')
     lazydeltabase = False
     if lazydelta:
         lazydeltabase = ui.configbool(
             b'storage', b'revlog.reuse-external-delta-parent'
         )
     if lazydeltabase is None:
         lazydeltabase = not scmutil.gddeltaconfig(ui)
     options[b'lazydelta'] = lazydelta
     options[b'lazydeltabase'] = lazydeltabase
 
     chainspan = ui.configbytes(b'experimental', b'maxdeltachainspan')
     if 0 <= chainspan:
         options[b'maxdeltachainspan'] = chainspan
 
     mmapindexthreshold = ui.configbytes(b'experimental', b'mmapindexthreshold')
     if mmapindexthreshold is not None:
         options[b'mmapindexthreshold'] = mmapindexthreshold
 
     withsparseread = ui.configbool(b'experimental', b'sparse-read')
     srdensitythres = float(
         ui.config(b'experimental', b'sparse-read.density-threshold')
     )
     srmingapsize = ui.configbytes(b'experimental', b'sparse-read.min-gap-size')
     options[b'with-sparse-read'] = withsparseread
     options[b'sparse-read-density-threshold'] = srdensitythres
     options[b'sparse-read-min-gap-size'] = srmingapsize
 
     sparserevlog = requirementsmod.SPARSEREVLOG_REQUIREMENT in requirements
     options[b'sparse-revlog'] = sparserevlog
     if sparserevlog:
         options[b'generaldelta'] = True
 
     maxchainlen = None
     if sparserevlog:
         maxchainlen = revlogconst.SPARSE_REVLOG_MAX_CHAIN_LENGTH
     # experimental config: format.maxchainlen
     maxchainlen = ui.configint(b'format', b'maxchainlen', maxchainlen)
     if maxchainlen is not None:
         options[b'maxchainlen'] = maxchainlen
 
     for r in requirements:
         # we allow multiple compression engine requirement to co-exist because
         # strickly speaking, revlog seems to support mixed compression style.
         #
         # The compression used for new entries will be "the last one"
         prefix = r.startswith
         if prefix(b'revlog-compression-') or prefix(b'exp-compression-'):
             options[b'compengine'] = r.split(b'-', 2)[2]
 
     options[b'zlib.level'] = ui.configint(b'storage', b'revlog.zlib.level')
     if options[b'zlib.level'] is not None:
         if not (0 <= options[b'zlib.level'] <= 9):
             msg = _(b'invalid value for `storage.revlog.zlib.level` config: %d')
             raise error.Abort(msg % options[b'zlib.level'])
     options[b'zstd.level'] = ui.configint(b'storage', b'revlog.zstd.level')
     if options[b'zstd.level'] is not None:
         if not (0 <= options[b'zstd.level'] <= 22):
             msg = _(b'invalid value for `storage.revlog.zstd.level` config: %d')
             raise error.Abort(msg % options[b'zstd.level'])
 
     if requirementsmod.NARROW_REQUIREMENT in requirements:
         options[b'enableellipsis'] = True
 
     if ui.configbool(b'experimental', b'rust.index'):
         options[b'rust.index'] = True
     if requirementsmod.NODEMAP_REQUIREMENT in requirements:
         slow_path = ui.config(
             b'storage', b'revlog.persistent-nodemap.slow-path'
         )
         if slow_path not in (b'allow', b'warn', b'abort'):
             default = ui.config_default(
                 b'storage', b'revlog.persistent-nodemap.slow-path'
             )
             msg = _(
                 b'unknown value for config '
                 b'"storage.revlog.persistent-nodemap.slow-path": "%s"\n'
             )
             ui.warn(msg % slow_path)
             if not ui.quiet:
                 ui.warn(_(b'falling back to default value: %s\n') % default)
             slow_path = default
 
         msg = _(
             b"accessing `persistent-nodemap` repository without associated "
             b"fast implementation."
         )
         hint = _(
             b"check `hg help config.format.use-persistent-nodemap` "
             b"for details"
         )
         if not revlog.HAS_FAST_PERSISTENT_NODEMAP:
             if slow_path == b'warn':
                 msg = b"warning: " + msg + b'\n'
                 ui.warn(msg)
                 if not ui.quiet:
                     hint = b'(' + hint + b')\n'
                     ui.warn(hint)
             if slow_path == b'abort':
                 raise error.Abort(msg, hint=hint)
         options[b'persistent-nodemap'] = True
     if requirementsmod.DIRSTATE_V2_REQUIREMENT in requirements:
         slow_path = ui.config(b'storage', b'dirstate-v2.slow-path')
         if slow_path not in (b'allow', b'warn', b'abort'):
             default = ui.config_default(b'storage', b'dirstate-v2.slow-path')
             msg = _(b'unknown value for config "dirstate-v2.slow-path": "%s"\n')
             ui.warn(msg % slow_path)
             if not ui.quiet:
                 ui.warn(_(b'falling back to default value: %s\n') % default)
             slow_path = default
 
         msg = _(
             b"accessing `dirstate-v2` repository without associated "
             b"fast implementation."
         )
         hint = _(
             b"check `hg help config.format.use-dirstate-v2` " b"for details"
         )
         if not dirstate.HAS_FAST_DIRSTATE_V2:
             if slow_path == b'warn':
                 msg = b"warning: " + msg + b'\n'
                 ui.warn(msg)
                 if not ui.quiet:
                     hint = b'(' + hint + b')\n'
                     ui.warn(hint)
             if slow_path == b'abort':
                 raise error.Abort(msg, hint=hint)
     if ui.configbool(b'storage', b'revlog.persistent-nodemap.mmap'):
         options[b'persistent-nodemap.mmap'] = True
     if ui.configbool(b'devel', b'persistent-nodemap'):
         options[b'devel-force-nodemap'] = True
 
     return options
 
 
 def makemain(**kwargs):
     """Produce a type conforming to ``ilocalrepositorymain``."""
     return localrepository
 
 
 @interfaceutil.implementer(repository.ilocalrepositoryfilestorage)
 class revlogfilestorage:
     """File storage when using revlogs."""
 
     def file(self, path):
         if path.startswith(b'/'):
             path = path[1:]
 
-        return filelog.filelog(self.svfs, path)
+        try_split = (
+            self.currenttransaction() is not None
+            or txnutil.mayhavepending(self.root)
+        )
+
+        return filelog.filelog(self.svfs, path, try_split=try_split)
 
 
 @interfaceutil.implementer(repository.ilocalrepositoryfilestorage)
 class revlognarrowfilestorage:
     """File storage when using revlogs and narrow files."""
 
     def file(self, path):
         if path.startswith(b'/'):
             path = path[1:]
 
-        return filelog.narrowfilelog(self.svfs, path, self._storenarrowmatch)
+        try_split = (
+            self.currenttransaction() is not None
+            or txnutil.mayhavepending(self.root)
+        )
+        return filelog.narrowfilelog(
+            self.svfs, path, self._storenarrowmatch, try_split=try_split
+        )
 
 
 def makefilestorage(requirements, features, **kwargs):
     """Produce a type conforming to ``ilocalrepositoryfilestorage``."""
     features.add(repository.REPO_FEATURE_REVLOG_FILE_STORAGE)
     features.add(repository.REPO_FEATURE_STREAM_CLONE)
 
     if requirementsmod.NARROW_REQUIREMENT in requirements:
         return revlognarrowfilestorage
     else:
         return revlogfilestorage
 
 
 # List of repository interfaces and factory functions for them. Each
 # will be called in order during ``makelocalrepository()`` to iteratively
 # derive the final type for a local repository instance. We capture the
 # function as a lambda so we don't hold a reference and the module-level
 # functions can be wrapped.
 REPO_INTERFACES = [
     (repository.ilocalrepositorymain, lambda: makemain),
     (repository.ilocalrepositoryfilestorage, lambda: makefilestorage),
 ]
 
 
 @interfaceutil.implementer(repository.ilocalrepositorymain)
 class localrepository:
     """Main class for representing local repositories.
 
     All local repositories are instances of this class.
 
     Constructed on its own, instances of this class are not usable as
     repository objects. To obtain a usable repository object, call
     ``hg.repository()``, ``localrepo.instance()``, or
     ``localrepo.makelocalrepository()``. The latter is the lowest-level.
     ``instance()`` adds support for creating new repositories.
     ``hg.repository()`` adds more extension integration, including calling
     ``reposetup()``. Generally speaking, ``hg.repository()`` should be
     used.
     """
 
     _basesupported = {
         requirementsmod.ARCHIVED_PHASE_REQUIREMENT,
         requirementsmod.BOOKMARKS_IN_STORE_REQUIREMENT,
         requirementsmod.CHANGELOGV2_REQUIREMENT,
         requirementsmod.COPIESSDC_REQUIREMENT,
         requirementsmod.DIRSTATE_TRACKED_HINT_V1,
         requirementsmod.DIRSTATE_V2_REQUIREMENT,
         requirementsmod.DOTENCODE_REQUIREMENT,
         requirementsmod.FNCACHE_REQUIREMENT,
         requirementsmod.GENERALDELTA_REQUIREMENT,
         requirementsmod.INTERNAL_PHASE_REQUIREMENT,
         requirementsmod.NODEMAP_REQUIREMENT,
         requirementsmod.RELATIVE_SHARED_REQUIREMENT,
         requirementsmod.REVLOGV1_REQUIREMENT,
         requirementsmod.REVLOGV2_REQUIREMENT,
         requirementsmod.SHARED_REQUIREMENT,
         requirementsmod.SHARESAFE_REQUIREMENT,
         requirementsmod.SPARSE_REQUIREMENT,
         requirementsmod.SPARSEREVLOG_REQUIREMENT,
         requirementsmod.STORE_REQUIREMENT,
         requirementsmod.TREEMANIFEST_REQUIREMENT,
     }
 
     # list of prefix for file which can be written without 'wlock'
     # Extensions should extend this list when needed
     _wlockfreeprefix = {
         # We migh consider requiring 'wlock' for the next
         # two, but pretty much all the existing code assume
         # wlock is not needed so we keep them excluded for
         # now.
         b'hgrc',
         b'requires',
         # XXX cache is a complicatged business someone
         # should investigate this in depth at some point
         b'cache/',
         # XXX bisect was still a bit too messy at the time
         # this changeset was introduced. Someone should fix
         # the remainig bit and drop this line
         b'bisect.state',
     }
 
     def __init__(
         self,
         baseui,
         ui,
         origroot: bytes,
         wdirvfs: vfsmod.vfs,
         hgvfs: vfsmod.vfs,
         requirements,
         supportedrequirements,
         sharedpath: bytes,
         store,
         cachevfs: vfsmod.vfs,
         wcachevfs: vfsmod.vfs,
         features,
         intents=None,
     ):
         """Create a new local repository instance.
 
         Most callers should use ``hg.repository()``, ``localrepo.instance()``,
         or ``localrepo.makelocalrepository()`` for obtaining a new repository
         object.
 
         Arguments:
 
         baseui
            ``ui.ui`` instance that ``ui`` argument was based off of.
 
         ui
            ``ui.ui`` instance for use by the repository.
 
         origroot
            ``bytes`` path to working directory root of this repository.
 
         wdirvfs
            ``vfs.vfs`` rooted at the working directory.
 
         hgvfs
            ``vfs.vfs`` rooted at .hg/
 
         requirements
            ``set`` of bytestrings representing repository opening requirements.
 
         supportedrequirements
            ``set`` of bytestrings representing repository requirements that we
            know how to open. May be a supetset of ``requirements``.
 
         sharedpath
            ``bytes`` Defining path to storage base directory. Points to a
            ``.hg/`` directory somewhere.
 
         store
            ``store.basicstore`` (or derived) instance providing access to
            versioned storage.
 
         cachevfs
            ``vfs.vfs`` used for cache files.
 
         wcachevfs
            ``vfs.vfs`` used for cache files related to the working copy.
 
         features
            ``set`` of bytestrings defining features/capabilities of this
            instance.
 
         intents
            ``set`` of system strings indicating what this repo will be used
            for.
         """
         self.baseui = baseui
         self.ui = ui
         self.origroot = origroot
         # vfs rooted at working directory.
         self.wvfs = wdirvfs
         self.root = wdirvfs.base
         # vfs rooted at .hg/. Used to access most non-store paths.
         self.vfs = hgvfs
         self.path = hgvfs.base
         self.requirements = requirements
         self.nodeconstants = sha1nodeconstants
         self.nullid = self.nodeconstants.nullid
         self.supported = supportedrequirements
         self.sharedpath = sharedpath
         self.store = store
         self.cachevfs = cachevfs
         self.wcachevfs = wcachevfs
         self.features = features
 
         self.filtername = None
 
         if self.ui.configbool(b'devel', b'all-warnings') or self.ui.configbool(
             b'devel', b'check-locks'
         ):
             self.vfs.audit = self._getvfsward(self.vfs.audit)
         # A list of callback to shape the phase if no data were found.
         # Callback are in the form: func(repo, roots) --> processed root.
         # This list it to be filled by extension during repo setup
         self._phasedefaults = []
 
         color.setup(self.ui)
 
         self.spath = self.store.path
         self.svfs = self.store.vfs
         self.sjoin = self.store.join
         if self.ui.configbool(b'devel', b'all-warnings') or self.ui.configbool(
             b'devel', b'check-locks'
         ):
             if util.safehasattr(self.svfs, b'vfs'):  # this is filtervfs
                 self.svfs.vfs.audit = self._getsvfsward(self.svfs.vfs.audit)
             else:  # standard vfs
                 self.svfs.audit = self._getsvfsward(self.svfs.audit)
 
         self._dirstatevalidatewarned = False
 
         self._branchcaches = branchmap.BranchMapCache()
         self._revbranchcache = None
         self._filterpats = {}
         self._datafilters = {}
         self._transref = self._lockref = self._wlockref = None
 
         # A cache for various files under .hg/ that tracks file changes,
         # (used by the filecache decorator)
         #
         # Maps a property name to its util.filecacheentry
         self._filecache = {}
 
         # hold sets of revision to be filtered
         # should be cleared when something might have changed the filter value:
         # - new changesets,
         # - phase change,
         # - new obsolescence marker,
         # - working directory parent change,
         # - bookmark changes
         self.filteredrevcache = {}
 
         self._dirstate = None
         # post-dirstate-status hooks
         self._postdsstatus = []
 
         self._pending_narrow_pats = None
         self._pending_narrow_pats_dirstate = None
 
         # generic mapping between names and nodes
         self.names = namespaces.namespaces()
 
         # Key to signature value.
         self._sparsesignaturecache = {}
         # Signature to cached matcher instance.
         self._sparsematchercache = {}
 
         self._extrafilterid = repoview.extrafilter(ui)
 
         self.filecopiesmode = None
         if requirementsmod.COPIESSDC_REQUIREMENT in self.requirements:
             self.filecopiesmode = b'changeset-sidedata'
 
         self._wanted_sidedata = set()
         self._sidedata_computers = {}
         sidedatamod.set_sidedata_spec_for_repo(self)
 
     def _getvfsward(self, origfunc):
         """build a ward for self.vfs"""
         rref = weakref.ref(self)
 
         def checkvfs(path, mode=None):
             ret = origfunc(path, mode=mode)
             repo = rref()
             if (
                 repo is None
                 or not util.safehasattr(repo, b'_wlockref')
                 or not util.safehasattr(repo, b'_lockref')
             ):
                 return
             if mode in (None, b'r', b'rb'):
                 return
             if path.startswith(repo.path):
                 # truncate name relative to the repository (.hg)
                 path = path[len(repo.path) + 1 :]
             if path.startswith(b'cache/'):
                 msg = b'accessing cache with vfs instead of cachevfs: "%s"'
                 repo.ui.develwarn(msg % path, stacklevel=3, config=b"cache-vfs")
             # path prefixes covered by 'lock'
             vfs_path_prefixes = (
                 b'journal.',
                 b'undo.',
                 b'strip-backup/',
                 b'cache/',
             )
             if any(path.startswith(prefix) for prefix in vfs_path_prefixes):
                 if repo._currentlock(repo._lockref) is None:
                     repo.ui.develwarn(
                         b'write with no lock: "%s"' % path,
                         stacklevel=3,
                         config=b'check-locks',
                     )
             elif repo._currentlock(repo._wlockref) is None:
                 # rest of vfs files are covered by 'wlock'
                 #
                 # exclude special files
                 for prefix in self._wlockfreeprefix:
                     if path.startswith(prefix):
                         return
                 repo.ui.develwarn(
                     b'write with no wlock: "%s"' % path,
                     stacklevel=3,
                     config=b'check-locks',
                 )
             return ret
 
         return checkvfs
 
     def _getsvfsward(self, origfunc):
         """build a ward for self.svfs"""
         rref = weakref.ref(self)
 
         def checksvfs(path, mode=None):
             ret = origfunc(path, mode=mode)
             repo = rref()
             if repo is None or not util.safehasattr(repo, b'_lockref'):
                 return
             if mode in (None, b'r', b'rb'):
                 return
             if path.startswith(repo.sharedpath):
                 # truncate name relative to the repository (.hg)
                 path = path[len(repo.sharedpath) + 1 :]
             if repo._currentlock(repo._lockref) is None:
                 repo.ui.develwarn(
                     b'write with no lock: "%s"' % path, stacklevel=4
                 )
             return ret
 
         return checksvfs
 
     @property
     def vfs_map(self):
         return {
             b'': self.svfs,
             b'plain': self.vfs,
             b'store': self.svfs,
         }
 
     def close(self):
         self._writecaches()
 
     def _writecaches(self):
         if self._revbranchcache:
             self._revbranchcache.write()
 
     def _restrictcapabilities(self, caps):
         if self.ui.configbool(b'experimental', b'bundle2-advertise'):
             caps = set(caps)
             capsblob = bundle2.encodecaps(
                 bundle2.getrepocaps(self, role=b'client')
             )
             caps.add(b'bundle2=' + urlreq.quote(capsblob))
         if self.ui.configbool(b'experimental', b'narrow'):
             caps.add(wireprototypes.NARROWCAP)
         return caps
 
     # Don't cache auditor/nofsauditor, or you'll end up with reference cycle:
     # self -> auditor -> self._checknested -> self
 
     @property
     def auditor(self):
         # This is only used by context.workingctx.match in order to
         # detect files in subrepos.
         return pathutil.pathauditor(self.root, callback=self._checknested)
 
     @property
     def nofsauditor(self):
         # This is only used by context.basectx.match in order to detect
         # files in subrepos.
         return pathutil.pathauditor(
             self.root, callback=self._checknested, realfs=False, cached=True
         )
 
     def _checknested(self, path):
         """Determine if path is a legal nested repository."""
         if not path.startswith(self.root):
             return False
         subpath = path[len(self.root) + 1 :]
         normsubpath = util.pconvert(subpath)
 
         # XXX: Checking against the current working copy is wrong in
         # the sense that it can reject things like
         #
         #   $ hg cat -r 10 sub/x.txt
         #
         # if sub/ is no longer a subrepository in the working copy
         # parent revision.
         #
         # However, it can of course also allow things that would have
         # been rejected before, such as the above cat command if sub/
         # is a subrepository now, but was a normal directory before.
         # The old path auditor would have rejected by mistake since it
         # panics when it sees sub/.hg/.
         #
         # All in all, checking against the working copy seems sensible
         # since we want to prevent access to nested repositories on
         # the filesystem *now*.
         ctx = self[None]
         parts = util.splitpath(subpath)
         while parts:
             prefix = b'/'.join(parts)
             if prefix in ctx.substate:
                 if prefix == normsubpath:
                     return True
                 else:
                     sub = ctx.sub(prefix)
                     return sub.checknested(subpath[len(prefix) + 1 :])
             else:
                 parts.pop()
         return False
 
     def peer(self, path=None):
         return localpeer(self, path=path)  # not cached to avoid reference cycle
 
     def unfiltered(self):
         """Return unfiltered version of the repository
 
         Intended to be overwritten by filtered repo."""
         return self
 
     def filtered(self, name, visibilityexceptions=None):
         """Return a filtered version of a repository
 
         The `name` parameter is the identifier of the requested view. This
         will return a repoview object set "exactly" to the specified view.
 
         This function does not apply recursive filtering to a repository. For
         example calling `repo.filtered("served")` will return a repoview using
         the "served" view, regardless of the initial view used by `repo`.
 
         In other word, there is always only one level of `repoview` "filtering".
         """
         if self._extrafilterid is not None and b'%' not in name:
             name = name + b'%' + self._extrafilterid
 
         cls = repoview.newtype(self.unfiltered().__class__)
         return cls(self, name, visibilityexceptions)
 
     @mixedrepostorecache(
         (b'bookmarks', b'plain'),
         (b'bookmarks.current', b'plain'),
         (b'bookmarks', b''),
         (b'00changelog.i', b''),
     )
     def _bookmarks(self):
         # Since the multiple files involved in the transaction cannot be
         # written atomically (with current repository format), there is a race
         # condition here.
         #
         # 1) changelog content A is read
         # 2) outside transaction update changelog to content B
         # 3) outside transaction update bookmark file referring to content B
         # 4) bookmarks file content is read and filtered against changelog-A
         #
         # When this happens, bookmarks against nodes missing from A are dropped.
         #
         # Having this happening during read is not great, but it become worse
         # when this happen during write because the bookmarks to the "unknown"
         # nodes will be dropped for good. However, writes happen within locks.
         # This locking makes it possible to have a race free consistent read.
         # For this purpose data read from disc before locking  are
         # "invalidated" right after the locks are taken. This invalidations are
         # "light", the `filecache` mechanism keep the data in memory and will
         # reuse them if the underlying files did not changed. Not parsing the
         # same data multiple times helps performances.
         #
         # Unfortunately in the case describe above, the files tracked by the
         # bookmarks file cache might not have changed, but the in-memory
         # content is still "wrong" because we used an older changelog content
         # to process the on-disk data. So after locking, the changelog would be
         # refreshed but `_bookmarks` would be preserved.
         # Adding `00changelog.i` to the list of tracked file is not
         # enough, because at the time we build the content for `_bookmarks` in
         # (4), the changelog file has already diverged from the content used
         # for loading `changelog` in (1)
         #
         # To prevent the issue, we force the changelog to be explicitly
         # reloaded while computing `_bookmarks`. The data race can still happen
         # without the lock (with a narrower window), but it would no longer go
         # undetected during the lock time refresh.
         #
         # The new schedule is as follow
         #
         # 1) filecache logic detect that `_bookmarks` needs to be computed
         # 2) cachestat for `bookmarks` and `changelog` are captured (for book)
         # 3) We force `changelog` filecache to be tested
         # 4) cachestat for `changelog` are captured (for changelog)
         # 5) `_bookmarks` is computed and cached
         #
         # The step in (3) ensure we have a changelog at least as recent as the
         # cache stat computed in (1). As a result at locking time:
         #  * if the changelog did not changed since (1) -> we can reuse the data
         #  * otherwise -> the bookmarks get refreshed.
         self._refreshchangelog()
         return bookmarks.bmstore(self)
 
     def _refreshchangelog(self):
         """make sure the in memory changelog match the on-disk one"""
         if 'changelog' in vars(self) and self.currenttransaction() is None:
             del self.changelog
 
     @property
     def _activebookmark(self):
         return self._bookmarks.active
 
     # _phasesets depend on changelog. what we need is to call
     # _phasecache.invalidate() if '00changelog.i' was changed, but it
     # can't be easily expressed in filecache mechanism.
     @storecache(b'phaseroots', b'00changelog.i')
     def _phasecache(self):
         return phases.phasecache(self, self._phasedefaults)
 
     @storecache(b'obsstore')
     def obsstore(self):
         return obsolete.makestore(self.ui, self)
 
     @changelogcache()
     def changelog(repo):
         # load dirstate before changelog to avoid race see issue6303
         repo.dirstate.prefetch_parents()
         return repo.store.changelog(
             txnutil.mayhavepending(repo.root),
             concurrencychecker=revlogchecker.get_checker(repo.ui, b'changelog'),
         )
 
     @manifestlogcache()
     def manifestlog(self):
         return self.store.manifestlog(self, self._storenarrowmatch)
 
     @unfilteredpropertycache
     def dirstate(self):
         if self._dirstate is None:
             self._dirstate = self._makedirstate()
         else:
             self._dirstate.refresh()
         return self._dirstate
 
     def _makedirstate(self):
         """Extension point for wrapping the dirstate per-repo."""
         sparsematchfn = None
         if sparse.use_sparse(self):
             sparsematchfn = lambda: sparse.matcher(self)
         v2_req = requirementsmod.DIRSTATE_V2_REQUIREMENT
         th = requirementsmod.DIRSTATE_TRACKED_HINT_V1
         use_dirstate_v2 = v2_req in self.requirements
         use_tracked_hint = th in self.requirements
 
         return dirstate.dirstate(
             self.vfs,
             self.ui,
             self.root,
             self._dirstatevalidate,
             sparsematchfn,
             self.nodeconstants,
             use_dirstate_v2,
             use_tracked_hint=use_tracked_hint,
         )
 
     def _dirstatevalidate(self, node):
         try:
             self.changelog.rev(node)
             return node
         except error.LookupError:
             if not self._dirstatevalidatewarned:
                 self._dirstatevalidatewarned = True
                 self.ui.warn(
                     _(b"warning: ignoring unknown working parent %s!\n")
                     % short(node)
                 )
             return self.nullid
 
     @storecache(narrowspec.FILENAME)
     def narrowpats(self):
         """matcher patterns for this repository's narrowspec
 
         A tuple of (includes, excludes).
         """
         # the narrow management should probably move into its own object
         val = self._pending_narrow_pats
         if val is None:
             val = narrowspec.load(self)
         return val
 
     @storecache(narrowspec.FILENAME)
     def _storenarrowmatch(self):
         if requirementsmod.NARROW_REQUIREMENT not in self.requirements:
             return matchmod.always()
         include, exclude = self.narrowpats
         return narrowspec.match(self.root, include=include, exclude=exclude)
 
     @storecache(narrowspec.FILENAME)
     def _narrowmatch(self):
         if requirementsmod.NARROW_REQUIREMENT not in self.requirements:
             return matchmod.always()
         narrowspec.checkworkingcopynarrowspec(self)
         include, exclude = self.narrowpats
         return narrowspec.match(self.root, include=include, exclude=exclude)
 
     def narrowmatch(self, match=None, includeexact=False):
         """matcher corresponding the the repo's narrowspec
 
         If `match` is given, then that will be intersected with the narrow
         matcher.
 
         If `includeexact` is True, then any exact matches from `match` will
         be included even if they're outside the narrowspec.
         """
         if match:
             if includeexact and not self._narrowmatch.always():
                 # do not exclude explicitly-specified paths so that they can
                 # be warned later on
                 em = matchmod.exact(match.files())
                 nm = matchmod.unionmatcher([self._narrowmatch, em])
                 return matchmod.intersectmatchers(match, nm)
             return matchmod.intersectmatchers(match, self._narrowmatch)
         return self._narrowmatch
 
     def setnarrowpats(self, newincludes, newexcludes):
         narrowspec.save(self, newincludes, newexcludes)
         self.invalidate(clearfilecache=True)
 
     @unfilteredpropertycache
     def _quick_access_changeid_null(self):
         return {
             b'null': (nullrev, self.nodeconstants.nullid),
             nullrev: (nullrev, self.nodeconstants.nullid),
             self.nullid: (nullrev, self.nullid),
         }
 
     @unfilteredpropertycache
     def _quick_access_changeid_wc(self):
         # also fast path access to the working copy parents
         # however, only do it for filter that ensure wc is visible.
         quick = self._quick_access_changeid_null.copy()
         cl = self.unfiltered().changelog
         for node in self.dirstate.parents():
             if node == self.nullid:
                 continue
             rev = cl.index.get_rev(node)
             if rev is None:
                 # unknown working copy parent case:
                 #
                 #   skip the fast path and let higher code deal with it
                 continue
             pair = (rev, node)
             quick[rev] = pair
             quick[node] = pair
             # also add the parents of the parents
             for r in cl.parentrevs(rev):
                 if r == nullrev:
                     continue
                 n = cl.node(r)
                 pair = (r, n)
                 quick[r] = pair
                 quick[n] = pair
         p1node = self.dirstate.p1()
         if p1node != self.nullid:
             quick[b'.'] = quick[p1node]
         return quick
 
     @unfilteredmethod
     def _quick_access_changeid_invalidate(self):
         if '_quick_access_changeid_wc' in vars(self):
             del self.__dict__['_quick_access_changeid_wc']
 
     @property
     def _quick_access_changeid(self):
         """an helper dictionnary for __getitem__ calls
 
         This contains a list of symbol we can recognise right away without
         further processing.
         """
         if self.filtername in repoview.filter_has_wc:
             return self._quick_access_changeid_wc
         return self._quick_access_changeid_null
 
     def __getitem__(self, changeid):
         # dealing with special cases
         if changeid is None:
             return context.workingctx(self)
         if isinstance(changeid, context.basectx):
             return changeid
 
         # dealing with multiple revisions
         if isinstance(changeid, slice):
             # wdirrev isn't contiguous so the slice shouldn't include it
             return [
                 self[i]
                 for i in range(*changeid.indices(len(self)))
                 if i not in self.changelog.filteredrevs
             ]
 
         # dealing with some special values
         quick_access = self._quick_access_changeid.get(changeid)
         if quick_access is not None:
             rev, node = quick_access
             return context.changectx(self, rev, node, maybe_filtered=False)
         if changeid == b'tip':
             node = self.changelog.tip()
             rev = self.changelog.rev(node)
             return context.changectx(self, rev, node)
 
         # dealing with arbitrary values
         try:
             if isinstance(changeid, int):
                 node = self.changelog.node(changeid)
                 rev = changeid
             elif changeid == b'.':
                 # this is a hack to delay/avoid loading obsmarkers
                 # when we know that '.' won't be hidden
                 node = self.dirstate.p1()
                 rev = self.unfiltered().changelog.rev(node)
             elif len(changeid) == self.nodeconstants.nodelen:
                 try:
                     node = changeid
                     rev = self.changelog.rev(changeid)
                 except error.FilteredLookupError:
                     changeid = hex(changeid)  # for the error message
                     raise
                 except LookupError:
                     # check if it might have come from damaged dirstate
                     #
                     # XXX we could avoid the unfiltered if we had a recognizable
                     # exception for filtered changeset access
                     if (
                         self.local()
                         and changeid in self.unfiltered().dirstate.parents()
                     ):
                         msg = _(b"working directory has unknown parent '%s'!")
                         raise error.Abort(msg % short(changeid))
                     changeid = hex(changeid)  # for the error message
                     raise
 
             elif len(changeid) == 2 * self.nodeconstants.nodelen:
                 node = bin(changeid)
                 rev = self.changelog.rev(node)
             else:
                 raise error.ProgrammingError(
                     b"unsupported changeid '%s' of type %s"
                     % (changeid, pycompat.bytestr(type(changeid)))
                 )
 
             return context.changectx(self, rev, node)
 
         except (error.FilteredIndexError, error.FilteredLookupError):
             raise error.FilteredRepoLookupError(
                 _(b"filtered revision '%s'") % pycompat.bytestr(changeid)
             )
         except (IndexError, LookupError):
             raise error.RepoLookupError(
                 _(b"unknown revision '%s'") % pycompat.bytestr(changeid)
             )
         except error.WdirUnsupported:
             return context.workingctx(self)
 
     def __contains__(self, changeid):
         """True if the given changeid exists"""
         try:
             self[changeid]
             return True
         except error.RepoLookupError:
             return False
 
     def __nonzero__(self):
         return True
 
     __bool__ = __nonzero__
 
     def __len__(self):
         # no need to pay the cost of repoview.changelog
         unfi = self.unfiltered()
         return len(unfi.changelog)
 
     def __iter__(self):
         return iter(self.changelog)
 
     def revs(self, expr: bytes, *args):
         """Find revisions matching a revset.
 
         The revset is specified as a string ``expr`` that may contain
         %-formatting to escape certain types. See ``revsetlang.formatspec``.
 
         Revset aliases from the configuration are not expanded. To expand
         user aliases, consider calling ``scmutil.revrange()`` or
         ``repo.anyrevs([expr], user=True)``.
 
         Returns a smartset.abstractsmartset, which is a list-like interface
         that contains integer revisions.
         """
         tree = revsetlang.spectree(expr, *args)
         return revset.makematcher(tree)(self)
 
     def set(self, expr: bytes, *args):
         """Find revisions matching a revset and emit changectx instances.
 
         This is a convenience wrapper around ``revs()`` that iterates the
         result and is a generator of changectx instances.
 
         Revset aliases from the configuration are not expanded. To expand
         user aliases, consider calling ``scmutil.revrange()``.
         """
         for r in self.revs(expr, *args):
             yield self[r]
 
     def anyrevs(self, specs: bytes, user=False, localalias=None):
         """Find revisions matching one of the given revsets.
 
         Revset aliases from the configuration are not expanded by default. To
         expand user aliases, specify ``user=True``. To provide some local
         definitions overriding user aliases, set ``localalias`` to
         ``{name: definitionstring}``.
         """
         if specs == [b'null']:
             return revset.baseset([nullrev])
         if specs == [b'.']:
             quick_data = self._quick_access_changeid.get(b'.')
             if quick_data is not None:
                 return revset.baseset([quick_data[0]])
         if user:
             m = revset.matchany(
                 self.ui,
                 specs,
                 lookup=revset.lookupfn(self),
                 localalias=localalias,
             )
         else:
             m = revset.matchany(None, specs, localalias=localalias)
         return m(self)
 
     def url(self) -> bytes:
         return b'file:' + self.root
 
     def hook(self, name, throw=False, **args):
         """Call a hook, passing this repo instance.
 
         This a convenience method to aid invoking hooks. Extensions likely
         won't call this unless they have registered a custom hook or are
         replacing code that is expected to call a hook.
         """
         return hook.hook(self.ui, self, name, throw, **args)
 
     @filteredpropertycache
     def _tagscache(self):
         """Returns a tagscache object that contains various tags related
         caches."""
 
         # This simplifies its cache management by having one decorated
         # function (this one) and the rest simply fetch things from it.
         class tagscache:
             def __init__(self):
                 # These two define the set of tags for this repository. tags
                 # maps tag name to node; tagtypes maps tag name to 'global' or
                 # 'local'. (Global tags are defined by .hgtags across all
                 # heads, and local tags are defined in .hg/localtags.)
                 # They constitute the in-memory cache of tags.
                 self.tags = self.tagtypes = None
 
                 self.nodetagscache = self.tagslist = None
 
         cache = tagscache()
         cache.tags, cache.tagtypes = self._findtags()
 
         return cache
 
     def tags(self):
         '''return a mapping of tag to node'''
         t = {}
         if self.changelog.filteredrevs:
             tags, tt = self._findtags()
         else:
             tags = self._tagscache.tags
         rev = self.changelog.rev
         for k, v in tags.items():
             try:
                 # ignore tags to unknown nodes
                 rev(v)
                 t[k] = v
             except (error.LookupError, ValueError):
                 pass
         return t
 
     def _findtags(self):
         """Do the hard work of finding tags.  Return a pair of dicts
         (tags, tagtypes) where tags maps tag name to node, and tagtypes
         maps tag name to a string like \'global\' or \'local\'.
         Subclasses or extensions are free to add their own tags, but
         should be aware that the returned dicts will be retained for the
         duration of the localrepo object."""
 
         # XXX what tagtype should subclasses/extensions use?  Currently
         # mq and bookmarks add tags, but do not set the tagtype at all.
         # Should each extension invent its own tag type?  Should there
         # be one tagtype for all such "virtual" tags?  Or is the status
         # quo fine?
 
         # map tag name to (node, hist)
         alltags = tagsmod.findglobaltags(self.ui, self)
         # map tag name to tag type
         tagtypes = {tag: b'global' for tag in alltags}
 
         tagsmod.readlocaltags(self.ui, self, alltags, tagtypes)
 
         # Build the return dicts.  Have to re-encode tag names because
         # the tags module always uses UTF-8 (in order not to lose info
         # writing to the cache), but the rest of Mercurial wants them in
         # local encoding.
         tags = {}
         for name, (node, hist) in alltags.items():
             if node != self.nullid:
                 tags[encoding.tolocal(name)] = node
         tags[b'tip'] = self.changelog.tip()
         tagtypes = {
             encoding.tolocal(name): value for (name, value) in tagtypes.items()
         }
         return (tags, tagtypes)
 
     def tagtype(self, tagname):
         """
         return the type of the given tag. result can be:
 
         'local'  : a local tag
         'global' : a global tag
         None     : tag does not exist
         """
 
         return self._tagscache.tagtypes.get(tagname)
 
     def tagslist(self):
         '''return a list of tags ordered by revision'''
         if not self._tagscache.tagslist:
             l = []
             for t, n in self.tags().items():
                 l.append((self.changelog.rev(n), t, n))
             self._tagscache.tagslist = [(t, n) for r, t, n in sorted(l)]
 
         return self._tagscache.tagslist
 
     def nodetags(self, node):
         '''return the tags associated with a node'''
         if not self._tagscache.nodetagscache:
             nodetagscache = {}
             for t, n in self._tagscache.tags.items():
                 nodetagscache.setdefault(n, []).append(t)
             for tags in nodetagscache.values():
                 tags.sort()
             self._tagscache.nodetagscache = nodetagscache
         return self._tagscache.nodetagscache.get(node, [])
 
     def nodebookmarks(self, node):
         """return the list of bookmarks pointing to the specified node"""
         return self._bookmarks.names(node)
 
     def branchmap(self):
         """returns a dictionary {branch: [branchheads]} with branchheads
         ordered by increasing revision number"""
         return self._branchcaches[self]
 
     @unfilteredmethod
     def revbranchcache(self):
         if not self._revbranchcache:
             self._revbranchcache = branchmap.revbranchcache(self.unfiltered())
         return self._revbranchcache
 
     def register_changeset(self, rev, changelogrevision):
         self.revbranchcache().setdata(rev, changelogrevision)
 
     def branchtip(self, branch, ignoremissing=False):
         """return the tip node for a given branch
 
         If ignoremissing is True, then this method will not raise an error.
         This is helpful for callers that only expect None for a missing branch
         (e.g. namespace).
 
         """
         try:
             return self.branchmap().branchtip(branch)
         except KeyError:
             if not ignoremissing:
                 raise error.RepoLookupError(_(b"unknown branch '%s'") % branch)
             else:
                 pass
 
     def lookup(self, key):
         node = scmutil.revsymbol(self, key).node()
         if node is None:
             raise error.RepoLookupError(_(b"unknown revision '%s'") % key)
         return node
 
     def lookupbranch(self, key):
         if self.branchmap().hasbranch(key):
             return key
 
         return scmutil.revsymbol(self, key).branch()
 
     def known(self, nodes):
         cl = self.changelog
         get_rev = cl.index.get_rev
         filtered = cl.filteredrevs
         result = []
         for n in nodes:
             r = get_rev(n)
             resp = not (r is None or r in filtered)
             result.append(resp)
         return result
 
     def local(self):
         return self
 
     def publishing(self):
         # it's safe (and desirable) to trust the publish flag unconditionally
         # so that we don't finalize changes shared between users via ssh or nfs
         return self.ui.configbool(b'phases', b'publish', untrusted=True)
 
     def cancopy(self):
         # so statichttprepo's override of local() works
         if not self.local():
             return False
         if not self.publishing():
             return True
         # if publishing we can't copy if there is filtered content
         return not self.filtered(b'visible').changelog.filteredrevs
 
     def shared(self):
         '''the type of shared repository (None if not shared)'''
         if self.sharedpath != self.path:
             return b'store'
         return None
 
     def wjoin(self, f: bytes, *insidef: bytes) -> bytes:
         return self.vfs.reljoin(self.root, f, *insidef)
 
     def setparents(self, p1, p2=None):
         if p2 is None:
             p2 = self.nullid
         self[None].setparents(p1, p2)
         self._quick_access_changeid_invalidate()
 
     def filectx(self, path: bytes, changeid=None, fileid=None, changectx=None):
         """changeid must be a changeset revision, if specified.
         fileid can be a file revision or node."""
         return context.filectx(
             self, path, changeid, fileid, changectx=changectx
         )
 
     def getcwd(self) -> bytes:
         return self.dirstate.getcwd()
 
     def pathto(self, f: bytes, cwd: Optional[bytes] = None) -> bytes:
         return self.dirstate.pathto(f, cwd)
 
     def _loadfilter(self, filter):
         if filter not in self._filterpats:
             l = []
             for pat, cmd in self.ui.configitems(filter):
                 if cmd == b'!':
                     continue
                 mf = matchmod.match(self.root, b'', [pat])
                 fn = None
                 params = cmd
                 for name, filterfn in self._datafilters.items():
                     if cmd.startswith(name):
                         fn = filterfn
                         params = cmd[len(name) :].lstrip()
                         break
                 if not fn:
                     fn = lambda s, c, **kwargs: procutil.filter(s, c)
                     fn.__name__ = 'commandfilter'
                 # Wrap old filters not supporting keyword arguments
                 if not pycompat.getargspec(fn)[2]:
                     oldfn = fn
                     fn = lambda s, c, oldfn=oldfn, **kwargs: oldfn(s, c)
                     fn.__name__ = 'compat-' + oldfn.__name__
                 l.append((mf, fn, params))
             self._filterpats[filter] = l
         return self._filterpats[filter]
 
     def _filter(self, filterpats, filename, data):
         for mf, fn, cmd in filterpats:
             if mf(filename):
                 self.ui.debug(
                     b"filtering %s through %s\n"
                     % (filename, cmd or pycompat.sysbytes(fn.__name__))
                 )
                 data = fn(data, cmd, ui=self.ui, repo=self, filename=filename)
                 break
 
         return data
 
     @unfilteredpropertycache
     def _encodefilterpats(self):
         return self._loadfilter(b'encode')
 
     @unfilteredpropertycache
     def _decodefilterpats(self):
         return self._loadfilter(b'decode')
 
     def adddatafilter(self, name, filter):
         self._datafilters[name] = filter
 
     def wread(self, filename: bytes) -> bytes:
         if self.wvfs.islink(filename):
             data = self.wvfs.readlink(filename)
         else:
             data = self.wvfs.read(filename)
         return self._filter(self._encodefilterpats, filename, data)
 
     def wwrite(
         self,
         filename: bytes,
         data: bytes,
         flags: bytes,
         backgroundclose=False,
         **kwargs
     ) -> int:
         """write ``data`` into ``filename`` in the working directory
 
         This returns length of written (maybe decoded) data.
         """
         data = self._filter(self._decodefilterpats, filename, data)
         if b'l' in flags:
             self.wvfs.symlink(data, filename)
         else:
             self.wvfs.write(
                 filename, data, backgroundclose=backgroundclose, **kwargs
             )
             if b'x' in flags:
                 self.wvfs.setflags(filename, False, True)
             else:
                 self.wvfs.setflags(filename, False, False)
         return len(data)
 
     def wwritedata(self, filename: bytes, data: bytes) -> bytes:
         return self._filter(self._decodefilterpats, filename, data)
 
     def currenttransaction(self):
         """return the current transaction or None if non exists"""
         if self._transref:
             tr = self._transref()
         else:
             tr = None
 
         if tr and tr.running():
             return tr
         return None
 
     def transaction(self, desc, report=None):
         if self.ui.configbool(b'devel', b'all-warnings') or self.ui.configbool(
             b'devel', b'check-locks'
         ):
             if self._currentlock(self._lockref) is None:
                 raise error.ProgrammingError(b'transaction requires locking')
         tr = self.currenttransaction()
         if tr is not None:
             return tr.nest(name=desc)
 
         # abort here if the journal already exists
         if self.svfs.exists(b"journal"):
             raise error.RepoError(
                 _(b"abandoned transaction found"),
                 hint=_(b"run 'hg recover' to clean up transaction"),
             )
 
         # At that point your dirstate should be clean:
         #
         # - If you don't have the wlock, why would you still have a dirty
         #   dirstate ?
         #
         # - If you hold the wlock, you should not be opening a transaction in
         #   the middle of a `distate.changing_*` block. The transaction needs to
         #   be open before that and wrap the change-context.
         #
         # - If you are not within a `dirstate.changing_*` context, why is our
         #   dirstate dirty?
         if self.dirstate._dirty:
             m = "cannot open a transaction with a dirty dirstate"
             raise error.ProgrammingError(m)
 
         idbase = b"%.40f#%f" % (random.random(), time.time())
         ha = hex(hashutil.sha1(idbase).digest())
         txnid = b'TXN:' + ha
         self.hook(b'pretxnopen', throw=True, txnname=desc, txnid=txnid)
 
         self._writejournal(desc)
         if report:
             rp = report
         else:
             rp = self.ui.warn
         vfsmap = self.vfs_map
         # we must avoid cyclic reference between repo and transaction.
         reporef = weakref.ref(self)
         # Code to track tag movement
         #
         # Since tags are all handled as file content, it is actually quite hard
         # to track these movement from a code perspective. So we fallback to a
         # tracking at the repository level. One could envision to track changes
         # to the '.hgtags' file through changegroup apply but that fails to
         # cope with case where transaction expose new heads without changegroup
         # being involved (eg: phase movement).
         #
         # For now, We gate the feature behind a flag since this likely comes
         # with performance impacts. The current code run more often than needed
         # and do not use caches as much as it could.  The current focus is on
         # the behavior of the feature so we disable it by default. The flag
         # will be removed when we are happy with the performance impact.
         #
         # Once this feature is no longer experimental move the following
         # documentation to the appropriate help section:
         #
         # The ``HG_TAG_MOVED`` variable will be set if the transaction touched
         # tags (new or changed or deleted tags). In addition the details of
         # these changes are made available in a file at:
         #     ``REPOROOT/.hg/changes/tags.changes``.
         # Make sure you check for HG_TAG_MOVED before reading that file as it
         # might exist from a previous transaction even if no tag were touched
         # in this one. Changes are recorded in a line base format::
         #
         #     <action> <hex-node> <tag-name>\n
         #
         # Actions are defined as follow:
         #   "-R": tag is removed,
         #   "+A": tag is added,
         #   "-M": tag is moved (old value),
         #   "+M": tag is moved (new value),
         tracktags = lambda x: None
         # experimental config: experimental.hook-track-tags
         shouldtracktags = self.ui.configbool(
             b'experimental', b'hook-track-tags'
         )
         if desc != b'strip' and shouldtracktags:
             oldheads = self.changelog.headrevs()
 
             def tracktags(tr2):
                 repo = reporef()
                 assert repo is not None  # help pytype
                 oldfnodes = tagsmod.fnoderevs(repo.ui, repo, oldheads)
                 newheads = repo.changelog.headrevs()
                 newfnodes = tagsmod.fnoderevs(repo.ui, repo, newheads)
                 # notes: we compare lists here.
                 # As we do it only once buiding set would not be cheaper
                 changes = tagsmod.difftags(repo.ui, repo, oldfnodes, newfnodes)
                 if changes:
                     tr2.hookargs[b'tag_moved'] = b'1'
                     with repo.vfs(
                         b'changes/tags.changes', b'w', atomictemp=True
                     ) as changesfile:
                         # note: we do not register the file to the transaction
                         # because we needs it to still exist on the transaction
                         # is close (for txnclose hooks)
                         tagsmod.writediff(changesfile, changes)
 
         def validate(tr2):
             """will run pre-closing hooks"""
             # XXX the transaction API is a bit lacking here so we take a hacky
             # path for now
             #
             # We cannot add this as a "pending" hooks since the 'tr.hookargs'
             # dict is copied before these run. In addition we needs the data
             # available to in memory hooks too.
             #
             # Moreover, we also need to make sure this runs before txnclose
             # hooks and there is no "pending" mechanism that would execute
             # logic only if hooks are about to run.
             #
             # Fixing this limitation of the transaction is also needed to track
             # other families of changes (bookmarks, phases, obsolescence).
             #
             # This will have to be fixed before we remove the experimental
             # gating.
             tracktags(tr2)
             repo = reporef()
             assert repo is not None  # help pytype
 
             singleheadopt = (b'experimental', b'single-head-per-branch')
             singlehead = repo.ui.configbool(*singleheadopt)
             if singlehead:
                 singleheadsub = repo.ui.configsuboptions(*singleheadopt)[1]
                 accountclosed = singleheadsub.get(
                     b"account-closed-heads", False
                 )
                 if singleheadsub.get(b"public-changes-only", False):
                     filtername = b"immutable"
                 else:
                     filtername = b"visible"
                 scmutil.enforcesinglehead(
                     repo, tr2, desc, accountclosed, filtername
                 )
             if hook.hashook(repo.ui, b'pretxnclose-bookmark'):
                 for name, (old, new) in sorted(
                     tr.changes[b'bookmarks'].items()
                 ):
                     args = tr.hookargs.copy()
                     args.update(bookmarks.preparehookargs(name, old, new))
                     repo.hook(
                         b'pretxnclose-bookmark',
                         throw=True,
                         **pycompat.strkwargs(args)
                     )
             if hook.hashook(repo.ui, b'pretxnclose-phase'):
                 cl = repo.unfiltered().changelog
                 for revs, (old, new) in tr.changes[b'phases']:
                     for rev in revs:
                         args = tr.hookargs.copy()
                         node = hex(cl.node(rev))
                         args.update(phases.preparehookargs(node, old, new))
                         repo.hook(
                             b'pretxnclose-phase',
                             throw=True,
                             **pycompat.strkwargs(args)
                         )
 
             repo.hook(
                 b'pretxnclose', throw=True, **pycompat.strkwargs(tr.hookargs)
             )
 
         def releasefn(tr, success):
             repo = reporef()
             if repo is None:
                 # If the repo has been GC'd (and this release function is being
                 # called from transaction.__del__), there's not much we can do,
                 # so just leave the unfinished transaction there and let the
                 # user run `hg recover`.
                 return
             if success:
                 # this should be explicitly invoked here, because
                 # in-memory changes aren't written out at closing
                 # transaction, if tr.addfilegenerator (via
                 # dirstate.write or so) isn't invoked while
                 # transaction running
                 repo.dirstate.write(None)
             else:
                 # discard all changes (including ones already written
                 # out) in this transaction
                 repo.invalidate(clearfilecache=True)
 
         tr = transaction.transaction(
             rp,
             self.svfs,
             vfsmap,
             b"journal",
             b"undo",
             lambda: None,
             self.store.createmode,
             validator=validate,
             releasefn=releasefn,
             checkambigfiles=_cachedfiles,
             name=desc,
         )
         for vfs_id, path in self._journalfiles():
             tr.add_journal(vfs_id, path)
         tr.changes[b'origrepolen'] = len(self)
         tr.changes[b'obsmarkers'] = set()
         tr.changes[b'phases'] = []
         tr.changes[b'bookmarks'] = {}
 
         tr.hookargs[b'txnid'] = txnid
         tr.hookargs[b'txnname'] = desc
         tr.hookargs[b'changes'] = tr.changes
         # note: writing the fncache only during finalize mean that the file is
         # outdated when running hooks. As fncache is used for streaming clone,
         # this is not expected to break anything that happen during the hooks.
         tr.addfinalize(b'flush-fncache', self.store.write)
 
         def txnclosehook(tr2):
             """To be run if transaction is successful, will schedule a hook run"""
             # Don't reference tr2 in hook() so we don't hold a reference.
             # This reduces memory consumption when there are multiple
             # transactions per lock. This can likely go away if issue5045
             # fixes the function accumulation.
             hookargs = tr2.hookargs
 
             def hookfunc(unused_success):
                 repo = reporef()
                 assert repo is not None  # help pytype
 
                 if hook.hashook(repo.ui, b'txnclose-bookmark'):
                     bmchanges = sorted(tr.changes[b'bookmarks'].items())
                     for name, (old, new) in bmchanges:
                         args = tr.hookargs.copy()
                         args.update(bookmarks.preparehookargs(name, old, new))
                         repo.hook(
                             b'txnclose-bookmark',
                             throw=False,
                             **pycompat.strkwargs(args)
                         )
 
                 if hook.hashook(repo.ui, b'txnclose-phase'):
                     cl = repo.unfiltered().changelog
                     phasemv = sorted(
                         tr.changes[b'phases'], key=lambda r: r[0][0]
                     )
                     for revs, (old, new) in phasemv:
                         for rev in revs:
                             args = tr.hookargs.copy()
                             node = hex(cl.node(rev))
                             args.update(phases.preparehookargs(node, old, new))
                             repo.hook(
                                 b'txnclose-phase',
                                 throw=False,
                                 **pycompat.strkwargs(args)
                             )
 
                 repo.hook(
                     b'txnclose', throw=False, **pycompat.strkwargs(hookargs)
                 )
 
             repo = reporef()
             assert repo is not None  # help pytype
             repo._afterlock(hookfunc)
 
         tr.addfinalize(b'txnclose-hook', txnclosehook)
         # Include a leading "-" to make it happen before the transaction summary
         # reports registered via scmutil.registersummarycallback() whose names
         # are 00-txnreport etc. That way, the caches will be warm when the
         # callbacks run.
         tr.addpostclose(b'-warm-cache', self._buildcacheupdater(tr))
 
         def txnaborthook(tr2):
             """To be run if transaction is aborted"""
             repo = reporef()
             assert repo is not None  # help pytype
             repo.hook(
                 b'txnabort', throw=False, **pycompat.strkwargs(tr2.hookargs)
             )
 
         tr.addabort(b'txnabort-hook', txnaborthook)
         # avoid eager cache invalidation. in-memory data should be identical
         # to stored data if transaction has no error.
         tr.addpostclose(b'refresh-filecachestats', self._refreshfilecachestats)
         self._transref = weakref.ref(tr)
         scmutil.registersummarycallback(self, tr, desc)
         # This only exist to deal with the need of rollback to have viable
         # parents at the end of the operation. So backup viable parents at the
         # time of this operation.
         #
         # We only do it when the `wlock` is taken, otherwise other might be
         # altering the dirstate under us.
         #
         # This is really not a great way to do this (first, because we cannot
         # always do it). There are more viable alternative that exists
         #
         # - backing only the working copy parent in a dedicated files and doing
         #   a clean "keep-update" to them on `hg rollback`.
         #
         # - slightly changing the behavior an applying a logic similar to "hg
         # strip" to pick a working copy destination on `hg rollback`
         if self.currentwlock() is not None:
             ds = self.dirstate
             if not self.vfs.exists(b'branch'):
                 # force a file to be written if None exist
                 ds.setbranch(b'default', None)
 
             def backup_dirstate(tr):
                 for f in ds.all_file_names():
                     # hardlink backup is okay because `dirstate` is always
                     # atomically written and possible data file are append only
                     # and resistant to trailing data.
                     tr.addbackup(f, hardlink=True, location=b'plain')
 
             tr.addvalidator(b'dirstate-backup', backup_dirstate)
         return tr
 
     def _journalfiles(self):
         return (
             (self.svfs, b'journal'),
             (self.vfs, b'journal.desc'),
         )
 
     def undofiles(self):
         return [(vfs, undoname(x)) for vfs, x in self._journalfiles()]
 
     @unfilteredmethod
     def _writejournal(self, desc):
         self.vfs.write(b"journal.desc", b"%d\n%s\n" % (len(self), desc))
 
     def recover(self):
         with self.lock():
             if self.svfs.exists(b"journal"):
                 self.ui.status(_(b"rolling back interrupted transaction\n"))
                 vfsmap = self.vfs_map
                 transaction.rollback(
                     self.svfs,
                     vfsmap,
                     b"journal",
                     self.ui.warn,
                     checkambigfiles=_cachedfiles,
                 )
                 self.invalidate()
                 return True
             else:
                 self.ui.warn(_(b"no interrupted transaction available\n"))
                 return False
 
     def rollback(self, dryrun=False, force=False):
         wlock = lock = None
         try:
             wlock = self.wlock()
             lock = self.lock()
             if self.svfs.exists(b"undo"):
                 return self._rollback(dryrun, force)
             else:
                 self.ui.warn(_(b"no rollback information available\n"))
                 return 1
         finally:
             release(lock, wlock)
 
     @unfilteredmethod  # Until we get smarter cache management
     def _rollback(self, dryrun, force):
         ui = self.ui
 
         parents = self.dirstate.parents()
         try:
             args = self.vfs.read(b'undo.desc').splitlines()
             (oldlen, desc, detail) = (int(args[0]), args[1], None)
             if len(args) >= 3:
                 detail = args[2]
             oldtip = oldlen - 1
 
             if detail and ui.verbose:
                 msg = _(
                     b'repository tip rolled back to revision %d'
                     b' (undo %s: %s)\n'
                 ) % (oldtip, desc, detail)
             else:
                 msg = _(
                     b'repository tip rolled back to revision %d (undo %s)\n'
                 ) % (oldtip, desc)
             parentgone = any(self[p].rev() > oldtip for p in parents)
         except IOError:
             msg = _(b'rolling back unknown transaction\n')
             desc = None
             parentgone = True
 
         if not force and self[b'.'] != self[b'tip'] and desc == b'commit':
             raise error.Abort(
                 _(
                     b'rollback of last commit while not checked out '
                     b'may lose data'
                 ),
                 hint=_(b'use -f to force'),
             )
 
         ui.status(msg)
         if dryrun:
             return 0
 
         self.destroying()
         vfsmap = self.vfs_map
         skip_journal_pattern = None
         if not parentgone:
             skip_journal_pattern = RE_SKIP_DIRSTATE_ROLLBACK
         transaction.rollback(
             self.svfs,
             vfsmap,
             b'undo',
             ui.warn,
             checkambigfiles=_cachedfiles,
             skip_journal_pattern=skip_journal_pattern,
         )
         self.invalidate()
         self.dirstate.invalidate()
 
         if parentgone:
             # replace this with some explicit parent update in the future.
             has_node = self.changelog.index.has_node
             if not all(has_node(p) for p in self.dirstate._pl):
                 # There was no dirstate to backup initially, we need to drop
                 # the existing one.
                 with self.dirstate.changing_parents(self):
                     self.dirstate.setparents(self.nullid)
                     self.dirstate.clear()
 
             parents = tuple([p.rev() for p in self[None].parents()])
             if len(parents) > 1:
                 ui.status(
                     _(
                         b'working directory now based on '
                         b'revisions %d and %d\n'
                     )
                     % parents
                 )
             else:
                 ui.status(
                     _(b'working directory now based on revision %d\n') % parents
                 )
             mergestatemod.mergestate.clean(self)
 
         # TODO: if we know which new heads may result from this rollback, pass
         # them to destroy(), which will prevent the branchhead cache from being
         # invalidated.
         self.destroyed()
         return 0
 
     def _buildcacheupdater(self, newtransaction):
         """called during transaction to build the callback updating cache
 
         Lives on the repository to help extension who might want to augment
         this logic. For this purpose, the created transaction is passed to the
         method.
         """
         # we must avoid cyclic reference between repo and transaction.
         reporef = weakref.ref(self)
 
         def updater(tr):
             repo = reporef()
             assert repo is not None  # help pytype
             repo.updatecaches(tr)
 
         return updater
 
     @unfilteredmethod
     def updatecaches(self, tr=None, full=False, caches=None):
         """warm appropriate caches
 
         If this function is called after a transaction closed. The transaction
         will be available in the 'tr' argument. This can be used to selectively
         update caches relevant to the changes in that transaction.
 
         If 'full' is set, make sure all caches the function knows about have
         up-to-date data. Even the ones usually loaded more lazily.
 
         The `full` argument can take a special "post-clone" value. In this case
         the cache warming is made after a clone and of the slower cache might
         be skipped, namely the `.fnodetags` one. This argument is 5.8 specific
         as we plan for a cleaner way to deal with this for 5.9.
         """
         if tr is not None and tr.hookargs.get(b'source') == b'strip':
             # During strip, many caches are invalid but
             # later call to `destroyed` will refresh them.
             return
 
         unfi = self.unfiltered()
 
         if full:
             msg = (
                 "`full` argument for `repo.updatecaches` is deprecated\n"
                 "(use `caches=repository.CACHE_ALL` instead)"
             )
             self.ui.deprecwarn(msg, b"5.9")
             caches = repository.CACHES_ALL
             if full == b"post-clone":
                 caches = repository.CACHES_POST_CLONE
             caches = repository.CACHES_ALL
         elif caches is None:
             caches = repository.CACHES_DEFAULT
 
         if repository.CACHE_BRANCHMAP_SERVED in caches:
             if tr is None or tr.changes[b'origrepolen'] < len(self):
                 # accessing the 'served' branchmap should refresh all the others,
                 self.ui.debug(b'updating the branch cache\n')
                 self.filtered(b'served').branchmap()
                 self.filtered(b'served.hidden').branchmap()
                 # flush all possibly delayed write.
                 self._branchcaches.write_delayed(self)
 
         if repository.CACHE_CHANGELOG_CACHE in caches:
             self.changelog.update_caches(transaction=tr)
 
         if repository.CACHE_MANIFESTLOG_CACHE in caches:
             self.manifestlog.update_caches(transaction=tr)
 
         if repository.CACHE_REV_BRANCH in caches:
             rbc = unfi.revbranchcache()
             for r in unfi.changelog:
                 rbc.branchinfo(r)
             rbc.write()
 
         if repository.CACHE_FULL_MANIFEST in caches:
             # ensure the working copy parents are in the manifestfulltextcache
             for ctx in self[b'.'].parents():
                 ctx.manifest()  # accessing the manifest is enough
 
         if repository.CACHE_FILE_NODE_TAGS in caches:
             # accessing fnode cache warms the cache
             tagsmod.fnoderevs(self.ui, unfi, unfi.changelog.revs())
 
         if repository.CACHE_TAGS_DEFAULT in caches:
             # accessing tags warm the cache
             self.tags()
         if repository.CACHE_TAGS_SERVED in caches:
             self.filtered(b'served').tags()
 
         if repository.CACHE_BRANCHMAP_ALL in caches:
             # The CACHE_BRANCHMAP_ALL updates lazily-loaded caches immediately,
             # so we're forcing a write to cause these caches to be warmed up
             # even if they haven't explicitly been requested yet (if they've
             # never been used by hg, they won't ever have been written, even if
             # they're a subset of another kind of cache that *has* been used).
             for filt in repoview.filtertable.keys():
                 filtered = self.filtered(filt)
                 filtered.branchmap().write(filtered)
 
     def invalidatecaches(self):
         if '_tagscache' in vars(self):
             # can't use delattr on proxy
             del self.__dict__['_tagscache']
 
         self._branchcaches.clear()
         self.invalidatevolatilesets()
         self._sparsesignaturecache.clear()
 
     def invalidatevolatilesets(self):
         self.filteredrevcache.clear()
         obsolete.clearobscaches(self)
         self._quick_access_changeid_invalidate()
 
     def invalidatedirstate(self):
         """Invalidates the dirstate, causing the next call to dirstate
         to check if it was modified since the last time it was read,
         rereading it if it has.
 
         This is different to dirstate.invalidate() that it doesn't always
         rereads the dirstate. Use dirstate.invalidate() if you want to
         explicitly read the dirstate again (i.e. restoring it to a previous
         known good state)."""
         unfi = self.unfiltered()
         if 'dirstate' in unfi.__dict__:
             assert not self.dirstate.is_changing_any
             del unfi.__dict__['dirstate']
 
     def invalidate(self, clearfilecache=False):
         """Invalidates both store and non-store parts other than dirstate
 
         If a transaction is running, invalidation of store is omitted,
         because discarding in-memory changes might cause inconsistency
         (e.g. incomplete fncache causes unintentional failure, but
         redundant one doesn't).
         """
         unfiltered = self.unfiltered()  # all file caches are stored unfiltered
         for k in list(self._filecache.keys()):
             if (
                 k == b'changelog'
                 and self.currenttransaction()
                 and self.changelog._delayed
             ):
                 # The changelog object may store unwritten revisions. We don't
                 # want to lose them.
                 # TODO: Solve the problem instead of working around it.
                 continue
 
             if clearfilecache:
                 del self._filecache[k]
             try:
                 delattr(unfiltered, k)
             except AttributeError:
                 pass
         self.invalidatecaches()
         if not self.currenttransaction():
             # TODO: Changing contents of store outside transaction
             # causes inconsistency. We should make in-memory store
             # changes detectable, and abort if changed.
             self.store.invalidatecaches()
 
     def invalidateall(self):
         """Fully invalidates both store and non-store parts, causing the
         subsequent operation to reread any outside changes."""
         # extension should hook this to invalidate its caches
         self.invalidate()
         self.invalidatedirstate()
 
     @unfilteredmethod
     def _refreshfilecachestats(self, tr):
         """Reload stats of cached files so that they are flagged as valid"""
         for k, ce in self._filecache.items():
             k = pycompat.sysstr(k)
             if k == 'dirstate' or k not in self.__dict__:
                 continue
             ce.refresh()
 
     def _lock(
         self,
         vfs,
         lockname,
         wait,
         releasefn,
         acquirefn,
         desc,
     ):
         timeout = 0
         warntimeout = 0
         if wait:
             timeout = self.ui.configint(b"ui", b"timeout")
             warntimeout = self.ui.configint(b"ui", b"timeout.warn")
         # internal config: ui.signal-safe-lock
         signalsafe = self.ui.configbool(b'ui', b'signal-safe-lock')
 
         l = lockmod.trylock(
             self.ui,
             vfs,
             lockname,
             timeout,
             warntimeout,
             releasefn=releasefn,
             acquirefn=acquirefn,
             desc=desc,
             signalsafe=signalsafe,
         )
         return l
 
     def _afterlock(self, callback):
         """add a callback to be run when the repository is fully unlocked
 
         The callback will be executed when the outermost lock is released
         (with wlock being higher level than 'lock')."""
         for ref in (self._wlockref, self._lockref):
             l = ref and ref()
             if l and l.held:
                 l.postrelease.append(callback)
                 break
         else:  # no lock have been found.
             callback(True)
 
     def lock(self, wait=True):
         """Lock the repository store (.hg/store) and return a weak reference
         to the lock. Use this before modifying the store (e.g. committing or
         stripping). If you are opening a transaction, get a lock as well.)
 
         If both 'lock' and 'wlock' must be acquired, ensure you always acquires
         'wlock' first to avoid a dead-lock hazard."""
         l = self._currentlock(self._lockref)
         if l is not None:
             l.lock()
             return l
 
         l = self._lock(
             vfs=self.svfs,
             lockname=b"lock",
             wait=wait,
             releasefn=None,
             acquirefn=self.invalidate,
             desc=_(b'repository %s') % self.origroot,
         )
         self._lockref = weakref.ref(l)
         return l
 
     def wlock(self, wait=True):
         """Lock the non-store parts of the repository (everything under
         .hg except .hg/store) and return a weak reference to the lock.
 
         Use this before modifying files in .hg.
 
         If both 'lock' and 'wlock' must be acquired, ensure you always acquires
         'wlock' first to avoid a dead-lock hazard."""
         l = self._wlockref() if self._wlockref else None
         if l is not None and l.held:
             l.lock()
             return l
 
         # We do not need to check for non-waiting lock acquisition.  Such
         # acquisition would not cause dead-lock as they would just fail.
         if wait and (
             self.ui.configbool(b'devel', b'all-warnings')
             or self.ui.configbool(b'devel', b'check-locks')
         ):
             if self._currentlock(self._lockref) is not None:
                 self.ui.develwarn(b'"wlock" acquired after "lock"')
 
         def unlock():
             if self.dirstate.is_changing_any:
                 msg = b"wlock release in the middle of a changing parents"
                 self.ui.develwarn(msg)
                 self.dirstate.invalidate()
             else:
                 if self.dirstate._dirty:
                     msg = b"dirty dirstate on wlock release"
                     self.ui.develwarn(msg)
                 self.dirstate.write(None)
 
             unfi = self.unfiltered()
             if 'dirstate' in unfi.__dict__:
                 del unfi.__dict__['dirstate']
 
         l = self._lock(
             self.vfs,
             b"wlock",
             wait,
             unlock,
             self.invalidatedirstate,
             _(b'working directory of %s') % self.origroot,
         )
         self._wlockref = weakref.ref(l)
         return l
 
     def _currentlock(self, lockref):
         """Returns the lock if it's held, or None if it's not."""
         if lockref is None:
             return None
         l = lockref()
         if l is None or not l.held:
             return None
         return l
 
     def currentwlock(self):
         """Returns the wlock if it's held, or None if it's not."""
         return self._currentlock(self._wlockref)
 
     def checkcommitpatterns(self, wctx, match, status, fail):
         """check for commit arguments that aren't committable"""
         if match.isexact() or match.prefix():
             matched = set(status.modified + status.added + status.removed)
 
             for f in match.files():
                 f = self.dirstate.normalize(f)
                 if f == b'.' or f in matched or f in wctx.substate:
                     continue
                 if f in status.deleted:
                     fail(f, _(b'file not found!'))
                 # Is it a directory that exists or used to exist?
                 if self.wvfs.isdir(f) or wctx.p1().hasdir(f):
                     d = f + b'/'
                     for mf in matched:
                         if mf.startswith(d):
                             break
                     else:
                         fail(f, _(b"no match under directory!"))
                 elif f not in self.dirstate:
                     fail(f, _(b"file not tracked!"))
 
     @unfilteredmethod
     def commit(
         self,
         text=b"",
         user=None,
         date=None,
         match=None,
         force=False,
         editor=None,
         extra=None,
     ):
         """Add a new revision to current repository.
 
         Revision information is gathered from the working directory,
         match can be used to filter the committed files. If editor is
         supplied, it is called to get a commit message.
         """
         if extra is None:
             extra = {}
 
         def fail(f, msg):
             raise error.InputError(b'%s: %s' % (f, msg))
 
         if not match:
             match = matchmod.always()
 
         if not force:
             match.bad = fail
 
         # lock() for recent changelog (see issue4368)
         with self.wlock(), self.lock():
             wctx = self[None]
             merge = len(wctx.parents()) > 1
 
             if not force and merge and not match.always():
                 raise error.Abort(
                     _(
                         b'cannot partially commit a merge '
                         b'(do not specify files or patterns)'
                     )
                 )
 
             status = self.status(match=match, clean=force)
             if force:
                 status.modified.extend(
                     status.clean
                 )  # mq may commit clean files
 
             # check subrepos
             subs, commitsubs, newstate = subrepoutil.precommit(
                 self.ui, wctx, status, match, force=force
             )
 
             # make sure all explicit patterns are matched
             if not force:
                 self.checkcommitpatterns(wctx, match, status, fail)
 
             cctx = context.workingcommitctx(
                 self, status, text, user, date, extra
             )
 
             ms = mergestatemod.mergestate.read(self)
             mergeutil.checkunresolved(ms)
 
             # internal config: ui.allowemptycommit
             if cctx.isempty() and not self.ui.configbool(
                 b'ui', b'allowemptycommit'
             ):
                 self.ui.debug(b'nothing to commit, clearing merge state\n')
                 ms.reset()
                 return None
 
             if merge and cctx.deleted():
                 raise error.Abort(_(b"cannot commit merge with missing files"))
 
             if editor:
                 cctx._text = editor(self, cctx, subs)
             edited = text != cctx._text
 
             # Save commit message in case this transaction gets rolled back
             # (e.g. by a pretxncommit hook).  Leave the content alone on
             # the assumption that the user will use the same editor again.
             msg_path = self.savecommitmessage(cctx._text)
 
             # commit subs and write new state
             if subs:
                 uipathfn = scmutil.getuipathfn(self)
                 for s in sorted(commitsubs):
                     sub = wctx.sub(s)
                     self.ui.status(
                         _(b'committing subrepository %s\n')
                         % uipathfn(subrepoutil.subrelpath(sub))
                     )
                     sr = sub.commit(cctx._text, user, date)
                     newstate[s] = (newstate[s][0], sr)
                 subrepoutil.writestate(self, newstate)
 
             p1, p2 = self.dirstate.parents()
             hookp1, hookp2 = hex(p1), (p2 != self.nullid and hex(p2) or b'')
             try:
                 self.hook(
                     b"precommit", throw=True, parent1=hookp1, parent2=hookp2
                 )
                 with self.transaction(b'commit'):
                     ret = self.commitctx(cctx, True)
                     # update bookmarks, dirstate and mergestate
                     bookmarks.update(self, [p1, p2], ret)
                     cctx.markcommitted(ret)
                     ms.reset()
             except:  # re-raises
                 if edited:
                     self.ui.write(
                         _(b'note: commit message saved in %s\n') % msg_path
                     )
                     self.ui.write(
                         _(
                             b"note: use 'hg commit --logfile "
                             b"%s --edit' to reuse it\n"
                         )
                         % msg_path
                     )
                 raise
 
         def commithook(unused_success):
             # hack for command that use a temporary commit (eg: histedit)
             # temporary commit got stripped before hook release
             if self.changelog.hasnode(ret):
                 self.hook(
                     b"commit", node=hex(ret), parent1=hookp1, parent2=hookp2
                 )
 
         self._afterlock(commithook)
         return ret
 
     @unfilteredmethod
     def commitctx(self, ctx, error=False, origctx=None):
         return commit.commitctx(self, ctx, error=error, origctx=origctx)
 
     @unfilteredmethod
     def destroying(self):
         """Inform the repository that nodes are about to be destroyed.
         Intended for use by strip and rollback, so there's a common
         place for anything that has to be done before destroying history.
 
         This is mostly useful for saving state that is in memory and waiting
         to be flushed when the current lock is released. Because a call to
         destroyed is imminent, the repo will be invalidated causing those
         changes to stay in memory (waiting for the next unlock), or vanish
         completely.
         """
         # When using the same lock to commit and strip, the phasecache is left
         # dirty after committing. Then when we strip, the repo is invalidated,
         # causing those changes to disappear.
         if '_phasecache' in vars(self):
             self._phasecache.write()
 
     @unfilteredmethod
     def destroyed(self):
         """Inform the repository that nodes have been destroyed.
         Intended for use by strip and rollback, so there's a common
         place for anything that has to be done after destroying history.
         """
         # When one tries to:
         # 1) destroy nodes thus calling this method (e.g. strip)
         # 2) use phasecache somewhere (e.g. commit)
         #
         # then 2) will fail because the phasecache contains nodes that were
         # removed. We can either remove phasecache from the filecache,
         # causing it to reload next time it is accessed, or simply filter
         # the removed nodes now and write the updated cache.
         self._phasecache.filterunknown(self)
         self._phasecache.write()
 
         # refresh all repository caches
         self.updatecaches()
 
         # Ensure the persistent tag cache is updated.  Doing it now
         # means that the tag cache only has to worry about destroyed
         # heads immediately after a strip/rollback.  That in turn
         # guarantees that "cachetip == currenttip" (comparing both rev
         # and node) always means no nodes have been added or destroyed.
 
         # XXX this is suboptimal when qrefresh'ing: we strip the current
         # head, refresh the tag cache, then immediately add a new head.
         # But I think doing it this way is necessary for the "instant
         # tag cache retrieval" case to work.
         self.invalidate()
 
     def status(
         self,
         node1=b'.',
         node2=None,
         match=None,
         ignored=False,
         clean=False,
         unknown=False,
         listsubrepos=False,
     ):
         '''a convenience method that calls node1.status(node2)'''
         return self[node1].status(
             node2, match, ignored, clean, unknown, listsubrepos
         )
 
     def addpostdsstatus(self, ps):
         """Add a callback to run within the wlock, at the point at which status
         fixups happen.
 
         On status completion, callback(wctx, status) will be called with the
         wlock held, unless the dirstate has changed from underneath or the wlock
         couldn't be grabbed.
 
         Callbacks should not capture and use a cached copy of the dirstate --
         it might change in the meanwhile. Instead, they should access the
         dirstate via wctx.repo().dirstate.
 
         This list is emptied out after each status run -- extensions should
         make sure it adds to this list each time dirstate.status is called.
         Extensions should also make sure they don't call this for statuses
         that don't involve the dirstate.
         """
 
         # The list is located here for uniqueness reasons -- it is actually
         # managed by the workingctx, but that isn't unique per-repo.
         self._postdsstatus.append(ps)
 
     def postdsstatus(self):
         """Used by workingctx to get the list of post-dirstate-status hooks."""
         return self._postdsstatus
 
     def clearpostdsstatus(self):
         """Used by workingctx to clear post-dirstate-status hooks."""
         del self._postdsstatus[:]
 
     def heads(self, start=None):
         if start is None:
             cl = self.changelog
             headrevs = reversed(cl.headrevs())
             return [cl.node(rev) for rev in headrevs]
 
         heads = self.changelog.heads(start)
         # sort the output in rev descending order
         return sorted(heads, key=self.changelog.rev, reverse=True)
 
     def branchheads(self, branch=None, start=None, closed=False):
         """return a (possibly filtered) list of heads for the given branch
 
         Heads are returned in topological order, from newest to oldest.
         If branch is None, use the dirstate branch.
         If start is not None, return only heads reachable from start.
         If closed is True, return heads that are marked as closed as well.
         """
         if branch is None:
             branch = self[None].branch()
         branches = self.branchmap()
         if not branches.hasbranch(branch):
             return []
         # the cache returns heads ordered lowest to highest
         bheads = list(reversed(branches.branchheads(branch, closed=closed)))
         if start is not None:
             # filter out the heads that cannot be reached from startrev
             fbheads = set(self.changelog.nodesbetween([start], bheads)[2])
             bheads = [h for h in bheads if h in fbheads]
         return bheads
 
     def branches(self, nodes):
         if not nodes:
             nodes = [self.changelog.tip()]
         b = []
         for n in nodes:
             t = n
             while True:
                 p = self.changelog.parents(n)
                 if p[1] != self.nullid or p[0] == self.nullid:
                     b.append((t, n, p[0], p[1]))
                     break
                 n = p[0]
         return b
 
     def between(self, pairs):
         r = []
 
         for top, bottom in pairs:
             n, l, i = top, [], 0
             f = 1
 
             while n != bottom and n != self.nullid:
                 p = self.changelog.parents(n)[0]
                 if i == f:
                     l.append(n)
                     f = f * 2
                 n = p
                 i += 1
 
             r.append(l)
 
         return r
 
     def checkpush(self, pushop):
         """Extensions can override this function if additional checks have
         to be performed before pushing, or call it if they override push
         command.
         """
 
     @unfilteredpropertycache
     def prepushoutgoinghooks(self):
         """Return util.hooks consists of a pushop with repo, remote, outgoing
         methods, which are called before pushing changesets.
         """
         return util.hooks()
 
     def pushkey(self, namespace, key, old, new):
         try:
             tr = self.currenttransaction()
             hookargs = {}
             if tr is not None:
                 hookargs.update(tr.hookargs)
             hookargs = pycompat.strkwargs(hookargs)
             hookargs['namespace'] = namespace
             hookargs['key'] = key
             hookargs['old'] = old
             hookargs['new'] = new
             self.hook(b'prepushkey', throw=True, **hookargs)
         except error.HookAbort as exc:
             self.ui.write_err(_(b"pushkey-abort: %s\n") % exc)
             if exc.hint:
                 self.ui.write_err(_(b"(%s)\n") % exc.hint)
             return False
         self.ui.debug(b'pushing key for "%s:%s"\n' % (namespace, key))
         ret = pushkey.push(self, namespace, key, old, new)
 
         def runhook(unused_success):
             self.hook(
                 b'pushkey',
                 namespace=namespace,
                 key=key,
                 old=old,
                 new=new,
                 ret=ret,
             )
 
         self._afterlock(runhook)
         return ret
 
     def listkeys(self, namespace):
         self.hook(b'prelistkeys', throw=True, namespace=namespace)
         self.ui.debug(b'listing keys for "%s"\n' % namespace)
         values = pushkey.list(self, namespace)
         self.hook(b'listkeys', namespace=namespace, values=values)
         return values
 
     def debugwireargs(self, one, two, three=None, four=None, five=None):
         '''used to test argument passing over the wire'''
         return b"%s %s %s %s %s" % (
             one,
             two,
             pycompat.bytestr(three),
             pycompat.bytestr(four),
             pycompat.bytestr(five),
         )
 
     def savecommitmessage(self, text):
         fp = self.vfs(b'last-message.txt', b'wb')
         try:
             fp.write(text)
         finally:
             fp.close()
         return self.pathto(fp.name[len(self.root) + 1 :])
 
     def register_wanted_sidedata(self, category):
         if repository.REPO_FEATURE_SIDE_DATA not in self.features:
             # Only revlogv2 repos can want sidedata.
             return
         self._wanted_sidedata.add(pycompat.bytestr(category))
 
     def register_sidedata_computer(
         self, kind, category, keys, computer, flags, replace=False
     ):
         if kind not in revlogconst.ALL_KINDS:
             msg = _(b"unexpected revlog kind '%s'.")
             raise error.ProgrammingError(msg % kind)
         category = pycompat.bytestr(category)
         already_registered = category in self._sidedata_computers.get(kind, [])
         if already_registered and not replace:
             msg = _(
                 b"cannot register a sidedata computer twice for category '%s'."
             )
             raise error.ProgrammingError(msg % category)
         if replace and not already_registered:
             msg = _(
                 b"cannot replace a sidedata computer that isn't registered "
                 b"for category '%s'."
             )
             raise error.ProgrammingError(msg % category)
         self._sidedata_computers.setdefault(kind, {})
         self._sidedata_computers[kind][category] = (keys, computer, flags)
 
 
 def undoname(fn: bytes) -> bytes:
     base, name = os.path.split(fn)
     assert name.startswith(b'journal')
     return os.path.join(base, name.replace(b'journal', b'undo', 1))
 
 
 def instance(ui, path: bytes, create, intents=None, createopts=None):
     # prevent cyclic import localrepo -> upgrade -> localrepo
     from . import upgrade
 
     localpath = urlutil.urllocalpath(path)
     if create:
         createrepository(ui, localpath, createopts=createopts)
 
     def repo_maker():
         return makelocalrepository(ui, localpath, intents=intents)
 
     repo = repo_maker()
     repo = upgrade.may_auto_upgrade(repo, repo_maker)
     return repo
 
 
 def islocal(path: bytes) -> bool:
     return True
 
 
 def defaultcreateopts(ui, createopts=None):
     """Populate the default creation options for a repository.
 
     A dictionary of explicitly requested creation options can be passed
     in. Missing keys will be populated.
     """
     createopts = dict(createopts or {})
 
     if b'backend' not in createopts:
         # experimental config: storage.new-repo-backend
         createopts[b'backend'] = ui.config(b'storage', b'new-repo-backend')
 
     return createopts
 
 
 def clone_requirements(ui, createopts, srcrepo):
     """clone the requirements of a local repo for a local clone
 
     The store requirements are unchanged while the working copy requirements
     depends on the configuration
     """
     target_requirements = set()
     if not srcrepo.requirements:
         # this is a legacy revlog "v0" repository, we cannot do anything fancy
         # with it.
         return target_requirements
     createopts = defaultcreateopts(ui, createopts=createopts)
     for r in newreporequirements(ui, createopts):
         if r in requirementsmod.WORKING_DIR_REQUIREMENTS:
             target_requirements.add(r)
 
     for r in srcrepo.requirements:
         if r not in requirementsmod.WORKING_DIR_REQUIREMENTS:
             target_requirements.add(r)
     return target_requirements
 
 
 def newreporequirements(ui, createopts):
     """Determine the set of requirements for a new local repository.
 
     Extensions can wrap this function to specify custom requirements for
     new repositories.
     """
 
     if b'backend' not in createopts:
         raise error.ProgrammingError(
             b'backend key not present in createopts; '
             b'was defaultcreateopts() called?'
         )
 
     if createopts[b'backend'] != b'revlogv1':
         raise error.Abort(
             _(
                 b'unable to determine repository requirements for '
                 b'storage backend: %s'
             )
             % createopts[b'backend']
         )
 
     requirements = {requirementsmod.REVLOGV1_REQUIREMENT}
     if ui.configbool(b'format', b'usestore'):
         requirements.add(requirementsmod.STORE_REQUIREMENT)
         if ui.configbool(b'format', b'usefncache'):
             requirements.add(requirementsmod.FNCACHE_REQUIREMENT)
             if ui.configbool(b'format', b'dotencode'):
                 requirements.add(requirementsmod.DOTENCODE_REQUIREMENT)
 
     compengines = ui.configlist(b'format', b'revlog-compression')
     for compengine in compengines:
         if compengine in util.compengines:
             engine = util.compengines[compengine]
             if engine.available() and engine.revlogheader():
                 break
     else:
         raise error.Abort(
             _(
                 b'compression engines %s defined by '
                 b'format.revlog-compression not available'
             )
             % b', '.join(b'"%s"' % e for e in compengines),
             hint=_(
                 b'run "hg debuginstall" to list available '
                 b'compression engines'
             ),
         )
 
     # zlib is the historical default and doesn't need an explicit requirement.
     if compengine == b'zstd':
         requirements.add(b'revlog-compression-zstd')
     elif compengine != b'zlib':
         requirements.add(b'exp-compression-%s' % compengine)
 
     if scmutil.gdinitconfig(ui):
         requirements.add(requirementsmod.GENERALDELTA_REQUIREMENT)
         if ui.configbool(b'format', b'sparse-revlog'):
             requirements.add(requirementsmod.SPARSEREVLOG_REQUIREMENT)
 
     # experimental config: format.use-dirstate-v2
     # Keep this logic in sync with `has_dirstate_v2()` in `tests/hghave.py`
     if ui.configbool(b'format', b'use-dirstate-v2'):
         requirements.add(requirementsmod.DIRSTATE_V2_REQUIREMENT)
 
     # experimental config: format.exp-use-copies-side-data-changeset
     if ui.configbool(b'format', b'exp-use-copies-side-data-changeset'):
         requirements.add(requirementsmod.CHANGELOGV2_REQUIREMENT)
         requirements.add(requirementsmod.COPIESSDC_REQUIREMENT)
     if ui.configbool(b'experimental', b'treemanifest'):
         requirements.add(requirementsmod.TREEMANIFEST_REQUIREMENT)
 
     changelogv2 = ui.config(b'format', b'exp-use-changelog-v2')
     if changelogv2 == b'enable-unstable-format-and-corrupt-my-data':
         requirements.add(requirementsmod.CHANGELOGV2_REQUIREMENT)
 
     revlogv2 = ui.config(b'experimental', b'revlogv2')
     if revlogv2 == b'enable-unstable-format-and-corrupt-my-data':
         requirements.discard(requirementsmod.REVLOGV1_REQUIREMENT)
         requirements.add(requirementsmod.REVLOGV2_REQUIREMENT)
     # experimental config: format.internal-phase
     if ui.configbool(b'format', b'use-internal-phase'):
         requirements.add(requirementsmod.INTERNAL_PHASE_REQUIREMENT)
 
     # experimental config: format.exp-archived-phase
     if ui.configbool(b'format', b'exp-archived-phase'):
         requirements.add(requirementsmod.ARCHIVED_PHASE_REQUIREMENT)
 
     if createopts.get(b'narrowfiles'):
         requirements.add(requirementsmod.NARROW_REQUIREMENT)
 
     if createopts.get(b'lfs'):
         requirements.add(b'lfs')
 
     if ui.configbool(b'format', b'bookmarks-in-store'):
         requirements.add(requirementsmod.BOOKMARKS_IN_STORE_REQUIREMENT)
 
     if ui.configbool(b'format', b'use-persistent-nodemap'):
         requirements.add(requirementsmod.NODEMAP_REQUIREMENT)
 
     # if share-safe is enabled, let's create the new repository with the new
     # requirement
     if ui.configbool(b'format', b'use-share-safe'):
         requirements.add(requirementsmod.SHARESAFE_REQUIREMENT)
 
     # if we are creating a share-repo¹  we have to handle requirement
     # differently.
     #
     # [1] (i.e. reusing the store from another repository, just having a
     # working copy)
     if b'sharedrepo' in createopts:
         source_requirements = set(createopts[b'sharedrepo'].requirements)
 
         if requirementsmod.SHARESAFE_REQUIREMENT not in source_requirements:
             # share to an old school repository, we have to copy the
             # requirements and hope for the best.
             requirements = source_requirements
         else:
             # We have control on the working copy only, so "copy" the non
             # working copy part over, ignoring previous logic.
             to_drop = set()
             for req in requirements:
                 if req in requirementsmod.WORKING_DIR_REQUIREMENTS:
                     continue
                 if req in source_requirements:
                     continue
                 to_drop.add(req)
             requirements -= to_drop
             requirements |= source_requirements
 
         if createopts.get(b'sharedrelative'):
             requirements.add(requirementsmod.RELATIVE_SHARED_REQUIREMENT)
         else:
             requirements.add(requirementsmod.SHARED_REQUIREMENT)
 
     if ui.configbool(b'format', b'use-dirstate-tracked-hint'):
         version = ui.configint(b'format', b'use-dirstate-tracked-hint.version')
         msg = _(b"ignoring unknown tracked key version: %d\n")
         hint = _(
             b"see `hg help config.format.use-dirstate-tracked-hint-version"
         )
         if version != 1:
             ui.warn(msg % version, hint=hint)
         else:
             requirements.add(requirementsmod.DIRSTATE_TRACKED_HINT_V1)
 
     return requirements
 
 
 def checkrequirementscompat(ui, requirements):
     """Checks compatibility of repository requirements enabled and disabled.
 
     Returns a set of requirements which needs to be dropped because dependend
     requirements are not enabled. Also warns users about it"""
 
     dropped = set()
 
     if requirementsmod.STORE_REQUIREMENT not in requirements:
         if requirementsmod.BOOKMARKS_IN_STORE_REQUIREMENT in requirements:
             ui.warn(
                 _(
                     b'ignoring enabled \'format.bookmarks-in-store\' config '
                     b'beacuse it is incompatible with disabled '
                     b'\'format.usestore\' config\n'
                 )
             )
             dropped.add(requirementsmod.BOOKMARKS_IN_STORE_REQUIREMENT)
 
         if (
             requirementsmod.SHARED_REQUIREMENT in requirements
             or requirementsmod.RELATIVE_SHARED_REQUIREMENT in requirements
         ):
             raise error.Abort(
                 _(
                     b"cannot create shared repository as source was created"
                     b" with 'format.usestore' config disabled"
                 )
             )
 
         if requirementsmod.SHARESAFE_REQUIREMENT in requirements:
             if ui.hasconfig(b'format', b'use-share-safe'):
                 msg = _(
                     b"ignoring enabled 'format.use-share-safe' config because "
                     b"it is incompatible with disabled 'format.usestore'"
                     b" config\n"
                 )
                 ui.warn(msg)
             dropped.add(requirementsmod.SHARESAFE_REQUIREMENT)
 
     return dropped
 
 
 def filterknowncreateopts(ui, createopts):
     """Filters a dict of repo creation options against options that are known.
 
     Receives a dict of repo creation options and returns a dict of those
     options that we don't know how to handle.
 
     This function is called as part of repository creation. If the
     returned dict contains any items, repository creation will not
     be allowed, as it means there was a request to create a repository
     with options not recognized by loaded code.
 
     Extensions can wrap this function to filter out creation options
     they know how to handle.
     """
     known = {
         b'backend',
         b'lfs',
         b'narrowfiles',
         b'sharedrepo',
         b'sharedrelative',
         b'shareditems',
         b'shallowfilestore',
     }
 
     return {k: v for k, v in createopts.items() if k not in known}
 
 
 def createrepository(ui, path: bytes, createopts=None, requirements=None):
     """Create a new repository in a vfs.
 
     ``path`` path to the new repo's working directory.
     ``createopts`` options for the new repository.
     ``requirement`` predefined set of requirements.
                     (incompatible with ``createopts``)
 
     The following keys for ``createopts`` are recognized:
 
     backend
        The storage backend to use.
     lfs
        Repository will be created with ``lfs`` requirement. The lfs extension
        will automatically be loaded when the repository is accessed.
     narrowfiles
        Set up repository to support narrow file storage.
     sharedrepo
        Repository object from which storage should be shared.
     sharedrelative
        Boolean indicating if the path to the shared repo should be
        stored as relative. By default, the pointer to the "parent" repo
        is stored as an absolute path.
     shareditems
        Set of items to share to the new repository (in addition to storage).
     shallowfilestore
        Indicates that storage for files should be shallow (not all ancestor
        revisions are known).
     """
 
     if requirements is not None:
         if createopts is not None:
             msg = b'cannot specify both createopts and requirements'
             raise error.ProgrammingError(msg)
         createopts = {}
     else:
         createopts = defaultcreateopts(ui, createopts=createopts)
 
         unknownopts = filterknowncreateopts(ui, createopts)
 
         if not isinstance(unknownopts, dict):
             raise error.ProgrammingError(
                 b'filterknowncreateopts() did not return a dict'
             )
 
         if unknownopts:
             raise error.Abort(
                 _(
                     b'unable to create repository because of unknown '
                     b'creation option: %s'
                 )
                 % b', '.join(sorted(unknownopts)),
                 hint=_(b'is a required extension not loaded?'),
             )
 
         requirements = newreporequirements(ui, createopts=createopts)
         requirements -= checkrequirementscompat(ui, requirements)
 
     wdirvfs = vfsmod.vfs(path, expandpath=True, realpath=True)
 
     hgvfs = vfsmod.vfs(wdirvfs.join(b'.hg'))
     if hgvfs.exists():
         raise error.RepoError(_(b'repository %s already exists') % path)
 
     if b'sharedrepo' in createopts:
         sharedpath = createopts[b'sharedrepo'].sharedpath
 
         if createopts.get(b'sharedrelative'):
             try:
                 sharedpath = os.path.relpath(sharedpath, hgvfs.base)
                 sharedpath = util.pconvert(sharedpath)
             except (IOError, ValueError) as e:
                 # ValueError is raised on Windows if the drive letters differ
                 # on each path.
                 raise error.Abort(
                     _(b'cannot calculate relative path'),
                     hint=stringutil.forcebytestr(e),
                 )
 
     if not wdirvfs.exists():
         wdirvfs.makedirs()
 
     hgvfs.makedir(notindexed=True)
     if b'sharedrepo' not in createopts:
         hgvfs.mkdir(b'cache')
     hgvfs.mkdir(b'wcache')
 
     has_store = requirementsmod.STORE_REQUIREMENT in requirements
     if has_store and b'sharedrepo' not in createopts:
         hgvfs.mkdir(b'store')
 
         # We create an invalid changelog outside the store so very old
         # Mercurial versions (which didn't know about the requirements
         # file) encounter an error on reading the changelog. This
         # effectively locks out old clients and prevents them from
         # mucking with a repo in an unknown format.
         #
         # The revlog header has version 65535, which won't be recognized by
         # such old clients.
         hgvfs.append(
             b'00changelog.i',
             b'\0\0\xFF\xFF dummy changelog to prevent using the old repo '
             b'layout',
         )
 
     # Filter the requirements into working copy and store ones
     wcreq, storereq = scmutil.filterrequirements(requirements)
     # write working copy ones
     scmutil.writerequires(hgvfs, wcreq)
     # If there are store requirements and the current repository
     # is not a shared one, write stored requirements
     # For new shared repository, we don't need to write the store
     # requirements as they are already present in store requires
     if storereq and b'sharedrepo' not in createopts:
         storevfs = vfsmod.vfs(hgvfs.join(b'store'), cacheaudited=True)
         scmutil.writerequires(storevfs, storereq)
 
     # Write out file telling readers where to find the shared store.
     if b'sharedrepo' in createopts:
         hgvfs.write(b'sharedpath', sharedpath)
 
     if createopts.get(b'shareditems'):
         shared = b'\n'.join(sorted(createopts[b'shareditems'])) + b'\n'
         hgvfs.write(b'shared', shared)
 
 
 def poisonrepository(repo):
     """Poison a repository instance so it can no longer be used."""
     # Perform any cleanup on the instance.
     repo.close()
 
     # Our strategy is to replace the type of the object with one that
     # has all attribute lookups result in error.
     #
     # But we have to allow the close() method because some constructors
     # of repos call close() on repo references.
     class poisonedrepository:
         def __getattribute__(self, item):
             if item == 'close':
                 return object.__getattribute__(self, item)
 
             raise error.ProgrammingError(
                 b'repo instances should not be used after unshare'
             )
 
         def close(self):
             pass
 
     # We may have a repoview, which intercepts __setattr__. So be sure
     # we operate at the lowest level possible.
     object.__setattr__(repo, '__class__', poisonedrepository)

mercurial/revlog.py

@@ -1,3385 +1,3399 b''
 # revlog.py - storage back-end for mercurial
 # coding: utf8
 #
 # Copyright 2005-2007 Olivia Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 """Storage back-end for Mercurial.
 
 This provides efficient delta storage with O(1) retrieve and append
 and O(changes) merge between branches.
 """
 
 
 import binascii
 import collections
 import contextlib
 import io
 import os
 import struct
 import zlib
 
 # import stuff from node for others to import from revlog
 from .node import (
     bin,
     hex,
     nullrev,
     sha1nodeconstants,
     short,
     wdirrev,
 )
 from .i18n import _
 from .pycompat import getattr
 from .revlogutils.constants import (
     ALL_KINDS,
     CHANGELOGV2,
     COMP_MODE_DEFAULT,
     COMP_MODE_INLINE,
     COMP_MODE_PLAIN,
     DELTA_BASE_REUSE_NO,
     DELTA_BASE_REUSE_TRY,
     ENTRY_RANK,
     FEATURES_BY_VERSION,
     FLAG_GENERALDELTA,
     FLAG_INLINE_DATA,
     INDEX_HEADER,
     KIND_CHANGELOG,
     KIND_FILELOG,
     RANK_UNKNOWN,
     REVLOGV0,
     REVLOGV1,
     REVLOGV1_FLAGS,
     REVLOGV2,
     REVLOGV2_FLAGS,
     REVLOG_DEFAULT_FLAGS,
     REVLOG_DEFAULT_FORMAT,
     REVLOG_DEFAULT_VERSION,
     SUPPORTED_FLAGS,
 )
 from .revlogutils.flagutil import (
     REVIDX_DEFAULT_FLAGS,
     REVIDX_ELLIPSIS,
     REVIDX_EXTSTORED,
     REVIDX_FLAGS_ORDER,
     REVIDX_HASCOPIESINFO,
     REVIDX_ISCENSORED,
     REVIDX_RAWTEXT_CHANGING_FLAGS,
 )
 from .thirdparty import attr
 from . import (
     ancestor,
     dagop,
     error,
     mdiff,
     policy,
     pycompat,
     revlogutils,
     templatefilters,
     util,
 )
 from .interfaces import (
     repository,
     util as interfaceutil,
 )
 from .revlogutils import (
     deltas as deltautil,
     docket as docketutil,
     flagutil,
     nodemap as nodemaputil,
     randomaccessfile,
     revlogv0,
     rewrite,
     sidedata as sidedatautil,
 )
 from .utils import (
     storageutil,
     stringutil,
 )
 
 # blanked usage of all the name to prevent pyflakes constraints
 # We need these name available in the module for extensions.
 
 REVLOGV0
 REVLOGV1
 REVLOGV2
 CHANGELOGV2
 FLAG_INLINE_DATA
 FLAG_GENERALDELTA
 REVLOG_DEFAULT_FLAGS
 REVLOG_DEFAULT_FORMAT
 REVLOG_DEFAULT_VERSION
 REVLOGV1_FLAGS
 REVLOGV2_FLAGS
 REVIDX_ISCENSORED
 REVIDX_ELLIPSIS
 REVIDX_HASCOPIESINFO
 REVIDX_EXTSTORED
 REVIDX_DEFAULT_FLAGS
 REVIDX_FLAGS_ORDER
 REVIDX_RAWTEXT_CHANGING_FLAGS
 
 parsers = policy.importmod('parsers')
 rustancestor = policy.importrust('ancestor')
 rustdagop = policy.importrust('dagop')
 rustrevlog = policy.importrust('revlog')
 
 # Aliased for performance.
 _zlibdecompress = zlib.decompress
 
 # max size of inline data embedded into a revlog
 _maxinline = 131072
 
 # Flag processors for REVIDX_ELLIPSIS.
 def ellipsisreadprocessor(rl, text):
     return text, False
 
 
 def ellipsiswriteprocessor(rl, text):
     return text, False
 
 
 def ellipsisrawprocessor(rl, text):
     return False
 
 
 ellipsisprocessor = (
     ellipsisreadprocessor,
     ellipsiswriteprocessor,
     ellipsisrawprocessor,
 )
 
 
 def _verify_revision(rl, skipflags, state, node):
     """Verify the integrity of the given revlog ``node`` while providing a hook
     point for extensions to influence the operation."""
     if skipflags:
         state[b'skipread'].add(node)
     else:
         # Side-effect: read content and verify hash.
         rl.revision(node)
 
 
 # True if a fast implementation for persistent-nodemap is available
 #
 # We also consider we have a "fast" implementation in "pure" python because
 # people using pure don't really have performance consideration (and a
 # wheelbarrow of other slowness source)
 HAS_FAST_PERSISTENT_NODEMAP = rustrevlog is not None or util.safehasattr(
     parsers, 'BaseIndexObject'
 )
 
 
 @interfaceutil.implementer(repository.irevisiondelta)
 @attr.s(slots=True)
 class revlogrevisiondelta:
     node = attr.ib()
     p1node = attr.ib()
     p2node = attr.ib()
     basenode = attr.ib()
     flags = attr.ib()
     baserevisionsize = attr.ib()
     revision = attr.ib()
     delta = attr.ib()
     sidedata = attr.ib()
     protocol_flags = attr.ib()
     linknode = attr.ib(default=None)
 
 
 @interfaceutil.implementer(repository.iverifyproblem)
 @attr.s(frozen=True)
 class revlogproblem:
     warning = attr.ib(default=None)
     error = attr.ib(default=None)
     node = attr.ib(default=None)
 
 
 def parse_index_v1(data, inline):
     # call the C implementation to parse the index data
     index, cache = parsers.parse_index2(data, inline)
     return index, cache
 
 
 def parse_index_v2(data, inline):
     # call the C implementation to parse the index data
     index, cache = parsers.parse_index2(data, inline, format=REVLOGV2)
     return index, cache
 
 
 def parse_index_cl_v2(data, inline):
     # call the C implementation to parse the index data
     index, cache = parsers.parse_index2(data, inline, format=CHANGELOGV2)
     return index, cache
 
 
 if util.safehasattr(parsers, 'parse_index_devel_nodemap'):
 
     def parse_index_v1_nodemap(data, inline):
         index, cache = parsers.parse_index_devel_nodemap(data, inline)
         return index, cache
 
 
 else:
     parse_index_v1_nodemap = None
 
 
 def parse_index_v1_mixed(data, inline):
     index, cache = parse_index_v1(data, inline)
     return rustrevlog.MixedIndex(index), cache
 
 
 # corresponds to uncompressed length of indexformatng (2 gigs, 4-byte
 # signed integer)
 _maxentrysize = 0x7FFFFFFF
 
 FILE_TOO_SHORT_MSG = _(
     b'cannot read from revlog %s;'
     b'  expected %d bytes from offset %d, data size is %d'
 )
 
 hexdigits = b'0123456789abcdefABCDEF'
 
 
 class revlog:
     """
     the underlying revision storage object
 
     A revlog consists of two parts, an index and the revision data.
 
     The index is a file with a fixed record size containing
     information on each revision, including its nodeid (hash), the
     nodeids of its parents, the position and offset of its data within
     the data file, and the revision it's based on. Finally, each entry
     contains a linkrev entry that can serve as a pointer to external
     data.
 
     The revision data itself is a linear collection of data chunks.
     Each chunk represents a revision and is usually represented as a
     delta against the previous chunk. To bound lookup time, runs of
     deltas are limited to about 2 times the length of the original
     version data. This makes retrieval of a version proportional to
     its size, or O(1) relative to the number of revisions.
 
     Both pieces of the revlog are written to in an append-only
     fashion, which means we never need to rewrite a file to insert or
     remove data, and can use some simple techniques to avoid the need
     for locking while reading.
 
     If checkambig, indexfile is opened with checkambig=True at
     writing, to avoid file stat ambiguity.
 
     If mmaplargeindex is True, and an mmapindexthreshold is set, the
     index will be mmapped rather than read if it is larger than the
     configured threshold.
 
     If censorable is True, the revlog can have censored revisions.
 
     If `upperboundcomp` is not None, this is the expected maximal gain from
     compression for the data content.
 
     `concurrencychecker` is an optional function that receives 3 arguments: a
     file handle, a filename, and an expected position. It should check whether
     the current position in the file handle is valid, and log/warn/fail (by
     raising).
 
     See mercurial/revlogutils/contants.py for details about the content of an
     index entry.
     """
 
     _flagserrorclass = error.RevlogError
 
     def __init__(
         self,
         opener,
         target,
         radix,
         postfix=None,  # only exist for `tmpcensored` now
         checkambig=False,
         mmaplargeindex=False,
         censorable=False,
         upperboundcomp=None,
         persistentnodemap=False,
         concurrencychecker=None,
         trypending=False,
+        try_split=False,
         canonical_parent_order=True,
     ):
         """
         create a revlog object
 
         opener is a function that abstracts the file opening operation
         and can be used to implement COW semantics or the like.
 
         `target`: a (KIND, ID) tuple that identify the content stored in
         this revlog. It help the rest of the code to understand what the revlog
         is about without having to resort to heuristic and index filename
         analysis. Note: that this must be reliably be set by normal code, but
         that test, debug, or performance measurement code might not set this to
         accurate value.
         """
         self.upperboundcomp = upperboundcomp
 
         self.radix = radix
 
         self._docket_file = None
         self._indexfile = None
         self._datafile = None
         self._sidedatafile = None
         self._nodemap_file = None
         self.postfix = postfix
         self._trypending = trypending
+        self._try_split = try_split
         self.opener = opener
         if persistentnodemap:
             self._nodemap_file = nodemaputil.get_nodemap_file(self)
 
         assert target[0] in ALL_KINDS
         assert len(target) == 2
         self.target = target
         #  When True, indexfile is opened with checkambig=True at writing, to
         #  avoid file stat ambiguity.
         self._checkambig = checkambig
         self._mmaplargeindex = mmaplargeindex
         self._censorable = censorable
         # 3-tuple of (node, rev, text) for a raw revision.
         self._revisioncache = None
         # Maps rev to chain base rev.
         self._chainbasecache = util.lrucachedict(100)
         # 2-tuple of (offset, data) of raw data from the revlog at an offset.
         self._chunkcache = (0, b'')
         # How much data to read and cache into the raw revlog data cache.
         self._chunkcachesize = 65536
         self._maxchainlen = None
         self._deltabothparents = True
         self._candidate_group_chunk_size = 0
         self._debug_delta = False
         self.index = None
         self._docket = None
         self._nodemap_docket = None
         # Mapping of partial identifiers to full nodes.
         self._pcache = {}
         # Mapping of revision integer to full node.
         self._compengine = b'zlib'
         self._compengineopts = {}
         self._maxdeltachainspan = -1
         self._withsparseread = False
         self._sparserevlog = False
         self.hassidedata = False
         self._srdensitythreshold = 0.50
         self._srmingapsize = 262144
 
         # other optionnals features
 
         # might remove rank configuration once the computation has no impact
         self._compute_rank = False
 
         # Make copy of flag processors so each revlog instance can support
         # custom flags.
         self._flagprocessors = dict(flagutil.flagprocessors)
 
         # 3-tuple of file handles being used for active writing.
         self._writinghandles = None
         # prevent nesting of addgroup
         self._adding_group = None
 
         self._loadindex()
 
         self._concurrencychecker = concurrencychecker
 
         # parent order is supposed to be semantically irrelevant, so we
         # normally resort parents to ensure that the first parent is non-null,
         # if there is a non-null parent at all.
         # filelog abuses the parent order as flag to mark some instances of
         # meta-encoded files, so allow it to disable this behavior.
         self.canonical_parent_order = canonical_parent_order
 
     def _init_opts(self):
         """process options (from above/config) to setup associated default revlog mode
 
         These values might be affected when actually reading on disk information.
 
         The relevant values are returned for use in _loadindex().
 
         * newversionflags:
             version header to use if we need to create a new revlog
 
         * mmapindexthreshold:
             minimal index size for start to use mmap
 
         * force_nodemap:
             force the usage of a "development" version of the nodemap code
         """
         mmapindexthreshold = None
         opts = self.opener.options
 
         if b'changelogv2' in opts and self.revlog_kind == KIND_CHANGELOG:
             new_header = CHANGELOGV2
             self._compute_rank = opts.get(b'changelogv2.compute-rank', True)
         elif b'revlogv2' in opts:
             new_header = REVLOGV2
         elif b'revlogv1' in opts:
             new_header = REVLOGV1 | FLAG_INLINE_DATA
             if b'generaldelta' in opts:
                 new_header |= FLAG_GENERALDELTA
         elif b'revlogv0' in self.opener.options:
             new_header = REVLOGV0
         else:
             new_header = REVLOG_DEFAULT_VERSION
 
         if b'chunkcachesize' in opts:
             self._chunkcachesize = opts[b'chunkcachesize']
         if b'maxchainlen' in opts:
             self._maxchainlen = opts[b'maxchainlen']
         if b'deltabothparents' in opts:
             self._deltabothparents = opts[b'deltabothparents']
         dps_cgds = opts.get(b'delta-parent-search.candidate-group-chunk-size')
         if dps_cgds:
             self._candidate_group_chunk_size = dps_cgds
         self._lazydelta = bool(opts.get(b'lazydelta', True))
         self._lazydeltabase = False
         if self._lazydelta:
             self._lazydeltabase = bool(opts.get(b'lazydeltabase', False))
         if b'debug-delta' in opts:
             self._debug_delta = opts[b'debug-delta']
         if b'compengine' in opts:
             self._compengine = opts[b'compengine']
         if b'zlib.level' in opts:
             self._compengineopts[b'zlib.level'] = opts[b'zlib.level']
         if b'zstd.level' in opts:
             self._compengineopts[b'zstd.level'] = opts[b'zstd.level']
         if b'maxdeltachainspan' in opts:
             self._maxdeltachainspan = opts[b'maxdeltachainspan']
         if self._mmaplargeindex and b'mmapindexthreshold' in opts:
             mmapindexthreshold = opts[b'mmapindexthreshold']
         self._sparserevlog = bool(opts.get(b'sparse-revlog', False))
         withsparseread = bool(opts.get(b'with-sparse-read', False))
         # sparse-revlog forces sparse-read
         self._withsparseread = self._sparserevlog or withsparseread
         if b'sparse-read-density-threshold' in opts:
             self._srdensitythreshold = opts[b'sparse-read-density-threshold']
         if b'sparse-read-min-gap-size' in opts:
             self._srmingapsize = opts[b'sparse-read-min-gap-size']
         if opts.get(b'enableellipsis'):
             self._flagprocessors[REVIDX_ELLIPSIS] = ellipsisprocessor
 
         # revlog v0 doesn't have flag processors
         for flag, processor in opts.get(b'flagprocessors', {}).items():
             flagutil.insertflagprocessor(flag, processor, self._flagprocessors)
 
         if self._chunkcachesize <= 0:
             raise error.RevlogError(
                 _(b'revlog chunk cache size %r is not greater than 0')
                 % self._chunkcachesize
             )
         elif self._chunkcachesize & (self._chunkcachesize - 1):
             raise error.RevlogError(
                 _(b'revlog chunk cache size %r is not a power of 2')
                 % self._chunkcachesize
             )
         force_nodemap = opts.get(b'devel-force-nodemap', False)
         return new_header, mmapindexthreshold, force_nodemap
 
     def _get_data(self, filepath, mmap_threshold, size=None):
         """return a file content with or without mmap
 
         If the file is missing return the empty string"""
         try:
             with self.opener(filepath) as fp:
                 if mmap_threshold is not None:
                     file_size = self.opener.fstat(fp).st_size
                     if file_size >= mmap_threshold:
                         if size is not None:
                             # avoid potentiel mmap crash
                             size = min(file_size, size)
                         # TODO: should .close() to release resources without
                         # relying on Python GC
                         if size is None:
                             return util.buffer(util.mmapread(fp))
                         else:
                             return util.buffer(util.mmapread(fp, size))
                 if size is None:
                     return fp.read()
                 else:
                     return fp.read(size)
         except FileNotFoundError:
             return b''
 
     def _loadindex(self, docket=None):
 
         new_header, mmapindexthreshold, force_nodemap = self._init_opts()
 
         if self.postfix is not None:
             entry_point = b'%s.i.%s' % (self.radix, self.postfix)
         elif self._trypending and self.opener.exists(b'%s.i.a' % self.radix):
             entry_point = b'%s.i.a' % self.radix
+        elif self._try_split and self.opener.exists(b'%s.i.s' % self.radix):
+            entry_point = b'%s.i.s' % self.radix
         else:
             entry_point = b'%s.i' % self.radix
 
         if docket is not None:
             self._docket = docket
             self._docket_file = entry_point
         else:
             self._initempty = True
             entry_data = self._get_data(entry_point, mmapindexthreshold)
             if len(entry_data) > 0:
                 header = INDEX_HEADER.unpack(entry_data[:4])[0]
                 self._initempty = False
             else:
                 header = new_header
 
             self._format_flags = header & ~0xFFFF
             self._format_version = header & 0xFFFF
 
             supported_flags = SUPPORTED_FLAGS.get(self._format_version)
             if supported_flags is None:
                 msg = _(b'unknown version (%d) in revlog %s')
                 msg %= (self._format_version, self.display_id)
                 raise error.RevlogError(msg)
             elif self._format_flags & ~supported_flags:
                 msg = _(b'unknown flags (%#04x) in version %d revlog %s')
                 display_flag = self._format_flags >> 16
                 msg %= (display_flag, self._format_version, self.display_id)
                 raise error.RevlogError(msg)
 
             features = FEATURES_BY_VERSION[self._format_version]
             self._inline = features[b'inline'](self._format_flags)
             self._generaldelta = features[b'generaldelta'](self._format_flags)
             self.hassidedata = features[b'sidedata']
 
             if not features[b'docket']:
                 self._indexfile = entry_point
                 index_data = entry_data
             else:
                 self._docket_file = entry_point
                 if self._initempty:
                     self._docket = docketutil.default_docket(self, header)
                 else:
                     self._docket = docketutil.parse_docket(
                         self, entry_data, use_pending=self._trypending
                     )
 
         if self._docket is not None:
             self._indexfile = self._docket.index_filepath()
             index_data = b''
             index_size = self._docket.index_end
             if index_size > 0:
                 index_data = self._get_data(
                     self._indexfile, mmapindexthreshold, size=index_size
                 )
                 if len(index_data) < index_size:
                     msg = _(b'too few index data for %s: got %d, expected %d')
                     msg %= (self.display_id, len(index_data), index_size)
                     raise error.RevlogError(msg)
 
             self._inline = False
             # generaldelta implied by version 2 revlogs.
             self._generaldelta = True
             # the logic for persistent nodemap will be dealt with within the
             # main docket, so disable it for now.
             self._nodemap_file = None
 
         if self._docket is not None:
             self._datafile = self._docket.data_filepath()
             self._sidedatafile = self._docket.sidedata_filepath()
         elif self.postfix is None:
             self._datafile = b'%s.d' % self.radix
         else:
             self._datafile = b'%s.d.%s' % (self.radix, self.postfix)
 
         self.nodeconstants = sha1nodeconstants
         self.nullid = self.nodeconstants.nullid
 
         # sparse-revlog can't be on without general-delta (issue6056)
         if not self._generaldelta:
             self._sparserevlog = False
 
         self._storedeltachains = True
 
         devel_nodemap = (
             self._nodemap_file
             and force_nodemap
             and parse_index_v1_nodemap is not None
         )
 
         use_rust_index = False
         if rustrevlog is not None:
             if self._nodemap_file is not None:
                 use_rust_index = True
             else:
                 use_rust_index = self.opener.options.get(b'rust.index')
 
         self._parse_index = parse_index_v1
         if self._format_version == REVLOGV0:
             self._parse_index = revlogv0.parse_index_v0
         elif self._format_version == REVLOGV2:
             self._parse_index = parse_index_v2
         elif self._format_version == CHANGELOGV2:
             self._parse_index = parse_index_cl_v2
         elif devel_nodemap:
             self._parse_index = parse_index_v1_nodemap
         elif use_rust_index:
             self._parse_index = parse_index_v1_mixed
         try:
             d = self._parse_index(index_data, self._inline)
             index, chunkcache = d
             use_nodemap = (
                 not self._inline
                 and self._nodemap_file is not None
                 and util.safehasattr(index, 'update_nodemap_data')
             )
             if use_nodemap:
                 nodemap_data = nodemaputil.persisted_data(self)
                 if nodemap_data is not None:
                     docket = nodemap_data[0]
                     if (
                         len(d[0]) > docket.tip_rev
                         and d[0][docket.tip_rev][7] == docket.tip_node
                     ):
                         # no changelog tampering
                         self._nodemap_docket = docket
                         index.update_nodemap_data(*nodemap_data)
         except (ValueError, IndexError):
             raise error.RevlogError(
                 _(b"index %s is corrupted") % self.display_id
             )
         self.index = index
         self._segmentfile = randomaccessfile.randomaccessfile(
             self.opener,
             (self._indexfile if self._inline else self._datafile),
             self._chunkcachesize,
             chunkcache,
         )
         self._segmentfile_sidedata = randomaccessfile.randomaccessfile(
             self.opener,
             self._sidedatafile,
             self._chunkcachesize,
         )
         # revnum -> (chain-length, sum-delta-length)
         self._chaininfocache = util.lrucachedict(500)
         # revlog header -> revlog compressor
         self._decompressors = {}
 
     @util.propertycache
     def revlog_kind(self):
         return self.target[0]
 
     @util.propertycache
     def display_id(self):
         """The public facing "ID" of the revlog that we use in message"""
         if self.revlog_kind == KIND_FILELOG:
             # Reference the file without the "data/" prefix, so it is familiar
             # to the user.
             return self.target[1]
         else:
             return self.radix
 
     def _get_decompressor(self, t):
         try:
             compressor = self._decompressors[t]
         except KeyError:
             try:
                 engine = util.compengines.forrevlogheader(t)
                 compressor = engine.revlogcompressor(self._compengineopts)
                 self._decompressors[t] = compressor
             except KeyError:
                 raise error.RevlogError(
                     _(b'unknown compression type %s') % binascii.hexlify(t)
                 )
         return compressor
 
     @util.propertycache
     def _compressor(self):
         engine = util.compengines[self._compengine]
         return engine.revlogcompressor(self._compengineopts)
 
     @util.propertycache
     def _decompressor(self):
         """the default decompressor"""
         if self._docket is None:
             return None
         t = self._docket.default_compression_header
         c = self._get_decompressor(t)
         return c.decompress
 
     def _indexfp(self):
         """file object for the revlog's index file"""
         return self.opener(self._indexfile, mode=b"r")
 
     def __index_write_fp(self):
         # You should not use this directly and use `_writing` instead
         try:
             f = self.opener(
                 self._indexfile, mode=b"r+", checkambig=self._checkambig
             )
             if self._docket is None:
                 f.seek(0, os.SEEK_END)
             else:
                 f.seek(self._docket.index_end, os.SEEK_SET)
             return f
         except FileNotFoundError:
             return self.opener(
                 self._indexfile, mode=b"w+", checkambig=self._checkambig
             )
 
     def __index_new_fp(self):
         # You should not use this unless you are upgrading from inline revlog
         return self.opener(
             self._indexfile,
             mode=b"w",
             checkambig=self._checkambig,
             atomictemp=True,
         )
 
     def _datafp(self, mode=b'r'):
         """file object for the revlog's data file"""
         return self.opener(self._datafile, mode=mode)
 
     @contextlib.contextmanager
     def _sidedatareadfp(self):
         """file object suitable to read sidedata"""
         if self._writinghandles:
             yield self._writinghandles[2]
         else:
             with self.opener(self._sidedatafile) as fp:
                 yield fp
 
     def tiprev(self):
         return len(self.index) - 1
 
     def tip(self):
         return self.node(self.tiprev())
 
     def __contains__(self, rev):
         return 0 <= rev < len(self)
 
     def __len__(self):
         return len(self.index)
 
     def __iter__(self):
         return iter(range(len(self)))
 
     def revs(self, start=0, stop=None):
         """iterate over all rev in this revlog (from start to stop)"""
         return storageutil.iterrevs(len(self), start=start, stop=stop)
 
     def hasnode(self, node):
         try:
             self.rev(node)
             return True
         except KeyError:
             return False
 
     def candelta(self, baserev, rev):
         """whether two revisions (baserev, rev) can be delta-ed or not"""
         # Disable delta if either rev requires a content-changing flag
         # processor (ex. LFS). This is because such flag processor can alter
         # the rawtext content that the delta will be based on, and two clients
         # could have a same revlog node with different flags (i.e. different
         # rawtext contents) and the delta could be incompatible.
         if (self.flags(baserev) & REVIDX_RAWTEXT_CHANGING_FLAGS) or (
             self.flags(rev) & REVIDX_RAWTEXT_CHANGING_FLAGS
         ):
             return False
         return True
 
     def update_caches(self, transaction):
         if self._nodemap_file is not None:
             if transaction is None:
                 nodemaputil.update_persistent_nodemap(self)
             else:
                 nodemaputil.setup_persistent_nodemap(transaction, self)
 
     def clearcaches(self):
         self._revisioncache = None
         self._chainbasecache.clear()
         self._segmentfile.clear_cache()
         self._segmentfile_sidedata.clear_cache()
         self._pcache = {}
         self._nodemap_docket = None
         self.index.clearcaches()
         # The python code is the one responsible for validating the docket, we
         # end up having to refresh it here.
         use_nodemap = (
             not self._inline
             and self._nodemap_file is not None
             and util.safehasattr(self.index, 'update_nodemap_data')
         )
         if use_nodemap:
             nodemap_data = nodemaputil.persisted_data(self)
             if nodemap_data is not None:
                 self._nodemap_docket = nodemap_data[0]
                 self.index.update_nodemap_data(*nodemap_data)
 
     def rev(self, node):
         try:
             return self.index.rev(node)
         except TypeError:
             raise
         except error.RevlogError:
             # parsers.c radix tree lookup failed
             if (
                 node == self.nodeconstants.wdirid
                 or node in self.nodeconstants.wdirfilenodeids
             ):
                 raise error.WdirUnsupported
             raise error.LookupError(node, self.display_id, _(b'no node'))
 
     # Accessors for index entries.
 
     # First tuple entry is 8 bytes. First 6 bytes are offset. Last 2 bytes
     # are flags.
     def start(self, rev):
         return int(self.index[rev][0] >> 16)
 
     def sidedata_cut_off(self, rev):
         sd_cut_off = self.index[rev][8]
         if sd_cut_off != 0:
             return sd_cut_off
         # This is some annoying dance, because entries without sidedata
         # currently use 0 as their ofsset. (instead of previous-offset +
         # previous-size)
         #
         # We should reconsider this sidedata → 0 sidata_offset policy.
         # In the meantime, we need this.
         while 0 <= rev:
             e = self.index[rev]
             if e[9] != 0:
                 return e[8] + e[9]
             rev -= 1
         return 0
 
     def flags(self, rev):
         return self.index[rev][0] & 0xFFFF
 
     def length(self, rev):
         return self.index[rev][1]
 
     def sidedata_length(self, rev):
         if not self.hassidedata:
             return 0
         return self.index[rev][9]
 
     def rawsize(self, rev):
         """return the length of the uncompressed text for a given revision"""
         l = self.index[rev][2]
         if l >= 0:
             return l
 
         t = self.rawdata(rev)
         return len(t)
 
     def size(self, rev):
         """length of non-raw text (processed by a "read" flag processor)"""
         # fast path: if no "read" flag processor could change the content,
         # size is rawsize. note: ELLIPSIS is known to not change the content.
         flags = self.flags(rev)
         if flags & (flagutil.REVIDX_KNOWN_FLAGS ^ REVIDX_ELLIPSIS) == 0:
             return self.rawsize(rev)
 
         return len(self.revision(rev))
 
     def fast_rank(self, rev):
         """Return the rank of a revision if already known, or None otherwise.
 
         The rank of a revision is the size of the sub-graph it defines as a
         head. Equivalently, the rank of a revision `r` is the size of the set
         `ancestors(r)`, `r` included.
 
         This method returns the rank retrieved from the revlog in constant
         time. It makes no attempt at computing unknown values for versions of
         the revlog which do not persist the rank.
         """
         rank = self.index[rev][ENTRY_RANK]
         if self._format_version != CHANGELOGV2 or rank == RANK_UNKNOWN:
             return None
         if rev == nullrev:
             return 0  # convention
         return rank
 
     def chainbase(self, rev):
         base = self._chainbasecache.get(rev)
         if base is not None:
             return base
 
         index = self.index
         iterrev = rev
         base = index[iterrev][3]
         while base != iterrev:
             iterrev = base
             base = index[iterrev][3]
 
         self._chainbasecache[rev] = base
         return base
 
     def linkrev(self, rev):
         return self.index[rev][4]
 
     def parentrevs(self, rev):
         try:
             entry = self.index[rev]
         except IndexError:
             if rev == wdirrev:
                 raise error.WdirUnsupported
             raise
 
         if self.canonical_parent_order and entry[5] == nullrev:
             return entry[6], entry[5]
         else:
             return entry[5], entry[6]
 
     # fast parentrevs(rev) where rev isn't filtered
     _uncheckedparentrevs = parentrevs
 
     def node(self, rev):
         try:
             return self.index[rev][7]
         except IndexError:
             if rev == wdirrev:
                 raise error.WdirUnsupported
             raise
 
     # Derived from index values.
 
     def end(self, rev):
         return self.start(rev) + self.length(rev)
 
     def parents(self, node):
         i = self.index
         d = i[self.rev(node)]
         # inline node() to avoid function call overhead
         if self.canonical_parent_order and d[5] == self.nullid:
             return i[d[6]][7], i[d[5]][7]
         else:
             return i[d[5]][7], i[d[6]][7]
 
     def chainlen(self, rev):
         return self._chaininfo(rev)[0]
 
     def _chaininfo(self, rev):
         chaininfocache = self._chaininfocache
         if rev in chaininfocache:
             return chaininfocache[rev]
         index = self.index
         generaldelta = self._generaldelta
         iterrev = rev
         e = index[iterrev]
         clen = 0
         compresseddeltalen = 0
         while iterrev != e[3]:
             clen += 1
             compresseddeltalen += e[1]
             if generaldelta:
                 iterrev = e[3]
             else:
                 iterrev -= 1
             if iterrev in chaininfocache:
                 t = chaininfocache[iterrev]
                 clen += t[0]
                 compresseddeltalen += t[1]
                 break
             e = index[iterrev]
         else:
             # Add text length of base since decompressing that also takes
             # work. For cache hits the length is already included.
             compresseddeltalen += e[1]
         r = (clen, compresseddeltalen)
         chaininfocache[rev] = r
         return r
 
     def _deltachain(self, rev, stoprev=None):
         """Obtain the delta chain for a revision.
 
         ``stoprev`` specifies a revision to stop at. If not specified, we
         stop at the base of the chain.
 
         Returns a 2-tuple of (chain, stopped) where ``chain`` is a list of
         revs in ascending order and ``stopped`` is a bool indicating whether
         ``stoprev`` was hit.
         """
         # Try C implementation.
         try:
             return self.index.deltachain(rev, stoprev, self._generaldelta)
         except AttributeError:
             pass
 
         chain = []
 
         # Alias to prevent attribute lookup in tight loop.
         index = self.index
         generaldelta = self._generaldelta
 
         iterrev = rev
         e = index[iterrev]
         while iterrev != e[3] and iterrev != stoprev:
             chain.append(iterrev)
             if generaldelta:
                 iterrev = e[3]
             else:
                 iterrev -= 1
             e = index[iterrev]
 
         if iterrev == stoprev:
             stopped = True
         else:
             chain.append(iterrev)
             stopped = False
 
         chain.reverse()
         return chain, stopped
 
     def ancestors(self, revs, stoprev=0, inclusive=False):
         """Generate the ancestors of 'revs' in reverse revision order.
         Does not generate revs lower than stoprev.
 
         See the documentation for ancestor.lazyancestors for more details."""
 
         # first, make sure start revisions aren't filtered
         revs = list(revs)
         checkrev = self.node
         for r in revs:
             checkrev(r)
         # and we're sure ancestors aren't filtered as well
 
         if rustancestor is not None and self.index.rust_ext_compat:
             lazyancestors = rustancestor.LazyAncestors
             arg = self.index
         else:
             lazyancestors = ancestor.lazyancestors
             arg = self._uncheckedparentrevs
         return lazyancestors(arg, revs, stoprev=stoprev, inclusive=inclusive)
 
     def descendants(self, revs):
         return dagop.descendantrevs(revs, self.revs, self.parentrevs)
 
     def findcommonmissing(self, common=None, heads=None):
         """Return a tuple of the ancestors of common and the ancestors of heads
         that are not ancestors of common. In revset terminology, we return the
         tuple:
 
           ::common, (::heads) - (::common)
 
         The list is sorted by revision number, meaning it is
         topologically sorted.
 
         'heads' and 'common' are both lists of node IDs.  If heads is
         not supplied, uses all of the revlog's heads.  If common is not
         supplied, uses nullid."""
         if common is None:
             common = [self.nullid]
         if heads is None:
             heads = self.heads()
 
         common = [self.rev(n) for n in common]
         heads = [self.rev(n) for n in heads]
 
         # we want the ancestors, but inclusive
         class lazyset:
             def __init__(self, lazyvalues):
                 self.addedvalues = set()
                 self.lazyvalues = lazyvalues
 
             def __contains__(self, value):
                 return value in self.addedvalues or value in self.lazyvalues
 
             def __iter__(self):
                 added = self.addedvalues
                 for r in added:
                     yield r
                 for r in self.lazyvalues:
                     if not r in added:
                         yield r
 
             def add(self, value):
                 self.addedvalues.add(value)
 
             def update(self, values):
                 self.addedvalues.update(values)
 
         has = lazyset(self.ancestors(common))
         has.add(nullrev)
         has.update(common)
 
         # take all ancestors from heads that aren't in has
         missing = set()
         visit = collections.deque(r for r in heads if r not in has)
         while visit:
             r = visit.popleft()
             if r in missing:
                 continue
             else:
                 missing.add(r)
                 for p in self.parentrevs(r):
                     if p not in has:
                         visit.append(p)
         missing = list(missing)
         missing.sort()
         return has, [self.node(miss) for miss in missing]
 
     def incrementalmissingrevs(self, common=None):
         """Return an object that can be used to incrementally compute the
         revision numbers of the ancestors of arbitrary sets that are not
         ancestors of common. This is an ancestor.incrementalmissingancestors
         object.
 
         'common' is a list of revision numbers. If common is not supplied, uses
         nullrev.
         """
         if common is None:
             common = [nullrev]
 
         if rustancestor is not None and self.index.rust_ext_compat:
             return rustancestor.MissingAncestors(self.index, common)
         return ancestor.incrementalmissingancestors(self.parentrevs, common)
 
     def findmissingrevs(self, common=None, heads=None):
         """Return the revision numbers of the ancestors of heads that
         are not ancestors of common.
 
         More specifically, return a list of revision numbers corresponding to
         nodes N such that every N satisfies the following constraints:
 
           1. N is an ancestor of some node in 'heads'
           2. N is not an ancestor of any node in 'common'
 
         The list is sorted by revision number, meaning it is
         topologically sorted.
 
         'heads' and 'common' are both lists of revision numbers.  If heads is
         not supplied, uses all of the revlog's heads.  If common is not
         supplied, uses nullid."""
         if common is None:
             common = [nullrev]
         if heads is None:
             heads = self.headrevs()
 
         inc = self.incrementalmissingrevs(common=common)
         return inc.missingancestors(heads)
 
     def findmissing(self, common=None, heads=None):
         """Return the ancestors of heads that are not ancestors of common.
 
         More specifically, return a list of nodes N such that every N
         satisfies the following constraints:
 
           1. N is an ancestor of some node in 'heads'
           2. N is not an ancestor of any node in 'common'
 
         The list is sorted by revision number, meaning it is
         topologically sorted.
 
         'heads' and 'common' are both lists of node IDs.  If heads is
         not supplied, uses all of the revlog's heads.  If common is not
         supplied, uses nullid."""
         if common is None:
             common = [self.nullid]
         if heads is None:
             heads = self.heads()
 
         common = [self.rev(n) for n in common]
         heads = [self.rev(n) for n in heads]
 
         inc = self.incrementalmissingrevs(common=common)
         return [self.node(r) for r in inc.missingancestors(heads)]
 
     def nodesbetween(self, roots=None, heads=None):
         """Return a topological path from 'roots' to 'heads'.
 
         Return a tuple (nodes, outroots, outheads) where 'nodes' is a
         topologically sorted list of all nodes N that satisfy both of
         these constraints:
 
           1. N is a descendant of some node in 'roots'
           2. N is an ancestor of some node in 'heads'
 
         Every node is considered to be both a descendant and an ancestor
         of itself, so every reachable node in 'roots' and 'heads' will be
         included in 'nodes'.
 
         'outroots' is the list of reachable nodes in 'roots', i.e., the
         subset of 'roots' that is returned in 'nodes'.  Likewise,
         'outheads' is the subset of 'heads' that is also in 'nodes'.
 
         'roots' and 'heads' are both lists of node IDs.  If 'roots' is
         unspecified, uses nullid as the only root.  If 'heads' is
         unspecified, uses list of all of the revlog's heads."""
         nonodes = ([], [], [])
         if roots is not None:
             roots = list(roots)
             if not roots:
                 return nonodes
             lowestrev = min([self.rev(n) for n in roots])
         else:
             roots = [self.nullid]  # Everybody's a descendant of nullid
             lowestrev = nullrev
         if (lowestrev == nullrev) and (heads is None):
             # We want _all_ the nodes!
             return (
                 [self.node(r) for r in self],
                 [self.nullid],
                 list(self.heads()),
             )
         if heads is None:
             # All nodes are ancestors, so the latest ancestor is the last
             # node.
             highestrev = len(self) - 1
             # Set ancestors to None to signal that every node is an ancestor.
             ancestors = None
             # Set heads to an empty dictionary for later discovery of heads
             heads = {}
         else:
             heads = list(heads)
             if not heads:
                 return nonodes
             ancestors = set()
             # Turn heads into a dictionary so we can remove 'fake' heads.
             # Also, later we will be using it to filter out the heads we can't
             # find from roots.
             heads = dict.fromkeys(heads, False)
             # Start at the top and keep marking parents until we're done.
             nodestotag = set(heads)
             # Remember where the top was so we can use it as a limit later.
             highestrev = max([self.rev(n) for n in nodestotag])
             while nodestotag:
                 # grab a node to tag
                 n = nodestotag.pop()
                 # Never tag nullid
                 if n == self.nullid:
                     continue
                 # A node's revision number represents its place in a
                 # topologically sorted list of nodes.
                 r = self.rev(n)
                 if r >= lowestrev:
                     if n not in ancestors:
                         # If we are possibly a descendant of one of the roots
                         # and we haven't already been marked as an ancestor
                         ancestors.add(n)  # Mark as ancestor
                         # Add non-nullid parents to list of nodes to tag.
                         nodestotag.update(
                             [p for p in self.parents(n) if p != self.nullid]
                         )
                     elif n in heads:  # We've seen it before, is it a fake head?
                         # So it is, real heads should not be the ancestors of
                         # any other heads.
                         heads.pop(n)
             if not ancestors:
                 return nonodes
             # Now that we have our set of ancestors, we want to remove any
             # roots that are not ancestors.
 
             # If one of the roots was nullid, everything is included anyway.
             if lowestrev > nullrev:
                 # But, since we weren't, let's recompute the lowest rev to not
                 # include roots that aren't ancestors.
 
                 # Filter out roots that aren't ancestors of heads
                 roots = [root for root in roots if root in ancestors]
                 # Recompute the lowest revision
                 if roots:
                     lowestrev = min([self.rev(root) for root in roots])
                 else:
                     # No more roots?  Return empty list
                     return nonodes
             else:
                 # We are descending from nullid, and don't need to care about
                 # any other roots.
                 lowestrev = nullrev
                 roots = [self.nullid]
         # Transform our roots list into a set.
         descendants = set(roots)
         # Also, keep the original roots so we can filter out roots that aren't
         # 'real' roots (i.e. are descended from other roots).
         roots = descendants.copy()
         # Our topologically sorted list of output nodes.
         orderedout = []
         # Don't start at nullid since we don't want nullid in our output list,
         # and if nullid shows up in descendants, empty parents will look like
         # they're descendants.
         for r in self.revs(start=max(lowestrev, 0), stop=highestrev + 1):
             n = self.node(r)
             isdescendant = False
             if lowestrev == nullrev:  # Everybody is a descendant of nullid
                 isdescendant = True
             elif n in descendants:
                 # n is already a descendant
                 isdescendant = True
                 # This check only needs to be done here because all the roots
                 # will start being marked is descendants before the loop.
                 if n in roots:
                     # If n was a root, check if it's a 'real' root.
                     p = tuple(self.parents(n))
                     # If any of its parents are descendants, it's not a root.
                     if (p[0] in descendants) or (p[1] in descendants):
                         roots.remove(n)
             else:
                 p = tuple(self.parents(n))
                 # A node is a descendant if either of its parents are
                 # descendants.  (We seeded the dependents list with the roots
                 # up there, remember?)
                 if (p[0] in descendants) or (p[1] in descendants):
                     descendants.add(n)
                     isdescendant = True
             if isdescendant and ((ancestors is None) or (n in ancestors)):
                 # Only include nodes that are both descendants and ancestors.
                 orderedout.append(n)
                 if (ancestors is not None) and (n in heads):
                     # We're trying to figure out which heads are reachable
                     # from roots.
                     # Mark this head as having been reached
                     heads[n] = True
                 elif ancestors is None:
                     # Otherwise, we're trying to discover the heads.
                     # Assume this is a head because if it isn't, the next step
                     # will eventually remove it.
                     heads[n] = True
                     # But, obviously its parents aren't.
                     for p in self.parents(n):
                         heads.pop(p, None)
         heads = [head for head, flag in heads.items() if flag]
         roots = list(roots)
         assert orderedout
         assert roots
         assert heads
         return (orderedout, roots, heads)
 
     def headrevs(self, revs=None):
         if revs is None:
             try:
                 return self.index.headrevs()
             except AttributeError:
                 return self._headrevs()
         if rustdagop is not None and self.index.rust_ext_compat:
             return rustdagop.headrevs(self.index, revs)
         return dagop.headrevs(revs, self._uncheckedparentrevs)
 
     def computephases(self, roots):
         return self.index.computephasesmapsets(roots)
 
     def _headrevs(self):
         count = len(self)
         if not count:
             return [nullrev]
         # we won't iter over filtered rev so nobody is a head at start
         ishead = [0] * (count + 1)
         index = self.index
         for r in self:
             ishead[r] = 1  # I may be an head
             e = index[r]
             ishead[e[5]] = ishead[e[6]] = 0  # my parent are not
         return [r for r, val in enumerate(ishead) if val]
 
     def heads(self, start=None, stop=None):
         """return the list of all nodes that have no children
 
         if start is specified, only heads that are descendants of
         start will be returned
         if stop is specified, it will consider all the revs from stop
         as if they had no children
         """
         if start is None and stop is None:
             if not len(self):
                 return [self.nullid]
             return [self.node(r) for r in self.headrevs()]
 
         if start is None:
             start = nullrev
         else:
             start = self.rev(start)
 
         stoprevs = {self.rev(n) for n in stop or []}
 
         revs = dagop.headrevssubset(
             self.revs, self.parentrevs, startrev=start, stoprevs=stoprevs
         )
 
         return [self.node(rev) for rev in revs]
 
     def children(self, node):
         """find the children of a given node"""
         c = []
         p = self.rev(node)
         for r in self.revs(start=p + 1):
             prevs = [pr for pr in self.parentrevs(r) if pr != nullrev]
             if prevs:
                 for pr in prevs:
                     if pr == p:
                         c.append(self.node(r))
             elif p == nullrev:
                 c.append(self.node(r))
         return c
 
     def commonancestorsheads(self, a, b):
         """calculate all the heads of the common ancestors of nodes a and b"""
         a, b = self.rev(a), self.rev(b)
         ancs = self._commonancestorsheads(a, b)
         return pycompat.maplist(self.node, ancs)
 
     def _commonancestorsheads(self, *revs):
         """calculate all the heads of the common ancestors of revs"""
         try:
             ancs = self.index.commonancestorsheads(*revs)
         except (AttributeError, OverflowError):  # C implementation failed
             ancs = ancestor.commonancestorsheads(self.parentrevs, *revs)
         return ancs
 
     def isancestor(self, a, b):
         """return True if node a is an ancestor of node b
 
         A revision is considered an ancestor of itself."""
         a, b = self.rev(a), self.rev(b)
         return self.isancestorrev(a, b)
 
     def isancestorrev(self, a, b):
         """return True if revision a is an ancestor of revision b
 
         A revision is considered an ancestor of itself.
 
         The implementation of this is trivial but the use of
         reachableroots is not."""
         if a == nullrev:
             return True
         elif a == b:
             return True
         elif a > b:
             return False
         return bool(self.reachableroots(a, [b], [a], includepath=False))
 
     def reachableroots(self, minroot, heads, roots, includepath=False):
         """return (heads(::(<roots> and <roots>::<heads>)))
 
         If includepath is True, return (<roots>::<heads>)."""
         try:
             return self.index.reachableroots2(
                 minroot, heads, roots, includepath
             )
         except AttributeError:
             return dagop._reachablerootspure(
                 self.parentrevs, minroot, roots, heads, includepath
             )
 
     def ancestor(self, a, b):
         """calculate the "best" common ancestor of nodes a and b"""
 
         a, b = self.rev(a), self.rev(b)
         try:
             ancs = self.index.ancestors(a, b)
         except (AttributeError, OverflowError):
             ancs = ancestor.ancestors(self.parentrevs, a, b)
         if ancs:
             # choose a consistent winner when there's a tie
             return min(map(self.node, ancs))
         return self.nullid
 
     def _match(self, id):
         if isinstance(id, int):
             # rev
             return self.node(id)
         if len(id) == self.nodeconstants.nodelen:
             # possibly a binary node
             # odds of a binary node being all hex in ASCII are 1 in 10**25
             try:
                 node = id
                 self.rev(node)  # quick search the index
                 return node
             except error.LookupError:
                 pass  # may be partial hex id
         try:
             # str(rev)
             rev = int(id)
             if b"%d" % rev != id:
                 raise ValueError
             if rev < 0:
                 rev = len(self) + rev
             if rev < 0 or rev >= len(self):
                 raise ValueError
             return self.node(rev)
         except (ValueError, OverflowError):
             pass
         if len(id) == 2 * self.nodeconstants.nodelen:
             try:
                 # a full hex nodeid?
                 node = bin(id)
                 self.rev(node)
                 return node
             except (binascii.Error, error.LookupError):
                 pass
 
     def _partialmatch(self, id):
         # we don't care wdirfilenodeids as they should be always full hash
         maybewdir = self.nodeconstants.wdirhex.startswith(id)
         ambiguous = False
         try:
             partial = self.index.partialmatch(id)
             if partial and self.hasnode(partial):
                 if maybewdir:
                     # single 'ff...' match in radix tree, ambiguous with wdir
                     ambiguous = True
                 else:
                     return partial
             elif maybewdir:
                 # no 'ff...' match in radix tree, wdir identified
                 raise error.WdirUnsupported
             else:
                 return None
         except error.RevlogError:
             # parsers.c radix tree lookup gave multiple matches
             # fast path: for unfiltered changelog, radix tree is accurate
             if not getattr(self, 'filteredrevs', None):
                 ambiguous = True
             # fall through to slow path that filters hidden revisions
         except (AttributeError, ValueError):
             # we are pure python, or key is not hex
             pass
         if ambiguous:
             raise error.AmbiguousPrefixLookupError(
                 id, self.display_id, _(b'ambiguous identifier')
             )
 
         if id in self._pcache:
             return self._pcache[id]
 
         if len(id) <= 40:
             # hex(node)[:...]
             l = len(id) // 2 * 2  # grab an even number of digits
             try:
                 # we're dropping the last digit, so let's check that it's hex,
                 # to avoid the expensive computation below if it's not
                 if len(id) % 2 > 0:
                     if not (id[-1] in hexdigits):
                         return None
                 prefix = bin(id[:l])
             except binascii.Error:
                 pass
             else:
                 nl = [e[7] for e in self.index if e[7].startswith(prefix)]
                 nl = [
                     n for n in nl if hex(n).startswith(id) and self.hasnode(n)
                 ]
                 if self.nodeconstants.nullhex.startswith(id):
                     nl.append(self.nullid)
                 if len(nl) > 0:
                     if len(nl) == 1 and not maybewdir:
                         self._pcache[id] = nl[0]
                         return nl[0]
                     raise error.AmbiguousPrefixLookupError(
                         id, self.display_id, _(b'ambiguous identifier')
                     )
                 if maybewdir:
                     raise error.WdirUnsupported
                 return None
 
     def lookup(self, id):
         """locate a node based on:
         - revision number or str(revision number)
         - nodeid or subset of hex nodeid
         """
         n = self._match(id)
         if n is not None:
             return n
         n = self._partialmatch(id)
         if n:
             return n
 
         raise error.LookupError(id, self.display_id, _(b'no match found'))
 
     def shortest(self, node, minlength=1):
         """Find the shortest unambiguous prefix that matches node."""
 
         def isvalid(prefix):
             try:
                 matchednode = self._partialmatch(prefix)
             except error.AmbiguousPrefixLookupError:
                 return False
             except error.WdirUnsupported:
                 # single 'ff...' match
                 return True
             if matchednode is None:
                 raise error.LookupError(node, self.display_id, _(b'no node'))
             return True
 
         def maybewdir(prefix):
             return all(c == b'f' for c in pycompat.iterbytestr(prefix))
 
         hexnode = hex(node)
 
         def disambiguate(hexnode, minlength):
             """Disambiguate against wdirid."""
             for length in range(minlength, len(hexnode) + 1):
                 prefix = hexnode[:length]
                 if not maybewdir(prefix):
                     return prefix
 
         if not getattr(self, 'filteredrevs', None):
             try:
                 length = max(self.index.shortest(node), minlength)
                 return disambiguate(hexnode, length)
             except error.RevlogError:
                 if node != self.nodeconstants.wdirid:
                     raise error.LookupError(
                         node, self.display_id, _(b'no node')
                     )
             except AttributeError:
                 # Fall through to pure code
                 pass
 
         if node == self.nodeconstants.wdirid:
             for length in range(minlength, len(hexnode) + 1):
                 prefix = hexnode[:length]
                 if isvalid(prefix):
                     return prefix
 
         for length in range(minlength, len(hexnode) + 1):
             prefix = hexnode[:length]
             if isvalid(prefix):
                 return disambiguate(hexnode, length)
 
     def cmp(self, node, text):
         """compare text with a given file revision
 
         returns True if text is different than what is stored.
         """
         p1, p2 = self.parents(node)
         return storageutil.hashrevisionsha1(text, p1, p2) != node
 
     def _getsegmentforrevs(self, startrev, endrev, df=None):
         """Obtain a segment of raw data corresponding to a range of revisions.
 
         Accepts the start and end revisions and an optional already-open
         file handle to be used for reading. If the file handle is read, its
         seek position will not be preserved.
 
         Requests for data may be satisfied by a cache.
 
         Returns a 2-tuple of (offset, data) for the requested range of
         revisions. Offset is the integer offset from the beginning of the
         revlog and data is a str or buffer of the raw byte data.
 
         Callers will need to call ``self.start(rev)`` and ``self.length(rev)``
         to determine where each revision's data begins and ends.
         """
         # Inlined self.start(startrev) & self.end(endrev) for perf reasons
         # (functions are expensive).
         index = self.index
         istart = index[startrev]
         start = int(istart[0] >> 16)
         if startrev == endrev:
             end = start + istart[1]
         else:
             iend = index[endrev]
             end = int(iend[0] >> 16) + iend[1]
 
         if self._inline:
             start += (startrev + 1) * self.index.entry_size
             end += (endrev + 1) * self.index.entry_size
         length = end - start
 
         return start, self._segmentfile.read_chunk(start, length, df)
 
     def _chunk(self, rev, df=None):
         """Obtain a single decompressed chunk for a revision.
 
         Accepts an integer revision and an optional already-open file handle
         to be used for reading. If used, the seek position of the file will not
         be preserved.
 
         Returns a str holding uncompressed data for the requested revision.
         """
         compression_mode = self.index[rev][10]
         data = self._getsegmentforrevs(rev, rev, df=df)[1]
         if compression_mode == COMP_MODE_PLAIN:
             return data
         elif compression_mode == COMP_MODE_DEFAULT:
             return self._decompressor(data)
         elif compression_mode == COMP_MODE_INLINE:
             return self.decompress(data)
         else:
             msg = b'unknown compression mode %d'
             msg %= compression_mode
             raise error.RevlogError(msg)
 
     def _chunks(self, revs, df=None, targetsize=None):
         """Obtain decompressed chunks for the specified revisions.
 
         Accepts an iterable of numeric revisions that are assumed to be in
         ascending order. Also accepts an optional already-open file handle
         to be used for reading. If used, the seek position of the file will
         not be preserved.
 
         This function is similar to calling ``self._chunk()`` multiple times,
         but is faster.
 
         Returns a list with decompressed data for each requested revision.
         """
         if not revs:
             return []
         start = self.start
         length = self.length
         inline = self._inline
         iosize = self.index.entry_size
         buffer = util.buffer
 
         l = []
         ladd = l.append
 
         if not self._withsparseread:
             slicedchunks = (revs,)
         else:
             slicedchunks = deltautil.slicechunk(
                 self, revs, targetsize=targetsize
             )
 
         for revschunk in slicedchunks:
             firstrev = revschunk[0]
             # Skip trailing revisions with empty diff
             for lastrev in revschunk[::-1]:
                 if length(lastrev) != 0:
                     break
 
             try:
                 offset, data = self._getsegmentforrevs(firstrev, lastrev, df=df)
             except OverflowError:
                 # issue4215 - we can't cache a run of chunks greater than
                 # 2G on Windows
                 return [self._chunk(rev, df=df) for rev in revschunk]
 
             decomp = self.decompress
             # self._decompressor might be None, but will not be used in that case
             def_decomp = self._decompressor
             for rev in revschunk:
                 chunkstart = start(rev)
                 if inline:
                     chunkstart += (rev + 1) * iosize
                 chunklength = length(rev)
                 comp_mode = self.index[rev][10]
                 c = buffer(data, chunkstart - offset, chunklength)
                 if comp_mode == COMP_MODE_PLAIN:
                     ladd(c)
                 elif comp_mode == COMP_MODE_INLINE:
                     ladd(decomp(c))
                 elif comp_mode == COMP_MODE_DEFAULT:
                     ladd(def_decomp(c))
                 else:
                     msg = b'unknown compression mode %d'
                     msg %= comp_mode
                     raise error.RevlogError(msg)
 
         return l
 
     def deltaparent(self, rev):
         """return deltaparent of the given revision"""
         base = self.index[rev][3]
         if base == rev:
             return nullrev
         elif self._generaldelta:
             return base
         else:
             return rev - 1
 
     def issnapshot(self, rev):
         """tells whether rev is a snapshot"""
         if not self._sparserevlog:
             return self.deltaparent(rev) == nullrev
         elif util.safehasattr(self.index, b'issnapshot'):
             # directly assign the method to cache the testing and access
             self.issnapshot = self.index.issnapshot
             return self.issnapshot(rev)
         if rev == nullrev:
             return True
         entry = self.index[rev]
         base = entry[3]
         if base == rev:
             return True
         if base == nullrev:
             return True
         p1 = entry[5]
         while self.length(p1) == 0:
             b = self.deltaparent(p1)
             if b == p1:
                 break
             p1 = b
         p2 = entry[6]
         while self.length(p2) == 0:
             b = self.deltaparent(p2)
             if b == p2:
                 break
             p2 = b
         if base == p1 or base == p2:
             return False
         return self.issnapshot(base)
 
     def snapshotdepth(self, rev):
         """number of snapshot in the chain before this one"""
         if not self.issnapshot(rev):
             raise error.ProgrammingError(b'revision %d not a snapshot')
         return len(self._deltachain(rev)[0]) - 1
 
     def revdiff(self, rev1, rev2):
         """return or calculate a delta between two revisions
 
         The delta calculated is in binary form and is intended to be written to
         revlog data directly. So this function needs raw revision data.
         """
         if rev1 != nullrev and self.deltaparent(rev2) == rev1:
             return bytes(self._chunk(rev2))
 
         return mdiff.textdiff(self.rawdata(rev1), self.rawdata(rev2))
 
     def revision(self, nodeorrev, _df=None):
         """return an uncompressed revision of a given node or revision
         number.
 
         _df - an existing file handle to read from. (internal-only)
         """
         return self._revisiondata(nodeorrev, _df)
 
     def sidedata(self, nodeorrev, _df=None):
         """a map of extra data related to the changeset but not part of the hash
 
         This function currently return a dictionary. However, more advanced
         mapping object will likely be used in the future for a more
         efficient/lazy code.
         """
         # deal with <nodeorrev> argument type
         if isinstance(nodeorrev, int):
             rev = nodeorrev
         else:
             rev = self.rev(nodeorrev)
         return self._sidedata(rev)
 
     def _revisiondata(self, nodeorrev, _df=None, raw=False):
         # deal with <nodeorrev> argument type
         if isinstance(nodeorrev, int):
             rev = nodeorrev
             node = self.node(rev)
         else:
             node = nodeorrev
             rev = None
 
         # fast path the special `nullid` rev
         if node == self.nullid:
             return b""
 
         # ``rawtext`` is the text as stored inside the revlog. Might be the
         # revision or might need to be processed to retrieve the revision.
         rev, rawtext, validated = self._rawtext(node, rev, _df=_df)
 
         if raw and validated:
             # if we don't want to process the raw text and that raw
             # text is cached, we can exit early.
             return rawtext
         if rev is None:
             rev = self.rev(node)
         # the revlog's flag for this revision
         # (usually alter its state or content)
         flags = self.flags(rev)
 
         if validated and flags == REVIDX_DEFAULT_FLAGS:
             # no extra flags set, no flag processor runs, text = rawtext
             return rawtext
 
         if raw:
             validatehash = flagutil.processflagsraw(self, rawtext, flags)
             text = rawtext
         else:
             r = flagutil.processflagsread(self, rawtext, flags)
             text, validatehash = r
         if validatehash:
             self.checkhash(text, node, rev=rev)
         if not validated:
             self._revisioncache = (node, rev, rawtext)
 
         return text
 
     def _rawtext(self, node, rev, _df=None):
         """return the possibly unvalidated rawtext for a revision
 
         returns (rev, rawtext, validated)
         """
 
         # revision in the cache (could be useful to apply delta)
         cachedrev = None
         # An intermediate text to apply deltas to
         basetext = None
 
         # Check if we have the entry in cache
         # The cache entry looks like (node, rev, rawtext)
         if self._revisioncache:
             if self._revisioncache[0] == node:
                 return (rev, self._revisioncache[2], True)
             cachedrev = self._revisioncache[1]
 
         if rev is None:
             rev = self.rev(node)
 
         chain, stopped = self._deltachain(rev, stoprev=cachedrev)
         if stopped:
             basetext = self._revisioncache[2]
 
         # drop cache to save memory, the caller is expected to
         # update self._revisioncache after validating the text
         self._revisioncache = None
 
         targetsize = None
         rawsize = self.index[rev][2]
         if 0 <= rawsize:
             targetsize = 4 * rawsize
 
         bins = self._chunks(chain, df=_df, targetsize=targetsize)
         if basetext is None:
             basetext = bytes(bins[0])
             bins = bins[1:]
 
         rawtext = mdiff.patches(basetext, bins)
         del basetext  # let us have a chance to free memory early
         return (rev, rawtext, False)
 
     def _sidedata(self, rev):
         """Return the sidedata for a given revision number."""
         index_entry = self.index[rev]
         sidedata_offset = index_entry[8]
         sidedata_size = index_entry[9]
 
         if self._inline:
             sidedata_offset += self.index.entry_size * (1 + rev)
         if sidedata_size == 0:
             return {}
 
         if self._docket.sidedata_end < sidedata_offset + sidedata_size:
             filename = self._sidedatafile
             end = self._docket.sidedata_end
             offset = sidedata_offset
             length = sidedata_size
             m = FILE_TOO_SHORT_MSG % (filename, length, offset, end)
             raise error.RevlogError(m)
 
         comp_segment = self._segmentfile_sidedata.read_chunk(
             sidedata_offset, sidedata_size
         )
 
         comp = self.index[rev][11]
         if comp == COMP_MODE_PLAIN:
             segment = comp_segment
         elif comp == COMP_MODE_DEFAULT:
             segment = self._decompressor(comp_segment)
         elif comp == COMP_MODE_INLINE:
             segment = self.decompress(comp_segment)
         else:
             msg = b'unknown compression mode %d'
             msg %= comp
             raise error.RevlogError(msg)
 
         sidedata = sidedatautil.deserialize_sidedata(segment)
         return sidedata
 
     def rawdata(self, nodeorrev, _df=None):
         """return an uncompressed raw data of a given node or revision number.
 
         _df - an existing file handle to read from. (internal-only)
         """
         return self._revisiondata(nodeorrev, _df, raw=True)
 
     def hash(self, text, p1, p2):
         """Compute a node hash.
 
         Available as a function so that subclasses can replace the hash
         as needed.
         """
         return storageutil.hashrevisionsha1(text, p1, p2)
 
     def checkhash(self, text, node, p1=None, p2=None, rev=None):
         """Check node hash integrity.
 
         Available as a function so that subclasses can extend hash mismatch
         behaviors as needed.
         """
         try:
             if p1 is None and p2 is None:
                 p1, p2 = self.parents(node)
             if node != self.hash(text, p1, p2):
                 # Clear the revision cache on hash failure. The revision cache
                 # only stores the raw revision and clearing the cache does have
                 # the side-effect that we won't have a cache hit when the raw
                 # revision data is accessed. But this case should be rare and
                 # it is extra work to teach the cache about the hash
                 # verification state.
                 if self._revisioncache and self._revisioncache[0] == node:
                     self._revisioncache = None
 
                 revornode = rev
                 if revornode is None:
                     revornode = templatefilters.short(hex(node))
                 raise error.RevlogError(
                     _(b"integrity check failed on %s:%s")
                     % (self.display_id, pycompat.bytestr(revornode))
                 )
         except error.RevlogError:
             if self._censorable and storageutil.iscensoredtext(text):
                 raise error.CensoredNodeError(self.display_id, node, text)
             raise
 
-    def _enforceinlinesize(self, tr):
+    def _enforceinlinesize(self, tr, side_write=True):
         """Check if the revlog is too big for inline and convert if so.
 
         This should be called after revisions are added to the revlog. If the
         revlog has grown too large to be an inline revlog, it will convert it
         to use multiple index and data files.
         """
         tiprev = len(self) - 1
         total_size = self.start(tiprev) + self.length(tiprev)
         if not self._inline or total_size < _maxinline:
             return
 
         troffset = tr.findoffset(self._indexfile)
         if troffset is None:
             raise error.RevlogError(
                 _(b"%s not found in the transaction") % self._indexfile
             )
-        trindex = None
+        if troffset:
+            tr.addbackup(self._indexfile, for_offset=True)
         tr.add(self._datafile, 0)
 
         existing_handles = False
         if self._writinghandles is not None:
             existing_handles = True
             fp = self._writinghandles[0]
             fp.flush()
             fp.close()
             # We can't use the cached file handle after close(). So prevent
             # its usage.
             self._writinghandles = None
             self._segmentfile.writing_handle = None
             # No need to deal with sidedata writing handle as it is only
             # relevant with revlog-v2 which is never inline, not reaching
             # this code
+        if side_write:
+            old_index_file_path = self._indexfile
+            new_index_file_path = self._indexfile + b'.s'
+            opener = self.opener
+
+            fncache = getattr(opener, 'fncache', None)
+            if fncache is not None:
+                fncache.addignore(new_index_file_path)
+
+            # the "split" index replace the real index when the transaction is finalized
+            def finalize_callback(tr):
+                opener.rename(
+                    new_index_file_path,
+                    old_index_file_path,
+                    checkambig=True,
+                )
+
+            tr.registertmp(new_index_file_path)
+            if self.target[1] is not None:
+                finalize_id = b'000-revlog-split-%d-%s' % self.target
+            else:
+                finalize_id = b'000-revlog-split-%d' % self.target[0]
+            tr.addfinalize(finalize_id, finalize_callback)
 
         new_dfh = self._datafp(b'w+')
         new_dfh.truncate(0)  # drop any potentially existing data
         try:
             with self._indexfp() as read_ifh:
                 for r in self:
                     new_dfh.write(self._getsegmentforrevs(r, r, df=read_ifh)[1])
-                    if (
-                        trindex is None
-                        and troffset
-                        <= self.start(r) + r * self.index.entry_size
-                    ):
-                        trindex = r
                 new_dfh.flush()
 
-            if trindex is None:
-                trindex = 0
-
+            if side_write:
+                self._indexfile = new_index_file_path
             with self.__index_new_fp() as fp:
                 self._format_flags &= ~FLAG_INLINE_DATA
                 self._inline = False
                 for i in self:
                     e = self.index.entry_binary(i)
                     if i == 0 and self._docket is None:
                         header = self._format_flags | self._format_version
                         header = self.index.pack_header(header)
                         e = header + e
                     fp.write(e)
                 if self._docket is not None:
                     self._docket.index_end = fp.tell()
 
-                # There is a small transactional race here. If the rename of
-                # the index fails, we should remove the datafile. It is more
-                # important to ensure that the data file is not truncated
-                # when the index is replaced as otherwise data is lost.
-                tr.replace(self._datafile, self.start(trindex))
-
-                # the temp file replace the real index when we exit the context
-                # manager
-
-            tr.replace(self._indexfile, trindex * self.index.entry_size)
+                # If we don't use side-write, the temp file replace the real
+                # index when we exit the context manager
+
             nodemaputil.setup_persistent_nodemap(tr, self)
             self._segmentfile = randomaccessfile.randomaccessfile(
                 self.opener,
                 self._datafile,
                 self._chunkcachesize,
             )
 
             if existing_handles:
                 # switched from inline to conventional reopen the index
                 ifh = self.__index_write_fp()
                 self._writinghandles = (ifh, new_dfh, None)
                 self._segmentfile.writing_handle = new_dfh
                 new_dfh = None
                 # No need to deal with sidedata writing handle as it is only
                 # relevant with revlog-v2 which is never inline, not reaching
                 # this code
         finally:
             if new_dfh is not None:
                 new_dfh.close()
 
     def _nodeduplicatecallback(self, transaction, node):
         """called when trying to add a node already stored."""
 
     @contextlib.contextmanager
     def reading(self):
         """Context manager that keeps data and sidedata files open for reading"""
         with self._segmentfile.reading():
             with self._segmentfile_sidedata.reading():
                 yield
 
     @contextlib.contextmanager
     def _writing(self, transaction):
         if self._trypending:
             msg = b'try to write in a `trypending` revlog: %s'
             msg %= self.display_id
             raise error.ProgrammingError(msg)
         if self._writinghandles is not None:
             yield
         else:
             ifh = dfh = sdfh = None
             try:
                 r = len(self)
                 # opening the data file.
                 dsize = 0
                 if r:
                     dsize = self.end(r - 1)
                 dfh = None
                 if not self._inline:
                     try:
                         dfh = self._datafp(b"r+")
                         if self._docket is None:
                             dfh.seek(0, os.SEEK_END)
                         else:
                             dfh.seek(self._docket.data_end, os.SEEK_SET)
                     except FileNotFoundError:
                         dfh = self._datafp(b"w+")
                     transaction.add(self._datafile, dsize)
                 if self._sidedatafile is not None:
                     # revlog-v2 does not inline, help Pytype
                     assert dfh is not None
                     try:
                         sdfh = self.opener(self._sidedatafile, mode=b"r+")
                         dfh.seek(self._docket.sidedata_end, os.SEEK_SET)
                     except FileNotFoundError:
                         sdfh = self.opener(self._sidedatafile, mode=b"w+")
                     transaction.add(
                         self._sidedatafile, self._docket.sidedata_end
                     )
 
                 # opening the index file.
                 isize = r * self.index.entry_size
                 ifh = self.__index_write_fp()
                 if self._inline:
                     transaction.add(self._indexfile, dsize + isize)
                 else:
                     transaction.add(self._indexfile, isize)
                 # exposing all file handle for writing.
                 self._writinghandles = (ifh, dfh, sdfh)
                 self._segmentfile.writing_handle = ifh if self._inline else dfh
                 self._segmentfile_sidedata.writing_handle = sdfh
                 yield
                 if self._docket is not None:
                     self._write_docket(transaction)
             finally:
                 self._writinghandles = None
                 self._segmentfile.writing_handle = None
                 self._segmentfile_sidedata.writing_handle = None
                 if dfh is not None:
                     dfh.close()
                 if sdfh is not None:
                     sdfh.close()
                 # closing the index file last to avoid exposing referent to
                 # potential unflushed data content.
                 if ifh is not None:
                     ifh.close()
 
     def _write_docket(self, transaction):
         """write the current docket on disk
 
         Exist as a method to help changelog to implement transaction logic
 
         We could also imagine using the same transaction logic for all revlog
         since docket are cheap."""
         self._docket.write(transaction)
 
     def addrevision(
         self,
         text,
         transaction,
         link,
         p1,
         p2,
         cachedelta=None,
         node=None,
         flags=REVIDX_DEFAULT_FLAGS,
         deltacomputer=None,
         sidedata=None,
     ):
         """add a revision to the log
 
         text - the revision data to add
         transaction - the transaction object used for rollback
         link - the linkrev data to add
         p1, p2 - the parent nodeids of the revision
         cachedelta - an optional precomputed delta
         node - nodeid of revision; typically node is not specified, and it is
             computed by default as hash(text, p1, p2), however subclasses might
             use different hashing method (and override checkhash() in such case)
         flags - the known flags to set on the revision
         deltacomputer - an optional deltacomputer instance shared between
             multiple calls
         """
         if link == nullrev:
             raise error.RevlogError(
                 _(b"attempted to add linkrev -1 to %s") % self.display_id
             )
 
         if sidedata is None:
             sidedata = {}
         elif sidedata and not self.hassidedata:
             raise error.ProgrammingError(
                 _(b"trying to add sidedata to a revlog who don't support them")
             )
 
         if flags:
             node = node or self.hash(text, p1, p2)
 
         rawtext, validatehash = flagutil.processflagswrite(self, text, flags)
 
         # If the flag processor modifies the revision data, ignore any provided
         # cachedelta.
         if rawtext != text:
             cachedelta = None
 
         if len(rawtext) > _maxentrysize:
             raise error.RevlogError(
                 _(
                     b"%s: size of %d bytes exceeds maximum revlog storage of 2GiB"
                 )
                 % (self.display_id, len(rawtext))
             )
 
         node = node or self.hash(rawtext, p1, p2)
         rev = self.index.get_rev(node)
         if rev is not None:
             return rev
 
         if validatehash:
             self.checkhash(rawtext, node, p1=p1, p2=p2)
 
         return self.addrawrevision(
             rawtext,
             transaction,
             link,
             p1,
             p2,
             node,
             flags,
             cachedelta=cachedelta,
             deltacomputer=deltacomputer,
             sidedata=sidedata,
         )
 
     def addrawrevision(
         self,
         rawtext,
         transaction,
         link,
         p1,
         p2,
         node,
         flags,
         cachedelta=None,
         deltacomputer=None,
         sidedata=None,
     ):
         """add a raw revision with known flags, node and parents
         useful when reusing a revision not stored in this revlog (ex: received
         over wire, or read from an external bundle).
         """
         with self._writing(transaction):
             return self._addrevision(
                 node,
                 rawtext,
                 transaction,
                 link,
                 p1,
                 p2,
                 flags,
                 cachedelta,
                 deltacomputer=deltacomputer,
                 sidedata=sidedata,
             )
 
     def compress(self, data):
         """Generate a possibly-compressed representation of data."""
         if not data:
             return b'', data
 
         compressed = self._compressor.compress(data)
 
         if compressed:
             # The revlog compressor added the header in the returned data.
             return b'', compressed
 
         if data[0:1] == b'\0':
             return b'', data
         return b'u', data
 
     def decompress(self, data):
         """Decompress a revlog chunk.
 
         The chunk is expected to begin with a header identifying the
         format type so it can be routed to an appropriate decompressor.
         """
         if not data:
             return data
 
         # Revlogs are read much more frequently than they are written and many
         # chunks only take microseconds to decompress, so performance is
         # important here.
         #
         # We can make a few assumptions about revlogs:
         #
         # 1) the majority of chunks will be compressed (as opposed to inline
         #    raw data).
         # 2) decompressing *any* data will likely by at least 10x slower than
         #    returning raw inline data.
         # 3) we want to prioritize common and officially supported compression
         #    engines
         #
         # It follows that we want to optimize for "decompress compressed data
         # when encoded with common and officially supported compression engines"
         # case over "raw data" and "data encoded by less common or non-official
         # compression engines." That is why we have the inline lookup first
         # followed by the compengines lookup.
         #
         # According to `hg perfrevlogchunks`, this is ~0.5% faster for zlib
         # compressed chunks. And this matters for changelog and manifest reads.
         t = data[0:1]
 
         if t == b'x':
             try:
                 return _zlibdecompress(data)
             except zlib.error as e:
                 raise error.RevlogError(
                     _(b'revlog decompress error: %s')
                     % stringutil.forcebytestr(e)
                 )
         # '\0' is more common than 'u' so it goes first.
         elif t == b'\0':
             return data
         elif t == b'u':
             return util.buffer(data, 1)
 
         compressor = self._get_decompressor(t)
 
         return compressor.decompress(data)
 
     def _addrevision(
         self,
         node,
         rawtext,
         transaction,
         link,
         p1,
         p2,
         flags,
         cachedelta,
         alwayscache=False,
         deltacomputer=None,
         sidedata=None,
     ):
         """internal function to add revisions to the log
 
         see addrevision for argument descriptions.
 
         note: "addrevision" takes non-raw text, "_addrevision" takes raw text.
 
         if "deltacomputer" is not provided or None, a defaultdeltacomputer will
         be used.
 
         invariants:
         - rawtext is optional (can be None); if not set, cachedelta must be set.
           if both are set, they must correspond to each other.
         """
         if node == self.nullid:
             raise error.RevlogError(
                 _(b"%s: attempt to add null revision") % self.display_id
             )
         if (
             node == self.nodeconstants.wdirid
             or node in self.nodeconstants.wdirfilenodeids
         ):
             raise error.RevlogError(
                 _(b"%s: attempt to add wdir revision") % self.display_id
             )
         if self._writinghandles is None:
             msg = b'adding revision outside `revlog._writing` context'
             raise error.ProgrammingError(msg)
 
         if self._inline:
             fh = self._writinghandles[0]
         else:
             fh = self._writinghandles[1]
 
         btext = [rawtext]
 
         curr = len(self)
         prev = curr - 1
 
         offset = self._get_data_offset(prev)
 
         if self._concurrencychecker:
             ifh, dfh, sdfh = self._writinghandles
             # XXX no checking for the sidedata file
             if self._inline:
                 # offset is "as if" it were in the .d file, so we need to add on
                 # the size of the entry metadata.
                 self._concurrencychecker(
                     ifh, self._indexfile, offset + curr * self.index.entry_size
                 )
             else:
                 # Entries in the .i are a consistent size.
                 self._concurrencychecker(
                     ifh, self._indexfile, curr * self.index.entry_size
                 )
                 self._concurrencychecker(dfh, self._datafile, offset)
 
         p1r, p2r = self.rev(p1), self.rev(p2)
 
         # full versions are inserted when the needed deltas
         # become comparable to the uncompressed text
         if rawtext is None:
             # need rawtext size, before changed by flag processors, which is
             # the non-raw size. use revlog explicitly to avoid filelog's extra
             # logic that might remove metadata size.
             textlen = mdiff.patchedsize(
                 revlog.size(self, cachedelta[0]), cachedelta[1]
             )
         else:
             textlen = len(rawtext)
 
         if deltacomputer is None:
             write_debug = None
             if self._debug_delta:
                 write_debug = transaction._report
             deltacomputer = deltautil.deltacomputer(
                 self, write_debug=write_debug
             )
 
         if cachedelta is not None and len(cachedelta) == 2:
             # If the cached delta has no information about how it should be
             # reused, add the default reuse instruction according to the
             # revlog's configuration.
             if self._generaldelta and self._lazydeltabase:
                 delta_base_reuse = DELTA_BASE_REUSE_TRY
             else:
                 delta_base_reuse = DELTA_BASE_REUSE_NO
             cachedelta = (cachedelta[0], cachedelta[1], delta_base_reuse)
 
         revinfo = revlogutils.revisioninfo(
             node,
             p1,
             p2,
             btext,
             textlen,
             cachedelta,
             flags,
         )
 
         deltainfo = deltacomputer.finddeltainfo(revinfo, fh)
 
         compression_mode = COMP_MODE_INLINE
         if self._docket is not None:
             default_comp = self._docket.default_compression_header
             r = deltautil.delta_compression(default_comp, deltainfo)
             compression_mode, deltainfo = r
 
         sidedata_compression_mode = COMP_MODE_INLINE
         if sidedata and self.hassidedata:
             sidedata_compression_mode = COMP_MODE_PLAIN
             serialized_sidedata = sidedatautil.serialize_sidedata(sidedata)
             sidedata_offset = self._docket.sidedata_end
             h, comp_sidedata = self.compress(serialized_sidedata)
             if (
                 h != b'u'
                 and comp_sidedata[0:1] != b'\0'
                 and len(comp_sidedata) < len(serialized_sidedata)
             ):
                 assert not h
                 if (
                     comp_sidedata[0:1]
                     == self._docket.default_compression_header
                 ):
                     sidedata_compression_mode = COMP_MODE_DEFAULT
                     serialized_sidedata = comp_sidedata
                 else:
                     sidedata_compression_mode = COMP_MODE_INLINE
                     serialized_sidedata = comp_sidedata
         else:
             serialized_sidedata = b""
             # Don't store the offset if the sidedata is empty, that way
             # we can easily detect empty sidedata and they will be no different
             # than ones we manually add.
             sidedata_offset = 0
 
         rank = RANK_UNKNOWN
         if self._compute_rank:
             if (p1r, p2r) == (nullrev, nullrev):
                 rank = 1
             elif p1r != nullrev and p2r == nullrev:
                 rank = 1 + self.fast_rank(p1r)
             elif p1r == nullrev and p2r != nullrev:
                 rank = 1 + self.fast_rank(p2r)
             else:  # merge node
                 if rustdagop is not None and self.index.rust_ext_compat:
                     rank = rustdagop.rank(self.index, p1r, p2r)
                 else:
                     pmin, pmax = sorted((p1r, p2r))
                     rank = 1 + self.fast_rank(pmax)
                     rank += sum(1 for _ in self.findmissingrevs([pmax], [pmin]))
 
         e = revlogutils.entry(
             flags=flags,
             data_offset=offset,
             data_compressed_length=deltainfo.deltalen,
             data_uncompressed_length=textlen,
             data_compression_mode=compression_mode,
             data_delta_base=deltainfo.base,
             link_rev=link,
             parent_rev_1=p1r,
             parent_rev_2=p2r,
             node_id=node,
             sidedata_offset=sidedata_offset,
             sidedata_compressed_length=len(serialized_sidedata),
             sidedata_compression_mode=sidedata_compression_mode,
             rank=rank,
         )
 
         self.index.append(e)
         entry = self.index.entry_binary(curr)
         if curr == 0 and self._docket is None:
             header = self._format_flags | self._format_version
             header = self.index.pack_header(header)
             entry = header + entry
         self._writeentry(
             transaction,
             entry,
             deltainfo.data,
             link,
             offset,
             serialized_sidedata,
             sidedata_offset,
         )
 
         rawtext = btext[0]
 
         if alwayscache and rawtext is None:
             rawtext = deltacomputer.buildtext(revinfo, fh)
 
         if type(rawtext) == bytes:  # only accept immutable objects
             self._revisioncache = (node, curr, rawtext)
         self._chainbasecache[curr] = deltainfo.chainbase
         return curr
 
     def _get_data_offset(self, prev):
         """Returns the current offset in the (in-transaction) data file.
         Versions < 2 of the revlog can get this 0(1), revlog v2 needs a docket
         file to store that information: since sidedata can be rewritten to the
         end of the data file within a transaction, you can have cases where, for
         example, rev `n` does not have sidedata while rev `n - 1` does, leading
         to `n - 1`'s sidedata being written after `n`'s data.
 
         TODO cache this in a docket file before getting out of experimental."""
         if self._docket is None:
             return self.end(prev)
         else:
             return self._docket.data_end
 
     def _writeentry(
         self, transaction, entry, data, link, offset, sidedata, sidedata_offset
     ):
         # Files opened in a+ mode have inconsistent behavior on various
         # platforms. Windows requires that a file positioning call be made
         # when the file handle transitions between reads and writes. See
         # 3686fa2b8eee and the mixedfilemodewrapper in windows.py. On other
         # platforms, Python or the platform itself can be buggy. Some versions
         # of Solaris have been observed to not append at the end of the file
         # if the file was seeked to before the end. See issue4943 for more.
         #
         # We work around this issue by inserting a seek() before writing.
         # Note: This is likely not necessary on Python 3. However, because
         # the file handle is reused for reads and may be seeked there, we need
         # to be careful before changing this.
         if self._writinghandles is None:
             msg = b'adding revision outside `revlog._writing` context'
             raise error.ProgrammingError(msg)
         ifh, dfh, sdfh = self._writinghandles
         if self._docket is None:
             ifh.seek(0, os.SEEK_END)
         else:
             ifh.seek(self._docket.index_end, os.SEEK_SET)
         if dfh:
             if self._docket is None:
                 dfh.seek(0, os.SEEK_END)
             else:
                 dfh.seek(self._docket.data_end, os.SEEK_SET)
         if sdfh:
             sdfh.seek(self._docket.sidedata_end, os.SEEK_SET)
 
         curr = len(self) - 1
         if not self._inline:
             transaction.add(self._datafile, offset)
             if self._sidedatafile:
                 transaction.add(self._sidedatafile, sidedata_offset)
             transaction.add(self._indexfile, curr * len(entry))
             if data[0]:
                 dfh.write(data[0])
             dfh.write(data[1])
             if sidedata:
                 sdfh.write(sidedata)
             ifh.write(entry)
         else:
             offset += curr * self.index.entry_size
             transaction.add(self._indexfile, offset)
             ifh.write(entry)
             ifh.write(data[0])
             ifh.write(data[1])
             assert not sidedata
             self._enforceinlinesize(transaction)
         if self._docket is not None:
             # revlog-v2 always has 3 writing handles, help Pytype
             wh1 = self._writinghandles[0]
             wh2 = self._writinghandles[1]
             wh3 = self._writinghandles[2]
             assert wh1 is not None
             assert wh2 is not None
             assert wh3 is not None
             self._docket.index_end = wh1.tell()
             self._docket.data_end = wh2.tell()
             self._docket.sidedata_end = wh3.tell()
 
         nodemaputil.setup_persistent_nodemap(transaction, self)
 
     def addgroup(
         self,
         deltas,
         linkmapper,
         transaction,
         alwayscache=False,
         addrevisioncb=None,
         duplicaterevisioncb=None,
         debug_info=None,
         delta_base_reuse_policy=None,
     ):
         """
         add a delta group
 
         given a set of deltas, add them to the revision log. the
         first delta is against its parent, which should be in our
         log, the rest are against the previous delta.
 
         If ``addrevisioncb`` is defined, it will be called with arguments of
         this revlog and the node that was added.
         """
 
         if self._adding_group:
             raise error.ProgrammingError(b'cannot nest addgroup() calls')
 
         # read the default delta-base reuse policy from revlog config if the
         # group did not specify one.
         if delta_base_reuse_policy is None:
             if self._generaldelta and self._lazydeltabase:
                 delta_base_reuse_policy = DELTA_BASE_REUSE_TRY
             else:
                 delta_base_reuse_policy = DELTA_BASE_REUSE_NO
 
         self._adding_group = True
         empty = True
         try:
             with self._writing(transaction):
                 write_debug = None
                 if self._debug_delta:
                     write_debug = transaction._report
                 deltacomputer = deltautil.deltacomputer(
                     self,
                     write_debug=write_debug,
                     debug_info=debug_info,
                 )
                 # loop through our set of deltas
                 for data in deltas:
                     (
                         node,
                         p1,
                         p2,
                         linknode,
                         deltabase,
                         delta,
                         flags,
                         sidedata,
                     ) = data
                     link = linkmapper(linknode)
                     flags = flags or REVIDX_DEFAULT_FLAGS
 
                     rev = self.index.get_rev(node)
                     if rev is not None:
                         # this can happen if two branches make the same change
                         self._nodeduplicatecallback(transaction, rev)
                         if duplicaterevisioncb:
                             duplicaterevisioncb(self, rev)
                         empty = False
                         continue
 
                     for p in (p1, p2):
                         if not self.index.has_node(p):
                             raise error.LookupError(
                                 p, self.radix, _(b'unknown parent')
                             )
 
                     if not self.index.has_node(deltabase):
                         raise error.LookupError(
                             deltabase, self.display_id, _(b'unknown delta base')
                         )
 
                     baserev = self.rev(deltabase)
 
                     if baserev != nullrev and self.iscensored(baserev):
                         # if base is censored, delta must be full replacement in a
                         # single patch operation
                         hlen = struct.calcsize(b">lll")
                         oldlen = self.rawsize(baserev)
                         newlen = len(delta) - hlen
                         if delta[:hlen] != mdiff.replacediffheader(
                             oldlen, newlen
                         ):
                             raise error.CensoredBaseError(
                                 self.display_id, self.node(baserev)
                             )
 
                     if not flags and self._peek_iscensored(baserev, delta):
                         flags |= REVIDX_ISCENSORED
 
                     # We assume consumers of addrevisioncb will want to retrieve
                     # the added revision, which will require a call to
                     # revision(). revision() will fast path if there is a cache
                     # hit. So, we tell _addrevision() to always cache in this case.
                     # We're only using addgroup() in the context of changegroup
                     # generation so the revision data can always be handled as raw
                     # by the flagprocessor.
                     rev = self._addrevision(
                         node,
                         None,
                         transaction,
                         link,
                         p1,
                         p2,
                         flags,
                         (baserev, delta, delta_base_reuse_policy),
                         alwayscache=alwayscache,
                         deltacomputer=deltacomputer,
                         sidedata=sidedata,
                     )
 
                     if addrevisioncb:
                         addrevisioncb(self, rev)
                     empty = False
         finally:
             self._adding_group = False
         return not empty
 
     def iscensored(self, rev):
         """Check if a file revision is censored."""
         if not self._censorable:
             return False
 
         return self.flags(rev) & REVIDX_ISCENSORED
 
     def _peek_iscensored(self, baserev, delta):
         """Quickly check if a delta produces a censored revision."""
         if not self._censorable:
             return False
 
         return storageutil.deltaiscensored(delta, baserev, self.rawsize)
 
     def getstrippoint(self, minlink):
         """find the minimum rev that must be stripped to strip the linkrev
 
         Returns a tuple containing the minimum rev and a set of all revs that
         have linkrevs that will be broken by this strip.
         """
         return storageutil.resolvestripinfo(
             minlink,
             len(self) - 1,
             self.headrevs(),
             self.linkrev,
             self.parentrevs,
         )
 
     def strip(self, minlink, transaction):
         """truncate the revlog on the first revision with a linkrev >= minlink
 
         This function is called when we're stripping revision minlink and
         its descendants from the repository.
 
         We have to remove all revisions with linkrev >= minlink, because
         the equivalent changelog revisions will be renumbered after the
         strip.
 
         So we truncate the revlog on the first of these revisions, and
         trust that the caller has saved the revisions that shouldn't be
         removed and that it'll re-add them after this truncation.
         """
         if len(self) == 0:
             return
 
         rev, _ = self.getstrippoint(minlink)
         if rev == len(self):
             return
 
         # first truncate the files on disk
         data_end = self.start(rev)
         if not self._inline:
             transaction.add(self._datafile, data_end)
             end = rev * self.index.entry_size
         else:
             end = data_end + (rev * self.index.entry_size)
 
         if self._sidedatafile:
             sidedata_end = self.sidedata_cut_off(rev)
             transaction.add(self._sidedatafile, sidedata_end)
 
         transaction.add(self._indexfile, end)
         if self._docket is not None:
             # XXX we could, leverage the docket while stripping. However it is
             # not powerfull enough at the time of this comment
             self._docket.index_end = end
             self._docket.data_end = data_end
             self._docket.sidedata_end = sidedata_end
             self._docket.write(transaction, stripping=True)
 
         # then reset internal state in memory to forget those revisions
         self._revisioncache = None
         self._chaininfocache = util.lrucachedict(500)
         self._segmentfile.clear_cache()
         self._segmentfile_sidedata.clear_cache()
 
         del self.index[rev:-1]
 
     def checksize(self):
         """Check size of index and data files
 
         return a (dd, di) tuple.
         - dd: extra bytes for the "data" file
         - di: extra bytes for the "index" file
 
         A healthy revlog will return (0, 0).
         """
         expected = 0
         if len(self):
             expected = max(0, self.end(len(self) - 1))
 
         try:
             with self._datafp() as f:
                 f.seek(0, io.SEEK_END)
                 actual = f.tell()
             dd = actual - expected
         except FileNotFoundError:
             dd = 0
 
         try:
             f = self.opener(self._indexfile)
             f.seek(0, io.SEEK_END)
             actual = f.tell()
             f.close()
             s = self.index.entry_size
             i = max(0, actual // s)
             di = actual - (i * s)
             if self._inline:
                 databytes = 0
                 for r in self:
                     databytes += max(0, self.length(r))
                 dd = 0
                 di = actual - len(self) * s - databytes
         except FileNotFoundError:
             di = 0
 
         return (dd, di)
 
     def files(self):
         res = [self._indexfile]
         if self._docket_file is None:
             if not self._inline:
                 res.append(self._datafile)
         else:
             res.append(self._docket_file)
             res.extend(self._docket.old_index_filepaths(include_empty=False))
             if self._docket.data_end:
                 res.append(self._datafile)
             res.extend(self._docket.old_data_filepaths(include_empty=False))
             if self._docket.sidedata_end:
                 res.append(self._sidedatafile)
             res.extend(self._docket.old_sidedata_filepaths(include_empty=False))
         return res
 
     def emitrevisions(
         self,
         nodes,
         nodesorder=None,
         revisiondata=False,
         assumehaveparentrevisions=False,
         deltamode=repository.CG_DELTAMODE_STD,
         sidedata_helpers=None,
         debug_info=None,
     ):
         if nodesorder not in (b'nodes', b'storage', b'linear', None):
             raise error.ProgrammingError(
                 b'unhandled value for nodesorder: %s' % nodesorder
             )
 
         if nodesorder is None and not self._generaldelta:
             nodesorder = b'storage'
 
         if (
             not self._storedeltachains
             and deltamode != repository.CG_DELTAMODE_PREV
         ):
             deltamode = repository.CG_DELTAMODE_FULL
 
         return storageutil.emitrevisions(
             self,
             nodes,
             nodesorder,
             revlogrevisiondelta,
             deltaparentfn=self.deltaparent,
             candeltafn=self.candelta,
             rawsizefn=self.rawsize,
             revdifffn=self.revdiff,
             flagsfn=self.flags,
             deltamode=deltamode,
             revisiondata=revisiondata,
             assumehaveparentrevisions=assumehaveparentrevisions,
             sidedata_helpers=sidedata_helpers,
             debug_info=debug_info,
         )
 
     DELTAREUSEALWAYS = b'always'
     DELTAREUSESAMEREVS = b'samerevs'
     DELTAREUSENEVER = b'never'
 
     DELTAREUSEFULLADD = b'fulladd'
 
     DELTAREUSEALL = {b'always', b'samerevs', b'never', b'fulladd'}
 
     def clone(
         self,
         tr,
         destrevlog,
         addrevisioncb=None,
         deltareuse=DELTAREUSESAMEREVS,
         forcedeltabothparents=None,
         sidedata_helpers=None,
     ):
         """Copy this revlog to another, possibly with format changes.
 
         The destination revlog will contain the same revisions and nodes.
         However, it may not be bit-for-bit identical due to e.g. delta encoding
         differences.
 
         The ``deltareuse`` argument control how deltas from the existing revlog
         are preserved in the destination revlog. The argument can have the
         following values:
 
         DELTAREUSEALWAYS
            Deltas will always be reused (if possible), even if the destination
            revlog would not select the same revisions for the delta. This is the
            fastest mode of operation.
         DELTAREUSESAMEREVS
            Deltas will be reused if the destination revlog would pick the same
            revisions for the delta. This mode strikes a balance between speed
            and optimization.
         DELTAREUSENEVER
            Deltas will never be reused. This is the slowest mode of execution.
            This mode can be used to recompute deltas (e.g. if the diff/delta
            algorithm changes).
         DELTAREUSEFULLADD
            Revision will be re-added as if their were new content. This is
            slower than DELTAREUSEALWAYS but allow more mechanism to kicks in.
            eg: large file detection and handling.
 
         Delta computation can be slow, so the choice of delta reuse policy can
         significantly affect run time.
 
         The default policy (``DELTAREUSESAMEREVS``) strikes a balance between
         two extremes. Deltas will be reused if they are appropriate. But if the
         delta could choose a better revision, it will do so. This means if you
         are converting a non-generaldelta revlog to a generaldelta revlog,
         deltas will be recomputed if the delta's parent isn't a parent of the
         revision.
 
         In addition to the delta policy, the ``forcedeltabothparents``
         argument controls whether to force compute deltas against both parents
         for merges. By default, the current default is used.
 
         See `revlogutil.sidedata.get_sidedata_helpers` for the doc on
         `sidedata_helpers`.
         """
         if deltareuse not in self.DELTAREUSEALL:
             raise ValueError(
                 _(b'value for deltareuse invalid: %s') % deltareuse
             )
 
         if len(destrevlog):
             raise ValueError(_(b'destination revlog is not empty'))
 
         if getattr(self, 'filteredrevs', None):
             raise ValueError(_(b'source revlog has filtered revisions'))
         if getattr(destrevlog, 'filteredrevs', None):
             raise ValueError(_(b'destination revlog has filtered revisions'))
 
         # lazydelta and lazydeltabase controls whether to reuse a cached delta,
         # if possible.
         oldlazydelta = destrevlog._lazydelta
         oldlazydeltabase = destrevlog._lazydeltabase
         oldamd = destrevlog._deltabothparents
 
         try:
             if deltareuse == self.DELTAREUSEALWAYS:
                 destrevlog._lazydeltabase = True
                 destrevlog._lazydelta = True
             elif deltareuse == self.DELTAREUSESAMEREVS:
                 destrevlog._lazydeltabase = False
                 destrevlog._lazydelta = True
             elif deltareuse == self.DELTAREUSENEVER:
                 destrevlog._lazydeltabase = False
                 destrevlog._lazydelta = False
 
             destrevlog._deltabothparents = forcedeltabothparents or oldamd
 
             self._clone(
                 tr,
                 destrevlog,
                 addrevisioncb,
                 deltareuse,
                 forcedeltabothparents,
                 sidedata_helpers,
             )
 
         finally:
             destrevlog._lazydelta = oldlazydelta
             destrevlog._lazydeltabase = oldlazydeltabase
             destrevlog._deltabothparents = oldamd
 
     def _clone(
         self,
         tr,
         destrevlog,
         addrevisioncb,
         deltareuse,
         forcedeltabothparents,
         sidedata_helpers,
     ):
         """perform the core duty of `revlog.clone` after parameter processing"""
         write_debug = None
         if self._debug_delta:
             write_debug = tr._report
         deltacomputer = deltautil.deltacomputer(
             destrevlog,
             write_debug=write_debug,
         )
         index = self.index
         for rev in self:
             entry = index[rev]
 
             # Some classes override linkrev to take filtered revs into
             # account. Use raw entry from index.
             flags = entry[0] & 0xFFFF
             linkrev = entry[4]
             p1 = index[entry[5]][7]
             p2 = index[entry[6]][7]
             node = entry[7]
 
             # (Possibly) reuse the delta from the revlog if allowed and
             # the revlog chunk is a delta.
             cachedelta = None
             rawtext = None
             if deltareuse == self.DELTAREUSEFULLADD:
                 text = self._revisiondata(rev)
                 sidedata = self.sidedata(rev)
 
                 if sidedata_helpers is not None:
                     (sidedata, new_flags) = sidedatautil.run_sidedata_helpers(
                         self, sidedata_helpers, sidedata, rev
                     )
                     flags = flags | new_flags[0] & ~new_flags[1]
 
                 destrevlog.addrevision(
                     text,
                     tr,
                     linkrev,
                     p1,
                     p2,
                     cachedelta=cachedelta,
                     node=node,
                     flags=flags,
                     deltacomputer=deltacomputer,
                     sidedata=sidedata,
                 )
             else:
                 if destrevlog._lazydelta:
                     dp = self.deltaparent(rev)
                     if dp != nullrev:
                         cachedelta = (dp, bytes(self._chunk(rev)))
 
                 sidedata = None
                 if not cachedelta:
                     rawtext = self._revisiondata(rev)
                     sidedata = self.sidedata(rev)
                 if sidedata is None:
                     sidedata = self.sidedata(rev)
 
                 if sidedata_helpers is not None:
                     (sidedata, new_flags) = sidedatautil.run_sidedata_helpers(
                         self, sidedata_helpers, sidedata, rev
                     )
                     flags = flags | new_flags[0] & ~new_flags[1]
 
                 with destrevlog._writing(tr):
                     destrevlog._addrevision(
                         node,
                         rawtext,
                         tr,
                         linkrev,
                         p1,
                         p2,
                         flags,
                         cachedelta,
                         deltacomputer=deltacomputer,
                         sidedata=sidedata,
                     )
 
             if addrevisioncb:
                 addrevisioncb(self, rev, node)
 
     def censorrevision(self, tr, censornode, tombstone=b''):
         if self._format_version == REVLOGV0:
             raise error.RevlogError(
                 _(b'cannot censor with version %d revlogs')
                 % self._format_version
             )
         elif self._format_version == REVLOGV1:
             rewrite.v1_censor(self, tr, censornode, tombstone)
         else:
             rewrite.v2_censor(self, tr, censornode, tombstone)
 
     def verifyintegrity(self, state):
         """Verifies the integrity of the revlog.
 
         Yields ``revlogproblem`` instances describing problems that are
         found.
         """
         dd, di = self.checksize()
         if dd:
             yield revlogproblem(error=_(b'data length off by %d bytes') % dd)
         if di:
             yield revlogproblem(error=_(b'index contains %d extra bytes') % di)
 
         version = self._format_version
 
         # The verifier tells us what version revlog we should be.
         if version != state[b'expectedversion']:
             yield revlogproblem(
                 warning=_(b"warning: '%s' uses revlog format %d; expected %d")
                 % (self.display_id, version, state[b'expectedversion'])
             )
 
         state[b'skipread'] = set()
         state[b'safe_renamed'] = set()
 
         for rev in self:
             node = self.node(rev)
 
             # Verify contents. 4 cases to care about:
             #
             #   common: the most common case
             #   rename: with a rename
             #   meta: file content starts with b'\1\n', the metadata
             #         header defined in filelog.py, but without a rename
             #   ext: content stored externally
             #
             # More formally, their differences are shown below:
             #
             #                       | common | rename | meta  | ext
             #  -------------------------------------------------------
             #   flags()             | 0      | 0      | 0     | not 0
             #   renamed()           | False  | True   | False | ?
             #   rawtext[0:2]=='\1\n'| False  | True   | True  | ?
             #
             # "rawtext" means the raw text stored in revlog data, which
             # could be retrieved by "rawdata(rev)". "text"
             # mentioned below is "revision(rev)".
             #
             # There are 3 different lengths stored physically:
             #  1. L1: rawsize, stored in revlog index
             #  2. L2: len(rawtext), stored in revlog data
             #  3. L3: len(text), stored in revlog data if flags==0, or
             #     possibly somewhere else if flags!=0
             #
             # L1 should be equal to L2. L3 could be different from them.
             # "text" may or may not affect commit hash depending on flag
             # processors (see flagutil.addflagprocessor).
             #
             #              | common  | rename | meta  | ext
             # -------------------------------------------------
             #    rawsize() | L1      | L1     | L1    | L1
             #       size() | L1      | L2-LM  | L1(*) | L1 (?)
             # len(rawtext) | L2      | L2     | L2    | L2
             #    len(text) | L2      | L2     | L2    | L3
             #  len(read()) | L2      | L2-LM  | L2-LM | L3 (?)
             #
             # LM:  length of metadata, depending on rawtext
             # (*): not ideal, see comment in filelog.size
             # (?): could be "- len(meta)" if the resolved content has
             #      rename metadata
             #
             # Checks needed to be done:
             #  1. length check: L1 == L2, in all cases.
             #  2. hash check: depending on flag processor, we may need to
             #     use either "text" (external), or "rawtext" (in revlog).
 
             try:
                 skipflags = state.get(b'skipflags', 0)
                 if skipflags:
                     skipflags &= self.flags(rev)
 
                 _verify_revision(self, skipflags, state, node)
 
                 l1 = self.rawsize(rev)
                 l2 = len(self.rawdata(node))
 
                 if l1 != l2:
                     yield revlogproblem(
                         error=_(b'unpacked size is %d, %d expected') % (l2, l1),
                         node=node,
                     )
 
             except error.CensoredNodeError:
                 if state[b'erroroncensored']:
                     yield revlogproblem(
                         error=_(b'censored file data'), node=node
                     )
                     state[b'skipread'].add(node)
             except Exception as e:
                 yield revlogproblem(
                     error=_(b'unpacking %s: %s')
                     % (short(node), stringutil.forcebytestr(e)),
                     node=node,
                 )
                 state[b'skipread'].add(node)
 
     def storageinfo(
         self,
         exclusivefiles=False,
         sharedfiles=False,
         revisionscount=False,
         trackedsize=False,
         storedsize=False,
     ):
         d = {}
 
         if exclusivefiles:
             d[b'exclusivefiles'] = [(self.opener, self._indexfile)]
             if not self._inline:
                 d[b'exclusivefiles'].append((self.opener, self._datafile))
 
         if sharedfiles:
             d[b'sharedfiles'] = []
 
         if revisionscount:
             d[b'revisionscount'] = len(self)
 
         if trackedsize:
             d[b'trackedsize'] = sum(map(self.rawsize, iter(self)))
 
         if storedsize:
             d[b'storedsize'] = sum(
                 self.opener.stat(path).st_size for path in self.files()
             )
 
         return d
 
     def rewrite_sidedata(self, transaction, helpers, startrev, endrev):
         if not self.hassidedata:
             return
         # revlog formats with sidedata support does not support inline
         assert not self._inline
         if not helpers[1] and not helpers[2]:
             # Nothing to generate or remove
             return
 
         new_entries = []
         # append the new sidedata
         with self._writing(transaction):
             ifh, dfh, sdfh = self._writinghandles
             dfh.seek(self._docket.sidedata_end, os.SEEK_SET)
 
             current_offset = sdfh.tell()
             for rev in range(startrev, endrev + 1):
                 entry = self.index[rev]
                 new_sidedata, flags = sidedatautil.run_sidedata_helpers(
                     store=self,
                     sidedata_helpers=helpers,
                     sidedata={},
                     rev=rev,
                 )
 
                 serialized_sidedata = sidedatautil.serialize_sidedata(
                     new_sidedata
                 )
 
                 sidedata_compression_mode = COMP_MODE_INLINE
                 if serialized_sidedata and self.hassidedata:
                     sidedata_compression_mode = COMP_MODE_PLAIN
                     h, comp_sidedata = self.compress(serialized_sidedata)
                     if (
                         h != b'u'
                         and comp_sidedata[0] != b'\0'
                         and len(comp_sidedata) < len(serialized_sidedata)
                     ):
                         assert not h
                         if (
                             comp_sidedata[0]
                             == self._docket.default_compression_header
                         ):
                             sidedata_compression_mode = COMP_MODE_DEFAULT
                             serialized_sidedata = comp_sidedata
                         else:
                             sidedata_compression_mode = COMP_MODE_INLINE
                             serialized_sidedata = comp_sidedata
                 if entry[8] != 0 or entry[9] != 0:
                     # rewriting entries that already have sidedata is not
                     # supported yet, because it introduces garbage data in the
                     # revlog.
                     msg = b"rewriting existing sidedata is not supported yet"
                     raise error.Abort(msg)
 
                 # Apply (potential) flags to add and to remove after running
                 # the sidedata helpers
                 new_offset_flags = entry[0] | flags[0] & ~flags[1]
                 entry_update = (
                     current_offset,
                     len(serialized_sidedata),
                     new_offset_flags,
                     sidedata_compression_mode,
                 )
 
                 # the sidedata computation might have move the file cursors around
                 sdfh.seek(current_offset, os.SEEK_SET)
                 sdfh.write(serialized_sidedata)
                 new_entries.append(entry_update)
                 current_offset += len(serialized_sidedata)
                 self._docket.sidedata_end = sdfh.tell()
 
             # rewrite the new index entries
             ifh.seek(startrev * self.index.entry_size)
             for i, e in enumerate(new_entries):
                 rev = startrev + i
                 self.index.replace_sidedata_info(rev, *e)
                 packed = self.index.entry_binary(rev)
                 if rev == 0 and self._docket is None:
                     header = self._format_flags | self._format_version
                     header = self.index.pack_header(header)
                     packed = header + packed
                 ifh.write(packed)

tests/test-transaction-rollback-on-revlog-split.t

@@ -1,433 +1,428 b''
 Test correctness of revlog inline -> non-inline transition
 ----------------------------------------------------------
 
 Helper extension to intercept renames and kill process
 
   $ cat > $TESTTMP/intercept_before_rename.py << EOF
   > import os
   > import signal
   > from mercurial import extensions, util
   > 
   > def extsetup(ui):
-  >     def close(orig, *args, **kwargs):
-  >         path = util.normpath(args[0]._atomictempfile__name)
-  >         if path.endswith(b'/.hg/store/data/file.i'):
+  >     def rename(orig, src, dest, *args, **kwargs):
+  >         path = util.normpath(dest)
+  >         if path.endswith(b'data/file.i'):
   >             os.kill(os.getpid(), signal.SIGKILL)
-  >         return orig(*args, **kwargs)
-  >     extensions.wrapfunction(util.atomictempfile, 'close', close)
+  >         return orig(src, dest, *args, **kwargs)
+  >     extensions.wrapfunction(util, 'rename', rename)
   > EOF
 
   $ cat > $TESTTMP/intercept_after_rename.py << EOF
   > import os
   > import signal
   > from mercurial import extensions, util
   > 
   > def extsetup(ui):
   >     def close(orig, *args, **kwargs):
   >         path = util.normpath(args[0]._atomictempfile__name)
   >         r = orig(*args, **kwargs)
   >         if path.endswith(b'/.hg/store/data/file.i'):
   >             os.kill(os.getpid(), signal.SIGKILL)
   >         return r
   >     extensions.wrapfunction(util.atomictempfile, 'close', close)
+  > def extsetup(ui):
+  >     def rename(orig, src, dest, *args, **kwargs):
+  >         path = util.normpath(dest)
+  >         r = orig(src, dest, *args, **kwargs)
+  >         if path.endswith(b'data/file.i'):
+  >             os.kill(os.getpid(), signal.SIGKILL)
+  >         return r
+  >     extensions.wrapfunction(util, 'rename', rename)
   > EOF
 
   $ cat > $TESTTMP/killme.py << EOF
   > import os
   > import signal
   > 
   > def killme(ui, repo, hooktype, **kwargs):
   >     os.kill(os.getpid(), signal.SIGKILL)
   > EOF
 
   $ cat > $TESTTMP/reader_wait_split.py << EOF
   > import os
   > import signal
   > from mercurial import extensions, revlog, testing
   > def _wait_post_load(orig, self, *args, **kwargs):
   >     wait = b'data/file' in self.radix
   >     if wait:
   >         testing.wait_file(b"$TESTTMP/writer-revlog-split")
   >     r = orig(self, *args, **kwargs)
   >     if wait:
   >         testing.write_file(b"$TESTTMP/reader-index-read")
   >         testing.wait_file(b"$TESTTMP/writer-revlog-unsplit")
   >     return r
   > 
   > def extsetup(ui):
   >     extensions.wrapfunction(revlog.revlog, '_loadindex', _wait_post_load)
   > EOF
 
 setup a repository for tests
 ----------------------------
 
   $ cat >> $HGRCPATH << EOF
   > [format]
   > revlog-compression=none
   > EOF
 
   $ hg init troffset-computation
   $ cd troffset-computation
   $ printf '%20d' '1' > file
   $ hg commit -Aqma
   $ printf '%1024d' '1' > file
   $ hg commit -Aqmb
   $ printf '%20d' '1' > file
   $ hg commit -Aqmc
   $ dd if=/dev/zero of=file bs=1k count=128 > /dev/null 2>&1
-  $ hg commit -AqmD
+  $ hg commit -AqmD --traceback
 
 Reference size:
   $ f -s file
   file: size=131072
   $ f -s .hg/store/data/file*
   .hg/store/data/file.d: size=132139
   .hg/store/data/file.i: size=256
 
   $ cd ..
 
 
 Test a hard crash after the file was split but before the transaction was committed
 ===================================================================================
 
 Test offset computation to correctly factor in the index entries themselves.
 Also test that the new data size has the correct size if the transaction is aborted
 after the index has been replaced.
 
 Test repo has commits a, b, c, D, where D is large (grows the revlog enough that it
 transitions to non-inline storage). The clone initially has changes a, b
 and will transition to non-inline storage when adding c, D.
 
 If the transaction adding c, D is rolled back, then we don't undo the revlog split,
 but truncate the index and the data to remove both c and D.
 
 
   $ hg clone --quiet --rev 1 troffset-computation troffset-computation-copy
   $ cd troffset-computation-copy
 
 Reference size:
   $ f -s file
   file: size=1024
   $ f -s .hg/store/data/file*
   .hg/store/data/file.i: size=1174
 
   $ cat > .hg/hgrc <<EOF
   > [hooks]
   > pretxnchangegroup = python:$TESTTMP/killme.py:killme
   > EOF
 #if chg
   $ hg pull ../troffset-computation
   pulling from ../troffset-computation
   [255]
 #else
   $ hg pull ../troffset-computation
   pulling from ../troffset-computation
   Killed
   [137]
 #endif
 
 
-The revlog have been split on disk
+The inline revlog still exist, but a split version exist next to it
 
   $ f -s .hg/store/data/file*
   .hg/store/data/file.d: size=132139
-  .hg/store/data/file.i: size=256
+  .hg/store/data/file.i: size=132395
+  .hg/store/data/file.i.s: size=256
 
-  $ cat .hg/store/journal | tr -s '\000' ' ' | grep data/file | tail -1
-  data/file.i 128
 
 The first file.i entry should match the "Reference size" above.
 The first file.d entry is the temporary record during the split,
 
-The second entry after the split happened. The sum of the second file.d
-and the second file.i entry should match the first file.i entry.
+A "temporary file" entry exist for the split index.
 
   $ cat .hg/store/journal | tr -s '\000' ' ' | grep data/file
   data/file.i 1174
   data/file.d 0
-  data/file.d 1046
-  data/file.i 128
+  $ cat .hg/store/journal.backupfiles | tr -s '\000' ' ' | tr -s '\00' ' '| grep data/file
+   data/file.i data/journal.backup.file.i 0
+   data/file.i.s 0
+
+recover is rolling the split back, the fncache is still valid
+
   $ hg recover
   rolling back interrupted transaction
   (verify step skipped, run `hg verify` to check your repository content)
   $ f -s .hg/store/data/file*
-  .hg/store/data/file.d: size=1046
-  .hg/store/data/file.i: size=128
+  .hg/store/data/file.i: size=1174
   $ hg tip
   changeset:   1:cfa8d6e60429
   tag:         tip
   user:        test
   date:        Thu Jan 01 00:00:00 1970 +0000
   summary:     b
   
   $ hg verify -q
-   warning: revlog 'data/file.d' not in fncache!
-  1 warnings encountered!
-  hint: run "hg debugrebuildfncache" to recover from corrupt fncache
   $ hg debugrebuildfncache --only-data
-  adding data/file.d
-  1 items added, 0 removed from fncache
+  fncache already up to date
   $ hg verify -q
   $ cd ..
 
 Test a hard crash right before the index is move into place
 ===========================================================
 
 Now retry the procedure but intercept the rename of the index and check that
 the journal does not contain the new index size. This demonstrates the edge case
 where the data file is left as garbage.
 
   $ hg clone --quiet --rev 1 troffset-computation troffset-computation-copy2
   $ cd troffset-computation-copy2
 
 Reference size:
   $ f -s file
   file: size=1024
   $ f -s .hg/store/data/file*
   .hg/store/data/file.i: size=1174
 
   $ cat > .hg/hgrc <<EOF
   > [extensions]
   > intercept_rename = $TESTTMP/intercept_before_rename.py
-  > [hooks]
-  > pretxnchangegroup = python:$TESTTMP/killme.py:killme
   > EOF
 #if chg
   $ hg pull ../troffset-computation
   pulling from ../troffset-computation
+  searching for changes
+  adding changesets
+  adding manifests
+  adding file changes
   [255]
 #else
   $ hg pull ../troffset-computation
   pulling from ../troffset-computation
+  searching for changes
+  adding changesets
+  adding manifests
+  adding file changes
   Killed
   [137]
 #endif
 
-The data file is created, but the revlog is still inline
+The inline revlog still exist, but a split version exist next to it
 
   $ f -s .hg/store/data/file*
   .hg/store/data/file.d: size=132139
   .hg/store/data/file.i: size=132395
+  .hg/store/data/file.i.s: size=256
 
   $ cat .hg/store/journal | tr -s '\000' ' ' | grep data/file
   data/file.i 1174
   data/file.d 0
-  data/file.d 1046
+
+recover is rolling the split back, the fncache is still valid
 
   $ hg recover
   rolling back interrupted transaction
   (verify step skipped, run `hg verify` to check your repository content)
   $ f -s .hg/store/data/file*
-  .hg/store/data/file.d: size=1046
   .hg/store/data/file.i: size=1174
   $ hg tip
   changeset:   1:cfa8d6e60429
   tag:         tip
   user:        test
   date:        Thu Jan 01 00:00:00 1970 +0000
   summary:     b
   
   $ hg verify -q
   $ cd ..
 
 Test a hard crash right after the index is move into place
 ===========================================================
 
 Now retry the procedure but intercept the rename of the index.
 
-Things get corrupted /o\
-
   $ hg clone --quiet --rev 1 troffset-computation troffset-computation-crash-after-rename
   $ cd troffset-computation-crash-after-rename
 
 Reference size:
   $ f -s file
   file: size=1024
   $ f -s .hg/store/data/file*
   .hg/store/data/file.i: size=1174
 
   $ cat > .hg/hgrc <<EOF
   > [extensions]
   > intercept_rename = $TESTTMP/intercept_after_rename.py
-  > [hooks]
-  > pretxnchangegroup = python:$TESTTMP/killme.py:killme
   > EOF
 #if chg
   $ hg pull ../troffset-computation
   pulling from ../troffset-computation
+  searching for changes
+  adding changesets
+  adding manifests
+  adding file changes
   [255]
 #else
   $ hg pull ../troffset-computation
   pulling from ../troffset-computation
+  searching for changes
+  adding changesets
+  adding manifests
+  adding file changes
   Killed
   [137]
 #endif
 
-the revlog has been split on disk
+The inline revlog was over written on disk
 
   $ f -s .hg/store/data/file*
   .hg/store/data/file.d: size=132139
   .hg/store/data/file.i: size=256
 
   $ cat .hg/store/journal | tr -s '\000' ' ' | grep data/file
   data/file.i 1174
   data/file.d 0
-  data/file.d 1046
+
+recover is rolling the split back, the fncache is still valid
 
   $ hg recover
   rolling back interrupted transaction
-  abort: attempted to truncate data/file.i to 1174 bytes, but it was already 256 bytes
-  
-  [255]
+  (verify step skipped, run `hg verify` to check your repository content)
   $ f -s .hg/store/data/file*
-  .hg/store/data/file.d: size=1046
-  .hg/store/data/file.i: size=256
+  .hg/store/data/file.i: size=1174
   $ hg tip
   changeset:   1:cfa8d6e60429
   tag:         tip
   user:        test
   date:        Thu Jan 01 00:00:00 1970 +0000
   summary:     b
   
   $ hg verify -q
-  abandoned transaction found - run hg recover
-   warning: revlog 'data/file.d' not in fncache!
-   file@0: data length off by -131093 bytes
-   file@2: unpacking fa1120531cc1: partial read of revlog data/file.d; expected 21 bytes from offset 1046, got 0
-   file@3: unpacking a631378adaa3: partial read of revlog data/file.d; expected 131072 bytes from offset 1067, got -21
-   file@?: rev 2 points to nonexistent changeset 2
-   (expected )
-   file@?: fa1120531cc1 not in manifests
-   file@?: rev 3 points to nonexistent changeset 3
-   (expected )
-   file@?: a631378adaa3 not in manifests
-  not checking dirstate because of previous errors
-  3 warnings encountered!
-  hint: run "hg debugrebuildfncache" to recover from corrupt fncache
-  7 integrity errors encountered!
-  (first damaged changeset appears to be 0)
-  [1]
   $ cd ..
 
 Have the transaction rollback itself without any hard crash
 ===========================================================
 
 
 Repeat the original test but let hg rollback the transaction.
 
   $ hg clone --quiet --rev 1 troffset-computation troffset-computation-copy-rb
   $ cd troffset-computation-copy-rb
   $ cat > .hg/hgrc <<EOF
   > [hooks]
   > pretxnchangegroup = false
   > EOF
   $ hg pull ../troffset-computation
   pulling from ../troffset-computation
   searching for changes
   adding changesets
   adding manifests
   adding file changes
   transaction abort!
   rollback completed
   abort: pretxnchangegroup hook exited with status 1
   [40]
 
-File are still split on disk, with the expected size.
+The split was rollback
 
   $ f -s .hg/store/data/file*
-  .hg/store/data/file.d: size=1046
-  .hg/store/data/file.i: size=128
+  .hg/store/data/file.d: size=0
+  .hg/store/data/file.i: size=1174
+
 
   $ hg tip
   changeset:   1:cfa8d6e60429
   tag:         tip
   user:        test
   date:        Thu Jan 01 00:00:00 1970 +0000
   summary:     b
   
   $ hg verify -q
-   warning: revlog 'data/file.d' not in fncache!
-  1 warnings encountered!
-  hint: run "hg debugrebuildfncache" to recover from corrupt fncache
   $ cd ..
 
 Read race
 =========
 
 We check that a client that started reading a revlog (its index) after the
 split and end reading (the data) after the rollback should be fine
 
   $ hg clone --quiet --rev 1 troffset-computation troffset-computation-race
   $ cd troffset-computation-race
   $ cat > .hg/hgrc <<EOF
   > [hooks]
   > pretxnchangegroup=$RUNTESTDIR/testlib/wait-on-file 5 $TESTTMP/reader-index-read $TESTTMP/writer-revlog-split
   > pretxnclose = false
   > EOF
 
 start a reader
 
   $ hg cat --rev 0 file \
   > --config "extensions.wait_read=$TESTTMP/reader_wait_split.py" \
   > 2> $TESTTMP/reader.stderr \
   > > $TESTTMP/reader.stdout &
 
 Do a failed pull in //
 
   $ hg pull ../troffset-computation
   pulling from ../troffset-computation
   searching for changes
   adding changesets
   adding manifests
   adding file changes
   transaction abort!
   rollback completed
   abort: pretxnclose hook exited with status 1
   [40]
   $ touch $TESTTMP/writer-revlog-unsplit
   $ wait
 
 The reader should be fine
   $ cat $TESTTMP/reader.stderr
   $ cat $TESTTMP/reader.stdout
                      1 (no-eol)
   $ cd ..
 
 pending hooks
 =============
 
 We checks that hooks properly see the inside of the transaction, while other process don't.
 
   $ hg clone --quiet --rev 1 troffset-computation troffset-computation-hooks
   $ cd troffset-computation-hooks
   $ cat > .hg/hgrc <<EOF
   > [hooks]
   > pretxnclose.01-echo = hg cat -r 'max(all())' file | f --size
   > pretxnclose.02-echo = $RUNTESTDIR/testlib/wait-on-file 5 $TESTTMP/hook-done $TESTTMP/hook-tr-ready
   > pretxnclose.03-abort = false
   > EOF
 
   $ (
   >   $RUNTESTDIR/testlib/wait-on-file 5 $TESTTMP/hook-tr-ready;\
   >   hg cat -r 'max(all())' file | f --size;\
   >   touch $TESTTMP/hook-done
   > ) >stdout 2>stderr &
 
   $ hg pull ../troffset-computation
   pulling from ../troffset-computation
   searching for changes
   adding changesets
   adding manifests
   adding file changes
   size=131072
   transaction abort!
   rollback completed
   abort: pretxnclose.03-abort hook exited with status 1
   [40]
 
   $ cat stdout
   size=1024
   $ cat stderr
 
 
   $ cd ..