patch: unify check_binary and binary flags

Follows up 079b27b5a869. If opts.text=True, check_binary is ignored, so we
can just pass the binary flag to unidiff().

perfunidiff now takes any inputs as text files, which I think is a desired
behavior.

contrib/perf.py

@@ -1,1730 +1,1730 b''
 # perf.py - performance test routines
 '''helper extension to measure performance'''
 
 # "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 functools
 import gc
 import os
 import random
 import struct
 import sys
 import threading
 import time
 from mercurial import (
     changegroup,
     cmdutil,
     commands,
     copies,
     error,
     extensions,
     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 import scmutil # since 1.9 (or 8b252e826c68)
 except ImportError:
     pass
 
 # 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, 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 == '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", [
         ('c', 'changelog', False, ('open changelog')),
         ('m', 'manifest', False, ('open manifest')),
         ('', 'dir', False, ('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.lstrip("^").split("|")
 
 if safehasattr(registrar, 'command'):
     command = registrar.command(cmdtable)
 elif safehasattr(cmdutil, 'command'):
     import inspect
     command = cmdutil.command(cmdtable)
     if 'norepo' not in inspect.getargspec(command)[0]:
         # 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 += ' %s' % ' '.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 += ' %s' % ' '.join(parsealiases(name))
             return func
         return decorator
 
 try:
     import mercurial.registrar
     import mercurial.configitems
     configtable = {}
     configitem = mercurial.registrar.configitem(configtable)
     configitem('perf', 'presleep',
         default=mercurial.configitems.dynamicdefault,
     )
     configitem('perf', 'stub',
         default=mercurial.configitems.dynamicdefault,
     )
     configitem('perf', 'parentscount',
         default=mercurial.configitems.dynamicdefault,
     )
 except (ImportError, AttributeError):
     pass
 
 def getlen(ui):
     if ui.configbool("perf", "stub", False):
         return lambda x: 1
     return len
 
 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, "perf", "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, '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('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, 'perf', opts)
 
     # stub function, runs code only once instead of in a loop
     # experimental config: perf.stub
     if ui.configbool("perf", "stub", False):
         return functools.partial(stub_timer, fm), fm
     return functools.partial(_timer, fm), fm
 
 def stub_timer(fm, func, title=None):
     func()
 
 def _timer(fm, func, title=None):
     gc.collect()
     results = []
     begin = util.timer()
     count = 0
     while True:
         ostart = os.times()
         cstart = util.timer()
         r = func()
         cstop = util.timer()
         ostop = os.times()
         count += 1
         a, b = ostart, ostop
         results.append((cstop - cstart, b[0] - a[0], b[1]-a[1]))
         if cstop - begin > 3 and count >= 100:
             break
         if cstop - begin > 10 and count >= 3:
             break
 
     fm.startitem()
 
     if title:
         fm.write('title', '! %s\n', title)
     if r:
         fm.write('result', '! result: %s\n', r)
     m = min(results)
     fm.plain('!')
     fm.write('wall', ' wall %f', m[0])
     fm.write('comb', ' comb %f', m[1] + m[2])
     fm.write('user', ' user %f', m[1])
     fm.write('sys',  ' sys %f', m[2])
     fm.write('count',  ' (best of %d)', count)
     fm.plain('\n')
 
 # 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(("%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(("missing attribute %s of %s might break assumption"
                            " of performance measurement") % (name, obj))
 
     origvalue = getattr(obj, name)
     class attrutil(object):
         def set(self, newvalue):
             setattr(obj, name, newvalue)
         def restore(self):
             setattr(obj, 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)
     for mod in (branchmap, repoview):
         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(("perfbranchmap not available with this Mercurial"),
                       hint="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, '_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 '_tagscache' in vars(repo):
                 del repo.__dict__['_tagscache']
         return clearcache
 
     repotags = safeattrsetter(repo, '_tags', ignoremissing=True)
     if repotags: # since 1.4 (or 5614a628d173)
         return lambda : repotags.set(None)
 
     repotagscache = safeattrsetter(repo, '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(("tags API of this hg command is unknown"))
 
 # utilities to clear cache
 
 def clearfilecache(repo, attrname):
     unfi = repo.unfiltered()
     if attrname in vars(unfi):
         delattr(unfi, attrname)
     unfi._filecache.pop(attrname, None)
 
 # perf commands
 
 @command('perfwalk', formatteropts)
 def perfwalk(ui, repo, *pats, **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('perfannotate', formatteropts)
 def perfannotate(ui, repo, f, **opts):
     timer, fm = gettimer(ui, opts)
     fc = repo['.'][f]
     timer(lambda: len(fc.annotate(True)))
     fm.end()
 
 @command('perfstatus',
          [('u', 'unknown', False,
            'ask status to look for unknown files')] + formatteropts)
 def perfstatus(ui, repo, **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['unknown']))))
     fm.end()
 
 @command('perfaddremove', formatteropts)
 def perfaddremove(ui, repo, **opts):
     timer, fm = gettimer(ui, opts)
     try:
         oldquiet = repo.ui.quiet
         repo.ui.quiet = True
         matcher = scmutil.match(repo[None])
         timer(lambda: scmutil.addremove(repo, matcher, "", dry_run=True))
     finally:
         repo.ui.quiet = oldquiet
         fm.end()
 
 def clearcaches(cl):
     # behave somewhat consistently across internal API changes
     if util.safehasattr(cl, 'clearcaches'):
         cl.clearcaches()
     elif util.safehasattr(cl, '_nodecache'):
         from mercurial.node import nullid, nullrev
         cl._nodecache = {nullid: nullrev}
         cl._nodepos = None
 
 @command('perfheads', formatteropts)
 def perfheads(ui, repo, **opts):
     timer, fm = gettimer(ui, opts)
     cl = repo.changelog
     def d():
         len(cl.headrevs())
         clearcaches(cl)
     timer(d)
     fm.end()
 
 @command('perftags', formatteropts)
 def perftags(ui, repo, **opts):
     import mercurial.changelog
     import mercurial.manifest
     timer, fm = gettimer(ui, opts)
     svfs = getsvfs(repo)
     repocleartagscache = repocleartagscachefunc(repo)
     def t():
         repo.changelog = mercurial.changelog.changelog(svfs)
         repo.manifestlog = mercurial.manifest.manifestlog(svfs, repo)
         repocleartagscache()
         return len(repo.tags())
     timer(t)
     fm.end()
 
 @command('perfancestors', formatteropts)
 def perfancestors(ui, repo, **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('perfancestorset', formatteropts)
 def perfancestorset(ui, repo, revset, **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('perfbookmarks', formatteropts)
 def perfbookmarks(ui, repo, **opts):
     """benchmark parsing bookmarks from disk to memory"""
     timer, fm = gettimer(ui, opts)
     def d():
         clearfilecache(repo, '_bookmarks')
         repo._bookmarks
     timer(d)
     fm.end()
 
 @command('perfbundleread', formatteropts, '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,
     )
 
     def makebench(fn):
         def run():
             with open(bundlepath, 'rb') as fh:
                 bundle = exchange.readbundle(ui, fh, bundlepath)
                 fn(bundle)
 
         return run
 
     def makereadnbytes(size):
         def run():
             with open(bundlepath, 'rb') as fh:
                 bundle = exchange.readbundle(ui, fh, bundlepath)
                 while bundle.read(size):
                     pass
 
         return run
 
     def makestdioread(size):
         def run():
             with open(bundlepath, '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, 'rb') as fh:
                 bundle = exchange.readbundle(ui, fh, bundlepath)
                 for part in bundle.iterparts():
                     while part.read(size):
                         pass
 
         return run
 
     benches = [
         (makestdioread(8192), 'read(8k)'),
         (makestdioread(16384), 'read(16k)'),
         (makestdioread(32768), 'read(32k)'),
         (makestdioread(131072), 'read(128k)'),
     ]
 
     with open(bundlepath, 'rb') as fh:
         bundle = exchange.readbundle(ui, fh, bundlepath)
 
         if isinstance(bundle, changegroup.cg1unpacker):
             benches.extend([
                 (makebench(deltaiter), 'cg1 deltaiter()'),
                 (makebench(iterchunks), 'cg1 getchunks()'),
                 (makereadnbytes(8192), 'cg1 read(8k)'),
                 (makereadnbytes(16384), 'cg1 read(16k)'),
                 (makereadnbytes(32768), 'cg1 read(32k)'),
                 (makereadnbytes(131072), 'cg1 read(128k)'),
             ])
         elif isinstance(bundle, bundle2.unbundle20):
             benches.extend([
                 (makebench(forwardchunks), 'bundle2 forwardchunks()'),
                 (makebench(iterparts), 'bundle2 iterparts()'),
                 (makebench(iterpartsseekable), 'bundle2 iterparts() seekable'),
                 (makebench(seek), 'bundle2 part seek()'),
                 (makepartreadnbytes(8192), 'bundle2 part read(8k)'),
                 (makepartreadnbytes(16384), 'bundle2 part read(16k)'),
                 (makepartreadnbytes(32768), 'bundle2 part read(32k)'),
                 (makepartreadnbytes(131072), 'bundle2 part read(128k)'),
             ])
         elif isinstance(bundle, streamclone.streamcloneapplier):
             raise error.Abort('stream clone bundles not supported')
         else:
             raise error.Abort('unhandled bundle type: %s' % type(bundle))
 
     for fn, title in benches:
         timer, fm = gettimer(ui, opts)
         timer(fn, title=title)
         fm.end()
 
 @command('perfchangegroupchangelog', formatteropts +
          [('', 'version', '02', 'changegroup version'),
           ('r', 'rev', '', 'revisions to add to changegroup')])
 def perfchangegroupchangelog(ui, repo, version='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.
     """
     cl = repo.changelog
     revs = [cl.lookup(r) for r in repo.revs(rev or 'all()')]
     bundler = changegroup.getbundler(version, repo)
 
     def lookup(node):
         # The real bundler reads the revision in order to access the
         # manifest node and files list. Do that here.
         cl.read(node)
         return node
 
     def d():
         for chunk in bundler.group(revs, cl, lookup):
             pass
 
     timer, fm = gettimer(ui, opts)
     timer(d)
     fm.end()
 
 @command('perfdirs', formatteropts)
 def perfdirs(ui, repo, **opts):
     timer, fm = gettimer(ui, opts)
     dirstate = repo.dirstate
     'a' in dirstate
     def d():
         dirstate.hasdir('a')
         del dirstate._map._dirs
     timer(d)
     fm.end()
 
 @command('perfdirstate', formatteropts)
 def perfdirstate(ui, repo, **opts):
     timer, fm = gettimer(ui, opts)
     "a" in repo.dirstate
     def d():
         repo.dirstate.invalidate()
         "a" in repo.dirstate
     timer(d)
     fm.end()
 
 @command('perfdirstatedirs', formatteropts)
 def perfdirstatedirs(ui, repo, **opts):
     timer, fm = gettimer(ui, opts)
     "a" in repo.dirstate
     def d():
         repo.dirstate.hasdir("a")
         del repo.dirstate._map._dirs
     timer(d)
     fm.end()
 
 @command('perfdirstatefoldmap', formatteropts)
 def perfdirstatefoldmap(ui, repo, **opts):
     timer, fm = gettimer(ui, opts)
     dirstate = repo.dirstate
     'a' in dirstate
     def d():
         dirstate._map.filefoldmap.get('a')
         del dirstate._map.filefoldmap
     timer(d)
     fm.end()
 
 @command('perfdirfoldmap', formatteropts)
 def perfdirfoldmap(ui, repo, **opts):
     timer, fm = gettimer(ui, opts)
     dirstate = repo.dirstate
     'a' in dirstate
     def d():
         dirstate._map.dirfoldmap.get('a')
         del dirstate._map.dirfoldmap
         del dirstate._map._dirs
     timer(d)
     fm.end()
 
 @command('perfdirstatewrite', formatteropts)
 def perfdirstatewrite(ui, repo, **opts):
     timer, fm = gettimer(ui, opts)
     ds = repo.dirstate
     "a" in ds
     def d():
         ds._dirty = True
         ds.write(repo.currenttransaction())
     timer(d)
     fm.end()
 
 @command('perfmergecalculate',
          [('r', 'rev', '.', 'rev to merge against')] + formatteropts)
 def perfmergecalculate(ui, repo, rev, **opts):
     timer, fm = gettimer(ui, opts)
     wctx = repo[None]
     rctx = scmutil.revsingle(repo, rev, rev)
     ancestor = wctx.ancestor(rctx)
     # we don't want working dir files to be stat'd in the benchmark, so prime
     # that cache
     wctx.dirty()
     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('perfpathcopies', [], "REV REV")
 def perfpathcopies(ui, repo, rev1, rev2, **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('perfphases',
          [('', 'full', False, 'include file reading time too'),
          ], "")
 def perfphases(ui, repo, **opts):
     """benchmark phasesets computation"""
     timer, fm = gettimer(ui, opts)
     _phases = repo._phasecache
     full = opts.get('full')
     def d():
         phases = _phases
         if full:
             clearfilecache(repo, '_phasecache')
             phases = repo._phasecache
         phases.invalidate()
         phases.loadphaserevs(repo)
     timer(d)
     fm.end()
 
 @command('perfmanifest', [], 'REV')
 def perfmanifest(ui, repo, rev, **opts):
     timer, fm = gettimer(ui, opts)
     ctx = scmutil.revsingle(repo, rev, rev)
     t = ctx.manifestnode()
     def d():
         repo.manifestlog.clearcaches()
         repo.manifestlog[t].read()
     timer(d)
     fm.end()
 
 @command('perfchangeset', formatteropts)
 def perfchangeset(ui, repo, rev, **opts):
     timer, fm = gettimer(ui, opts)
     n = repo[rev].node()
     def d():
         repo.changelog.read(n)
         #repo.changelog._cache = None
     timer(d)
     fm.end()
 
 @command('perfindex', formatteropts)
 def perfindex(ui, repo, **opts):
     import mercurial.revlog
     timer, fm = gettimer(ui, opts)
     mercurial.revlog._prereadsize = 2**24 # disable lazy parser in old hg
     n = repo["tip"].node()
     svfs = getsvfs(repo)
     def d():
         cl = mercurial.revlog.revlog(svfs, "00changelog.i")
         cl.rev(n)
     timer(d)
     fm.end()
 
 @command('perfstartup', formatteropts)
 def perfstartup(ui, repo, **opts):
     timer, fm = gettimer(ui, opts)
     cmd = sys.argv[0]
     def d():
         if os.name != 'nt':
             os.system("HGRCPATH= %s version -q > /dev/null" % cmd)
         else:
             os.environ['HGRCPATH'] = ' '
             os.system("%s version -q > NUL" % cmd)
     timer(d)
     fm.end()
 
 @command('perfparents', formatteropts)
 def perfparents(ui, repo, **opts):
     timer, fm = gettimer(ui, opts)
     # control the number of commits perfparents iterates over
     # experimental config: perf.parentscount
     count = getint(ui, "perf", "parentscount", 1000)
     if len(repo.changelog) < count:
         raise error.Abort("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('perfctxfiles', formatteropts)
 def perfctxfiles(ui, repo, x, **opts):
     x = int(x)
     timer, fm = gettimer(ui, opts)
     def d():
         len(repo[x].files())
     timer(d)
     fm.end()
 
 @command('perfrawfiles', formatteropts)
 def perfrawfiles(ui, repo, x, **opts):
     x = int(x)
     timer, fm = gettimer(ui, opts)
     cl = repo.changelog
     def d():
         len(cl.read(x)[3])
     timer(d)
     fm.end()
 
 @command('perflookup', formatteropts)
 def perflookup(ui, repo, rev, **opts):
     timer, fm = gettimer(ui, opts)
     timer(lambda: len(repo.lookup(rev)))
     fm.end()
 
 @command('perfrevrange', formatteropts)
 def perfrevrange(ui, repo, *specs, **opts):
     timer, fm = gettimer(ui, opts)
     revrange = scmutil.revrange
     timer(lambda: len(revrange(repo, specs)))
     fm.end()
 
 @command('perfnodelookup', formatteropts)
 def perfnodelookup(ui, repo, rev, **opts):
     timer, fm = gettimer(ui, opts)
     import mercurial.revlog
     mercurial.revlog._prereadsize = 2**24 # disable lazy parser in old hg
     n = repo[rev].node()
     cl = mercurial.revlog.revlog(getsvfs(repo), "00changelog.i")
     def d():
         cl.rev(n)
         clearcaches(cl)
     timer(d)
     fm.end()
 
 @command('perflog',
          [('', 'rename', False, 'ask log to follow renames')] + formatteropts)
 def perflog(ui, repo, rev=None, **opts):
     if rev is None:
         rev=[]
     timer, fm = gettimer(ui, opts)
     ui.pushbuffer()
     timer(lambda: commands.log(ui, repo, rev=rev, date='', user='',
                                copies=opts.get('rename')))
     ui.popbuffer()
     fm.end()
 
 @command('perfmoonwalk', formatteropts)
 def perfmoonwalk(ui, repo, **opts):
     """benchmark walking the changelog backwards
 
     This also loads the changelog data for each revision in the changelog.
     """
     timer, fm = gettimer(ui, opts)
     def moonwalk():
         for i in xrange(len(repo), -1, -1):
             ctx = repo[i]
             ctx.branch() # read changelog data (in addition to the index)
     timer(moonwalk)
     fm.end()
 
 @command('perftemplating', formatteropts)
 def perftemplating(ui, repo, rev=None, **opts):
     if rev is None:
         rev=[]
     timer, fm = gettimer(ui, opts)
     ui.pushbuffer()
     timer(lambda: commands.log(ui, repo, rev=rev, date='', user='',
                                template='{date|shortdate} [{rev}:{node|short}]'
                                ' {author|person}: {desc|firstline}\n'))
     ui.popbuffer()
     fm.end()
 
 @command('perfcca', formatteropts)
 def perfcca(ui, repo, **opts):
     timer, fm = gettimer(ui, opts)
     timer(lambda: scmutil.casecollisionauditor(ui, False, repo.dirstate))
     fm.end()
 
 @command('perffncacheload', formatteropts)
 def perffncacheload(ui, repo, **opts):
     timer, fm = gettimer(ui, opts)
     s = repo.store
     def d():
         s.fncache._load()
     timer(d)
     fm.end()
 
 @command('perffncachewrite', formatteropts)
 def perffncachewrite(ui, repo, **opts):
     timer, fm = gettimer(ui, opts)
     s = repo.store
     s.fncache._load()
     lock = repo.lock()
     tr = repo.transaction('perffncachewrite')
     def d():
         s.fncache._dirty = True
         s.fncache.write(tr)
     timer(d)
     tr.close()
     lock.release()
     fm.end()
 
 @command('perffncacheencode', formatteropts)
 def perffncacheencode(ui, repo, **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, ready, done):
     while not done.is_set():
         pair = q.get()
         while pair is not None:
             mdiff.textdiff(*pair)
             q.task_done()
             pair = q.get()
         q.task_done() # for the None one
         with ready:
             ready.wait()
 
 @command('perfbdiff', revlogopts + formatteropts + [
     ('', 'count', 1, 'number of revisions to test (when using --startrev)'),
     ('', 'alldata', False, 'test bdiffs for all associated revisions'),
     ('', 'threads', 0, 'number of thread to use (disable with 0)'),
     ],
 
     '-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).
     """
     if opts['alldata']:
         opts['changelog'] = True
 
     if opts.get('changelog') or opts.get('manifest'):
         file_, rev = None, file_
     elif rev is None:
         raise error.CommandError('perfbdiff', 'invalid arguments')
 
     textpairs = []
 
     r = cmdutil.openrevlog(repo, 'perfbdiff', file_, opts)
 
     startrev = r.rev(r.lookup(rev))
     for rev in range(startrev, min(startrev + count, len(r) - 1)):
         if opts['alldata']:
             # Load revisions associated with changeset.
             ctx = repo[rev]
             mtext = repo.manifestlog._revlog.revision(ctx.manifestnode())
             for pctx in ctx.parents():
                 pman = repo.manifestlog._revlog.revision(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:
                 mdiff.textdiff(*pair)
     else:
         q = util.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, 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('perfunidiff', revlogopts + formatteropts + [
     ('', 'count', 1, 'number of revisions to test (when using --startrev)'),
     ('', 'alldata', False, 'test unidiffs for all associated revisions'),
     ], '-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).
     """
     if opts['alldata']:
         opts['changelog'] = True
 
     if opts.get('changelog') or opts.get('manifest'):
         file_, rev = None, file_
     elif rev is None:
         raise error.CommandError('perfunidiff', 'invalid arguments')
 
     textpairs = []
 
     r = cmdutil.openrevlog(repo, 'perfunidiff', file_, opts)
 
     startrev = r.rev(r.lookup(rev))
     for rev in range(startrev, min(startrev + count, len(r) - 1)):
         if opts['alldata']:
             # Load revisions associated with changeset.
             ctx = repo[rev]
             mtext = repo.manifestlog._revlog.revision(ctx.manifestnode())
             for pctx in ctx.parents():
                 pman = repo.manifestlog._revlog.revision(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, '', right, '', 'left', 'right')
+                left, '', right, '', 'left', '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('perfdiffwd', formatteropts)
 def perfdiffwd(ui, repo, **opts):
     """Profile diff of working directory changes"""
     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], '1') for c in diffopt)
         def d():
             ui.pushbuffer()
             commands.diff(ui, repo, **opts)
             ui.popbuffer()
         title = 'diffopts: %s' % (diffopt and ('-' + diffopt) or 'none')
         timer(d, title)
     fm.end()
 
 @command('perfrevlogindex', revlogopts + formatteropts,
          '-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.
     """
 
     rl = cmdutil.openrevlog(repo, 'perfrevlogindex', file_, opts)
 
     opener = getattr(rl, 'opener')  # trick linter
     indexfile = rl.indexfile
     data = opener.read(indexfile)
 
     header = struct.unpack('>I', data[0:4])[0]
     version = header & 0xFFFF
     if version == 1:
         revlogio = revlog.revlogio()
         inline = header & (1 << 16)
     else:
         raise error.Abort(('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, 'revlog constructor'),
         (read, 'read'),
         (parseindex, 'create index object'),
         (lambda: getentry(0), 'retrieve index entry for rev 0'),
         (lambda: resolvenode('a' * 20), 'look up missing node'),
         (lambda: resolvenode(node0), 'look up node at rev 0'),
         (lambda: resolvenode(node25), 'look up node at 1/4 len'),
         (lambda: resolvenode(node50), 'look up node at 1/2 len'),
         (lambda: resolvenode(node75), 'look up node at 3/4 len'),
         (lambda: resolvenode(node100), 'look up node at tip'),
         # 2x variation is to measure caching impact.
         (lambda: resolvenodes(allnodes),
          'look up all nodes (forward)'),
         (lambda: resolvenodes(allnodes, 2),
          'look up all nodes 2x (forward)'),
         (lambda: resolvenodes(allnodesrev),
          'look up all nodes (reverse)'),
         (lambda: resolvenodes(allnodesrev, 2),
          'look up all nodes 2x (reverse)'),
         (lambda: getentries(allrevs),
          'retrieve all index entries (forward)'),
         (lambda: getentries(allrevs, 2),
          'retrieve all index entries 2x (forward)'),
         (lambda: getentries(allrevsrev),
          'retrieve all index entries (reverse)'),
         (lambda: getentries(allrevsrev, 2),
          'retrieve all index entries 2x (reverse)'),
     ]
 
     for fn, title in benches:
         timer, fm = gettimer(ui, opts)
         timer(fn, title=title)
         fm.end()
 
 @command('perfrevlogrevisions', revlogopts + formatteropts +
          [('d', 'dist', 100, 'distance between the revisions'),
           ('s', 'startrev', 0, 'revision to start reading at'),
           ('', 'reverse', False, 'read in reverse')],
          '-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``.
     """
     rl = cmdutil.openrevlog(repo, 'perfrevlogrevisions', file_, opts)
     rllen = getlen(ui)(rl)
 
     def d():
         rl.clearcaches()
 
         beginrev = startrev
         endrev = rllen
         dist = opts['dist']
 
         if reverse:
             beginrev, endrev = endrev, beginrev
             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('perfrevlogchunks', revlogopts + formatteropts +
          [('e', 'engines', '', 'compression engines to use'),
           ('s', 'startrev', 0, 'revision to start at')],
          '-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``.
     """
     rl = cmdutil.openrevlog(repo, '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(','))
         for engine in engines:
             try:
                 util.compressionengines[engine]
             except KeyError:
                 raise error.Abort('unknown compression engine: %s' % engine)
     else:
         engines = []
         for e in util.compengines:
             engine = util.compengines[e]
             try:
                 if engine.available():
                     engine.revlogcompressor().compress('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(), 'read'),
         (lambda: doreadcachedfh(), 'read w/ reused fd'),
         (lambda: doreadbatch(), 'read batch'),
         (lambda: doreadbatchcachedfh(), 'read batch w/ reused fd'),
         (lambda: dochunk(), 'chunk'),
         (lambda: dochunkbatch(), 'chunk batch'),
     ]
 
     for engine in sorted(engines):
         compressor = util.compengines[engine].revlogcompressor()
         benches.append((functools.partial(docompress, compressor),
                         'compress w/ %s' % engine))
 
     for fn, title in benches:
         timer, fm = gettimer(ui, opts)
         timer(fn, title=title)
         fm.end()
 
 @command('perfrevlogrevision', revlogopts + formatteropts +
          [('', 'cache', False, 'use caches instead of clearing')],
          '-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. Obtain the raw chunks for that delta chain
     3. Decompress each raw chunk
     4. Apply binary patches to obtain fulltext
     5. Verify hash of fulltext
 
     This command measures the time spent in each of these phases.
     """
     if opts.get('changelog') or opts.get('manifest'):
         file_, rev = None, file_
     elif rev is None:
         raise error.CommandError('perfrevlogrevision', 'invalid arguments')
 
     r = cmdutil.openrevlog(repo, '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
         offset = start(chain[0])
 
         chunks = []
         ladd = chunks.append
 
         for rev in chain:
             chunkstart = start(rev)
             if inline:
                 chunkstart += (rev + 1) * iosize
             chunklength = length(rev)
             ladd(buffer(data, chunkstart - offset, chunklength))
 
         return chunks
 
     def dodeltachain(rev):
         if not cache:
             r.clearcaches()
         r._deltachain(rev)
 
     def doread(chain):
         if not cache:
             r.clearcaches()
         segmentforrevs(chain[0], chain[-1])
 
     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)
 
     chain = r._deltachain(rev)[0]
     data = segmentforrevs(chain[0], chain[-1])[1]
     rawchunks = getrawchunks(data, chain)
     bins = r._chunks(chain)
     text = str(bins[0])
     bins = bins[1:]
     text = mdiff.patches(text, bins)
 
     benches = [
         (lambda: dorevision(), 'full'),
         (lambda: dodeltachain(rev), 'deltachain'),
         (lambda: doread(chain), 'read'),
         (lambda: dorawchunks(data, chain), 'rawchunks'),
         (lambda: dodecompress(rawchunks), 'decompress'),
         (lambda: dopatch(text, bins), 'patch'),
         (lambda: dohash(text), 'hash'),
     ]
 
     for fn, title in benches:
         timer, fm = gettimer(ui, opts)
         timer(fn, title=title)
         fm.end()
 
 @command('perfrevset',
          [('C', 'clear', False, 'clear volatile cache between each call.'),
           ('', 'contexts', False, 'obtain changectx for each revision')]
          + formatteropts, "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."""
     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('perfvolatilesets',
          [('', 'clear-obsstore', False, '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."""
     timer, fm = gettimer(ui, opts)
     repo = repo.unfiltered()
 
     def getobs(name):
         def d():
             repo.invalidatevolatilesets()
             if opts['clear_obsstore']:
                 clearfilecache(repo, '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['clear_obsstore']:
                 clearfilecache(repo, '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('perfbranchmap',
          [('f', 'full', False,
            'Includes build time of subset'),
           ('', 'clear-revbranch', False,
            'purge the revbranch cache between computation'),
          ] + formatteropts)
 def perfbranchmap(ui, repo, full=False, clear_revbranch=False, **opts):
     """benchmark the update of a branchmap
 
     This benchmarks the full repo.branchmap() call with read and write disabled
     """
     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)
         def d():
             if clear_revbranch:
                 repo.revbranchcache()._clear()
             if full:
                 view._branchcaches.clear()
             else:
                 view._branchcaches.pop(filtername, None)
             view.branchmap()
         return d
     # add filter in smaller subset to bigger subset
     possiblefilters = set(repoview.filtertable)
     subsettable = getbranchmapsubsettable()
     allfilters = []
     while possiblefilters:
         for name in possiblefilters:
             subset = subsettable.get(name)
             if subset not in possiblefilters:
                 break
         else:
             assert False, 'subset cycle %s!' % possiblefilters
         allfilters.append(name)
         possiblefilters.remove(name)
 
     # warm the cache
     if not full:
         for name in allfilters:
             repo.filtered(name).branchmap()
     # add unfiltered
     allfilters.append(None)
 
     branchcacheread = safeattrsetter(branchmap, 'read')
     branchcachewrite = safeattrsetter(branchmap.branchcache, 'write')
     branchcacheread.set(lambda repo: None)
     branchcachewrite.set(lambda bc, repo: None)
     try:
         for name in allfilters:
             timer(getbranchmap(name), title=str(name))
     finally:
         branchcacheread.restore()
         branchcachewrite.restore()
     fm.end()
 
 @command('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('perflrucachedict', formatteropts +
     [('', 'size', 4, 'size of cache'),
      ('', 'gets', 10000, 'number of key lookups'),
      ('', 'sets', 10000, 'number of key sets'),
      ('', 'mixed', 10000, 'number of mixed mode operations'),
      ('', 'mixedgetfreq', 50, 'frequency of get vs set ops in mixed mode')],
     norepo=True)
 def perflrucache(ui, size=4, gets=10000, sets=10000, mixed=10000,
                  mixedgetfreq=50, **opts):
     def doinit():
         for i in xrange(10000):
             util.lrucachedict(size)
 
     values = []
     for i in xrange(size):
         values.append(random.randint(0, sys.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
 
     # Set mode tests insertion speed with cache eviction.
     setseq = []
     for i in xrange(sets):
         setseq.append(random.randint(0, sys.maxint))
 
     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)))
 
     def domixed():
         d = util.lrucachedict(size)
 
         for op, v in mixedops:
             if op == 0:
                 try:
                     d[v]
                 except KeyError:
                     pass
             else:
                 d[v] = v
 
     benches = [
         (doinit, 'init'),
         (dogets, 'gets'),
         (dosets, 'sets'),
         (domixed, 'mixed')
     ]
 
     for fn, title in benches:
         timer, fm = gettimer(ui, opts)
         timer(fn, title=title)
         fm.end()
 
 @command('perfwrite', formatteropts)
 def perfwrite(ui, repo, **opts):
     """microbenchmark ui.write
     """
     timer, fm = gettimer(ui, opts)
     def write():
         for i in range(100000):
             ui.write(('Testing write performance\n'))
     timer(write)
     fm.end()
 
 def uisetup(ui):
     if (util.safehasattr(cmdutil, 'openrevlog') and
         not util.safehasattr(commands, '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('dir') and not util.safehasattr(repo, 'dirlog'):
                 raise error.Abort("This version doesn't support --dir option",
                                   hint="use 3.5 or later")
             return orig(repo, cmd, file_, opts)
         extensions.wrapfunction(cmdutil, 'openrevlog', openrevlog)

mercurial/mdiff.py

@@ -1,522 +1,521 b''
 # mdiff.py - diff and patch routines for mercurial
 #
 # Copyright 2005, 2006 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 re
 import struct
 import zlib
 
 from .i18n import _
 from . import (
     error,
     policy,
     pycompat,
     util,
 )
 
 _missing_newline_marker = "\\ No newline at end of file\n"
 
 bdiff = policy.importmod(r'bdiff')
 mpatch = policy.importmod(r'mpatch')
 
 blocks = bdiff.blocks
 fixws = bdiff.fixws
 patches = mpatch.patches
 patchedsize = mpatch.patchedsize
 textdiff = bdiff.bdiff
 
 def splitnewlines(text):
     '''like str.splitlines, but only split on newlines.'''
     lines = [l + '\n' for l in text.split('\n')]
     if lines:
         if lines[-1] == '\n':
             lines.pop()
         else:
             lines[-1] = lines[-1][:-1]
     return lines
 
 class diffopts(object):
     '''context is the number of context lines
     text treats all files as text
     showfunc enables diff -p output
     git enables the git extended patch format
     nodates removes dates from diff headers
     nobinary ignores binary files
     noprefix disables the 'a/' and 'b/' prefixes (ignored in plain mode)
     ignorews ignores all whitespace changes in the diff
     ignorewsamount ignores changes in the amount of whitespace
     ignoreblanklines ignores changes whose lines are all blank
     upgrade generates git diffs to avoid data loss
     '''
 
     defaults = {
         'context': 3,
         'text': False,
         'showfunc': False,
         'git': False,
         'nodates': False,
         'nobinary': False,
         'noprefix': False,
         'index': 0,
         'ignorews': False,
         'ignorewsamount': False,
         'ignorewseol': False,
         'ignoreblanklines': False,
         'upgrade': False,
         'showsimilarity': False,
         'worddiff': False,
         }
 
     def __init__(self, **opts):
         opts = pycompat.byteskwargs(opts)
         for k in self.defaults.keys():
             v = opts.get(k)
             if v is None:
                 v = self.defaults[k]
             setattr(self, k, v)
 
         try:
             self.context = int(self.context)
         except ValueError:
             raise error.Abort(_('diff context lines count must be '
                                'an integer, not %r') % self.context)
 
     def copy(self, **kwargs):
         opts = dict((k, getattr(self, k)) for k in self.defaults)
         opts = pycompat.strkwargs(opts)
         opts.update(kwargs)
         return diffopts(**opts)
 
 defaultopts = diffopts()
 
 def wsclean(opts, text, blank=True):
     if opts.ignorews:
         text = bdiff.fixws(text, 1)
     elif opts.ignorewsamount:
         text = bdiff.fixws(text, 0)
     if blank and opts.ignoreblanklines:
         text = re.sub('\n+', '\n', text).strip('\n')
     if opts.ignorewseol:
         text = re.sub(br'[ \t\r\f]+\n', r'\n', text)
     return text
 
 def splitblock(base1, lines1, base2, lines2, opts):
     # The input lines matches except for interwoven blank lines. We
     # transform it into a sequence of matching blocks and blank blocks.
     lines1 = [(wsclean(opts, l) and 1 or 0) for l in lines1]
     lines2 = [(wsclean(opts, l) and 1 or 0) for l in lines2]
     s1, e1 = 0, len(lines1)
     s2, e2 = 0, len(lines2)
     while s1 < e1 or s2 < e2:
         i1, i2, btype = s1, s2, '='
         if (i1 >= e1 or lines1[i1] == 0
             or i2 >= e2 or lines2[i2] == 0):
             # Consume the block of blank lines
             btype = '~'
             while i1 < e1 and lines1[i1] == 0:
                 i1 += 1
             while i2 < e2 and lines2[i2] == 0:
                 i2 += 1
         else:
             # Consume the matching lines
             while i1 < e1 and lines1[i1] == 1 and lines2[i2] == 1:
                 i1 += 1
                 i2 += 1
         yield [base1 + s1, base1 + i1, base2 + s2, base2 + i2], btype
         s1 = i1
         s2 = i2
 
 def hunkinrange(hunk, linerange):
     """Return True if `hunk` defined as (start, length) is in `linerange`
     defined as (lowerbound, upperbound).
 
     >>> hunkinrange((5, 10), (2, 7))
     True
     >>> hunkinrange((5, 10), (6, 12))
     True
     >>> hunkinrange((5, 10), (13, 17))
     True
     >>> hunkinrange((5, 10), (3, 17))
     True
     >>> hunkinrange((5, 10), (1, 3))
     False
     >>> hunkinrange((5, 10), (18, 20))
     False
     >>> hunkinrange((5, 10), (1, 5))
     False
     >>> hunkinrange((5, 10), (15, 27))
     False
     """
     start, length = hunk
     lowerbound, upperbound = linerange
     return lowerbound < start + length and start < upperbound
 
 def blocksinrange(blocks, rangeb):
     """filter `blocks` like (a1, a2, b1, b2) from items outside line range
     `rangeb` from ``(b1, b2)`` point of view.
 
     Return `filteredblocks, rangea` where:
 
     * `filteredblocks` is list of ``block = (a1, a2, b1, b2), stype`` items of
       `blocks` that are inside `rangeb` from ``(b1, b2)`` point of view; a
       block ``(b1, b2)`` being inside `rangeb` if
       ``rangeb[0] < b2 and b1 < rangeb[1]``;
     * `rangea` is the line range w.r.t. to ``(a1, a2)`` parts of `blocks`.
     """
     lbb, ubb = rangeb
     lba, uba = None, None
     filteredblocks = []
     for block in blocks:
         (a1, a2, b1, b2), stype = block
         if lbb >= b1 and ubb <= b2 and stype == '=':
             # rangeb is within a single "=" hunk, restrict back linerange1
             # by offsetting rangeb
             lba = lbb - b1 + a1
             uba = ubb - b1 + a1
         else:
             if b1 <= lbb < b2:
                 if stype == '=':
                     lba = a2 - (b2 - lbb)
                 else:
                     lba = a1
             if b1 < ubb <= b2:
                 if stype == '=':
                     uba = a1 + (ubb - b1)
                 else:
                     uba = a2
         if hunkinrange((b1, (b2 - b1)), rangeb):
             filteredblocks.append(block)
     if lba is None or uba is None or uba < lba:
         raise error.Abort(_('line range exceeds file size'))
     return filteredblocks, (lba, uba)
 
 def allblocks(text1, text2, opts=None, lines1=None, lines2=None):
     """Return (block, type) tuples, where block is an mdiff.blocks
     line entry. type is '=' for blocks matching exactly one another
     (bdiff blocks), '!' for non-matching blocks and '~' for blocks
     matching only after having filtered blank lines.
     line1 and line2 are text1 and text2 split with splitnewlines() if
     they are already available.
     """
     if opts is None:
         opts = defaultopts
     if opts.ignorews or opts.ignorewsamount or opts.ignorewseol:
         text1 = wsclean(opts, text1, False)
         text2 = wsclean(opts, text2, False)
     diff = bdiff.blocks(text1, text2)
     for i, s1 in enumerate(diff):
         # The first match is special.
         # we've either found a match starting at line 0 or a match later
         # in the file.  If it starts later, old and new below will both be
         # empty and we'll continue to the next match.
         if i > 0:
             s = diff[i - 1]
         else:
             s = [0, 0, 0, 0]
         s = [s[1], s1[0], s[3], s1[2]]
 
         # bdiff sometimes gives huge matches past eof, this check eats them,
         # and deals with the special first match case described above
         if s[0] != s[1] or s[2] != s[3]:
             type = '!'
             if opts.ignoreblanklines:
                 if lines1 is None:
                     lines1 = splitnewlines(text1)
                 if lines2 is None:
                     lines2 = splitnewlines(text2)
                 old = wsclean(opts, "".join(lines1[s[0]:s[1]]))
                 new = wsclean(opts, "".join(lines2[s[2]:s[3]]))
                 if old == new:
                     type = '~'
             yield s, type
         yield s1, '='
 
-def unidiff(a, ad, b, bd, fn1, fn2, opts=defaultopts, check_binary=True):
+def unidiff(a, ad, b, bd, fn1, fn2, binary, opts=defaultopts):
     """Return a unified diff as a (headers, hunks) tuple.
 
     If the diff is not null, `headers` is a list with unified diff header
     lines "--- <original>" and "+++ <new>" and `hunks` is a generator yielding
     (hunkrange, hunklines) coming from _unidiff().
     Otherwise, `headers` and `hunks` are empty.
 
-    Setting `check_binary` to false will skip the binary check, i.e. when
-    it has been done in advance. Files are expected to be text in this case.
+    Set binary=True if either a or b should be taken as a binary file.
     """
     def datetag(date, fn=None):
         if not opts.git and not opts.nodates:
             return '\t%s' % date
         if fn and ' ' in fn:
             return '\t'
         return ''
 
     sentinel = [], ()
     if not a and not b:
         return sentinel
 
     if opts.noprefix:
         aprefix = bprefix = ''
     else:
         aprefix = 'a/'
         bprefix = 'b/'
 
     epoch = util.datestr((0, 0))
 
     fn1 = util.pconvert(fn1)
     fn2 = util.pconvert(fn2)
 
-    if not opts.text and check_binary and (util.binary(a) or util.binary(b)):
+    if binary:
         if a and b and len(a) == len(b) and a == b:
             return sentinel
         headerlines = []
         hunks = (None, ['Binary file %s has changed\n' % fn1]),
     elif not a:
         without_newline = b[-1:] != '\n'
         b = splitnewlines(b)
         if a is None:
             l1 = '--- /dev/null%s' % datetag(epoch)
         else:
             l1 = "--- %s%s%s" % (aprefix, fn1, datetag(ad, fn1))
         l2 = "+++ %s%s" % (bprefix + fn2, datetag(bd, fn2))
         headerlines = [l1, l2]
         size = len(b)
         hunkrange = (0, 0, 1, size)
         hunklines = ["@@ -0,0 +1,%d @@\n" % size] + ["+" + e for e in b]
         if without_newline:
             hunklines[-1] += '\n'
             hunklines.append(_missing_newline_marker)
         hunks = (hunkrange, hunklines),
     elif not b:
         without_newline = a[-1:] != '\n'
         a = splitnewlines(a)
         l1 = "--- %s%s%s" % (aprefix, fn1, datetag(ad, fn1))
         if b is None:
             l2 = '+++ /dev/null%s' % datetag(epoch)
         else:
             l2 = "+++ %s%s%s" % (bprefix, fn2, datetag(bd, fn2))
         headerlines = [l1, l2]
         size = len(a)
         hunkrange = (1, size, 0, 0)
         hunklines = ["@@ -1,%d +0,0 @@\n" % size] + ["-" + e for e in a]
         if without_newline:
             hunklines[-1] += '\n'
             hunklines.append(_missing_newline_marker)
         hunks = (hunkrange, hunklines),
     else:
         hunks = _unidiff(a, b, opts=opts)
         if not next(hunks):
             return sentinel
 
         headerlines = [
             "--- %s%s%s" % (aprefix, fn1, datetag(ad, fn1)),
             "+++ %s%s%s" % (bprefix, fn2, datetag(bd, fn2)),
         ]
 
     return headerlines, hunks
 
 def _unidiff(t1, t2, opts=defaultopts):
     """Yield hunks of a headerless unified diff from t1 and t2 texts.
 
     Each hunk consists of a (hunkrange, hunklines) tuple where `hunkrange` is a
     tuple (s1, l1, s2, l2) representing the range information of the hunk to
     form the '@@ -s1,l1 +s2,l2 @@' header and `hunklines` is a list of lines
     of the hunk combining said header followed by line additions and
     deletions.
 
     The hunks are prefixed with a bool.
     """
     l1 = splitnewlines(t1)
     l2 = splitnewlines(t2)
     def contextend(l, len):
         ret = l + opts.context
         if ret > len:
             ret = len
         return ret
 
     def contextstart(l):
         ret = l - opts.context
         if ret < 0:
             return 0
         return ret
 
     lastfunc = [0, '']
     def yieldhunk(hunk):
         (astart, a2, bstart, b2, delta) = hunk
         aend = contextend(a2, len(l1))
         alen = aend - astart
         blen = b2 - bstart + aend - a2
 
         func = ""
         if opts.showfunc:
             lastpos, func = lastfunc
             # walk backwards from the start of the context up to the start of
             # the previous hunk context until we find a line starting with an
             # alphanumeric char.
             for i in xrange(astart - 1, lastpos - 1, -1):
                 if l1[i][0:1].isalnum():
                     func = ' ' + l1[i].rstrip()[:40]
                     lastfunc[1] = func
                     break
             # by recording this hunk's starting point as the next place to
             # start looking for function lines, we avoid reading any line in
             # the file more than once.
             lastfunc[0] = astart
 
         # zero-length hunk ranges report their start line as one less
         if alen:
             astart += 1
         if blen:
             bstart += 1
 
         hunkrange = astart, alen, bstart, blen
         hunklines = (
             ["@@ -%d,%d +%d,%d @@%s\n" % (hunkrange + (func,))]
             + delta
             + [' ' + l1[x] for x in xrange(a2, aend)]
         )
         # If either file ends without a newline and the last line of
         # that file is part of a hunk, a marker is printed. If the
         # last line of both files is identical and neither ends in
         # a newline, print only one marker. That's the only case in
         # which the hunk can end in a shared line without a newline.
         skip = False
         if t1[-1:] != '\n' and astart + alen == len(l1) + 1:
             for i in xrange(len(hunklines) - 1, -1, -1):
                 if hunklines[i][0:1] in ('-', ' '):
                     if hunklines[i][0:1] == ' ':
                         skip = True
                     hunklines[i] += '\n'
                     hunklines.insert(i + 1, _missing_newline_marker)
                     break
         if not skip and t2[-1:] != '\n' and bstart + blen == len(l2) + 1:
             for i in xrange(len(hunklines) - 1, -1, -1):
                 if hunklines[i][0:1] == '+':
                     hunklines[i] += '\n'
                     hunklines.insert(i + 1, _missing_newline_marker)
                     break
         yield hunkrange, hunklines
 
     # bdiff.blocks gives us the matching sequences in the files.  The loop
     # below finds the spaces between those matching sequences and translates
     # them into diff output.
     #
     hunk = None
     ignoredlines = 0
     has_hunks = False
     for s, stype in allblocks(t1, t2, opts, l1, l2):
         a1, a2, b1, b2 = s
         if stype != '!':
             if stype == '~':
                 # The diff context lines are based on t1 content. When
                 # blank lines are ignored, the new lines offsets must
                 # be adjusted as if equivalent blocks ('~') had the
                 # same sizes on both sides.
                 ignoredlines += (b2 - b1) - (a2 - a1)
             continue
         delta = []
         old = l1[a1:a2]
         new = l2[b1:b2]
 
         b1 -= ignoredlines
         b2 -= ignoredlines
         astart = contextstart(a1)
         bstart = contextstart(b1)
         prev = None
         if hunk:
             # join with the previous hunk if it falls inside the context
             if astart < hunk[1] + opts.context + 1:
                 prev = hunk
                 astart = hunk[1]
                 bstart = hunk[3]
             else:
                 if not has_hunks:
                     has_hunks = True
                     yield True
                 for x in yieldhunk(hunk):
                     yield x
         if prev:
             # we've joined the previous hunk, record the new ending points.
             hunk[1] = a2
             hunk[3] = b2
             delta = hunk[4]
         else:
             # create a new hunk
             hunk = [astart, a2, bstart, b2, delta]
 
         delta[len(delta):] = [' ' + x for x in l1[astart:a1]]
         delta[len(delta):] = ['-' + x for x in old]
         delta[len(delta):] = ['+' + x for x in new]
 
     if hunk:
         if not has_hunks:
             has_hunks = True
             yield True
         for x in yieldhunk(hunk):
             yield x
     elif not has_hunks:
         yield False
 
 def b85diff(to, tn):
     '''print base85-encoded binary diff'''
     def fmtline(line):
         l = len(line)
         if l <= 26:
             l = pycompat.bytechr(ord('A') + l - 1)
         else:
             l = pycompat.bytechr(l - 26 + ord('a') - 1)
         return '%c%s\n' % (l, util.b85encode(line, True))
 
     def chunk(text, csize=52):
         l = len(text)
         i = 0
         while i < l:
             yield text[i:i + csize]
             i += csize
 
     if to is None:
         to = ''
     if tn is None:
         tn = ''
 
     if to == tn:
         return ''
 
     # TODO: deltas
     ret = []
     ret.append('GIT binary patch\n')
     ret.append('literal %d\n' % len(tn))
     for l in chunk(zlib.compress(tn)):
         ret.append(fmtline(l))
     ret.append('\n')
 
     return ''.join(ret)
 
 def patchtext(bin):
     pos = 0
     t = []
     while pos < len(bin):
         p1, p2, l = struct.unpack(">lll", bin[pos:pos + 12])
         pos += 12
         t.append(bin[pos:pos + l])
         pos += l
     return "".join(t)
 
 def patch(a, bin):
     if len(a) == 0:
         # skip over trivial delta header
         return util.buffer(bin, 12)
     return mpatch.patches(a, [bin])
 
 # similar to difflib.SequenceMatcher.get_matching_blocks
 def get_matching_blocks(a, b):
     return [(d[0], d[2], d[1] - d[0]) for d in bdiff.blocks(a, b)]
 
 def trivialdiffheader(length):
     return struct.pack(">lll", 0, 0, length) if length else ''
 
 def replacediffheader(oldlen, newlen):
     return struct.pack(">lll", 0, oldlen, newlen)

mercurial/patch.py

@@ -1,2907 +1,2904 b''
 # patch.py - patch file parsing routines
 #
 # Copyright 2006 Brendan Cully <[e-mail]>
 # Copyright 2007 Chris Mason <[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, print_function
 
 import collections
 import copy
 import difflib
 import email
 import email.parser as emailparser
 import errno
 import hashlib
 import os
 import posixpath
 import re
 import shutil
 import tempfile
 import zlib
 
 from .i18n import _
 from .node import (
     hex,
     short,
 )
 from . import (
     copies,
     encoding,
     error,
     mail,
     mdiff,
     pathutil,
     policy,
     pycompat,
     scmutil,
     similar,
     util,
     vfs as vfsmod,
 )
 
 diffhelpers = policy.importmod(r'diffhelpers')
 stringio = util.stringio
 
 gitre = re.compile(br'diff --git a/(.*) b/(.*)')
 tabsplitter = re.compile(br'(\t+|[^\t]+)')
 _nonwordre = re.compile(br'([^a-zA-Z0-9_\x80-\xff])')
 
 PatchError = error.PatchError
 
 # public functions
 
 def split(stream):
     '''return an iterator of individual patches from a stream'''
     def isheader(line, inheader):
         if inheader and line[0] in (' ', '\t'):
             # continuation
             return True
         if line[0] in (' ', '-', '+'):
             # diff line - don't check for header pattern in there
             return False
         l = line.split(': ', 1)
         return len(l) == 2 and ' ' not in l[0]
 
     def chunk(lines):
         return stringio(''.join(lines))
 
     def hgsplit(stream, cur):
         inheader = True
 
         for line in stream:
             if not line.strip():
                 inheader = False
             if not inheader and line.startswith('# HG changeset patch'):
                 yield chunk(cur)
                 cur = []
                 inheader = True
 
             cur.append(line)
 
         if cur:
             yield chunk(cur)
 
     def mboxsplit(stream, cur):
         for line in stream:
             if line.startswith('From '):
                 for c in split(chunk(cur[1:])):
                     yield c
                 cur = []
 
             cur.append(line)
 
         if cur:
             for c in split(chunk(cur[1:])):
                 yield c
 
     def mimesplit(stream, cur):
         def msgfp(m):
             fp = stringio()
             g = email.Generator.Generator(fp, mangle_from_=False)
             g.flatten(m)
             fp.seek(0)
             return fp
 
         for line in stream:
             cur.append(line)
         c = chunk(cur)
 
         m = emailparser.Parser().parse(c)
         if not m.is_multipart():
             yield msgfp(m)
         else:
             ok_types = ('text/plain', 'text/x-diff', 'text/x-patch')
             for part in m.walk():
                 ct = part.get_content_type()
                 if ct not in ok_types:
                     continue
                 yield msgfp(part)
 
     def headersplit(stream, cur):
         inheader = False
 
         for line in stream:
             if not inheader and isheader(line, inheader):
                 yield chunk(cur)
                 cur = []
                 inheader = True
             if inheader and not isheader(line, inheader):
                 inheader = False
 
             cur.append(line)
 
         if cur:
             yield chunk(cur)
 
     def remainder(cur):
         yield chunk(cur)
 
     class fiter(object):
         def __init__(self, fp):
             self.fp = fp
 
         def __iter__(self):
             return self
 
         def next(self):
             l = self.fp.readline()
             if not l:
                 raise StopIteration
             return l
 
         __next__ = next
 
     inheader = False
     cur = []
 
     mimeheaders = ['content-type']
 
     if not util.safehasattr(stream, 'next'):
         # http responses, for example, have readline but not next
         stream = fiter(stream)
 
     for line in stream:
         cur.append(line)
         if line.startswith('# HG changeset patch'):
             return hgsplit(stream, cur)
         elif line.startswith('From '):
             return mboxsplit(stream, cur)
         elif isheader(line, inheader):
             inheader = True
             if line.split(':', 1)[0].lower() in mimeheaders:
                 # let email parser handle this
                 return mimesplit(stream, cur)
         elif line.startswith('--- ') and inheader:
             # No evil headers seen by diff start, split by hand
             return headersplit(stream, cur)
         # Not enough info, keep reading
 
     # if we are here, we have a very plain patch
     return remainder(cur)
 
 ## Some facility for extensible patch parsing:
 # list of pairs ("header to match", "data key")
 patchheadermap = [('Date', 'date'),
                   ('Branch', 'branch'),
                   ('Node ID', 'nodeid'),
                  ]
 
 def extract(ui, fileobj):
     '''extract patch from data read from fileobj.
 
     patch can be a normal patch or contained in an email message.
 
     return a dictionary. Standard keys are:
       - filename,
       - message,
       - user,
       - date,
       - branch,
       - node,
       - p1,
       - p2.
     Any item can be missing from the dictionary. If filename is missing,
     fileobj did not contain a patch. Caller must unlink filename when done.'''
 
     # attempt to detect the start of a patch
     # (this heuristic is borrowed from quilt)
     diffre = re.compile(br'^(?:Index:[ \t]|diff[ \t]-|RCS file: |'
                         br'retrieving revision [0-9]+(\.[0-9]+)*$|'
                         br'---[ \t].*?^\+\+\+[ \t]|'
                         br'\*\*\*[ \t].*?^---[ \t])',
                         re.MULTILINE | re.DOTALL)
 
     data = {}
     fd, tmpname = tempfile.mkstemp(prefix='hg-patch-')
     tmpfp = os.fdopen(fd, pycompat.sysstr('w'))
     try:
         msg = emailparser.Parser().parse(fileobj)
 
         subject = msg['Subject'] and mail.headdecode(msg['Subject'])
         data['user'] = msg['From'] and mail.headdecode(msg['From'])
         if not subject and not data['user']:
             # Not an email, restore parsed headers if any
             subject = '\n'.join(': '.join(h) for h in msg.items()) + '\n'
 
         # should try to parse msg['Date']
         parents = []
 
         if subject:
             if subject.startswith('[PATCH'):
                 pend = subject.find(']')
                 if pend >= 0:
                     subject = subject[pend + 1:].lstrip()
             subject = re.sub(br'\n[ \t]+', ' ', subject)
             ui.debug('Subject: %s\n' % subject)
         if data['user']:
             ui.debug('From: %s\n' % data['user'])
         diffs_seen = 0
         ok_types = ('text/plain', 'text/x-diff', 'text/x-patch')
         message = ''
         for part in msg.walk():
             content_type = part.get_content_type()
             ui.debug('Content-Type: %s\n' % content_type)
             if content_type not in ok_types:
                 continue
             payload = part.get_payload(decode=True)
             m = diffre.search(payload)
             if m:
                 hgpatch = False
                 hgpatchheader = False
                 ignoretext = False
 
                 ui.debug('found patch at byte %d\n' % m.start(0))
                 diffs_seen += 1
                 cfp = stringio()
                 for line in payload[:m.start(0)].splitlines():
                     if line.startswith('# HG changeset patch') and not hgpatch:
                         ui.debug('patch generated by hg export\n')
                         hgpatch = True
                         hgpatchheader = True
                         # drop earlier commit message content
                         cfp.seek(0)
                         cfp.truncate()
                         subject = None
                     elif hgpatchheader:
                         if line.startswith('# User '):
                             data['user'] = line[7:]
                             ui.debug('From: %s\n' % data['user'])
                         elif line.startswith("# Parent "):
                             parents.append(line[9:].lstrip())
                         elif line.startswith("# "):
                             for header, key in patchheadermap:
                                 prefix = '# %s ' % header
                                 if line.startswith(prefix):
                                     data[key] = line[len(prefix):]
                         else:
                             hgpatchheader = False
                     elif line == '---':
                         ignoretext = True
                     if not hgpatchheader and not ignoretext:
                         cfp.write(line)
                         cfp.write('\n')
                 message = cfp.getvalue()
                 if tmpfp:
                     tmpfp.write(payload)
                     if not payload.endswith('\n'):
                         tmpfp.write('\n')
             elif not diffs_seen and message and content_type == 'text/plain':
                 message += '\n' + payload
     except: # re-raises
         tmpfp.close()
         os.unlink(tmpname)
         raise
 
     if subject and not message.startswith(subject):
         message = '%s\n%s' % (subject, message)
     data['message'] = message
     tmpfp.close()
     if parents:
         data['p1'] = parents.pop(0)
         if parents:
             data['p2'] = parents.pop(0)
 
     if diffs_seen:
         data['filename'] = tmpname
     else:
         os.unlink(tmpname)
     return data
 
 class patchmeta(object):
     """Patched file metadata
 
     'op' is the performed operation within ADD, DELETE, RENAME, MODIFY
     or COPY.  'path' is patched file path. 'oldpath' is set to the
     origin file when 'op' is either COPY or RENAME, None otherwise. If
     file mode is changed, 'mode' is a tuple (islink, isexec) where
     'islink' is True if the file is a symlink and 'isexec' is True if
     the file is executable. Otherwise, 'mode' is None.
     """
     def __init__(self, path):
         self.path = path
         self.oldpath = None
         self.mode = None
         self.op = 'MODIFY'
         self.binary = False
 
     def setmode(self, mode):
         islink = mode & 0o20000
         isexec = mode & 0o100
         self.mode = (islink, isexec)
 
     def copy(self):
         other = patchmeta(self.path)
         other.oldpath = self.oldpath
         other.mode = self.mode
         other.op = self.op
         other.binary = self.binary
         return other
 
     def _ispatchinga(self, afile):
         if afile == '/dev/null':
             return self.op == 'ADD'
         return afile == 'a/' + (self.oldpath or self.path)
 
     def _ispatchingb(self, bfile):
         if bfile == '/dev/null':
             return self.op == 'DELETE'
         return bfile == 'b/' + self.path
 
     def ispatching(self, afile, bfile):
         return self._ispatchinga(afile) and self._ispatchingb(bfile)
 
     def __repr__(self):
         return "<patchmeta %s %r>" % (self.op, self.path)
 
 def readgitpatch(lr):
     """extract git-style metadata about patches from <patchname>"""
 
     # Filter patch for git information
     gp = None
     gitpatches = []
     for line in lr:
         line = line.rstrip(' \r\n')
         if line.startswith('diff --git a/'):
             m = gitre.match(line)
             if m:
                 if gp:
                     gitpatches.append(gp)
                 dst = m.group(2)
                 gp = patchmeta(dst)
         elif gp:
             if line.startswith('--- '):
                 gitpatches.append(gp)
                 gp = None
                 continue
             if line.startswith('rename from '):
                 gp.op = 'RENAME'
                 gp.oldpath = line[12:]
             elif line.startswith('rename to '):
                 gp.path = line[10:]
             elif line.startswith('copy from '):
                 gp.op = 'COPY'
                 gp.oldpath = line[10:]
             elif line.startswith('copy to '):
                 gp.path = line[8:]
             elif line.startswith('deleted file'):
                 gp.op = 'DELETE'
             elif line.startswith('new file mode '):
                 gp.op = 'ADD'
                 gp.setmode(int(line[-6:], 8))
             elif line.startswith('new mode '):
                 gp.setmode(int(line[-6:], 8))
             elif line.startswith('GIT binary patch'):
                 gp.binary = True
     if gp:
         gitpatches.append(gp)
 
     return gitpatches
 
 class linereader(object):
     # simple class to allow pushing lines back into the input stream
     def __init__(self, fp):
         self.fp = fp
         self.buf = []
 
     def push(self, line):
         if line is not None:
             self.buf.append(line)
 
     def readline(self):
         if self.buf:
             l = self.buf[0]
             del self.buf[0]
             return l
         return self.fp.readline()
 
     def __iter__(self):
         return iter(self.readline, '')
 
 class abstractbackend(object):
     def __init__(self, ui):
         self.ui = ui
 
     def getfile(self, fname):
         """Return target file data and flags as a (data, (islink,
         isexec)) tuple. Data is None if file is missing/deleted.
         """
         raise NotImplementedError
 
     def setfile(self, fname, data, mode, copysource):
         """Write data to target file fname and set its mode. mode is a
         (islink, isexec) tuple. If data is None, the file content should
         be left unchanged. If the file is modified after being copied,
         copysource is set to the original file name.
         """
         raise NotImplementedError
 
     def unlink(self, fname):
         """Unlink target file."""
         raise NotImplementedError
 
     def writerej(self, fname, failed, total, lines):
         """Write rejected lines for fname. total is the number of hunks
         which failed to apply and total the total number of hunks for this
         files.
         """
 
     def exists(self, fname):
         raise NotImplementedError
 
     def close(self):
         raise NotImplementedError
 
 class fsbackend(abstractbackend):
     def __init__(self, ui, basedir):
         super(fsbackend, self).__init__(ui)
         self.opener = vfsmod.vfs(basedir)
 
     def getfile(self, fname):
         if self.opener.islink(fname):
             return (self.opener.readlink(fname), (True, False))
 
         isexec = False
         try:
             isexec = self.opener.lstat(fname).st_mode & 0o100 != 0
         except OSError as e:
             if e.errno != errno.ENOENT:
                 raise
         try:
             return (self.opener.read(fname), (False, isexec))
         except IOError as e:
             if e.errno != errno.ENOENT:
                 raise
             return None, None
 
     def setfile(self, fname, data, mode, copysource):
         islink, isexec = mode
         if data is None:
             self.opener.setflags(fname, islink, isexec)
             return
         if islink:
             self.opener.symlink(data, fname)
         else:
             self.opener.write(fname, data)
             if isexec:
                 self.opener.setflags(fname, False, True)
 
     def unlink(self, fname):
         self.opener.unlinkpath(fname, ignoremissing=True)
 
     def writerej(self, fname, failed, total, lines):
         fname = fname + ".rej"
         self.ui.warn(
             _("%d out of %d hunks FAILED -- saving rejects to file %s\n") %
             (failed, total, fname))
         fp = self.opener(fname, 'w')
         fp.writelines(lines)
         fp.close()
 
     def exists(self, fname):
         return self.opener.lexists(fname)
 
 class workingbackend(fsbackend):
     def __init__(self, ui, repo, similarity):
         super(workingbackend, self).__init__(ui, repo.root)
         self.repo = repo
         self.similarity = similarity
         self.removed = set()
         self.changed = set()
         self.copied = []
 
     def _checkknown(self, fname):
         if self.repo.dirstate[fname] == '?' and self.exists(fname):
             raise PatchError(_('cannot patch %s: file is not tracked') % fname)
 
     def setfile(self, fname, data, mode, copysource):
         self._checkknown(fname)
         super(workingbackend, self).setfile(fname, data, mode, copysource)
         if copysource is not None:
             self.copied.append((copysource, fname))
         self.changed.add(fname)
 
     def unlink(self, fname):
         self._checkknown(fname)
         super(workingbackend, self).unlink(fname)
         self.removed.add(fname)
         self.changed.add(fname)
 
     def close(self):
         wctx = self.repo[None]
         changed = set(self.changed)
         for src, dst in self.copied:
             scmutil.dirstatecopy(self.ui, self.repo, wctx, src, dst)
         if self.removed:
             wctx.forget(sorted(self.removed))
             for f in self.removed:
                 if f not in self.repo.dirstate:
                     # File was deleted and no longer belongs to the
                     # dirstate, it was probably marked added then
                     # deleted, and should not be considered by
                     # marktouched().
                     changed.discard(f)
         if changed:
             scmutil.marktouched(self.repo, changed, self.similarity)
         return sorted(self.changed)
 
 class filestore(object):
     def __init__(self, maxsize=None):
         self.opener = None
         self.files = {}
         self.created = 0
         self.maxsize = maxsize
         if self.maxsize is None:
             self.maxsize = 4*(2**20)
         self.size = 0
         self.data = {}
 
     def setfile(self, fname, data, mode, copied=None):
         if self.maxsize < 0 or (len(data) + self.size) <= self.maxsize:
             self.data[fname] = (data, mode, copied)
             self.size += len(data)
         else:
             if self.opener is None:
                 root = tempfile.mkdtemp(prefix='hg-patch-')
                 self.opener = vfsmod.vfs(root)
             # Avoid filename issues with these simple names
             fn = str(self.created)
             self.opener.write(fn, data)
             self.created += 1
             self.files[fname] = (fn, mode, copied)
 
     def getfile(self, fname):
         if fname in self.data:
             return self.data[fname]
         if not self.opener or fname not in self.files:
             return None, None, None
         fn, mode, copied = self.files[fname]
         return self.opener.read(fn), mode, copied
 
     def close(self):
         if self.opener:
             shutil.rmtree(self.opener.base)
 
 class repobackend(abstractbackend):
     def __init__(self, ui, repo, ctx, store):
         super(repobackend, self).__init__(ui)
         self.repo = repo
         self.ctx = ctx
         self.store = store
         self.changed = set()
         self.removed = set()
         self.copied = {}
 
     def _checkknown(self, fname):
         if fname not in self.ctx:
             raise PatchError(_('cannot patch %s: file is not tracked') % fname)
 
     def getfile(self, fname):
         try:
             fctx = self.ctx[fname]
         except error.LookupError:
             return None, None
         flags = fctx.flags()
         return fctx.data(), ('l' in flags, 'x' in flags)
 
     def setfile(self, fname, data, mode, copysource):
         if copysource:
             self._checkknown(copysource)
         if data is None:
             data = self.ctx[fname].data()
         self.store.setfile(fname, data, mode, copysource)
         self.changed.add(fname)
         if copysource:
             self.copied[fname] = copysource
 
     def unlink(self, fname):
         self._checkknown(fname)
         self.removed.add(fname)
 
     def exists(self, fname):
         return fname in self.ctx
 
     def close(self):
         return self.changed | self.removed
 
 # @@ -start,len +start,len @@ or @@ -start +start @@ if len is 1
 unidesc = re.compile('@@ -(\d+)(?:,(\d+))? \+(\d+)(?:,(\d+))? @@')
 contextdesc = re.compile('(?:---|\*\*\*) (\d+)(?:,(\d+))? (?:---|\*\*\*)')
 eolmodes = ['strict', 'crlf', 'lf', 'auto']
 
 class patchfile(object):
     def __init__(self, ui, gp, backend, store, eolmode='strict'):
         self.fname = gp.path
         self.eolmode = eolmode
         self.eol = None
         self.backend = backend
         self.ui = ui
         self.lines = []
         self.exists = False
         self.missing = True
         self.mode = gp.mode
         self.copysource = gp.oldpath
         self.create = gp.op in ('ADD', 'COPY', 'RENAME')
         self.remove = gp.op == 'DELETE'
         if self.copysource is None:
             data, mode = backend.getfile(self.fname)
         else:
             data, mode = store.getfile(self.copysource)[:2]
         if data is not None:
             self.exists = self.copysource is None or backend.exists(self.fname)
             self.missing = False
             if data:
                 self.lines = mdiff.splitnewlines(data)
             if self.mode is None:
                 self.mode = mode
             if self.lines:
                 # Normalize line endings
                 if self.lines[0].endswith('\r\n'):
                     self.eol = '\r\n'
                 elif self.lines[0].endswith('\n'):
                     self.eol = '\n'
                 if eolmode != 'strict':
                     nlines = []
                     for l in self.lines:
                         if l.endswith('\r\n'):
                             l = l[:-2] + '\n'
                         nlines.append(l)
                     self.lines = nlines
         else:
             if self.create:
                 self.missing = False
             if self.mode is None:
                 self.mode = (False, False)
         if self.missing:
             self.ui.warn(_("unable to find '%s' for patching\n") % self.fname)
             self.ui.warn(_("(use '--prefix' to apply patch relative to the "
                            "current directory)\n"))
 
         self.hash = {}
         self.dirty = 0
         self.offset = 0
         self.skew = 0
         self.rej = []
         self.fileprinted = False
         self.printfile(False)
         self.hunks = 0
 
     def writelines(self, fname, lines, mode):
         if self.eolmode == 'auto':
             eol = self.eol
         elif self.eolmode == 'crlf':
             eol = '\r\n'
         else:
             eol = '\n'
 
         if self.eolmode != 'strict' and eol and eol != '\n':
             rawlines = []
             for l in lines:
                 if l and l[-1] == '\n':
                     l = l[:-1] + eol
                 rawlines.append(l)
             lines = rawlines
 
         self.backend.setfile(fname, ''.join(lines), mode, self.copysource)
 
     def printfile(self, warn):
         if self.fileprinted:
             return
         if warn or self.ui.verbose:
             self.fileprinted = True
         s = _("patching file %s\n") % self.fname
         if warn:
             self.ui.warn(s)
         else:
             self.ui.note(s)
 
 
     def findlines(self, l, linenum):
         # looks through the hash and finds candidate lines.  The
         # result is a list of line numbers sorted based on distance
         # from linenum
 
         cand = self.hash.get(l, [])
         if len(cand) > 1:
             # resort our list of potentials forward then back.
             cand.sort(key=lambda x: abs(x - linenum))
         return cand
 
     def write_rej(self):
         # our rejects are a little different from patch(1).  This always
         # creates rejects in the same form as the original patch.  A file
         # header is inserted so that you can run the reject through patch again
         # without having to type the filename.
         if not self.rej:
             return
         base = os.path.basename(self.fname)
         lines = ["--- %s\n+++ %s\n" % (base, base)]
         for x in self.rej:
             for l in x.hunk:
                 lines.append(l)
                 if l[-1:] != '\n':
                     lines.append("\n\ No newline at end of file\n")
         self.backend.writerej(self.fname, len(self.rej), self.hunks, lines)
 
     def apply(self, h):
         if not h.complete():
             raise PatchError(_("bad hunk #%d %s (%d %d %d %d)") %
                             (h.number, h.desc, len(h.a), h.lena, len(h.b),
                             h.lenb))
 
         self.hunks += 1
 
         if self.missing:
             self.rej.append(h)
             return -1
 
         if self.exists and self.create:
             if self.copysource:
                 self.ui.warn(_("cannot create %s: destination already "
                                "exists\n") % self.fname)
             else:
                 self.ui.warn(_("file %s already exists\n") % self.fname)
             self.rej.append(h)
             return -1
 
         if isinstance(h, binhunk):
             if self.remove:
                 self.backend.unlink(self.fname)
             else:
                 l = h.new(self.lines)
                 self.lines[:] = l
                 self.offset += len(l)
                 self.dirty = True
             return 0
 
         horig = h
         if (self.eolmode in ('crlf', 'lf')
             or self.eolmode == 'auto' and self.eol):
             # If new eols are going to be normalized, then normalize
             # hunk data before patching. Otherwise, preserve input
             # line-endings.
             h = h.getnormalized()
 
         # fast case first, no offsets, no fuzz
         old, oldstart, new, newstart = h.fuzzit(0, False)
         oldstart += self.offset
         orig_start = oldstart
         # if there's skew we want to emit the "(offset %d lines)" even
         # when the hunk cleanly applies at start + skew, so skip the
         # fast case code
         if (self.skew == 0 and
             diffhelpers.testhunk(old, self.lines, oldstart) == 0):
             if self.remove:
                 self.backend.unlink(self.fname)
             else:
                 self.lines[oldstart:oldstart + len(old)] = new
                 self.offset += len(new) - len(old)
                 self.dirty = True
             return 0
 
         # ok, we couldn't match the hunk. Lets look for offsets and fuzz it
         self.hash = {}
         for x, s in enumerate(self.lines):
             self.hash.setdefault(s, []).append(x)
 
         for fuzzlen in xrange(self.ui.configint("patch", "fuzz") + 1):
             for toponly in [True, False]:
                 old, oldstart, new, newstart = h.fuzzit(fuzzlen, toponly)
                 oldstart = oldstart + self.offset + self.skew
                 oldstart = min(oldstart, len(self.lines))
                 if old:
                     cand = self.findlines(old[0][1:], oldstart)
                 else:
                     # Only adding lines with no or fuzzed context, just
                     # take the skew in account
                     cand = [oldstart]
 
                 for l in cand:
                     if not old or diffhelpers.testhunk(old, self.lines, l) == 0:
                         self.lines[l : l + len(old)] = new
                         self.offset += len(new) - len(old)
                         self.skew = l - orig_start
                         self.dirty = True
                         offset = l - orig_start - fuzzlen
                         if fuzzlen:
                             msg = _("Hunk #%d succeeded at %d "
                                     "with fuzz %d "
                                     "(offset %d lines).\n")
                             self.printfile(True)
                             self.ui.warn(msg %
                                 (h.number, l + 1, fuzzlen, offset))
                         else:
                             msg = _("Hunk #%d succeeded at %d "
                                     "(offset %d lines).\n")
                             self.ui.note(msg % (h.number, l + 1, offset))
                         return fuzzlen
         self.printfile(True)
         self.ui.warn(_("Hunk #%d FAILED at %d\n") % (h.number, orig_start))
         self.rej.append(horig)
         return -1
 
     def close(self):
         if self.dirty:
             self.writelines(self.fname, self.lines, self.mode)
         self.write_rej()
         return len(self.rej)
 
 class header(object):
     """patch header
     """
     diffgit_re = re.compile('diff --git a/(.*) b/(.*)$')
     diff_re = re.compile('diff -r .* (.*)$')
     allhunks_re = re.compile('(?:index|deleted file) ')
     pretty_re = re.compile('(?:new file|deleted file) ')
     special_re = re.compile('(?:index|deleted|copy|rename) ')
     newfile_re = re.compile('(?:new file)')
 
     def __init__(self, header):
         self.header = header
         self.hunks = []
 
     def binary(self):
         return any(h.startswith('index ') for h in self.header)
 
     def pretty(self, fp):
         for h in self.header:
             if h.startswith('index '):
                 fp.write(_('this modifies a binary file (all or nothing)\n'))
                 break
             if self.pretty_re.match(h):
                 fp.write(h)
                 if self.binary():
                     fp.write(_('this is a binary file\n'))
                 break
             if h.startswith('---'):
                 fp.write(_('%d hunks, %d lines changed\n') %
                          (len(self.hunks),
                           sum([max(h.added, h.removed) for h in self.hunks])))
                 break
             fp.write(h)
 
     def write(self, fp):
         fp.write(''.join(self.header))
 
     def allhunks(self):
         return any(self.allhunks_re.match(h) for h in self.header)
 
     def files(self):
         match = self.diffgit_re.match(self.header[0])
         if match:
             fromfile, tofile = match.groups()
             if fromfile == tofile:
                 return [fromfile]
             return [fromfile, tofile]
         else:
             return self.diff_re.match(self.header[0]).groups()
 
     def filename(self):
         return self.files()[-1]
 
     def __repr__(self):
         return '<header %s>' % (' '.join(map(repr, self.files())))
 
     def isnewfile(self):
         return any(self.newfile_re.match(h) for h in self.header)
 
     def special(self):
         # Special files are shown only at the header level and not at the hunk
         # level for example a file that has been deleted is a special file.
         # The user cannot change the content of the operation, in the case of
         # the deleted file he has to take the deletion or not take it, he
         # cannot take some of it.
         # Newly added files are special if they are empty, they are not special
         # if they have some content as we want to be able to change it
         nocontent = len(self.header) == 2
         emptynewfile = self.isnewfile() and nocontent
         return emptynewfile or \
                 any(self.special_re.match(h) for h in self.header)
 
 class recordhunk(object):
     """patch hunk
 
     XXX shouldn't we merge this with the other hunk class?
     """
 
     def __init__(self, header, fromline, toline, proc, before, hunk, after,
                  maxcontext=None):
         def trimcontext(lines, reverse=False):
             if maxcontext is not None:
                 delta = len(lines) - maxcontext
                 if delta > 0:
                     if reverse:
                         return delta, lines[delta:]
                     else:
                         return delta, lines[:maxcontext]
             return 0, lines
 
         self.header = header
         trimedbefore, self.before = trimcontext(before, True)
         self.fromline = fromline + trimedbefore
         self.toline = toline + trimedbefore
         _trimedafter, self.after = trimcontext(after, False)
         self.proc = proc
         self.hunk = hunk
         self.added, self.removed = self.countchanges(self.hunk)
 
     def __eq__(self, v):
         if not isinstance(v, recordhunk):
             return False
 
         return ((v.hunk == self.hunk) and
                 (v.proc == self.proc) and
                 (self.fromline == v.fromline) and
                 (self.header.files() == v.header.files()))
 
     def __hash__(self):
         return hash((tuple(self.hunk),
             tuple(self.header.files()),
             self.fromline,
             self.proc))
 
     def countchanges(self, hunk):
         """hunk -> (n+,n-)"""
         add = len([h for h in hunk if h.startswith('+')])
         rem = len([h for h in hunk if h.startswith('-')])
         return add, rem
 
     def reversehunk(self):
         """return another recordhunk which is the reverse of the hunk
 
         If this hunk is diff(A, B), the returned hunk is diff(B, A). To do
         that, swap fromline/toline and +/- signs while keep other things
         unchanged.
         """
         m = {'+': '-', '-': '+', '\\': '\\'}
         hunk = ['%s%s' % (m[l[0:1]], l[1:]) for l in self.hunk]
         return recordhunk(self.header, self.toline, self.fromline, self.proc,
                           self.before, hunk, self.after)
 
     def write(self, fp):
         delta = len(self.before) + len(self.after)
         if self.after and self.after[-1] == '\\ No newline at end of file\n':
             delta -= 1
         fromlen = delta + self.removed
         tolen = delta + self.added
         fp.write('@@ -%d,%d +%d,%d @@%s\n' %
                  (self.fromline, fromlen, self.toline, tolen,
                   self.proc and (' ' + self.proc)))
         fp.write(''.join(self.before + self.hunk + self.after))
 
     pretty = write
 
     def filename(self):
         return self.header.filename()
 
     def __repr__(self):
         return '<hunk %r@%d>' % (self.filename(), self.fromline)
 
 def getmessages():
     return {
         'multiple': {
             'apply': _("apply change %d/%d to '%s'?"),
             'discard': _("discard change %d/%d to '%s'?"),
             'record': _("record change %d/%d to '%s'?"),
         },
         'single': {
             'apply': _("apply this change to '%s'?"),
             'discard': _("discard this change to '%s'?"),
             'record': _("record this change to '%s'?"),
         },
         'help': {
             'apply': _('[Ynesfdaq?]'
                          '$$ &Yes, apply this change'
                          '$$ &No, skip this change'
                          '$$ &Edit this change manually'
                          '$$ &Skip remaining changes to this file'
                          '$$ Apply remaining changes to this &file'
                          '$$ &Done, skip remaining changes and files'
                          '$$ Apply &all changes to all remaining files'
                          '$$ &Quit, applying no changes'
                          '$$ &? (display help)'),
             'discard': _('[Ynesfdaq?]'
                          '$$ &Yes, discard this change'
                          '$$ &No, skip this change'
                          '$$ &Edit this change manually'
                          '$$ &Skip remaining changes to this file'
                          '$$ Discard remaining changes to this &file'
                          '$$ &Done, skip remaining changes and files'
                          '$$ Discard &all changes to all remaining files'
                          '$$ &Quit, discarding no changes'
                          '$$ &? (display help)'),
             'record': _('[Ynesfdaq?]'
                         '$$ &Yes, record this change'
                         '$$ &No, skip this change'
                         '$$ &Edit this change manually'
                         '$$ &Skip remaining changes to this file'
                         '$$ Record remaining changes to this &file'
                         '$$ &Done, skip remaining changes and files'
                         '$$ Record &all changes to all remaining files'
                         '$$ &Quit, recording no changes'
                         '$$ &? (display help)'),
         }
     }
 
 def filterpatch(ui, headers, operation=None):
     """Interactively filter patch chunks into applied-only chunks"""
     messages = getmessages()
 
     if operation is None:
         operation = 'record'
 
     def prompt(skipfile, skipall, query, chunk):
         """prompt query, and process base inputs
 
         - y/n for the rest of file
         - y/n for the rest
         - ? (help)
         - q (quit)
 
         Return True/False and possibly updated skipfile and skipall.
         """
         newpatches = None
         if skipall is not None:
             return skipall, skipfile, skipall, newpatches
         if skipfile is not None:
             return skipfile, skipfile, skipall, newpatches
         while True:
             resps = messages['help'][operation]
             r = ui.promptchoice("%s %s" % (query, resps))
             ui.write("\n")
             if r == 8: # ?
                 for c, t in ui.extractchoices(resps)[1]:
                     ui.write('%s - %s\n' % (c, encoding.lower(t)))
                 continue
             elif r == 0: # yes
                 ret = True
             elif r == 1: # no
                 ret = False
             elif r == 2: # Edit patch
                 if chunk is None:
                     ui.write(_('cannot edit patch for whole file'))
                     ui.write("\n")
                     continue
                 if chunk.header.binary():
                     ui.write(_('cannot edit patch for binary file'))
                     ui.write("\n")
                     continue
                 # Patch comment based on the Git one (based on comment at end of
                 # https://mercurial-scm.org/wiki/RecordExtension)
                 phelp = '---' + _("""
 To remove '-' lines, make them ' ' lines (context).
 To remove '+' lines, delete them.
 Lines starting with # will be removed from the patch.
 
 If the patch applies cleanly, the edited hunk will immediately be
 added to the record list. If it does not apply cleanly, a rejects
 file will be generated: you can use that when you try again. If
 all lines of the hunk are removed, then the edit is aborted and
 the hunk is left unchanged.
 """)
                 (patchfd, patchfn) = tempfile.mkstemp(prefix="hg-editor-",
                         suffix=".diff", text=True)
                 ncpatchfp = None
                 try:
                     # Write the initial patch
                     f = os.fdopen(patchfd, pycompat.sysstr("w"))
                     chunk.header.write(f)
                     chunk.write(f)
                     f.write('\n'.join(['# ' + i for i in phelp.splitlines()]))
                     f.close()
                     # Start the editor and wait for it to complete
                     editor = ui.geteditor()
                     ret = ui.system("%s \"%s\"" % (editor, patchfn),
                                     environ={'HGUSER': ui.username()},
                                     blockedtag='filterpatch')
                     if ret != 0:
                         ui.warn(_("editor exited with exit code %d\n") % ret)
                         continue
                     # Remove comment lines
                     patchfp = open(patchfn)
                     ncpatchfp = stringio()
                     for line in util.iterfile(patchfp):
                         if not line.startswith('#'):
                             ncpatchfp.write(line)
                     patchfp.close()
                     ncpatchfp.seek(0)
                     newpatches = parsepatch(ncpatchfp)
                 finally:
                     os.unlink(patchfn)
                     del ncpatchfp
                 # Signal that the chunk shouldn't be applied as-is, but
                 # provide the new patch to be used instead.
                 ret = False
             elif r == 3: # Skip
                 ret = skipfile = False
             elif r == 4: # file (Record remaining)
                 ret = skipfile = True
             elif r == 5: # done, skip remaining
                 ret = skipall = False
             elif r == 6: # all
                 ret = skipall = True
             elif r == 7: # quit
                 raise error.Abort(_('user quit'))
             return ret, skipfile, skipall, newpatches
 
     seen = set()
     applied = {}        # 'filename' -> [] of chunks
     skipfile, skipall = None, None
     pos, total = 1, sum(len(h.hunks) for h in headers)
     for h in headers:
         pos += len(h.hunks)
         skipfile = None
         fixoffset = 0
         hdr = ''.join(h.header)
         if hdr in seen:
             continue
         seen.add(hdr)
         if skipall is None:
             h.pretty(ui)
         msg = (_('examine changes to %s?') %
                _(' and ').join("'%s'" % f for f in h.files()))
         r, skipfile, skipall, np = prompt(skipfile, skipall, msg, None)
         if not r:
             continue
         applied[h.filename()] = [h]
         if h.allhunks():
             applied[h.filename()] += h.hunks
             continue
         for i, chunk in enumerate(h.hunks):
             if skipfile is None and skipall is None:
                 chunk.pretty(ui)
             if total == 1:
                 msg = messages['single'][operation] % chunk.filename()
             else:
                 idx = pos - len(h.hunks) + i
                 msg = messages['multiple'][operation] % (idx, total,
                                                          chunk.filename())
             r, skipfile, skipall, newpatches = prompt(skipfile,
                     skipall, msg, chunk)
             if r:
                 if fixoffset:
                     chunk = copy.copy(chunk)
                     chunk.toline += fixoffset
                 applied[chunk.filename()].append(chunk)
             elif newpatches is not None:
                 for newpatch in newpatches:
                     for newhunk in newpatch.hunks:
                         if fixoffset:
                             newhunk.toline += fixoffset
                         applied[newhunk.filename()].append(newhunk)
             else:
                 fixoffset += chunk.removed - chunk.added
     return (sum([h for h in applied.itervalues()
                if h[0].special() or len(h) > 1], []), {})
 class hunk(object):
     def __init__(self, desc, num, lr, context):
         self.number = num
         self.desc = desc
         self.hunk = [desc]
         self.a = []
         self.b = []
         self.starta = self.lena = None
         self.startb = self.lenb = None
         if lr is not None:
             if context:
                 self.read_context_hunk(lr)
             else:
                 self.read_unified_hunk(lr)
 
     def getnormalized(self):
         """Return a copy with line endings normalized to LF."""
 
         def normalize(lines):
             nlines = []
             for line in lines:
                 if line.endswith('\r\n'):
                     line = line[:-2] + '\n'
                 nlines.append(line)
             return nlines
 
         # Dummy object, it is rebuilt manually
         nh = hunk(self.desc, self.number, None, None)
         nh.number = self.number
         nh.desc = self.desc
         nh.hunk = self.hunk
         nh.a = normalize(self.a)
         nh.b = normalize(self.b)
         nh.starta = self.starta
         nh.startb = self.startb
         nh.lena = self.lena
         nh.lenb = self.lenb
         return nh
 
     def read_unified_hunk(self, lr):
         m = unidesc.match(self.desc)
         if not m:
             raise PatchError(_("bad hunk #%d") % self.number)
         self.starta, self.lena, self.startb, self.lenb = m.groups()
         if self.lena is None:
             self.lena = 1
         else:
             self.lena = int(self.lena)
         if self.lenb is None:
             self.lenb = 1
         else:
             self.lenb = int(self.lenb)
         self.starta = int(self.starta)
         self.startb = int(self.startb)
         diffhelpers.addlines(lr, self.hunk, self.lena, self.lenb, self.a,
                              self.b)
         # if we hit eof before finishing out the hunk, the last line will
         # be zero length.  Lets try to fix it up.
         while len(self.hunk[-1]) == 0:
             del self.hunk[-1]
             del self.a[-1]
             del self.b[-1]
             self.lena -= 1
             self.lenb -= 1
         self._fixnewline(lr)
 
     def read_context_hunk(self, lr):
         self.desc = lr.readline()
         m = contextdesc.match(self.desc)
         if not m:
             raise PatchError(_("bad hunk #%d") % self.number)
         self.starta, aend = m.groups()
         self.starta = int(self.starta)
         if aend is None:
             aend = self.starta
         self.lena = int(aend) - self.starta
         if self.starta:
             self.lena += 1
         for x in xrange(self.lena):
             l = lr.readline()
             if l.startswith('---'):
                 # lines addition, old block is empty
                 lr.push(l)
                 break
             s = l[2:]
             if l.startswith('- ') or l.startswith('! '):
                 u = '-' + s
             elif l.startswith('  '):
                 u = ' ' + s
             else:
                 raise PatchError(_("bad hunk #%d old text line %d") %
                                  (self.number, x))
             self.a.append(u)
             self.hunk.append(u)
 
         l = lr.readline()
         if l.startswith('\ '):
             s = self.a[-1][:-1]
             self.a[-1] = s
             self.hunk[-1] = s
             l = lr.readline()
         m = contextdesc.match(l)
         if not m:
             raise PatchError(_("bad hunk #%d") % self.number)
         self.startb, bend = m.groups()
         self.startb = int(self.startb)
         if bend is None:
             bend = self.startb
         self.lenb = int(bend) - self.startb
         if self.startb:
             self.lenb += 1
         hunki = 1
         for x in xrange(self.lenb):
             l = lr.readline()
             if l.startswith('\ '):
                 # XXX: the only way to hit this is with an invalid line range.
                 # The no-eol marker is not counted in the line range, but I
                 # guess there are diff(1) out there which behave differently.
                 s = self.b[-1][:-1]
                 self.b[-1] = s
                 self.hunk[hunki - 1] = s
                 continue
             if not l:
                 # line deletions, new block is empty and we hit EOF
                 lr.push(l)
                 break
             s = l[2:]
             if l.startswith('+ ') or l.startswith('! '):
                 u = '+' + s
             elif l.startswith('  '):
                 u = ' ' + s
             elif len(self.b) == 0:
                 # line deletions, new block is empty
                 lr.push(l)
                 break
             else:
                 raise PatchError(_("bad hunk #%d old text line %d") %
                                  (self.number, x))
             self.b.append(s)
             while True:
                 if hunki >= len(self.hunk):
                     h = ""
                 else:
                     h = self.hunk[hunki]
                 hunki += 1
                 if h == u:
                     break
                 elif h.startswith('-'):
                     continue
                 else:
                     self.hunk.insert(hunki - 1, u)
                     break
 
         if not self.a:
             # this happens when lines were only added to the hunk
             for x in self.hunk:
                 if x.startswith('-') or x.startswith(' '):
                     self.a.append(x)
         if not self.b:
             # this happens when lines were only deleted from the hunk
             for x in self.hunk:
                 if x.startswith('+') or x.startswith(' '):
                     self.b.append(x[1:])
         # @@ -start,len +start,len @@
         self.desc = "@@ -%d,%d +%d,%d @@\n" % (self.starta, self.lena,
                                              self.startb, self.lenb)
         self.hunk[0] = self.desc
         self._fixnewline(lr)
 
     def _fixnewline(self, lr):
         l = lr.readline()
         if l.startswith('\ '):
             diffhelpers.fix_newline(self.hunk, self.a, self.b)
         else:
             lr.push(l)
 
     def complete(self):
         return len(self.a) == self.lena and len(self.b) == self.lenb
 
     def _fuzzit(self, old, new, fuzz, toponly):
         # this removes context lines from the top and bottom of list 'l'.  It
         # checks the hunk to make sure only context lines are removed, and then
         # returns a new shortened list of lines.
         fuzz = min(fuzz, len(old))
         if fuzz:
             top = 0
             bot = 0
             hlen = len(self.hunk)
             for x in xrange(hlen - 1):
                 # the hunk starts with the @@ line, so use x+1
                 if self.hunk[x + 1][0] == ' ':
                     top += 1
                 else:
                     break
             if not toponly:
                 for x in xrange(hlen - 1):
                     if self.hunk[hlen - bot - 1][0] == ' ':
                         bot += 1
                     else:
                         break
 
             bot = min(fuzz, bot)
             top = min(fuzz, top)
             return old[top:len(old) - bot], new[top:len(new) - bot], top
         return old, new, 0
 
     def fuzzit(self, fuzz, toponly):
         old, new, top = self._fuzzit(self.a, self.b, fuzz, toponly)
         oldstart = self.starta + top
         newstart = self.startb + top
         # zero length hunk ranges already have their start decremented
         if self.lena and oldstart > 0:
             oldstart -= 1
         if self.lenb and newstart > 0:
             newstart -= 1
         return old, oldstart, new, newstart
 
 class binhunk(object):
     'A binary patch file.'
     def __init__(self, lr, fname):
         self.text = None
         self.delta = False
         self.hunk = ['GIT binary patch\n']
         self._fname = fname
         self._read(lr)
 
     def complete(self):
         return self.text is not None
 
     def new(self, lines):
         if self.delta:
             return [applybindelta(self.text, ''.join(lines))]
         return [self.text]
 
     def _read(self, lr):
         def getline(lr, hunk):
             l = lr.readline()
             hunk.append(l)
             return l.rstrip('\r\n')
 
         size = 0
         while True:
             line = getline(lr, self.hunk)
             if not line:
                 raise PatchError(_('could not extract "%s" binary data')
                                  % self._fname)
             if line.startswith('literal '):
                 size = int(line[8:].rstrip())
                 break
             if line.startswith('delta '):
                 size = int(line[6:].rstrip())
                 self.delta = True
                 break
         dec = []
         line = getline(lr, self.hunk)
         while len(line) > 1:
             l = line[0]
             if l <= 'Z' and l >= 'A':
                 l = ord(l) - ord('A') + 1
             else:
                 l = ord(l) - ord('a') + 27
             try:
                 dec.append(util.b85decode(line[1:])[:l])
             except ValueError as e:
                 raise PatchError(_('could not decode "%s" binary patch: %s')
                                  % (self._fname, str(e)))
             line = getline(lr, self.hunk)
         text = zlib.decompress(''.join(dec))
         if len(text) != size:
             raise PatchError(_('"%s" length is %d bytes, should be %d')
                              % (self._fname, len(text), size))
         self.text = text
 
 def parsefilename(str):
     # --- filename \t|space stuff
     s = str[4:].rstrip('\r\n')
     i = s.find('\t')
     if i < 0:
         i = s.find(' ')
         if i < 0:
             return s
     return s[:i]
 
 def reversehunks(hunks):
     '''reverse the signs in the hunks given as argument
 
     This function operates on hunks coming out of patch.filterpatch, that is
     a list of the form: [header1, hunk1, hunk2, header2...]. Example usage:
 
     >>> rawpatch = b"""diff --git a/folder1/g b/folder1/g
     ... --- a/folder1/g
     ... +++ b/folder1/g
     ... @@ -1,7 +1,7 @@
     ... +firstline
     ...  c
     ...  1
     ...  2
     ... + 3
     ... -4
     ...  5
     ...  d
     ... +lastline"""
     >>> hunks = parsepatch([rawpatch])
     >>> hunkscomingfromfilterpatch = []
     >>> for h in hunks:
     ...     hunkscomingfromfilterpatch.append(h)
     ...     hunkscomingfromfilterpatch.extend(h.hunks)
 
     >>> reversedhunks = reversehunks(hunkscomingfromfilterpatch)
     >>> from . import util
     >>> fp = util.stringio()
     >>> for c in reversedhunks:
     ...      c.write(fp)
     >>> fp.seek(0) or None
     >>> reversedpatch = fp.read()
     >>> print(pycompat.sysstr(reversedpatch))
     diff --git a/folder1/g b/folder1/g
     --- a/folder1/g
     +++ b/folder1/g
     @@ -1,4 +1,3 @@
     -firstline
      c
      1
      2
     @@ -2,6 +1,6 @@
      c
      1
      2
     - 3
     +4
      5
      d
     @@ -6,3 +5,2 @@
      5
      d
     -lastline
 
     '''
 
     newhunks = []
     for c in hunks:
         if util.safehasattr(c, 'reversehunk'):
             c = c.reversehunk()
         newhunks.append(c)
     return newhunks
 
 def parsepatch(originalchunks, maxcontext=None):
     """patch -> [] of headers -> [] of hunks
 
     If maxcontext is not None, trim context lines if necessary.
 
     >>> rawpatch = b'''diff --git a/folder1/g b/folder1/g
     ... --- a/folder1/g
     ... +++ b/folder1/g
     ... @@ -1,8 +1,10 @@
     ...  1
     ...  2
     ... -3
     ...  4
     ...  5
     ...  6
     ... +6.1
     ... +6.2
     ...  7
     ...  8
     ... +9'''
     >>> out = util.stringio()
     >>> headers = parsepatch([rawpatch], maxcontext=1)
     >>> for header in headers:
     ...     header.write(out)
     ...     for hunk in header.hunks:
     ...         hunk.write(out)
     >>> print(pycompat.sysstr(out.getvalue()))
     diff --git a/folder1/g b/folder1/g
     --- a/folder1/g
     +++ b/folder1/g
     @@ -2,3 +2,2 @@
      2
     -3
      4
     @@ -6,2 +5,4 @@
      6
     +6.1
     +6.2
      7
     @@ -8,1 +9,2 @@
      8
     +9
     """
     class parser(object):
         """patch parsing state machine"""
         def __init__(self):
             self.fromline = 0
             self.toline = 0
             self.proc = ''
             self.header = None
             self.context = []
             self.before = []
             self.hunk = []
             self.headers = []
 
         def addrange(self, limits):
             fromstart, fromend, tostart, toend, proc = limits
             self.fromline = int(fromstart)
             self.toline = int(tostart)
             self.proc = proc
 
         def addcontext(self, context):
             if self.hunk:
                 h = recordhunk(self.header, self.fromline, self.toline,
                         self.proc, self.before, self.hunk, context, maxcontext)
                 self.header.hunks.append(h)
                 self.fromline += len(self.before) + h.removed
                 self.toline += len(self.before) + h.added
                 self.before = []
                 self.hunk = []
             self.context = context
 
         def addhunk(self, hunk):
             if self.context:
                 self.before = self.context
                 self.context = []
             self.hunk = hunk
 
         def newfile(self, hdr):
             self.addcontext([])
             h = header(hdr)
             self.headers.append(h)
             self.header = h
 
         def addother(self, line):
             pass # 'other' lines are ignored
 
         def finished(self):
             self.addcontext([])
             return self.headers
 
         transitions = {
             'file': {'context': addcontext,
                      'file': newfile,
                      'hunk': addhunk,
                      'range': addrange},
             'context': {'file': newfile,
                         'hunk': addhunk,
                         'range': addrange,
                         'other': addother},
             'hunk': {'context': addcontext,
                      'file': newfile,
                      'range': addrange},
             'range': {'context': addcontext,
                       'hunk': addhunk},
             'other': {'other': addother},
             }
 
     p = parser()
     fp = stringio()
     fp.write(''.join(originalchunks))
     fp.seek(0)
 
     state = 'context'
     for newstate, data in scanpatch(fp):
         try:
             p.transitions[state][newstate](p, data)
         except KeyError:
             raise PatchError('unhandled transition: %s -> %s' %
                                    (state, newstate))
         state = newstate
     del fp
     return p.finished()
 
 def pathtransform(path, strip, prefix):
     '''turn a path from a patch into a path suitable for the repository
 
     prefix, if not empty, is expected to be normalized with a / at the end.
 
     Returns (stripped components, path in repository).
 
     >>> pathtransform(b'a/b/c', 0, b'')
     ('', 'a/b/c')
     >>> pathtransform(b'   a/b/c   ', 0, b'')
     ('', '   a/b/c')
     >>> pathtransform(b'   a/b/c   ', 2, b'')
     ('a/b/', 'c')
     >>> pathtransform(b'a/b/c', 0, b'd/e/')
     ('', 'd/e/a/b/c')
     >>> pathtransform(b'   a//b/c   ', 2, b'd/e/')
     ('a//b/', 'd/e/c')
     >>> pathtransform(b'a/b/c', 3, b'')
     Traceback (most recent call last):
     PatchError: unable to strip away 1 of 3 dirs from a/b/c
     '''
     pathlen = len(path)
     i = 0
     if strip == 0:
         return '', prefix + path.rstrip()
     count = strip
     while count > 0:
         i = path.find('/', i)
         if i == -1:
             raise PatchError(_("unable to strip away %d of %d dirs from %s") %
                              (count, strip, path))
         i += 1
         # consume '//' in the path
         while i < pathlen - 1 and path[i:i + 1] == '/':
             i += 1
         count -= 1
     return path[:i].lstrip(), prefix + path[i:].rstrip()
 
 def makepatchmeta(backend, afile_orig, bfile_orig, hunk, strip, prefix):
     nulla = afile_orig == "/dev/null"
     nullb = bfile_orig == "/dev/null"
     create = nulla and hunk.starta == 0 and hunk.lena == 0
     remove = nullb and hunk.startb == 0 and hunk.lenb == 0
     abase, afile = pathtransform(afile_orig, strip, prefix)
     gooda = not nulla and backend.exists(afile)
     bbase, bfile = pathtransform(bfile_orig, strip, prefix)
     if afile == bfile:
         goodb = gooda
     else:
         goodb = not nullb and backend.exists(bfile)
     missing = not goodb and not gooda and not create
 
     # some diff programs apparently produce patches where the afile is
     # not /dev/null, but afile starts with bfile
     abasedir = afile[:afile.rfind('/') + 1]
     bbasedir = bfile[:bfile.rfind('/') + 1]
     if (missing and abasedir == bbasedir and afile.startswith(bfile)
         and hunk.starta == 0 and hunk.lena == 0):
         create = True
         missing = False
 
     # If afile is "a/b/foo" and bfile is "a/b/foo.orig" we assume the
     # diff is between a file and its backup. In this case, the original
     # file should be patched (see original mpatch code).
     isbackup = (abase == bbase and bfile.startswith(afile))
     fname = None
     if not missing:
         if gooda and goodb:
             if isbackup:
                 fname = afile
             else:
                 fname = bfile
         elif gooda:
             fname = afile
 
     if not fname:
         if not nullb:
             if isbackup:
                 fname = afile
             else:
                 fname = bfile
         elif not nulla:
             fname = afile
         else:
             raise PatchError(_("undefined source and destination files"))
 
     gp = patchmeta(fname)
     if create:
         gp.op = 'ADD'
     elif remove:
         gp.op = 'DELETE'
     return gp
 
 def scanpatch(fp):
     """like patch.iterhunks, but yield different events
 
     - ('file',    [header_lines + fromfile + tofile])
     - ('context', [context_lines])
     - ('hunk',    [hunk_lines])
     - ('range',   (-start,len, +start,len, proc))
     """
     lines_re = re.compile(br'@@ -(\d+),(\d+) \+(\d+),(\d+) @@\s*(.*)')
     lr = linereader(fp)
 
     def scanwhile(first, p):
         """scan lr while predicate holds"""
         lines = [first]
         for line in iter(lr.readline, ''):
             if p(line):
                 lines.append(line)
             else:
                 lr.push(line)
                 break
         return lines
 
     for line in iter(lr.readline, ''):
         if line.startswith('diff --git a/') or line.startswith('diff -r '):
             def notheader(line):
                 s = line.split(None, 1)
                 return not s or s[0] not in ('---', 'diff')
             header = scanwhile(line, notheader)
             fromfile = lr.readline()
             if fromfile.startswith('---'):
                 tofile = lr.readline()
                 header += [fromfile, tofile]
             else:
                 lr.push(fromfile)
             yield 'file', header
         elif line[0:1] == ' ':
             yield 'context', scanwhile(line, lambda l: l[0] in ' \\')
         elif line[0] in '-+':
             yield 'hunk', scanwhile(line, lambda l: l[0] in '-+\\')
         else:
             m = lines_re.match(line)
             if m:
                 yield 'range', m.groups()
             else:
                 yield 'other', line
 
 def scangitpatch(lr, firstline):
     """
     Git patches can emit:
     - rename a to b
     - change b
     - copy a to c
     - change c
 
     We cannot apply this sequence as-is, the renamed 'a' could not be
     found for it would have been renamed already. And we cannot copy
     from 'b' instead because 'b' would have been changed already. So
     we scan the git patch for copy and rename commands so we can
     perform the copies ahead of time.
     """
     pos = 0
     try:
         pos = lr.fp.tell()
         fp = lr.fp
     except IOError:
         fp = stringio(lr.fp.read())
     gitlr = linereader(fp)
     gitlr.push(firstline)
     gitpatches = readgitpatch(gitlr)
     fp.seek(pos)
     return gitpatches
 
 def iterhunks(fp):
     """Read a patch and yield the following events:
     - ("file", afile, bfile, firsthunk): select a new target file.
     - ("hunk", hunk): a new hunk is ready to be applied, follows a
     "file" event.
     - ("git", gitchanges): current diff is in git format, gitchanges
     maps filenames to gitpatch records. Unique event.
     """
     afile = ""
     bfile = ""
     state = None
     hunknum = 0
     emitfile = newfile = False
     gitpatches = None
 
     # our states
     BFILE = 1
     context = None
     lr = linereader(fp)
 
     for x in iter(lr.readline, ''):
         if state == BFILE and (
             (not context and x[0] == '@')
             or (context is not False and x.startswith('***************'))
             or x.startswith('GIT binary patch')):
             gp = None
             if (gitpatches and
                 gitpatches[-1].ispatching(afile, bfile)):
                 gp = gitpatches.pop()
             if x.startswith('GIT binary patch'):
                 h = binhunk(lr, gp.path)
             else:
                 if context is None and x.startswith('***************'):
                     context = True
                 h = hunk(x, hunknum + 1, lr, context)
             hunknum += 1
             if emitfile:
                 emitfile = False
                 yield 'file', (afile, bfile, h, gp and gp.copy() or None)
             yield 'hunk', h
         elif x.startswith('diff --git a/'):
             m = gitre.match(x.rstrip(' \r\n'))
             if not m:
                 continue
             if gitpatches is None:
                 # scan whole input for git metadata
                 gitpatches = scangitpatch(lr, x)
                 yield 'git', [g.copy() for g in gitpatches
                               if g.op in ('COPY', 'RENAME')]
                 gitpatches.reverse()
             afile = 'a/' + m.group(1)
             bfile = 'b/' + m.group(2)
             while gitpatches and not gitpatches[-1].ispatching(afile, bfile):
                 gp = gitpatches.pop()
                 yield 'file', ('a/' + gp.path, 'b/' + gp.path, None, gp.copy())
             if not gitpatches:
                 raise PatchError(_('failed to synchronize metadata for "%s"')
                                  % afile[2:])
             gp = gitpatches[-1]
             newfile = True
         elif x.startswith('---'):
             # check for a unified diff
             l2 = lr.readline()
             if not l2.startswith('+++'):
                 lr.push(l2)
                 continue
             newfile = True
             context = False
             afile = parsefilename(x)
             bfile = parsefilename(l2)
         elif x.startswith('***'):
             # check for a context diff
             l2 = lr.readline()
             if not l2.startswith('---'):
                 lr.push(l2)
                 continue
             l3 = lr.readline()
             lr.push(l3)
             if not l3.startswith("***************"):
                 lr.push(l2)
                 continue
             newfile = True
             context = True
             afile = parsefilename(x)
             bfile = parsefilename(l2)
 
         if newfile:
             newfile = False
             emitfile = True
             state = BFILE
             hunknum = 0
 
     while gitpatches:
         gp = gitpatches.pop()
         yield 'file', ('a/' + gp.path, 'b/' + gp.path, None, gp.copy())
 
 def applybindelta(binchunk, data):
     """Apply a binary delta hunk
     The algorithm used is the algorithm from git's patch-delta.c
     """
     def deltahead(binchunk):
         i = 0
         for c in binchunk:
             i += 1
             if not (ord(c) & 0x80):
                 return i
         return i
     out = ""
     s = deltahead(binchunk)
     binchunk = binchunk[s:]
     s = deltahead(binchunk)
     binchunk = binchunk[s:]
     i = 0
     while i < len(binchunk):
         cmd = ord(binchunk[i])
         i += 1
         if (cmd & 0x80):
             offset = 0
             size = 0
             if (cmd & 0x01):
                 offset = ord(binchunk[i])
                 i += 1
             if (cmd & 0x02):
                 offset |= ord(binchunk[i]) << 8
                 i += 1
             if (cmd & 0x04):
                 offset |= ord(binchunk[i]) << 16
                 i += 1
             if (cmd & 0x08):
                 offset |= ord(binchunk[i]) << 24
                 i += 1
             if (cmd & 0x10):
                 size = ord(binchunk[i])
                 i += 1
             if (cmd & 0x20):
                 size |= ord(binchunk[i]) << 8
                 i += 1
             if (cmd & 0x40):
                 size |= ord(binchunk[i]) << 16
                 i += 1
             if size == 0:
                 size = 0x10000
             offset_end = offset + size
             out += data[offset:offset_end]
         elif cmd != 0:
             offset_end = i + cmd
             out += binchunk[i:offset_end]
             i += cmd
         else:
             raise PatchError(_('unexpected delta opcode 0'))
     return out
 
 def applydiff(ui, fp, backend, store, strip=1, prefix='', eolmode='strict'):
     """Reads a patch from fp and tries to apply it.
 
     Returns 0 for a clean patch, -1 if any rejects were found and 1 if
     there was any fuzz.
 
     If 'eolmode' is 'strict', the patch content and patched file are
     read in binary mode. Otherwise, line endings are ignored when
     patching then normalized according to 'eolmode'.
     """
     return _applydiff(ui, fp, patchfile, backend, store, strip=strip,
                       prefix=prefix, eolmode=eolmode)
 
 def _canonprefix(repo, prefix):
     if prefix:
         prefix = pathutil.canonpath(repo.root, repo.getcwd(), prefix)
         if prefix != '':
             prefix += '/'
     return prefix
 
 def _applydiff(ui, fp, patcher, backend, store, strip=1, prefix='',
                eolmode='strict'):
     prefix = _canonprefix(backend.repo, prefix)
     def pstrip(p):
         return pathtransform(p, strip - 1, prefix)[1]
 
     rejects = 0
     err = 0
     current_file = None
 
     for state, values in iterhunks(fp):
         if state == 'hunk':
             if not current_file:
                 continue
             ret = current_file.apply(values)
             if ret > 0:
                 err = 1
         elif state == 'file':
             if current_file:
                 rejects += current_file.close()
                 current_file = None
             afile, bfile, first_hunk, gp = values
             if gp:
                 gp.path = pstrip(gp.path)
                 if gp.oldpath:
                     gp.oldpath = pstrip(gp.oldpath)
             else:
                 gp = makepatchmeta(backend, afile, bfile, first_hunk, strip,
                                    prefix)
             if gp.op == 'RENAME':
                 backend.unlink(gp.oldpath)
             if not first_hunk:
                 if gp.op == 'DELETE':
                     backend.unlink(gp.path)
                     continue
                 data, mode = None, None
                 if gp.op in ('RENAME', 'COPY'):
                     data, mode = store.getfile(gp.oldpath)[:2]
                     if data is None:
                         # This means that the old path does not exist
                         raise PatchError(_("source file '%s' does not exist")
                                            % gp.oldpath)
                 if gp.mode:
                     mode = gp.mode
                     if gp.op == 'ADD':
                         # Added files without content have no hunk and
                         # must be created
                         data = ''
                 if data or mode:
                     if (gp.op in ('ADD', 'RENAME', 'COPY')
                         and backend.exists(gp.path)):
                         raise PatchError(_("cannot create %s: destination "
                                            "already exists") % gp.path)
                     backend.setfile(gp.path, data, mode, gp.oldpath)
                 continue
             try:
                 current_file = patcher(ui, gp, backend, store,
                                        eolmode=eolmode)
             except PatchError as inst:
                 ui.warn(str(inst) + '\n')
                 current_file = None
                 rejects += 1
                 continue
         elif state == 'git':
             for gp in values:
                 path = pstrip(gp.oldpath)
                 data, mode = backend.getfile(path)
                 if data is None:
                     # The error ignored here will trigger a getfile()
                     # error in a place more appropriate for error
                     # handling, and will not interrupt the patching
                     # process.
                     pass
                 else:
                     store.setfile(path, data, mode)
         else:
             raise error.Abort(_('unsupported parser state: %s') % state)
 
     if current_file:
         rejects += current_file.close()
 
     if rejects:
         return -1
     return err
 
 def _externalpatch(ui, repo, patcher, patchname, strip, files,
                    similarity):
     """use <patcher> to apply <patchname> to the working directory.
     returns whether patch was applied with fuzz factor."""
 
     fuzz = False
     args = []
     cwd = repo.root
     if cwd:
         args.append('-d %s' % util.shellquote(cwd))
     fp = util.popen('%s %s -p%d < %s' % (patcher, ' '.join(args), strip,
                                        util.shellquote(patchname)))
     try:
         for line in util.iterfile(fp):
             line = line.rstrip()
             ui.note(line + '\n')
             if line.startswith('patching file '):
                 pf = util.parsepatchoutput(line)
                 printed_file = False
                 files.add(pf)
             elif line.find('with fuzz') >= 0:
                 fuzz = True
                 if not printed_file:
                     ui.warn(pf + '\n')
                     printed_file = True
                 ui.warn(line + '\n')
             elif line.find('saving rejects to file') >= 0:
                 ui.warn(line + '\n')
             elif line.find('FAILED') >= 0:
                 if not printed_file:
                     ui.warn(pf + '\n')
                     printed_file = True
                 ui.warn(line + '\n')
     finally:
         if files:
             scmutil.marktouched(repo, files, similarity)
     code = fp.close()
     if code:
         raise PatchError(_("patch command failed: %s") %
                          util.explainexit(code)[0])
     return fuzz
 
 def patchbackend(ui, backend, patchobj, strip, prefix, files=None,
                  eolmode='strict'):
     if files is None:
         files = set()
     if eolmode is None:
         eolmode = ui.config('patch', 'eol')
     if eolmode.lower() not in eolmodes:
         raise error.Abort(_('unsupported line endings type: %s') % eolmode)
     eolmode = eolmode.lower()
 
     store = filestore()
     try:
         fp = open(patchobj, 'rb')
     except TypeError:
         fp = patchobj
     try:
         ret = applydiff(ui, fp, backend, store, strip=strip, prefix=prefix,
                         eolmode=eolmode)
     finally:
         if fp != patchobj:
             fp.close()
         files.update(backend.close())
         store.close()
     if ret < 0:
         raise PatchError(_('patch failed to apply'))
     return ret > 0
 
 def internalpatch(ui, repo, patchobj, strip, prefix='', files=None,
                   eolmode='strict', similarity=0):
     """use builtin patch to apply <patchobj> to the working directory.
     returns whether patch was applied with fuzz factor."""
     backend = workingbackend(ui, repo, similarity)
     return patchbackend(ui, backend, patchobj, strip, prefix, files, eolmode)
 
 def patchrepo(ui, repo, ctx, store, patchobj, strip, prefix, files=None,
               eolmode='strict'):
     backend = repobackend(ui, repo, ctx, store)
     return patchbackend(ui, backend, patchobj, strip, prefix, files, eolmode)
 
 def patch(ui, repo, patchname, strip=1, prefix='', files=None, eolmode='strict',
           similarity=0):
     """Apply <patchname> to the working directory.
 
     'eolmode' specifies how end of lines should be handled. It can be:
     - 'strict': inputs are read in binary mode, EOLs are preserved
     - 'crlf': EOLs are ignored when patching and reset to CRLF
     - 'lf': EOLs are ignored when patching and reset to LF
     - None: get it from user settings, default to 'strict'
     'eolmode' is ignored when using an external patcher program.
 
     Returns whether patch was applied with fuzz factor.
     """
     patcher = ui.config('ui', 'patch')
     if files is None:
         files = set()
     if patcher:
         return _externalpatch(ui, repo, patcher, patchname, strip,
                               files, similarity)
     return internalpatch(ui, repo, patchname, strip, prefix, files, eolmode,
                          similarity)
 
 def changedfiles(ui, repo, patchpath, strip=1, prefix=''):
     backend = fsbackend(ui, repo.root)
     prefix = _canonprefix(repo, prefix)
     with open(patchpath, 'rb') as fp:
         changed = set()
         for state, values in iterhunks(fp):
             if state == 'file':
                 afile, bfile, first_hunk, gp = values
                 if gp:
                     gp.path = pathtransform(gp.path, strip - 1, prefix)[1]
                     if gp.oldpath:
                         gp.oldpath = pathtransform(gp.oldpath, strip - 1,
                                                    prefix)[1]
                 else:
                     gp = makepatchmeta(backend, afile, bfile, first_hunk, strip,
                                        prefix)
                 changed.add(gp.path)
                 if gp.op == 'RENAME':
                     changed.add(gp.oldpath)
             elif state not in ('hunk', 'git'):
                 raise error.Abort(_('unsupported parser state: %s') % state)
         return changed
 
 class GitDiffRequired(Exception):
     pass
 
 def diffallopts(ui, opts=None, untrusted=False, section='diff'):
     '''return diffopts with all features supported and parsed'''
     return difffeatureopts(ui, opts=opts, untrusted=untrusted, section=section,
                            git=True, whitespace=True, formatchanging=True)
 
 diffopts = diffallopts
 
 def difffeatureopts(ui, opts=None, untrusted=False, section='diff', git=False,
                     whitespace=False, formatchanging=False):
     '''return diffopts with only opted-in features parsed
 
     Features:
     - git: git-style diffs
     - whitespace: whitespace options like ignoreblanklines and ignorews
     - formatchanging: options that will likely break or cause correctness issues
       with most diff parsers
     '''
     def get(key, name=None, getter=ui.configbool, forceplain=None):
         if opts:
             v = opts.get(key)
             # diffopts flags are either None-default (which is passed
             # through unchanged, so we can identify unset values), or
             # some other falsey default (eg --unified, which defaults
             # to an empty string). We only want to override the config
             # entries from hgrc with command line values if they
             # appear to have been set, which is any truthy value,
             # True, or False.
             if v or isinstance(v, bool):
                 return v
         if forceplain is not None and ui.plain():
             return forceplain
         return getter(section, name or key, untrusted=untrusted)
 
     # core options, expected to be understood by every diff parser
     buildopts = {
         'nodates': get('nodates'),
         'showfunc': get('show_function', 'showfunc'),
         'context': get('unified', getter=ui.config),
     }
     buildopts['worddiff'] = ui.configbool('experimental', 'worddiff')
 
     if git:
         buildopts['git'] = get('git')
 
         # since this is in the experimental section, we need to call
         # ui.configbool directory
         buildopts['showsimilarity'] = ui.configbool('experimental',
                                                     'extendedheader.similarity')
 
         # need to inspect the ui object instead of using get() since we want to
         # test for an int
         hconf = ui.config('experimental', 'extendedheader.index')
         if hconf is not None:
             hlen = None
             try:
                 # the hash config could be an integer (for length of hash) or a
                 # word (e.g. short, full, none)
                 hlen = int(hconf)
                 if hlen < 0 or hlen > 40:
                     msg = _("invalid length for extendedheader.index: '%d'\n")
                     ui.warn(msg % hlen)
             except ValueError:
                 # default value
                 if hconf == 'short' or hconf == '':
                     hlen = 12
                 elif hconf == 'full':
                     hlen = 40
                 elif hconf != 'none':
                     msg = _("invalid value for extendedheader.index: '%s'\n")
                     ui.warn(msg % hconf)
             finally:
                 buildopts['index'] = hlen
 
     if whitespace:
         buildopts['ignorews'] = get('ignore_all_space', 'ignorews')
         buildopts['ignorewsamount'] = get('ignore_space_change',
                                           'ignorewsamount')
         buildopts['ignoreblanklines'] = get('ignore_blank_lines',
                                             'ignoreblanklines')
         buildopts['ignorewseol'] = get('ignore_space_at_eol', 'ignorewseol')
     if formatchanging:
         buildopts['text'] = opts and opts.get('text')
         binary = None if opts is None else opts.get('binary')
         buildopts['nobinary'] = (not binary if binary is not None
                                  else get('nobinary', forceplain=False))
         buildopts['noprefix'] = get('noprefix', forceplain=False)
 
     return mdiff.diffopts(**pycompat.strkwargs(buildopts))
 
 def diff(repo, node1=None, node2=None, match=None, changes=None,
          opts=None, losedatafn=None, prefix='', relroot='', copy=None,
          hunksfilterfn=None):
     '''yields diff of changes to files between two nodes, or node and
     working directory.
 
     if node1 is None, use first dirstate parent instead.
     if node2 is None, compare node1 with working directory.
 
     losedatafn(**kwarg) is a callable run when opts.upgrade=True and
     every time some change cannot be represented with the current
     patch format. Return False to upgrade to git patch format, True to
     accept the loss or raise an exception to abort the diff. It is
     called with the name of current file being diffed as 'fn'. If set
     to None, patches will always be upgraded to git format when
     necessary.
 
     prefix is a filename prefix that is prepended to all filenames on
     display (used for subrepos).
 
     relroot, if not empty, must be normalized with a trailing /. Any match
     patterns that fall outside it will be ignored.
 
     copy, if not empty, should contain mappings {dst@y: src@x} of copy
     information.
 
     hunksfilterfn, if not None, should be a function taking a filectx and
     hunks generator that may yield filtered hunks.
     '''
     for fctx1, fctx2, hdr, hunks in diffhunks(
             repo, node1=node1, node2=node2,
             match=match, changes=changes, opts=opts,
             losedatafn=losedatafn, prefix=prefix, relroot=relroot, copy=copy,
     ):
         if hunksfilterfn is not None:
             # If the file has been removed, fctx2 is None; but this should
             # not occur here since we catch removed files early in
             # logcmdutil.getlinerangerevs() for 'hg log -L'.
             assert fctx2 is not None, \
                 'fctx2 unexpectly None in diff hunks filtering'
             hunks = hunksfilterfn(fctx2, hunks)
         text = ''.join(sum((list(hlines) for hrange, hlines in hunks), []))
         if hdr and (text or len(hdr) > 1):
             yield '\n'.join(hdr) + '\n'
         if text:
             yield text
 
 def diffhunks(repo, node1=None, node2=None, match=None, changes=None,
               opts=None, losedatafn=None, prefix='', relroot='', copy=None):
     """Yield diff of changes to files in the form of (`header`, `hunks`) tuples
     where `header` is a list of diff headers and `hunks` is an iterable of
     (`hunkrange`, `hunklines`) tuples.
 
     See diff() for the meaning of parameters.
     """
 
     if opts is None:
         opts = mdiff.defaultopts
 
     if not node1 and not node2:
         node1 = repo.dirstate.p1()
 
     def lrugetfilectx():
         cache = {}
         order = collections.deque()
         def getfilectx(f, ctx):
             fctx = ctx.filectx(f, filelog=cache.get(f))
             if f not in cache:
                 if len(cache) > 20:
                     del cache[order.popleft()]
                 cache[f] = fctx.filelog()
             else:
                 order.remove(f)
             order.append(f)
             return fctx
         return getfilectx
     getfilectx = lrugetfilectx()
 
     ctx1 = repo[node1]
     ctx2 = repo[node2]
 
     relfiltered = False
     if relroot != '' and match.always():
         # as a special case, create a new matcher with just the relroot
         pats = [relroot]
         match = scmutil.match(ctx2, pats, default='path')
         relfiltered = True
 
     if not changes:
         changes = repo.status(ctx1, ctx2, match=match)
     modified, added, removed = changes[:3]
 
     if not modified and not added and not removed:
         return []
 
     if repo.ui.debugflag:
         hexfunc = hex
     else:
         hexfunc = short
     revs = [hexfunc(node) for node in [ctx1.node(), ctx2.node()] if node]
 
     if copy is None:
         copy = {}
         if opts.git or opts.upgrade:
             copy = copies.pathcopies(ctx1, ctx2, match=match)
 
     if relroot is not None:
         if not relfiltered:
             # XXX this would ideally be done in the matcher, but that is
             # generally meant to 'or' patterns, not 'and' them. In this case we
             # need to 'and' all the patterns from the matcher with relroot.
             def filterrel(l):
                 return [f for f in l if f.startswith(relroot)]
             modified = filterrel(modified)
             added = filterrel(added)
             removed = filterrel(removed)
             relfiltered = True
         # filter out copies where either side isn't inside the relative root
         copy = dict(((dst, src) for (dst, src) in copy.iteritems()
                      if dst.startswith(relroot)
                      and src.startswith(relroot)))
 
     modifiedset = set(modified)
     addedset = set(added)
     removedset = set(removed)
     for f in modified:
         if f not in ctx1:
             # Fix up added, since merged-in additions appear as
             # modifications during merges
             modifiedset.remove(f)
             addedset.add(f)
     for f in removed:
         if f not in ctx1:
             # Merged-in additions that are then removed are reported as removed.
             # They are not in ctx1, so We don't want to show them in the diff.
             removedset.remove(f)
     modified = sorted(modifiedset)
     added = sorted(addedset)
     removed = sorted(removedset)
     for dst, src in list(copy.items()):
         if src not in ctx1:
             # Files merged in during a merge and then copied/renamed are
             # reported as copies. We want to show them in the diff as additions.
             del copy[dst]
 
     def difffn(opts, losedata):
         return trydiff(repo, revs, ctx1, ctx2, modified, added, removed,
                        copy, getfilectx, opts, losedata, prefix, relroot)
     if opts.upgrade and not opts.git:
         try:
             def losedata(fn):
                 if not losedatafn or not losedatafn(fn=fn):
                     raise GitDiffRequired
             # Buffer the whole output until we are sure it can be generated
             return list(difffn(opts.copy(git=False), losedata))
         except GitDiffRequired:
             return difffn(opts.copy(git=True), None)
     else:
         return difffn(opts, None)
 
 def difflabel(func, *args, **kw):
     '''yields 2-tuples of (output, label) based on the output of func()'''
     inlinecolor = False
     if kw.get(r'opts'):
         inlinecolor = kw[r'opts'].worddiff
     headprefixes = [('diff', 'diff.diffline'),
                     ('copy', 'diff.extended'),
                     ('rename', 'diff.extended'),
                     ('old', 'diff.extended'),
                     ('new', 'diff.extended'),
                     ('deleted', 'diff.extended'),
                     ('index', 'diff.extended'),
                     ('similarity', 'diff.extended'),
                     ('---', 'diff.file_a'),
                     ('+++', 'diff.file_b')]
     textprefixes = [('@', 'diff.hunk'),
                     ('-', 'diff.deleted'),
                     ('+', 'diff.inserted')]
     head = False
     for chunk in func(*args, **kw):
         lines = chunk.split('\n')
         matches = {}
         if inlinecolor:
             matches = _findmatches(lines)
         for i, line in enumerate(lines):
             if i != 0:
                 yield ('\n', '')
             if head:
                 if line.startswith('@'):
                     head = False
             else:
                 if line and line[0] not in ' +-@\\':
                     head = True
             stripline = line
             diffline = False
             if not head and line and line[0] in '+-':
                 # highlight tabs and trailing whitespace, but only in
                 # changed lines
                 stripline = line.rstrip()
                 diffline = True
 
             prefixes = textprefixes
             if head:
                 prefixes = headprefixes
             for prefix, label in prefixes:
                 if stripline.startswith(prefix):
                     if diffline:
                         if i in matches:
                             for t, l in _inlinediff(lines[i].rstrip(),
                                                     lines[matches[i]].rstrip(),
                                                     label):
                                 yield (t, l)
                         else:
                             for token in tabsplitter.findall(stripline):
                                 if '\t' == token[0]:
                                     yield (token, 'diff.tab')
                                 else:
                                     yield (token, label)
                     else:
                         yield (stripline, label)
                     break
             else:
                 yield (line, '')
             if line != stripline:
                 yield (line[len(stripline):], 'diff.trailingwhitespace')
 
 def _findmatches(slist):
     '''Look for insertion matches to deletion and returns a dict of
     correspondences.
     '''
     lastmatch = 0
     matches = {}
     for i, line in enumerate(slist):
         if line == '':
             continue
         if line[0] == '-':
             lastmatch = max(lastmatch, i)
             newgroup = False
             for j, newline in enumerate(slist[lastmatch + 1:]):
                 if newline == '':
                     continue
                 if newline[0] == '-' and newgroup: # too far, no match
                     break
                 if newline[0] == '+': # potential match
                     newgroup = True
                     sim = difflib.SequenceMatcher(None, line, newline).ratio()
                     if sim > 0.7:
                         lastmatch = lastmatch + 1 + j
                         matches[i] = lastmatch
                         matches[lastmatch] = i
                         break
     return matches
 
 def _inlinediff(s1, s2, operation):
     '''Perform string diff to highlight specific changes.'''
     operation_skip = '+?' if operation == 'diff.deleted' else '-?'
     if operation == 'diff.deleted':
         s2, s1 = s1, s2
 
     buff = []
     # we never want to higlight the leading +-
     if operation == 'diff.deleted' and s2.startswith('-'):
         label = operation
         token = '-'
         s2 = s2[1:]
         s1 = s1[1:]
     elif operation == 'diff.inserted' and s1.startswith('+'):
         label = operation
         token = '+'
         s2 = s2[1:]
         s1 = s1[1:]
     else:
         raise error.ProgrammingError("Case not expected, operation = %s" %
                                      operation)
 
     s = difflib.ndiff(_nonwordre.split(s2), _nonwordre.split(s1))
     for part in s:
         if part[0] in operation_skip or len(part) == 2:
             continue
         l = operation + '.highlight'
         if part[0] in ' ':
             l = operation
         if part[2:] == '\t':
             l = 'diff.tab'
         if l == label: # contiguous token with same label
             token += part[2:]
             continue
         else:
             buff.append((token, label))
             label = l
             token = part[2:]
     buff.append((token, label))
 
     return buff
 
 def diffui(*args, **kw):
     '''like diff(), but yields 2-tuples of (output, label) for ui.write()'''
     return difflabel(diff, *args, **kw)
 
 def _filepairs(modified, added, removed, copy, opts):
     '''generates tuples (f1, f2, copyop), where f1 is the name of the file
     before and f2 is the the name after. For added files, f1 will be None,
     and for removed files, f2 will be None. copyop may be set to None, 'copy'
     or 'rename' (the latter two only if opts.git is set).'''
     gone = set()
 
     copyto = dict([(v, k) for k, v in copy.items()])
 
     addedset, removedset = set(added), set(removed)
 
     for f in sorted(modified + added + removed):
         copyop = None
         f1, f2 = f, f
         if f in addedset:
             f1 = None
             if f in copy:
                 if opts.git:
                     f1 = copy[f]
                     if f1 in removedset and f1 not in gone:
                         copyop = 'rename'
                         gone.add(f1)
                     else:
                         copyop = 'copy'
         elif f in removedset:
             f2 = None
             if opts.git:
                 # have we already reported a copy above?
                 if (f in copyto and copyto[f] in addedset
                     and copy[copyto[f]] == f):
                     continue
         yield f1, f2, copyop
 
 def trydiff(repo, revs, ctx1, ctx2, modified, added, removed,
             copy, getfilectx, opts, losedatafn, prefix, relroot):
     '''given input data, generate a diff and yield it in blocks
 
     If generating a diff would lose data like flags or binary data and
     losedatafn is not None, it will be called.
 
     relroot is removed and prefix is added to every path in the diff output.
 
     If relroot is not empty, this function expects every path in modified,
     added, removed and copy to start with it.'''
 
     def gitindex(text):
         if not text:
             text = ""
         l = len(text)
         s = hashlib.sha1('blob %d\0' % l)
         s.update(text)
         return hex(s.digest())
 
     if opts.noprefix:
         aprefix = bprefix = ''
     else:
         aprefix = 'a/'
         bprefix = 'b/'
 
     def diffline(f, revs):
         revinfo = ' '.join(["-r %s" % rev for rev in revs])
         return 'diff %s %s' % (revinfo, f)
 
     def isempty(fctx):
         return fctx is None or fctx.size() == 0
 
     date1 = util.datestr(ctx1.date())
     date2 = util.datestr(ctx2.date())
 
     gitmode = {'l': '120000', 'x': '100755', '': '100644'}
 
     if relroot != '' and (repo.ui.configbool('devel', 'all-warnings')
                           or repo.ui.configbool('devel', 'check-relroot')):
         for f in modified + added + removed + list(copy) + list(copy.values()):
             if f is not None and not f.startswith(relroot):
                 raise AssertionError(
                     "file %s doesn't start with relroot %s" % (f, relroot))
 
     for f1, f2, copyop in _filepairs(modified, added, removed, copy, opts):
         content1 = None
         content2 = None
         fctx1 = None
         fctx2 = None
         flag1 = None
         flag2 = None
         if f1:
             fctx1 = getfilectx(f1, ctx1)
             if opts.git or losedatafn:
                 flag1 = ctx1.flags(f1)
         if f2:
             fctx2 = getfilectx(f2, ctx2)
             if opts.git or losedatafn:
                 flag2 = ctx2.flags(f2)
         # if binary is True, output "summary" or "base85", but not "text diff"
         if opts.text:
-            check_binary = True
             binary = False
         else:
-            check_binary = any(f.isbinary()
-                               for f in [fctx1, fctx2] if f is not None)
-            binary = check_binary
+            binary = any(f.isbinary() for f in [fctx1, fctx2] if f is not None)
 
         if losedatafn and not opts.git:
             if (binary or
                 # copy/rename
                 f2 in copy or
                 # empty file creation
                 (not f1 and isempty(fctx2)) or
                 # empty file deletion
                 (isempty(fctx1) and not f2) or
                 # create with flags
                 (not f1 and flag2) or
                 # change flags
                 (f1 and f2 and flag1 != flag2)):
                 losedatafn(f2 or f1)
 
         path1 = f1 or f2
         path2 = f2 or f1
         path1 = posixpath.join(prefix, path1[len(relroot):])
         path2 = posixpath.join(prefix, path2[len(relroot):])
         header = []
         if opts.git:
             header.append('diff --git %s%s %s%s' %
                           (aprefix, path1, bprefix, path2))
             if not f1: # added
                 header.append('new file mode %s' % gitmode[flag2])
             elif not f2: # removed
                 header.append('deleted file mode %s' % gitmode[flag1])
             else:  # modified/copied/renamed
                 mode1, mode2 = gitmode[flag1], gitmode[flag2]
                 if mode1 != mode2:
                     header.append('old mode %s' % mode1)
                     header.append('new mode %s' % mode2)
                 if copyop is not None:
                     if opts.showsimilarity:
                         sim = similar.score(ctx1[path1], ctx2[path2]) * 100
                         header.append('similarity index %d%%' % sim)
                     header.append('%s from %s' % (copyop, path1))
                     header.append('%s to %s' % (copyop, path2))
         elif revs and not repo.ui.quiet:
             header.append(diffline(path1, revs))
 
         #  fctx.is  | diffopts                | what to   | is fctx.data()
         #  binary() | text nobinary git index | output?   | outputted?
         # ------------------------------------|----------------------------
         #  yes      | no   no       no  *     | summary   | no
         #  yes      | no   no       yes *     | base85    | yes
         #  yes      | no   yes      no  *     | summary   | no
         #  yes      | no   yes      yes 0     | summary   | no
         #  yes      | no   yes      yes >0    | summary   | semi [1]
         #  yes      | yes  *        *   *     | text diff | yes
         #  no       | *    *        *   *     | text diff | yes
         # [1]: hash(fctx.data()) is outputted. so fctx.data() cannot be faked
         if binary and (not opts.git or (opts.git and opts.nobinary and not
                                         opts.index)):
             # fast path: no binary content will be displayed, content1 and
             # content2 are only used for equivalent test. cmp() could have a
             # fast path.
             if fctx1 is not None:
                 content1 = b'\0'
             if fctx2 is not None:
                 if fctx1 is not None and not fctx1.cmp(fctx2):
                     content2 = b'\0' # not different
                 else:
                     content2 = b'\0\0'
         else:
             # normal path: load contents
             if fctx1 is not None:
                 content1 = fctx1.data()
             if fctx2 is not None:
                 content2 = fctx2.data()
 
         if binary and opts.git and not opts.nobinary:
             text = mdiff.b85diff(content1, content2)
             if text:
                 header.append('index %s..%s' %
                               (gitindex(content1), gitindex(content2)))
             hunks = (None, [text]),
         else:
             if opts.git and opts.index > 0:
                 flag = flag1
                 if flag is None:
                     flag = flag2
                 header.append('index %s..%s %s' %
                               (gitindex(content1)[0:opts.index],
                                gitindex(content2)[0:opts.index],
                                gitmode[flag]))
 
             uheaders, hunks = mdiff.unidiff(content1, date1,
                                             content2, date2,
-                                            path1, path2, opts=opts,
-                                            check_binary=check_binary)
+                                            path1, path2,
+                                            binary=binary, opts=opts)
             header.extend(uheaders)
         yield fctx1, fctx2, header, hunks
 
 def diffstatsum(stats):
     maxfile, maxtotal, addtotal, removetotal, binary = 0, 0, 0, 0, False
     for f, a, r, b in stats:
         maxfile = max(maxfile, encoding.colwidth(f))
         maxtotal = max(maxtotal, a + r)
         addtotal += a
         removetotal += r
         binary = binary or b
 
     return maxfile, maxtotal, addtotal, removetotal, binary
 
 def diffstatdata(lines):
     diffre = re.compile('^diff .*-r [a-z0-9]+\s(.*)$')
 
     results = []
     filename, adds, removes, isbinary = None, 0, 0, False
 
     def addresult():
         if filename:
             results.append((filename, adds, removes, isbinary))
 
     # inheader is used to track if a line is in the
     # header portion of the diff.  This helps properly account
     # for lines that start with '--' or '++'
     inheader = False
 
     for line in lines:
         if line.startswith('diff'):
             addresult()
             # starting a new file diff
             # set numbers to 0 and reset inheader
             inheader = True
             adds, removes, isbinary = 0, 0, False
             if line.startswith('diff --git a/'):
                 filename = gitre.search(line).group(2)
             elif line.startswith('diff -r'):
                 # format: "diff -r ... -r ... filename"
                 filename = diffre.search(line).group(1)
         elif line.startswith('@@'):
             inheader = False
         elif line.startswith('+') and not inheader:
             adds += 1
         elif line.startswith('-') and not inheader:
             removes += 1
         elif (line.startswith('GIT binary patch') or
               line.startswith('Binary file')):
             isbinary = True
     addresult()
     return results
 
 def diffstat(lines, width=80):
     output = []
     stats = diffstatdata(lines)
     maxname, maxtotal, totaladds, totalremoves, hasbinary = diffstatsum(stats)
 
     countwidth = len(str(maxtotal))
     if hasbinary and countwidth < 3:
         countwidth = 3
     graphwidth = width - countwidth - maxname - 6
     if graphwidth < 10:
         graphwidth = 10
 
     def scale(i):
         if maxtotal <= graphwidth:
             return i
         # If diffstat runs out of room it doesn't print anything,
         # which isn't very useful, so always print at least one + or -
         # if there were at least some changes.
         return max(i * graphwidth // maxtotal, int(bool(i)))
 
     for filename, adds, removes, isbinary in stats:
         if isbinary:
             count = 'Bin'
         else:
             count = '%d' % (adds + removes)
         pluses = '+' * scale(adds)
         minuses = '-' * scale(removes)
         output.append(' %s%s |  %*s %s%s\n' %
                       (filename, ' ' * (maxname - encoding.colwidth(filename)),
                        countwidth, count, pluses, minuses))
 
     if stats:
         output.append(_(' %d files changed, %d insertions(+), '
                         '%d deletions(-)\n')
                       % (len(stats), totaladds, totalremoves))
 
     return ''.join(output)
 
 def diffstatui(*args, **kw):
     '''like diffstat(), but yields 2-tuples of (output, label) for
     ui.write()
     '''
 
     for line in diffstat(*args, **kw).splitlines():
         if line and line[-1] in '+-':
             name, graph = line.rsplit(' ', 1)
             yield (name + ' ', '')
             m = re.search(br'\++', graph)
             if m:
                 yield (m.group(0), 'diffstat.inserted')
             m = re.search(br'-+', graph)
             if m:
                 yield (m.group(0), 'diffstat.deleted')
         else:
             yield (line, '')
         yield ('\n', '')