revlog: add the option to track the expected compression upper bound

There are various optimization we can do if we can estimate the size of delta
before actually spending CPU compressing them. So we add a attributed dedicated
to tracking that.

We only use it on Manifest because (1) it structure is quite stable across all
Mercurial repository so its compression ratio is fairly universal. This is the
revlog with most extreme delta (cf the sparse-revlog optimization).

This will be put to use in later changesets.

Right now the compression upper bound is set to 10. This is a fairly
conservative value (observed value is more around 3), but I prefer to be safe
while introducing the optimization principles. We can tune the optimization
threshold later.

contrib/perf.py

@@ -1,3059 +1,3063 b''
 # perf.py - performance test routines
 '''helper extension to measure performance
 
 Configurations
 ==============
 
 ``perf``
 --------
 
 ``all-timing``
     When set, additional statistics will be reported for each benchmark: best,
     worst, median average. If not set only the best timing is reported
     (default: off).
 
 ``presleep``
   number of second to wait before any group of runs (default: 1)
 
 ``pre-run``
   number of run to perform before starting measurement.
 
 ``profile-benchmark``
   Enable profiling for the benchmarked section.
   (The first iteration is benchmarked)
 
 ``run-limits``
   Control the number of runs each benchmark will perform. The option value
   should be a list of `<time>-<numberofrun>` pairs. After each run the
   conditions are considered in order with the following logic:
 
       If benchmark has been running for <time> seconds, and we have performed
       <numberofrun> iterations, stop the benchmark,
 
   The default value is: `3.0-100, 10.0-3`
 
 ``stub``
     When set, benchmarks will only be run once, useful for testing
     (default: off)
 '''
 
 # "historical portability" policy of perf.py:
 #
 # We have to do:
 # - make perf.py "loadable" with as wide Mercurial version as possible
 #   This doesn't mean that perf commands work correctly with that Mercurial.
 #   BTW, perf.py itself has been available since 1.1 (or eb240755386d).
 # - make historical perf command work correctly with as wide Mercurial
 #   version as possible
 #
 # We have to do, if possible with reasonable cost:
 # - make recent perf command for historical feature work correctly
 #   with early Mercurial
 #
 # We don't have to do:
 # - make perf command for recent feature work correctly with early
 #   Mercurial
 
 from __future__ import absolute_import
 import contextlib
 import functools
 import gc
 import os
 import random
 import shutil
 import struct
 import sys
 import tempfile
 import threading
 import time
 from mercurial import (
     changegroup,
     cmdutil,
     commands,
     copies,
     error,
     extensions,
     hg,
     mdiff,
     merge,
     revlog,
     util,
 )
 
 # for "historical portability":
 # try to import modules separately (in dict order), and ignore
 # failure, because these aren't available with early Mercurial
 try:
     from mercurial import branchmap # since 2.5 (or bcee63733aad)
 except ImportError:
     pass
 try:
     from mercurial import obsolete # since 2.3 (or ad0d6c2b3279)
 except ImportError:
     pass
 try:
     from mercurial import registrar # since 3.7 (or 37d50250b696)
     dir(registrar) # forcibly load it
 except ImportError:
     registrar = None
 try:
     from mercurial import repoview # since 2.5 (or 3a6ddacb7198)
 except ImportError:
     pass
 try:
     from mercurial.utils import repoviewutil # since 5.0
 except ImportError:
     repoviewutil = None
 try:
     from mercurial import scmutil # since 1.9 (or 8b252e826c68)
 except ImportError:
     pass
 try:
     from mercurial import setdiscovery # since 1.9 (or cb98fed52495)
 except ImportError:
     pass
 
 try:
     from mercurial import profiling
 except ImportError:
     profiling = None
 
 def identity(a):
     return a
 
 try:
     from mercurial import pycompat
     getargspec = pycompat.getargspec  # added to module after 4.5
     _byteskwargs = pycompat.byteskwargs  # since 4.1 (or fbc3f73dc802)
     _sysstr = pycompat.sysstr         # since 4.0 (or 2219f4f82ede)
     _xrange = pycompat.xrange         # since 4.8 (or 7eba8f83129b)
     fsencode = pycompat.fsencode      # since 3.9 (or f4a5e0e86a7e)
     if pycompat.ispy3:
         _maxint = sys.maxsize  # per py3 docs for replacing maxint
     else:
         _maxint = sys.maxint
 except (ImportError, AttributeError):
     import inspect
     getargspec = inspect.getargspec
     _byteskwargs = identity
     fsencode = identity               # no py3 support
     _maxint = sys.maxint              # no py3 support
     _sysstr = lambda x: x             # no py3 support
     _xrange = xrange
 
 try:
     # 4.7+
     queue = pycompat.queue.Queue
 except (AttributeError, ImportError):
     # <4.7.
     try:
         queue = pycompat.queue
     except (AttributeError, ImportError):
         queue = util.queue
 
 try:
     from mercurial import logcmdutil
     makelogtemplater = logcmdutil.maketemplater
 except (AttributeError, ImportError):
     try:
         makelogtemplater = cmdutil.makelogtemplater
     except (AttributeError, ImportError):
         makelogtemplater = None
 
 # for "historical portability":
 # define util.safehasattr forcibly, because util.safehasattr has been
 # available since 1.9.3 (or 94b200a11cf7)
 _undefined = object()
 def safehasattr(thing, attr):
     return getattr(thing, _sysstr(attr), _undefined) is not _undefined
 setattr(util, 'safehasattr', safehasattr)
 
 # for "historical portability":
 # define util.timer forcibly, because util.timer has been available
 # since ae5d60bb70c9
 if safehasattr(time, 'perf_counter'):
     util.timer = time.perf_counter
 elif os.name == b'nt':
     util.timer = time.clock
 else:
     util.timer = time.time
 
 # for "historical portability":
 # use locally defined empty option list, if formatteropts isn't
 # available, because commands.formatteropts has been available since
 # 3.2 (or 7a7eed5176a4), even though formatting itself has been
 # available since 2.2 (or ae5f92e154d3)
 formatteropts = getattr(cmdutil, "formatteropts",
                         getattr(commands, "formatteropts", []))
 
 # for "historical portability":
 # use locally defined option list, if debugrevlogopts isn't available,
 # because commands.debugrevlogopts has been available since 3.7 (or
 # 5606f7d0d063), even though cmdutil.openrevlog() has been available
 # since 1.9 (or a79fea6b3e77).
 revlogopts = getattr(cmdutil, "debugrevlogopts",
                      getattr(commands, "debugrevlogopts", [
         (b'c', b'changelog', False, (b'open changelog')),
         (b'm', b'manifest', False, (b'open manifest')),
         (b'', b'dir', False, (b'open directory manifest')),
         ]))
 
 cmdtable = {}
 
 # for "historical portability":
 # define parsealiases locally, because cmdutil.parsealiases has been
 # available since 1.5 (or 6252852b4332)
 def parsealiases(cmd):
     return cmd.split(b"|")
 
 if safehasattr(registrar, 'command'):
     command = registrar.command(cmdtable)
 elif safehasattr(cmdutil, 'command'):
     command = cmdutil.command(cmdtable)
     if b'norepo' not in getargspec(command).args:
         # for "historical portability":
         # wrap original cmdutil.command, because "norepo" option has
         # been available since 3.1 (or 75a96326cecb)
         _command = command
         def command(name, options=(), synopsis=None, norepo=False):
             if norepo:
                 commands.norepo += b' %s' % b' '.join(parsealiases(name))
             return _command(name, list(options), synopsis)
 else:
     # for "historical portability":
     # define "@command" annotation locally, because cmdutil.command
     # has been available since 1.9 (or 2daa5179e73f)
     def command(name, options=(), synopsis=None, norepo=False):
         def decorator(func):
             if synopsis:
                 cmdtable[name] = func, list(options), synopsis
             else:
                 cmdtable[name] = func, list(options)
             if norepo:
                 commands.norepo += b' %s' % b' '.join(parsealiases(name))
             return func
         return decorator
 
 try:
     import mercurial.registrar
     import mercurial.configitems
     configtable = {}
     configitem = mercurial.registrar.configitem(configtable)
     configitem(b'perf', b'presleep',
         default=mercurial.configitems.dynamicdefault,
     )
     configitem(b'perf', b'stub',
         default=mercurial.configitems.dynamicdefault,
     )
     configitem(b'perf', b'parentscount',
         default=mercurial.configitems.dynamicdefault,
     )
     configitem(b'perf', b'all-timing',
         default=mercurial.configitems.dynamicdefault,
     )
     configitem(b'perf', b'pre-run',
         default=mercurial.configitems.dynamicdefault,
     )
     configitem(b'perf', b'profile-benchmark',
         default=mercurial.configitems.dynamicdefault,
     )
     configitem(b'perf', b'run-limits',
         default=mercurial.configitems.dynamicdefault,
     )
 except (ImportError, AttributeError):
     pass
 
 def getlen(ui):
     if ui.configbool(b"perf", b"stub", False):
         return lambda x: 1
     return len
 
 class noop(object):
     """dummy context manager"""
     def __enter__(self):
         pass
     def __exit__(self, *args):
         pass
 
 NOOPCTX = noop()
 
 def gettimer(ui, opts=None):
     """return a timer function and formatter: (timer, formatter)
 
     This function exists to gather the creation of formatter in a single
     place instead of duplicating it in all performance commands."""
 
     # enforce an idle period before execution to counteract power management
     # experimental config: perf.presleep
     time.sleep(getint(ui, b"perf", b"presleep", 1))
 
     if opts is None:
         opts = {}
     # redirect all to stderr unless buffer api is in use
     if not ui._buffers:
         ui = ui.copy()
         uifout = safeattrsetter(ui, b'fout', ignoremissing=True)
         if uifout:
             # for "historical portability":
             # ui.fout/ferr have been available since 1.9 (or 4e1ccd4c2b6d)
             uifout.set(ui.ferr)
 
     # get a formatter
     uiformatter = getattr(ui, 'formatter', None)
     if uiformatter:
         fm = uiformatter(b'perf', opts)
     else:
         # for "historical portability":
         # define formatter locally, because ui.formatter has been
         # available since 2.2 (or ae5f92e154d3)
         from mercurial import node
         class defaultformatter(object):
             """Minimized composition of baseformatter and plainformatter
             """
             def __init__(self, ui, topic, opts):
                 self._ui = ui
                 if ui.debugflag:
                     self.hexfunc = node.hex
                 else:
                     self.hexfunc = node.short
             def __nonzero__(self):
                 return False
             __bool__ = __nonzero__
             def startitem(self):
                 pass
             def data(self, **data):
                 pass
             def write(self, fields, deftext, *fielddata, **opts):
                 self._ui.write(deftext % fielddata, **opts)
             def condwrite(self, cond, fields, deftext, *fielddata, **opts):
                 if cond:
                     self._ui.write(deftext % fielddata, **opts)
             def plain(self, text, **opts):
                 self._ui.write(text, **opts)
             def end(self):
                 pass
         fm = defaultformatter(ui, b'perf', opts)
 
     # stub function, runs code only once instead of in a loop
     # experimental config: perf.stub
     if ui.configbool(b"perf", b"stub", False):
         return functools.partial(stub_timer, fm), fm
 
     # experimental config: perf.all-timing
     displayall = ui.configbool(b"perf", b"all-timing", False)
 
     # experimental config: perf.run-limits
     limitspec = ui.configlist(b"perf", b"run-limits", [])
     limits = []
     for item in limitspec:
         parts = item.split(b'-', 1)
         if len(parts) < 2:
             ui.warn((b'malformatted run limit entry, missing "-": %s\n'
                      % item))
             continue
         try:
             time_limit = float(pycompat.sysstr(parts[0]))
         except ValueError as e:
             ui.warn((b'malformatted run limit entry, %s: %s\n'
                      % (pycompat.bytestr(e), item)))
             continue
         try:
             run_limit = int(pycompat.sysstr(parts[1]))
         except ValueError as e:
             ui.warn((b'malformatted run limit entry, %s: %s\n'
                      % (pycompat.bytestr(e), item)))
             continue
         limits.append((time_limit, run_limit))
     if not limits:
         limits = DEFAULTLIMITS
 
     profiler = None
     if profiling is not None:
         if ui.configbool(b"perf", b"profile-benchmark", False):
             profiler = profiling.profile(ui)
 
     prerun = getint(ui, b"perf", b"pre-run", 0)
     t = functools.partial(_timer, fm, displayall=displayall, limits=limits,
                           prerun=prerun, profiler=profiler)
     return t, fm
 
 def stub_timer(fm, func, setup=None, title=None):
     if setup is not None:
         setup()
     func()
 
 @contextlib.contextmanager
 def timeone():
     r = []
     ostart = os.times()
     cstart = util.timer()
     yield r
     cstop = util.timer()
     ostop = os.times()
     a, b = ostart, ostop
     r.append((cstop - cstart, b[0] - a[0], b[1]-a[1]))
 
 
 # list of stop condition (elapsed time, minimal run count)
 DEFAULTLIMITS = (
     (3.0, 100),
     (10.0, 3),
 )
 
 def _timer(fm, func, setup=None, title=None, displayall=False,
            limits=DEFAULTLIMITS, prerun=0, profiler=None):
     gc.collect()
     results = []
     begin = util.timer()
     count = 0
     if profiler is None:
         profiler = NOOPCTX
     for i in range(prerun):
         if setup is not None:
             setup()
         func()
     keepgoing = True
     while keepgoing:
         if setup is not None:
             setup()
         with profiler:
             with timeone() as item:
                 r = func()
         profiler = NOOPCTX
         count += 1
         results.append(item[0])
         cstop = util.timer()
         # Look for a stop condition.
         elapsed = cstop - begin
         for t, mincount in limits:
             if elapsed >= t and count >= mincount:
                 keepgoing = False
                 break
 
     formatone(fm, results, title=title, result=r,
               displayall=displayall)
 
 def formatone(fm, timings, title=None, result=None, displayall=False):
 
     count = len(timings)
 
     fm.startitem()
 
     if title:
         fm.write(b'title', b'! %s\n', title)
     if result:
         fm.write(b'result', b'! result: %s\n', result)
     def display(role, entry):
         prefix = b''
         if role != b'best':
             prefix = b'%s.' % role
         fm.plain(b'!')
         fm.write(prefix + b'wall', b' wall %f', entry[0])
         fm.write(prefix + b'comb', b' comb %f', entry[1] + entry[2])
         fm.write(prefix + b'user', b' user %f', entry[1])
         fm.write(prefix + b'sys',  b' sys %f', entry[2])
         fm.write(prefix + b'count',  b' (%s of %%d)' % role, count)
         fm.plain(b'\n')
     timings.sort()
     min_val = timings[0]
     display(b'best', min_val)
     if displayall:
         max_val = timings[-1]
         display(b'max', max_val)
         avg = tuple([sum(x) / count for x in zip(*timings)])
         display(b'avg', avg)
         median = timings[len(timings) // 2]
         display(b'median', median)
 
 # utilities for historical portability
 
 def getint(ui, section, name, default):
     # for "historical portability":
     # ui.configint has been available since 1.9 (or fa2b596db182)
     v = ui.config(section, name, None)
     if v is None:
         return default
     try:
         return int(v)
     except ValueError:
         raise error.ConfigError((b"%s.%s is not an integer ('%s')")
                                 % (section, name, v))
 
 def safeattrsetter(obj, name, ignoremissing=False):
     """Ensure that 'obj' has 'name' attribute before subsequent setattr
 
     This function is aborted, if 'obj' doesn't have 'name' attribute
     at runtime. This avoids overlooking removal of an attribute, which
     breaks assumption of performance measurement, in the future.
 
     This function returns the object to (1) assign a new value, and
     (2) restore an original value to the attribute.
 
     If 'ignoremissing' is true, missing 'name' attribute doesn't cause
     abortion, and this function returns None. This is useful to
     examine an attribute, which isn't ensured in all Mercurial
     versions.
     """
     if not util.safehasattr(obj, name):
         if ignoremissing:
             return None
         raise error.Abort((b"missing attribute %s of %s might break assumption"
                            b" of performance measurement") % (name, obj))
 
     origvalue = getattr(obj, _sysstr(name))
     class attrutil(object):
         def set(self, newvalue):
             setattr(obj, _sysstr(name), newvalue)
         def restore(self):
             setattr(obj, _sysstr(name), origvalue)
 
     return attrutil()
 
 # utilities to examine each internal API changes
 
 def getbranchmapsubsettable():
     # for "historical portability":
     # subsettable is defined in:
     # - branchmap since 2.9 (or 175c6fd8cacc)
     # - repoview since 2.5 (or 59a9f18d4587)
     # - repoviewutil since 5.0
     for mod in (branchmap, repoview, repoviewutil):
         subsettable = getattr(mod, 'subsettable', None)
         if subsettable:
             return subsettable
 
     # bisecting in bcee63733aad::59a9f18d4587 can reach here (both
     # branchmap and repoview modules exist, but subsettable attribute
     # doesn't)
     raise error.Abort((b"perfbranchmap not available with this Mercurial"),
                       hint=b"use 2.5 or later")
 
 def getsvfs(repo):
     """Return appropriate object to access files under .hg/store
     """
     # for "historical portability":
     # repo.svfs has been available since 2.3 (or 7034365089bf)
     svfs = getattr(repo, 'svfs', None)
     if svfs:
         return svfs
     else:
         return getattr(repo, 'sopener')
 
 def getvfs(repo):
     """Return appropriate object to access files under .hg
     """
     # for "historical portability":
     # repo.vfs has been available since 2.3 (or 7034365089bf)
     vfs = getattr(repo, 'vfs', None)
     if vfs:
         return vfs
     else:
         return getattr(repo, 'opener')
 
 def repocleartagscachefunc(repo):
     """Return the function to clear tags cache according to repo internal API
     """
     if util.safehasattr(repo, b'_tagscache'): # since 2.0 (or 9dca7653b525)
         # in this case, setattr(repo, '_tagscache', None) or so isn't
         # correct way to clear tags cache, because existing code paths
         # expect _tagscache to be a structured object.
         def clearcache():
             # _tagscache has been filteredpropertycache since 2.5 (or
             # 98c867ac1330), and delattr() can't work in such case
             if b'_tagscache' in vars(repo):
                 del repo.__dict__[b'_tagscache']
         return clearcache
 
     repotags = safeattrsetter(repo, b'_tags', ignoremissing=True)
     if repotags: # since 1.4 (or 5614a628d173)
         return lambda : repotags.set(None)
 
     repotagscache = safeattrsetter(repo, b'tagscache', ignoremissing=True)
     if repotagscache: # since 0.6 (or d7df759d0e97)
         return lambda : repotagscache.set(None)
 
     # Mercurial earlier than 0.6 (or d7df759d0e97) logically reaches
     # this point, but it isn't so problematic, because:
     # - repo.tags of such Mercurial isn't "callable", and repo.tags()
     #   in perftags() causes failure soon
     # - perf.py itself has been available since 1.1 (or eb240755386d)
     raise error.Abort((b"tags API of this hg command is unknown"))
 
 # utilities to clear cache
 
 def clearfilecache(obj, attrname):
     unfiltered = getattr(obj, 'unfiltered', None)
     if unfiltered is not None:
         obj = obj.unfiltered()
     if attrname in vars(obj):
         delattr(obj, attrname)
     obj._filecache.pop(attrname, None)
 
 def clearchangelog(repo):
     if repo is not repo.unfiltered():
         object.__setattr__(repo, r'_clcachekey', None)
         object.__setattr__(repo, r'_clcache', None)
     clearfilecache(repo.unfiltered(), 'changelog')
 
 # perf commands
 
 @command(b'perfwalk', formatteropts)
 def perfwalk(ui, repo, *pats, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     m = scmutil.match(repo[None], pats, {})
     timer(lambda: len(list(repo.dirstate.walk(m, subrepos=[], unknown=True,
                                               ignored=False))))
     fm.end()
 
 @command(b'perfannotate', formatteropts)
 def perfannotate(ui, repo, f, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     fc = repo[b'.'][f]
     timer(lambda: len(fc.annotate(True)))
     fm.end()
 
 @command(b'perfstatus',
          [(b'u', b'unknown', False,
            b'ask status to look for unknown files')] + formatteropts)
 def perfstatus(ui, repo, **opts):
     opts = _byteskwargs(opts)
     #m = match.always(repo.root, repo.getcwd())
     #timer(lambda: sum(map(len, repo.dirstate.status(m, [], False, False,
     #                                                False))))
     timer, fm = gettimer(ui, opts)
     timer(lambda: sum(map(len, repo.status(unknown=opts[b'unknown']))))
     fm.end()
 
 @command(b'perfaddremove', formatteropts)
 def perfaddremove(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     try:
         oldquiet = repo.ui.quiet
         repo.ui.quiet = True
         matcher = scmutil.match(repo[None])
         opts[b'dry_run'] = True
         if b'uipathfn' in getargspec(scmutil.addremove).args:
             uipathfn = scmutil.getuipathfn(repo)
             timer(lambda: scmutil.addremove(repo, matcher, b"", uipathfn, opts))
         else:
             timer(lambda: scmutil.addremove(repo, matcher, b"", opts))
     finally:
         repo.ui.quiet = oldquiet
         fm.end()
 
 def clearcaches(cl):
     # behave somewhat consistently across internal API changes
     if util.safehasattr(cl, b'clearcaches'):
         cl.clearcaches()
     elif util.safehasattr(cl, b'_nodecache'):
         from mercurial.node import nullid, nullrev
         cl._nodecache = {nullid: nullrev}
         cl._nodepos = None
 
 @command(b'perfheads', formatteropts)
 def perfheads(ui, repo, **opts):
     """benchmark the computation of a changelog heads"""
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     cl = repo.changelog
     def s():
         clearcaches(cl)
     def d():
         len(cl.headrevs())
     timer(d, setup=s)
     fm.end()
 
 @command(b'perftags', formatteropts+
         [
             (b'', b'clear-revlogs', False, b'refresh changelog and manifest'),
         ])
 def perftags(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     repocleartagscache = repocleartagscachefunc(repo)
     clearrevlogs = opts[b'clear_revlogs']
     def s():
         if clearrevlogs:
             clearchangelog(repo)
             clearfilecache(repo.unfiltered(), 'manifest')
         repocleartagscache()
     def t():
         return len(repo.tags())
     timer(t, setup=s)
     fm.end()
 
 @command(b'perfancestors', formatteropts)
 def perfancestors(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     heads = repo.changelog.headrevs()
     def d():
         for a in repo.changelog.ancestors(heads):
             pass
     timer(d)
     fm.end()
 
 @command(b'perfancestorset', formatteropts)
 def perfancestorset(ui, repo, revset, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     revs = repo.revs(revset)
     heads = repo.changelog.headrevs()
     def d():
         s = repo.changelog.ancestors(heads)
         for rev in revs:
             rev in s
     timer(d)
     fm.end()
 
 @command(b'perfdiscovery', formatteropts, b'PATH')
 def perfdiscovery(ui, repo, path, **opts):
     """benchmark discovery between local repo and the peer at given path
     """
     repos = [repo, None]
     timer, fm = gettimer(ui, opts)
     path = ui.expandpath(path)
 
     def s():
         repos[1] = hg.peer(ui, opts, path)
     def d():
         setdiscovery.findcommonheads(ui, *repos)
     timer(d, setup=s)
     fm.end()
 
 @command(b'perfbookmarks', formatteropts +
         [
             (b'', b'clear-revlogs', False, b'refresh changelog and manifest'),
         ])
 def perfbookmarks(ui, repo, **opts):
     """benchmark parsing bookmarks from disk to memory"""
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
 
     clearrevlogs = opts[b'clear_revlogs']
     def s():
         if clearrevlogs:
             clearchangelog(repo)
         clearfilecache(repo, b'_bookmarks')
     def d():
         repo._bookmarks
     timer(d, setup=s)
     fm.end()
 
 @command(b'perfbundleread', formatteropts, b'BUNDLE')
 def perfbundleread(ui, repo, bundlepath, **opts):
     """Benchmark reading of bundle files.
 
     This command is meant to isolate the I/O part of bundle reading as
     much as possible.
     """
     from mercurial import (
         bundle2,
         exchange,
         streamclone,
     )
 
     opts = _byteskwargs(opts)
 
     def makebench(fn):
         def run():
             with open(bundlepath, b'rb') as fh:
                 bundle = exchange.readbundle(ui, fh, bundlepath)
                 fn(bundle)
 
         return run
 
     def makereadnbytes(size):
         def run():
             with open(bundlepath, b'rb') as fh:
                 bundle = exchange.readbundle(ui, fh, bundlepath)
                 while bundle.read(size):
                     pass
 
         return run
 
     def makestdioread(size):
         def run():
             with open(bundlepath, b'rb') as fh:
                 while fh.read(size):
                     pass
 
         return run
 
     # bundle1
 
     def deltaiter(bundle):
         for delta in bundle.deltaiter():
             pass
 
     def iterchunks(bundle):
         for chunk in bundle.getchunks():
             pass
 
     # bundle2
 
     def forwardchunks(bundle):
         for chunk in bundle._forwardchunks():
             pass
 
     def iterparts(bundle):
         for part in bundle.iterparts():
             pass
 
     def iterpartsseekable(bundle):
         for part in bundle.iterparts(seekable=True):
             pass
 
     def seek(bundle):
         for part in bundle.iterparts(seekable=True):
             part.seek(0, os.SEEK_END)
 
     def makepartreadnbytes(size):
         def run():
             with open(bundlepath, b'rb') as fh:
                 bundle = exchange.readbundle(ui, fh, bundlepath)
                 for part in bundle.iterparts():
                     while part.read(size):
                         pass
 
         return run
 
     benches = [
         (makestdioread(8192), b'read(8k)'),
         (makestdioread(16384), b'read(16k)'),
         (makestdioread(32768), b'read(32k)'),
         (makestdioread(131072), b'read(128k)'),
     ]
 
     with open(bundlepath, b'rb') as fh:
         bundle = exchange.readbundle(ui, fh, bundlepath)
 
         if isinstance(bundle, changegroup.cg1unpacker):
             benches.extend([
                 (makebench(deltaiter), b'cg1 deltaiter()'),
                 (makebench(iterchunks), b'cg1 getchunks()'),
                 (makereadnbytes(8192), b'cg1 read(8k)'),
                 (makereadnbytes(16384), b'cg1 read(16k)'),
                 (makereadnbytes(32768), b'cg1 read(32k)'),
                 (makereadnbytes(131072), b'cg1 read(128k)'),
             ])
         elif isinstance(bundle, bundle2.unbundle20):
             benches.extend([
                 (makebench(forwardchunks), b'bundle2 forwardchunks()'),
                 (makebench(iterparts), b'bundle2 iterparts()'),
                 (makebench(iterpartsseekable), b'bundle2 iterparts() seekable'),
                 (makebench(seek), b'bundle2 part seek()'),
                 (makepartreadnbytes(8192), b'bundle2 part read(8k)'),
                 (makepartreadnbytes(16384), b'bundle2 part read(16k)'),
                 (makepartreadnbytes(32768), b'bundle2 part read(32k)'),
                 (makepartreadnbytes(131072), b'bundle2 part read(128k)'),
             ])
         elif isinstance(bundle, streamclone.streamcloneapplier):
             raise error.Abort(b'stream clone bundles not supported')
         else:
             raise error.Abort(b'unhandled bundle type: %s' % type(bundle))
 
     for fn, title in benches:
         timer, fm = gettimer(ui, opts)
         timer(fn, title=title)
         fm.end()
 
 @command(b'perfchangegroupchangelog', formatteropts +
          [(b'', b'cgversion', b'02', b'changegroup version'),
           (b'r', b'rev', b'', b'revisions to add to changegroup')])
 def perfchangegroupchangelog(ui, repo, cgversion=b'02', rev=None, **opts):
     """Benchmark producing a changelog group for a changegroup.
 
     This measures the time spent processing the changelog during a
     bundle operation. This occurs during `hg bundle` and on a server
     processing a `getbundle` wire protocol request (handles clones
     and pull requests).
 
     By default, all revisions are added to the changegroup.
     """
     opts = _byteskwargs(opts)
     cl = repo.changelog
     nodes = [cl.lookup(r) for r in repo.revs(rev or b'all()')]
     bundler = changegroup.getbundler(cgversion, repo)
 
     def d():
         state, chunks = bundler._generatechangelog(cl, nodes)
         for chunk in chunks:
             pass
 
     timer, fm = gettimer(ui, opts)
 
     # Terminal printing can interfere with timing. So disable it.
     with ui.configoverride({(b'progress', b'disable'): True}):
         timer(d)
 
     fm.end()
 
 @command(b'perfdirs', formatteropts)
 def perfdirs(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     dirstate = repo.dirstate
     b'a' in dirstate
     def d():
         dirstate.hasdir(b'a')
         del dirstate._map._dirs
     timer(d)
     fm.end()
 
 @command(b'perfdirstate', formatteropts)
 def perfdirstate(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     b"a" in repo.dirstate
     def d():
         repo.dirstate.invalidate()
         b"a" in repo.dirstate
     timer(d)
     fm.end()
 
 @command(b'perfdirstatedirs', formatteropts)
 def perfdirstatedirs(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     b"a" in repo.dirstate
     def d():
         repo.dirstate.hasdir(b"a")
         del repo.dirstate._map._dirs
     timer(d)
     fm.end()
 
 @command(b'perfdirstatefoldmap', formatteropts)
 def perfdirstatefoldmap(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     dirstate = repo.dirstate
     b'a' in dirstate
     def d():
         dirstate._map.filefoldmap.get(b'a')
         del dirstate._map.filefoldmap
     timer(d)
     fm.end()
 
 @command(b'perfdirfoldmap', formatteropts)
 def perfdirfoldmap(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     dirstate = repo.dirstate
     b'a' in dirstate
     def d():
         dirstate._map.dirfoldmap.get(b'a')
         del dirstate._map.dirfoldmap
         del dirstate._map._dirs
     timer(d)
     fm.end()
 
 @command(b'perfdirstatewrite', formatteropts)
 def perfdirstatewrite(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     ds = repo.dirstate
     b"a" in ds
     def d():
         ds._dirty = True
         ds.write(repo.currenttransaction())
     timer(d)
     fm.end()
 
 def _getmergerevs(repo, opts):
     """parse command argument to return rev involved in merge
 
     input: options dictionnary with `rev`, `from` and `bse`
     output: (localctx, otherctx, basectx)
     """
     if opts[b'from']:
         fromrev = scmutil.revsingle(repo, opts[b'from'])
         wctx = repo[fromrev]
     else:
         wctx = repo[None]
         # we don't want working dir files to be stat'd in the benchmark, so
         # prime that cache
         wctx.dirty()
     rctx = scmutil.revsingle(repo, opts[b'rev'], opts[b'rev'])
     if opts[b'base']:
         fromrev = scmutil.revsingle(repo, opts[b'base'])
         ancestor = repo[fromrev]
     else:
         ancestor = wctx.ancestor(rctx)
     return (wctx, rctx, ancestor)
 
 @command(b'perfmergecalculate',
          [
              (b'r', b'rev', b'.', b'rev to merge against'),
              (b'', b'from', b'', b'rev to merge from'),
              (b'', b'base', b'', b'the revision to use as base'),
          ] + formatteropts)
 def perfmergecalculate(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
 
     wctx, rctx, ancestor = _getmergerevs(repo, opts)
     def d():
         # acceptremote is True because we don't want prompts in the middle of
         # our benchmark
         merge.calculateupdates(repo, wctx, rctx, [ancestor], False, False,
                                acceptremote=True, followcopies=True)
     timer(d)
     fm.end()
 
 @command(b'perfmergecopies',
          [
              (b'r', b'rev', b'.', b'rev to merge against'),
              (b'', b'from', b'', b'rev to merge from'),
              (b'', b'base', b'', b'the revision to use as base'),
          ] + formatteropts)
 def perfmergecopies(ui, repo, **opts):
     """measure runtime of `copies.mergecopies`"""
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     wctx, rctx, ancestor = _getmergerevs(repo, opts)
     def d():
         # acceptremote is True because we don't want prompts in the middle of
         # our benchmark
         copies.mergecopies(repo, wctx, rctx, ancestor)
     timer(d)
     fm.end()
 
 @command(b'perfpathcopies', [], b"REV REV")
 def perfpathcopies(ui, repo, rev1, rev2, **opts):
     """benchmark the copy tracing logic"""
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     ctx1 = scmutil.revsingle(repo, rev1, rev1)
     ctx2 = scmutil.revsingle(repo, rev2, rev2)
     def d():
         copies.pathcopies(ctx1, ctx2)
     timer(d)
     fm.end()
 
 @command(b'perfphases',
          [(b'', b'full', False, b'include file reading time too'),
           ], b"")
 def perfphases(ui, repo, **opts):
     """benchmark phasesets computation"""
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     _phases = repo._phasecache
     full = opts.get(b'full')
     def d():
         phases = _phases
         if full:
             clearfilecache(repo, b'_phasecache')
             phases = repo._phasecache
         phases.invalidate()
         phases.loadphaserevs(repo)
     timer(d)
     fm.end()
 
 @command(b'perfphasesremote',
          [], b"[DEST]")
 def perfphasesremote(ui, repo, dest=None, **opts):
     """benchmark time needed to analyse phases of the remote server"""
     from mercurial.node import (
         bin,
     )
     from mercurial import (
         exchange,
         hg,
         phases,
     )
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
 
     path = ui.paths.getpath(dest, default=(b'default-push', b'default'))
     if not path:
         raise error.Abort((b'default repository not configured!'),
                           hint=(b"see 'hg help config.paths'"))
     dest = path.pushloc or path.loc
     ui.status((b'analysing phase of %s\n') % util.hidepassword(dest))
     other = hg.peer(repo, opts, dest)
 
     # easier to perform discovery through the operation
     op = exchange.pushoperation(repo, other)
     exchange._pushdiscoverychangeset(op)
 
     remotesubset = op.fallbackheads
 
     with other.commandexecutor() as e:
         remotephases = e.callcommand(b'listkeys',
                        {b'namespace': b'phases'}).result()
     del other
     publishing = remotephases.get(b'publishing', False)
     if publishing:
         ui.status((b'publishing: yes\n'))
     else:
         ui.status((b'publishing: no\n'))
 
     nodemap = repo.changelog.nodemap
     nonpublishroots = 0
     for nhex, phase in remotephases.iteritems():
         if nhex == b'publishing': # ignore data related to publish option
             continue
         node = bin(nhex)
         if node in nodemap and int(phase):
             nonpublishroots += 1
     ui.status((b'number of roots: %d\n') % len(remotephases))
     ui.status((b'number of known non public roots: %d\n') % nonpublishroots)
     def d():
         phases.remotephasessummary(repo,
                                    remotesubset,
                                    remotephases)
     timer(d)
     fm.end()
 
 @command(b'perfmanifest',[
             (b'm', b'manifest-rev', False, b'Look up a manifest node revision'),
             (b'', b'clear-disk', False, b'clear on-disk caches too'),
          ] + formatteropts, b'REV|NODE')
 def perfmanifest(ui, repo, rev, manifest_rev=False, clear_disk=False, **opts):
     """benchmark the time to read a manifest from disk and return a usable
     dict-like object
 
     Manifest caches are cleared before retrieval."""
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     if not manifest_rev:
         ctx = scmutil.revsingle(repo, rev, rev)
         t = ctx.manifestnode()
     else:
         from mercurial.node import bin
 
         if len(rev) == 40:
             t = bin(rev)
         else:
             try:
                 rev = int(rev)
 
                 if util.safehasattr(repo.manifestlog, b'getstorage'):
                     t = repo.manifestlog.getstorage(b'').node(rev)
                 else:
                     t = repo.manifestlog._revlog.lookup(rev)
             except ValueError:
                 raise error.Abort(b'manifest revision must be integer or full '
                                   b'node')
     def d():
         repo.manifestlog.clearcaches(clear_persisted_data=clear_disk)
         repo.manifestlog[t].read()
     timer(d)
     fm.end()
 
 @command(b'perfchangeset', formatteropts)
 def perfchangeset(ui, repo, rev, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     n = scmutil.revsingle(repo, rev).node()
     def d():
         repo.changelog.read(n)
         #repo.changelog._cache = None
     timer(d)
     fm.end()
 
 @command(b'perfignore', formatteropts)
 def perfignore(ui, repo, **opts):
     """benchmark operation related to computing ignore"""
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     dirstate = repo.dirstate
 
     def setupone():
         dirstate.invalidate()
         clearfilecache(dirstate, b'_ignore')
 
     def runone():
         dirstate._ignore
 
     timer(runone, setup=setupone, title=b"load")
     fm.end()
 
 @command(b'perfindex', [
             (b'', b'rev', [], b'revision to be looked up (default tip)'),
             (b'', b'no-lookup', None, b'do not revision lookup post creation'),
          ] + formatteropts)
 def perfindex(ui, repo, **opts):
     """benchmark index creation time followed by a lookup
 
     The default is to look `tip` up. Depending on the index implementation,
     the revision looked up can matters. For example, an implementation
     scanning the index will have a faster lookup time for `--rev tip` than for
     `--rev 0`. The number of looked up revisions and their order can also
     matters.
 
     Example of useful set to test:
     * tip
     * 0
     * -10:
     * :10
     * -10: + :10
     * :10: + -10:
     * -10000:
     * -10000: + 0
 
     It is not currently possible to check for lookup of a missing node. For
     deeper lookup benchmarking, checkout the `perfnodemap` command."""
     import mercurial.revlog
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     mercurial.revlog._prereadsize = 2**24 # disable lazy parser in old hg
     if opts[b'no_lookup']:
         if opts['rev']:
             raise error.Abort('--no-lookup and --rev are mutually exclusive')
         nodes = []
     elif not opts[b'rev']:
         nodes = [repo[b"tip"].node()]
     else:
         revs = scmutil.revrange(repo, opts[b'rev'])
         cl = repo.changelog
         nodes = [cl.node(r) for r in revs]
 
     unfi = repo.unfiltered()
     # find the filecache func directly
     # This avoid polluting the benchmark with the filecache logic
     makecl = unfi.__class__.changelog.func
     def setup():
         # probably not necessary, but for good measure
         clearchangelog(unfi)
     def d():
         cl = makecl(unfi)
         for n in nodes:
             cl.rev(n)
     timer(d, setup=setup)
     fm.end()
 
 @command(b'perfnodemap', [
           (b'', b'rev', [], b'revision to be looked up (default tip)'),
           (b'', b'clear-caches', True, b'clear revlog cache between calls'),
     ] + formatteropts)
 def perfnodemap(ui, repo, **opts):
     """benchmark the time necessary to look up revision from a cold nodemap
 
     Depending on the implementation, the amount and order of revision we look
     up can varies. Example of useful set to test:
     * tip
     * 0
     * -10:
     * :10
     * -10: + :10
     * :10: + -10:
     * -10000:
     * -10000: + 0
 
     The command currently focus on valid binary lookup. Benchmarking for
     hexlookup, prefix lookup and missing lookup would also be valuable.
     """
     import mercurial.revlog
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     mercurial.revlog._prereadsize = 2**24 # disable lazy parser in old hg
 
     unfi = repo.unfiltered()
     clearcaches = opts['clear_caches']
     # find the filecache func directly
     # This avoid polluting the benchmark with the filecache logic
     makecl = unfi.__class__.changelog.func
     if not opts[b'rev']:
         raise error.Abort('use --rev to specify revisions to look up')
     revs = scmutil.revrange(repo, opts[b'rev'])
     cl = repo.changelog
     nodes = [cl.node(r) for r in revs]
 
     # use a list to pass reference to a nodemap from one closure to the next
     nodeget = [None]
     def setnodeget():
         # probably not necessary, but for good measure
         clearchangelog(unfi)
         nodeget[0] = makecl(unfi).nodemap.get
 
     def d():
         get = nodeget[0]
         for n in nodes:
             get(n)
 
     setup = None
     if clearcaches:
         def setup():
             setnodeget()
     else:
         setnodeget()
         d() # prewarm the data structure
     timer(d, setup=setup)
     fm.end()
 
 @command(b'perfstartup', formatteropts)
 def perfstartup(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     def d():
         if os.name != r'nt':
             os.system(b"HGRCPATH= %s version -q > /dev/null" %
                       fsencode(sys.argv[0]))
         else:
             os.environ[r'HGRCPATH'] = r' '
             os.system(r"%s version -q > NUL" % sys.argv[0])
     timer(d)
     fm.end()
 
 @command(b'perfparents', formatteropts)
 def perfparents(ui, repo, **opts):
     """benchmark the time necessary to fetch one changeset's parents.
 
     The fetch is done using the `node identifier`, traversing all object layers
     from the repository object. The first N revisions will be used for this
     benchmark. N is controlled by the ``perf.parentscount`` config option
     (default: 1000).
     """
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     # control the number of commits perfparents iterates over
     # experimental config: perf.parentscount
     count = getint(ui, b"perf", b"parentscount", 1000)
     if len(repo.changelog) < count:
         raise error.Abort(b"repo needs %d commits for this test" % count)
     repo = repo.unfiltered()
     nl = [repo.changelog.node(i) for i in _xrange(count)]
     def d():
         for n in nl:
             repo.changelog.parents(n)
     timer(d)
     fm.end()
 
 @command(b'perfctxfiles', formatteropts)
 def perfctxfiles(ui, repo, x, **opts):
     opts = _byteskwargs(opts)
     x = int(x)
     timer, fm = gettimer(ui, opts)
     def d():
         len(repo[x].files())
     timer(d)
     fm.end()
 
 @command(b'perfrawfiles', formatteropts)
 def perfrawfiles(ui, repo, x, **opts):
     opts = _byteskwargs(opts)
     x = int(x)
     timer, fm = gettimer(ui, opts)
     cl = repo.changelog
     def d():
         len(cl.read(x)[3])
     timer(d)
     fm.end()
 
 @command(b'perflookup', formatteropts)
 def perflookup(ui, repo, rev, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     timer(lambda: len(repo.lookup(rev)))
     fm.end()
 
 @command(b'perflinelogedits',
          [(b'n', b'edits', 10000, b'number of edits'),
           (b'', b'max-hunk-lines', 10, b'max lines in a hunk'),
           ], norepo=True)
 def perflinelogedits(ui, **opts):
     from mercurial import linelog
 
     opts = _byteskwargs(opts)
 
     edits = opts[b'edits']
     maxhunklines = opts[b'max_hunk_lines']
 
     maxb1 = 100000
     random.seed(0)
     randint = random.randint
     currentlines = 0
     arglist = []
     for rev in _xrange(edits):
         a1 = randint(0, currentlines)
         a2 = randint(a1, min(currentlines, a1 + maxhunklines))
         b1 = randint(0, maxb1)
         b2 = randint(b1, b1 + maxhunklines)
         currentlines += (b2 - b1) - (a2 - a1)
         arglist.append((rev, a1, a2, b1, b2))
 
     def d():
         ll = linelog.linelog()
         for args in arglist:
             ll.replacelines(*args)
 
     timer, fm = gettimer(ui, opts)
     timer(d)
     fm.end()
 
 @command(b'perfrevrange', formatteropts)
 def perfrevrange(ui, repo, *specs, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     revrange = scmutil.revrange
     timer(lambda: len(revrange(repo, specs)))
     fm.end()
 
 @command(b'perfnodelookup', formatteropts)
 def perfnodelookup(ui, repo, rev, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     import mercurial.revlog
     mercurial.revlog._prereadsize = 2**24 # disable lazy parser in old hg
     n = scmutil.revsingle(repo, rev).node()
     cl = mercurial.revlog.revlog(getsvfs(repo), b"00changelog.i")
     def d():
         cl.rev(n)
         clearcaches(cl)
     timer(d)
     fm.end()
 
 @command(b'perflog',
          [(b'', b'rename', False, b'ask log to follow renames')
          ] + formatteropts)
 def perflog(ui, repo, rev=None, **opts):
     opts = _byteskwargs(opts)
     if rev is None:
         rev=[]
     timer, fm = gettimer(ui, opts)
     ui.pushbuffer()
     timer(lambda: commands.log(ui, repo, rev=rev, date=b'', user=b'',
                                copies=opts.get(b'rename')))
     ui.popbuffer()
     fm.end()
 
 @command(b'perfmoonwalk', formatteropts)
 def perfmoonwalk(ui, repo, **opts):
     """benchmark walking the changelog backwards
 
     This also loads the changelog data for each revision in the changelog.
     """
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     def moonwalk():
         for i in repo.changelog.revs(start=(len(repo) - 1), stop=-1):
             ctx = repo[i]
             ctx.branch() # read changelog data (in addition to the index)
     timer(moonwalk)
     fm.end()
 
 @command(b'perftemplating',
          [(b'r', b'rev', [], b'revisions to run the template on'),
           ] + formatteropts)
 def perftemplating(ui, repo, testedtemplate=None, **opts):
     """test the rendering time of a given template"""
     if makelogtemplater is None:
         raise error.Abort((b"perftemplating not available with this Mercurial"),
                           hint=b"use 4.3 or later")
 
     opts = _byteskwargs(opts)
 
     nullui = ui.copy()
     nullui.fout = open(os.devnull, r'wb')
     nullui.disablepager()
     revs = opts.get(b'rev')
     if not revs:
         revs = [b'all()']
     revs = list(scmutil.revrange(repo, revs))
 
     defaulttemplate = (b'{date|shortdate} [{rev}:{node|short}]'
                        b' {author|person}: {desc|firstline}\n')
     if testedtemplate is None:
         testedtemplate = defaulttemplate
     displayer = makelogtemplater(nullui, repo, testedtemplate)
     def format():
         for r in revs:
             ctx = repo[r]
             displayer.show(ctx)
             displayer.flush(ctx)
 
     timer, fm = gettimer(ui, opts)
     timer(format)
     fm.end()
 
 @command(b'perfhelper-mergecopies', formatteropts +
          [
           (b'r', b'revs', [], b'restrict search to these revisions'),
           (b'', b'timing', False, b'provides extra data (costly)'),
          ])
 def perfhelpermergecopies(ui, repo, revs=[], **opts):
     """find statistics about potential parameters for `perfmergecopies`
 
     This command find (base, p1, p2) triplet relevant for copytracing
     benchmarking in the context of a merge.  It reports values for some of the
     parameters that impact merge copy tracing time during merge.
 
     If `--timing` is set, rename detection is run and the associated timing
     will be reported. The extra details come at the cost of slower command
     execution.
 
     Since rename detection is only run once, other factors might easily
     affect the precision of the timing. However it should give a good
     approximation of which revision triplets are very costly.
     """
     opts = _byteskwargs(opts)
     fm = ui.formatter(b'perf', opts)
     dotiming = opts[b'timing']
 
     output_template = [
         ("base", "%(base)12s"),
         ("p1", "%(p1.node)12s"),
         ("p2", "%(p2.node)12s"),
         ("p1.nb-revs", "%(p1.nbrevs)12d"),
         ("p1.nb-files", "%(p1.nbmissingfiles)12d"),
         ("p1.renames", "%(p1.renamedfiles)12d"),
         ("p1.time", "%(p1.time)12.3f"),
         ("p2.nb-revs", "%(p2.nbrevs)12d"),
         ("p2.nb-files", "%(p2.nbmissingfiles)12d"),
         ("p2.renames", "%(p2.renamedfiles)12d"),
         ("p2.time", "%(p2.time)12.3f"),
         ("renames", "%(nbrenamedfiles)12d"),
         ("total.time", "%(time)12.3f"),
         ]
     if not dotiming:
         output_template = [i for i in output_template
                            if not ('time' in i[0] or 'renames' in i[0])]
     header_names = [h for (h, v) in output_template]
     output = ' '.join([v for (h, v) in output_template]) + '\n'
     header = ' '.join(['%12s'] * len(header_names)) + '\n'
     fm.plain(header % tuple(header_names))
 
     if not revs:
         revs = ['all()']
     revs = scmutil.revrange(repo, revs)
 
     roi = repo.revs('merge() and %ld', revs)
     for r in roi:
         ctx = repo[r]
         p1 = ctx.p1()
         p2 = ctx.p2()
         bases = repo.changelog._commonancestorsheads(p1.rev(), p2.rev())
         for b in bases:
             b = repo[b]
             p1missing = copies._computeforwardmissing(b, p1)
             p2missing = copies._computeforwardmissing(b, p2)
             data = {
                 b'base': b.hex(),
                 b'p1.node': p1.hex(),
                 b'p1.nbrevs': len(repo.revs('%d::%d', b.rev(), p1.rev())),
                 b'p1.nbmissingfiles': len(p1missing),
                 b'p2.node': p2.hex(),
                 b'p2.nbrevs': len(repo.revs('%d::%d', b.rev(), p2.rev())),
                 b'p2.nbmissingfiles': len(p2missing),
             }
             if dotiming:
                 begin = util.timer()
                 mergedata = copies.mergecopies(repo, p1, p2, b)
                 end = util.timer()
                 # not very stable timing since we did only one run
                 data['time'] = end - begin
                 # mergedata contains five dicts: "copy", "movewithdir",
                 # "diverge", "renamedelete" and "dirmove".
                 # The first 4 are about renamed file so lets count that.
                 renames = len(mergedata[0])
                 renames += len(mergedata[1])
                 renames += len(mergedata[2])
                 renames += len(mergedata[3])
                 data['nbrenamedfiles'] = renames
                 begin = util.timer()
                 p1renames = copies.pathcopies(b, p1)
                 end = util.timer()
                 data['p1.time'] = end - begin
                 begin = util.timer()
                 p2renames = copies.pathcopies(b, p2)
                 data['p2.time'] = end - begin
                 end = util.timer()
                 data['p1.renamedfiles'] = len(p1renames)
                 data['p2.renamedfiles'] = len(p2renames)
             fm.startitem()
             fm.data(**data)
             # make node pretty for the human output
             out = data.copy()
             out['base'] = fm.hexfunc(b.node())
             out['p1.node'] = fm.hexfunc(p1.node())
             out['p2.node'] = fm.hexfunc(p2.node())
             fm.plain(output % out)
 
     fm.end()
 
 @command(b'perfhelper-pathcopies', formatteropts +
          [
           (b'r', b'revs', [], b'restrict search to these revisions'),
           (b'', b'timing', False, b'provides extra data (costly)'),
          ])
 def perfhelperpathcopies(ui, repo, revs=[], **opts):
     """find statistic about potential parameters for the `perftracecopies`
 
     This command find source-destination pair relevant for copytracing testing.
     It report value for some of the parameters that impact copy tracing time.
 
     If `--timing` is set, rename detection is run and the associated timing
     will be reported. The extra details comes at the cost of a slower command
     execution.
 
     Since the rename detection is only run once, other factors might easily
     affect the precision of the timing. However it should give a good
     approximation of which revision pairs are very costly.
     """
     opts = _byteskwargs(opts)
     fm = ui.formatter(b'perf', opts)
     dotiming = opts[b'timing']
 
     if dotiming:
         header = '%12s %12s %12s %12s %12s %12s\n'
         output = ("%(source)12s %(destination)12s "
                   "%(nbrevs)12d %(nbmissingfiles)12d "
                   "%(nbrenamedfiles)12d %(time)18.5f\n")
         header_names = ("source", "destination", "nb-revs", "nb-files",
                         "nb-renames", "time")
         fm.plain(header % header_names)
     else:
         header = '%12s %12s %12s %12s\n'
         output = ("%(source)12s %(destination)12s "
                   "%(nbrevs)12d %(nbmissingfiles)12d\n")
         fm.plain(header % ("source", "destination", "nb-revs", "nb-files"))
 
     if not revs:
         revs = ['all()']
     revs = scmutil.revrange(repo, revs)
 
     roi = repo.revs('merge() and %ld', revs)
     for r in roi:
         ctx = repo[r]
         p1 = ctx.p1().rev()
         p2 = ctx.p2().rev()
         bases = repo.changelog._commonancestorsheads(p1, p2)
         for p in (p1, p2):
             for b in bases:
                 base = repo[b]
                 parent = repo[p]
                 missing = copies._computeforwardmissing(base, parent)
                 if not missing:
                     continue
                 data = {
                     b'source': base.hex(),
                     b'destination': parent.hex(),
                     b'nbrevs': len(repo.revs('%d::%d', b, p)),
                     b'nbmissingfiles': len(missing),
                 }
                 if dotiming:
                     begin = util.timer()
                     renames = copies.pathcopies(base, parent)
                     end = util.timer()
                     # not very stable timing since we did only one run
                     data['time'] = end - begin
                     data['nbrenamedfiles'] = len(renames)
                 fm.startitem()
                 fm.data(**data)
                 out = data.copy()
                 out['source'] = fm.hexfunc(base.node())
                 out['destination'] = fm.hexfunc(parent.node())
                 fm.plain(output % out)
 
     fm.end()
 
 @command(b'perfcca', formatteropts)
 def perfcca(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     timer(lambda: scmutil.casecollisionauditor(ui, False, repo.dirstate))
     fm.end()
 
 @command(b'perffncacheload', formatteropts)
 def perffncacheload(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     s = repo.store
     def d():
         s.fncache._load()
     timer(d)
     fm.end()
 
 @command(b'perffncachewrite', formatteropts)
 def perffncachewrite(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     s = repo.store
     lock = repo.lock()
     s.fncache._load()
     tr = repo.transaction(b'perffncachewrite')
     tr.addbackup(b'fncache')
     def d():
         s.fncache._dirty = True
         s.fncache.write(tr)
     timer(d)
     tr.close()
     lock.release()
     fm.end()
 
 @command(b'perffncacheencode', formatteropts)
 def perffncacheencode(ui, repo, **opts):
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     s = repo.store
     s.fncache._load()
     def d():
         for p in s.fncache.entries:
             s.encode(p)
     timer(d)
     fm.end()
 
 def _bdiffworker(q, blocks, xdiff, ready, done):
     while not done.is_set():
         pair = q.get()
         while pair is not None:
             if xdiff:
                 mdiff.bdiff.xdiffblocks(*pair)
             elif blocks:
                 mdiff.bdiff.blocks(*pair)
             else:
                 mdiff.textdiff(*pair)
             q.task_done()
             pair = q.get()
         q.task_done() # for the None one
         with ready:
             ready.wait()
 
 def _manifestrevision(repo, mnode):
     ml = repo.manifestlog
 
     if util.safehasattr(ml, b'getstorage'):
         store = ml.getstorage(b'')
     else:
         store = ml._revlog
 
     return store.revision(mnode)
 
 @command(b'perfbdiff', revlogopts + formatteropts + [
     (b'', b'count', 1, b'number of revisions to test (when using --startrev)'),
     (b'', b'alldata', False, b'test bdiffs for all associated revisions'),
     (b'', b'threads', 0, b'number of thread to use (disable with 0)'),
     (b'', b'blocks', False, b'test computing diffs into blocks'),
     (b'', b'xdiff', False, b'use xdiff algorithm'),
     ],
 
     b'-c|-m|FILE REV')
 def perfbdiff(ui, repo, file_, rev=None, count=None, threads=0, **opts):
     """benchmark a bdiff between revisions
 
     By default, benchmark a bdiff between its delta parent and itself.
 
     With ``--count``, benchmark bdiffs between delta parents and self for N
     revisions starting at the specified revision.
 
     With ``--alldata``, assume the requested revision is a changeset and
     measure bdiffs for all changes related to that changeset (manifest
     and filelogs).
     """
     opts = _byteskwargs(opts)
 
     if opts[b'xdiff'] and not opts[b'blocks']:
         raise error.CommandError(b'perfbdiff', b'--xdiff requires --blocks')
 
     if opts[b'alldata']:
         opts[b'changelog'] = True
 
     if opts.get(b'changelog') or opts.get(b'manifest'):
         file_, rev = None, file_
     elif rev is None:
         raise error.CommandError(b'perfbdiff', b'invalid arguments')
 
     blocks = opts[b'blocks']
     xdiff = opts[b'xdiff']
     textpairs = []
 
     r = cmdutil.openrevlog(repo, b'perfbdiff', file_, opts)
 
     startrev = r.rev(r.lookup(rev))
     for rev in range(startrev, min(startrev + count, len(r) - 1)):
         if opts[b'alldata']:
             # Load revisions associated with changeset.
             ctx = repo[rev]
             mtext = _manifestrevision(repo, ctx.manifestnode())
             for pctx in ctx.parents():
                 pman = _manifestrevision(repo, pctx.manifestnode())
                 textpairs.append((pman, mtext))
 
             # Load filelog revisions by iterating manifest delta.
             man = ctx.manifest()
             pman = ctx.p1().manifest()
             for filename, change in pman.diff(man).items():
                 fctx = repo.file(filename)
                 f1 = fctx.revision(change[0][0] or -1)
                 f2 = fctx.revision(change[1][0] or -1)
                 textpairs.append((f1, f2))
         else:
             dp = r.deltaparent(rev)
             textpairs.append((r.revision(dp), r.revision(rev)))
 
     withthreads = threads > 0
     if not withthreads:
         def d():
             for pair in textpairs:
                 if xdiff:
                     mdiff.bdiff.xdiffblocks(*pair)
                 elif blocks:
                     mdiff.bdiff.blocks(*pair)
                 else:
                     mdiff.textdiff(*pair)
     else:
         q = queue()
         for i in _xrange(threads):
             q.put(None)
         ready = threading.Condition()
         done = threading.Event()
         for i in _xrange(threads):
             threading.Thread(target=_bdiffworker,
                              args=(q, blocks, xdiff, ready, done)).start()
         q.join()
         def d():
             for pair in textpairs:
                 q.put(pair)
             for i in _xrange(threads):
                 q.put(None)
             with ready:
                 ready.notify_all()
             q.join()
     timer, fm = gettimer(ui, opts)
     timer(d)
     fm.end()
 
     if withthreads:
         done.set()
         for i in _xrange(threads):
             q.put(None)
         with ready:
             ready.notify_all()
 
 @command(b'perfunidiff', revlogopts + formatteropts + [
     (b'', b'count', 1, b'number of revisions to test (when using --startrev)'),
     (b'', b'alldata', False, b'test unidiffs for all associated revisions'),
     ], b'-c|-m|FILE REV')
 def perfunidiff(ui, repo, file_, rev=None, count=None, **opts):
     """benchmark a unified diff between revisions
 
     This doesn't include any copy tracing - it's just a unified diff
     of the texts.
 
     By default, benchmark a diff between its delta parent and itself.
 
     With ``--count``, benchmark diffs between delta parents and self for N
     revisions starting at the specified revision.
 
     With ``--alldata``, assume the requested revision is a changeset and
     measure diffs for all changes related to that changeset (manifest
     and filelogs).
     """
     opts = _byteskwargs(opts)
     if opts[b'alldata']:
         opts[b'changelog'] = True
 
     if opts.get(b'changelog') or opts.get(b'manifest'):
         file_, rev = None, file_
     elif rev is None:
         raise error.CommandError(b'perfunidiff', b'invalid arguments')
 
     textpairs = []
 
     r = cmdutil.openrevlog(repo, b'perfunidiff', file_, opts)
 
     startrev = r.rev(r.lookup(rev))
     for rev in range(startrev, min(startrev + count, len(r) - 1)):
         if opts[b'alldata']:
             # Load revisions associated with changeset.
             ctx = repo[rev]
             mtext = _manifestrevision(repo, ctx.manifestnode())
             for pctx in ctx.parents():
                 pman = _manifestrevision(repo, pctx.manifestnode())
                 textpairs.append((pman, mtext))
 
             # Load filelog revisions by iterating manifest delta.
             man = ctx.manifest()
             pman = ctx.p1().manifest()
             for filename, change in pman.diff(man).items():
                 fctx = repo.file(filename)
                 f1 = fctx.revision(change[0][0] or -1)
                 f2 = fctx.revision(change[1][0] or -1)
                 textpairs.append((f1, f2))
         else:
             dp = r.deltaparent(rev)
             textpairs.append((r.revision(dp), r.revision(rev)))
 
     def d():
         for left, right in textpairs:
             # The date strings don't matter, so we pass empty strings.
             headerlines, hunks = mdiff.unidiff(
                 left, b'', right, b'', b'left', b'right', binary=False)
             # consume iterators in roughly the way patch.py does
             b'\n'.join(headerlines)
             b''.join(sum((list(hlines) for hrange, hlines in hunks), []))
     timer, fm = gettimer(ui, opts)
     timer(d)
     fm.end()
 
 @command(b'perfdiffwd', formatteropts)
 def perfdiffwd(ui, repo, **opts):
     """Profile diff of working directory changes"""
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     options = {
         'w': 'ignore_all_space',
         'b': 'ignore_space_change',
         'B': 'ignore_blank_lines',
         }
 
     for diffopt in ('', 'w', 'b', 'B', 'wB'):
         opts = dict((options[c], b'1') for c in diffopt)
         def d():
             ui.pushbuffer()
             commands.diff(ui, repo, **opts)
             ui.popbuffer()
         diffopt = diffopt.encode('ascii')
         title = b'diffopts: %s' % (diffopt and (b'-' + diffopt) or b'none')
         timer(d, title=title)
     fm.end()
 
 @command(b'perfrevlogindex', revlogopts + formatteropts,
          b'-c|-m|FILE')
 def perfrevlogindex(ui, repo, file_=None, **opts):
     """Benchmark operations against a revlog index.
 
     This tests constructing a revlog instance, reading index data,
     parsing index data, and performing various operations related to
     index data.
     """
 
     opts = _byteskwargs(opts)
 
     rl = cmdutil.openrevlog(repo, b'perfrevlogindex', file_, opts)
 
     opener = getattr(rl, 'opener')  # trick linter
     indexfile = rl.indexfile
     data = opener.read(indexfile)
 
     header = struct.unpack(b'>I', data[0:4])[0]
     version = header & 0xFFFF
     if version == 1:
         revlogio = revlog.revlogio()
         inline = header & (1 << 16)
     else:
         raise error.Abort((b'unsupported revlog version: %d') % version)
 
     rllen = len(rl)
 
     node0 = rl.node(0)
     node25 = rl.node(rllen // 4)
     node50 = rl.node(rllen // 2)
     node75 = rl.node(rllen // 4 * 3)
     node100 = rl.node(rllen - 1)
 
     allrevs = range(rllen)
     allrevsrev = list(reversed(allrevs))
     allnodes = [rl.node(rev) for rev in range(rllen)]
     allnodesrev = list(reversed(allnodes))
 
     def constructor():
         revlog.revlog(opener, indexfile)
 
     def read():
         with opener(indexfile) as fh:
             fh.read()
 
     def parseindex():
         revlogio.parseindex(data, inline)
 
     def getentry(revornode):
         index = revlogio.parseindex(data, inline)[0]
         index[revornode]
 
     def getentries(revs, count=1):
         index = revlogio.parseindex(data, inline)[0]
 
         for i in range(count):
             for rev in revs:
                 index[rev]
 
     def resolvenode(node):
         nodemap = revlogio.parseindex(data, inline)[1]
         # This only works for the C code.
         if nodemap is None:
             return
 
         try:
             nodemap[node]
         except error.RevlogError:
             pass
 
     def resolvenodes(nodes, count=1):
         nodemap = revlogio.parseindex(data, inline)[1]
         if nodemap is None:
             return
 
         for i in range(count):
             for node in nodes:
                 try:
                     nodemap[node]
                 except error.RevlogError:
                     pass
 
     benches = [
         (constructor, b'revlog constructor'),
         (read, b'read'),
         (parseindex, b'create index object'),
         (lambda: getentry(0), b'retrieve index entry for rev 0'),
         (lambda: resolvenode(b'a' * 20), b'look up missing node'),
         (lambda: resolvenode(node0), b'look up node at rev 0'),
         (lambda: resolvenode(node25), b'look up node at 1/4 len'),
         (lambda: resolvenode(node50), b'look up node at 1/2 len'),
         (lambda: resolvenode(node75), b'look up node at 3/4 len'),
         (lambda: resolvenode(node100), b'look up node at tip'),
         # 2x variation is to measure caching impact.
         (lambda: resolvenodes(allnodes),
          b'look up all nodes (forward)'),
         (lambda: resolvenodes(allnodes, 2),
          b'look up all nodes 2x (forward)'),
         (lambda: resolvenodes(allnodesrev),
          b'look up all nodes (reverse)'),
         (lambda: resolvenodes(allnodesrev, 2),
          b'look up all nodes 2x (reverse)'),
         (lambda: getentries(allrevs),
          b'retrieve all index entries (forward)'),
         (lambda: getentries(allrevs, 2),
          b'retrieve all index entries 2x (forward)'),
         (lambda: getentries(allrevsrev),
          b'retrieve all index entries (reverse)'),
         (lambda: getentries(allrevsrev, 2),
          b'retrieve all index entries 2x (reverse)'),
     ]
 
     for fn, title in benches:
         timer, fm = gettimer(ui, opts)
         timer(fn, title=title)
         fm.end()
 
 @command(b'perfrevlogrevisions', revlogopts + formatteropts +
          [(b'd', b'dist', 100, b'distance between the revisions'),
           (b's', b'startrev', 0, b'revision to start reading at'),
           (b'', b'reverse', False, b'read in reverse')],
          b'-c|-m|FILE')
 def perfrevlogrevisions(ui, repo, file_=None, startrev=0, reverse=False,
                         **opts):
     """Benchmark reading a series of revisions from a revlog.
 
     By default, we read every ``-d/--dist`` revision from 0 to tip of
     the specified revlog.
 
     The start revision can be defined via ``-s/--startrev``.
     """
     opts = _byteskwargs(opts)
 
     rl = cmdutil.openrevlog(repo, b'perfrevlogrevisions', file_, opts)
     rllen = getlen(ui)(rl)
 
     if startrev < 0:
         startrev = rllen + startrev
 
     def d():
         rl.clearcaches()
 
         beginrev = startrev
         endrev = rllen
         dist = opts[b'dist']
 
         if reverse:
             beginrev, endrev = endrev - 1, beginrev - 1
             dist = -1 * dist
 
         for x in _xrange(beginrev, endrev, dist):
             # Old revisions don't support passing int.
             n = rl.node(x)
             rl.revision(n)
 
     timer, fm = gettimer(ui, opts)
     timer(d)
     fm.end()
 
 @command(b'perfrevlogwrite', revlogopts + formatteropts +
          [(b's', b'startrev', 1000, b'revision to start writing at'),
           (b'', b'stoprev', -1, b'last revision to write'),
           (b'', b'count', 3, b'number of passes to perform'),
           (b'', b'details', False, b'print timing for every revisions tested'),
           (b'', b'source', b'full', b'the kind of data feed in the revlog'),
           (b'', b'lazydeltabase', True, b'try the provided delta first'),
           (b'', b'clear-caches', True, b'clear revlog cache between calls'),
          ],
          b'-c|-m|FILE')
 def perfrevlogwrite(ui, repo, file_=None, startrev=1000, stoprev=-1, **opts):
     """Benchmark writing a series of revisions to a revlog.
 
     Possible source values are:
     * `full`: add from a full text (default).
     * `parent-1`: add from a delta to the first parent
     * `parent-2`: add from a delta to the second parent if it exists
                   (use a delta from the first parent otherwise)
     * `parent-smallest`: add from the smallest delta (either p1 or p2)
     * `storage`: add from the existing precomputed deltas
 
     Note: This performance command measures performance in a custom way. As a
     result some of the global configuration of the 'perf' command does not
     apply to it:
 
     * ``pre-run``: disabled
 
     * ``profile-benchmark``: disabled
 
     * ``run-limits``: disabled use --count instead
     """
     opts = _byteskwargs(opts)
 
     rl = cmdutil.openrevlog(repo, b'perfrevlogwrite', file_, opts)
     rllen = getlen(ui)(rl)
     if startrev < 0:
         startrev = rllen + startrev
     if stoprev < 0:
         stoprev = rllen + stoprev
 
     lazydeltabase = opts['lazydeltabase']
     source = opts['source']
     clearcaches = opts['clear_caches']
     validsource = (b'full', b'parent-1', b'parent-2', b'parent-smallest',
                    b'storage')
     if source not in validsource:
         raise error.Abort('invalid source type: %s' % source)
 
     ### actually gather results
     count = opts['count']
     if count <= 0:
         raise error.Abort('invalide run count: %d' % count)
     allresults = []
     for c in range(count):
         timing = _timeonewrite(ui, rl, source, startrev, stoprev, c + 1,
                                lazydeltabase=lazydeltabase,
                                clearcaches=clearcaches)
         allresults.append(timing)
 
     ### consolidate the results in a single list
     results = []
     for idx, (rev, t) in enumerate(allresults[0]):
         ts = [t]
         for other in allresults[1:]:
             orev, ot = other[idx]
             assert orev == rev
             ts.append(ot)
         results.append((rev, ts))
     resultcount = len(results)
 
     ### Compute and display relevant statistics
 
     # get a formatter
     fm = ui.formatter(b'perf', opts)
     displayall = ui.configbool(b"perf", b"all-timing", False)
 
     # print individual details if requested
     if opts['details']:
         for idx, item in enumerate(results, 1):
             rev, data = item
             title = 'revisions #%d of %d, rev %d' % (idx, resultcount, rev)
             formatone(fm, data, title=title, displayall=displayall)
 
     # sorts results by median time
     results.sort(key=lambda x: sorted(x[1])[len(x[1]) // 2])
     # list of (name, index) to display)
     relevants = [
         ("min", 0),
         ("10%", resultcount * 10 // 100),
         ("25%", resultcount * 25 // 100),
         ("50%", resultcount * 70 // 100),
         ("75%", resultcount * 75 // 100),
         ("90%", resultcount * 90 // 100),
         ("95%", resultcount * 95 // 100),
         ("99%", resultcount * 99 // 100),
         ("99.9%", resultcount * 999 // 1000),
         ("99.99%", resultcount * 9999 // 10000),
         ("99.999%", resultcount * 99999 // 100000),
         ("max", -1),
     ]
     if not ui.quiet:
         for name, idx in relevants:
             data = results[idx]
             title = '%s of %d, rev %d' % (name, resultcount, data[0])
             formatone(fm, data[1], title=title, displayall=displayall)
 
     # XXX summing that many float will not be very precise, we ignore this fact
     # for now
     totaltime = []
     for item in allresults:
         totaltime.append((sum(x[1][0] for x in item),
                           sum(x[1][1] for x in item),
                           sum(x[1][2] for x in item),)
         )
     formatone(fm, totaltime, title="total time (%d revs)" % resultcount,
               displayall=displayall)
     fm.end()
 
 class _faketr(object):
     def add(s, x, y, z=None):
         return None
 
 def _timeonewrite(ui, orig, source, startrev, stoprev, runidx=None,
                   lazydeltabase=True, clearcaches=True):
     timings = []
     tr = _faketr()
     with _temprevlog(ui, orig, startrev) as dest:
         dest._lazydeltabase = lazydeltabase
         revs = list(orig.revs(startrev, stoprev))
         total = len(revs)
         topic = 'adding'
         if runidx is not None:
             topic += ' (run #%d)' % runidx
          # Support both old and new progress API
         if util.safehasattr(ui, 'makeprogress'):
             progress = ui.makeprogress(topic, unit='revs', total=total)
             def updateprogress(pos):
                 progress.update(pos)
             def completeprogress():
                 progress.complete()
         else:
             def updateprogress(pos):
                 ui.progress(topic, pos, unit='revs', total=total)
             def completeprogress():
                 ui.progress(topic, None, unit='revs', total=total)
 
         for idx, rev in enumerate(revs):
             updateprogress(idx)
             addargs, addkwargs = _getrevisionseed(orig, rev, tr, source)
             if clearcaches:
                 dest.index.clearcaches()
                 dest.clearcaches()
             with timeone() as r:
                 dest.addrawrevision(*addargs, **addkwargs)
             timings.append((rev, r[0]))
         updateprogress(total)
         completeprogress()
     return timings
 
 def _getrevisionseed(orig, rev, tr, source):
     from mercurial.node import nullid
 
     linkrev = orig.linkrev(rev)
     node = orig.node(rev)
     p1, p2 = orig.parents(node)
     flags = orig.flags(rev)
     cachedelta = None
     text = None
 
     if source == b'full':
         text = orig.revision(rev)
     elif source == b'parent-1':
         baserev = orig.rev(p1)
         cachedelta = (baserev, orig.revdiff(p1, rev))
     elif source == b'parent-2':
         parent = p2
         if p2 == nullid:
             parent = p1
         baserev = orig.rev(parent)
         cachedelta = (baserev, orig.revdiff(parent, rev))
     elif source == b'parent-smallest':
         p1diff = orig.revdiff(p1, rev)
         parent = p1
         diff = p1diff
         if p2 != nullid:
             p2diff = orig.revdiff(p2, rev)
             if len(p1diff) > len(p2diff):
                 parent = p2
                 diff = p2diff
         baserev = orig.rev(parent)
         cachedelta = (baserev, diff)
     elif source == b'storage':
         baserev = orig.deltaparent(rev)
         cachedelta = (baserev, orig.revdiff(orig.node(baserev), rev))
 
     return ((text, tr, linkrev, p1, p2),
             {'node': node, 'flags': flags, 'cachedelta': cachedelta})
 
 @contextlib.contextmanager
 def _temprevlog(ui, orig, truncaterev):
     from mercurial import vfs as vfsmod
 
     if orig._inline:
         raise error.Abort('not supporting inline revlog (yet)')
+    revlogkwargs = {}
+    k = 'upperboundcomp'
+    if util.safehasattr(orig, k):
+        revlogkwargs[k] = getattr(orig, k)
 
     origindexpath = orig.opener.join(orig.indexfile)
     origdatapath = orig.opener.join(orig.datafile)
     indexname = 'revlog.i'
     dataname = 'revlog.d'
 
     tmpdir = tempfile.mkdtemp(prefix='tmp-hgperf-')
     try:
         # copy the data file in a temporary directory
         ui.debug('copying data in %s\n' % tmpdir)
         destindexpath = os.path.join(tmpdir, 'revlog.i')
         destdatapath = os.path.join(tmpdir, 'revlog.d')
         shutil.copyfile(origindexpath, destindexpath)
         shutil.copyfile(origdatapath, destdatapath)
 
         # remove the data we want to add again
         ui.debug('truncating data to be rewritten\n')
         with open(destindexpath, 'ab') as index:
             index.seek(0)
             index.truncate(truncaterev * orig._io.size)
         with open(destdatapath, 'ab') as data:
             data.seek(0)
             data.truncate(orig.start(truncaterev))
 
         # instantiate a new revlog from the temporary copy
         ui.debug('truncating adding to be rewritten\n')
         vfs = vfsmod.vfs(tmpdir)
         vfs.options = getattr(orig.opener, 'options', None)
 
         dest = revlog.revlog(vfs,
                              indexfile=indexname,
-                             datafile=dataname)
+                             datafile=dataname, **revlogkwargs)
         if dest._inline:
             raise error.Abort('not supporting inline revlog (yet)')
         # make sure internals are initialized
         dest.revision(len(dest) - 1)
         yield dest
         del dest, vfs
     finally:
         shutil.rmtree(tmpdir, True)
 
 @command(b'perfrevlogchunks', revlogopts + formatteropts +
          [(b'e', b'engines', b'', b'compression engines to use'),
           (b's', b'startrev', 0, b'revision to start at')],
          b'-c|-m|FILE')
 def perfrevlogchunks(ui, repo, file_=None, engines=None, startrev=0, **opts):
     """Benchmark operations on revlog chunks.
 
     Logically, each revlog is a collection of fulltext revisions. However,
     stored within each revlog are "chunks" of possibly compressed data. This
     data needs to be read and decompressed or compressed and written.
 
     This command measures the time it takes to read+decompress and recompress
     chunks in a revlog. It effectively isolates I/O and compression performance.
     For measurements of higher-level operations like resolving revisions,
     see ``perfrevlogrevisions`` and ``perfrevlogrevision``.
     """
     opts = _byteskwargs(opts)
 
     rl = cmdutil.openrevlog(repo, b'perfrevlogchunks', file_, opts)
 
     # _chunkraw was renamed to _getsegmentforrevs.
     try:
         segmentforrevs = rl._getsegmentforrevs
     except AttributeError:
         segmentforrevs = rl._chunkraw
 
     # Verify engines argument.
     if engines:
         engines = set(e.strip() for e in engines.split(b','))
         for engine in engines:
             try:
                 util.compressionengines[engine]
             except KeyError:
                 raise error.Abort(b'unknown compression engine: %s' % engine)
     else:
         engines = []
         for e in util.compengines:
             engine = util.compengines[e]
             try:
                 if engine.available():
                     engine.revlogcompressor().compress(b'dummy')
                     engines.append(e)
             except NotImplementedError:
                 pass
 
     revs = list(rl.revs(startrev, len(rl) - 1))
 
     def rlfh(rl):
         if rl._inline:
             return getsvfs(repo)(rl.indexfile)
         else:
             return getsvfs(repo)(rl.datafile)
 
     def doread():
         rl.clearcaches()
         for rev in revs:
             segmentforrevs(rev, rev)
 
     def doreadcachedfh():
         rl.clearcaches()
         fh = rlfh(rl)
         for rev in revs:
             segmentforrevs(rev, rev, df=fh)
 
     def doreadbatch():
         rl.clearcaches()
         segmentforrevs(revs[0], revs[-1])
 
     def doreadbatchcachedfh():
         rl.clearcaches()
         fh = rlfh(rl)
         segmentforrevs(revs[0], revs[-1], df=fh)
 
     def dochunk():
         rl.clearcaches()
         fh = rlfh(rl)
         for rev in revs:
             rl._chunk(rev, df=fh)
 
     chunks = [None]
 
     def dochunkbatch():
         rl.clearcaches()
         fh = rlfh(rl)
         # Save chunks as a side-effect.
         chunks[0] = rl._chunks(revs, df=fh)
 
     def docompress(compressor):
         rl.clearcaches()
 
         try:
             # Swap in the requested compression engine.
             oldcompressor = rl._compressor
             rl._compressor = compressor
             for chunk in chunks[0]:
                 rl.compress(chunk)
         finally:
             rl._compressor = oldcompressor
 
     benches = [
         (lambda: doread(), b'read'),
         (lambda: doreadcachedfh(), b'read w/ reused fd'),
         (lambda: doreadbatch(), b'read batch'),
         (lambda: doreadbatchcachedfh(), b'read batch w/ reused fd'),
         (lambda: dochunk(), b'chunk'),
         (lambda: dochunkbatch(), b'chunk batch'),
     ]
 
     for engine in sorted(engines):
         compressor = util.compengines[engine].revlogcompressor()
         benches.append((functools.partial(docompress, compressor),
                         b'compress w/ %s' % engine))
 
     for fn, title in benches:
         timer, fm = gettimer(ui, opts)
         timer(fn, title=title)
         fm.end()
 
 @command(b'perfrevlogrevision', revlogopts + formatteropts +
          [(b'', b'cache', False, b'use caches instead of clearing')],
          b'-c|-m|FILE REV')
 def perfrevlogrevision(ui, repo, file_, rev=None, cache=None, **opts):
     """Benchmark obtaining a revlog revision.
 
     Obtaining a revlog revision consists of roughly the following steps:
 
     1. Compute the delta chain
     2. Slice the delta chain if applicable
     3. Obtain the raw chunks for that delta chain
     4. Decompress each raw chunk
     5. Apply binary patches to obtain fulltext
     6. Verify hash of fulltext
 
     This command measures the time spent in each of these phases.
     """
     opts = _byteskwargs(opts)
 
     if opts.get(b'changelog') or opts.get(b'manifest'):
         file_, rev = None, file_
     elif rev is None:
         raise error.CommandError(b'perfrevlogrevision', b'invalid arguments')
 
     r = cmdutil.openrevlog(repo, b'perfrevlogrevision', file_, opts)
 
     # _chunkraw was renamed to _getsegmentforrevs.
     try:
         segmentforrevs = r._getsegmentforrevs
     except AttributeError:
         segmentforrevs = r._chunkraw
 
     node = r.lookup(rev)
     rev = r.rev(node)
 
     def getrawchunks(data, chain):
         start = r.start
         length = r.length
         inline = r._inline
         iosize = r._io.size
         buffer = util.buffer
 
         chunks = []
         ladd = chunks.append
         for idx, item in enumerate(chain):
             offset = start(item[0])
             bits = data[idx]
             for rev in item:
                 chunkstart = start(rev)
                 if inline:
                     chunkstart += (rev + 1) * iosize
                 chunklength = length(rev)
                 ladd(buffer(bits, chunkstart - offset, chunklength))
 
         return chunks
 
     def dodeltachain(rev):
         if not cache:
             r.clearcaches()
         r._deltachain(rev)
 
     def doread(chain):
         if not cache:
             r.clearcaches()
         for item in slicedchain:
             segmentforrevs(item[0], item[-1])
 
     def doslice(r, chain, size):
         for s in slicechunk(r, chain, targetsize=size):
             pass
 
     def dorawchunks(data, chain):
         if not cache:
             r.clearcaches()
         getrawchunks(data, chain)
 
     def dodecompress(chunks):
         decomp = r.decompress
         for chunk in chunks:
             decomp(chunk)
 
     def dopatch(text, bins):
         if not cache:
             r.clearcaches()
         mdiff.patches(text, bins)
 
     def dohash(text):
         if not cache:
             r.clearcaches()
         r.checkhash(text, node, rev=rev)
 
     def dorevision():
         if not cache:
             r.clearcaches()
         r.revision(node)
 
     try:
         from mercurial.revlogutils.deltas import slicechunk
     except ImportError:
         slicechunk = getattr(revlog, '_slicechunk', None)
 
     size = r.length(rev)
     chain = r._deltachain(rev)[0]
     if not getattr(r, '_withsparseread', False):
         slicedchain = (chain,)
     else:
         slicedchain = tuple(slicechunk(r, chain, targetsize=size))
     data = [segmentforrevs(seg[0], seg[-1])[1] for seg in slicedchain]
     rawchunks = getrawchunks(data, slicedchain)
     bins = r._chunks(chain)
     text = bytes(bins[0])
     bins = bins[1:]
     text = mdiff.patches(text, bins)
 
     benches = [
         (lambda: dorevision(), b'full'),
         (lambda: dodeltachain(rev), b'deltachain'),
         (lambda: doread(chain), b'read'),
     ]
 
     if getattr(r, '_withsparseread', False):
         slicing = (lambda: doslice(r, chain, size), b'slice-sparse-chain')
         benches.append(slicing)
 
     benches.extend([
         (lambda: dorawchunks(data, slicedchain), b'rawchunks'),
         (lambda: dodecompress(rawchunks), b'decompress'),
         (lambda: dopatch(text, bins), b'patch'),
         (lambda: dohash(text), b'hash'),
     ])
 
     timer, fm = gettimer(ui, opts)
     for fn, title in benches:
         timer(fn, title=title)
     fm.end()
 
 @command(b'perfrevset',
          [(b'C', b'clear', False, b'clear volatile cache between each call.'),
           (b'', b'contexts', False, b'obtain changectx for each revision')]
          + formatteropts, b"REVSET")
 def perfrevset(ui, repo, expr, clear=False, contexts=False, **opts):
     """benchmark the execution time of a revset
 
     Use the --clean option if need to evaluate the impact of build volatile
     revisions set cache on the revset execution. Volatile cache hold filtered
     and obsolete related cache."""
     opts = _byteskwargs(opts)
 
     timer, fm = gettimer(ui, opts)
     def d():
         if clear:
             repo.invalidatevolatilesets()
         if contexts:
             for ctx in repo.set(expr): pass
         else:
             for r in repo.revs(expr): pass
     timer(d)
     fm.end()
 
 @command(b'perfvolatilesets',
          [(b'', b'clear-obsstore', False, b'drop obsstore between each call.'),
           ] + formatteropts)
 def perfvolatilesets(ui, repo, *names, **opts):
     """benchmark the computation of various volatile set
 
     Volatile set computes element related to filtering and obsolescence."""
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     repo = repo.unfiltered()
 
     def getobs(name):
         def d():
             repo.invalidatevolatilesets()
             if opts[b'clear_obsstore']:
                 clearfilecache(repo, b'obsstore')
             obsolete.getrevs(repo, name)
         return d
 
     allobs = sorted(obsolete.cachefuncs)
     if names:
         allobs = [n for n in allobs if n in names]
 
     for name in allobs:
         timer(getobs(name), title=name)
 
     def getfiltered(name):
         def d():
             repo.invalidatevolatilesets()
             if opts[b'clear_obsstore']:
                 clearfilecache(repo, b'obsstore')
             repoview.filterrevs(repo, name)
         return d
 
     allfilter = sorted(repoview.filtertable)
     if names:
         allfilter = [n for n in allfilter if n in names]
 
     for name in allfilter:
         timer(getfiltered(name), title=name)
     fm.end()
 
 @command(b'perfbranchmap',
          [(b'f', b'full', False,
            b'Includes build time of subset'),
           (b'', b'clear-revbranch', False,
            b'purge the revbranch cache between computation'),
           ] + formatteropts)
 def perfbranchmap(ui, repo, *filternames, **opts):
     """benchmark the update of a branchmap
 
     This benchmarks the full repo.branchmap() call with read and write disabled
     """
     opts = _byteskwargs(opts)
     full = opts.get(b"full", False)
     clear_revbranch = opts.get(b"clear_revbranch", False)
     timer, fm = gettimer(ui, opts)
     def getbranchmap(filtername):
         """generate a benchmark function for the filtername"""
         if filtername is None:
             view = repo
         else:
             view = repo.filtered(filtername)
         if util.safehasattr(view._branchcaches, '_per_filter'):
             filtered = view._branchcaches._per_filter
         else:
             # older versions
             filtered = view._branchcaches
         def d():
             if clear_revbranch:
                 repo.revbranchcache()._clear()
             if full:
                 view._branchcaches.clear()
             else:
                 filtered.pop(filtername, None)
             view.branchmap()
         return d
     # add filter in smaller subset to bigger subset
     possiblefilters = set(repoview.filtertable)
     if filternames:
         possiblefilters &= set(filternames)
     subsettable = getbranchmapsubsettable()
     allfilters = []
     while possiblefilters:
         for name in possiblefilters:
             subset = subsettable.get(name)
             if subset not in possiblefilters:
                 break
         else:
             assert False, b'subset cycle %s!' % possiblefilters
         allfilters.append(name)
         possiblefilters.remove(name)
 
     # warm the cache
     if not full:
         for name in allfilters:
             repo.filtered(name).branchmap()
     if not filternames or b'unfiltered' in filternames:
         # add unfiltered
         allfilters.append(None)
 
     if util.safehasattr(branchmap.branchcache, 'fromfile'):
         branchcacheread = safeattrsetter(branchmap.branchcache, b'fromfile')
         branchcacheread.set(classmethod(lambda *args: None))
     else:
         # older versions
         branchcacheread = safeattrsetter(branchmap, b'read')
         branchcacheread.set(lambda *args: None)
     branchcachewrite = safeattrsetter(branchmap.branchcache, b'write')
     branchcachewrite.set(lambda *args: None)
     try:
         for name in allfilters:
             printname = name
             if name is None:
                 printname = b'unfiltered'
             timer(getbranchmap(name), title=str(printname))
     finally:
         branchcacheread.restore()
         branchcachewrite.restore()
     fm.end()
 
 @command(b'perfbranchmapupdate', [
      (b'', b'base', [], b'subset of revision to start from'),
      (b'', b'target', [], b'subset of revision to end with'),
      (b'', b'clear-caches', False, b'clear cache between each runs')
     ] + formatteropts)
 def perfbranchmapupdate(ui, repo, base=(), target=(), **opts):
     """benchmark branchmap update from for <base> revs to <target> revs
 
     If `--clear-caches` is passed, the following items will be reset before
     each update:
         * the changelog instance and associated indexes
         * the rev-branch-cache instance
 
     Examples:
 
        # update for the one last revision
        $ hg perfbranchmapupdate --base 'not tip' --target 'tip'
 
        $ update for change coming with a new branch
        $ hg perfbranchmapupdate --base 'stable' --target 'default'
     """
     from mercurial import branchmap
     from mercurial import repoview
     opts = _byteskwargs(opts)
     timer, fm = gettimer(ui, opts)
     clearcaches = opts[b'clear_caches']
     unfi = repo.unfiltered()
     x = [None] # used to pass data between closure
 
     # we use a `list` here to avoid possible side effect from smartset
     baserevs = list(scmutil.revrange(repo, base))
     targetrevs = list(scmutil.revrange(repo, target))
     if not baserevs:
         raise error.Abort(b'no revisions selected for --base')
     if not targetrevs:
         raise error.Abort(b'no revisions selected for --target')
 
     # make sure the target branchmap also contains the one in the base
     targetrevs = list(set(baserevs) | set(targetrevs))
     targetrevs.sort()
 
     cl = repo.changelog
     allbaserevs = list(cl.ancestors(baserevs, inclusive=True))
     allbaserevs.sort()
     alltargetrevs = frozenset(cl.ancestors(targetrevs, inclusive=True))
 
     newrevs = list(alltargetrevs.difference(allbaserevs))
     newrevs.sort()
 
     allrevs = frozenset(unfi.changelog.revs())
     basefilterrevs = frozenset(allrevs.difference(allbaserevs))
     targetfilterrevs = frozenset(allrevs.difference(alltargetrevs))
 
     def basefilter(repo, visibilityexceptions=None):
         return basefilterrevs
 
     def targetfilter(repo, visibilityexceptions=None):
         return targetfilterrevs
 
     msg = b'benchmark of branchmap with %d revisions with %d new ones\n'
     ui.status(msg % (len(allbaserevs), len(newrevs)))
     if targetfilterrevs:
         msg = b'(%d revisions still filtered)\n'
         ui.status(msg % len(targetfilterrevs))
 
     try:
         repoview.filtertable[b'__perf_branchmap_update_base'] = basefilter
         repoview.filtertable[b'__perf_branchmap_update_target'] = targetfilter
 
         baserepo = repo.filtered(b'__perf_branchmap_update_base')
         targetrepo = repo.filtered(b'__perf_branchmap_update_target')
 
         # try to find an existing branchmap to reuse
         subsettable = getbranchmapsubsettable()
         candidatefilter = subsettable.get(None)
         while candidatefilter is not None:
             candidatebm = repo.filtered(candidatefilter).branchmap()
             if candidatebm.validfor(baserepo):
                 filtered = repoview.filterrevs(repo, candidatefilter)
                 missing = [r for r in allbaserevs if r in filtered]
                 base = candidatebm.copy()
                 base.update(baserepo, missing)
                 break
             candidatefilter = subsettable.get(candidatefilter)
         else:
             # no suitable subset where found
             base = branchmap.branchcache()
             base.update(baserepo, allbaserevs)
 
         def setup():
             x[0] = base.copy()
             if clearcaches:
                 unfi._revbranchcache = None
                 clearchangelog(repo)
 
         def bench():
             x[0].update(targetrepo, newrevs)
 
         timer(bench, setup=setup)
         fm.end()
     finally:
         repoview.filtertable.pop(b'__perf_branchmap_update_base', None)
         repoview.filtertable.pop(b'__perf_branchmap_update_target', None)
 
 @command(b'perfbranchmapload', [
      (b'f', b'filter', b'', b'Specify repoview filter'),
      (b'', b'list', False, b'List brachmap filter caches'),
      (b'', b'clear-revlogs', False, b'refresh changelog and manifest'),
 
     ] + formatteropts)
 def perfbranchmapload(ui, repo, filter=b'', list=False, **opts):
     """benchmark reading the branchmap"""
     opts = _byteskwargs(opts)
     clearrevlogs = opts[b'clear_revlogs']
 
     if list:
         for name, kind, st in repo.cachevfs.readdir(stat=True):
             if name.startswith(b'branch2'):
                 filtername = name.partition(b'-')[2] or b'unfiltered'
                 ui.status(b'%s - %s\n'
                           % (filtername, util.bytecount(st.st_size)))
         return
     if not filter:
         filter = None
     subsettable = getbranchmapsubsettable()
     if filter is None:
         repo = repo.unfiltered()
     else:
         repo = repoview.repoview(repo, filter)
 
     repo.branchmap() # make sure we have a relevant, up to date branchmap
 
     try:
         fromfile = branchmap.branchcache.fromfile
     except AttributeError:
         # older versions
         fromfile = branchmap.read
 
     currentfilter = filter
     # try once without timer, the filter may not be cached
     while fromfile(repo) is None:
         currentfilter = subsettable.get(currentfilter)
         if currentfilter is None:
             raise error.Abort(b'No branchmap cached for %s repo'
                               % (filter or b'unfiltered'))
         repo = repo.filtered(currentfilter)
     timer, fm = gettimer(ui, opts)
     def setup():
         if clearrevlogs:
             clearchangelog(repo)
     def bench():
         fromfile(repo)
     timer(bench, setup=setup)
     fm.end()
 
 @command(b'perfloadmarkers')
 def perfloadmarkers(ui, repo):
     """benchmark the time to parse the on-disk markers for a repo
 
     Result is the number of markers in the repo."""
     timer, fm = gettimer(ui)
     svfs = getsvfs(repo)
     timer(lambda: len(obsolete.obsstore(svfs)))
     fm.end()
 
 @command(b'perflrucachedict', formatteropts +
     [(b'', b'costlimit', 0, b'maximum total cost of items in cache'),
      (b'', b'mincost', 0, b'smallest cost of items in cache'),
      (b'', b'maxcost', 100, b'maximum cost of items in cache'),
      (b'', b'size', 4, b'size of cache'),
      (b'', b'gets', 10000, b'number of key lookups'),
      (b'', b'sets', 10000, b'number of key sets'),
      (b'', b'mixed', 10000, b'number of mixed mode operations'),
      (b'', b'mixedgetfreq', 50, b'frequency of get vs set ops in mixed mode')],
     norepo=True)
 def perflrucache(ui, mincost=0, maxcost=100, costlimit=0, size=4,
                  gets=10000, sets=10000, mixed=10000, mixedgetfreq=50, **opts):
     opts = _byteskwargs(opts)
 
     def doinit():
         for i in _xrange(10000):
             util.lrucachedict(size)
 
     costrange = list(range(mincost, maxcost + 1))
 
     values = []
     for i in _xrange(size):
         values.append(random.randint(0, _maxint))
 
     # Get mode fills the cache and tests raw lookup performance with no
     # eviction.
     getseq = []
     for i in _xrange(gets):
         getseq.append(random.choice(values))
 
     def dogets():
         d = util.lrucachedict(size)
         for v in values:
             d[v] = v
         for key in getseq:
             value = d[key]
             value # silence pyflakes warning
 
     def dogetscost():
         d = util.lrucachedict(size, maxcost=costlimit)
         for i, v in enumerate(values):
             d.insert(v, v, cost=costs[i])
         for key in getseq:
             try:
                 value = d[key]
                 value # silence pyflakes warning
             except KeyError:
                 pass
 
     # Set mode tests insertion speed with cache eviction.
     setseq = []
     costs = []
     for i in _xrange(sets):
         setseq.append(random.randint(0, _maxint))
         costs.append(random.choice(costrange))
 
     def doinserts():
         d = util.lrucachedict(size)
         for v in setseq:
             d.insert(v, v)
 
     def doinsertscost():
         d = util.lrucachedict(size, maxcost=costlimit)
         for i, v in enumerate(setseq):
             d.insert(v, v, cost=costs[i])
 
     def dosets():
         d = util.lrucachedict(size)
         for v in setseq:
             d[v] = v
 
     # Mixed mode randomly performs gets and sets with eviction.
     mixedops = []
     for i in _xrange(mixed):
         r = random.randint(0, 100)
         if r < mixedgetfreq:
             op = 0
         else:
             op = 1
 
         mixedops.append((op,
                          random.randint(0, size * 2),
                          random.choice(costrange)))
 
     def domixed():
         d = util.lrucachedict(size)
 
         for op, v, cost in mixedops:
             if op == 0:
                 try:
                     d[v]
                 except KeyError:
                     pass
             else:
                 d[v] = v
 
     def domixedcost():
         d = util.lrucachedict(size, maxcost=costlimit)
 
         for op, v, cost in mixedops:
             if op == 0:
                 try:
                     d[v]
                 except KeyError:
                     pass
             else:
                 d.insert(v, v, cost=cost)
 
     benches = [
         (doinit, b'init'),
     ]
 
     if costlimit:
         benches.extend([
             (dogetscost, b'gets w/ cost limit'),
             (doinsertscost, b'inserts w/ cost limit'),
             (domixedcost, b'mixed w/ cost limit'),
         ])
     else:
         benches.extend([
             (dogets, b'gets'),
             (doinserts, b'inserts'),
             (dosets, b'sets'),
             (domixed, b'mixed')
         ])
 
     for fn, title in benches:
         timer, fm = gettimer(ui, opts)
         timer(fn, title=title)
         fm.end()
 
 @command(b'perfwrite', formatteropts)
 def perfwrite(ui, repo, **opts):
     """microbenchmark ui.write
     """
     opts = _byteskwargs(opts)
 
     timer, fm = gettimer(ui, opts)
     def write():
         for i in range(100000):
             ui.write((b'Testing write performance\n'))
     timer(write)
     fm.end()
 
 def uisetup(ui):
     if (util.safehasattr(cmdutil, b'openrevlog') and
         not util.safehasattr(commands, b'debugrevlogopts')):
         # for "historical portability":
         # In this case, Mercurial should be 1.9 (or a79fea6b3e77) -
         # 3.7 (or 5606f7d0d063). Therefore, '--dir' option for
         # openrevlog() should cause failure, because it has been
         # available since 3.5 (or 49c583ca48c4).
         def openrevlog(orig, repo, cmd, file_, opts):
             if opts.get(b'dir') and not util.safehasattr(repo, b'dirlog'):
                 raise error.Abort(b"This version doesn't support --dir option",
                                   hint=b"use 3.5 or later")
             return orig(repo, cmd, file_, opts)
         extensions.wrapfunction(cmdutil, b'openrevlog', openrevlog)
 
 @command(b'perfprogress', formatteropts + [
     (b'', b'topic', b'topic', b'topic for progress messages'),
     (b'c', b'total', 1000000, b'total value we are progressing to'),
 ], norepo=True)
 def perfprogress(ui, topic=None, total=None, **opts):
     """printing of progress bars"""
     opts = _byteskwargs(opts)
 
     timer, fm = gettimer(ui, opts)
 
     def doprogress():
         with ui.makeprogress(topic, total=total) as progress:
             for i in pycompat.xrange(total):
                 progress.increment()
 
     timer(doprogress)
     fm.end()

mercurial/manifest.py

@@ -1,2095 +1,2100 b''
 # manifest.py - manifest revision class for mercurial
 #
 # Copyright 2005-2007 Matt Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 import heapq
 import itertools
 import struct
 import weakref
 
 from .i18n import _
 from .node import (
     bin,
     hex,
     nullid,
     nullrev,
 )
 from . import (
     error,
     mdiff,
     policy,
     pycompat,
     repository,
     revlog,
     util,
 )
 from .utils import (
     interfaceutil,
 )
 
 parsers = policy.importmod(r'parsers')
 propertycache = util.propertycache
 
 # Allow tests to more easily test the alternate path in manifestdict.fastdelta()
 FASTDELTA_TEXTDIFF_THRESHOLD = 1000
 
 def _parse(data):
     # This method does a little bit of excessive-looking
     # precondition checking. This is so that the behavior of this
     # class exactly matches its C counterpart to try and help
     # prevent surprise breakage for anyone that develops against
     # the pure version.
     if data and data[-1:] != '\n':
         raise ValueError('Manifest did not end in a newline.')
     prev = None
     for l in data.splitlines():
         if prev is not None and prev > l:
             raise ValueError('Manifest lines not in sorted order.')
         prev = l
         f, n = l.split('\0')
         if len(n) > 40:
             yield f, bin(n[:40]), n[40:]
         else:
             yield f, bin(n), ''
 
 def _text(it):
     files = []
     lines = []
     for f, n, fl in it:
         files.append(f)
         # if this is changed to support newlines in filenames,
         # be sure to check the templates/ dir again (especially *-raw.tmpl)
         lines.append("%s\0%s%s\n" % (f, hex(n), fl))
 
     _checkforbidden(files)
     return ''.join(lines)
 
 class lazymanifestiter(object):
     def __init__(self, lm):
         self.pos = 0
         self.lm = lm
 
     def __iter__(self):
         return self
 
     def next(self):
         try:
             data, pos = self.lm._get(self.pos)
         except IndexError:
             raise StopIteration
         if pos == -1:
             self.pos += 1
             return data[0]
         self.pos += 1
         zeropos = data.find('\x00', pos)
         return data[pos:zeropos]
 
     __next__ = next
 
 class lazymanifestiterentries(object):
     def __init__(self, lm):
         self.lm = lm
         self.pos = 0
 
     def __iter__(self):
         return self
 
     def next(self):
         try:
             data, pos = self.lm._get(self.pos)
         except IndexError:
             raise StopIteration
         if pos == -1:
             self.pos += 1
             return data
         zeropos = data.find('\x00', pos)
         hashval = unhexlify(data, self.lm.extrainfo[self.pos],
                             zeropos + 1, 40)
         flags = self.lm._getflags(data, self.pos, zeropos)
         self.pos += 1
         return (data[pos:zeropos], hashval, flags)
 
     __next__ = next
 
 def unhexlify(data, extra, pos, length):
     s = bin(data[pos:pos + length])
     if extra:
         s += chr(extra & 0xff)
     return s
 
 def _cmp(a, b):
     return (a > b) - (a < b)
 
 class _lazymanifest(object):
     """A pure python manifest backed by a byte string.  It is supplimented with
     internal lists as it is modified, until it is compacted back to a pure byte
     string.
 
     ``data`` is the initial manifest data.
 
     ``positions`` is a list of offsets, one per manifest entry.  Positive
     values are offsets into ``data``, negative values are offsets into the
     ``extradata`` list.  When an entry is removed, its entry is dropped from
     ``positions``.  The values are encoded such that when walking the list and
     indexing into ``data`` or ``extradata`` as appropriate, the entries are
     sorted by filename.
 
     ``extradata`` is a list of (key, hash, flags) for entries that were added or
     modified since the manifest was created or compacted.
     """
     def __init__(self, data, positions=None, extrainfo=None, extradata=None,
                  hasremovals=False):
         if positions is None:
             self.positions = self.findlines(data)
             self.extrainfo = [0] * len(self.positions)
             self.data = data
             self.extradata = []
             self.hasremovals = False
         else:
             self.positions = positions[:]
             self.extrainfo = extrainfo[:]
             self.extradata = extradata[:]
             self.data = data
             self.hasremovals = hasremovals
 
     def findlines(self, data):
         if not data:
             return []
         pos = data.find("\n")
         if pos == -1 or data[-1:] != '\n':
             raise ValueError("Manifest did not end in a newline.")
         positions = [0]
         prev = data[:data.find('\x00')]
         while pos < len(data) - 1 and pos != -1:
             positions.append(pos + 1)
             nexts = data[pos + 1:data.find('\x00', pos + 1)]
             if nexts < prev:
                 raise ValueError("Manifest lines not in sorted order.")
             prev = nexts
             pos = data.find("\n", pos + 1)
         return positions
 
     def _get(self, index):
         # get the position encoded in pos:
         #   positive number is an index in 'data'
         #   negative number is in extrapieces
         pos = self.positions[index]
         if pos >= 0:
             return self.data, pos
         return self.extradata[-pos - 1], -1
 
     def _getkey(self, pos):
         if pos >= 0:
             return self.data[pos:self.data.find('\x00', pos + 1)]
         return self.extradata[-pos - 1][0]
 
     def bsearch(self, key):
         first = 0
         last = len(self.positions) - 1
 
         while first <= last:
             midpoint = (first + last)//2
             nextpos = self.positions[midpoint]
             candidate = self._getkey(nextpos)
             r = _cmp(key, candidate)
             if r == 0:
                 return midpoint
             else:
                 if r < 0:
                     last = midpoint - 1
                 else:
                     first = midpoint + 1
         return -1
 
     def bsearch2(self, key):
         # same as the above, but will always return the position
         # done for performance reasons
         first = 0
         last = len(self.positions) - 1
 
         while first <= last:
             midpoint = (first + last)//2
             nextpos = self.positions[midpoint]
             candidate = self._getkey(nextpos)
             r = _cmp(key, candidate)
             if r == 0:
                 return (midpoint, True)
             else:
                 if r < 0:
                     last = midpoint - 1
                 else:
                     first = midpoint + 1
         return (first, False)
 
     def __contains__(self, key):
         return self.bsearch(key) != -1
 
     def _getflags(self, data, needle, pos):
         start = pos + 41
         end = data.find("\n", start)
         if end == -1:
             end = len(data) - 1
         if start == end:
             return ''
         return self.data[start:end]
 
     def __getitem__(self, key):
         if not isinstance(key, bytes):
             raise TypeError("getitem: manifest keys must be a bytes.")
         needle = self.bsearch(key)
         if needle == -1:
             raise KeyError
         data, pos = self._get(needle)
         if pos == -1:
             return (data[1], data[2])
         zeropos = data.find('\x00', pos)
         assert 0 <= needle <= len(self.positions)
         assert len(self.extrainfo) == len(self.positions)
         hashval = unhexlify(data, self.extrainfo[needle], zeropos + 1, 40)
         flags = self._getflags(data, needle, zeropos)
         return (hashval, flags)
 
     def __delitem__(self, key):
         needle, found = self.bsearch2(key)
         if not found:
             raise KeyError
         cur = self.positions[needle]
         self.positions = self.positions[:needle] + self.positions[needle + 1:]
         self.extrainfo = self.extrainfo[:needle] + self.extrainfo[needle + 1:]
         if cur >= 0:
             # This does NOT unsort the list as far as the search functions are
             # concerned, as they only examine lines mapped by self.positions.
             self.data = self.data[:cur] + '\x00' + self.data[cur + 1:]
             self.hasremovals = True
 
     def __setitem__(self, key, value):
         if not isinstance(key, bytes):
             raise TypeError("setitem: manifest keys must be a byte string.")
         if not isinstance(value, tuple) or len(value) != 2:
             raise TypeError("Manifest values must be a tuple of (node, flags).")
         hashval = value[0]
         if not isinstance(hashval, bytes) or not 20 <= len(hashval) <= 22:
             raise TypeError("node must be a 20-byte byte string")
         flags = value[1]
         if len(hashval) == 22:
             hashval = hashval[:-1]
         if not isinstance(flags, bytes) or len(flags) > 1:
             raise TypeError("flags must a 0 or 1 byte string, got %r", flags)
         needle, found = self.bsearch2(key)
         if found:
             # put the item
             pos = self.positions[needle]
             if pos < 0:
                 self.extradata[-pos - 1] = (key, hashval, value[1])
             else:
                 # just don't bother
                 self.extradata.append((key, hashval, value[1]))
                 self.positions[needle] = -len(self.extradata)
         else:
             # not found, put it in with extra positions
             self.extradata.append((key, hashval, value[1]))
             self.positions = (self.positions[:needle] + [-len(self.extradata)]
                               + self.positions[needle:])
             self.extrainfo = (self.extrainfo[:needle] + [0] +
                               self.extrainfo[needle:])
 
     def copy(self):
         # XXX call _compact like in C?
         return _lazymanifest(self.data, self.positions, self.extrainfo,
             self.extradata, self.hasremovals)
 
     def _compact(self):
         # hopefully not called TOO often
         if len(self.extradata) == 0 and not self.hasremovals:
             return
         l = []
         i = 0
         offset = 0
         self.extrainfo = [0] * len(self.positions)
         while i < len(self.positions):
             if self.positions[i] >= 0:
                 cur = self.positions[i]
                 last_cut = cur
 
                 # Collect all contiguous entries in the buffer at the current
                 # offset, breaking out only for added/modified items held in
                 # extradata, or a deleted line prior to the next position.
                 while True:
                     self.positions[i] = offset
                     i += 1
                     if i == len(self.positions) or self.positions[i] < 0:
                         break
 
                     # A removed file has no positions[] entry, but does have an
                     # overwritten first byte.  Break out and find the end of the
                     # current good entry/entries if there is a removed file
                     # before the next position.
                     if (self.hasremovals
                         and self.data.find('\n\x00', cur,
                                            self.positions[i]) != -1):
                         break
 
                     offset += self.positions[i] - cur
                     cur = self.positions[i]
                 end_cut = self.data.find('\n', cur)
                 if end_cut != -1:
                     end_cut += 1
                 offset += end_cut - cur
                 l.append(self.data[last_cut:end_cut])
             else:
                 while i < len(self.positions) and self.positions[i] < 0:
                     cur = self.positions[i]
                     t = self.extradata[-cur - 1]
                     l.append(self._pack(t))
                     self.positions[i] = offset
                     if len(t[1]) > 20:
                         self.extrainfo[i] = ord(t[1][21])
                     offset += len(l[-1])
                     i += 1
         self.data = ''.join(l)
         self.hasremovals = False
         self.extradata = []
 
     def _pack(self, d):
         return d[0] + '\x00' + hex(d[1][:20]) + d[2] + '\n'
 
     def text(self):
         self._compact()
         return self.data
 
     def diff(self, m2, clean=False):
         '''Finds changes between the current manifest and m2.'''
         # XXX think whether efficiency matters here
         diff = {}
 
         for fn, e1, flags in self.iterentries():
             if fn not in m2:
                 diff[fn] = (e1, flags), (None, '')
             else:
                 e2 = m2[fn]
                 if (e1, flags) != e2:
                     diff[fn] = (e1, flags), e2
                 elif clean:
                     diff[fn] = None
 
         for fn, e2, flags in m2.iterentries():
             if fn not in self:
                 diff[fn] = (None, ''), (e2, flags)
 
         return diff
 
     def iterentries(self):
         return lazymanifestiterentries(self)
 
     def iterkeys(self):
         return lazymanifestiter(self)
 
     def __iter__(self):
         return lazymanifestiter(self)
 
     def __len__(self):
         return len(self.positions)
 
     def filtercopy(self, filterfn):
         # XXX should be optimized
         c = _lazymanifest('')
         for f, n, fl in self.iterentries():
             if filterfn(f):
                 c[f] = n, fl
         return c
 
 try:
     _lazymanifest = parsers.lazymanifest
 except AttributeError:
     pass
 
 @interfaceutil.implementer(repository.imanifestdict)
 class manifestdict(object):
     def __init__(self, data=''):
         self._lm = _lazymanifest(data)
 
     def __getitem__(self, key):
         return self._lm[key][0]
 
     def find(self, key):
         return self._lm[key]
 
     def __len__(self):
         return len(self._lm)
 
     def __nonzero__(self):
         # nonzero is covered by the __len__ function, but implementing it here
         # makes it easier for extensions to override.
         return len(self._lm) != 0
 
     __bool__ = __nonzero__
 
     def __setitem__(self, key, node):
         self._lm[key] = node, self.flags(key, '')
 
     def __contains__(self, key):
         if key is None:
             return False
         return key in self._lm
 
     def __delitem__(self, key):
         del self._lm[key]
 
     def __iter__(self):
         return self._lm.__iter__()
 
     def iterkeys(self):
         return self._lm.iterkeys()
 
     def keys(self):
         return list(self.iterkeys())
 
     def filesnotin(self, m2, match=None):
         '''Set of files in this manifest that are not in the other'''
         if match:
             m1 = self.matches(match)
             m2 = m2.matches(match)
             return m1.filesnotin(m2)
         diff = self.diff(m2)
         files = set(filepath
                     for filepath, hashflags in diff.iteritems()
                     if hashflags[1][0] is None)
         return files
 
     @propertycache
     def _dirs(self):
         return util.dirs(self)
 
     def dirs(self):
         return self._dirs
 
     def hasdir(self, dir):
         return dir in self._dirs
 
     def _filesfastpath(self, match):
         '''Checks whether we can correctly and quickly iterate over matcher
         files instead of over manifest files.'''
         files = match.files()
         return (len(files) < 100 and (match.isexact() or
             (match.prefix() and all(fn in self for fn in files))))
 
     def walk(self, match):
         '''Generates matching file names.
 
         Equivalent to manifest.matches(match).iterkeys(), but without creating
         an entirely new manifest.
 
         It also reports nonexistent files by marking them bad with match.bad().
         '''
         if match.always():
             for f in iter(self):
                 yield f
             return
 
         fset = set(match.files())
 
         # avoid the entire walk if we're only looking for specific files
         if self._filesfastpath(match):
             for fn in sorted(fset):
                 yield fn
             return
 
         for fn in self:
             if fn in fset:
                 # specified pattern is the exact name
                 fset.remove(fn)
             if match(fn):
                 yield fn
 
         # for dirstate.walk, files=[''] means "walk the whole tree".
         # follow that here, too
         fset.discard('')
 
         for fn in sorted(fset):
             if not self.hasdir(fn):
                 match.bad(fn, None)
 
     def matches(self, match):
         '''generate a new manifest filtered by the match argument'''
         if match.always():
             return self.copy()
 
         if self._filesfastpath(match):
             m = manifestdict()
             lm = self._lm
             for fn in match.files():
                 if fn in lm:
                     m._lm[fn] = lm[fn]
             return m
 
         m = manifestdict()
         m._lm = self._lm.filtercopy(match)
         return m
 
     def diff(self, m2, match=None, clean=False):
         '''Finds changes between the current manifest and m2.
 
         Args:
           m2: the manifest to which this manifest should be compared.
           clean: if true, include files unchanged between these manifests
                  with a None value in the returned dictionary.
 
         The result is returned as a dict with filename as key and
         values of the form ((n1,fl1),(n2,fl2)), where n1/n2 is the
         nodeid in the current/other manifest and fl1/fl2 is the flag
         in the current/other manifest. Where the file does not exist,
         the nodeid will be None and the flags will be the empty
         string.
         '''
         if match:
             m1 = self.matches(match)
             m2 = m2.matches(match)
             return m1.diff(m2, clean=clean)
         return self._lm.diff(m2._lm, clean)
 
     def setflag(self, key, flag):
         self._lm[key] = self[key], flag
 
     def get(self, key, default=None):
         try:
             return self._lm[key][0]
         except KeyError:
             return default
 
     def flags(self, key, default=''):
         try:
             return self._lm[key][1]
         except KeyError:
             return default
 
     def copy(self):
         c = manifestdict()
         c._lm = self._lm.copy()
         return c
 
     def items(self):
         return (x[:2] for x in self._lm.iterentries())
 
     def iteritems(self):
         return (x[:2] for x in self._lm.iterentries())
 
     def iterentries(self):
         return self._lm.iterentries()
 
     def text(self):
         # most likely uses native version
         return self._lm.text()
 
     def fastdelta(self, base, changes):
         """Given a base manifest text as a bytearray and a list of changes
         relative to that text, compute a delta that can be used by revlog.
         """
         delta = []
         dstart = None
         dend = None
         dline = [""]
         start = 0
         # zero copy representation of base as a buffer
         addbuf = util.buffer(base)
 
         changes = list(changes)
         if len(changes) < FASTDELTA_TEXTDIFF_THRESHOLD:
             # start with a readonly loop that finds the offset of
             # each line and creates the deltas
             for f, todelete in changes:
                 # bs will either be the index of the item or the insert point
                 start, end = _msearch(addbuf, f, start)
                 if not todelete:
                     h, fl = self._lm[f]
                     l = "%s\0%s%s\n" % (f, hex(h), fl)
                 else:
                     if start == end:
                         # item we want to delete was not found, error out
                         raise AssertionError(
                                 _("failed to remove %s from manifest") % f)
                     l = ""
                 if dstart is not None and dstart <= start and dend >= start:
                     if dend < end:
                         dend = end
                     if l:
                         dline.append(l)
                 else:
                     if dstart is not None:
                         delta.append([dstart, dend, "".join(dline)])
                     dstart = start
                     dend = end
                     dline = [l]
 
             if dstart is not None:
                 delta.append([dstart, dend, "".join(dline)])
             # apply the delta to the base, and get a delta for addrevision
             deltatext, arraytext = _addlistdelta(base, delta)
         else:
             # For large changes, it's much cheaper to just build the text and
             # diff it.
             arraytext = bytearray(self.text())
             deltatext = mdiff.textdiff(
                 util.buffer(base), util.buffer(arraytext))
 
         return arraytext, deltatext
 
 def _msearch(m, s, lo=0, hi=None):
     '''return a tuple (start, end) that says where to find s within m.
 
     If the string is found m[start:end] are the line containing
     that string.  If start == end the string was not found and
     they indicate the proper sorted insertion point.
 
     m should be a buffer, a memoryview or a byte string.
     s is a byte string'''
     def advance(i, c):
         while i < lenm and m[i:i + 1] != c:
             i += 1
         return i
     if not s:
         return (lo, lo)
     lenm = len(m)
     if not hi:
         hi = lenm
     while lo < hi:
         mid = (lo + hi) // 2
         start = mid
         while start > 0 and m[start - 1:start] != '\n':
             start -= 1
         end = advance(start, '\0')
         if bytes(m[start:end]) < s:
             # we know that after the null there are 40 bytes of sha1
             # this translates to the bisect lo = mid + 1
             lo = advance(end + 40, '\n') + 1
         else:
             # this translates to the bisect hi = mid
             hi = start
     end = advance(lo, '\0')
     found = m[lo:end]
     if s == found:
         # we know that after the null there are 40 bytes of sha1
         end = advance(end + 40, '\n')
         return (lo, end + 1)
     else:
         return (lo, lo)
 
 def _checkforbidden(l):
     """Check filenames for illegal characters."""
     for f in l:
         if '\n' in f or '\r' in f:
             raise error.StorageError(
                 _("'\\n' and '\\r' disallowed in filenames: %r")
                 % pycompat.bytestr(f))
 
 
 # apply the changes collected during the bisect loop to our addlist
 # return a delta suitable for addrevision
 def _addlistdelta(addlist, x):
     # for large addlist arrays, building a new array is cheaper
     # than repeatedly modifying the existing one
     currentposition = 0
     newaddlist = bytearray()
 
     for start, end, content in x:
         newaddlist += addlist[currentposition:start]
         if content:
             newaddlist += bytearray(content)
 
         currentposition = end
 
     newaddlist += addlist[currentposition:]
 
     deltatext = "".join(struct.pack(">lll", start, end, len(content))
                    + content for start, end, content in x)
     return deltatext, newaddlist
 
 def _splittopdir(f):
     if '/' in f:
         dir, subpath = f.split('/', 1)
         return dir + '/', subpath
     else:
         return '', f
 
 _noop = lambda s: None
 
 class treemanifest(object):
     def __init__(self, dir='', text=''):
         self._dir = dir
         self._node = nullid
         self._loadfunc = _noop
         self._copyfunc = _noop
         self._dirty = False
         self._dirs = {}
         self._lazydirs = {}
         # Using _lazymanifest here is a little slower than plain old dicts
         self._files = {}
         self._flags = {}
         if text:
             def readsubtree(subdir, subm):
                 raise AssertionError('treemanifest constructor only accepts '
                                      'flat manifests')
             self.parse(text, readsubtree)
             self._dirty = True # Mark flat manifest dirty after parsing
 
     def _subpath(self, path):
         return self._dir + path
 
     def _loadalllazy(self):
         selfdirs = self._dirs
         for d, (path, node, readsubtree, docopy) in self._lazydirs.iteritems():
             if docopy:
                 selfdirs[d] = readsubtree(path, node).copy()
             else:
                 selfdirs[d] = readsubtree(path, node)
         self._lazydirs = {}
 
     def _loadlazy(self, d):
         v = self._lazydirs.get(d)
         if v:
             path, node, readsubtree, docopy = v
             if docopy:
                 self._dirs[d] = readsubtree(path, node).copy()
             else:
                 self._dirs[d] = readsubtree(path, node)
             del self._lazydirs[d]
 
     def _loadchildrensetlazy(self, visit):
         if not visit:
             return None
         if visit == 'all' or visit == 'this':
             self._loadalllazy()
             return None
 
         loadlazy = self._loadlazy
         for k in visit:
             loadlazy(k + '/')
         return visit
 
     def _loaddifflazy(self, t1, t2):
         """load items in t1 and t2 if they're needed for diffing.
 
         The criteria currently is:
         - if it's not present in _lazydirs in either t1 or t2, load it in the
           other (it may already be loaded or it may not exist, doesn't matter)
         - if it's present in _lazydirs in both, compare the nodeid; if it
           differs, load it in both
         """
         toloadlazy = []
         for d, v1 in t1._lazydirs.iteritems():
             v2 = t2._lazydirs.get(d)
             if not v2 or v2[1] != v1[1]:
                 toloadlazy.append(d)
         for d, v1 in t2._lazydirs.iteritems():
             if d not in t1._lazydirs:
                 toloadlazy.append(d)
 
         for d in toloadlazy:
             t1._loadlazy(d)
             t2._loadlazy(d)
 
     def __len__(self):
         self._load()
         size = len(self._files)
         self._loadalllazy()
         for m in self._dirs.values():
             size += m.__len__()
         return size
 
     def __nonzero__(self):
         # Faster than "__len() != 0" since it avoids loading sub-manifests
         return not self._isempty()
 
     __bool__ = __nonzero__
 
     def _isempty(self):
         self._load() # for consistency; already loaded by all callers
         # See if we can skip loading everything.
         if self._files or (self._dirs and
                            any(not m._isempty() for m in self._dirs.values())):
             return False
         self._loadalllazy()
         return (not self._dirs or
                 all(m._isempty() for m in self._dirs.values()))
 
     def __repr__(self):
         return ('<treemanifest dir=%s, node=%s, loaded=%s, dirty=%s at 0x%x>' %
                 (self._dir, hex(self._node),
                  bool(self._loadfunc is _noop),
                  self._dirty, id(self)))
 
     def dir(self):
         '''The directory that this tree manifest represents, including a
         trailing '/'. Empty string for the repo root directory.'''
         return self._dir
 
     def node(self):
         '''This node of this instance. nullid for unsaved instances. Should
         be updated when the instance is read or written from a revlog.
         '''
         assert not self._dirty
         return self._node
 
     def setnode(self, node):
         self._node = node
         self._dirty = False
 
     def iterentries(self):
         self._load()
         self._loadalllazy()
         for p, n in sorted(itertools.chain(self._dirs.items(),
                                            self._files.items())):
             if p in self._files:
                 yield self._subpath(p), n, self._flags.get(p, '')
             else:
                 for x in n.iterentries():
                     yield x
 
     def items(self):
         self._load()
         self._loadalllazy()
         for p, n in sorted(itertools.chain(self._dirs.items(),
                                            self._files.items())):
             if p in self._files:
                 yield self._subpath(p), n
             else:
                 for f, sn in n.iteritems():
                     yield f, sn
 
     iteritems = items
 
     def iterkeys(self):
         self._load()
         self._loadalllazy()
         for p in sorted(itertools.chain(self._dirs, self._files)):
             if p in self._files:
                 yield self._subpath(p)
             else:
                 for f in self._dirs[p]:
                     yield f
 
     def keys(self):
         return list(self.iterkeys())
 
     def __iter__(self):
         return self.iterkeys()
 
     def __contains__(self, f):
         if f is None:
             return False
         self._load()
         dir, subpath = _splittopdir(f)
         if dir:
             self._loadlazy(dir)
 
             if dir not in self._dirs:
                 return False
 
             return self._dirs[dir].__contains__(subpath)
         else:
             return f in self._files
 
     def get(self, f, default=None):
         self._load()
         dir, subpath = _splittopdir(f)
         if dir:
             self._loadlazy(dir)
 
             if dir not in self._dirs:
                 return default
             return self._dirs[dir].get(subpath, default)
         else:
             return self._files.get(f, default)
 
     def __getitem__(self, f):
         self._load()
         dir, subpath = _splittopdir(f)
         if dir:
             self._loadlazy(dir)
 
             return self._dirs[dir].__getitem__(subpath)
         else:
             return self._files[f]
 
     def flags(self, f):
         self._load()
         dir, subpath = _splittopdir(f)
         if dir:
             self._loadlazy(dir)
 
             if dir not in self._dirs:
                 return ''
             return self._dirs[dir].flags(subpath)
         else:
             if f in self._lazydirs or f in self._dirs:
                 return ''
             return self._flags.get(f, '')
 
     def find(self, f):
         self._load()
         dir, subpath = _splittopdir(f)
         if dir:
             self._loadlazy(dir)
 
             return self._dirs[dir].find(subpath)
         else:
             return self._files[f], self._flags.get(f, '')
 
     def __delitem__(self, f):
         self._load()
         dir, subpath = _splittopdir(f)
         if dir:
             self._loadlazy(dir)
 
             self._dirs[dir].__delitem__(subpath)
             # If the directory is now empty, remove it
             if self._dirs[dir]._isempty():
                 del self._dirs[dir]
         else:
             del self._files[f]
             if f in self._flags:
                 del self._flags[f]
         self._dirty = True
 
     def __setitem__(self, f, n):
         assert n is not None
         self._load()
         dir, subpath = _splittopdir(f)
         if dir:
             self._loadlazy(dir)
             if dir not in self._dirs:
                 self._dirs[dir] = treemanifest(self._subpath(dir))
             self._dirs[dir].__setitem__(subpath, n)
         else:
             self._files[f] = n[:21] # to match manifestdict's behavior
         self._dirty = True
 
     def _load(self):
         if self._loadfunc is not _noop:
             lf, self._loadfunc = self._loadfunc, _noop
             lf(self)
         elif self._copyfunc is not _noop:
             cf, self._copyfunc = self._copyfunc, _noop
             cf(self)
 
     def setflag(self, f, flags):
         """Set the flags (symlink, executable) for path f."""
         self._load()
         dir, subpath = _splittopdir(f)
         if dir:
             self._loadlazy(dir)
             if dir not in self._dirs:
                 self._dirs[dir] = treemanifest(self._subpath(dir))
             self._dirs[dir].setflag(subpath, flags)
         else:
             self._flags[f] = flags
         self._dirty = True
 
     def copy(self):
         copy = treemanifest(self._dir)
         copy._node = self._node
         copy._dirty = self._dirty
         if self._copyfunc is _noop:
             def _copyfunc(s):
                 self._load()
                 s._lazydirs = {d: (p, n, r, True) for
                                d, (p, n, r, c) in self._lazydirs.iteritems()}
                 sdirs = s._dirs
                 for d, v in self._dirs.iteritems():
                     sdirs[d] = v.copy()
                 s._files = dict.copy(self._files)
                 s._flags = dict.copy(self._flags)
             if self._loadfunc is _noop:
                 _copyfunc(copy)
             else:
                 copy._copyfunc = _copyfunc
         else:
             copy._copyfunc = self._copyfunc
         return copy
 
     def filesnotin(self, m2, match=None):
         '''Set of files in this manifest that are not in the other'''
         if match and not match.always():
             m1 = self.matches(match)
             m2 = m2.matches(match)
             return m1.filesnotin(m2)
 
         files = set()
         def _filesnotin(t1, t2):
             if t1._node == t2._node and not t1._dirty and not t2._dirty:
                 return
             t1._load()
             t2._load()
             self._loaddifflazy(t1, t2)
             for d, m1 in t1._dirs.iteritems():
                 if d in t2._dirs:
                     m2 = t2._dirs[d]
                     _filesnotin(m1, m2)
                 else:
                     files.update(m1.iterkeys())
 
             for fn in t1._files:
                 if fn not in t2._files:
                     files.add(t1._subpath(fn))
 
         _filesnotin(self, m2)
         return files
 
     @propertycache
     def _alldirs(self):
         return util.dirs(self)
 
     def dirs(self):
         return self._alldirs
 
     def hasdir(self, dir):
         self._load()
         topdir, subdir = _splittopdir(dir)
         if topdir:
             self._loadlazy(topdir)
             if topdir in self._dirs:
                 return self._dirs[topdir].hasdir(subdir)
             return False
         dirslash = dir + '/'
         return dirslash in self._dirs or dirslash in self._lazydirs
 
     def walk(self, match):
         '''Generates matching file names.
 
         Equivalent to manifest.matches(match).iterkeys(), but without creating
         an entirely new manifest.
 
         It also reports nonexistent files by marking them bad with match.bad().
         '''
         if match.always():
             for f in iter(self):
                 yield f
             return
 
         fset = set(match.files())
 
         for fn in self._walk(match):
             if fn in fset:
                 # specified pattern is the exact name
                 fset.remove(fn)
             yield fn
 
         # for dirstate.walk, files=[''] means "walk the whole tree".
         # follow that here, too
         fset.discard('')
 
         for fn in sorted(fset):
             if not self.hasdir(fn):
                 match.bad(fn, None)
 
     def _walk(self, match):
         '''Recursively generates matching file names for walk().'''
         visit = match.visitchildrenset(self._dir[:-1])
         if not visit:
             return
 
         # yield this dir's files and walk its submanifests
         self._load()
         visit = self._loadchildrensetlazy(visit)
         for p in sorted(list(self._dirs) + list(self._files)):
             if p in self._files:
                 fullp = self._subpath(p)
                 if match(fullp):
                     yield fullp
             else:
                 if not visit or p[:-1] in visit:
                     for f in self._dirs[p]._walk(match):
                         yield f
 
     def matches(self, match):
         '''generate a new manifest filtered by the match argument'''
         if match.always():
             return self.copy()
 
         return self._matches(match)
 
     def _matches(self, match):
         '''recursively generate a new manifest filtered by the match argument.
         '''
 
         visit = match.visitchildrenset(self._dir[:-1])
         if visit == 'all':
             return self.copy()
         ret = treemanifest(self._dir)
         if not visit:
             return ret
 
         self._load()
         for fn in self._files:
             # While visitchildrenset *usually* lists only subdirs, this is
             # actually up to the matcher and may have some files in the set().
             # If visit == 'this', we should obviously look at the files in this
             # directory; if visit is a set, and fn is in it, we should inspect
             # fn (but no need to inspect things not in the set).
             if visit != 'this' and fn not in visit:
                 continue
             fullp = self._subpath(fn)
             # visitchildrenset isn't perfect, we still need to call the regular
             # matcher code to further filter results.
             if not match(fullp):
                 continue
             ret._files[fn] = self._files[fn]
             if fn in self._flags:
                 ret._flags[fn] = self._flags[fn]
 
         visit = self._loadchildrensetlazy(visit)
         for dir, subm in self._dirs.iteritems():
             if visit and dir[:-1] not in visit:
                 continue
             m = subm._matches(match)
             if not m._isempty():
                 ret._dirs[dir] = m
 
         if not ret._isempty():
             ret._dirty = True
         return ret
 
     def diff(self, m2, match=None, clean=False):
         '''Finds changes between the current manifest and m2.
 
         Args:
           m2: the manifest to which this manifest should be compared.
           clean: if true, include files unchanged between these manifests
                  with a None value in the returned dictionary.
 
         The result is returned as a dict with filename as key and
         values of the form ((n1,fl1),(n2,fl2)), where n1/n2 is the
         nodeid in the current/other manifest and fl1/fl2 is the flag
         in the current/other manifest. Where the file does not exist,
         the nodeid will be None and the flags will be the empty
         string.
         '''
         if match and not match.always():
             m1 = self.matches(match)
             m2 = m2.matches(match)
             return m1.diff(m2, clean=clean)
         result = {}
         emptytree = treemanifest()
 
         def _iterativediff(t1, t2, stack):
             """compares two tree manifests and append new tree-manifests which
             needs to be compared to stack"""
             if t1._node == t2._node and not t1._dirty and not t2._dirty:
                 return
             t1._load()
             t2._load()
             self._loaddifflazy(t1, t2)
 
             for d, m1 in t1._dirs.iteritems():
                 m2 = t2._dirs.get(d, emptytree)
                 stack.append((m1, m2))
 
             for d, m2 in t2._dirs.iteritems():
                 if d not in t1._dirs:
                     stack.append((emptytree, m2))
 
             for fn, n1 in t1._files.iteritems():
                 fl1 = t1._flags.get(fn, '')
                 n2 = t2._files.get(fn, None)
                 fl2 = t2._flags.get(fn, '')
                 if n1 != n2 or fl1 != fl2:
                     result[t1._subpath(fn)] = ((n1, fl1), (n2, fl2))
                 elif clean:
                     result[t1._subpath(fn)] = None
 
             for fn, n2 in t2._files.iteritems():
                 if fn not in t1._files:
                     fl2 = t2._flags.get(fn, '')
                     result[t2._subpath(fn)] = ((None, ''), (n2, fl2))
 
         stackls = []
         _iterativediff(self, m2, stackls)
         while stackls:
             t1, t2 = stackls.pop()
             # stackls is populated in the function call
             _iterativediff(t1, t2, stackls)
         return result
 
     def unmodifiedsince(self, m2):
         return not self._dirty and not m2._dirty and self._node == m2._node
 
     def parse(self, text, readsubtree):
         selflazy = self._lazydirs
         subpath = self._subpath
         for f, n, fl in _parse(text):
             if fl == 't':
                 f = f + '/'
                 # False below means "doesn't need to be copied" and can use the
                 # cached value from readsubtree directly.
                 selflazy[f] = (subpath(f), n, readsubtree, False)
             elif '/' in f:
                 # This is a flat manifest, so use __setitem__ and setflag rather
                 # than assigning directly to _files and _flags, so we can
                 # assign a path in a subdirectory, and to mark dirty (compared
                 # to nullid).
                 self[f] = n
                 if fl:
                     self.setflag(f, fl)
             else:
                 # Assigning to _files and _flags avoids marking as dirty,
                 # and should be a little faster.
                 self._files[f] = n
                 if fl:
                     self._flags[f] = fl
 
     def text(self):
         """Get the full data of this manifest as a bytestring."""
         self._load()
         return _text(self.iterentries())
 
     def dirtext(self):
         """Get the full data of this directory as a bytestring. Make sure that
         any submanifests have been written first, so their nodeids are correct.
         """
         self._load()
         flags = self.flags
         lazydirs = [(d[:-1], v[1], 't') for d, v in self._lazydirs.iteritems()]
         dirs = [(d[:-1], self._dirs[d]._node, 't') for d in self._dirs]
         files = [(f, self._files[f], flags(f)) for f in self._files]
         return _text(sorted(dirs + files + lazydirs))
 
     def read(self, gettext, readsubtree):
         def _load_for_read(s):
             s.parse(gettext(), readsubtree)
             s._dirty = False
         self._loadfunc = _load_for_read
 
     def writesubtrees(self, m1, m2, writesubtree, match):
         self._load() # for consistency; should never have any effect here
         m1._load()
         m2._load()
         emptytree = treemanifest()
         def getnode(m, d):
             ld = m._lazydirs.get(d)
             if ld:
                 return ld[1]
             return m._dirs.get(d, emptytree)._node
 
         # let's skip investigating things that `match` says we do not need.
         visit = match.visitchildrenset(self._dir[:-1])
         visit = self._loadchildrensetlazy(visit)
         if visit == 'this' or visit == 'all':
             visit = None
         for d, subm in self._dirs.iteritems():
             if visit and d[:-1] not in visit:
                 continue
             subp1 = getnode(m1, d)
             subp2 = getnode(m2, d)
             if subp1 == nullid:
                 subp1, subp2 = subp2, subp1
             writesubtree(subm, subp1, subp2, match)
 
     def walksubtrees(self, matcher=None):
         """Returns an iterator of the subtrees of this manifest, including this
         manifest itself.
 
         If `matcher` is provided, it only returns subtrees that match.
         """
         if matcher and not matcher.visitdir(self._dir[:-1]):
             return
         if not matcher or matcher(self._dir[:-1]):
             yield self
 
         self._load()
         # OPT: use visitchildrenset to avoid loading everything.
         self._loadalllazy()
         for d, subm in self._dirs.iteritems():
             for subtree in subm.walksubtrees(matcher=matcher):
                 yield subtree
 
 class manifestfulltextcache(util.lrucachedict):
     """File-backed LRU cache for the manifest cache
 
     File consists of entries, up to EOF:
 
     - 20 bytes node, 4 bytes length, <length> manifest data
 
     These are written in reverse cache order (oldest to newest).
 
     """
 
     _file = 'manifestfulltextcache'
 
     def __init__(self, max):
         super(manifestfulltextcache, self).__init__(max)
         self._dirty = False
         self._read = False
         self._opener = None
 
     def read(self):
         if self._read or self._opener is None:
             return
 
         try:
             with self._opener(self._file) as fp:
                 set = super(manifestfulltextcache, self).__setitem__
                 # ignore trailing data, this is a cache, corruption is skipped
                 while True:
                     node = fp.read(20)
                     if len(node) < 20:
                         break
                     try:
                         size = struct.unpack('>L', fp.read(4))[0]
                     except struct.error:
                         break
                     value = bytearray(fp.read(size))
                     if len(value) != size:
                         break
                     set(node, value)
         except IOError:
             # the file is allowed to be missing
             pass
 
         self._read = True
         self._dirty = False
 
     def write(self):
         if not self._dirty or self._opener is None:
             return
         # rotate backwards to the first used node
         with self._opener(self._file, 'w', atomictemp=True, checkambig=True
             ) as fp:
             node = self._head.prev
             while True:
                 if node.key in self._cache:
                     fp.write(node.key)
                     fp.write(struct.pack('>L', len(node.value)))
                     fp.write(node.value)
                 if node is self._head:
                     break
                 node = node.prev
 
     def __len__(self):
         if not self._read:
             self.read()
         return super(manifestfulltextcache, self).__len__()
 
     def __contains__(self, k):
         if not self._read:
             self.read()
         return super(manifestfulltextcache, self).__contains__(k)
 
     def __iter__(self):
         if not self._read:
             self.read()
         return super(manifestfulltextcache, self).__iter__()
 
     def __getitem__(self, k):
         if not self._read:
             self.read()
         # the cache lru order can change on read
         setdirty = self._cache.get(k) is not self._head
         value = super(manifestfulltextcache, self).__getitem__(k)
         if setdirty:
             self._dirty = True
         return value
 
     def __setitem__(self, k, v):
         if not self._read:
             self.read()
         super(manifestfulltextcache, self).__setitem__(k, v)
         self._dirty = True
 
     def __delitem__(self, k):
         if not self._read:
             self.read()
         super(manifestfulltextcache, self).__delitem__(k)
         self._dirty = True
 
     def get(self, k, default=None):
         if not self._read:
             self.read()
         return super(manifestfulltextcache, self).get(k, default=default)
 
     def clear(self, clear_persisted_data=False):
         super(manifestfulltextcache, self).clear()
         if clear_persisted_data:
             self._dirty = True
             self.write()
         self._read = False
 
+# and upper bound of what we expect from compression
+# (real live value seems to be "3")
+MAXCOMPRESSION = 10
+
 @interfaceutil.implementer(repository.imanifeststorage)
 class manifestrevlog(object):
     '''A revlog that stores manifest texts. This is responsible for caching the
     full-text manifest contents.
     '''
     def __init__(self, opener, tree='', dirlogcache=None, indexfile=None,
                  treemanifest=False):
         """Constructs a new manifest revlog
 
         `indexfile` - used by extensions to have two manifests at once, like
         when transitioning between flatmanifeset and treemanifests.
 
         `treemanifest` - used to indicate this is a tree manifest revlog. Opener
         options can also be used to make this a tree manifest revlog. The opener
         option takes precedence, so if it is set to True, we ignore whatever
         value is passed in to the constructor.
         """
         # During normal operations, we expect to deal with not more than four
         # revs at a time (such as during commit --amend). When rebasing large
         # stacks of commits, the number can go up, hence the config knob below.
         cachesize = 4
         optiontreemanifest = False
         opts = getattr(opener, 'options', None)
         if opts is not None:
             cachesize = opts.get('manifestcachesize', cachesize)
             optiontreemanifest = opts.get('treemanifest', False)
 
         self._treeondisk = optiontreemanifest or treemanifest
 
         self._fulltextcache = manifestfulltextcache(cachesize)
 
         if tree:
             assert self._treeondisk, 'opts is %r' % opts
 
         if indexfile is None:
             indexfile = '00manifest.i'
             if tree:
                 indexfile = "meta/" + tree + indexfile
 
         self.tree = tree
 
         # The dirlogcache is kept on the root manifest log
         if tree:
             self._dirlogcache = dirlogcache
         else:
             self._dirlogcache = {'': self}
 
         self._revlog = revlog.revlog(opener, indexfile,
                                      # only root indexfile is cached
                                      checkambig=not bool(tree),
-                                     mmaplargeindex=True)
+                                     mmaplargeindex=True,
+                                     upperboundcomp=MAXCOMPRESSION)
 
         self.index = self._revlog.index
         self.version = self._revlog.version
         self._generaldelta = self._revlog._generaldelta
 
     def _setupmanifestcachehooks(self, repo):
         """Persist the manifestfulltextcache on lock release"""
         if not util.safehasattr(repo, '_wlockref'):
             return
 
         self._fulltextcache._opener = repo.wcachevfs
         if repo._currentlock(repo._wlockref) is None:
             return
 
         reporef = weakref.ref(repo)
         manifestrevlogref = weakref.ref(self)
 
         def persistmanifestcache():
             repo = reporef()
             self = manifestrevlogref()
             if repo is None or self is None:
                 return
             if repo.manifestlog.getstorage(b'') is not self:
                 # there's a different manifest in play now, abort
                 return
             self._fulltextcache.write()
 
         repo._afterlock(persistmanifestcache)
 
     @property
     def fulltextcache(self):
         return self._fulltextcache
 
     def clearcaches(self, clear_persisted_data=False):
         self._revlog.clearcaches()
         self._fulltextcache.clear(clear_persisted_data=clear_persisted_data)
         self._dirlogcache = {self.tree: self}
 
     def dirlog(self, d):
         if d:
             assert self._treeondisk
         if d not in self._dirlogcache:
             mfrevlog = manifestrevlog(self.opener, d,
                                       self._dirlogcache,
                                       treemanifest=self._treeondisk)
             self._dirlogcache[d] = mfrevlog
         return self._dirlogcache[d]
 
     def add(self, m, transaction, link, p1, p2, added, removed, readtree=None,
             match=None):
         if p1 in self.fulltextcache and util.safehasattr(m, 'fastdelta'):
             # If our first parent is in the manifest cache, we can
             # compute a delta here using properties we know about the
             # manifest up-front, which may save time later for the
             # revlog layer.
 
             _checkforbidden(added)
             # combine the changed lists into one sorted iterator
             work = heapq.merge([(x, False) for x in sorted(added)],
                                [(x, True) for x in sorted(removed)])
 
             arraytext, deltatext = m.fastdelta(self.fulltextcache[p1], work)
             cachedelta = self._revlog.rev(p1), deltatext
             text = util.buffer(arraytext)
             n = self._revlog.addrevision(text, transaction, link, p1, p2,
                                          cachedelta)
         else:
             # The first parent manifest isn't already loaded, so we'll
             # just encode a fulltext of the manifest and pass that
             # through to the revlog layer, and let it handle the delta
             # process.
             if self._treeondisk:
                 assert readtree, "readtree must be set for treemanifest writes"
                 assert match, "match must be specified for treemanifest writes"
                 m1 = readtree(self.tree, p1)
                 m2 = readtree(self.tree, p2)
                 n = self._addtree(m, transaction, link, m1, m2, readtree,
                                   match=match)
                 arraytext = None
             else:
                 text = m.text()
                 n = self._revlog.addrevision(text, transaction, link, p1, p2)
                 arraytext = bytearray(text)
 
         if arraytext is not None:
             self.fulltextcache[n] = arraytext
 
         return n
 
     def _addtree(self, m, transaction, link, m1, m2, readtree, match):
         # If the manifest is unchanged compared to one parent,
         # don't write a new revision
         if self.tree != '' and (m.unmodifiedsince(m1) or m.unmodifiedsince(
             m2)):
             return m.node()
         def writesubtree(subm, subp1, subp2, match):
             sublog = self.dirlog(subm.dir())
             sublog.add(subm, transaction, link, subp1, subp2, None, None,
                        readtree=readtree, match=match)
         m.writesubtrees(m1, m2, writesubtree, match)
         text = m.dirtext()
         n = None
         if self.tree != '':
             # Double-check whether contents are unchanged to one parent
             if text == m1.dirtext():
                 n = m1.node()
             elif text == m2.dirtext():
                 n = m2.node()
 
         if not n:
             n = self._revlog.addrevision(text, transaction, link, m1.node(),
                                          m2.node())
 
         # Save nodeid so parent manifest can calculate its nodeid
         m.setnode(n)
         return n
 
     def __len__(self):
         return len(self._revlog)
 
     def __iter__(self):
         return self._revlog.__iter__()
 
     def rev(self, node):
         return self._revlog.rev(node)
 
     def node(self, rev):
         return self._revlog.node(rev)
 
     def lookup(self, value):
         return self._revlog.lookup(value)
 
     def parentrevs(self, rev):
         return self._revlog.parentrevs(rev)
 
     def parents(self, node):
         return self._revlog.parents(node)
 
     def linkrev(self, rev):
         return self._revlog.linkrev(rev)
 
     def checksize(self):
         return self._revlog.checksize()
 
     def revision(self, node, _df=None, raw=False):
         return self._revlog.revision(node, _df=_df, raw=raw)
 
     def revdiff(self, rev1, rev2):
         return self._revlog.revdiff(rev1, rev2)
 
     def cmp(self, node, text):
         return self._revlog.cmp(node, text)
 
     def deltaparent(self, rev):
         return self._revlog.deltaparent(rev)
 
     def emitrevisions(self, nodes, nodesorder=None,
                       revisiondata=False, assumehaveparentrevisions=False,
                       deltamode=repository.CG_DELTAMODE_STD):
         return self._revlog.emitrevisions(
             nodes, nodesorder=nodesorder, revisiondata=revisiondata,
             assumehaveparentrevisions=assumehaveparentrevisions,
             deltamode=deltamode)
 
     def addgroup(self, deltas, linkmapper, transaction, addrevisioncb=None):
         return self._revlog.addgroup(deltas, linkmapper, transaction,
                                      addrevisioncb=addrevisioncb)
 
     def rawsize(self, rev):
         return self._revlog.rawsize(rev)
 
     def getstrippoint(self, minlink):
         return self._revlog.getstrippoint(minlink)
 
     def strip(self, minlink, transaction):
         return self._revlog.strip(minlink, transaction)
 
     def files(self):
         return self._revlog.files()
 
     def clone(self, tr, destrevlog, **kwargs):
         if not isinstance(destrevlog, manifestrevlog):
             raise error.ProgrammingError('expected manifestrevlog to clone()')
 
         return self._revlog.clone(tr, destrevlog._revlog, **kwargs)
 
     def storageinfo(self, exclusivefiles=False, sharedfiles=False,
                     revisionscount=False, trackedsize=False,
                     storedsize=False):
         return self._revlog.storageinfo(
             exclusivefiles=exclusivefiles, sharedfiles=sharedfiles,
             revisionscount=revisionscount, trackedsize=trackedsize,
             storedsize=storedsize)
 
     @property
     def indexfile(self):
         return self._revlog.indexfile
 
     @indexfile.setter
     def indexfile(self, value):
         self._revlog.indexfile = value
 
     @property
     def opener(self):
         return self._revlog.opener
 
     @opener.setter
     def opener(self, value):
         self._revlog.opener = value
 
 @interfaceutil.implementer(repository.imanifestlog)
 class manifestlog(object):
     """A collection class representing the collection of manifest snapshots
     referenced by commits in the repository.
 
     In this situation, 'manifest' refers to the abstract concept of a snapshot
     of the list of files in the given commit. Consumers of the output of this
     class do not care about the implementation details of the actual manifests
     they receive (i.e. tree or flat or lazily loaded, etc)."""
     def __init__(self, opener, repo, rootstore, narrowmatch):
         usetreemanifest = False
         cachesize = 4
 
         opts = getattr(opener, 'options', None)
         if opts is not None:
             usetreemanifest = opts.get('treemanifest', usetreemanifest)
             cachesize = opts.get('manifestcachesize', cachesize)
 
         self._treemanifests = usetreemanifest
 
         self._rootstore = rootstore
         self._rootstore._setupmanifestcachehooks(repo)
         self._narrowmatch = narrowmatch
 
         # A cache of the manifestctx or treemanifestctx for each directory
         self._dirmancache = {}
         self._dirmancache[''] = util.lrucachedict(cachesize)
 
         self._cachesize = cachesize
 
     def __getitem__(self, node):
         """Retrieves the manifest instance for the given node. Throws a
         LookupError if not found.
         """
         return self.get('', node)
 
     def get(self, tree, node, verify=True):
         """Retrieves the manifest instance for the given node. Throws a
         LookupError if not found.
 
         `verify` - if True an exception will be thrown if the node is not in
                    the revlog
         """
         if node in self._dirmancache.get(tree, ()):
             return self._dirmancache[tree][node]
 
         if not self._narrowmatch.always():
             if not self._narrowmatch.visitdir(tree[:-1]):
                 return excludeddirmanifestctx(tree, node)
         if tree:
             if self._rootstore._treeondisk:
                 if verify:
                     # Side-effect is LookupError is raised if node doesn't
                     # exist.
                     self.getstorage(tree).rev(node)
 
                 m = treemanifestctx(self, tree, node)
             else:
                 raise error.Abort(
                         _("cannot ask for manifest directory '%s' in a flat "
                           "manifest") % tree)
         else:
             if verify:
                 # Side-effect is LookupError is raised if node doesn't exist.
                 self._rootstore.rev(node)
 
             if self._treemanifests:
                 m = treemanifestctx(self, '', node)
             else:
                 m = manifestctx(self, node)
 
         if node != nullid:
             mancache = self._dirmancache.get(tree)
             if not mancache:
                 mancache = util.lrucachedict(self._cachesize)
                 self._dirmancache[tree] = mancache
             mancache[node] = m
         return m
 
     def getstorage(self, tree):
         return self._rootstore.dirlog(tree)
 
     def clearcaches(self, clear_persisted_data=False):
         self._dirmancache.clear()
         self._rootstore.clearcaches(clear_persisted_data=clear_persisted_data)
 
     def rev(self, node):
         return self._rootstore.rev(node)
 
 @interfaceutil.implementer(repository.imanifestrevisionwritable)
 class memmanifestctx(object):
     def __init__(self, manifestlog):
         self._manifestlog = manifestlog
         self._manifestdict = manifestdict()
 
     def _storage(self):
         return self._manifestlog.getstorage(b'')
 
     def new(self):
         return memmanifestctx(self._manifestlog)
 
     def copy(self):
         memmf = memmanifestctx(self._manifestlog)
         memmf._manifestdict = self.read().copy()
         return memmf
 
     def read(self):
         return self._manifestdict
 
     def write(self, transaction, link, p1, p2, added, removed, match=None):
         return self._storage().add(self._manifestdict, transaction, link,
                                    p1, p2, added, removed, match=match)
 
 @interfaceutil.implementer(repository.imanifestrevisionstored)
 class manifestctx(object):
     """A class representing a single revision of a manifest, including its
     contents, its parent revs, and its linkrev.
     """
     def __init__(self, manifestlog, node):
         self._manifestlog = manifestlog
         self._data = None
 
         self._node = node
 
         # TODO: We eventually want p1, p2, and linkrev exposed on this class,
         # but let's add it later when something needs it and we can load it
         # lazily.
         #self.p1, self.p2 = store.parents(node)
         #rev = store.rev(node)
         #self.linkrev = store.linkrev(rev)
 
     def _storage(self):
         return self._manifestlog.getstorage(b'')
 
     def node(self):
         return self._node
 
     def new(self):
         return memmanifestctx(self._manifestlog)
 
     def copy(self):
         memmf = memmanifestctx(self._manifestlog)
         memmf._manifestdict = self.read().copy()
         return memmf
 
     @propertycache
     def parents(self):
         return self._storage().parents(self._node)
 
     def read(self):
         if self._data is None:
             if self._node == nullid:
                 self._data = manifestdict()
             else:
                 store = self._storage()
                 if self._node in store.fulltextcache:
                     text = pycompat.bytestr(store.fulltextcache[self._node])
                 else:
                     text = store.revision(self._node)
                     arraytext = bytearray(text)
                     store.fulltextcache[self._node] = arraytext
                 self._data = manifestdict(text)
         return self._data
 
     def readfast(self, shallow=False):
         '''Calls either readdelta or read, based on which would be less work.
         readdelta is called if the delta is against the p1, and therefore can be
         read quickly.
 
         If `shallow` is True, nothing changes since this is a flat manifest.
         '''
         store = self._storage()
         r = store.rev(self._node)
         deltaparent = store.deltaparent(r)
         if deltaparent != nullrev and deltaparent in store.parentrevs(r):
             return self.readdelta()
         return self.read()
 
     def readdelta(self, shallow=False):
         '''Returns a manifest containing just the entries that are present
         in this manifest, but not in its p1 manifest. This is efficient to read
         if the revlog delta is already p1.
 
         Changing the value of `shallow` has no effect on flat manifests.
         '''
         store = self._storage()
         r = store.rev(self._node)
         d = mdiff.patchtext(store.revdiff(store.deltaparent(r), r))
         return manifestdict(d)
 
     def find(self, key):
         return self.read().find(key)
 
 @interfaceutil.implementer(repository.imanifestrevisionwritable)
 class memtreemanifestctx(object):
     def __init__(self, manifestlog, dir=''):
         self._manifestlog = manifestlog
         self._dir = dir
         self._treemanifest = treemanifest()
 
     def _storage(self):
         return self._manifestlog.getstorage(b'')
 
     def new(self, dir=''):
         return memtreemanifestctx(self._manifestlog, dir=dir)
 
     def copy(self):
         memmf = memtreemanifestctx(self._manifestlog, dir=self._dir)
         memmf._treemanifest = self._treemanifest.copy()
         return memmf
 
     def read(self):
         return self._treemanifest
 
     def write(self, transaction, link, p1, p2, added, removed, match=None):
         def readtree(dir, node):
             return self._manifestlog.get(dir, node).read()
         return self._storage().add(self._treemanifest, transaction, link,
                                    p1, p2, added, removed, readtree=readtree,
                                    match=match)
 
 @interfaceutil.implementer(repository.imanifestrevisionstored)
 class treemanifestctx(object):
     def __init__(self, manifestlog, dir, node):
         self._manifestlog = manifestlog
         self._dir = dir
         self._data = None
 
         self._node = node
 
         # TODO: Load p1/p2/linkrev lazily. They need to be lazily loaded so that
         # we can instantiate treemanifestctx objects for directories we don't
         # have on disk.
         #self.p1, self.p2 = store.parents(node)
         #rev = store.rev(node)
         #self.linkrev = store.linkrev(rev)
 
     def _storage(self):
         narrowmatch = self._manifestlog._narrowmatch
         if not narrowmatch.always():
             if not narrowmatch.visitdir(self._dir[:-1]):
                 return excludedmanifestrevlog(self._dir)
         return self._manifestlog.getstorage(self._dir)
 
     def read(self):
         if self._data is None:
             store = self._storage()
             if self._node == nullid:
                 self._data = treemanifest()
             # TODO accessing non-public API
             elif store._treeondisk:
                 m = treemanifest(dir=self._dir)
                 def gettext():
                     return store.revision(self._node)
                 def readsubtree(dir, subm):
                     # Set verify to False since we need to be able to create
                     # subtrees for trees that don't exist on disk.
                     return self._manifestlog.get(dir, subm, verify=False).read()
                 m.read(gettext, readsubtree)
                 m.setnode(self._node)
                 self._data = m
             else:
                 if self._node in store.fulltextcache:
                     text = pycompat.bytestr(store.fulltextcache[self._node])
                 else:
                     text = store.revision(self._node)
                     arraytext = bytearray(text)
                     store.fulltextcache[self._node] = arraytext
                 self._data = treemanifest(dir=self._dir, text=text)
 
         return self._data
 
     def node(self):
         return self._node
 
     def new(self, dir=''):
         return memtreemanifestctx(self._manifestlog, dir=dir)
 
     def copy(self):
         memmf = memtreemanifestctx(self._manifestlog, dir=self._dir)
         memmf._treemanifest = self.read().copy()
         return memmf
 
     @propertycache
     def parents(self):
         return self._storage().parents(self._node)
 
     def readdelta(self, shallow=False):
         '''Returns a manifest containing just the entries that are present
         in this manifest, but not in its p1 manifest. This is efficient to read
         if the revlog delta is already p1.
 
         If `shallow` is True, this will read the delta for this directory,
         without recursively reading subdirectory manifests. Instead, any
         subdirectory entry will be reported as it appears in the manifest, i.e.
         the subdirectory will be reported among files and distinguished only by
         its 't' flag.
         '''
         store = self._storage()
         if shallow:
             r = store.rev(self._node)
             d = mdiff.patchtext(store.revdiff(store.deltaparent(r), r))
             return manifestdict(d)
         else:
             # Need to perform a slow delta
             r0 = store.deltaparent(store.rev(self._node))
             m0 = self._manifestlog.get(self._dir, store.node(r0)).read()
             m1 = self.read()
             md = treemanifest(dir=self._dir)
             for f, ((n0, fl0), (n1, fl1)) in m0.diff(m1).iteritems():
                 if n1:
                     md[f] = n1
                     if fl1:
                         md.setflag(f, fl1)
             return md
 
     def readfast(self, shallow=False):
         '''Calls either readdelta or read, based on which would be less work.
         readdelta is called if the delta is against the p1, and therefore can be
         read quickly.
 
         If `shallow` is True, it only returns the entries from this manifest,
         and not any submanifests.
         '''
         store = self._storage()
         r = store.rev(self._node)
         deltaparent = store.deltaparent(r)
         if (deltaparent != nullrev and
             deltaparent in store.parentrevs(r)):
             return self.readdelta(shallow=shallow)
 
         if shallow:
             return manifestdict(store.revision(self._node))
         else:
             return self.read()
 
     def find(self, key):
         return self.read().find(key)
 
 class excludeddir(treemanifest):
     """Stand-in for a directory that is excluded from the repository.
 
     With narrowing active on a repository that uses treemanifests,
     some of the directory revlogs will be excluded from the resulting
     clone. This is a huge storage win for clients, but means we need
     some sort of pseudo-manifest to surface to internals so we can
     detect a merge conflict outside the narrowspec. That's what this
     class is: it stands in for a directory whose node is known, but
     whose contents are unknown.
     """
     def __init__(self, dir, node):
         super(excludeddir, self).__init__(dir)
         self._node = node
         # Add an empty file, which will be included by iterators and such,
         # appearing as the directory itself (i.e. something like "dir/")
         self._files[''] = node
         self._flags[''] = 't'
 
     # Manifests outside the narrowspec should never be modified, so avoid
     # copying. This makes a noticeable difference when there are very many
     # directories outside the narrowspec. Also, it makes sense for the copy to
     # be of the same type as the original, which would not happen with the
     # super type's copy().
     def copy(self):
         return self
 
 class excludeddirmanifestctx(treemanifestctx):
     """context wrapper for excludeddir - see that docstring for rationale"""
     def __init__(self, dir, node):
         self._dir = dir
         self._node = node
 
     def read(self):
         return excludeddir(self._dir, self._node)
 
     def write(self, *args):
         raise error.ProgrammingError(
             'attempt to write manifest from excluded dir %s' % self._dir)
 
 class excludedmanifestrevlog(manifestrevlog):
     """Stand-in for excluded treemanifest revlogs.
 
     When narrowing is active on a treemanifest repository, we'll have
     references to directories we can't see due to the revlog being
     skipped. This class exists to conform to the manifestrevlog
     interface for those directories and proactively prevent writes to
     outside the narrowspec.
     """
 
     def __init__(self, dir):
         self._dir = dir
 
     def __len__(self):
         raise error.ProgrammingError(
             'attempt to get length of excluded dir %s' % self._dir)
 
     def rev(self, node):
         raise error.ProgrammingError(
             'attempt to get rev from excluded dir %s' % self._dir)
 
     def linkrev(self, node):
         raise error.ProgrammingError(
             'attempt to get linkrev from excluded dir %s' % self._dir)
 
     def node(self, rev):
         raise error.ProgrammingError(
             'attempt to get node from excluded dir %s' % self._dir)
 
     def add(self, *args, **kwargs):
         # We should never write entries in dirlogs outside the narrow clone.
         # However, the method still gets called from writesubtree() in
         # _addtree(), so we need to handle it. We should possibly make that
         # avoid calling add() with a clean manifest (_dirty is always False
         # in excludeddir instances).
         pass

mercurial/revlog.py

@@ -1,2684 +1,2690 b''
 # revlog.py - storage back-end for mercurial
 #
 # Copyright 2005-2007 Matt Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 """Storage back-end for Mercurial.
 
 This provides efficient delta storage with O(1) retrieve and append
 and O(changes) merge between branches.
 """
 
 from __future__ import absolute_import
 
 import collections
 import contextlib
 import errno
 import os
 import struct
 import zlib
 
 # import stuff from node for others to import from revlog
 from .node import (
     bin,
     hex,
     nullhex,
     nullid,
     nullrev,
     short,
     wdirfilenodeids,
     wdirhex,
     wdirid,
     wdirrev,
 )
 from .i18n import _
 from .revlogutils.constants import (
     FLAG_GENERALDELTA,
     FLAG_INLINE_DATA,
     REVIDX_DEFAULT_FLAGS,
     REVIDX_ELLIPSIS,
     REVIDX_EXTSTORED,
     REVIDX_FLAGS_ORDER,
     REVIDX_ISCENSORED,
     REVIDX_KNOWN_FLAGS,
     REVIDX_RAWTEXT_CHANGING_FLAGS,
     REVLOGV0,
     REVLOGV1,
     REVLOGV1_FLAGS,
     REVLOGV2,
     REVLOGV2_FLAGS,
     REVLOG_DEFAULT_FLAGS,
     REVLOG_DEFAULT_FORMAT,
     REVLOG_DEFAULT_VERSION,
 )
 from .thirdparty import (
     attr,
 )
 from . import (
     ancestor,
     dagop,
     error,
     mdiff,
     policy,
     pycompat,
     repository,
     templatefilters,
     util,
 )
 from .revlogutils import (
     deltas as deltautil,
 )
 from .utils import (
     interfaceutil,
     storageutil,
     stringutil,
 )
 
 # blanked usage of all the name to prevent pyflakes constraints
 # We need these name available in the module for extensions.
 REVLOGV0
 REVLOGV1
 REVLOGV2
 FLAG_INLINE_DATA
 FLAG_GENERALDELTA
 REVLOG_DEFAULT_FLAGS
 REVLOG_DEFAULT_FORMAT
 REVLOG_DEFAULT_VERSION
 REVLOGV1_FLAGS
 REVLOGV2_FLAGS
 REVIDX_ISCENSORED
 REVIDX_ELLIPSIS
 REVIDX_EXTSTORED
 REVIDX_DEFAULT_FLAGS
 REVIDX_FLAGS_ORDER
 REVIDX_KNOWN_FLAGS
 REVIDX_RAWTEXT_CHANGING_FLAGS
 
 parsers = policy.importmod(r'parsers')
 rustancestor = policy.importrust(r'ancestor')
 rustdagop = policy.importrust(r'dagop')
 
 # Aliased for performance.
 _zlibdecompress = zlib.decompress
 
 # max size of revlog with inline data
 _maxinline = 131072
 _chunksize = 1048576
 
 # Store flag processors (cf. 'addflagprocessor()' to register)
 _flagprocessors = {
     REVIDX_ISCENSORED: None,
 }
 
 # 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 addflagprocessor(flag, processor):
     """Register a flag processor on a revision data flag.
 
     Invariant:
     - Flags need to be defined in REVIDX_KNOWN_FLAGS and REVIDX_FLAGS_ORDER,
       and REVIDX_RAWTEXT_CHANGING_FLAGS if they can alter rawtext.
     - Only one flag processor can be registered on a specific flag.
     - flagprocessors must be 3-tuples of functions (read, write, raw) with the
       following signatures:
           - (read)  f(self, rawtext) -> text, bool
           - (write) f(self, text) -> rawtext, bool
           - (raw)   f(self, rawtext) -> bool
       "text" is presented to the user. "rawtext" is stored in revlog data, not
       directly visible to the user.
       The boolean returned by these transforms is used to determine whether
       the returned text can be used for hash integrity checking. For example,
       if "write" returns False, then "text" is used to generate hash. If
       "write" returns True, that basically means "rawtext" returned by "write"
       should be used to generate hash. Usually, "write" and "read" return
       different booleans. And "raw" returns a same boolean as "write".
 
       Note: The 'raw' transform is used for changegroup generation and in some
       debug commands. In this case the transform only indicates whether the
       contents can be used for hash integrity checks.
     """
     _insertflagprocessor(flag, processor, _flagprocessors)
 
 def _insertflagprocessor(flag, processor, flagprocessors):
     if not flag & REVIDX_KNOWN_FLAGS:
         msg = _("cannot register processor on unknown flag '%#x'.") % (flag)
         raise error.ProgrammingError(msg)
     if flag not in REVIDX_FLAGS_ORDER:
         msg = _("flag '%#x' undefined in REVIDX_FLAGS_ORDER.") % (flag)
         raise error.ProgrammingError(msg)
     if flag in flagprocessors:
         msg = _("cannot register multiple processors on flag '%#x'.") % (flag)
         raise error.Abort(msg)
     flagprocessors[flag] = processor
 
 def getoffset(q):
     return int(q >> 16)
 
 def gettype(q):
     return int(q & 0xFFFF)
 
 def offset_type(offset, type):
     if (type & ~REVIDX_KNOWN_FLAGS) != 0:
         raise ValueError('unknown revlog index flags')
     return int(int(offset) << 16 | type)
 
 @attr.s(slots=True, frozen=True)
 class _revisioninfo(object):
     """Information about a revision that allows building its fulltext
     node:       expected hash of the revision
     p1, p2:     parent revs of the revision
     btext:      built text cache consisting of a one-element list
     cachedelta: (baserev, uncompressed_delta) or None
     flags:      flags associated to the revision storage
 
     One of btext[0] or cachedelta must be set.
     """
     node = attr.ib()
     p1 = attr.ib()
     p2 = attr.ib()
     btext = attr.ib()
     textlen = attr.ib()
     cachedelta = attr.ib()
     flags = attr.ib()
 
 @interfaceutil.implementer(repository.irevisiondelta)
 @attr.s(slots=True)
 class revlogrevisiondelta(object):
     node = attr.ib()
     p1node = attr.ib()
     p2node = attr.ib()
     basenode = attr.ib()
     flags = attr.ib()
     baserevisionsize = attr.ib()
     revision = attr.ib()
     delta = attr.ib()
     linknode = attr.ib(default=None)
 
 @interfaceutil.implementer(repository.iverifyproblem)
 @attr.s(frozen=True)
 class revlogproblem(object):
     warning = attr.ib(default=None)
     error = attr.ib(default=None)
     node = attr.ib(default=None)
 
 # index v0:
 #  4 bytes: offset
 #  4 bytes: compressed length
 #  4 bytes: base rev
 #  4 bytes: link rev
 # 20 bytes: parent 1 nodeid
 # 20 bytes: parent 2 nodeid
 # 20 bytes: nodeid
 indexformatv0 = struct.Struct(">4l20s20s20s")
 indexformatv0_pack = indexformatv0.pack
 indexformatv0_unpack = indexformatv0.unpack
 
 class revlogoldindex(list):
     def __getitem__(self, i):
         if i == -1:
             return (0, 0, 0, -1, -1, -1, -1, nullid)
         return list.__getitem__(self, i)
 
 class revlogoldio(object):
     def __init__(self):
         self.size = indexformatv0.size
 
     def parseindex(self, data, inline):
         s = self.size
         index = []
         nodemap = {nullid: nullrev}
         n = off = 0
         l = len(data)
         while off + s <= l:
             cur = data[off:off + s]
             off += s
             e = indexformatv0_unpack(cur)
             # transform to revlogv1 format
             e2 = (offset_type(e[0], 0), e[1], -1, e[2], e[3],
                   nodemap.get(e[4], nullrev), nodemap.get(e[5], nullrev), e[6])
             index.append(e2)
             nodemap[e[6]] = n
             n += 1
 
         return revlogoldindex(index), nodemap, None
 
     def packentry(self, entry, node, version, rev):
         if gettype(entry[0]):
             raise error.RevlogError(_('index entry flags need revlog '
                                       'version 1'))
         e2 = (getoffset(entry[0]), entry[1], entry[3], entry[4],
               node(entry[5]), node(entry[6]), entry[7])
         return indexformatv0_pack(*e2)
 
 # index ng:
 #  6 bytes: offset
 #  2 bytes: flags
 #  4 bytes: compressed length
 #  4 bytes: uncompressed length
 #  4 bytes: base rev
 #  4 bytes: link rev
 #  4 bytes: parent 1 rev
 #  4 bytes: parent 2 rev
 # 32 bytes: nodeid
 indexformatng = struct.Struct(">Qiiiiii20s12x")
 indexformatng_pack = indexformatng.pack
 versionformat = struct.Struct(">I")
 versionformat_pack = versionformat.pack
 versionformat_unpack = versionformat.unpack
 
 # corresponds to uncompressed length of indexformatng (2 gigs, 4-byte
 # signed integer)
 _maxentrysize = 0x7fffffff
 
 class revlogio(object):
     def __init__(self):
         self.size = indexformatng.size
 
     def parseindex(self, data, inline):
         # call the C implementation to parse the index data
         index, cache = parsers.parse_index2(data, inline)
         return index, getattr(index, 'nodemap', None), cache
 
     def packentry(self, entry, node, version, rev):
         p = indexformatng_pack(*entry)
         if rev == 0:
             p = versionformat_pack(version) + p[4:]
         return p
 
 class revlog(object):
     """
     the underlying revision storage object
 
     A revlog consists of two parts, an index and the revision data.
 
     The index is a file with a fixed record size containing
     information on each revision, including its nodeid (hash), the
     nodeids of its parents, the position and offset of its data within
     the data file, and the revision it's based on. Finally, each entry
     contains a linkrev entry that can serve as a pointer to external
     data.
 
     The revision data itself is a linear collection of data chunks.
     Each chunk represents a revision and is usually represented as a
     delta against the previous chunk. To bound lookup time, runs of
     deltas are limited to about 2 times the length of the original
     version data. This makes retrieval of a version proportional to
     its size, or O(1) relative to the number of revisions.
 
     Both pieces of the revlog are written to in an append-only
     fashion, which means we never need to rewrite a file to insert or
     remove data, and can use some simple techniques to avoid the need
     for locking while reading.
 
     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.
     """
     def __init__(self, opener, indexfile, datafile=None, checkambig=False,
-                 mmaplargeindex=False, censorable=False):
+                 mmaplargeindex=False, censorable=False,
+                 upperboundcomp=None):
         """
         create a revlog object
 
         opener is a function that abstracts the file opening operation
         and can be used to implement COW semantics or the like.
+
         """
+        self.upperboundcomp = upperboundcomp
         self.indexfile = indexfile
         self.datafile = datafile or (indexfile[:-2] + ".d")
         self.opener = opener
         #  When True, indexfile is opened with checkambig=True at writing, to
         #  avoid file stat ambiguity.
         self._checkambig = checkambig
         self._mmaplargeindex = mmaplargeindex
         self._censorable = censorable
         # 3-tuple of (node, rev, text) for a raw revision.
         self._revisioncache = None
         # Maps rev to chain base rev.
         self._chainbasecache = util.lrucachedict(100)
         # 2-tuple of (offset, data) of raw data from the revlog at an offset.
         self._chunkcache = (0, '')
         # How much data to read and cache into the raw revlog data cache.
         self._chunkcachesize = 65536
         self._maxchainlen = None
         self._deltabothparents = True
         self.index = []
         # Mapping of partial identifiers to full nodes.
         self._pcache = {}
         # Mapping of revision integer to full node.
         self._nodecache = {nullid: nullrev}
         self._nodepos = None
         self._compengine = 'zlib'
         self._compengineopts = {}
         self._maxdeltachainspan = -1
         self._withsparseread = False
         self._sparserevlog = False
         self._srdensitythreshold = 0.50
         self._srmingapsize = 262144
 
         # Make copy of flag processors so each revlog instance can support
         # custom flags.
         self._flagprocessors = dict(_flagprocessors)
 
         # 2-tuple of file handles being used for active writing.
         self._writinghandles = None
 
         self._loadindex()
 
     def _loadindex(self):
         mmapindexthreshold = None
         opts = getattr(self.opener, 'options', {}) or {}
 
         if 'revlogv2' in opts:
             newversionflags = REVLOGV2 | FLAG_INLINE_DATA
         elif 'revlogv1' in opts:
             newversionflags = REVLOGV1 | FLAG_INLINE_DATA
             if 'generaldelta' in opts:
                 newversionflags |= FLAG_GENERALDELTA
         elif getattr(self.opener, 'options', None) is not None:
             # If options provided but no 'revlog*' found, the repository
             # would have no 'requires' file in it, which means we have to
             # stick to the old format.
             newversionflags = REVLOGV0
         else:
             newversionflags = REVLOG_DEFAULT_VERSION
 
         if 'chunkcachesize' in opts:
             self._chunkcachesize = opts['chunkcachesize']
         if 'maxchainlen' in opts:
             self._maxchainlen = opts['maxchainlen']
         if 'deltabothparents' in opts:
             self._deltabothparents = opts['deltabothparents']
         self._lazydelta = bool(opts.get('lazydelta', True))
         self._lazydeltabase = False
         if self._lazydelta:
             self._lazydeltabase = bool(opts.get('lazydeltabase', False))
         if 'compengine' in opts:
             self._compengine = opts['compengine']
         if 'zlib.level' in opts:
             self._compengineopts['zlib.level'] = opts['zlib.level']
         if 'zstd.level' in opts:
             self._compengineopts['zstd.level'] = opts['zstd.level']
         if 'maxdeltachainspan' in opts:
             self._maxdeltachainspan = opts['maxdeltachainspan']
         if self._mmaplargeindex and 'mmapindexthreshold' in opts:
             mmapindexthreshold = opts['mmapindexthreshold']
         self._sparserevlog = bool(opts.get('sparse-revlog', False))
         withsparseread = bool(opts.get('with-sparse-read', False))
         # sparse-revlog forces sparse-read
         self._withsparseread = self._sparserevlog or withsparseread
         if 'sparse-read-density-threshold' in opts:
             self._srdensitythreshold = opts['sparse-read-density-threshold']
         if 'sparse-read-min-gap-size' in opts:
             self._srmingapsize = opts['sparse-read-min-gap-size']
         if opts.get('enableellipsis'):
             self._flagprocessors[REVIDX_ELLIPSIS] = ellipsisprocessor
 
         # revlog v0 doesn't have flag processors
         for flag, processor in opts.get(b'flagprocessors', {}).iteritems():
             _insertflagprocessor(flag, processor, self._flagprocessors)
 
         if self._chunkcachesize <= 0:
             raise error.RevlogError(_('revlog chunk cache size %r is not '
                                       'greater than 0') % self._chunkcachesize)
         elif self._chunkcachesize & (self._chunkcachesize - 1):
             raise error.RevlogError(_('revlog chunk cache size %r is not a '
                                       'power of 2') % self._chunkcachesize)
 
         indexdata = ''
         self._initempty = True
         try:
             with self._indexfp() as f:
                 if (mmapindexthreshold is not None and
                     self.opener.fstat(f).st_size >= mmapindexthreshold):
                     # TODO: should .close() to release resources without
                     # relying on Python GC
                     indexdata = util.buffer(util.mmapread(f))
                 else:
                     indexdata = f.read()
             if len(indexdata) > 0:
                 versionflags = versionformat_unpack(indexdata[:4])[0]
                 self._initempty = False
             else:
                 versionflags = newversionflags
         except IOError as inst:
             if inst.errno != errno.ENOENT:
                 raise
 
             versionflags = newversionflags
 
         self.version = versionflags
 
         flags = versionflags & ~0xFFFF
         fmt = versionflags & 0xFFFF
 
         if fmt == REVLOGV0:
             if flags:
                 raise error.RevlogError(_('unknown flags (%#04x) in version %d '
                                           'revlog %s') %
                                         (flags >> 16, fmt, self.indexfile))
 
             self._inline = False
             self._generaldelta = False
 
         elif fmt == REVLOGV1:
             if flags & ~REVLOGV1_FLAGS:
                 raise error.RevlogError(_('unknown flags (%#04x) in version %d '
                                           'revlog %s') %
                                         (flags >> 16, fmt, self.indexfile))
 
             self._inline = versionflags & FLAG_INLINE_DATA
             self._generaldelta = versionflags & FLAG_GENERALDELTA
 
         elif fmt == REVLOGV2:
             if flags & ~REVLOGV2_FLAGS:
                 raise error.RevlogError(_('unknown flags (%#04x) in version %d '
                                           'revlog %s') %
                                         (flags >> 16, fmt, self.indexfile))
 
             self._inline = versionflags & FLAG_INLINE_DATA
             # generaldelta implied by version 2 revlogs.
             self._generaldelta = True
 
         else:
             raise error.RevlogError(_('unknown version (%d) in revlog %s') %
                                     (fmt, self.indexfile))
         # sparse-revlog can't be on without general-delta (issue6056)
         if not self._generaldelta:
             self._sparserevlog = False
 
         self._storedeltachains = True
 
         self._io = revlogio()
         if self.version == REVLOGV0:
             self._io = revlogoldio()
         try:
             d = self._io.parseindex(indexdata, self._inline)
         except (ValueError, IndexError):
             raise error.RevlogError(_("index %s is corrupted") %
                                     self.indexfile)
         self.index, nodemap, self._chunkcache = d
         if nodemap is not None:
             self.nodemap = self._nodecache = nodemap
         if not self._chunkcache:
             self._chunkclear()
         # revnum -> (chain-length, sum-delta-length)
         self._chaininfocache = {}
         # revlog header -> revlog compressor
         self._decompressors = {}
 
     @util.propertycache
     def _compressor(self):
         engine = util.compengines[self._compengine]
         return engine.revlogcompressor(self._compengineopts)
 
     def _indexfp(self, mode='r'):
         """file object for the revlog's index file"""
         args = {r'mode': mode}
         if mode != 'r':
             args[r'checkambig'] = self._checkambig
         if mode == 'w':
             args[r'atomictemp'] = True
         return self.opener(self.indexfile, **args)
 
     def _datafp(self, mode='r'):
         """file object for the revlog's data file"""
         return self.opener(self.datafile, mode=mode)
 
     @contextlib.contextmanager
     def _datareadfp(self, existingfp=None):
         """file object suitable to read data"""
         # Use explicit file handle, if given.
         if existingfp is not None:
             yield existingfp
 
         # Use a file handle being actively used for writes, if available.
         # There is some danger to doing this because reads will seek the
         # file. However, _writeentry() performs a SEEK_END before all writes,
         # so we should be safe.
         elif self._writinghandles:
             if self._inline:
                 yield self._writinghandles[0]
             else:
                 yield self._writinghandles[1]
 
         # Otherwise open a new file handle.
         else:
             if self._inline:
                 func = self._indexfp
             else:
                 func = self._datafp
             with func() as fp:
                 yield fp
 
     def tip(self):
         return self.node(len(self.index) - 1)
     def __contains__(self, rev):
         return 0 <= rev < len(self)
     def __len__(self):
         return len(self.index)
     def __iter__(self):
         return iter(pycompat.xrange(len(self)))
     def revs(self, start=0, stop=None):
         """iterate over all rev in this revlog (from start to stop)"""
         return storageutil.iterrevs(len(self), start=start, stop=stop)
 
     @util.propertycache
     def nodemap(self):
         if self.index:
             # populate mapping down to the initial node
             node0 = self.index[0][7]  # get around changelog filtering
             self.rev(node0)
         return self._nodecache
 
     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 clearcaches(self):
         self._revisioncache = None
         self._chainbasecache.clear()
         self._chunkcache = (0, '')
         self._pcache = {}
 
         try:
             # If we are using the native C version, you are in a fun case
             # where self.index, self.nodemap and self._nodecaches is the same
             # object.
             self._nodecache.clearcaches()
         except AttributeError:
             self._nodecache = {nullid: nullrev}
             self._nodepos = None
 
     def rev(self, node):
         try:
             return self._nodecache[node]
         except TypeError:
             raise
         except error.RevlogError:
             # parsers.c radix tree lookup failed
             if node == wdirid or node in wdirfilenodeids:
                 raise error.WdirUnsupported
             raise error.LookupError(node, self.indexfile, _('no node'))
         except KeyError:
             # pure python cache lookup failed
             n = self._nodecache
             i = self.index
             p = self._nodepos
             if p is None:
                 p = len(i) - 1
             else:
                 assert p < len(i)
             for r in pycompat.xrange(p, -1, -1):
                 v = i[r][7]
                 n[v] = r
                 if v == node:
                     self._nodepos = r - 1
                     return r
             if node == wdirid or node in wdirfilenodeids:
                 raise error.WdirUnsupported
             raise error.LookupError(node, self.indexfile, _('no node'))
 
     # Accessors for index entries.
 
     # First tuple entry is 8 bytes. First 6 bytes are offset. Last 2 bytes
     # are flags.
     def start(self, rev):
         return int(self.index[rev][0] >> 16)
 
     def flags(self, rev):
         return self.index[rev][0] & 0xFFFF
 
     def length(self, rev):
         return self.index[rev][1]
 
     def rawsize(self, rev):
         """return the length of the uncompressed text for a given revision"""
         l = self.index[rev][2]
         if l >= 0:
             return l
 
         t = self.revision(rev, raw=True)
         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 & (REVIDX_KNOWN_FLAGS ^ REVIDX_ELLIPSIS) == 0:
             return self.rawsize(rev)
 
         return len(self.revision(rev, raw=False))
 
     def chainbase(self, rev):
         base = self._chainbasecache.get(rev)
         if base is not None:
             return base
 
         index = self.index
         iterrev = rev
         base = index[iterrev][3]
         while base != iterrev:
             iterrev = base
             base = index[iterrev][3]
 
         self._chainbasecache[rev] = base
         return base
 
     def linkrev(self, rev):
         return self.index[rev][4]
 
     def parentrevs(self, rev):
         try:
             entry = self.index[rev]
         except IndexError:
             if rev == wdirrev:
                 raise error.WdirUnsupported
             raise
 
         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)]
         return i[d[5]][7], i[d[6]][7] # map revisions to nodes inline
 
     def chainlen(self, rev):
         return self._chaininfo(rev)[0]
 
     def _chaininfo(self, rev):
         chaininfocache = self._chaininfocache
         if rev in chaininfocache:
             return chaininfocache[rev]
         index = self.index
         generaldelta = self._generaldelta
         iterrev = rev
         e = index[iterrev]
         clen = 0
         compresseddeltalen = 0
         while iterrev != e[3]:
             clen += 1
             compresseddeltalen += e[1]
             if generaldelta:
                 iterrev = e[3]
             else:
                 iterrev -= 1
             if iterrev in chaininfocache:
                 t = chaininfocache[iterrev]
                 clen += t[0]
                 compresseddeltalen += t[1]
                 break
             e = index[iterrev]
         else:
             # Add text length of base since decompressing that also takes
             # work. For cache hits the length is already included.
             compresseddeltalen += e[1]
         r = (clen, compresseddeltalen)
         chaininfocache[rev] = r
         return r
 
     def _deltachain(self, rev, stoprev=None):
         """Obtain the delta chain for a revision.
 
         ``stoprev`` specifies a revision to stop at. If not specified, we
         stop at the base of the chain.
 
         Returns a 2-tuple of (chain, stopped) where ``chain`` is a list of
         revs in ascending order and ``stopped`` is a bool indicating whether
         ``stoprev`` was hit.
         """
         # Try C implementation.
         try:
             return self.index.deltachain(rev, stoprev, self._generaldelta)
         except AttributeError:
             pass
 
         chain = []
 
         # Alias to prevent attribute lookup in tight loop.
         index = self.index
         generaldelta = self._generaldelta
 
         iterrev = rev
         e = index[iterrev]
         while iterrev != e[3] and iterrev != stoprev:
             chain.append(iterrev)
             if generaldelta:
                 iterrev = e[3]
             else:
                 iterrev -= 1
             e = index[iterrev]
 
         if iterrev == stoprev:
             stopped = True
         else:
             chain.append(iterrev)
             stopped = False
 
         chain.reverse()
         return chain, stopped
 
     def ancestors(self, revs, stoprev=0, inclusive=False):
         """Generate the ancestors of 'revs' in reverse revision order.
         Does not generate revs lower than stoprev.
 
         See the documentation for ancestor.lazyancestors for more details."""
 
         # first, make sure start revisions aren't filtered
         revs = list(revs)
         checkrev = self.node
         for r in revs:
             checkrev(r)
         # and we're sure ancestors aren't filtered as well
 
         if rustancestor is not None:
             lazyancestors = rustancestor.LazyAncestors
             arg = self.index
         elif util.safehasattr(parsers, 'rustlazyancestors'):
             lazyancestors = ancestor.rustlazyancestors
             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 = [nullid]
         if heads is None:
             heads = self.heads()
 
         common = [self.rev(n) for n in common]
         heads = [self.rev(n) for n in heads]
 
         # we want the ancestors, but inclusive
         class lazyset(object):
             def __init__(self, lazyvalues):
                 self.addedvalues = set()
                 self.lazyvalues = lazyvalues
 
             def __contains__(self, value):
                 return value in self.addedvalues or value in self.lazyvalues
 
             def __iter__(self):
                 added = self.addedvalues
                 for r in added:
                     yield r
                 for r in self.lazyvalues:
                     if not r in added:
                         yield r
 
             def add(self, value):
                 self.addedvalues.add(value)
 
             def update(self, values):
                 self.addedvalues.update(values)
 
         has = lazyset(self.ancestors(common))
         has.add(nullrev)
         has.update(common)
 
         # take all ancestors from heads that aren't in has
         missing = set()
         visit = collections.deque(r for r in heads if r not in has)
         while visit:
             r = visit.popleft()
             if r in missing:
                 continue
             else:
                 missing.add(r)
                 for p in self.parentrevs(r):
                     if p not in has:
                         visit.append(p)
         missing = list(missing)
         missing.sort()
         return has, [self.node(miss) for miss in missing]
 
     def incrementalmissingrevs(self, common=None):
         """Return an object that can be used to incrementally compute the
         revision numbers of the ancestors of arbitrary sets that are not
         ancestors of common. This is an ancestor.incrementalmissingancestors
         object.
 
         'common' is a list of revision numbers. If common is not supplied, uses
         nullrev.
         """
         if common is None:
             common = [nullrev]
 
         if rustancestor is not None:
             return rustancestor.MissingAncestors(self.index, common)
         return ancestor.incrementalmissingancestors(self.parentrevs, common)
 
     def findmissingrevs(self, common=None, heads=None):
         """Return the revision numbers of the ancestors of heads that
         are not ancestors of common.
 
         More specifically, return a list of revision numbers corresponding to
         nodes N such that every N satisfies the following constraints:
 
           1. N is an ancestor of some node in 'heads'
           2. N is not an ancestor of any node in 'common'
 
         The list is sorted by revision number, meaning it is
         topologically sorted.
 
         'heads' and 'common' are both lists of revision numbers.  If heads is
         not supplied, uses all of the revlog's heads.  If common is not
         supplied, uses nullid."""
         if common is None:
             common = [nullrev]
         if heads is None:
             heads = self.headrevs()
 
         inc = self.incrementalmissingrevs(common=common)
         return inc.missingancestors(heads)
 
     def findmissing(self, common=None, heads=None):
         """Return the ancestors of heads that are not ancestors of common.
 
         More specifically, return a list of nodes N such that every N
         satisfies the following constraints:
 
           1. N is an ancestor of some node in 'heads'
           2. N is not an ancestor of any node in 'common'
 
         The list is sorted by revision number, meaning it is
         topologically sorted.
 
         'heads' and 'common' are both lists of node IDs.  If heads is
         not supplied, uses all of the revlog's heads.  If common is not
         supplied, uses nullid."""
         if common is None:
             common = [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 = [nullid] # Everybody's a descendant of nullid
             lowestrev = nullrev
         if (lowestrev == nullrev) and (heads is None):
             # We want _all_ the nodes!
             return ([self.node(r) for r in self], [nullid], list(self.heads()))
         if heads is None:
             # All nodes are ancestors, so the latest ancestor is the last
             # node.
             highestrev = len(self) - 1
             # Set ancestors to None to signal that every node is an ancestor.
             ancestors = None
             # Set heads to an empty dictionary for later discovery of heads
             heads = {}
         else:
             heads = list(heads)
             if not heads:
                 return nonodes
             ancestors = set()
             # Turn heads into a dictionary so we can remove 'fake' heads.
             # Also, later we will be using it to filter out the heads we can't
             # find from roots.
             heads = dict.fromkeys(heads, False)
             # Start at the top and keep marking parents until we're done.
             nodestotag = set(heads)
             # Remember where the top was so we can use it as a limit later.
             highestrev = max([self.rev(n) for n in nodestotag])
             while nodestotag:
                 # grab a node to tag
                 n = nodestotag.pop()
                 # Never tag nullid
                 if n == nullid:
                     continue
                 # A node's revision number represents its place in a
                 # topologically sorted list of nodes.
                 r = self.rev(n)
                 if r >= lowestrev:
                     if n not in ancestors:
                         # If we are possibly a descendant of one of the roots
                         # and we haven't already been marked as an ancestor
                         ancestors.add(n) # Mark as ancestor
                         # Add non-nullid parents to list of nodes to tag.
                         nodestotag.update([p for p in self.parents(n) if
                                            p != nullid])
                     elif n in heads: # We've seen it before, is it a fake head?
                         # So it is, real heads should not be the ancestors of
                         # any other heads.
                         heads.pop(n)
             if not ancestors:
                 return nonodes
             # Now that we have our set of ancestors, we want to remove any
             # roots that are not ancestors.
 
             # If one of the roots was nullid, everything is included anyway.
             if lowestrev > nullrev:
                 # But, since we weren't, let's recompute the lowest rev to not
                 # include roots that aren't ancestors.
 
                 # Filter out roots that aren't ancestors of heads
                 roots = [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 = [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.iteritems() if flag]
         roots = list(roots)
         assert orderedout
         assert roots
         assert heads
         return (orderedout, roots, heads)
 
     def headrevs(self, revs=None):
         if revs is None:
             try:
                 return self.index.headrevs()
             except AttributeError:
                 return self._headrevs()
         if rustdagop is not None:
             return rustdagop.headrevs(self.index, revs)
         return dagop.headrevs(revs, self._uncheckedparentrevs)
 
     def computephases(self, roots):
         return self.index.computephasesmapsets(roots)
 
     def _headrevs(self):
         count = len(self)
         if not count:
             return [nullrev]
         # we won't iter over filtered rev so nobody is a head at start
         ishead = [0] * (count + 1)
         index = self.index
         for r in self:
             ishead[r] = 1  # I may be an head
             e = index[r]
             ishead[e[5]] = ishead[e[6]] = 0  # my parent are not
         return [r for r, val in enumerate(ishead) if val]
 
     def heads(self, start=None, stop=None):
         """return the list of all nodes that have no children
 
         if start is specified, only heads that are descendants of
         start will be returned
         if stop is specified, it will consider all the revs from stop
         as if they had no children
         """
         if start is None and stop is None:
             if not len(self):
                 return [nullid]
             return [self.node(r) for r in self.headrevs()]
 
         if start is None:
             start = nullrev
         else:
             start = self.rev(start)
 
         stoprevs = set(self.rev(n) for n in stop or [])
 
         revs = dagop.headrevssubset(self.revs, self.parentrevs, startrev=start,
                                     stoprevs=stoprevs)
 
         return [self.node(rev) for rev in revs]
 
     def children(self, node):
         """find the children of a given node"""
         c = []
         p = self.rev(node)
         for r in self.revs(start=p + 1):
             prevs = [pr for pr in self.parentrevs(r) if pr != nullrev]
             if prevs:
                 for pr in prevs:
                     if pr == p:
                         c.append(self.node(r))
             elif p == nullrev:
                 c.append(self.node(r))
         return c
 
     def commonancestorsheads(self, a, b):
         """calculate all the heads of the common ancestors of nodes a and b"""
         a, b = self.rev(a), self.rev(b)
         ancs = self._commonancestorsheads(a, b)
         return pycompat.maplist(self.node, ancs)
 
     def _commonancestorsheads(self, *revs):
         """calculate all the heads of the common ancestors of revs"""
         try:
             ancs = self.index.commonancestorsheads(*revs)
         except (AttributeError, OverflowError): # C implementation failed
             ancs = ancestor.commonancestorsheads(self.parentrevs, *revs)
         return ancs
 
     def isancestor(self, a, b):
         """return True if node a is an ancestor of node b
 
         A revision is considered an ancestor of itself."""
         a, b = self.rev(a), self.rev(b)
         return self.isancestorrev(a, b)
 
     def isancestorrev(self, a, b):
         """return True if revision a is an ancestor of revision b
 
         A revision is considered an ancestor of itself.
 
         The implementation of this is trivial but the use of
         reachableroots is not."""
         if a == nullrev:
             return True
         elif a == b:
             return True
         elif a > b:
             return False
         return bool(self.reachableroots(a, [b], [a], includepath=False))
 
     def reachableroots(self, minroot, heads, roots, includepath=False):
         """return (heads(::<roots> and <roots>::<heads>))
 
         If includepath is True, return (<roots>::<heads>)."""
         try:
             return self.index.reachableroots2(minroot, heads, roots,
                                               includepath)
         except AttributeError:
             return dagop._reachablerootspure(self.parentrevs,
                                              minroot, roots, heads, includepath)
 
     def ancestor(self, a, b):
         """calculate the "best" common ancestor of nodes a and b"""
 
         a, b = self.rev(a), self.rev(b)
         try:
             ancs = self.index.ancestors(a, b)
         except (AttributeError, OverflowError):
             ancs = ancestor.ancestors(self.parentrevs, a, b)
         if ancs:
             # choose a consistent winner when there's a tie
             return min(map(self.node, ancs))
         return nullid
 
     def _match(self, id):
         if isinstance(id, int):
             # rev
             return self.node(id)
         if len(id) == 20:
             # possibly a binary node
             # odds of a binary node being all hex in ASCII are 1 in 10**25
             try:
                 node = id
                 self.rev(node) # quick search the index
                 return node
             except error.LookupError:
                 pass # may be partial hex id
         try:
             # str(rev)
             rev = int(id)
             if "%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) == 40:
             try:
                 # a full hex nodeid?
                 node = bin(id)
                 self.rev(node)
                 return node
             except (TypeError, error.LookupError):
                 pass
 
     def _partialmatch(self, id):
         # we don't care wdirfilenodeids as they should be always full hash
         maybewdir = wdirhex.startswith(id)
         try:
             partial = self.index.partialmatch(id)
             if partial and self.hasnode(partial):
                 if maybewdir:
                     # single 'ff...' match in radix tree, ambiguous with wdir
                     raise error.RevlogError
                 return partial
             if maybewdir:
                 # no 'ff...' match in radix tree, wdir identified
                 raise error.WdirUnsupported
             return None
         except error.RevlogError:
             # parsers.c radix tree lookup gave multiple matches
             # fast path: for unfiltered changelog, radix tree is accurate
             if not getattr(self, 'filteredrevs', None):
                 raise error.AmbiguousPrefixLookupError(
                     id, self.indexfile, _('ambiguous identifier'))
             # fall through to slow path that filters hidden revisions
         except (AttributeError, ValueError):
             # we are pure python, or key was too short to search radix tree
             pass
 
         if id in self._pcache:
             return self._pcache[id]
 
         if len(id) <= 40:
             try:
                 # hex(node)[:...]
                 l = len(id) // 2  # grab an even number of digits
                 prefix = bin(id[:l * 2])
                 nl = [e[7] for e in self.index if e[7].startswith(prefix)]
                 nl = [n for n in nl if hex(n).startswith(id) and
                       self.hasnode(n)]
                 if nullhex.startswith(id):
                     nl.append(nullid)
                 if len(nl) > 0:
                     if len(nl) == 1 and not maybewdir:
                         self._pcache[id] = nl[0]
                         return nl[0]
                     raise error.AmbiguousPrefixLookupError(
                         id, self.indexfile, _('ambiguous identifier'))
                 if maybewdir:
                     raise error.WdirUnsupported
                 return None
             except TypeError:
                 pass
 
     def lookup(self, id):
         """locate a node based on:
             - revision number or str(revision number)
             - nodeid or subset of hex nodeid
         """
         n = self._match(id)
         if n is not None:
             return n
         n = self._partialmatch(id)
         if n:
             return n
 
         raise error.LookupError(id, self.indexfile, _('no match found'))
 
     def shortest(self, node, minlength=1):
         """Find the shortest unambiguous prefix that matches node."""
         def isvalid(prefix):
             try:
                 node = self._partialmatch(prefix)
             except error.AmbiguousPrefixLookupError:
                 return False
             except error.WdirUnsupported:
                 # single 'ff...' match
                 return True
             if node is None:
                 raise error.LookupError(node, self.indexfile, _('no node'))
             return True
 
         def maybewdir(prefix):
             return all(c == 'f' for c in pycompat.iterbytestr(prefix))
 
         hexnode = hex(node)
 
         def disambiguate(hexnode, minlength):
             """Disambiguate against wdirid."""
             for length in range(minlength, 41):
                 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 != wdirid:
                     raise error.LookupError(node, self.indexfile, _('no node'))
             except AttributeError:
                 # Fall through to pure code
                 pass
 
         if node == wdirid:
             for length in range(minlength, 41):
                 prefix = hexnode[:length]
                 if isvalid(prefix):
                     return prefix
 
         for length in range(minlength, 41):
             prefix = hexnode[:length]
             if isvalid(prefix):
                 return disambiguate(hexnode, length)
 
     def cmp(self, node, text):
         """compare text with a given file revision
 
         returns True if text is different than what is stored.
         """
         p1, p2 = self.parents(node)
         return storageutil.hashrevisionsha1(text, p1, p2) != node
 
     def _cachesegment(self, offset, data):
         """Add a segment to the revlog cache.
 
         Accepts an absolute offset and the data that is at that location.
         """
         o, d = self._chunkcache
         # try to add to existing cache
         if o + len(d) == offset and len(d) + len(data) < _chunksize:
             self._chunkcache = o, d + data
         else:
             self._chunkcache = offset, data
 
     def _readsegment(self, offset, length, df=None):
         """Load a segment of raw data from the revlog.
 
         Accepts an absolute offset, length to read, and an optional existing
         file handle to read from.
 
         If an existing file handle is passed, it will be seeked and the
         original seek position will NOT be restored.
 
         Returns a str or buffer of raw byte data.
 
         Raises if the requested number of bytes could not be read.
         """
         # Cache data both forward and backward around the requested
         # data, in a fixed size window. This helps speed up operations
         # involving reading the revlog backwards.
         cachesize = self._chunkcachesize
         realoffset = offset & ~(cachesize - 1)
         reallength = (((offset + length + cachesize) & ~(cachesize - 1))
                       - realoffset)
         with self._datareadfp(df) as df:
             df.seek(realoffset)
             d = df.read(reallength)
 
         self._cachesegment(realoffset, d)
         if offset != realoffset or reallength != length:
             startoffset = offset - realoffset
             if len(d) - startoffset < length:
                 raise error.RevlogError(
                     _('partial read of revlog %s; expected %d bytes from '
                       'offset %d, got %d') %
                     (self.indexfile if self._inline else self.datafile,
                      length, realoffset, len(d) - startoffset))
 
             return util.buffer(d, startoffset, length)
 
         if len(d) < length:
             raise error.RevlogError(
                 _('partial read of revlog %s; expected %d bytes from offset '
                   '%d, got %d') %
                 (self.indexfile if self._inline else self.datafile,
                  length, offset, len(d)))
 
         return d
 
     def _getsegment(self, offset, length, df=None):
         """Obtain a segment of raw data from the revlog.
 
         Accepts an absolute offset, length of bytes to obtain, and an
         optional file handle to the already-opened revlog. If the file
         handle is used, it's original seek position will not be preserved.
 
         Requests for data may be returned from a cache.
 
         Returns a str or a buffer instance of raw byte data.
         """
         o, d = self._chunkcache
         l = len(d)
 
         # is it in the cache?
         cachestart = offset - o
         cacheend = cachestart + length
         if cachestart >= 0 and cacheend <= l:
             if cachestart == 0 and cacheend == l:
                 return d # avoid a copy
             return util.buffer(d, cachestart, cacheend - cachestart)
 
         return self._readsegment(offset, length, df=df)
 
     def _getsegmentforrevs(self, startrev, endrev, df=None):
         """Obtain a segment of raw data corresponding to a range of revisions.
 
         Accepts the start and end revisions and an optional already-open
         file handle to be used for reading. If the file handle is read, its
         seek position will not be preserved.
 
         Requests for data may be satisfied by a cache.
 
         Returns a 2-tuple of (offset, data) for the requested range of
         revisions. Offset is the integer offset from the beginning of the
         revlog and data is a str or buffer of the raw byte data.
 
         Callers will need to call ``self.start(rev)`` and ``self.length(rev)``
         to determine where each revision's data begins and ends.
         """
         # Inlined self.start(startrev) & self.end(endrev) for perf reasons
         # (functions are expensive).
         index = self.index
         istart = index[startrev]
         start = int(istart[0] >> 16)
         if startrev == endrev:
             end = start + istart[1]
         else:
             iend = index[endrev]
             end = int(iend[0] >> 16) + iend[1]
 
         if self._inline:
             start += (startrev + 1) * self._io.size
             end += (endrev + 1) * self._io.size
         length = end - start
 
         return start, self._getsegment(start, length, df=df)
 
     def _chunk(self, rev, df=None):
         """Obtain a single decompressed chunk for a revision.
 
         Accepts an integer revision and an optional already-open file handle
         to be used for reading. If used, the seek position of the file will not
         be preserved.
 
         Returns a str holding uncompressed data for the requested revision.
         """
         return self.decompress(self._getsegmentforrevs(rev, rev, df=df)[1])
 
     def _chunks(self, revs, df=None, targetsize=None):
         """Obtain decompressed chunks for the specified revisions.
 
         Accepts an iterable of numeric revisions that are assumed to be in
         ascending order. Also accepts an optional already-open file handle
         to be used for reading. If used, the seek position of the file will
         not be preserved.
 
         This function is similar to calling ``self._chunk()`` multiple times,
         but is faster.
 
         Returns a list with decompressed data for each requested revision.
         """
         if not revs:
             return []
         start = self.start
         length = self.length
         inline = self._inline
         iosize = self._io.size
         buffer = util.buffer
 
         l = []
         ladd = l.append
 
         if not self._withsparseread:
             slicedchunks = (revs,)
         else:
             slicedchunks = deltautil.slicechunk(self, revs,
                                                 targetsize=targetsize)
 
         for revschunk in slicedchunks:
             firstrev = revschunk[0]
             # Skip trailing revisions with empty diff
             for lastrev in revschunk[::-1]:
                 if length(lastrev) != 0:
                     break
 
             try:
                 offset, data = self._getsegmentforrevs(firstrev, lastrev, df=df)
             except OverflowError:
                 # issue4215 - we can't cache a run of chunks greater than
                 # 2G on Windows
                 return [self._chunk(rev, df=df) for rev in revschunk]
 
             decomp = self.decompress
             for rev in revschunk:
                 chunkstart = start(rev)
                 if inline:
                     chunkstart += (rev + 1) * iosize
                 chunklength = length(rev)
                 ladd(decomp(buffer(data, chunkstart - offset, chunklength)))
 
         return l
 
     def _chunkclear(self):
         """Clear the raw chunk cache."""
         self._chunkcache = (0, '')
 
     def deltaparent(self, rev):
         """return deltaparent of the given revision"""
         base = self.index[rev][3]
         if base == rev:
             return nullrev
         elif self._generaldelta:
             return base
         else:
             return rev - 1
 
     def issnapshot(self, rev):
         """tells whether rev is a snapshot
         """
         if not self._sparserevlog:
             return self.deltaparent(rev) == nullrev
         elif util.safehasattr(self.index, 'issnapshot'):
             # directly assign the method to cache the testing and access
             self.issnapshot = self.index.issnapshot
             return self.issnapshot(rev)
         if rev == nullrev:
             return True
         entry = self.index[rev]
         base = entry[3]
         if base == rev:
             return True
         if base == nullrev:
             return True
         p1 = entry[5]
         p2 = entry[6]
         if base == p1 or base == p2:
             return False
         return self.issnapshot(base)
 
     def snapshotdepth(self, rev):
         """number of snapshot in the chain before this one"""
         if not self.issnapshot(rev):
             raise error.ProgrammingError('revision %d not a snapshot')
         return len(self._deltachain(rev)[0]) - 1
 
     def revdiff(self, rev1, rev2):
         """return or calculate a delta between two revisions
 
         The delta calculated is in binary form and is intended to be written to
         revlog data directly. So this function needs raw revision data.
         """
         if rev1 != nullrev and self.deltaparent(rev2) == rev1:
             return bytes(self._chunk(rev2))
 
         return mdiff.textdiff(self.revision(rev1, raw=True),
                               self.revision(rev2, raw=True))
 
     def revision(self, nodeorrev, _df=None, raw=False):
         """return an uncompressed revision of a given node or revision
         number.
 
         _df - an existing file handle to read from. (internal-only)
         raw - an optional argument specifying if the revision data is to be
         treated as raw data when applying flag transforms. 'raw' should be set
         to True when generating changegroups or in debug commands.
         """
         if isinstance(nodeorrev, int):
             rev = nodeorrev
             node = self.node(rev)
         else:
             node = nodeorrev
             rev = None
 
         cachedrev = None
         flags = None
         rawtext = None
         if node == nullid:
             return ""
         if self._revisioncache:
             if self._revisioncache[0] == node:
                 # _cache only stores rawtext
                 if raw:
                     return self._revisioncache[2]
                 # duplicated, but good for perf
                 if rev is None:
                     rev = self.rev(node)
                 if flags is None:
                     flags = self.flags(rev)
                 # no extra flags set, no flag processor runs, text = rawtext
                 if flags == REVIDX_DEFAULT_FLAGS:
                     return self._revisioncache[2]
                 # rawtext is reusable. need to run flag processor
                 rawtext = self._revisioncache[2]
 
             cachedrev = self._revisioncache[1]
 
         # look up what we need to read
         if rawtext is None:
             if rev is None:
                 rev = self.rev(node)
 
             chain, stopped = self._deltachain(rev, stoprev=cachedrev)
             if stopped:
                 rawtext = self._revisioncache[2]
 
             # drop cache to save memory
             self._revisioncache = None
 
             targetsize = None
             rawsize = self.index[rev][2]
             if 0 <= rawsize:
                 targetsize = 4 * rawsize
 
             bins = self._chunks(chain, df=_df, targetsize=targetsize)
             if rawtext is None:
                 rawtext = bytes(bins[0])
                 bins = bins[1:]
 
             rawtext = mdiff.patches(rawtext, bins)
             self._revisioncache = (node, rev, rawtext)
 
         if flags is None:
             if rev is None:
                 rev = self.rev(node)
             flags = self.flags(rev)
 
         text, validatehash = self._processflags(rawtext, flags, 'read', raw=raw)
         if validatehash:
             self.checkhash(text, node, rev=rev)
 
         return text
 
     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 _processflags(self, text, flags, operation, raw=False):
         """Inspect revision data flags and applies transforms defined by
         registered flag processors.
 
         ``text`` - the revision data to process
         ``flags`` - the revision flags
         ``operation`` - the operation being performed (read or write)
         ``raw`` - an optional argument describing if the raw transform should be
         applied.
 
         This method processes the flags in the order (or reverse order if
         ``operation`` is 'write') defined by REVIDX_FLAGS_ORDER, applying the
         flag processors registered for present flags. The order of flags defined
         in REVIDX_FLAGS_ORDER needs to be stable to allow non-commutativity.
 
         Returns a 2-tuple of ``(text, validatehash)`` where ``text`` is the
         processed text and ``validatehash`` is a bool indicating whether the
         returned text should be checked for hash integrity.
 
         Note: If the ``raw`` argument is set, it has precedence over the
         operation and will only update the value of ``validatehash``.
         """
         # fast path: no flag processors will run
         if flags == 0:
             return text, True
         if not operation in ('read', 'write'):
             raise error.ProgrammingError(_("invalid '%s' operation") %
                                          operation)
         # Check all flags are known.
         if flags & ~REVIDX_KNOWN_FLAGS:
             raise error.RevlogError(_("incompatible revision flag '%#x'") %
                                     (flags & ~REVIDX_KNOWN_FLAGS))
         validatehash = True
         # Depending on the operation (read or write), the order might be
         # reversed due to non-commutative transforms.
         orderedflags = REVIDX_FLAGS_ORDER
         if operation == 'write':
             orderedflags = reversed(orderedflags)
 
         for flag in orderedflags:
             # If a flagprocessor has been registered for a known flag, apply the
             # related operation transform and update result tuple.
             if flag & flags:
                 vhash = True
 
                 if flag not in self._flagprocessors:
                     message = _("missing processor for flag '%#x'") % (flag)
                     raise error.RevlogError(message)
 
                 processor = self._flagprocessors[flag]
                 if processor is not None:
                     readtransform, writetransform, rawtransform = processor
 
                     if raw:
                         vhash = rawtransform(self, text)
                     elif operation == 'read':
                         text, vhash = readtransform(self, text)
                     else: # write operation
                         text, vhash = writetransform(self, text)
                 validatehash = validatehash and vhash
 
         return text, validatehash
 
     def checkhash(self, text, node, p1=None, p2=None, rev=None):
         """Check node hash integrity.
 
         Available as a function so that subclasses can extend hash mismatch
         behaviors as needed.
         """
         try:
             if p1 is None and p2 is None:
                 p1, p2 = self.parents(node)
             if node != self.hash(text, p1, p2):
                 # Clear the revision cache on hash failure. The revision cache
                 # only stores the raw revision and clearing the cache does have
                 # the side-effect that we won't have a cache hit when the raw
                 # revision data is accessed. But this case should be rare and
                 # it is extra work to teach the cache about the hash
                 # verification state.
                 if self._revisioncache and self._revisioncache[0] == node:
                     self._revisioncache = None
 
                 revornode = rev
                 if revornode is None:
                     revornode = templatefilters.short(hex(node))
                 raise error.RevlogError(_("integrity check failed on %s:%s")
                     % (self.indexfile, pycompat.bytestr(revornode)))
         except error.RevlogError:
             if self._censorable and storageutil.iscensoredtext(text):
                 raise error.CensoredNodeError(self.indexfile, node, text)
             raise
 
     def _enforceinlinesize(self, tr, fp=None):
         """Check if the revlog is too big for inline and convert if so.
 
         This should be called after revisions are added to the revlog. If the
         revlog has grown too large to be an inline revlog, it will convert it
         to use multiple index and data files.
         """
         tiprev = len(self) - 1
         if (not self._inline or
             (self.start(tiprev) + self.length(tiprev)) < _maxinline):
             return
 
         trinfo = tr.find(self.indexfile)
         if trinfo is None:
             raise error.RevlogError(_("%s not found in the transaction")
                                     % self.indexfile)
 
         trindex = trinfo[2]
         if trindex is not None:
             dataoff = self.start(trindex)
         else:
             # revlog was stripped at start of transaction, use all leftover data
             trindex = len(self) - 1
             dataoff = self.end(tiprev)
 
         tr.add(self.datafile, dataoff)
 
         if fp:
             fp.flush()
             fp.close()
             # We can't use the cached file handle after close(). So prevent
             # its usage.
             self._writinghandles = None
 
         with self._indexfp('r') as ifh, self._datafp('w') as dfh:
             for r in self:
                 dfh.write(self._getsegmentforrevs(r, r, df=ifh)[1])
 
         with self._indexfp('w') as fp:
             self.version &= ~FLAG_INLINE_DATA
             self._inline = False
             io = self._io
             for i in self:
                 e = io.packentry(self.index[i], self.node, self.version, i)
                 fp.write(e)
 
             # the temp file replace the real index when we exit the context
             # manager
 
         tr.replace(self.indexfile, trindex * self._io.size)
         self._chunkclear()
 
     def _nodeduplicatecallback(self, transaction, node):
         """called when trying to add a node already stored.
         """
 
     def addrevision(self, text, transaction, link, p1, p2, cachedelta=None,
                     node=None, flags=REVIDX_DEFAULT_FLAGS, deltacomputer=None):
         """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(_("attempted to add linkrev -1 to %s")
                                     % self.indexfile)
 
         if flags:
             node = node or self.hash(text, p1, p2)
 
         rawtext, validatehash = self._processflags(text, flags, 'write')
 
         # If the flag processor modifies the revision data, ignore any provided
         # cachedelta.
         if rawtext != text:
             cachedelta = None
 
         if len(rawtext) > _maxentrysize:
             raise error.RevlogError(
                 _("%s: size of %d bytes exceeds maximum revlog storage of 2GiB")
                 % (self.indexfile, len(rawtext)))
 
         node = node or self.hash(rawtext, p1, p2)
         if node in self.nodemap:
             return node
 
         if validatehash:
             self.checkhash(rawtext, node, p1=p1, p2=p2)
 
         return self.addrawrevision(rawtext, transaction, link, p1, p2, node,
                                    flags, cachedelta=cachedelta,
                                    deltacomputer=deltacomputer)
 
     def addrawrevision(self, rawtext, transaction, link, p1, p2, node, flags,
                        cachedelta=None, deltacomputer=None):
         """add a raw revision with known flags, node and parents
         useful when reusing a revision not stored in this revlog (ex: received
         over wire, or read from an external bundle).
         """
         dfh = None
         if not self._inline:
             dfh = self._datafp("a+")
         ifh = self._indexfp("a+")
         try:
             return self._addrevision(node, rawtext, transaction, link, p1, p2,
                                      flags, cachedelta, ifh, dfh,
                                      deltacomputer=deltacomputer)
         finally:
             if dfh:
                 dfh.close()
             ifh.close()
 
     def compress(self, data):
         """Generate a possibly-compressed representation of data."""
         if not data:
             return '', data
 
         compressed = self._compressor.compress(data)
 
         if compressed:
             # The revlog compressor added the header in the returned data.
             return '', compressed
 
         if data[0:1] == '\0':
             return '', data
         return '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 == 'x':
             try:
                 return _zlibdecompress(data)
             except zlib.error as e:
                 raise error.RevlogError(_('revlog decompress error: %s') %
                                         stringutil.forcebytestr(e))
         # '\0' is more common than 'u' so it goes first.
         elif t == '\0':
             return data
         elif t == 'u':
             return util.buffer(data, 1)
 
         try:
             compressor = self._decompressors[t]
         except KeyError:
             try:
                 engine = util.compengines.forrevlogheader(t)
                 compressor = engine.revlogcompressor(self._compengineopts)
                 self._decompressors[t] = compressor
             except KeyError:
                 raise error.RevlogError(_('unknown compression type %r') % t)
 
         return compressor.decompress(data)
 
     def _addrevision(self, node, rawtext, transaction, link, p1, p2, flags,
                      cachedelta, ifh, dfh, alwayscache=False,
                      deltacomputer=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 == nullid:
             raise error.RevlogError(_("%s: attempt to add null revision") %
                                     self.indexfile)
         if node == wdirid or node in wdirfilenodeids:
             raise error.RevlogError(_("%s: attempt to add wdir revision") %
                                     self.indexfile)
 
         if self._inline:
             fh = ifh
         else:
             fh = dfh
 
         btext = [rawtext]
 
         curr = len(self)
         prev = curr - 1
         offset = self.end(prev)
         p1r, p2r = self.rev(p1), self.rev(p2)
 
         # full versions are inserted when the needed deltas
         # become comparable to the uncompressed text
         if rawtext is None:
             # need rawtext size, before changed by flag processors, which is
             # the non-raw size. use revlog explicitly to avoid filelog's extra
             # logic that might remove metadata size.
             textlen = mdiff.patchedsize(revlog.size(self, cachedelta[0]),
                                         cachedelta[1])
         else:
             textlen = len(rawtext)
 
         if deltacomputer is None:
             deltacomputer = deltautil.deltacomputer(self)
 
         revinfo = _revisioninfo(node, p1, p2, btext, textlen, cachedelta, flags)
 
         deltainfo = deltacomputer.finddeltainfo(revinfo, fh)
 
         e = (offset_type(offset, flags), deltainfo.deltalen, textlen,
              deltainfo.base, link, p1r, p2r, node)
         self.index.append(e)
         self.nodemap[node] = curr
 
         # Reset the pure node cache start lookup offset to account for new
         # revision.
         if self._nodepos is not None:
             self._nodepos = curr
 
         entry = self._io.packentry(e, self.node, self.version, curr)
         self._writeentry(transaction, ifh, dfh, entry, deltainfo.data,
                          link, offset)
 
         rawtext = btext[0]
 
         if alwayscache and rawtext is None:
             rawtext = deltacomputer.buildtext(revinfo, fh)
 
         if type(rawtext) == bytes: # only accept immutable objects
             self._revisioncache = (node, curr, rawtext)
         self._chainbasecache[curr] = deltainfo.chainbase
         return node
 
     def _writeentry(self, transaction, ifh, dfh, entry, data, link, 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.
         ifh.seek(0, os.SEEK_END)
         if dfh:
             dfh.seek(0, os.SEEK_END)
 
         curr = len(self) - 1
         if not self._inline:
             transaction.add(self.datafile, offset)
             transaction.add(self.indexfile, curr * len(entry))
             if data[0]:
                 dfh.write(data[0])
             dfh.write(data[1])
             ifh.write(entry)
         else:
             offset += curr * self._io.size
             transaction.add(self.indexfile, offset, curr)
             ifh.write(entry)
             ifh.write(data[0])
             ifh.write(data[1])
             self._enforceinlinesize(transaction, ifh)
 
     def addgroup(self, deltas, linkmapper, transaction, addrevisioncb=None):
         """
         add a delta group
 
         given a set of deltas, add them to the revision log. the
         first delta is against its parent, which should be in our
         log, the rest are against the previous delta.
 
         If ``addrevisioncb`` is defined, it will be called with arguments of
         this revlog and the node that was added.
         """
 
         if self._writinghandles:
             raise error.ProgrammingError('cannot nest addgroup() calls')
 
         nodes = []
 
         r = len(self)
         end = 0
         if r:
             end = self.end(r - 1)
         ifh = self._indexfp("a+")
         isize = r * self._io.size
         if self._inline:
             transaction.add(self.indexfile, end + isize, r)
             dfh = None
         else:
             transaction.add(self.indexfile, isize, r)
             transaction.add(self.datafile, end)
             dfh = self._datafp("a+")
         def flush():
             if dfh:
                 dfh.flush()
             ifh.flush()
 
         self._writinghandles = (ifh, dfh)
 
         try:
             deltacomputer = deltautil.deltacomputer(self)
             # loop through our set of deltas
             for data in deltas:
                 node, p1, p2, linknode, deltabase, delta, flags = data
                 link = linkmapper(linknode)
                 flags = flags or REVIDX_DEFAULT_FLAGS
 
                 nodes.append(node)
 
                 if node in self.nodemap:
                     self._nodeduplicatecallback(transaction, node)
                     # this can happen if two branches make the same change
                     continue
 
                 for p in (p1, p2):
                     if p not in self.nodemap:
                         raise error.LookupError(p, self.indexfile,
                                                 _('unknown parent'))
 
                 if deltabase not in self.nodemap:
                     raise error.LookupError(deltabase, self.indexfile,
                                             _('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(">lll")
                     oldlen = self.rawsize(baserev)
                     newlen = len(delta) - hlen
                     if delta[:hlen] != mdiff.replacediffheader(oldlen, newlen):
                         raise error.CensoredBaseError(self.indexfile,
                                                       self.node(baserev))
 
                 if not flags and self._peek_iscensored(baserev, delta, flush):
                     flags |= REVIDX_ISCENSORED
 
                 # We assume consumers of addrevisioncb will want to retrieve
                 # the added revision, which will require a call to
                 # revision(). revision() will fast path if there is a cache
                 # hit. So, we tell _addrevision() to always cache in this case.
                 # We're only using addgroup() in the context of changegroup
                 # generation so the revision data can always be handled as raw
                 # by the flagprocessor.
                 self._addrevision(node, None, transaction, link,
                                   p1, p2, flags, (baserev, delta),
                                   ifh, dfh,
                                   alwayscache=bool(addrevisioncb),
                                   deltacomputer=deltacomputer)
 
                 if addrevisioncb:
                     addrevisioncb(self, node)
 
                 if not dfh and not self._inline:
                     # addrevision switched from inline to conventional
                     # reopen the index
                     ifh.close()
                     dfh = self._datafp("a+")
                     ifh = self._indexfp("a+")
                     self._writinghandles = (ifh, dfh)
         finally:
             self._writinghandles = None
 
             if dfh:
                 dfh.close()
             ifh.close()
 
         return nodes
 
     def iscensored(self, rev):
         """Check if a file revision is censored."""
         if not self._censorable:
             return False
 
         return self.flags(rev) & REVIDX_ISCENSORED
 
     def _peek_iscensored(self, baserev, delta, flush):
         """Quickly check if a delta produces a censored revision."""
         if not self._censorable:
             return False
 
         return storageutil.deltaiscensored(delta, baserev, self.rawsize)
 
     def getstrippoint(self, minlink):
         """find the minimum rev that must be stripped to strip the linkrev
 
         Returns a tuple containing the minimum rev and a set of all revs that
         have linkrevs that will be broken by this strip.
         """
         return storageutil.resolvestripinfo(minlink, len(self) - 1,
                                             self.headrevs(),
                                             self.linkrev, self.parentrevs)
 
     def strip(self, minlink, transaction):
         """truncate the revlog on the first revision with a linkrev >= minlink
 
         This function is called when we're stripping revision minlink and
         its descendants from the repository.
 
         We have to remove all revisions with linkrev >= minlink, because
         the equivalent changelog revisions will be renumbered after the
         strip.
 
         So we truncate the revlog on the first of these revisions, and
         trust that the caller has saved the revisions that shouldn't be
         removed and that it'll re-add them after this truncation.
         """
         if len(self) == 0:
             return
 
         rev, _ = self.getstrippoint(minlink)
         if rev == len(self):
             return
 
         # first truncate the files on disk
         end = self.start(rev)
         if not self._inline:
             transaction.add(self.datafile, end)
             end = rev * self._io.size
         else:
             end += rev * self._io.size
 
         transaction.add(self.indexfile, end)
 
         # then reset internal state in memory to forget those revisions
         self._revisioncache = None
         self._chaininfocache = {}
         self._chunkclear()
         for x in pycompat.xrange(rev, len(self)):
             del self.nodemap[self.node(x)]
 
         del self.index[rev:-1]
         self._nodepos = None
 
     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, 2)
                 actual = f.tell()
             dd = actual - expected
         except IOError as inst:
             if inst.errno != errno.ENOENT:
                 raise
             dd = 0
 
         try:
             f = self.opener(self.indexfile)
             f.seek(0, 2)
             actual = f.tell()
             f.close()
             s = self._io.size
             i = max(0, actual // s)
             di = actual - (i * s)
             if self._inline:
                 databytes = 0
                 for r in self:
                     databytes += max(0, self.length(r))
                 dd = 0
                 di = actual - len(self) * s - databytes
         except IOError as inst:
             if inst.errno != errno.ENOENT:
                 raise
             di = 0
 
         return (dd, di)
 
     def files(self):
         res = [self.indexfile]
         if not self._inline:
             res.append(self.datafile)
         return res
 
     def emitrevisions(self, nodes, nodesorder=None, revisiondata=False,
                       assumehaveparentrevisions=False,
                       deltamode=repository.CG_DELTAMODE_STD):
         if nodesorder not in ('nodes', 'storage', 'linear', None):
             raise error.ProgrammingError('unhandled value for nodesorder: %s' %
                                          nodesorder)
 
         if nodesorder is None and not self._generaldelta:
             nodesorder = '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)
 
     DELTAREUSEALWAYS = 'always'
     DELTAREUSESAMEREVS = 'samerevs'
     DELTAREUSENEVER = 'never'
 
     DELTAREUSEFULLADD = 'fulladd'
 
     DELTAREUSEALL = {'always', 'samerevs', 'never', 'fulladd'}
 
     def clone(self, tr, destrevlog, addrevisioncb=None,
               deltareuse=DELTAREUSESAMEREVS, forcedeltabothparents=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).
 
         Delta computation can be slow, so the choice of delta reuse policy can
         significantly affect run time.
 
         The default policy (``DELTAREUSESAMEREVS``) strikes a balance between
         two extremes. Deltas will be reused if they are appropriate. But if the
         delta could choose a better revision, it will do so. This means if you
         are converting a non-generaldelta revlog to a generaldelta revlog,
         deltas will be recomputed if the delta's parent isn't a parent of the
         revision.
 
         In addition to the delta policy, the ``forcedeltabothparents``
         argument controls whether to force compute deltas against both parents
         for merges. By default, the current default is used.
         """
         if deltareuse not in self.DELTAREUSEALL:
             raise ValueError(_('value for deltareuse invalid: %s') % deltareuse)
 
         if len(destrevlog):
             raise ValueError(_('destination revlog is not empty'))
 
         if getattr(self, 'filteredrevs', None):
             raise ValueError(_('source revlog has filtered revisions'))
         if getattr(destrevlog, 'filteredrevs', None):
             raise ValueError(_('destination revlog has filtered revisions'))
 
         # lazydelta and lazydeltabase controls whether to reuse a cached delta,
         # if possible.
         oldlazydelta = destrevlog._lazydelta
         oldlazydeltabase = destrevlog._lazydeltabase
         oldamd = destrevlog._deltabothparents
 
         try:
             if deltareuse == self.DELTAREUSEALWAYS:
                 destrevlog._lazydeltabase = True
                 destrevlog._lazydelta = True
             elif deltareuse == self.DELTAREUSESAMEREVS:
                 destrevlog._lazydeltabase = False
                 destrevlog._lazydelta = True
             elif deltareuse == self.DELTAREUSENEVER:
                 destrevlog._lazydeltabase = False
                 destrevlog._lazydelta = False
 
             destrevlog._deltabothparents = forcedeltabothparents or oldamd
 
             deltacomputer = deltautil.deltacomputer(destrevlog)
             index = self.index
             for rev in self:
                 entry = index[rev]
 
                 # Some classes override linkrev to take filtered revs into
                 # account. Use raw entry from index.
                 flags = entry[0] & 0xffff
                 linkrev = entry[4]
                 p1 = index[entry[5]][7]
                 p2 = index[entry[6]][7]
                 node = entry[7]
 
                 # (Possibly) reuse the delta from the revlog if allowed and
                 # the revlog chunk is a delta.
                 cachedelta = None
                 rawtext = None
                 if destrevlog._lazydelta:
                     dp = self.deltaparent(rev)
                     if dp != nullrev:
                         cachedelta = (dp, bytes(self._chunk(rev)))
 
                 if not cachedelta:
                     rawtext = self.revision(rev, raw=True)
 
 
                 if deltareuse == self.DELTAREUSEFULLADD:
                     destrevlog.addrevision(rawtext, tr, linkrev, p1, p2,
                                            cachedelta=cachedelta,
                                            node=node, flags=flags,
                                            deltacomputer=deltacomputer)
                 else:
                     ifh = destrevlog.opener(destrevlog.indexfile, 'a+',
                                             checkambig=False)
                     dfh = None
                     if not destrevlog._inline:
                         dfh = destrevlog.opener(destrevlog.datafile, 'a+')
                     try:
                         destrevlog._addrevision(node, rawtext, tr, linkrev, p1,
                                                 p2, flags, cachedelta, ifh, dfh,
                                                 deltacomputer=deltacomputer)
                     finally:
                         if dfh:
                             dfh.close()
                         ifh.close()
 
                 if addrevisioncb:
                     addrevisioncb(self, rev, node)
         finally:
             destrevlog._lazydelta = oldlazydelta
             destrevlog._lazydeltabase = oldlazydeltabase
             destrevlog._deltabothparents = oldamd
 
     def censorrevision(self, tr, censornode, tombstone=b''):
         if (self.version & 0xFFFF) == REVLOGV0:
             raise error.RevlogError(_('cannot censor with version %d revlogs') %
                                     self.version)
 
         censorrev = self.rev(censornode)
         tombstone = storageutil.packmeta({b'censored': tombstone}, b'')
 
         if len(tombstone) > self.rawsize(censorrev):
             raise error.Abort(_('censor tombstone must be no longer than '
                                 'censored data'))
 
         # Rewriting the revlog in place is hard. Our strategy for censoring is
         # to create a new revlog, copy all revisions to it, then replace the
         # revlogs on transaction close.
 
         newindexfile = self.indexfile + b'.tmpcensored'
         newdatafile = self.datafile + b'.tmpcensored'
 
         # This is a bit dangerous. We could easily have a mismatch of state.
         newrl = revlog(self.opener, newindexfile, newdatafile,
                        censorable=True)
         newrl.version = self.version
         newrl._generaldelta = self._generaldelta
         newrl._io = self._io
 
         for rev in self.revs():
             node = self.node(rev)
             p1, p2 = self.parents(node)
 
             if rev == censorrev:
                 newrl.addrawrevision(tombstone, tr, self.linkrev(censorrev),
                                      p1, p2, censornode, REVIDX_ISCENSORED)
 
                 if newrl.deltaparent(rev) != nullrev:
                     raise error.Abort(_('censored revision stored as delta; '
                                         'cannot censor'),
                                       hint=_('censoring of revlogs is not '
                                              'fully implemented; please report '
                                              'this bug'))
                 continue
 
             if self.iscensored(rev):
                 if self.deltaparent(rev) != nullrev:
                     raise error.Abort(_('cannot censor due to censored '
                                         'revision having delta stored'))
                 rawtext = self._chunk(rev)
             else:
                 rawtext = self.revision(rev, raw=True)
 
             newrl.addrawrevision(rawtext, tr, self.linkrev(rev), p1, p2, node,
                                  self.flags(rev))
 
         tr.addbackup(self.indexfile, location='store')
         if not self._inline:
             tr.addbackup(self.datafile, location='store')
 
         self.opener.rename(newrl.indexfile, self.indexfile)
         if not self._inline:
             self.opener.rename(newrl.datafile, self.datafile)
 
         self.clearcaches()
         self._loadindex()
 
     def verifyintegrity(self, state):
         """Verifies the integrity of the revlog.
 
         Yields ``revlogproblem`` instances describing problems that are
         found.
         """
         dd, di = self.checksize()
         if dd:
             yield revlogproblem(error=_('data length off by %d bytes') % dd)
         if di:
             yield revlogproblem(error=_('index contains %d extra bytes') % di)
 
         version = self.version & 0xFFFF
 
         # The verifier tells us what version revlog we should be.
         if version != state['expectedversion']:
             yield revlogproblem(
                 warning=_("warning: '%s' uses revlog format %d; expected %d") %
                         (self.indexfile, version, state['expectedversion']))
 
         state['skipread'] = 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 "revision(rev, raw=True)". "text"
             # mentioned below is "revision(rev, raw=False)".
             #
             # 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 revlog.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('skipflags', 0)
                 if skipflags:
                     skipflags &= self.flags(rev)
 
                 if skipflags:
                     state['skipread'].add(node)
                 else:
                     # Side-effect: read content and verify hash.
                     self.revision(node)
 
                 l1 = self.rawsize(rev)
                 l2 = len(self.revision(node, raw=True))
 
                 if l1 != l2:
                     yield revlogproblem(
                         error=_('unpacked size is %d, %d expected') % (l2, l1),
                         node=node)
 
             except error.CensoredNodeError:
                 if state['erroroncensored']:
                     yield revlogproblem(error=_('censored file data'),
                                         node=node)
                     state['skipread'].add(node)
             except Exception as e:
                 yield revlogproblem(
                     error=_('unpacking %s: %s') % (short(node),
                                                    stringutil.forcebytestr(e)),
                     node=node)
                 state['skipread'].add(node)
 
     def storageinfo(self, exclusivefiles=False, sharedfiles=False,
                     revisionscount=False, trackedsize=False,
                     storedsize=False):
         d = {}
 
         if exclusivefiles:
             d['exclusivefiles'] = [(self.opener, self.indexfile)]
             if not self._inline:
                 d['exclusivefiles'].append((self.opener, self.datafile))
 
         if sharedfiles:
             d['sharedfiles'] = []
 
         if revisionscount:
             d['revisionscount'] = len(self)
 
         if trackedsize:
             d['trackedsize'] = sum(map(self.rawsize, iter(self)))
 
         if storedsize:
             d['storedsize'] = sum(self.opener.stat(path).st_size
                                   for path in self.files())
 
         return d