unbundle: faster computation of changed heads

To compute the set of changed heads it's sufficient to look at the recent commits,
instead of looking at all heads currently in existence.

mercurial/changegroup.py

@@ -1,2442 +1,2441 b''
 # changegroup.py - Mercurial changegroup manipulation functions
 #
 #  Copyright 2006 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 os
 import struct
 import weakref
 
 from .i18n import _
 from .node import (
     hex,
     nullrev,
     short,
 )
 from .pycompat import open
 
 from . import (
     error,
     match as matchmod,
     mdiff,
     phases,
     pycompat,
     requirements,
     scmutil,
     util,
 )
 
 from .interfaces import repository
 from .revlogutils import sidedata as sidedatamod
 from .revlogutils import constants as revlog_constants
 from .utils import storageutil
 
 _CHANGEGROUPV1_DELTA_HEADER = struct.Struct(b"20s20s20s20s")
 _CHANGEGROUPV2_DELTA_HEADER = struct.Struct(b"20s20s20s20s20s")
 _CHANGEGROUPV3_DELTA_HEADER = struct.Struct(b">20s20s20s20s20sH")
 _CHANGEGROUPV4_DELTA_HEADER = struct.Struct(b">B20s20s20s20s20sH")
 
 LFS_REQUIREMENT = b'lfs'
 
 readexactly = util.readexactly
 
 
 def getchunk(stream):
     """return the next chunk from stream as a string"""
     d = readexactly(stream, 4)
     l = struct.unpack(b">l", d)[0]
     if l <= 4:
         if l:
             raise error.Abort(_(b"invalid chunk length %d") % l)
         return b""
     return readexactly(stream, l - 4)
 
 
 def chunkheader(length):
     """return a changegroup chunk header (string)"""
     return struct.pack(b">l", length + 4)
 
 
 def closechunk():
     """return a changegroup chunk header (string) for a zero-length chunk"""
     return struct.pack(b">l", 0)
 
 
 def _fileheader(path):
     """Obtain a changegroup chunk header for a named path."""
     return chunkheader(len(path)) + path
 
 
 def writechunks(ui, chunks, filename, vfs=None):
     """Write chunks to a file and return its filename.
 
     The stream is assumed to be a bundle file.
     Existing files will not be overwritten.
     If no filename is specified, a temporary file is created.
     """
     fh = None
     cleanup = None
     try:
         if filename:
             if vfs:
                 fh = vfs.open(filename, b"wb")
             else:
                 # Increase default buffer size because default is usually
                 # small (4k is common on Linux).
                 fh = open(filename, b"wb", 131072)
         else:
             fd, filename = pycompat.mkstemp(prefix=b"hg-bundle-", suffix=b".hg")
             fh = os.fdopen(fd, "wb")
         cleanup = filename
         for c in chunks:
             fh.write(c)
         cleanup = None
         return filename
     finally:
         if fh is not None:
             fh.close()
         if cleanup is not None:
             if filename and vfs:
                 vfs.unlink(cleanup)
             else:
                 os.unlink(cleanup)
 
 
 def _dbg_ubdl_line(
     ui,
     indent,
     key,
     base_value=None,
     percentage_base=None,
     percentage_key=None,
 ):
     """Print one line of debug_unbundle_debug_info"""
     line = b"DEBUG-UNBUNDLING: "
     line += b' ' * (2 * indent)
     key += b":"
     padding = b''
     if base_value is not None:
         assert len(key) + 1 + (2 * indent) <= _KEY_PART_WIDTH
         line += key.ljust(_KEY_PART_WIDTH - (2 * indent))
         if isinstance(base_value, float):
             line += b"%14.3f seconds" % base_value
         else:
             line += b"%10d" % base_value
             padding = b'            '
     else:
         line += key
 
     if percentage_base is not None:
         line += padding
         padding = b''
         assert base_value is not None
         percentage = base_value * 100 // percentage_base
         if percentage_key is not None:
             line += b" (%3d%% of %s)" % (
                 percentage,
                 percentage_key,
             )
         else:
             line += b" (%3d%%)" % percentage
 
     line += b'\n'
     ui.write_err(line)
 
 
 def _sumf(items):
     # python < 3.8 does not support a `start=0.0` argument to sum
     # So we have to cheat a bit until we drop support for those version
     if not items:
         return 0.0
     return sum(items)
 
 
 def display_unbundle_debug_info(ui, debug_info):
     """display an unbundling report from debug information"""
     cl_info = []
     mn_info = []
     fl_info = []
     _dispatch = [
         (b'CHANGELOG:', cl_info),
         (b'MANIFESTLOG:', mn_info),
         (b'FILELOG:', fl_info),
     ]
     for e in debug_info:
         for prefix, info in _dispatch:
             if e["target-revlog"].startswith(prefix):
                 info.append(e)
                 break
         else:
             assert False, 'unreachable'
     each_info = [
         (b'changelog', cl_info),
         (b'manifests', mn_info),
         (b'files', fl_info),
     ]
 
     # General Revision Countss
     _dbg_ubdl_line(ui, 0, b'revisions', len(debug_info))
     for key, info in each_info:
         if not info:
             continue
         _dbg_ubdl_line(ui, 1, key, len(info), len(debug_info))
 
     # General Time spent
     all_durations = [e['duration'] for e in debug_info]
     all_durations.sort()
     total_duration = _sumf(all_durations)
     _dbg_ubdl_line(ui, 0, b'total-time', total_duration)
 
     for key, info in each_info:
         if not info:
             continue
         durations = [e['duration'] for e in info]
         durations.sort()
         _dbg_ubdl_line(ui, 1, key, _sumf(durations), total_duration)
 
     # Count and cache reuse per delta types
     each_types = {}
     for key, info in each_info:
         each_types[key] = types = {
             b'full': 0,
             b'full-cached': 0,
             b'snapshot': 0,
             b'snapshot-cached': 0,
             b'delta': 0,
             b'delta-cached': 0,
             b'unknown': 0,
             b'unknown-cached': 0,
         }
         for e in info:
             types[e['type']] += 1
             if e['using-cached-base']:
                 types[e['type'] + b'-cached'] += 1
 
     EXPECTED_TYPES = (b'full', b'snapshot', b'delta', b'unknown')
     if debug_info:
         _dbg_ubdl_line(ui, 0, b'type-count')
     for key, info in each_info:
         if info:
             _dbg_ubdl_line(ui, 1, key)
         t = each_types[key]
         for tn in EXPECTED_TYPES:
             if t[tn]:
                 tc = tn + b'-cached'
                 _dbg_ubdl_line(ui, 2, tn, t[tn])
                 _dbg_ubdl_line(ui, 3, b'cached', t[tc], t[tn])
 
     # time perf delta types and reuse
     each_type_time = {}
     for key, info in each_info:
         each_type_time[key] = t = {
             b'full': [],
             b'full-cached': [],
             b'snapshot': [],
             b'snapshot-cached': [],
             b'delta': [],
             b'delta-cached': [],
             b'unknown': [],
             b'unknown-cached': [],
         }
         for e in info:
             t[e['type']].append(e['duration'])
             if e['using-cached-base']:
                 t[e['type'] + b'-cached'].append(e['duration'])
         for t_key, value in list(t.items()):
             value.sort()
             t[t_key] = _sumf(value)
 
     if debug_info:
         _dbg_ubdl_line(ui, 0, b'type-time')
     for key, info in each_info:
         if info:
             _dbg_ubdl_line(ui, 1, key)
         t = each_type_time[key]
         td = total_duration  # to same space on next lines
         for tn in EXPECTED_TYPES:
             if t[tn]:
                 tc = tn + b'-cached'
                 _dbg_ubdl_line(ui, 2, tn, t[tn], td, b"total")
                 _dbg_ubdl_line(ui, 3, b'cached', t[tc], td, b"total")
 
 
 class cg1unpacker:
     """Unpacker for cg1 changegroup streams.
 
     A changegroup unpacker handles the framing of the revision data in
     the wire format. Most consumers will want to use the apply()
     method to add the changes from the changegroup to a repository.
 
     If you're forwarding a changegroup unmodified to another consumer,
     use getchunks(), which returns an iterator of changegroup
     chunks. This is mostly useful for cases where you need to know the
     data stream has ended by observing the end of the changegroup.
 
     deltachunk() is useful only if you're applying delta data. Most
     consumers should prefer apply() instead.
 
     A few other public methods exist. Those are used only for
     bundlerepo and some debug commands - their use is discouraged.
     """
 
     deltaheader = _CHANGEGROUPV1_DELTA_HEADER
     deltaheadersize = deltaheader.size
     version = b'01'
     _grouplistcount = 1  # One list of files after the manifests
 
     def __init__(self, fh, alg, extras=None):
         if alg is None:
             alg = b'UN'
         if alg not in util.compengines.supportedbundletypes:
             raise error.Abort(_(b'unknown stream compression type: %s') % alg)
         if alg == b'BZ':
             alg = b'_truncatedBZ'
 
         compengine = util.compengines.forbundletype(alg)
         self._stream = compengine.decompressorreader(fh)
         self._type = alg
         self.extras = extras or {}
         self.callback = None
 
     # These methods (compressed, read, seek, tell) all appear to only
     # be used by bundlerepo, but it's a little hard to tell.
     def compressed(self):
         return self._type is not None and self._type != b'UN'
 
     def read(self, l):
         return self._stream.read(l)
 
     def seek(self, pos):
         return self._stream.seek(pos)
 
     def tell(self):
         return self._stream.tell()
 
     def close(self):
         return self._stream.close()
 
     def _chunklength(self):
         d = readexactly(self._stream, 4)
         l = struct.unpack(b">l", d)[0]
         if l <= 4:
             if l:
                 raise error.Abort(_(b"invalid chunk length %d") % l)
             return 0
         if self.callback:
             self.callback()
         return l - 4
 
     def changelogheader(self):
         """v10 does not have a changelog header chunk"""
         return {}
 
     def manifestheader(self):
         """v10 does not have a manifest header chunk"""
         return {}
 
     def filelogheader(self):
         """return the header of the filelogs chunk, v10 only has the filename"""
         l = self._chunklength()
         if not l:
             return {}
         fname = readexactly(self._stream, l)
         return {b'filename': fname}
 
     def _deltaheader(self, headertuple, prevnode):
         node, p1, p2, cs = headertuple
         if prevnode is None:
             deltabase = p1
         else:
             deltabase = prevnode
         flags = 0
         protocol_flags = 0
         return node, p1, p2, deltabase, cs, flags, protocol_flags
 
     def deltachunk(self, prevnode):
         # Chunkdata: (node, p1, p2, cs, deltabase, delta, flags, sidedata, proto_flags)
         l = self._chunklength()
         if not l:
             return {}
         headerdata = readexactly(self._stream, self.deltaheadersize)
         header = self.deltaheader.unpack(headerdata)
         delta = readexactly(self._stream, l - self.deltaheadersize)
         header = self._deltaheader(header, prevnode)
         node, p1, p2, deltabase, cs, flags, protocol_flags = header
         return node, p1, p2, cs, deltabase, delta, flags, {}, protocol_flags
 
     def getchunks(self):
         """returns all the chunks contains in the bundle
 
         Used when you need to forward the binary stream to a file or another
         network API. To do so, it parse the changegroup data, otherwise it will
         block in case of sshrepo because it don't know the end of the stream.
         """
         # For changegroup 1 and 2, we expect 3 parts: changelog, manifestlog,
         # and a list of filelogs. For changegroup 3, we expect 4 parts:
         # changelog, manifestlog, a list of tree manifestlogs, and a list of
         # filelogs.
         #
         # Changelog and manifestlog parts are terminated with empty chunks. The
         # tree and file parts are a list of entry sections. Each entry section
         # is a series of chunks terminating in an empty chunk. The list of these
         # entry sections is terminated in yet another empty chunk, so we know
         # we've reached the end of the tree/file list when we reach an empty
         # chunk that was proceeded by no non-empty chunks.
 
         parts = 0
         while parts < 2 + self._grouplistcount:
             noentries = True
             while True:
                 chunk = getchunk(self)
                 if not chunk:
                     # The first two empty chunks represent the end of the
                     # changelog and the manifestlog portions. The remaining
                     # empty chunks represent either A) the end of individual
                     # tree or file entries in the file list, or B) the end of
                     # the entire list. It's the end of the entire list if there
                     # were no entries (i.e. noentries is True).
                     if parts < 2:
                         parts += 1
                     elif noentries:
                         parts += 1
                     break
                 noentries = False
                 yield chunkheader(len(chunk))
                 pos = 0
                 while pos < len(chunk):
                     next = pos + 2 ** 20
                     yield chunk[pos:next]
                     pos = next
             yield closechunk()
 
     def _unpackmanifests(
         self,
         repo,
         revmap,
         trp,
         prog,
         addrevisioncb=None,
         debug_info=None,
         delta_base_reuse_policy=None,
     ):
         self.callback = prog.increment
         # no need to check for empty manifest group here:
         # if the result of the merge of 1 and 2 is the same in 3 and 4,
         # no new manifest will be created and the manifest group will
         # be empty during the pull
         self.manifestheader()
         deltas = self.deltaiter()
         storage = repo.manifestlog.getstorage(b'')
         storage.addgroup(
             deltas,
             revmap,
             trp,
             addrevisioncb=addrevisioncb,
             debug_info=debug_info,
             delta_base_reuse_policy=delta_base_reuse_policy,
         )
         prog.complete()
         self.callback = None
 
     def apply(
         self,
         repo,
         tr,
         srctype,
         url,
         targetphase=phases.draft,
         expectedtotal=None,
         sidedata_categories=None,
         delta_base_reuse_policy=None,
     ):
         """Add the changegroup returned by source.read() to this repo.
         srctype is a string like 'push', 'pull', or 'unbundle'.  url is
         the URL of the repo where this changegroup is coming from.
 
         Return an integer summarizing the change to this repo:
         - nothing changed or no source: 0
         - more heads than before: 1+added heads (2..n)
         - fewer heads than before: -1-removed heads (-2..-n)
         - number of heads stays the same: 1
 
         `sidedata_categories` is an optional set of the remote's sidedata wanted
         categories.
 
         `delta_base_reuse_policy` is an optional argument, when set to a value
         it will control the way the delta contained into the bundle are reused
         when applied in the revlog.
 
         See `DELTA_BASE_REUSE_*` entry in mercurial.revlogutils.constants.
         """
         repo = repo.unfiltered()
 
         debug_info = None
         if repo.ui.configbool(b'debug', b'unbundling-stats'):
             debug_info = []
 
         # Only useful if we're adding sidedata categories. If both peers have
         # the same categories, then we simply don't do anything.
         adding_sidedata = (
             (
                 requirements.REVLOGV2_REQUIREMENT in repo.requirements
                 or requirements.CHANGELOGV2_REQUIREMENT in repo.requirements
             )
             and self.version == b'04'
             and srctype == b'pull'
         )
         if adding_sidedata:
             sidedata_helpers = sidedatamod.get_sidedata_helpers(
                 repo,
                 sidedata_categories or set(),
                 pull=True,
             )
         else:
             sidedata_helpers = None
 
         def csmap(x):
             repo.ui.debug(b"add changeset %s\n" % short(x))
             return len(cl)
 
         def revmap(x):
             return cl.rev(x)
 
         try:
             # The transaction may already carry source information. In this
             # case we use the top level data. We overwrite the argument
             # because we need to use the top level value (if they exist)
             # in this function.
             srctype = tr.hookargs.setdefault(b'source', srctype)
             tr.hookargs.setdefault(b'url', url)
             repo.hook(
                 b'prechangegroup', throw=True, **pycompat.strkwargs(tr.hookargs)
             )
 
             # write changelog data to temp files so concurrent readers
             # will not see an inconsistent view
             cl = repo.changelog
             cl.delayupdate(tr)
-            oldheads = set(cl.heads())
+            oldrevcount = len(cl)
 
             trp = weakref.proxy(tr)
             # pull off the changeset group
             repo.ui.status(_(b"adding changesets\n"))
             clstart = len(cl)
             progress = repo.ui.makeprogress(
                 _(b'changesets'), unit=_(b'chunks'), total=expectedtotal
             )
             self.callback = progress.increment
 
             efilesset = set()
             duprevs = []
 
             def ondupchangelog(cl, rev):
                 if rev < clstart:
                     duprevs.append(rev)  # pytype: disable=attribute-error
 
             def onchangelog(cl, rev):
                 ctx = cl.changelogrevision(rev)
                 assert efilesset is not None  # help pytype
                 efilesset.update(ctx.files)
                 repo.register_changeset(rev, ctx)
 
             self.changelogheader()
             deltas = self.deltaiter()
             if not cl.addgroup(
                 deltas,
                 csmap,
                 trp,
                 alwayscache=True,
                 addrevisioncb=onchangelog,
                 duplicaterevisioncb=ondupchangelog,
                 debug_info=debug_info,
                 delta_base_reuse_policy=delta_base_reuse_policy,
             ):
                 repo.ui.develwarn(
                     b'applied empty changelog from changegroup',
                     config=b'warn-empty-changegroup',
                 )
             efiles = len(efilesset)
             clend = len(cl)
             changesets = clend - clstart
             progress.complete()
             del deltas
             # TODO Python 2.7 removal
             # del efilesset
             efilesset = None
             self.callback = None
 
             # Keep track of the (non-changelog) revlogs we've updated and their
             # range of new revisions for sidedata rewrite.
             # TODO do something more efficient than keeping the reference to
             # the revlogs, especially memory-wise.
             touched_manifests = {}
             touched_filelogs = {}
 
             # pull off the manifest group
             repo.ui.status(_(b"adding manifests\n"))
             # We know that we'll never have more manifests than we had
             # changesets.
             progress = repo.ui.makeprogress(
                 _(b'manifests'), unit=_(b'chunks'), total=changesets
             )
             on_manifest_rev = None
             if sidedata_helpers:
                 if revlog_constants.KIND_MANIFESTLOG in sidedata_helpers[1]:
 
                     def on_manifest_rev(manifest, rev):
                         range = touched_manifests.get(manifest)
                         if not range:
                             touched_manifests[manifest] = (rev, rev)
                         else:
                             assert rev == range[1] + 1
                             touched_manifests[manifest] = (range[0], rev)
 
             self._unpackmanifests(
                 repo,
                 revmap,
                 trp,
                 progress,
                 addrevisioncb=on_manifest_rev,
                 debug_info=debug_info,
                 delta_base_reuse_policy=delta_base_reuse_policy,
             )
 
             needfiles = {}
             if repo.ui.configbool(b'server', b'validate'):
                 cl = repo.changelog
                 ml = repo.manifestlog
                 # validate incoming csets have their manifests
                 for cset in range(clstart, clend):
                     mfnode = cl.changelogrevision(cset).manifest
                     mfest = ml[mfnode].readdelta()
                     # store file nodes we must see
                     for f, n in mfest.items():
                         needfiles.setdefault(f, set()).add(n)
 
             on_filelog_rev = None
             if sidedata_helpers:
                 if revlog_constants.KIND_FILELOG in sidedata_helpers[1]:
 
                     def on_filelog_rev(filelog, rev):
                         range = touched_filelogs.get(filelog)
                         if not range:
                             touched_filelogs[filelog] = (rev, rev)
                         else:
                             assert rev == range[1] + 1
                             touched_filelogs[filelog] = (range[0], rev)
 
             # process the files
             repo.ui.status(_(b"adding file changes\n"))
             newrevs, newfiles = _addchangegroupfiles(
                 repo,
                 self,
                 revmap,
                 trp,
                 efiles,
                 needfiles,
                 addrevisioncb=on_filelog_rev,
                 debug_info=debug_info,
                 delta_base_reuse_policy=delta_base_reuse_policy,
             )
 
             if sidedata_helpers:
                 if revlog_constants.KIND_CHANGELOG in sidedata_helpers[1]:
                     cl.rewrite_sidedata(
                         trp, sidedata_helpers, clstart, clend - 1
                     )
                 for mf, (startrev, endrev) in touched_manifests.items():
                     mf.rewrite_sidedata(trp, sidedata_helpers, startrev, endrev)
                 for fl, (startrev, endrev) in touched_filelogs.items():
                     fl.rewrite_sidedata(trp, sidedata_helpers, startrev, endrev)
 
             # making sure the value exists
             tr.changes.setdefault(b'changegroup-count-changesets', 0)
             tr.changes.setdefault(b'changegroup-count-revisions', 0)
             tr.changes.setdefault(b'changegroup-count-files', 0)
             tr.changes.setdefault(b'changegroup-count-heads', 0)
 
             # some code use bundle operation for internal purpose. They usually
             # set `ui.quiet` to do this outside of user sight. Size the report
             # of such operation now happens at the end of the transaction, that
             # ui.quiet has not direct effect on the output.
             #
             # To preserve this intend use an inelegant hack, we fail to report
             # the change if `quiet` is set. We should probably move to
             # something better, but this is a good first step to allow the "end
             # of transaction report" to pass tests.
             if not repo.ui.quiet:
                 tr.changes[b'changegroup-count-changesets'] += changesets
                 tr.changes[b'changegroup-count-revisions'] += newrevs
                 tr.changes[b'changegroup-count-files'] += newfiles
 
             deltaheads = 0
-            if oldheads:
-                heads = cl.heads()
-                deltaheads += len(heads) - len(oldheads)
-                for h in heads:
-                    if h not in oldheads and repo[h].closesbranch():
-                        deltaheads -= 1
+            newrevcount = len(cl)
+            heads_removed, heads_added = cl.diffheads(oldrevcount, newrevcount)
+            deltaheads += len(heads_added) - len(heads_removed)
+            for h in heads_added:
+                if repo[h].closesbranch():
+                    deltaheads -= 1
 
             # see previous comment about checking ui.quiet
             if not repo.ui.quiet:
                 tr.changes[b'changegroup-count-heads'] += deltaheads
             repo.invalidatevolatilesets()
 
             if changesets > 0:
                 if b'node' not in tr.hookargs:
                     tr.hookargs[b'node'] = hex(cl.node(clstart))
                     tr.hookargs[b'node_last'] = hex(cl.node(clend - 1))
                     hookargs = dict(tr.hookargs)
                 else:
                     hookargs = dict(tr.hookargs)
                     hookargs[b'node'] = hex(cl.node(clstart))
                     hookargs[b'node_last'] = hex(cl.node(clend - 1))
                 repo.hook(
                     b'pretxnchangegroup',
                     throw=True,
                     **pycompat.strkwargs(hookargs)
                 )
 
             added = range(clstart, clend)
             phaseall = None
             if srctype in (b'push', b'serve'):
                 # Old servers can not push the boundary themselves.
                 # New servers won't push the boundary if changeset already
                 # exists locally as secret
                 #
                 # We should not use added here but the list of all change in
                 # the bundle
                 if repo.publishing():
                     targetphase = phaseall = phases.public
                 else:
                     # closer target phase computation
 
                     # Those changesets have been pushed from the
                     # outside, their phases are going to be pushed
                     # alongside. Therefor `targetphase` is
                     # ignored.
                     targetphase = phaseall = phases.draft
             if added:
                 phases.registernew(repo, tr, targetphase, added)
             if phaseall is not None:
                 if duprevs:
                     duprevs.extend(added)
                 else:
                     duprevs = added
                 phases.advanceboundary(repo, tr, phaseall, [], revs=duprevs)
                 duprevs = []
 
             if changesets > 0:
 
                 def runhooks(unused_success):
                     # These hooks run when the lock releases, not when the
                     # transaction closes. So it's possible for the changelog
                     # to have changed since we last saw it.
                     if clstart >= len(repo):
                         return
 
                     repo.hook(b"changegroup", **pycompat.strkwargs(hookargs))
 
                     for rev in added:
                         args = hookargs.copy()
                         args[b'node'] = hex(cl.node(rev))
                         del args[b'node_last']
                         repo.hook(b"incoming", **pycompat.strkwargs(args))
 
-                    newheads = [h for h in repo.heads() if h not in oldheads]
                     repo.ui.log(
                         b"incoming",
                         b"%d incoming changes - new heads: %s\n",
                         len(added),
-                        b', '.join([hex(c[:6]) for c in newheads]),
+                        b', '.join([hex(c[:6]) for c in heads_added]),
                     )
 
                 tr.addpostclose(
                     b'changegroup-runhooks-%020i' % clstart,
                     lambda tr: repo._afterlock(runhooks),
                 )
             if debug_info is not None:
                 display_unbundle_debug_info(repo.ui, debug_info)
         finally:
             repo.ui.flush()
         # never return 0 here:
         if deltaheads < 0:
             ret = deltaheads - 1
         else:
             ret = deltaheads + 1
         return ret
 
     def deltaiter(self):
         """
         returns an iterator of the deltas in this changegroup
 
         Useful for passing to the underlying storage system to be stored.
         """
         chain = None
         for chunkdata in iter(lambda: self.deltachunk(chain), {}):
             # Chunkdata: (node, p1, p2, cs, deltabase, delta, flags, sidedata, proto_flags)
             yield chunkdata[:8]
             chain = chunkdata[0]
 
 
 class cg2unpacker(cg1unpacker):
     """Unpacker for cg2 streams.
 
     cg2 streams add support for generaldelta, so the delta header
     format is slightly different. All other features about the data
     remain the same.
     """
 
     deltaheader = _CHANGEGROUPV2_DELTA_HEADER
     deltaheadersize = deltaheader.size
     version = b'02'
 
     def _deltaheader(self, headertuple, prevnode):
         node, p1, p2, deltabase, cs = headertuple
         flags = 0
         protocol_flags = 0
         return node, p1, p2, deltabase, cs, flags, protocol_flags
 
 
 class cg3unpacker(cg2unpacker):
     """Unpacker for cg3 streams.
 
     cg3 streams add support for exchanging treemanifests and revlog
     flags. It adds the revlog flags to the delta header and an empty chunk
     separating manifests and files.
     """
 
     deltaheader = _CHANGEGROUPV3_DELTA_HEADER
     deltaheadersize = deltaheader.size
     version = b'03'
     _grouplistcount = 2  # One list of manifests and one list of files
 
     def _deltaheader(self, headertuple, prevnode):
         node, p1, p2, deltabase, cs, flags = headertuple
         protocol_flags = 0
         return node, p1, p2, deltabase, cs, flags, protocol_flags
 
     def _unpackmanifests(
         self,
         repo,
         revmap,
         trp,
         prog,
         addrevisioncb=None,
         debug_info=None,
         delta_base_reuse_policy=None,
     ):
         super(cg3unpacker, self)._unpackmanifests(
             repo,
             revmap,
             trp,
             prog,
             addrevisioncb=addrevisioncb,
             debug_info=debug_info,
             delta_base_reuse_policy=delta_base_reuse_policy,
         )
         for chunkdata in iter(self.filelogheader, {}):
             # If we get here, there are directory manifests in the changegroup
             d = chunkdata[b"filename"]
             repo.ui.debug(b"adding %s revisions\n" % d)
             deltas = self.deltaiter()
             if not repo.manifestlog.getstorage(d).addgroup(
                 deltas,
                 revmap,
                 trp,
                 addrevisioncb=addrevisioncb,
                 debug_info=debug_info,
                 delta_base_reuse_policy=delta_base_reuse_policy,
             ):
                 raise error.Abort(_(b"received dir revlog group is empty"))
 
 
 class cg4unpacker(cg3unpacker):
     """Unpacker for cg4 streams.
 
     cg4 streams add support for exchanging sidedata.
     """
 
     deltaheader = _CHANGEGROUPV4_DELTA_HEADER
     deltaheadersize = deltaheader.size
     version = b'04'
 
     def _deltaheader(self, headertuple, prevnode):
         protocol_flags, node, p1, p2, deltabase, cs, flags = headertuple
         return node, p1, p2, deltabase, cs, flags, protocol_flags
 
     def deltachunk(self, prevnode):
         res = super(cg4unpacker, self).deltachunk(prevnode)
         if not res:
             return res
 
         (
             node,
             p1,
             p2,
             cs,
             deltabase,
             delta,
             flags,
             sidedata,
             protocol_flags,
         ) = res
         assert not sidedata
 
         sidedata = {}
         if protocol_flags & storageutil.CG_FLAG_SIDEDATA:
             sidedata_raw = getchunk(self._stream)
             sidedata = sidedatamod.deserialize_sidedata(sidedata_raw)
 
         return (
             node,
             p1,
             p2,
             cs,
             deltabase,
             delta,
             flags,
             sidedata,
             protocol_flags,
         )
 
 
 class headerlessfixup:
     def __init__(self, fh, h):
         self._h = h
         self._fh = fh
 
     def read(self, n):
         if self._h:
             d, self._h = self._h[:n], self._h[n:]
             if len(d) < n:
                 d += readexactly(self._fh, n - len(d))
             return d
         return readexactly(self._fh, n)
 
 
 def _revisiondeltatochunks(repo, delta, headerfn):
     """Serialize a revisiondelta to changegroup chunks."""
 
     # The captured revision delta may be encoded as a delta against
     # a base revision or as a full revision. The changegroup format
     # requires that everything on the wire be deltas. So for full
     # revisions, we need to invent a header that says to rewrite
     # data.
 
     if delta.delta is not None:
         prefix, data = b'', delta.delta
     elif delta.basenode == repo.nullid:
         data = delta.revision
         prefix = mdiff.trivialdiffheader(len(data))
     else:
         data = delta.revision
         prefix = mdiff.replacediffheader(delta.baserevisionsize, len(data))
 
     meta = headerfn(delta)
 
     yield chunkheader(len(meta) + len(prefix) + len(data))
     yield meta
     if prefix:
         yield prefix
     yield data
 
     if delta.protocol_flags & storageutil.CG_FLAG_SIDEDATA:
         # Need a separate chunk for sidedata to be able to differentiate
         # "raw delta" length and sidedata length
         sidedata = delta.sidedata
         yield chunkheader(len(sidedata))
         yield sidedata
 
 
 def _sortnodesellipsis(store, nodes, cl, lookup):
     """Sort nodes for changegroup generation."""
     # Ellipses serving mode.
     #
     # In a perfect world, we'd generate better ellipsis-ified graphs
     # for non-changelog revlogs. In practice, we haven't started doing
     # that yet, so the resulting DAGs for the manifestlog and filelogs
     # are actually full of bogus parentage on all the ellipsis
     # nodes. This has the side effect that, while the contents are
     # correct, the individual DAGs might be completely out of whack in
     # a case like 882681bc3166 and its ancestors (back about 10
     # revisions or so) in the main hg repo.
     #
     # The one invariant we *know* holds is that the new (potentially
     # bogus) DAG shape will be valid if we order the nodes in the
     # order that they're introduced in dramatis personae by the
     # changelog, so what we do is we sort the non-changelog histories
     # by the order in which they are used by the changelog.
     key = lambda n: cl.rev(lookup(n))
     return sorted(nodes, key=key)
 
 
 def _resolvenarrowrevisioninfo(
     cl,
     store,
     ischangelog,
     rev,
     linkrev,
     linknode,
     clrevtolocalrev,
     fullclnodes,
     precomputedellipsis,
 ):
     linkparents = precomputedellipsis[linkrev]
 
     def local(clrev):
         """Turn a changelog revnum into a local revnum.
 
         The ellipsis dag is stored as revnums on the changelog,
         but when we're producing ellipsis entries for
         non-changelog revlogs, we need to turn those numbers into
         something local. This does that for us, and during the
         changelog sending phase will also expand the stored
         mappings as needed.
         """
         if clrev == nullrev:
             return nullrev
 
         if ischangelog:
             return clrev
 
         # Walk the ellipsis-ized changelog breadth-first looking for a
         # change that has been linked from the current revlog.
         #
         # For a flat manifest revlog only a single step should be necessary
         # as all relevant changelog entries are relevant to the flat
         # manifest.
         #
         # For a filelog or tree manifest dirlog however not every changelog
         # entry will have been relevant, so we need to skip some changelog
         # nodes even after ellipsis-izing.
         walk = [clrev]
         while walk:
             p = walk[0]
             walk = walk[1:]
             if p in clrevtolocalrev:
                 return clrevtolocalrev[p]
             elif p in fullclnodes:
                 walk.extend([pp for pp in cl.parentrevs(p) if pp != nullrev])
             elif p in precomputedellipsis:
                 walk.extend(
                     [pp for pp in precomputedellipsis[p] if pp != nullrev]
                 )
             else:
                 # In this case, we've got an ellipsis with parents
                 # outside the current bundle (likely an
                 # incremental pull). We "know" that we can use the
                 # value of this same revlog at whatever revision
                 # is pointed to by linknode. "Know" is in scare
                 # quotes because I haven't done enough examination
                 # of edge cases to convince myself this is really
                 # a fact - it works for all the (admittedly
                 # thorough) cases in our testsuite, but I would be
                 # somewhat unsurprised to find a case in the wild
                 # where this breaks down a bit. That said, I don't
                 # know if it would hurt anything.
                 for i in range(rev, 0, -1):
                     if store.linkrev(i) == clrev:
                         return i
                 # We failed to resolve a parent for this node, so
                 # we crash the changegroup construction.
                 if hasattr(store, 'target'):
                     target = store.display_id
                 else:
                     # some revlog not actually a revlog
                     target = store._revlog.display_id
 
                 raise error.Abort(
                     b"unable to resolve parent while packing '%s' %r"
                     b' for changeset %r' % (target, rev, clrev)
                 )
 
         return nullrev
 
     if not linkparents or (store.parentrevs(rev) == (nullrev, nullrev)):
         p1, p2 = nullrev, nullrev
     elif len(linkparents) == 1:
         (p1,) = sorted(local(p) for p in linkparents)
         p2 = nullrev
     else:
         p1, p2 = sorted(local(p) for p in linkparents)
 
     p1node, p2node = store.node(p1), store.node(p2)
 
     return p1node, p2node, linknode
 
 
 def deltagroup(
     repo,
     store,
     nodes,
     ischangelog,
     lookup,
     forcedeltaparentprev,
     topic=None,
     ellipses=False,
     clrevtolocalrev=None,
     fullclnodes=None,
     precomputedellipsis=None,
     sidedata_helpers=None,
     debug_info=None,
 ):
     """Calculate deltas for a set of revisions.
 
     Is a generator of ``revisiondelta`` instances.
 
     If topic is not None, progress detail will be generated using this
     topic name (e.g. changesets, manifests, etc).
 
     See `revlogutil.sidedata.get_sidedata_helpers` for the doc on
     `sidedata_helpers`.
     """
     if not nodes:
         return
 
     cl = repo.changelog
 
     if ischangelog:
         # `hg log` shows changesets in storage order. To preserve order
         # across clones, send out changesets in storage order.
         nodesorder = b'storage'
     elif ellipses:
         nodes = _sortnodesellipsis(store, nodes, cl, lookup)
         nodesorder = b'nodes'
     else:
         nodesorder = None
 
     # Perform ellipses filtering and revision massaging. We do this before
     # emitrevisions() because a) filtering out revisions creates less work
     # for emitrevisions() b) dropping revisions would break emitrevisions()'s
     # assumptions about delta choices and we would possibly send a delta
     # referencing a missing base revision.
     #
     # Also, calling lookup() has side-effects with regards to populating
     # data structures. If we don't call lookup() for each node or if we call
     # lookup() after the first pass through each node, things can break -
     # possibly intermittently depending on the python hash seed! For that
     # reason, we store a mapping of all linknodes during the initial node
     # pass rather than use lookup() on the output side.
     if ellipses:
         filtered = []
         adjustedparents = {}
         linknodes = {}
 
         for node in nodes:
             rev = store.rev(node)
             linknode = lookup(node)
             linkrev = cl.rev(linknode)
             clrevtolocalrev[linkrev] = rev
 
             # If linknode is in fullclnodes, it means the corresponding
             # changeset was a full changeset and is being sent unaltered.
             if linknode in fullclnodes:
                 linknodes[node] = linknode
 
             # If the corresponding changeset wasn't in the set computed
             # as relevant to us, it should be dropped outright.
             elif linkrev not in precomputedellipsis:
                 continue
 
             else:
                 # We could probably do this later and avoid the dict
                 # holding state. But it likely doesn't matter.
                 p1node, p2node, linknode = _resolvenarrowrevisioninfo(
                     cl,
                     store,
                     ischangelog,
                     rev,
                     linkrev,
                     linknode,
                     clrevtolocalrev,
                     fullclnodes,
                     precomputedellipsis,
                 )
 
                 adjustedparents[node] = (p1node, p2node)
                 linknodes[node] = linknode
 
             filtered.append(node)
 
         nodes = filtered
 
     # We expect the first pass to be fast, so we only engage the progress
     # meter for constructing the revision deltas.
     progress = None
     if topic is not None:
         progress = repo.ui.makeprogress(
             topic, unit=_(b'chunks'), total=len(nodes)
         )
 
     configtarget = repo.ui.config(b'devel', b'bundle.delta')
     if configtarget not in (b'', b'p1', b'full'):
         msg = _(b"""config "devel.bundle.delta" as unknown value: %s""")
         repo.ui.warn(msg % configtarget)
 
     deltamode = repository.CG_DELTAMODE_STD
     if forcedeltaparentprev:
         deltamode = repository.CG_DELTAMODE_PREV
     elif configtarget == b'p1':
         deltamode = repository.CG_DELTAMODE_P1
     elif configtarget == b'full':
         deltamode = repository.CG_DELTAMODE_FULL
 
     revisions = store.emitrevisions(
         nodes,
         nodesorder=nodesorder,
         revisiondata=True,
         assumehaveparentrevisions=not ellipses,
         deltamode=deltamode,
         sidedata_helpers=sidedata_helpers,
         debug_info=debug_info,
     )
 
     for i, revision in enumerate(revisions):
         if progress:
             progress.update(i + 1)
 
         if ellipses:
             linknode = linknodes[revision.node]
 
             if revision.node in adjustedparents:
                 p1node, p2node = adjustedparents[revision.node]
                 revision.p1node = p1node
                 revision.p2node = p2node
                 revision.flags |= repository.REVISION_FLAG_ELLIPSIS
 
         else:
             linknode = lookup(revision.node)
 
         revision.linknode = linknode
         yield revision
 
     if progress:
         progress.complete()
 
 
 def make_debug_info():
     """ "build a "new" debug_info dictionnary
 
     That dictionnary can be used to gather information about the bundle process
     """
     return {
         'revision-total': 0,
         'revision-changelog': 0,
         'revision-manifest': 0,
         'revision-files': 0,
         'file-count': 0,
         'merge-total': 0,
         'available-delta': 0,
         'available-full': 0,
         'delta-against-prev': 0,
         'delta-full': 0,
         'delta-against-p1': 0,
         'denied-delta-candeltafn': 0,
         'denied-base-not-available': 0,
         'reused-storage-delta': 0,
         'computed-delta': 0,
     }
 
 
 def merge_debug_info(base, other):
     """merge the debug information from <other> into <base>
 
     This function can be used to gather lower level information into higher level ones.
     """
     for key in (
         'revision-total',
         'revision-changelog',
         'revision-manifest',
         'revision-files',
         'merge-total',
         'available-delta',
         'available-full',
         'delta-against-prev',
         'delta-full',
         'delta-against-p1',
         'denied-delta-candeltafn',
         'denied-base-not-available',
         'reused-storage-delta',
         'computed-delta',
     ):
         base[key] += other[key]
 
 
 _KEY_PART_WIDTH = 17
 
 
 def _dbg_bdl_line(
     ui,
     indent,
     key,
     base_value=None,
     percentage_base=None,
     percentage_key=None,
     percentage_ref=None,
     extra=None,
 ):
     """Print one line of debug_bundle_debug_info"""
     line = b"DEBUG-BUNDLING: "
     line += b' ' * (2 * indent)
     key += b":"
     if base_value is not None:
         assert len(key) + 1 + (2 * indent) <= _KEY_PART_WIDTH
         line += key.ljust(_KEY_PART_WIDTH - (2 * indent))
         line += b"%10d" % base_value
     else:
         line += key
 
     if percentage_base is not None:
         assert base_value is not None
         percentage = base_value * 100 // percentage_base
         if percentage_key is not None:
             line += b" (%d%% of %s %d)" % (
                 percentage,
                 percentage_key,
                 percentage_ref,
             )
         else:
             line += b" (%d%%)" % percentage
 
     if extra:
         line += b" "
         line += extra
 
     line += b'\n'
     ui.write_err(line)
 
 
 def display_bundling_debug_info(
     ui,
     debug_info,
     cl_debug_info,
     mn_debug_info,
     fl_debug_info,
 ):
     """display debug information gathered during a bundling through `ui`"""
     d = debug_info
     c = cl_debug_info
     m = mn_debug_info
     f = fl_debug_info
     all_info = [
         (b"changelog", b"cl", c),
         (b"manifests", b"mn", m),
         (b"files", b"fl", f),
     ]
     _dbg_bdl_line(ui, 0, b'revisions', d['revision-total'])
     _dbg_bdl_line(ui, 1, b'changelog', d['revision-changelog'])
     _dbg_bdl_line(ui, 1, b'manifest', d['revision-manifest'])
     extra = b'(for %d revlogs)' % d['file-count']
     _dbg_bdl_line(ui, 1, b'files', d['revision-files'], extra=extra)
     if d['merge-total']:
         _dbg_bdl_line(ui, 1, b'merge', d['merge-total'], d['revision-total'])
     for k, __, v in all_info:
         if v['merge-total']:
             _dbg_bdl_line(ui, 2, k, v['merge-total'], v['revision-total'])
 
     _dbg_bdl_line(ui, 0, b'deltas')
     _dbg_bdl_line(
         ui,
         1,
         b'from-storage',
         d['reused-storage-delta'],
         percentage_base=d['available-delta'],
         percentage_key=b"available",
         percentage_ref=d['available-delta'],
     )
 
     if d['denied-delta-candeltafn']:
         _dbg_bdl_line(ui, 2, b'denied-fn', d['denied-delta-candeltafn'])
     for __, k, v in all_info:
         if v['denied-delta-candeltafn']:
             _dbg_bdl_line(ui, 3, k, v['denied-delta-candeltafn'])
 
     if d['denied-base-not-available']:
         _dbg_bdl_line(ui, 2, b'denied-nb', d['denied-base-not-available'])
     for k, __, v in all_info:
         if v['denied-base-not-available']:
             _dbg_bdl_line(ui, 3, k, v['denied-base-not-available'])
 
     if d['computed-delta']:
         _dbg_bdl_line(ui, 1, b'computed', d['computed-delta'])
 
     if d['available-full']:
         _dbg_bdl_line(
             ui,
             2,
             b'full',
             d['delta-full'],
             percentage_base=d['available-full'],
             percentage_key=b"native",
             percentage_ref=d['available-full'],
         )
     for k, __, v in all_info:
         if v['available-full']:
             _dbg_bdl_line(
                 ui,
                 3,
                 k,
                 v['delta-full'],
                 percentage_base=v['available-full'],
                 percentage_key=b"native",
                 percentage_ref=v['available-full'],
             )
 
     if d['delta-against-prev']:
         _dbg_bdl_line(ui, 2, b'previous', d['delta-against-prev'])
     for k, __, v in all_info:
         if v['delta-against-prev']:
             _dbg_bdl_line(ui, 3, k, v['delta-against-prev'])
 
     if d['delta-against-p1']:
         _dbg_bdl_line(ui, 2, b'parent-1', d['delta-against-prev'])
     for k, __, v in all_info:
         if v['delta-against-p1']:
             _dbg_bdl_line(ui, 3, k, v['delta-against-p1'])
 
 
 class cgpacker:
     def __init__(
         self,
         repo,
         oldmatcher,
         matcher,
         version,
         builddeltaheader,
         manifestsend,
         forcedeltaparentprev=False,
         bundlecaps=None,
         ellipses=False,
         shallow=False,
         ellipsisroots=None,
         fullnodes=None,
         remote_sidedata=None,
     ):
         """Given a source repo, construct a bundler.
 
         oldmatcher is a matcher that matches on files the client already has.
         These will not be included in the changegroup.
 
         matcher is a matcher that matches on files to include in the
         changegroup. Used to facilitate sparse changegroups.
 
         forcedeltaparentprev indicates whether delta parents must be against
         the previous revision in a delta group. This should only be used for
         compatibility with changegroup version 1.
 
         builddeltaheader is a callable that constructs the header for a group
         delta.
 
         manifestsend is a chunk to send after manifests have been fully emitted.
 
         ellipses indicates whether ellipsis serving mode is enabled.
 
         bundlecaps is optional and can be used to specify the set of
         capabilities which can be used to build the bundle. While bundlecaps is
         unused in core Mercurial, extensions rely on this feature to communicate
         capabilities to customize the changegroup packer.
 
         shallow indicates whether shallow data might be sent. The packer may
         need to pack file contents not introduced by the changes being packed.
 
         fullnodes is the set of changelog nodes which should not be ellipsis
         nodes. We store this rather than the set of nodes that should be
         ellipsis because for very large histories we expect this to be
         significantly smaller.
 
         remote_sidedata is the set of sidedata categories wanted by the remote.
         """
         assert oldmatcher
         assert matcher
         self._oldmatcher = oldmatcher
         self._matcher = matcher
 
         self.version = version
         self._forcedeltaparentprev = forcedeltaparentprev
         self._builddeltaheader = builddeltaheader
         self._manifestsend = manifestsend
         self._ellipses = ellipses
 
         # Set of capabilities we can use to build the bundle.
         if bundlecaps is None:
             bundlecaps = set()
         self._bundlecaps = bundlecaps
         if remote_sidedata is None:
             remote_sidedata = set()
         self._remote_sidedata = remote_sidedata
         self._isshallow = shallow
         self._fullclnodes = fullnodes
 
         # Maps ellipsis revs to their roots at the changelog level.
         self._precomputedellipsis = ellipsisroots
 
         self._repo = repo
 
         if self._repo.ui.verbose and not self._repo.ui.debugflag:
             self._verbosenote = self._repo.ui.note
         else:
             self._verbosenote = lambda s: None
 
     def generate(
         self,
         commonrevs,
         clnodes,
         fastpathlinkrev,
         source,
         changelog=True,
     ):
         """Yield a sequence of changegroup byte chunks.
         If changelog is False, changelog data won't be added to changegroup
         """
 
         debug_info = None
         repo = self._repo
         if repo.ui.configbool(b'debug', b'bundling-stats'):
             debug_info = make_debug_info()
         cl = repo.changelog
 
         self._verbosenote(_(b'uncompressed size of bundle content:\n'))
         size = 0
 
         sidedata_helpers = None
         if self.version == b'04':
             remote_sidedata = self._remote_sidedata
             if source == b'strip':
                 # We're our own remote when stripping, get the no-op helpers
                 # TODO a better approach would be for the strip bundle to
                 # correctly advertise its sidedata categories directly.
                 remote_sidedata = repo._wanted_sidedata
             sidedata_helpers = sidedatamod.get_sidedata_helpers(
                 repo,
                 remote_sidedata,
             )
 
         cl_debug_info = None
         if debug_info is not None:
             cl_debug_info = make_debug_info()
         clstate, deltas = self._generatechangelog(
             cl,
             clnodes,
             generate=changelog,
             sidedata_helpers=sidedata_helpers,
             debug_info=cl_debug_info,
         )
         for delta in deltas:
             for chunk in _revisiondeltatochunks(
                 self._repo, delta, self._builddeltaheader
             ):
                 size += len(chunk)
                 yield chunk
 
         close = closechunk()
         size += len(close)
         yield closechunk()
         if debug_info is not None:
             merge_debug_info(debug_info, cl_debug_info)
             debug_info['revision-changelog'] = cl_debug_info['revision-total']
 
         self._verbosenote(_(b'%8.i (changelog)\n') % size)
 
         clrevorder = clstate[b'clrevorder']
         manifests = clstate[b'manifests']
         changedfiles = clstate[b'changedfiles']
 
         if debug_info is not None:
             debug_info['file-count'] = len(changedfiles)
 
         # We need to make sure that the linkrev in the changegroup refers to
         # the first changeset that introduced the manifest or file revision.
         # The fastpath is usually safer than the slowpath, because the filelogs
         # are walked in revlog order.
         #
         # When taking the slowpath when the manifest revlog uses generaldelta,
         # the manifest may be walked in the "wrong" order. Without 'clrevorder',
         # we would get an incorrect linkrev (see fix in cc0ff93d0c0c).
         #
         # When taking the fastpath, we are only vulnerable to reordering
         # of the changelog itself. The changelog never uses generaldelta and is
         # never reordered. To handle this case, we simply take the slowpath,
         # which already has the 'clrevorder' logic. This was also fixed in
         # cc0ff93d0c0c.
 
         # Treemanifests don't work correctly with fastpathlinkrev
         # either, because we don't discover which directory nodes to
         # send along with files. This could probably be fixed.
         fastpathlinkrev = fastpathlinkrev and not scmutil.istreemanifest(repo)
 
         fnodes = {}  # needed file nodes
 
         size = 0
         mn_debug_info = None
         if debug_info is not None:
             mn_debug_info = make_debug_info()
         it = self.generatemanifests(
             commonrevs,
             clrevorder,
             fastpathlinkrev,
             manifests,
             fnodes,
             source,
             clstate[b'clrevtomanifestrev'],
             sidedata_helpers=sidedata_helpers,
             debug_info=mn_debug_info,
         )
 
         for tree, deltas in it:
             if tree:
                 assert self.version in (b'03', b'04')
                 chunk = _fileheader(tree)
                 size += len(chunk)
                 yield chunk
 
             for delta in deltas:
                 chunks = _revisiondeltatochunks(
                     self._repo, delta, self._builddeltaheader
                 )
                 for chunk in chunks:
                     size += len(chunk)
                     yield chunk
 
             close = closechunk()
             size += len(close)
             yield close
         if debug_info is not None:
             merge_debug_info(debug_info, mn_debug_info)
             debug_info['revision-manifest'] = mn_debug_info['revision-total']
 
         self._verbosenote(_(b'%8.i (manifests)\n') % size)
         yield self._manifestsend
 
         mfdicts = None
         if self._ellipses and self._isshallow:
             mfdicts = [
                 (repo.manifestlog[n].read(), lr)
                 for (n, lr) in pycompat.iteritems(manifests)
             ]
 
         manifests.clear()
         clrevs = {cl.rev(x) for x in clnodes}
 
         fl_debug_info = None
         if debug_info is not None:
             fl_debug_info = make_debug_info()
         it = self.generatefiles(
             changedfiles,
             commonrevs,
             source,
             mfdicts,
             fastpathlinkrev,
             fnodes,
             clrevs,
             sidedata_helpers=sidedata_helpers,
             debug_info=fl_debug_info,
         )
 
         for path, deltas in it:
             h = _fileheader(path)
             size = len(h)
             yield h
 
             for delta in deltas:
                 chunks = _revisiondeltatochunks(
                     self._repo, delta, self._builddeltaheader
                 )
                 for chunk in chunks:
                     size += len(chunk)
                     yield chunk
 
             close = closechunk()
             size += len(close)
             yield close
 
             self._verbosenote(_(b'%8.i  %s\n') % (size, path))
 
         yield closechunk()
         if debug_info is not None:
             merge_debug_info(debug_info, fl_debug_info)
             debug_info['revision-files'] = fl_debug_info['revision-total']
 
         if debug_info is not None:
             display_bundling_debug_info(
                 repo.ui,
                 debug_info,
                 cl_debug_info,
                 mn_debug_info,
                 fl_debug_info,
             )
 
         if clnodes:
             repo.hook(b'outgoing', node=hex(clnodes[0]), source=source)
 
     def _generatechangelog(
         self,
         cl,
         nodes,
         generate=True,
         sidedata_helpers=None,
         debug_info=None,
     ):
         """Generate data for changelog chunks.
 
         Returns a 2-tuple of a dict containing state and an iterable of
         byte chunks. The state will not be fully populated until the
         chunk stream has been fully consumed.
 
         if generate is False, the state will be fully populated and no chunk
         stream will be yielded
 
         See `revlogutil.sidedata.get_sidedata_helpers` for the doc on
         `sidedata_helpers`.
         """
         clrevorder = {}
         manifests = {}
         mfl = self._repo.manifestlog
         changedfiles = set()
         clrevtomanifestrev = {}
 
         state = {
             b'clrevorder': clrevorder,
             b'manifests': manifests,
             b'changedfiles': changedfiles,
             b'clrevtomanifestrev': clrevtomanifestrev,
         }
 
         if not (generate or self._ellipses):
             # sort the nodes in storage order
             nodes = sorted(nodes, key=cl.rev)
             for node in nodes:
                 c = cl.changelogrevision(node)
                 clrevorder[node] = len(clrevorder)
                 # record the first changeset introducing this manifest version
                 manifests.setdefault(c.manifest, node)
                 # Record a complete list of potentially-changed files in
                 # this manifest.
                 changedfiles.update(c.files)
 
             return state, ()
 
         # Callback for the changelog, used to collect changed files and
         # manifest nodes.
         # Returns the linkrev node (identity in the changelog case).
         def lookupcl(x):
             c = cl.changelogrevision(x)
             clrevorder[x] = len(clrevorder)
 
             if self._ellipses:
                 # Only update manifests if x is going to be sent. Otherwise we
                 # end up with bogus linkrevs specified for manifests and
                 # we skip some manifest nodes that we should otherwise
                 # have sent.
                 if (
                     x in self._fullclnodes
                     or cl.rev(x) in self._precomputedellipsis
                 ):
-
                     manifestnode = c.manifest
                     # Record the first changeset introducing this manifest
                     # version.
                     manifests.setdefault(manifestnode, x)
                     # Set this narrow-specific dict so we have the lowest
                     # manifest revnum to look up for this cl revnum. (Part of
                     # mapping changelog ellipsis parents to manifest ellipsis
                     # parents)
                     clrevtomanifestrev.setdefault(
                         cl.rev(x), mfl.rev(manifestnode)
                     )
                 # We can't trust the changed files list in the changeset if the
                 # client requested a shallow clone.
                 if self._isshallow:
                     changedfiles.update(mfl[c.manifest].read().keys())
                 else:
                     changedfiles.update(c.files)
             else:
                 # record the first changeset introducing this manifest version
                 manifests.setdefault(c.manifest, x)
                 # Record a complete list of potentially-changed files in
                 # this manifest.
                 changedfiles.update(c.files)
 
             return x
 
         gen = deltagroup(
             self._repo,
             cl,
             nodes,
             True,
             lookupcl,
             self._forcedeltaparentprev,
             ellipses=self._ellipses,
             topic=_(b'changesets'),
             clrevtolocalrev={},
             fullclnodes=self._fullclnodes,
             precomputedellipsis=self._precomputedellipsis,
             sidedata_helpers=sidedata_helpers,
             debug_info=debug_info,
         )
 
         return state, gen
 
     def generatemanifests(
         self,
         commonrevs,
         clrevorder,
         fastpathlinkrev,
         manifests,
         fnodes,
         source,
         clrevtolocalrev,
         sidedata_helpers=None,
         debug_info=None,
     ):
         """Returns an iterator of changegroup chunks containing manifests.
 
         `source` is unused here, but is used by extensions like remotefilelog to
         change what is sent based in pulls vs pushes, etc.
 
         See `revlogutil.sidedata.get_sidedata_helpers` for the doc on
         `sidedata_helpers`.
         """
         repo = self._repo
         mfl = repo.manifestlog
         tmfnodes = {b'': manifests}
 
         # Callback for the manifest, used to collect linkrevs for filelog
         # revisions.
         # Returns the linkrev node (collected in lookupcl).
         def makelookupmflinknode(tree, nodes):
             if fastpathlinkrev:
                 assert not tree
 
                 # pytype: disable=unsupported-operands
                 return manifests.__getitem__
                 # pytype: enable=unsupported-operands
 
             def lookupmflinknode(x):
                 """Callback for looking up the linknode for manifests.
 
                 Returns the linkrev node for the specified manifest.
 
                 SIDE EFFECT:
 
                 1) fclnodes gets populated with the list of relevant
                    file nodes if we're not using fastpathlinkrev
                 2) When treemanifests are in use, collects treemanifest nodes
                    to send
 
                 Note that this means manifests must be completely sent to
                 the client before you can trust the list of files and
                 treemanifests to send.
                 """
                 clnode = nodes[x]
                 mdata = mfl.get(tree, x).readfast(shallow=True)
                 for p, n, fl in mdata.iterentries():
                     if fl == b't':  # subdirectory manifest
                         subtree = tree + p + b'/'
                         tmfclnodes = tmfnodes.setdefault(subtree, {})
                         tmfclnode = tmfclnodes.setdefault(n, clnode)
                         if clrevorder[clnode] < clrevorder[tmfclnode]:
                             tmfclnodes[n] = clnode
                     else:
                         f = tree + p
                         fclnodes = fnodes.setdefault(f, {})
                         fclnode = fclnodes.setdefault(n, clnode)
                         if clrevorder[clnode] < clrevorder[fclnode]:
                             fclnodes[n] = clnode
                 return clnode
 
             return lookupmflinknode
 
         while tmfnodes:
             tree, nodes = tmfnodes.popitem()
 
             should_visit = self._matcher.visitdir(tree[:-1])
             if tree and not should_visit:
                 continue
 
             store = mfl.getstorage(tree)
 
             if not should_visit:
                 # No nodes to send because this directory is out of
                 # the client's view of the repository (probably
                 # because of narrow clones). Do this even for the root
                 # directory (tree=='')
                 prunednodes = []
             else:
                 # Avoid sending any manifest nodes we can prove the
                 # client already has by checking linkrevs. See the
                 # related comment in generatefiles().
                 prunednodes = self._prunemanifests(store, nodes, commonrevs)
 
             if tree and not prunednodes:
                 continue
 
             lookupfn = makelookupmflinknode(tree, nodes)
 
             deltas = deltagroup(
                 self._repo,
                 store,
                 prunednodes,
                 False,
                 lookupfn,
                 self._forcedeltaparentprev,
                 ellipses=self._ellipses,
                 topic=_(b'manifests'),
                 clrevtolocalrev=clrevtolocalrev,
                 fullclnodes=self._fullclnodes,
                 precomputedellipsis=self._precomputedellipsis,
                 sidedata_helpers=sidedata_helpers,
                 debug_info=debug_info,
             )
 
             if not self._oldmatcher.visitdir(store.tree[:-1]):
                 yield tree, deltas
             else:
                 # 'deltas' is a generator and we need to consume it even if
                 # we are not going to send it because a side-effect is that
                 # it updates tmdnodes (via lookupfn)
                 for d in deltas:
                     pass
                 if not tree:
                     yield tree, []
 
     def _prunemanifests(self, store, nodes, commonrevs):
         if not self._ellipses:
             # In non-ellipses case and large repositories, it is better to
             # prevent calling of store.rev and store.linkrev on a lot of
             # nodes as compared to sending some extra data
             return nodes.copy()
         # This is split out as a separate method to allow filtering
         # commonrevs in extension code.
         #
         # TODO(augie): this shouldn't be required, instead we should
         # make filtering of revisions to send delegated to the store
         # layer.
         frev, flr = store.rev, store.linkrev
         return [n for n in nodes if flr(frev(n)) not in commonrevs]
 
     # The 'source' parameter is useful for extensions
     def generatefiles(
         self,
         changedfiles,
         commonrevs,
         source,
         mfdicts,
         fastpathlinkrev,
         fnodes,
         clrevs,
         sidedata_helpers=None,
         debug_info=None,
     ):
         changedfiles = [
             f
             for f in changedfiles
             if self._matcher(f) and not self._oldmatcher(f)
         ]
 
         if not fastpathlinkrev:
 
             def normallinknodes(unused, fname):
                 return fnodes.get(fname, {})
 
         else:
             cln = self._repo.changelog.node
 
             def normallinknodes(store, fname):
                 flinkrev = store.linkrev
                 fnode = store.node
                 revs = ((r, flinkrev(r)) for r in store)
                 return {fnode(r): cln(lr) for r, lr in revs if lr in clrevs}
 
         clrevtolocalrev = {}
 
         if self._isshallow:
             # In a shallow clone, the linknodes callback needs to also include
             # those file nodes that are in the manifests we sent but weren't
             # introduced by those manifests.
             commonctxs = [self._repo[c] for c in commonrevs]
             clrev = self._repo.changelog.rev
 
             def linknodes(flog, fname):
                 for c in commonctxs:
                     try:
                         fnode = c.filenode(fname)
                         clrevtolocalrev[c.rev()] = flog.rev(fnode)
                     except error.ManifestLookupError:
                         pass
                 links = normallinknodes(flog, fname)
                 if len(links) != len(mfdicts):
                     for mf, lr in mfdicts:
                         fnode = mf.get(fname, None)
                         if fnode in links:
                             links[fnode] = min(links[fnode], lr, key=clrev)
                         elif fnode:
                             links[fnode] = lr
                 return links
 
         else:
             linknodes = normallinknodes
 
         repo = self._repo
         progress = repo.ui.makeprogress(
             _(b'files'), unit=_(b'files'), total=len(changedfiles)
         )
         for i, fname in enumerate(sorted(changedfiles)):
             filerevlog = repo.file(fname)
             if not filerevlog:
                 raise error.Abort(
                     _(b"empty or missing file data for %s") % fname
                 )
 
             clrevtolocalrev.clear()
 
             linkrevnodes = linknodes(filerevlog, fname)
+
             # Lookup for filenodes, we collected the linkrev nodes above in the
             # fastpath case and with lookupmf in the slowpath case.
             def lookupfilelog(x):
                 return linkrevnodes[x]
 
             frev, flr = filerevlog.rev, filerevlog.linkrev
             # Skip sending any filenode we know the client already
             # has. This avoids over-sending files relatively
             # inexpensively, so it's not a problem if we under-filter
             # here.
             filenodes = [
                 n for n in linkrevnodes if flr(frev(n)) not in commonrevs
             ]
 
             if not filenodes:
                 continue
 
             progress.update(i + 1, item=fname)
 
             deltas = deltagroup(
                 self._repo,
                 filerevlog,
                 filenodes,
                 False,
                 lookupfilelog,
                 self._forcedeltaparentprev,
                 ellipses=self._ellipses,
                 clrevtolocalrev=clrevtolocalrev,
                 fullclnodes=self._fullclnodes,
                 precomputedellipsis=self._precomputedellipsis,
                 sidedata_helpers=sidedata_helpers,
                 debug_info=debug_info,
             )
 
             yield fname, deltas
 
         progress.complete()
 
 
 def _makecg1packer(
     repo,
     oldmatcher,
     matcher,
     bundlecaps,
     ellipses=False,
     shallow=False,
     ellipsisroots=None,
     fullnodes=None,
     remote_sidedata=None,
 ):
     builddeltaheader = lambda d: _CHANGEGROUPV1_DELTA_HEADER.pack(
         d.node, d.p1node, d.p2node, d.linknode
     )
 
     return cgpacker(
         repo,
         oldmatcher,
         matcher,
         b'01',
         builddeltaheader=builddeltaheader,
         manifestsend=b'',
         forcedeltaparentprev=True,
         bundlecaps=bundlecaps,
         ellipses=ellipses,
         shallow=shallow,
         ellipsisroots=ellipsisroots,
         fullnodes=fullnodes,
     )
 
 
 def _makecg2packer(
     repo,
     oldmatcher,
     matcher,
     bundlecaps,
     ellipses=False,
     shallow=False,
     ellipsisroots=None,
     fullnodes=None,
     remote_sidedata=None,
 ):
     builddeltaheader = lambda d: _CHANGEGROUPV2_DELTA_HEADER.pack(
         d.node, d.p1node, d.p2node, d.basenode, d.linknode
     )
 
     return cgpacker(
         repo,
         oldmatcher,
         matcher,
         b'02',
         builddeltaheader=builddeltaheader,
         manifestsend=b'',
         bundlecaps=bundlecaps,
         ellipses=ellipses,
         shallow=shallow,
         ellipsisroots=ellipsisroots,
         fullnodes=fullnodes,
     )
 
 
 def _makecg3packer(
     repo,
     oldmatcher,
     matcher,
     bundlecaps,
     ellipses=False,
     shallow=False,
     ellipsisroots=None,
     fullnodes=None,
     remote_sidedata=None,
 ):
     builddeltaheader = lambda d: _CHANGEGROUPV3_DELTA_HEADER.pack(
         d.node, d.p1node, d.p2node, d.basenode, d.linknode, d.flags
     )
 
     return cgpacker(
         repo,
         oldmatcher,
         matcher,
         b'03',
         builddeltaheader=builddeltaheader,
         manifestsend=closechunk(),
         bundlecaps=bundlecaps,
         ellipses=ellipses,
         shallow=shallow,
         ellipsisroots=ellipsisroots,
         fullnodes=fullnodes,
     )
 
 
 def _makecg4packer(
     repo,
     oldmatcher,
     matcher,
     bundlecaps,
     ellipses=False,
     shallow=False,
     ellipsisroots=None,
     fullnodes=None,
     remote_sidedata=None,
 ):
     # Sidedata is in a separate chunk from the delta to differentiate
     # "raw delta" and sidedata.
     def builddeltaheader(d):
         return _CHANGEGROUPV4_DELTA_HEADER.pack(
             d.protocol_flags,
             d.node,
             d.p1node,
             d.p2node,
             d.basenode,
             d.linknode,
             d.flags,
         )
 
     return cgpacker(
         repo,
         oldmatcher,
         matcher,
         b'04',
         builddeltaheader=builddeltaheader,
         manifestsend=closechunk(),
         bundlecaps=bundlecaps,
         ellipses=ellipses,
         shallow=shallow,
         ellipsisroots=ellipsisroots,
         fullnodes=fullnodes,
         remote_sidedata=remote_sidedata,
     )
 
 
 _packermap = {
     b'01': (_makecg1packer, cg1unpacker),
     # cg2 adds support for exchanging generaldelta
     b'02': (_makecg2packer, cg2unpacker),
     # cg3 adds support for exchanging revlog flags and treemanifests
     b'03': (_makecg3packer, cg3unpacker),
     # ch4 adds support for exchanging sidedata
     b'04': (_makecg4packer, cg4unpacker),
 }
 
 
 def allsupportedversions(repo):
     versions = set(_packermap.keys())
     needv03 = False
     if (
         repo.ui.configbool(b'experimental', b'changegroup3')
         or repo.ui.configbool(b'experimental', b'treemanifest')
         or scmutil.istreemanifest(repo)
     ):
         # we keep version 03 because we need to to exchange treemanifest data
         #
         # we also keep vresion 01 and 02, because it is possible for repo to
         # contains both normal and tree manifest at the same time. so using
         # older version to pull data is viable
         #
         # (or even to push subset of history)
         needv03 = True
     if not needv03:
         versions.discard(b'03')
     want_v4 = (
         repo.ui.configbool(b'experimental', b'changegroup4')
         or requirements.REVLOGV2_REQUIREMENT in repo.requirements
         or requirements.CHANGELOGV2_REQUIREMENT in repo.requirements
     )
     if not want_v4:
         versions.discard(b'04')
     return versions
 
 
 # Changegroup versions that can be applied to the repo
 def supportedincomingversions(repo):
     return allsupportedversions(repo)
 
 
 # Changegroup versions that can be created from the repo
 def supportedoutgoingversions(repo):
     versions = allsupportedversions(repo)
     if scmutil.istreemanifest(repo):
         # Versions 01 and 02 support only flat manifests and it's just too
         # expensive to convert between the flat manifest and tree manifest on
         # the fly. Since tree manifests are hashed differently, all of history
         # would have to be converted. Instead, we simply don't even pretend to
         # support versions 01 and 02.
         versions.discard(b'01')
         versions.discard(b'02')
     if requirements.NARROW_REQUIREMENT in repo.requirements:
         # Versions 01 and 02 don't support revlog flags, and we need to
         # support that for stripping and unbundling to work.
         versions.discard(b'01')
         versions.discard(b'02')
     if LFS_REQUIREMENT in repo.requirements:
         # Versions 01 and 02 don't support revlog flags, and we need to
         # mark LFS entries with REVIDX_EXTSTORED.
         versions.discard(b'01')
         versions.discard(b'02')
 
     return versions
 
 
 def localversion(repo):
     # Finds the best version to use for bundles that are meant to be used
     # locally, such as those from strip and shelve, and temporary bundles.
     return max(supportedoutgoingversions(repo))
 
 
 def safeversion(repo):
     # Finds the smallest version that it's safe to assume clients of the repo
     # will support. For example, all hg versions that support generaldelta also
     # support changegroup 02.
     versions = supportedoutgoingversions(repo)
     if requirements.GENERALDELTA_REQUIREMENT in repo.requirements:
         versions.discard(b'01')
     assert versions
     return min(versions)
 
 
 def getbundler(
     version,
     repo,
     bundlecaps=None,
     oldmatcher=None,
     matcher=None,
     ellipses=False,
     shallow=False,
     ellipsisroots=None,
     fullnodes=None,
     remote_sidedata=None,
 ):
     assert version in supportedoutgoingversions(repo)
 
     if matcher is None:
         matcher = matchmod.always()
     if oldmatcher is None:
         oldmatcher = matchmod.never()
 
     if version == b'01' and not matcher.always():
         raise error.ProgrammingError(
             b'version 01 changegroups do not support sparse file matchers'
         )
 
     if ellipses and version in (b'01', b'02'):
         raise error.Abort(
             _(
                 b'ellipsis nodes require at least cg3 on client and server, '
                 b'but negotiated version %s'
             )
             % version
         )
 
     # Requested files could include files not in the local store. So
     # filter those out.
     matcher = repo.narrowmatch(matcher)
 
     fn = _packermap[version][0]
     return fn(
         repo,
         oldmatcher,
         matcher,
         bundlecaps,
         ellipses=ellipses,
         shallow=shallow,
         ellipsisroots=ellipsisroots,
         fullnodes=fullnodes,
         remote_sidedata=remote_sidedata,
     )
 
 
 def getunbundler(version, fh, alg, extras=None):
     return _packermap[version][1](fh, alg, extras=extras)
 
 
 def _changegroupinfo(repo, nodes, source):
     if repo.ui.verbose or source == b'bundle':
         repo.ui.status(_(b"%d changesets found\n") % len(nodes))
     if repo.ui.debugflag:
         repo.ui.debug(b"list of changesets:\n")
         for node in nodes:
             repo.ui.debug(b"%s\n" % hex(node))
 
 
 def makechangegroup(
     repo,
     outgoing,
     version,
     source,
     fastpath=False,
     bundlecaps=None,
 ):
     cgstream = makestream(
         repo,
         outgoing,
         version,
         source,
         fastpath=fastpath,
         bundlecaps=bundlecaps,
     )
     return getunbundler(
         version,
         util.chunkbuffer(cgstream),
         None,
         {b'clcount': len(outgoing.missing)},
     )
 
 
 def makestream(
     repo,
     outgoing,
     version,
     source,
     fastpath=False,
     bundlecaps=None,
     matcher=None,
     remote_sidedata=None,
 ):
     bundler = getbundler(
         version,
         repo,
         bundlecaps=bundlecaps,
         matcher=matcher,
         remote_sidedata=remote_sidedata,
     )
 
     repo = repo.unfiltered()
     commonrevs = outgoing.common
     csets = outgoing.missing
     heads = outgoing.ancestorsof
     # We go through the fast path if we get told to, or if all (unfiltered
     # heads have been requested (since we then know there all linkrevs will
     # be pulled by the client).
     heads.sort()
     fastpathlinkrev = fastpath or (
         repo.filtername is None and heads == sorted(repo.heads())
     )
 
     repo.hook(b'preoutgoing', throw=True, source=source)
     _changegroupinfo(repo, csets, source)
     return bundler.generate(
         commonrevs,
         csets,
         fastpathlinkrev,
         source,
     )
 
 
 def _addchangegroupfiles(
     repo,
     source,
     revmap,
     trp,
     expectedfiles,
     needfiles,
     addrevisioncb=None,
     debug_info=None,
     delta_base_reuse_policy=None,
 ):
     revisions = 0
     files = 0
     progress = repo.ui.makeprogress(
         _(b'files'), unit=_(b'files'), total=expectedfiles
     )
     for chunkdata in iter(source.filelogheader, {}):
         files += 1
         f = chunkdata[b"filename"]
         repo.ui.debug(b"adding %s revisions\n" % f)
         progress.increment()
         fl = repo.file(f)
         o = len(fl)
         try:
             deltas = source.deltaiter()
             added = fl.addgroup(
                 deltas,
                 revmap,
                 trp,
                 addrevisioncb=addrevisioncb,
                 debug_info=debug_info,
                 delta_base_reuse_policy=delta_base_reuse_policy,
             )
             if not added:
                 raise error.Abort(_(b"received file revlog group is empty"))
         except error.CensoredBaseError as e:
             raise error.Abort(_(b"received delta base is censored: %s") % e)
         revisions += len(fl) - o
         if f in needfiles:
             needs = needfiles[f]
             for new in range(o, len(fl)):
                 n = fl.node(new)
                 if n in needs:
                     needs.remove(n)
                 else:
                     raise error.Abort(_(b"received spurious file revlog entry"))
             if not needs:
                 del needfiles[f]
     progress.complete()
 
     for f, needs in needfiles.items():
         fl = repo.file(f)
         for n in needs:
             try:
                 fl.rev(n)
             except error.LookupError:
                 raise error.Abort(
                     _(b'missing file data for %s:%s - run hg verify')
                     % (f, hex(n))
                 )
 
     return revisions, files

mercurial/dagop.py

@@ -1,1110 +1,1141 b''
 # dagop.py - graph ancestry and topology algorithm for revset
 #
 # Copyright 2010 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 heapq
 
 from .thirdparty import attr
 from .node import nullrev
 from . import (
     error,
     mdiff,
     patch,
     pycompat,
     scmutil,
     smartset,
 )
 
 baseset = smartset.baseset
 generatorset = smartset.generatorset
 
 # possible maximum depth between null and wdir()
 maxlogdepth = 0x80000000
 
 
 def _walkrevtree(pfunc, revs, startdepth, stopdepth, reverse):
     """Walk DAG using 'pfunc' from the given 'revs' nodes
 
     'pfunc(rev)' should return the parent/child revisions of the given 'rev'
     if 'reverse' is True/False respectively.
 
     Scan ends at the stopdepth (exlusive) if specified. Revisions found
     earlier than the startdepth are omitted.
     """
     if startdepth is None:
         startdepth = 0
     if stopdepth is None:
         stopdepth = maxlogdepth
     if stopdepth == 0:
         return
     if stopdepth < 0:
         raise error.ProgrammingError(b'negative stopdepth')
     if reverse:
         heapsign = -1  # max heap
     else:
         heapsign = +1  # min heap
 
     # load input revs lazily to heap so earlier revisions can be yielded
     # without fully computing the input revs
     revs.sort(reverse)
     irevs = iter(revs)
     pendingheap = []  # [(heapsign * rev, depth), ...] (i.e. lower depth first)
 
     inputrev = next(irevs, None)
     if inputrev is not None:
         heapq.heappush(pendingheap, (heapsign * inputrev, 0))
 
     lastrev = None
     while pendingheap:
         currev, curdepth = heapq.heappop(pendingheap)
         currev = heapsign * currev
         if currev == inputrev:
             inputrev = next(irevs, None)
             if inputrev is not None:
                 heapq.heappush(pendingheap, (heapsign * inputrev, 0))
         # rescan parents until curdepth >= startdepth because queued entries
         # of the same revision are iterated from the lowest depth
         foundnew = currev != lastrev
         if foundnew and curdepth >= startdepth:
             lastrev = currev
             yield currev
         pdepth = curdepth + 1
         if foundnew and pdepth < stopdepth:
             for prev in pfunc(currev):
                 if prev != nullrev:
                     heapq.heappush(pendingheap, (heapsign * prev, pdepth))
 
 
 def filectxancestors(fctxs, followfirst=False):
     """Like filectx.ancestors(), but can walk from multiple files/revisions,
     and includes the given fctxs themselves
 
     Yields (rev, {fctx, ...}) pairs in descending order.
     """
     visit = {}
     visitheap = []
 
     def addvisit(fctx):
         rev = scmutil.intrev(fctx)
         if rev not in visit:
             visit[rev] = set()
             heapq.heappush(visitheap, -rev)  # max heap
         visit[rev].add(fctx)
 
     if followfirst:
         cut = 1
     else:
         cut = None
 
     for c in fctxs:
         addvisit(c)
     while visit:
         currev = -(heapq.heappop(visitheap))
         curfctxs = visit.pop(currev)
         yield currev, curfctxs
         for c in curfctxs:
             for parent in c.parents()[:cut]:
                 addvisit(parent)
     assert not visitheap
 
 
 def filerevancestors(fctxs, followfirst=False):
     """Like filectx.ancestors(), but can walk from multiple files/revisions,
     and includes the given fctxs themselves
 
     Returns a smartset.
     """
     gen = (rev for rev, _cs in filectxancestors(fctxs, followfirst))
     return generatorset(gen, iterasc=False)
 
 
 def _genrevancestors(repo, revs, followfirst, startdepth, stopdepth, cutfunc):
     if followfirst:
         cut = 1
     else:
         cut = None
     cl = repo.changelog
 
     def plainpfunc(rev):
         try:
             return cl.parentrevs(rev)[:cut]
         except error.WdirUnsupported:
             return (pctx.rev() for pctx in repo[rev].parents()[:cut])
 
     if cutfunc is None:
         pfunc = plainpfunc
     else:
         pfunc = lambda rev: [r for r in plainpfunc(rev) if not cutfunc(r)]
         revs = revs.filter(lambda rev: not cutfunc(rev))
     return _walkrevtree(pfunc, revs, startdepth, stopdepth, reverse=True)
 
 
 def revancestors(
     repo, revs, followfirst=False, startdepth=None, stopdepth=None, cutfunc=None
 ):
     r"""Like revlog.ancestors(), but supports additional options, includes
     the given revs themselves, and returns a smartset
 
     Scan ends at the stopdepth (exlusive) if specified. Revisions found
     earlier than the startdepth are omitted.
 
     If cutfunc is provided, it will be used to cut the traversal of the DAG.
     When cutfunc(X) returns True, the DAG traversal stops - revision X and
     X's ancestors in the traversal path will be skipped. This could be an
     optimization sometimes.
 
     Note: if Y is an ancestor of X, cutfunc(X) returning True does not
     necessarily mean Y will also be cut. Usually cutfunc(Y) also wants to
     return True in this case. For example,
 
         D     # revancestors(repo, D, cutfunc=lambda rev: rev == B)
         |\    # will include "A", because the path D -> C -> A was not cut.
         B C   # If "B" gets cut, "A" might want to be cut too.
         |/
         A
     """
     gen = _genrevancestors(
         repo, revs, followfirst, startdepth, stopdepth, cutfunc
     )
     return generatorset(gen, iterasc=False)
 
 
 def _genrevdescendants(repo, revs, followfirst):
     if followfirst:
         cut = 1
     else:
         cut = None
 
     cl = repo.changelog
     first = revs.min()
     if first == nullrev:
         # Are there nodes with a null first parent and a non-null
         # second one? Maybe. Do we care? Probably not.
         yield first
         for i in cl:
             yield i
     else:
         seen = set(revs)
         for i in cl.revs(first):
             if i in seen:
                 yield i
                 continue
             for x in cl.parentrevs(i)[:cut]:
                 if x != nullrev and x in seen:
                     seen.add(i)
                     yield i
                     break
 
 
 def _builddescendantsmap(repo, startrev, followfirst):
     """Build map of 'rev -> child revs', offset from startrev"""
     cl = repo.changelog
     descmap = [[] for _rev in range(startrev, len(cl))]
     for currev in cl.revs(startrev + 1):
         p1rev, p2rev = cl.parentrevs(currev)
         if p1rev >= startrev:
             descmap[p1rev - startrev].append(currev)
         if not followfirst and p2rev != nullrev and p2rev >= startrev:
             descmap[p2rev - startrev].append(currev)
     return descmap
 
 
 def _genrevdescendantsofdepth(repo, revs, followfirst, startdepth, stopdepth):
     startrev = revs.min()
     descmap = _builddescendantsmap(repo, startrev, followfirst)
 
     def pfunc(rev):
         return descmap[rev - startrev]
 
     return _walkrevtree(pfunc, revs, startdepth, stopdepth, reverse=False)
 
 
 def revdescendants(repo, revs, followfirst, startdepth=None, stopdepth=None):
     """Like revlog.descendants() but supports additional options, includes
     the given revs themselves, and returns a smartset
 
     Scan ends at the stopdepth (exlusive) if specified. Revisions found
     earlier than the startdepth are omitted.
     """
     if startdepth is None and (stopdepth is None or stopdepth >= maxlogdepth):
         gen = _genrevdescendants(repo, revs, followfirst)
     else:
         gen = _genrevdescendantsofdepth(
             repo, revs, followfirst, startdepth, stopdepth
         )
     return generatorset(gen, iterasc=True)
 
 
 def descendantrevs(revs, revsfn, parentrevsfn):
     """Generate revision number descendants in revision order.
 
     Yields revision numbers starting with a child of some rev in
     ``revs``. Results are ordered by revision number and are
     therefore topological. Each revision is not considered a descendant
     of itself.
 
     ``revsfn`` is a callable that with no argument iterates over all
     revision numbers and with a ``start`` argument iterates over revision
     numbers beginning with that value.
 
     ``parentrevsfn`` is a callable that receives a revision number and
     returns an iterable of parent revision numbers, whose values may include
     nullrev.
     """
     first = min(revs)
 
     if first == nullrev:
         for rev in revsfn():
             yield rev
         return
 
     seen = set(revs)
     for rev in revsfn(start=first + 1):
         for prev in parentrevsfn(rev):
             if prev != nullrev and prev in seen:
                 seen.add(rev)
                 yield rev
                 break
 
 
 class subsetparentswalker:
     r"""Scan adjacent ancestors in the graph given by the subset
 
     This computes parent-child relations in the sub graph filtered by
     a revset. Primary use case is to draw a revisions graph.
 
     In the following example, we consider that the node 'f' has edges to all
     ancestor nodes, but redundant paths are eliminated. The edge 'f'->'b'
     is eliminated because there is a path 'f'->'c'->'b' for example.
 
           - d - e -
          /         \
         a - b - c - f
 
     If the node 'c' is filtered out, the edge 'f'->'b' is activated.
 
           - d - e -
          /         \
         a - b -(c)- f
 
     Likewise, if 'd' and 'e' are filtered out, this edge is fully eliminated
     since there is a path 'f'->'c'->'b'->'a' for 'f'->'a'.
 
            (d) (e)
 
         a - b - c - f
 
     Implementation-wise, 'f' is passed down to 'a' as unresolved through the
     'f'->'e'->'d'->'a' path, whereas we do also remember that 'f' has already
     been resolved while walking down the 'f'->'c'->'b'->'a' path. When
     processing the node 'a', the unresolved 'f'->'a' path is eliminated as
     the 'f' end is marked as resolved.
 
     Ancestors are searched from the tipmost revision in the subset so the
     results can be cached. You should specify startrev to narrow the search
     space to ':startrev'.
     """
 
     def __init__(self, repo, subset, startrev=None):
         if startrev is not None:
             subset = repo.revs(b'%d:null', startrev) & subset
 
         # equivalent to 'subset = subset.sorted(reverse=True)', but there's
         # no such function.
         fastdesc = subset.fastdesc
         if fastdesc:
             desciter = fastdesc()
         else:
             if not subset.isdescending() and not subset.istopo():
                 subset = smartset.baseset(subset)
                 subset.sort(reverse=True)
             desciter = iter(subset)
 
         self._repo = repo
         self._changelog = repo.changelog
         self._subset = subset
 
         # scanning state (see _scanparents):
         self._tovisit = []
         self._pendingcnt = {}
         self._pointers = {}
         self._parents = {}
         self._inputhead = nullrev  # reassigned by self._advanceinput()
         self._inputtail = desciter
         self._bottomrev = nullrev
         self._advanceinput()
 
     def parentsset(self, rev):
         """Look up parents of the given revision in the subset, and returns
         as a smartset"""
         return smartset.baseset(self.parents(rev))
 
     def parents(self, rev):
         """Look up parents of the given revision in the subset
 
         The returned revisions are sorted by parent index (p1/p2).
         """
         self._scanparents(rev)
         return [r for _c, r in sorted(self._parents.get(rev, []))]
 
     def _parentrevs(self, rev):
         try:
             revs = self._changelog.parentrevs(rev)
             if revs[-1] == nullrev:
                 return revs[:-1]
             return revs
         except error.WdirUnsupported:
             return tuple(pctx.rev() for pctx in self._repo[None].parents())
 
     def _advanceinput(self):
         """Advance the input iterator and set the next revision to _inputhead"""
         if self._inputhead < nullrev:
             return
         try:
             self._inputhead = next(self._inputtail)
         except StopIteration:
             self._bottomrev = self._inputhead
             self._inputhead = nullrev - 1
 
     def _scanparents(self, stoprev):
         """Scan ancestors until the parents of the specified stoprev are
         resolved"""
 
         # 'tovisit' is the queue of the input revisions and their ancestors.
         # It will be populated incrementally to minimize the initial cost
         # of computing the given subset.
         #
         # For to-visit revisions, we keep track of
         # - the number of the unresolved paths: pendingcnt[rev],
         # - dict of the unresolved descendants and chains: pointers[rev][0],
         # - set of the already resolved descendants: pointers[rev][1].
         #
         # When a revision is visited, 'pointers[rev]' should be popped and
         # propagated to its parents accordingly.
         #
         # Once all pending paths have been resolved, 'pendingcnt[rev]' becomes
         # 0 and 'parents[rev]' contains the unsorted list of parent revisions
         # and p1/p2 chains (excluding linear paths.) The p1/p2 chains will be
         # used as a sort key preferring p1. 'len(chain)' should be the number
         # of merges between two revisions.
 
         subset = self._subset
         tovisit = self._tovisit  # heap queue of [-rev]
         pendingcnt = self._pendingcnt  # {rev: count} for visited revisions
         pointers = self._pointers  # {rev: [{unresolved_rev: chain}, resolved]}
         parents = self._parents  # {rev: [(chain, rev)]}
 
         while tovisit or self._inputhead >= nullrev:
             if pendingcnt.get(stoprev) == 0:
                 return
 
             # feed greater revisions from input set to queue
             if not tovisit:
                 heapq.heappush(tovisit, -self._inputhead)
                 self._advanceinput()
             while self._inputhead >= -tovisit[0]:
                 heapq.heappush(tovisit, -self._inputhead)
                 self._advanceinput()
 
             rev = -heapq.heappop(tovisit)
             if rev < self._bottomrev:
                 return
             if rev in pendingcnt and rev not in pointers:
                 continue  # already visited
 
             curactive = rev in subset
             pendingcnt.setdefault(rev, 0)  # mark as visited
             if curactive:
                 assert rev not in parents
                 parents[rev] = []
             unresolved, resolved = pointers.pop(rev, ({}, set()))
 
             if curactive:
                 # reached to active rev, resolve pending descendants' parents
                 for r, c in unresolved.items():
                     pendingcnt[r] -= 1
                     assert pendingcnt[r] >= 0
                     if r in resolved:
                         continue  # eliminate redundant path
                     parents[r].append((c, rev))
                     # mark the descendant 'r' as resolved through this path if
                     # there are still pending pointers. the 'resolved' set may
                     # be concatenated later at a fork revision.
                     if pendingcnt[r] > 0:
                         resolved.add(r)
                 unresolved.clear()
                 # occasionally clean resolved markers. otherwise the set
                 # would grow indefinitely.
                 resolved = {r for r in resolved if pendingcnt[r] > 0}
 
             parentrevs = self._parentrevs(rev)
             bothparentsactive = all(p in subset for p in parentrevs)
 
             # set up or propagate tracking pointers if
             # - one of the parents is not active,
             # - or descendants' parents are unresolved.
             if not bothparentsactive or unresolved or resolved:
                 if len(parentrevs) <= 1:
                     # can avoid copying the tracking pointer
                     parentpointers = [(unresolved, resolved)]
                 else:
                     parentpointers = [
                         (unresolved, resolved),
                         (unresolved.copy(), resolved.copy()),
                     ]
                     # 'rev' is a merge revision. increment the pending count
                     # as the 'unresolved' dict will be duplicated, and append
                     # p1/p2 code to the existing chains.
                     for r in unresolved:
                         pendingcnt[r] += 1
                         parentpointers[0][0][r] += b'1'
                         parentpointers[1][0][r] += b'2'
                 for i, p in enumerate(parentrevs):
                     assert p < rev
                     heapq.heappush(tovisit, -p)
                     if p in pointers:
                         # 'p' is a fork revision. concatenate tracking pointers
                         # and decrement the pending count accordingly.
                         knownunresolved, knownresolved = pointers[p]
                         unresolved, resolved = parentpointers[i]
                         for r, c in unresolved.items():
                             if r in knownunresolved:
                                 # unresolved at both paths
                                 pendingcnt[r] -= 1
                                 assert pendingcnt[r] > 0
                                 # take shorter chain
                                 knownunresolved[r] = min(c, knownunresolved[r])
                             else:
                                 knownunresolved[r] = c
                         # simply propagate the 'resolved' set as deduplicating
                         # 'unresolved' here would be slightly complicated.
                         knownresolved.update(resolved)
                     else:
                         pointers[p] = parentpointers[i]
 
             # then, populate the active parents directly and add the current
             # 'rev' to the tracking pointers of the inactive parents.
             # 'pointers[p]' may be optimized out if both parents are active.
             chaincodes = [b''] if len(parentrevs) <= 1 else [b'1', b'2']
             if curactive and bothparentsactive:
                 for i, p in enumerate(parentrevs):
                     c = chaincodes[i]
                     parents[rev].append((c, p))
                     # no need to mark 'rev' as resolved since the 'rev' should
                     # be fully resolved (i.e. pendingcnt[rev] == 0)
                 assert pendingcnt[rev] == 0
             elif curactive:
                 for i, p in enumerate(parentrevs):
                     unresolved, resolved = pointers[p]
                     assert rev not in unresolved
                     c = chaincodes[i]
                     if p in subset:
                         parents[rev].append((c, p))
                         # mark 'rev' as resolved through this path
                         resolved.add(rev)
                     else:
                         pendingcnt[rev] += 1
                         unresolved[rev] = c
                 assert 0 < pendingcnt[rev] <= 2
 
 
 def _reachablerootspure(pfunc, minroot, roots, heads, includepath):
     """See revlog.reachableroots"""
     if not roots:
         return []
     roots = set(roots)
     visit = list(heads)
     reachable = set()
     seen = {}
     # prefetch all the things! (because python is slow)
     reached = reachable.add
     dovisit = visit.append
     nextvisit = visit.pop
     # open-code the post-order traversal due to the tiny size of
     # sys.getrecursionlimit()
     while visit:
         rev = nextvisit()
         if rev in roots:
             reached(rev)
             if not includepath:
                 continue
         parents = pfunc(rev)
         seen[rev] = parents
         for parent in parents:
             if parent >= minroot and parent not in seen:
                 dovisit(parent)
     if not reachable:
         return baseset()
     if not includepath:
         return reachable
     for rev in sorted(seen):
         for parent in seen[rev]:
             if parent in reachable:
                 reached(rev)
     return reachable
 
 
 def reachableroots(repo, roots, heads, includepath=False):
     """See revlog.reachableroots"""
     if not roots:
         return baseset()
     minroot = roots.min()
     roots = list(roots)
     heads = list(heads)
     revs = repo.changelog.reachableroots(minroot, heads, roots, includepath)
     revs = baseset(revs)
     revs.sort()
     return revs
 
 
 def _changesrange(fctx1, fctx2, linerange2, diffopts):
     """Return `(diffinrange, linerange1)` where `diffinrange` is True
     if diff from fctx2 to fctx1 has changes in linerange2 and
     `linerange1` is the new line range for fctx1.
     """
     blocks = mdiff.allblocks(fctx1.data(), fctx2.data(), diffopts)
     filteredblocks, linerange1 = mdiff.blocksinrange(blocks, linerange2)
     diffinrange = any(stype == b'!' for _, stype in filteredblocks)
     return diffinrange, linerange1
 
 
 def blockancestors(fctx, fromline, toline, followfirst=False):
     """Yield ancestors of `fctx` with respect to the block of lines within
     `fromline`-`toline` range.
     """
     diffopts = patch.diffopts(fctx._repo.ui)
     fctx = fctx.introfilectx()
     visit = {(fctx.linkrev(), fctx.filenode()): (fctx, (fromline, toline))}
     while visit:
         c, linerange2 = visit.pop(max(visit))
         pl = c.parents()
         if followfirst:
             pl = pl[:1]
         if not pl:
             # The block originates from the initial revision.
             yield c, linerange2
             continue
         inrange = False
         for p in pl:
             inrangep, linerange1 = _changesrange(p, c, linerange2, diffopts)
             inrange = inrange or inrangep
             if linerange1[0] == linerange1[1]:
                 # Parent's linerange is empty, meaning that the block got
                 # introduced in this revision; no need to go futher in this
                 # branch.
                 continue
             # Set _descendantrev with 'c' (a known descendant) so that, when
             # _adjustlinkrev is called for 'p', it receives this descendant
             # (as srcrev) instead possibly topmost introrev.
             p._descendantrev = c.rev()
             visit[p.linkrev(), p.filenode()] = p, linerange1
         if inrange:
             yield c, linerange2
 
 
 def blockdescendants(fctx, fromline, toline):
     """Yield descendants of `fctx` with respect to the block of lines within
     `fromline`-`toline` range.
     """
     # First possibly yield 'fctx' if it has changes in range with respect to
     # its parents.
     try:
         c, linerange1 = next(blockancestors(fctx, fromline, toline))
     except StopIteration:
         pass
     else:
         if c == fctx:
             yield c, linerange1
 
     diffopts = patch.diffopts(fctx._repo.ui)
     fl = fctx.filelog()
     seen = {fctx.filerev(): (fctx, (fromline, toline))}
     for i in fl.descendants([fctx.filerev()]):
         c = fctx.filectx(i)
         inrange = False
         for x in fl.parentrevs(i):
             try:
                 p, linerange2 = seen[x]
             except KeyError:
                 # nullrev or other branch
                 continue
             inrangep, linerange1 = _changesrange(c, p, linerange2, diffopts)
             inrange = inrange or inrangep
             # If revision 'i' has been seen (it's a merge) and the line range
             # previously computed differs from the one we just got, we take the
             # surrounding interval. This is conservative but avoids loosing
             # information.
             if i in seen and seen[i][1] != linerange1:
                 lbs, ubs = zip(linerange1, seen[i][1])
                 linerange1 = min(lbs), max(ubs)
             seen[i] = c, linerange1
         if inrange:
             yield c, linerange1
 
 
 @attr.s(slots=True, frozen=True)
 class annotateline:
     fctx = attr.ib()
     lineno = attr.ib()
     # Whether this annotation was the result of a skip-annotate.
     skip = attr.ib(default=False)
     text = attr.ib(default=None)
 
 
 @attr.s(slots=True, frozen=True)
 class _annotatedfile:
     # list indexed by lineno - 1
     fctxs = attr.ib()
     linenos = attr.ib()
     skips = attr.ib()
     # full file content
     text = attr.ib()
 
 
 def _countlines(text):
     if text.endswith(b"\n"):
         return text.count(b"\n")
     return text.count(b"\n") + int(bool(text))
 
 
 def _decoratelines(text, fctx):
     n = _countlines(text)
     linenos = pycompat.rangelist(1, n + 1)
     return _annotatedfile([fctx] * n, linenos, [False] * n, text)
 
 
 def _annotatepair(parents, childfctx, child, skipchild, diffopts):
     r"""
     Given parent and child fctxes and annotate data for parents, for all lines
     in either parent that match the child, annotate the child with the parent's
     data.
 
     Additionally, if `skipchild` is True, replace all other lines with parent
     annotate data as well such that child is never blamed for any lines.
 
     See test-annotate.py for unit tests.
     """
     pblocks = [
         (parent, mdiff.allblocks(parent.text, child.text, opts=diffopts))
         for parent in parents
     ]
 
     if skipchild:
         # Need to iterate over the blocks twice -- make it a list
         pblocks = [(p, list(blocks)) for (p, blocks) in pblocks]
     # Mercurial currently prefers p2 over p1 for annotate.
     # TODO: change this?
     for parent, blocks in pblocks:
         for (a1, a2, b1, b2), t in blocks:
             # Changed blocks ('!') or blocks made only of blank lines ('~')
             # belong to the child.
             if t == b'=':
                 child.fctxs[b1:b2] = parent.fctxs[a1:a2]
                 child.linenos[b1:b2] = parent.linenos[a1:a2]
                 child.skips[b1:b2] = parent.skips[a1:a2]
 
     if skipchild:
         # Now try and match up anything that couldn't be matched,
         # Reversing pblocks maintains bias towards p2, matching above
         # behavior.
         pblocks.reverse()
 
         # The heuristics are:
         # * Work on blocks of changed lines (effectively diff hunks with -U0).
         # This could potentially be smarter but works well enough.
         # * For a non-matching section, do a best-effort fit. Match lines in
         #   diff hunks 1:1, dropping lines as necessary.
         # * Repeat the last line as a last resort.
 
         # First, replace as much as possible without repeating the last line.
         remaining = [(parent, []) for parent, _blocks in pblocks]
         for idx, (parent, blocks) in enumerate(pblocks):
             for (a1, a2, b1, b2), _t in blocks:
                 if a2 - a1 >= b2 - b1:
                     for bk in range(b1, b2):
                         if child.fctxs[bk] == childfctx:
                             ak = min(a1 + (bk - b1), a2 - 1)
                             child.fctxs[bk] = parent.fctxs[ak]
                             child.linenos[bk] = parent.linenos[ak]
                             child.skips[bk] = True
                 else:
                     remaining[idx][1].append((a1, a2, b1, b2))
 
         # Then, look at anything left, which might involve repeating the last
         # line.
         for parent, blocks in remaining:
             for a1, a2, b1, b2 in blocks:
                 for bk in range(b1, b2):
                     if child.fctxs[bk] == childfctx:
                         ak = min(a1 + (bk - b1), a2 - 1)
                         child.fctxs[bk] = parent.fctxs[ak]
                         child.linenos[bk] = parent.linenos[ak]
                         child.skips[bk] = True
     return child
 
 
 def annotate(base, parents, skiprevs=None, diffopts=None):
     """Core algorithm for filectx.annotate()
 
     `parents(fctx)` is a function returning a list of parent filectxs.
     """
 
     # This algorithm would prefer to be recursive, but Python is a
     # bit recursion-hostile. Instead we do an iterative
     # depth-first search.
 
     # 1st DFS pre-calculates pcache and needed
     visit = [base]
     pcache = {}
     needed = {base: 1}
     while visit:
         f = visit.pop()
         if f in pcache:
             continue
         pl = parents(f)
         pcache[f] = pl
         for p in pl:
             needed[p] = needed.get(p, 0) + 1
             if p not in pcache:
                 visit.append(p)
 
     # 2nd DFS does the actual annotate
     visit[:] = [base]
     hist = {}
     while visit:
         f = visit[-1]
         if f in hist:
             visit.pop()
             continue
 
         ready = True
         pl = pcache[f]
         for p in pl:
             if p not in hist:
                 ready = False
                 visit.append(p)
         if ready:
             visit.pop()
             curr = _decoratelines(f.data(), f)
             skipchild = False
             if skiprevs is not None:
                 skipchild = f._changeid in skiprevs
             curr = _annotatepair(
                 [hist[p] for p in pl], f, curr, skipchild, diffopts
             )
             for p in pl:
                 if needed[p] == 1:
                     del hist[p]
                     del needed[p]
                 else:
                     needed[p] -= 1
 
             hist[f] = curr
             del pcache[f]
 
     a = hist[base]
     return [
         annotateline(*r)
         for r in zip(a.fctxs, a.linenos, a.skips, mdiff.splitnewlines(a.text))
     ]
 
 
 def toposort(revs, parentsfunc, firstbranch=()):
     """Yield revisions from heads to roots one (topo) branch at a time.
 
     This function aims to be used by a graph generator that wishes to minimize
     the number of parallel branches and their interleaving.
 
     Example iteration order (numbers show the "true" order in a changelog):
 
       o  4
       |
       o  1
       |
       | o  3
       | |
       | o  2
       |/
       o  0
 
     Note that the ancestors of merges are understood by the current
     algorithm to be on the same branch. This means no reordering will
     occur behind a merge.
     """
 
     ### Quick summary of the algorithm
     #
     # This function is based around a "retention" principle. We keep revisions
     # in memory until we are ready to emit a whole branch that immediately
     # "merges" into an existing one. This reduces the number of parallel
     # branches with interleaved revisions.
     #
     # During iteration revs are split into two groups:
     # A) revision already emitted
     # B) revision in "retention". They are stored as different subgroups.
     #
     # for each REV, we do the following logic:
     #
     #   1) if REV is a parent of (A), we will emit it. If there is a
     #   retention group ((B) above) that is blocked on REV being
     #   available, we emit all the revisions out of that retention
     #   group first.
     #
     #   2) else, we'll search for a subgroup in (B) awaiting for REV to be
     #   available, if such subgroup exist, we add REV to it and the subgroup is
     #   now awaiting for REV.parents() to be available.
     #
     #   3) finally if no such group existed in (B), we create a new subgroup.
     #
     #
     # To bootstrap the algorithm, we emit the tipmost revision (which
     # puts it in group (A) from above).
 
     revs.sort(reverse=True)
 
     # Set of parents of revision that have been emitted. They can be considered
     # unblocked as the graph generator is already aware of them so there is no
     # need to delay the revisions that reference them.
     #
     # If someone wants to prioritize a branch over the others, pre-filling this
     # set will force all other branches to wait until this branch is ready to be
     # emitted.
     unblocked = set(firstbranch)
 
     # list of groups waiting to be displayed, each group is defined by:
     #
     #   (revs:    lists of revs waiting to be displayed,
     #    blocked: set of that cannot be displayed before those in 'revs')
     #
     # The second value ('blocked') correspond to parents of any revision in the
     # group ('revs') that is not itself contained in the group. The main idea
     # of this algorithm is to delay as much as possible the emission of any
     # revision.  This means waiting for the moment we are about to display
     # these parents to display the revs in a group.
     #
     # This first implementation is smart until it encounters a merge: it will
     # emit revs as soon as any parent is about to be emitted and can grow an
     # arbitrary number of revs in 'blocked'. In practice this mean we properly
     # retains new branches but gives up on any special ordering for ancestors
     # of merges. The implementation can be improved to handle this better.
     #
     # The first subgroup is special. It corresponds to all the revision that
     # were already emitted. The 'revs' lists is expected to be empty and the
     # 'blocked' set contains the parents revisions of already emitted revision.
     #
     # You could pre-seed the <parents> set of groups[0] to a specific
     # changesets to select what the first emitted branch should be.
     groups = [([], unblocked)]
     pendingheap = []
     pendingset = set()
 
     heapq.heapify(pendingheap)
     heappop = heapq.heappop
     heappush = heapq.heappush
     for currentrev in revs:
         # Heap works with smallest element, we want highest so we invert
         if currentrev not in pendingset:
             heappush(pendingheap, -currentrev)
             pendingset.add(currentrev)
         # iterates on pending rev until after the current rev have been
         # processed.
         rev = None
         while rev != currentrev:
             rev = -heappop(pendingheap)
             pendingset.remove(rev)
 
             # Seek for a subgroup blocked, waiting for the current revision.
             matching = [i for i, g in enumerate(groups) if rev in g[1]]
 
             if matching:
                 # The main idea is to gather together all sets that are blocked
                 # on the same revision.
                 #
                 # Groups are merged when a common blocking ancestor is
                 # observed. For example, given two groups:
                 #
                 # revs [5, 4] waiting for 1
                 # revs [3, 2] waiting for 1
                 #
                 # These two groups will be merged when we process
                 # 1. In theory, we could have merged the groups when
                 # we added 2 to the group it is now in (we could have
                 # noticed the groups were both blocked on 1 then), but
                 # the way it works now makes the algorithm simpler.
                 #
                 # We also always keep the oldest subgroup first. We can
                 # probably improve the behavior by having the longest set
                 # first. That way, graph algorithms could minimise the length
                 # of parallel lines their drawing. This is currently not done.
                 targetidx = matching.pop(0)
                 trevs, tparents = groups[targetidx]
                 for i in matching:
                     gr = groups[i]
                     trevs.extend(gr[0])
                     tparents |= gr[1]
                 # delete all merged subgroups (except the one we kept)
                 # (starting from the last subgroup for performance and
                 # sanity reasons)
                 for i in reversed(matching):
                     del groups[i]
             else:
                 # This is a new head. We create a new subgroup for it.
                 targetidx = len(groups)
                 groups.append(([], {rev}))
 
             gr = groups[targetidx]
 
             # We now add the current nodes to this subgroups. This is done
             # after the subgroup merging because all elements from a subgroup
             # that relied on this rev must precede it.
             #
             # we also update the <parents> set to include the parents of the
             # new nodes.
             if rev == currentrev:  # only display stuff in rev
                 gr[0].append(rev)
             gr[1].remove(rev)
             parents = [p for p in parentsfunc(rev) if p > nullrev]
             gr[1].update(parents)
             for p in parents:
                 if p not in pendingset:
                     pendingset.add(p)
                     heappush(pendingheap, -p)
 
             # Look for a subgroup to display
             #
             # When unblocked is empty (if clause), we were not waiting for any
             # revisions during the first iteration (if no priority was given) or
             # if we emitted a whole disconnected set of the graph (reached a
             # root).  In that case we arbitrarily take the oldest known
             # subgroup. The heuristic could probably be better.
             #
             # Otherwise (elif clause) if the subgroup is blocked on
             # a revision we just emitted, we can safely emit it as
             # well.
             if not unblocked:
                 if len(groups) > 1:  # display other subset
                     targetidx = 1
                     gr = groups[1]
             elif not gr[1] & unblocked:
                 gr = None
 
             if gr is not None:
                 # update the set of awaited revisions with the one from the
                 # subgroup
                 unblocked |= gr[1]
                 # output all revisions in the subgroup
                 for r in gr[0]:
                     yield r
                 # delete the subgroup that you just output
                 # unless it is groups[0] in which case you just empty it.
                 if targetidx:
                     del groups[targetidx]
                 else:
                     gr[0][:] = []
     # Check if we have some subgroup waiting for revisions we are not going to
     # iterate over
     for g in groups:
         for r in g[0]:
             yield r
 
 
 def headrevs(revs, parentsfn):
     """Resolve the set of heads from a set of revisions.
 
     Receives an iterable of revision numbers and a callbable that receives a
     revision number and returns an iterable of parent revision numbers, possibly
     including nullrev.
 
     Returns a set of revision numbers that are DAG heads within the passed
     subset.
 
     ``nullrev`` is never included in the returned set, even if it is provided in
     the input set.
     """
     headrevs = set(revs)
     parents = {nullrev}
     up = parents.update
 
     for rev in revs:
         up(parentsfn(rev))
     headrevs.difference_update(parents)
     return headrevs
 
 
+def headrevsdiff(parentsfn, start, stop):
+    """Compute how the set of heads changed between
+    revisions `start-1` and `stop-1`.
+    """
+    parents = set()
+
+    heads_added = set()
+    heads_removed = set()
+
+    for rev in range(stop - 1, start - 1, -1):
+        if rev in parents:
+            parents.remove(rev)
+        else:
+            heads_added.add(rev)
+        for p in parentsfn(rev):
+            parents.add(p)
+
+    # now `parents` is the collection of candidate removed heads
+    rev = start - 1
+    while parents:
+        if rev in parents:
+            heads_removed.add(rev)
+            parents.remove(rev)
+
+        for p in parentsfn(rev):
+            parents.discard(p)
+        rev = rev - 1
+
+    return (heads_removed, heads_added)
+
+
 def headrevssubset(revsfn, parentrevsfn, startrev=None, stoprevs=None):
     """Returns the set of all revs that have no children with control.
 
     ``revsfn`` is a callable that with no arguments returns an iterator over
     all revision numbers in topological order. With a ``start`` argument, it
     returns revision numbers starting at that number.
 
     ``parentrevsfn`` is a callable receiving a revision number and returns an
     iterable of parent revision numbers, where values can include nullrev.
 
     ``startrev`` is a revision number at which to start the search.
 
     ``stoprevs`` is an iterable of revision numbers that, when encountered,
     will stop DAG traversal beyond them. Parents of revisions in this
     collection will be heads.
     """
     if startrev is None:
         startrev = nullrev
 
     stoprevs = set(stoprevs or [])
 
     reachable = {startrev}
     heads = {startrev}
 
     for rev in revsfn(start=startrev + 1):
         for prev in parentrevsfn(rev):
             if prev in reachable:
                 if rev not in stoprevs:
                     reachable.add(rev)
                 heads.add(rev)
 
             if prev in heads and prev not in stoprevs:
                 heads.remove(prev)
 
     return heads
 
 
 def linearize(revs, parentsfn):
     """Linearize and topologically sort a list of revisions.
 
     The linearization process tries to create long runs of revs where a child
     rev comes immediately after its first parent. This is done by visiting the
     heads of the revs in inverse topological order, and for each visited rev,
     visiting its second parent, then its first parent, then adding the rev
     itself to the output list.
 
     Returns a list of revision numbers.
     """
     visit = list(sorted(headrevs(revs, parentsfn), reverse=True))
     finished = set()
     result = []
 
     while visit:
         rev = visit.pop()
         if rev < 0:
             rev = -rev - 1
 
             if rev not in finished:
                 result.append(rev)
                 finished.add(rev)
 
         else:
             visit.append(-rev - 1)
 
             for prev in parentsfn(rev):
                 if prev == nullrev or prev not in revs or prev in finished:
                     continue
 
                 visit.append(prev)
 
     assert len(result) == len(revs)
 
     return result

mercurial/revlog.py

@@ -1,4065 +1,4077 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 functools
 import io
 import os
 import struct
 import weakref
 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 .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 hasattr(
     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 hasattr(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_rust(data, inline, default_header):
     cache = (0, data) if inline else None
     return rustrevlog.Index(data, default_header), 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 _Config:
     def copy(self):
         return self.__class__(**self.__dict__)
 
 
 @attr.s()
 class FeatureConfig(_Config):
     """Hold configuration values about the available revlog features"""
 
     # the default compression engine
     compression_engine = attr.ib(default=b'zlib')
     # compression engines options
     compression_engine_options = attr.ib(default=attr.Factory(dict))
 
     # can we use censor on this revlog
     censorable = attr.ib(default=False)
     # does this revlog use the "side data" feature
     has_side_data = attr.ib(default=False)
     # might remove rank configuration once the computation has no impact
     compute_rank = attr.ib(default=False)
     # 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.
     canonical_parent_order = attr.ib(default=False)
     # can ellipsis commit be used
     enable_ellipsis = attr.ib(default=False)
 
     def copy(self):
         new = super().copy()
         new.compression_engine_options = self.compression_engine_options.copy()
         return new
 
 
 @attr.s()
 class DataConfig(_Config):
     """Hold configuration value about how the revlog data are read"""
 
     # should we try to open the "pending" version of the revlog
     try_pending = attr.ib(default=False)
     # should we try to open the "splitted" version of the revlog
     try_split = attr.ib(default=False)
     #  When True, indexfile should be opened with checkambig=True at writing,
     #  to avoid file stat ambiguity.
     check_ambig = attr.ib(default=False)
 
     # If true, use mmap instead of reading to deal with large index
     mmap_large_index = attr.ib(default=False)
     # how much data is large
     mmap_index_threshold = attr.ib(default=None)
     # How much data to read and cache into the raw revlog data cache.
     chunk_cache_size = attr.ib(default=65536)
 
     # The size of the uncompressed cache compared to the largest revision seen.
     uncompressed_cache_factor = attr.ib(default=None)
 
     # The number of chunk cached
     uncompressed_cache_count = attr.ib(default=None)
 
     # Allow sparse reading of the revlog data
     with_sparse_read = attr.ib(default=False)
     # minimal density of a sparse read chunk
     sr_density_threshold = attr.ib(default=0.50)
     # minimal size of data we skip when performing sparse read
     sr_min_gap_size = attr.ib(default=262144)
 
     # are delta encoded against arbitrary bases.
     generaldelta = attr.ib(default=False)
 
 
 @attr.s()
 class DeltaConfig(_Config):
     """Hold configuration value about how new delta are computed
 
     Some attributes are duplicated from DataConfig to help havign each object
     self contained.
     """
 
     # can delta be encoded against arbitrary bases.
     general_delta = attr.ib(default=False)
     # Allow sparse writing of the revlog data
     sparse_revlog = attr.ib(default=False)
     # maximum length of a delta chain
     max_chain_len = attr.ib(default=None)
     # Maximum distance between delta chain base start and end
     max_deltachain_span = attr.ib(default=-1)
     # If `upper_bound_comp` is not None, this is the expected maximal gain from
     # compression for the data content.
     upper_bound_comp = attr.ib(default=None)
     # Should we try a delta against both parent
     delta_both_parents = attr.ib(default=True)
     # Test delta base candidate group by chunk of this maximal size.
     candidate_group_chunk_size = attr.ib(default=0)
     # Should we display debug information about delta computation
     debug_delta = attr.ib(default=False)
     # trust incoming delta by default
     lazy_delta = attr.ib(default=True)
     # trust the base of incoming delta by default
     lazy_delta_base = attr.ib(default=False)
 
 
 class _InnerRevlog:
     """An inner layer of the revlog object
 
     That layer exist to be able to delegate some operation to Rust, its
     boundaries are arbitrary and based on what we can delegate to Rust.
     """
 
     def __init__(
         self,
         opener,
         index,
         index_file,
         data_file,
         sidedata_file,
         inline,
         data_config,
         delta_config,
         feature_config,
         chunk_cache,
         default_compression_header,
     ):
         self.opener = opener
         self.index = index
 
         self.index_file = index_file
         self.data_file = data_file
         self.sidedata_file = sidedata_file
         self.inline = inline
         self.data_config = data_config
         self.delta_config = delta_config
         self.feature_config = feature_config
 
         # used during diverted write.
         self._orig_index_file = None
 
         self._default_compression_header = default_compression_header
 
         # index
 
         # 3-tuple of file handles being used for active writing.
         self._writinghandles = None
 
         self._segmentfile = randomaccessfile.randomaccessfile(
             self.opener,
             (self.index_file if self.inline else self.data_file),
             self.data_config.chunk_cache_size,
             chunk_cache,
         )
         self._segmentfile_sidedata = randomaccessfile.randomaccessfile(
             self.opener,
             self.sidedata_file,
             self.data_config.chunk_cache_size,
         )
 
         # revlog header -> revlog compressor
         self._decompressors = {}
         # 3-tuple of (node, rev, text) for a raw revision.
         self._revisioncache = None
 
         # cache some uncompressed chunks
         # rev → uncompressed_chunk
         #
         # the max cost is dynamically updated to be proportionnal to the
         # size of revision we actually encounter.
         self._uncompressed_chunk_cache = None
         if self.data_config.uncompressed_cache_factor is not None:
             self._uncompressed_chunk_cache = util.lrucachedict(
                 self.data_config.uncompressed_cache_count,
                 maxcost=65536,  # some arbitrary initial value
             )
 
         self._delay_buffer = None
 
     def __len__(self):
         return len(self.index)
 
     def clear_cache(self):
         assert not self.is_delaying
         self._revisioncache = None
         if self._uncompressed_chunk_cache is not None:
             self._uncompressed_chunk_cache.clear()
         self._segmentfile.clear_cache()
         self._segmentfile_sidedata.clear_cache()
 
     @property
     def canonical_index_file(self):
         if self._orig_index_file is not None:
             return self._orig_index_file
         return self.index_file
 
     @property
     def is_delaying(self):
         """is the revlog is currently delaying the visibility of written data?
 
         The delaying mechanism can be either in-memory or written on disk in a
         side-file."""
         return (self._delay_buffer is not None) or (
             self._orig_index_file is not None
         )
 
     # Derived from index values.
 
     def start(self, rev):
         """the offset of the data chunk for this revision"""
         return int(self.index[rev][0] >> 16)
 
     def length(self, rev):
         """the length of the data chunk for this revision"""
         return self.index[rev][1]
 
     def end(self, rev):
         """the end of the data chunk for this revision"""
         return self.start(rev) + self.length(rev)
 
     def deltaparent(self, rev):
         """return deltaparent of the given revision"""
         base = self.index[rev][3]
         if base == rev:
             return nullrev
         elif self.delta_config.general_delta:
             return base
         else:
             return rev - 1
 
     def issnapshot(self, rev):
         """tells whether rev is a snapshot"""
         if not self.delta_config.sparse_revlog:
             return self.deltaparent(rev) == nullrev
         elif hasattr(self.index, '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 _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.
         """
         generaldelta = self.delta_config.general_delta
         # Try C implementation.
         try:
             return self.index.deltachain(rev, stoprev, generaldelta)
         except AttributeError:
             pass
 
         chain = []
 
         # Alias to prevent attribute lookup in tight loop.
         index = self.index
 
         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
 
     @util.propertycache
     def _compressor(self):
         engine = util.compengines[self.feature_config.compression_engine]
         return engine.revlogcompressor(
             self.feature_config.compression_engine_options
         )
 
     @util.propertycache
     def _decompressor(self):
         """the default decompressor"""
         if self._default_compression_header is None:
             return None
         t = self._default_compression_header
         c = self._get_decompressor(t)
         return c.decompress
 
     def _get_decompressor(self, t):
         try:
             compressor = self._decompressors[t]
         except KeyError:
             try:
                 engine = util.compengines.forrevlogheader(t)
                 compressor = engine.revlogcompressor(
                     self.feature_config.compression_engine_options
                 )
                 self._decompressors[t] = compressor
             except KeyError:
                 raise error.RevlogError(
                     _(b'unknown compression type %s') % binascii.hexlify(t)
                 )
         return compressor
 
     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)
 
     @contextlib.contextmanager
     def reading(self):
         """Context manager that keeps data and sidedata files open for reading"""
         if len(self.index) == 0:
             yield  # nothing to be read
         elif self._delay_buffer is not None and self.inline:
             msg = "revlog with delayed write should not be inline"
             raise error.ProgrammingError(msg)
         else:
             with self._segmentfile.reading():
                 with self._segmentfile_sidedata.reading():
                     yield
 
     @property
     def is_writing(self):
         """True is a writing context is open"""
         return self._writinghandles is not None
 
     @property
     def is_open(self):
         """True if any file handle is being held
 
         Used for assert and debug in the python code"""
         return self._segmentfile.is_open or self._segmentfile_sidedata.is_open
 
     @contextlib.contextmanager
     def writing(self, transaction, data_end=None, sidedata_end=None):
         """Open the revlog files for writing
 
         Add content to a revlog should be done within such context.
         """
         if self.is_writing:
             yield
         else:
             ifh = dfh = sdfh = None
             try:
                 r = len(self.index)
                 # opening the data file.
                 dsize = 0
                 if r:
                     dsize = self.end(r - 1)
                 dfh = None
                 if not self.inline:
                     try:
                         dfh = self.opener(self.data_file, mode=b"r+")
                         if data_end is None:
                             dfh.seek(0, os.SEEK_END)
                         else:
                             dfh.seek(data_end, os.SEEK_SET)
                     except FileNotFoundError:
                         dfh = self.opener(self.data_file, mode=b"w+")
                     transaction.add(self.data_file, dsize)
                 if self.sidedata_file is not None:
                     assert sidedata_end is not None
                     # revlog-v2 does not inline, help Pytype
                     assert dfh is not None
                     try:
                         sdfh = self.opener(self.sidedata_file, mode=b"r+")
                         dfh.seek(sidedata_end, os.SEEK_SET)
                     except FileNotFoundError:
                         sdfh = self.opener(self.sidedata_file, mode=b"w+")
                     transaction.add(self.sidedata_file, sidedata_end)
 
                 # opening the index file.
                 isize = r * self.index.entry_size
                 ifh = self.__index_write_fp()
                 if self.inline:
                     transaction.add(self.index_file, dsize + isize)
                 else:
                     transaction.add(self.index_file, 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
             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 __index_write_fp(self, index_end=None):
         """internal method to open the index file for writing
 
         You should not use this directly and use `_writing` instead
         """
         try:
             if self._delay_buffer is None:
                 f = self.opener(
                     self.index_file,
                     mode=b"r+",
                     checkambig=self.data_config.check_ambig,
                 )
             else:
                 # check_ambig affect we way we open file for writing, however
                 # here, we do not actually open a file for writting as write
                 # will appened to a delay_buffer. So check_ambig is not
                 # meaningful and unneeded here.
                 f = randomaccessfile.appender(
                     self.opener, self.index_file, b"r+", self._delay_buffer
                 )
             if index_end is None:
                 f.seek(0, os.SEEK_END)
             else:
                 f.seek(index_end, os.SEEK_SET)
             return f
         except FileNotFoundError:
             if self._delay_buffer is None:
                 return self.opener(
                     self.index_file,
                     mode=b"w+",
                     checkambig=self.data_config.check_ambig,
                 )
             else:
                 return randomaccessfile.appender(
                     self.opener, self.index_file, b"w+", self._delay_buffer
                 )
 
     def __index_new_fp(self):
         """internal method to create a new index file for writing
 
         You should not use this unless you are upgrading from inline revlog
         """
         return self.opener(
             self.index_file,
             mode=b"w",
             checkambig=self.data_config.check_ambig,
         )
 
     def split_inline(self, tr, header, new_index_file_path=None):
         """split the data of an inline revlog into an index and a data file"""
         assert self._delay_buffer is None
         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
 
         new_dfh = self.opener(self.data_file, mode=b"w+")
         new_dfh.truncate(0)  # drop any potentially existing data
         try:
             with self.reading():
                 for r in range(len(self.index)):
                     new_dfh.write(self.get_segment_for_revs(r, r)[1])
                 new_dfh.flush()
 
             if new_index_file_path is not None:
                 self.index_file = new_index_file_path
             with self.__index_new_fp() as fp:
                 self.inline = False
                 for i in range(len(self.index)):
                     e = self.index.entry_binary(i)
                     if i == 0:
                         packed_header = self.index.pack_header(header)
                         e = packed_header + e
                     fp.write(e)
 
                 # If we don't use side-write, the temp file replace the real
                 # index when we exit the context manager
 
             self._segmentfile = randomaccessfile.randomaccessfile(
                 self.opener,
                 self.data_file,
                 self.data_config.chunk_cache_size,
             )
 
             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()
         return self.index_file
 
     def get_segment_for_revs(self, startrev, endrev):
         """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.
 
         API: we should consider making this a private part of the InnerRevlog
         at some point.
         """
         # 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)
 
     def _chunk(self, rev):
         """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.
         """
         if self._uncompressed_chunk_cache is not None:
             uncomp = self._uncompressed_chunk_cache.get(rev)
             if uncomp is not None:
                 return uncomp
 
         compression_mode = self.index[rev][10]
         data = self.get_segment_for_revs(rev, rev)[1]
         if compression_mode == COMP_MODE_PLAIN:
             uncomp = data
         elif compression_mode == COMP_MODE_DEFAULT:
             uncomp = self._decompressor(data)
         elif compression_mode == COMP_MODE_INLINE:
             uncomp = self.decompress(data)
         else:
             msg = b'unknown compression mode %d'
             msg %= compression_mode
             raise error.RevlogError(msg)
         if self._uncompressed_chunk_cache is not None:
             self._uncompressed_chunk_cache.insert(rev, uncomp, cost=len(uncomp))
         return uncomp
 
     def _chunks(self, revs, 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
 
         fetched_revs = []
         fadd = fetched_revs.append
 
         chunks = []
         ladd = chunks.append
 
         if self._uncompressed_chunk_cache is None:
             fetched_revs = revs
         else:
             for rev in revs:
                 cached_value = self._uncompressed_chunk_cache.get(rev)
                 if cached_value is None:
                     fadd(rev)
                 else:
                     ladd((rev, cached_value))
 
         if not fetched_revs:
             slicedchunks = ()
         elif not self.data_config.with_sparse_read:
             slicedchunks = (fetched_revs,)
         else:
             slicedchunks = deltautil.slicechunk(
                 self,
                 fetched_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.get_segment_for_revs(firstrev, lastrev)
             except OverflowError:
                 # issue4215 - we can't cache a run of chunks greater than
                 # 2G on Windows
                 for rev in revschunk:
                     ladd((rev, self._chunk(rev)))
 
             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:
                     c = c
                 elif comp_mode == COMP_MODE_INLINE:
                     c = decomp(c)
                 elif comp_mode == COMP_MODE_DEFAULT:
                     c = def_decomp(c)
                 else:
                     msg = b'unknown compression mode %d'
                     msg %= comp_mode
                     raise error.RevlogError(msg)
                 ladd((rev, c))
                 if self._uncompressed_chunk_cache is not None:
                     self._uncompressed_chunk_cache.insert(rev, c, len(c))
 
         chunks.sort()
         return [x[1] for x in chunks]
 
     def raw_text(self, node, rev):
         """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:
             cachedrev = self._revisioncache[1]
 
         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._inner._revisioncache after validating the text
         self._revisioncache = None
 
         targetsize = None
         rawsize = self.index[rev][2]
         if 0 <= rawsize:
             targetsize = 4 * rawsize
 
         if self._uncompressed_chunk_cache is not None:
             # dynamically update the uncompressed_chunk_cache size to the
             # largest revision we saw in this revlog.
             factor = self.data_config.uncompressed_cache_factor
             candidate_size = rawsize * factor
             if candidate_size > self._uncompressed_chunk_cache.maxcost:
                 self._uncompressed_chunk_cache.maxcost = candidate_size
 
         bins = self._chunks(chain, 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, sidedata_end):
         """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 sidedata_end < sidedata_offset + sidedata_size:
             filename = self.sidedata_file
             end = 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 write_entry(
         self,
         transaction,
         entry,
         data,
         link,
         offset,
         sidedata,
         sidedata_offset,
         index_end,
         data_end,
         sidedata_end,
     ):
         # 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 index_end is None:
             ifh.seek(0, os.SEEK_END)
         else:
             ifh.seek(index_end, os.SEEK_SET)
         if dfh:
             if data_end is None:
                 dfh.seek(0, os.SEEK_END)
             else:
                 dfh.seek(data_end, os.SEEK_SET)
         if sdfh:
             sdfh.seek(sidedata_end, os.SEEK_SET)
 
         curr = len(self.index) - 1
         if not self.inline:
             transaction.add(self.data_file, offset)
             if self.sidedata_file:
                 transaction.add(self.sidedata_file, sidedata_offset)
             transaction.add(self.canonical_index_file, curr * len(entry))
             if data[0]:
                 dfh.write(data[0])
             dfh.write(data[1])
             if sidedata:
                 sdfh.write(sidedata)
             if self._delay_buffer is None:
                 ifh.write(entry)
             else:
                 self._delay_buffer.append(entry)
         elif self._delay_buffer is not None:
             msg = b'invalid delayed write on inline revlog'
             raise error.ProgrammingError(msg)
         else:
             offset += curr * self.index.entry_size
             transaction.add(self.canonical_index_file, offset)
             assert not sidedata
             ifh.write(entry)
             ifh.write(data[0])
             ifh.write(data[1])
         return (
             ifh.tell(),
             dfh.tell() if dfh else None,
             sdfh.tell() if sdfh else None,
         )
 
     def _divert_index(self):
         return self.index_file + b'.a'
 
     def delay(self):
         assert not self.is_open
         if self.inline:
             msg = "revlog with delayed write should not be inline"
             raise error.ProgrammingError(msg)
         if self._delay_buffer is not None or self._orig_index_file is not None:
             # delay or divert already in place
             return None
         elif len(self.index) == 0:
             self._orig_index_file = self.index_file
             self.index_file = self._divert_index()
             assert self._orig_index_file is not None
             assert self.index_file is not None
             if self.opener.exists(self.index_file):
                 self.opener.unlink(self.index_file)
             return self.index_file
         else:
             self._delay_buffer = []
             return None
 
     def write_pending(self):
         assert not self.is_open
         if self.inline:
             msg = "revlog with delayed write should not be inline"
             raise error.ProgrammingError(msg)
         if self._orig_index_file is not None:
             return None, True
         any_pending = False
         pending_index_file = self._divert_index()
         if self.opener.exists(pending_index_file):
             self.opener.unlink(pending_index_file)
         util.copyfile(
             self.opener.join(self.index_file),
             self.opener.join(pending_index_file),
         )
         if self._delay_buffer:
             with self.opener(pending_index_file, b'r+') as ifh:
                 ifh.seek(0, os.SEEK_END)
                 ifh.write(b"".join(self._delay_buffer))
             any_pending = True
         self._delay_buffer = None
         self._orig_index_file = self.index_file
         self.index_file = pending_index_file
         return self.index_file, any_pending
 
     def finalize_pending(self):
         assert not self.is_open
         if self.inline:
             msg = "revlog with delayed write should not be inline"
             raise error.ProgrammingError(msg)
 
         delay = self._delay_buffer is not None
         divert = self._orig_index_file is not None
 
         if delay and divert:
             assert False, "unreachable"
         elif delay:
             if self._delay_buffer:
                 with self.opener(self.index_file, b'r+') as ifh:
                     ifh.seek(0, os.SEEK_END)
                     ifh.write(b"".join(self._delay_buffer))
             self._delay_buffer = None
         elif divert:
             if self.opener.exists(self.index_file):
                 self.opener.rename(
                     self.index_file,
                     self._orig_index_file,
                     checkambig=True,
                 )
             self.index_file = self._orig_index_file
             self._orig_index_file = None
         else:
             msg = b"not delay or divert found on this revlog"
             raise error.ProgrammingError(msg)
         return self.canonical_index_file
 
 
 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
 
     @staticmethod
     def is_inline_index(header_bytes):
         """Determine if a revlog is inline from the initial bytes of the index"""
         if len(header_bytes) == 0:
             return True
 
         header = INDEX_HEADER.unpack(header_bytes)[0]
 
         _format_flags = header & ~0xFFFF
         _format_version = header & 0xFFFF
 
         features = FEATURES_BY_VERSION[_format_version]
         return features[b'inline'](_format_flags)
 
     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,
         data_config=None,
         delta_config=None,
         feature_config=None,
         may_inline=True,  # may inline new revlog
     ):
         """
         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.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._may_inline = may_inline
         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
         if feature_config is not None:
             self.feature_config = feature_config.copy()
         elif b'feature-config' in self.opener.options:
             self.feature_config = self.opener.options[b'feature-config'].copy()
         else:
             self.feature_config = FeatureConfig()
         self.feature_config.censorable = censorable
         self.feature_config.canonical_parent_order = canonical_parent_order
         if data_config is not None:
             self.data_config = data_config.copy()
         elif b'data-config' in self.opener.options:
             self.data_config = self.opener.options[b'data-config'].copy()
         else:
             self.data_config = DataConfig()
         self.data_config.check_ambig = checkambig
         self.data_config.mmap_large_index = mmaplargeindex
         if delta_config is not None:
             self.delta_config = delta_config.copy()
         elif b'delta-config' in self.opener.options:
             self.delta_config = self.opener.options[b'delta-config'].copy()
         else:
             self.delta_config = DeltaConfig()
         self.delta_config.upper_bound_comp = upperboundcomp
 
         # Maps rev to chain base rev.
         self._chainbasecache = util.lrucachedict(100)
 
         self.index = None
         self._docket = None
         self._nodemap_docket = None
         # Mapping of partial identifiers to full nodes.
         self._pcache = {}
 
         # other optionnals features
 
         # Make copy of flag processors so each revlog instance can support
         # custom flags.
         self._flagprocessors = dict(flagutil.flagprocessors)
         # prevent nesting of addgroup
         self._adding_group = None
 
         chunk_cache = self._loadindex()
         self._load_inner(chunk_cache)
         self._concurrencychecker = concurrencychecker
 
     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
         """
         opts = self.opener.options
 
         if b'changelogv2' in opts and self.revlog_kind == KIND_CHANGELOG:
             new_header = CHANGELOGV2
             compute_rank = opts.get(b'changelogv2.compute-rank', True)
             self.feature_config.compute_rank = compute_rank
         elif b'revlogv2' in opts:
             new_header = REVLOGV2
         elif b'revlogv1' in opts:
             new_header = REVLOGV1
             if self._may_inline:
                 new_header |= 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
 
         mmapindexthreshold = None
         if self.data_config.mmap_large_index:
             mmapindexthreshold = self.data_config.mmap_index_threshold
         if self.feature_config.enable_ellipsis:
             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)
 
         chunk_cache_size = self.data_config.chunk_cache_size
         if chunk_cache_size <= 0:
             raise error.RevlogError(
                 _(b'revlog chunk cache size %r is not greater than 0')
                 % chunk_cache_size
             )
         elif chunk_cache_size & (chunk_cache_size - 1):
             raise error.RevlogError(
                 _(b'revlog chunk cache size %r is not a power of 2')
                 % chunk_cache_size
             )
         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 get_streams(self, max_linkrev, force_inline=False):
         """return a list of streams that represent this revlog
 
         This is used by stream-clone to do bytes to bytes copies of a repository.
 
         This streams data for all revisions that refer to a changelog revision up
         to `max_linkrev`.
 
         If `force_inline` is set, it enforces that the stream will represent an inline revlog.
 
         It returns is a list of three-tuple:
 
             [
                 (filename, bytes_stream, stream_size),
                 …
             ]
         """
         n = len(self)
         index = self.index
         while n > 0:
             linkrev = index[n - 1][4]
             if linkrev < max_linkrev:
                 break
             # note: this loop will rarely go through multiple iterations, since
             # it only traverses commits created during the current streaming
             # pull operation.
             #
             # If this become a problem, using a binary search should cap the
             # runtime of this.
             n = n - 1
         if n == 0:
             # no data to send
             return []
         index_size = n * index.entry_size
         data_size = self.end(n - 1)
 
         # XXX we might have been split (or stripped) since the object
         # initialization, We need to close this race too, but having a way to
         # pre-open the file we feed to the revlog and never closing them before
         # we are done streaming.
 
         if self._inline:
 
             def get_stream():
                 with self.opener(self._indexfile, mode=b"r") as fp:
                     yield None
                     size = index_size + data_size
                     if size <= 65536:
                         yield fp.read(size)
                     else:
                         yield from util.filechunkiter(fp, limit=size)
 
             inline_stream = get_stream()
             next(inline_stream)
             return [
                 (self._indexfile, inline_stream, index_size + data_size),
             ]
         elif force_inline:
 
             def get_stream():
                 with self.reading():
                     yield None
 
                     for rev in range(n):
                         idx = self.index.entry_binary(rev)
                         if rev == 0 and self._docket is None:
                             # re-inject the inline flag
                             header = self._format_flags
                             header |= self._format_version
                             header |= FLAG_INLINE_DATA
                             header = self.index.pack_header(header)
                             idx = header + idx
                         yield idx
                         yield self._inner.get_segment_for_revs(rev, rev)[1]
 
             inline_stream = get_stream()
             next(inline_stream)
             return [
                 (self._indexfile, inline_stream, index_size + data_size),
             ]
         else:
 
             def get_index_stream():
                 with self.opener(self._indexfile, mode=b"r") as fp:
                     yield None
                     if index_size <= 65536:
                         yield fp.read(index_size)
                     else:
                         yield from util.filechunkiter(fp, limit=index_size)
 
             def get_data_stream():
                 with self._datafp() as fp:
                     yield None
                     if data_size <= 65536:
                         yield fp.read(data_size)
                     else:
                         yield from util.filechunkiter(fp, limit=data_size)
 
             index_stream = get_index_stream()
             next(index_stream)
             data_stream = get_data_stream()
             next(data_stream)
             return [
                 (self._datafile, data_stream, data_size),
                 (self._indexfile, index_stream, index_size),
             ]
 
     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(self._split_index_file):
             entry_point = self._split_index_file
         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.delta_config.general_delta = features[b'generaldelta'](
                 self._format_flags
             )
             self.feature_config.has_side_data = 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.delta_config.general_delta = 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.delta_config.general_delta:
             self.delta_config.sparse_revlog = 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:
                 # Using the CIndex is not longer possible, as the
                 # `AncestorsIterator` and `LazyAncestors` classes now require
                 # a Rust index for instantiation.
                 use_rust_index = True
 
         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 = functools.partial(
                 parse_index_v1_rust, default_header=new_header
             )
         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 hasattr(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
         # revnum -> (chain-length, sum-delta-length)
         self._chaininfocache = util.lrucachedict(500)
 
         return chunkcache
 
     def _load_inner(self, chunk_cache):
         if self._docket is None:
             default_compression_header = None
         else:
             default_compression_header = self._docket.default_compression_header
 
         self._inner = _InnerRevlog(
             opener=self.opener,
             index=self.index,
             index_file=self._indexfile,
             data_file=self._datafile,
             sidedata_file=self._sidedatafile,
             inline=self._inline,
             data_config=self.data_config,
             delta_config=self.delta_config,
             feature_config=self.feature_config,
             chunk_cache=chunk_cache,
             default_compression_header=default_compression_header,
         )
 
     def get_revlog(self):
         """simple function to mirror API of other not-really-revlog API"""
         return self
 
     @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 _datafp(self, mode=b'r'):
         """file object for the revlog's data file"""
         return self.opener(self._datafile, mode=mode)
 
     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):
         """update on disk cache
 
         If a transaction is passed, the update may be delayed to transaction
         commit."""
         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):
         """Clear in-memory caches"""
         self._chainbasecache.clear()
         self._inner.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 hasattr(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):
         """return the revision number associated with a <nodeid>"""
         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.feature_config.has_side_data:
             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.feature_config.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.feature_config.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.delta_config.general_delta
         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):
         return self._inner._deltachain(rev, stoprev=stoprev)
 
     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 headrevsdiff(self, start, stop):
+        try:
+            return self.index.headrevsdiff(start, stop)
+        except AttributeError:
+            return dagop.headrevsdiff(self._uncheckedparentrevs, start, stop)
+
     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 _head_node_ids(self):
         try:
             return self.index.head_node_ids()
         except AttributeError:
             return [self.node(r) for r in self.headrevs()]
 
     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._head_node_ids()
         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 diffheads(self, start, stop):
+        """return the nodes that make up the difference between
+        heads of revs before `start` and heads of revs before `stop`"""
+        removed, added = self.headrevsdiff(start, stop)
+        return [self.node(r) for r in removed], [self.node(r) for r in added]
+
     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 deltaparent(self, rev):
         """return deltaparent of the given revision"""
         base = self.index[rev][3]
         if base == rev:
             return nullrev
         elif self.delta_config.general_delta:
             return base
         else:
             return rev - 1
 
     def issnapshot(self, rev):
         """tells whether rev is a snapshot"""
         ret = self._inner.issnapshot(rev)
         self.issnapshot = self._inner.issnapshot
         return ret
 
     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._inner._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._inner._chunk(rev2))
 
         return mdiff.textdiff(self.rawdata(rev1), self.rawdata(rev2))
 
     def revision(self, nodeorrev):
         """return an uncompressed revision of a given node or revision
         number.
         """
         return self._revisiondata(nodeorrev)
 
     def sidedata(self, nodeorrev):
         """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 _rawtext(self, node, rev):
         """return the possibly unvalidated rawtext for a revision
 
         returns (rev, rawtext, validated)
         """
         # Check if we have the entry in cache
         # The cache entry looks like (node, rev, rawtext)
         if self._inner._revisioncache:
             if self._inner._revisioncache[0] == node:
                 return (rev, self._inner._revisioncache[2], True)
 
         if rev is None:
             rev = self.rev(node)
 
         return self._inner.raw_text(node, rev)
 
     def _revisiondata(self, nodeorrev, 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)
 
         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._inner._revisioncache = (node, rev, rawtext)
 
         return text
 
     def _sidedata(self, rev):
         """Return the sidedata for a given revision number."""
         sidedata_end = None
         if self._docket is not None:
             sidedata_end = self._docket.sidedata_end
         return self._inner.sidedata(rev, sidedata_end)
 
     def rawdata(self, nodeorrev):
         """return an uncompressed raw data of a given node or revision number."""
         return self._revisiondata(nodeorrev, 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._inner._revisioncache
                     and self._inner._revisioncache[0] == node
                 ):
                     self._inner._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.feature_config.censorable and storageutil.iscensoredtext(
                 text
             ):
                 raise error.CensoredNodeError(self.display_id, node, text)
             raise
 
     @property
     def _split_index_file(self):
         """the path where to expect the index of an ongoing splitting operation
 
         The file will only exist if a splitting operation is in progress, but
         it is always expected at the same location."""
         parts = self.radix.split(b'/')
         if len(parts) > 1:
             # adds a '-s' prefix to the ``data/` or `meta/` base
             head = parts[0] + b'-s'
             mids = parts[1:-1]
             tail = parts[-1] + b'.i'
             pieces = [head] + mids + [tail]
             return b'/'.join(pieces)
         else:
             # the revlog is stored at the root of the store (changelog or
             # manifest), no risk of collision.
             return self.radix + b'.i.s'
 
     def _enforceinlinesize(self, tr):
         """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 (self._may_inline and total_size < _maxinline):
             return
 
         if self._docket is not None:
             msg = b"inline revlog should not have a docket"
             raise error.ProgrammingError(msg)
 
         # In the common case, we enforce inline size because the revlog has
         # been appened too. And in such case, it must have an initial offset
         # recorded in the transaction.
         troffset = tr.findoffset(self._inner.canonical_index_file)
         pre_touched = troffset is not None
         if not pre_touched and self.target[0] != KIND_CHANGELOG:
             raise error.RevlogError(
                 _(b"%s not found in the transaction") % self._indexfile
             )
 
         tr.addbackup(self._inner.canonical_index_file, for_offset=pre_touched)
         tr.add(self._datafile, 0)
 
         new_index_file_path = None
         old_index_file_path = self._indexfile
         new_index_file_path = self._split_index_file
         opener = self.opener
         weak_self = weakref.ref(self)
 
         # 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,
             )
             maybe_self = weak_self()
             if maybe_self is not None:
                 maybe_self._indexfile = old_index_file_path
                 maybe_self._inner.index_file = maybe_self._indexfile
 
         def abort_callback(tr):
             maybe_self = weak_self()
             if maybe_self is not None:
                 maybe_self._indexfile = old_index_file_path
                 maybe_self._inner.inline = True
                 maybe_self._inner.index_file = old_index_file_path
 
         tr.registertmp(new_index_file_path)
         if self.target[1] is not None:
             callback_id = b'000-revlog-split-%d-%s' % self.target
         else:
             callback_id = b'000-revlog-split-%d' % self.target[0]
         tr.addfinalize(callback_id, finalize_callback)
         tr.addabort(callback_id, abort_callback)
 
         self._format_flags &= ~FLAG_INLINE_DATA
         self._inner.split_inline(
             tr,
             self._format_flags | self._format_version,
             new_index_file_path=new_index_file_path,
         )
 
         self._inline = False
         if new_index_file_path is not None:
             self._indexfile = new_index_file_path
 
         nodemaputil.setup_persistent_nodemap(tr, self)
 
     def _nodeduplicatecallback(self, transaction, node):
         """called when trying to add a node already stored."""
 
     @contextlib.contextmanager
     def reading(self):
         with self._inner.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._inner.is_writing:
             yield
         else:
             data_end = None
             sidedata_end = None
             if self._docket is not None:
                 data_end = self._docket.data_end
                 sidedata_end = self._docket.sidedata_end
             with self._inner.writing(
                 transaction,
                 data_end=data_end,
                 sidedata_end=sidedata_end,
             ):
                 yield
                 if self._docket is not None:
                     self._write_docket(transaction)
 
     @property
     def is_delaying(self):
         return self._inner.is_delaying
 
     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.feature_config.has_side_data:
             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):
         return self._inner.compress(data)
 
     def decompress(self, data):
         return self._inner.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._inner._writinghandles is None:
             msg = b'adding revision outside `revlog._writing` context'
             raise error.ProgrammingError(msg)
 
         btext = [rawtext]
 
         curr = len(self)
         prev = curr - 1
 
         offset = self._get_data_offset(prev)
 
         if self._concurrencychecker:
             ifh, dfh, sdfh = self._inner._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.delta_config.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.delta_config.general_delta
                 and self.delta_config.lazy_delta_base
             ):
                 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)
 
         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.feature_config.has_side_data:
             sidedata_compression_mode = COMP_MODE_PLAIN
             serialized_sidedata = sidedatautil.serialize_sidedata(sidedata)
             sidedata_offset = self._docket.sidedata_end
             h, comp_sidedata = self._inner.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.feature_config.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)
 
         if type(rawtext) == bytes:  # only accept immutable objects
             self._inner._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.
         index_end = data_end = sidedata_end = None
         if self._docket is not None:
             index_end = self._docket.index_end
             data_end = self._docket.data_end
             sidedata_end = self._docket.sidedata_end
 
         files_end = self._inner.write_entry(
             transaction,
             entry,
             data,
             link,
             offset,
             sidedata,
             sidedata_offset,
             index_end,
             data_end,
             sidedata_end,
         )
         self._enforceinlinesize(transaction)
         if self._docket is not None:
             self._docket.index_end = files_end[0]
             self._docket.data_end = files_end[1]
             self._docket.sidedata_end = files_end[2]
 
         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.delta_config.general_delta
                 and self.delta_config.lazy_delta_base
             ):
                 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.delta_config.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.feature_config.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.feature_config.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._chaininfocache = util.lrucachedict(500)
         self._inner.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):
         """return list of files that compose this revlog"""
         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.delta_config.general_delta:
             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.
         old_delta_config = destrevlog.delta_config
         destrevlog.delta_config = destrevlog.delta_config.copy()
 
         try:
             if deltareuse == self.DELTAREUSEALWAYS:
                 destrevlog.delta_config.lazy_delta_base = True
                 destrevlog.delta_config.lazy_delta = True
             elif deltareuse == self.DELTAREUSESAMEREVS:
                 destrevlog.delta_config.lazy_delta_base = False
                 destrevlog.delta_config.lazy_delta = True
             elif deltareuse == self.DELTAREUSENEVER:
                 destrevlog.delta_config.lazy_delta_base = False
                 destrevlog.delta_config.lazy_delta = False
 
             delta_both_parents = (
                 forcedeltabothparents or old_delta_config.delta_both_parents
             )
             destrevlog.delta_config.delta_both_parents = delta_both_parents
 
             with self.reading(), destrevlog._writing(tr):
                 self._clone(
                     tr,
                     destrevlog,
                     addrevisioncb,
                     deltareuse,
                     forcedeltabothparents,
                     sidedata_helpers,
                 )
 
         finally:
             destrevlog.delta_config = old_delta_config
 
     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.delta_config.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.delta_config.lazy_delta:
                     dp = self.deltaparent(rev)
                     if dp != nullrev:
                         cachedelta = (dp, bytes(self._inner._chunk(rev)))
 
                 sidedata = None
                 if not cachedelta:
                     try:
                         rawtext = self._revisiondata(rev)
                     except error.CensoredNodeError as censored:
                         assert flags & REVIDX_ISCENSORED
                         rawtext = censored.tombstone
                     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]
 
                 destrevlog._addrevision(
                     node,
                     rawtext,
                     tr,
                     linkrev,
                     p1,
                     p2,
                     flags,
                     cachedelta,
                     deltacomputer=deltacomputer,
                     sidedata=sidedata,
                 )
 
             if addrevisioncb:
                 addrevisioncb(self, rev, node)
 
     def censorrevision(self, tr, censor_nodes, 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, censor_nodes, tombstone)
         else:
             rewrite.v2_censor(self, tr, censor_nodes, 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.feature_config.has_side_data:
             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._inner._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.feature_config.has_side_data:
                     sidedata_compression_mode = COMP_MODE_PLAIN
                     h, comp_sidedata = self._inner.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)