revlog: add an aggressivemergedelta option

This adds an option for delta'ing against both p1 and p2 when applying merge
revisions and picking whichever is smallest.

Some before and after stats on manifest.d size:

internal large repo:
before: 1.2 GB
after: 930 MB

mozilla-central:
before: 261 MB
after: 92 MB

mercurial/localrepo.py

@@ -1,1952 +1,1956
 # localrepo.py - read/write repository class for mercurial
 #
 # Copyright 2005-2007 Matt Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 from node import hex, nullid, wdirrev, short
 from i18n import _
 import urllib
 import peer, changegroup, subrepo, pushkey, obsolete, repoview
 import changelog, dirstate, filelog, manifest, context, bookmarks, phases
 import lock as lockmod
 import transaction, store, encoding, exchange, bundle2
 import scmutil, util, extensions, hook, error, revset
 import match as matchmod
 import merge as mergemod
 import tags as tagsmod
 from lock import release
 import weakref, errno, os, time, inspect, random
 import branchmap, pathutil
 import namespaces
 propertycache = util.propertycache
 filecache = scmutil.filecache
 
 class repofilecache(filecache):
     """All filecache usage on repo are done for logic that should be unfiltered
     """
 
     def __get__(self, repo, type=None):
         return super(repofilecache, self).__get__(repo.unfiltered(), type)
     def __set__(self, repo, value):
         return super(repofilecache, self).__set__(repo.unfiltered(), value)
     def __delete__(self, repo):
         return super(repofilecache, self).__delete__(repo.unfiltered())
 
 class storecache(repofilecache):
     """filecache for files in the store"""
     def join(self, obj, fname):
         return obj.sjoin(fname)
 
 class unfilteredpropertycache(propertycache):
     """propertycache that apply to unfiltered repo only"""
 
     def __get__(self, repo, type=None):
         unfi = repo.unfiltered()
         if unfi is repo:
             return super(unfilteredpropertycache, self).__get__(unfi)
         return getattr(unfi, self.name)
 
 class filteredpropertycache(propertycache):
     """propertycache that must take filtering in account"""
 
     def cachevalue(self, obj, value):
         object.__setattr__(obj, self.name, value)
 
 
 def hasunfilteredcache(repo, name):
     """check if a repo has an unfilteredpropertycache value for <name>"""
     return name in vars(repo.unfiltered())
 
 def unfilteredmethod(orig):
     """decorate method that always need to be run on unfiltered version"""
     def wrapper(repo, *args, **kwargs):
         return orig(repo.unfiltered(), *args, **kwargs)
     return wrapper
 
 moderncaps = set(('lookup', 'branchmap', 'pushkey', 'known', 'getbundle',
                   'unbundle'))
 legacycaps = moderncaps.union(set(['changegroupsubset']))
 
 class localpeer(peer.peerrepository):
     '''peer for a local repo; reflects only the most recent API'''
 
     def __init__(self, repo, caps=moderncaps):
         peer.peerrepository.__init__(self)
         self._repo = repo.filtered('served')
         self.ui = repo.ui
         self._caps = repo._restrictcapabilities(caps)
         self.requirements = repo.requirements
         self.supportedformats = repo.supportedformats
 
     def close(self):
         self._repo.close()
 
     def _capabilities(self):
         return self._caps
 
     def local(self):
         return self._repo
 
     def canpush(self):
         return True
 
     def url(self):
         return self._repo.url()
 
     def lookup(self, key):
         return self._repo.lookup(key)
 
     def branchmap(self):
         return self._repo.branchmap()
 
     def heads(self):
         return self._repo.heads()
 
     def known(self, nodes):
         return self._repo.known(nodes)
 
     def getbundle(self, source, heads=None, common=None, bundlecaps=None,
                   **kwargs):
         cg = exchange.getbundle(self._repo, source, heads=heads,
                                 common=common, bundlecaps=bundlecaps, **kwargs)
         if bundlecaps is not None and 'HG20' in bundlecaps:
             # When requesting a bundle2, getbundle returns a stream to make the
             # wire level function happier. We need to build a proper object
             # from it in local peer.
             cg = bundle2.getunbundler(self.ui, cg)
         return cg
 
     # TODO We might want to move the next two calls into legacypeer and add
     # unbundle instead.
 
     def unbundle(self, cg, heads, url):
         """apply a bundle on a repo
 
         This function handles the repo locking itself."""
         try:
             try:
                 cg = exchange.readbundle(self.ui, cg, None)
                 ret = exchange.unbundle(self._repo, cg, heads, 'push', url)
                 if util.safehasattr(ret, 'getchunks'):
                     # This is a bundle20 object, turn it into an unbundler.
                     # This little dance should be dropped eventually when the
                     # API is finally improved.
                     stream = util.chunkbuffer(ret.getchunks())
                     ret = bundle2.getunbundler(self.ui, stream)
                 return ret
             except Exception as exc:
                 # If the exception contains output salvaged from a bundle2
                 # reply, we need to make sure it is printed before continuing
                 # to fail. So we build a bundle2 with such output and consume
                 # it directly.
                 #
                 # This is not very elegant but allows a "simple" solution for
                 # issue4594
                 output = getattr(exc, '_bundle2salvagedoutput', ())
                 if output:
                     bundler = bundle2.bundle20(self._repo.ui)
                     for out in output:
                         bundler.addpart(out)
                     stream = util.chunkbuffer(bundler.getchunks())
                     b = bundle2.getunbundler(self.ui, stream)
                     bundle2.processbundle(self._repo, b)
                 raise
         except error.PushRaced as exc:
             raise error.ResponseError(_('push failed:'), str(exc))
 
     def lock(self):
         return self._repo.lock()
 
     def addchangegroup(self, cg, source, url):
         return changegroup.addchangegroup(self._repo, cg, source, url)
 
     def pushkey(self, namespace, key, old, new):
         return self._repo.pushkey(namespace, key, old, new)
 
     def listkeys(self, namespace):
         return self._repo.listkeys(namespace)
 
     def debugwireargs(self, one, two, three=None, four=None, five=None):
         '''used to test argument passing over the wire'''
         return "%s %s %s %s %s" % (one, two, three, four, five)
 
 class locallegacypeer(localpeer):
     '''peer extension which implements legacy methods too; used for tests with
     restricted capabilities'''
 
     def __init__(self, repo):
         localpeer.__init__(self, repo, caps=legacycaps)
 
     def branches(self, nodes):
         return self._repo.branches(nodes)
 
     def between(self, pairs):
         return self._repo.between(pairs)
 
     def changegroup(self, basenodes, source):
         return changegroup.changegroup(self._repo, basenodes, source)
 
     def changegroupsubset(self, bases, heads, source):
         return changegroup.changegroupsubset(self._repo, bases, heads, source)
 
 class localrepository(object):
 
     supportedformats = set(('revlogv1', 'generaldelta', 'treemanifest',
                             'manifestv2'))
     _basesupported = supportedformats | set(('store', 'fncache', 'shared',
                                              'dotencode'))
     openerreqs = set(('revlogv1', 'generaldelta', 'treemanifest', 'manifestv2'))
     filtername = None
 
     # a list of (ui, featureset) functions.
     # only functions defined in module of enabled extensions are invoked
     featuresetupfuncs = set()
 
     def _baserequirements(self, create):
         return ['revlogv1']
 
     def __init__(self, baseui, path=None, create=False):
         self.requirements = set()
         self.wvfs = scmutil.vfs(path, expandpath=True, realpath=True)
         self.wopener = self.wvfs
         self.root = self.wvfs.base
         self.path = self.wvfs.join(".hg")
         self.origroot = path
         self.auditor = pathutil.pathauditor(self.root, self._checknested)
         self.vfs = scmutil.vfs(self.path)
         self.opener = self.vfs
         self.baseui = baseui
         self.ui = baseui.copy()
         self.ui.copy = baseui.copy # prevent copying repo configuration
         # A list of callback to shape the phase if no data were found.
         # Callback are in the form: func(repo, roots) --> processed root.
         # This list it to be filled by extension during repo setup
         self._phasedefaults = []
         try:
             self.ui.readconfig(self.join("hgrc"), self.root)
             extensions.loadall(self.ui)
         except IOError:
             pass
 
         if self.featuresetupfuncs:
             self.supported = set(self._basesupported) # use private copy
             extmods = set(m.__name__ for n, m
                           in extensions.extensions(self.ui))
             for setupfunc in self.featuresetupfuncs:
                 if setupfunc.__module__ in extmods:
                     setupfunc(self.ui, self.supported)
         else:
             self.supported = self._basesupported
 
         if not self.vfs.isdir():
             if create:
                 if not self.wvfs.exists():
                     self.wvfs.makedirs()
                 self.vfs.makedir(notindexed=True)
                 self.requirements.update(self._baserequirements(create))
                 if self.ui.configbool('format', 'usestore', True):
                     self.vfs.mkdir("store")
                     self.requirements.add("store")
                     if self.ui.configbool('format', 'usefncache', True):
                         self.requirements.add("fncache")
                         if self.ui.configbool('format', 'dotencode', True):
                             self.requirements.add('dotencode')
                     # create an invalid changelog
                     self.vfs.append(
                         "00changelog.i",
                         '\0\0\0\2' # represents revlogv2
                         ' dummy changelog to prevent using the old repo layout'
                     )
                 # experimental config: format.generaldelta
                 if self.ui.configbool('format', 'generaldelta', False):
                     self.requirements.add("generaldelta")
                 if self.ui.configbool('experimental', 'treemanifest', False):
                     self.requirements.add("treemanifest")
                 if self.ui.configbool('experimental', 'manifestv2', False):
                     self.requirements.add("manifestv2")
             else:
                 raise error.RepoError(_("repository %s not found") % path)
         elif create:
             raise error.RepoError(_("repository %s already exists") % path)
         else:
             try:
                 self.requirements = scmutil.readrequires(
                         self.vfs, self.supported)
             except IOError as inst:
                 if inst.errno != errno.ENOENT:
                     raise
 
         self.sharedpath = self.path
         try:
             vfs = scmutil.vfs(self.vfs.read("sharedpath").rstrip('\n'),
                               realpath=True)
             s = vfs.base
             if not vfs.exists():
                 raise error.RepoError(
                     _('.hg/sharedpath points to nonexistent directory %s') % s)
             self.sharedpath = s
         except IOError as inst:
             if inst.errno != errno.ENOENT:
                 raise
 
         self.store = store.store(
                 self.requirements, self.sharedpath, scmutil.vfs)
         self.spath = self.store.path
         self.svfs = self.store.vfs
         self.sjoin = self.store.join
         self.vfs.createmode = self.store.createmode
         self._applyopenerreqs()
         if create:
             self._writerequirements()
 
 
         self._branchcaches = {}
         self._revbranchcache = None
         self.filterpats = {}
         self._datafilters = {}
         self._transref = self._lockref = self._wlockref = None
 
         # A cache for various files under .hg/ that tracks file changes,
         # (used by the filecache decorator)
         #
         # Maps a property name to its util.filecacheentry
         self._filecache = {}
 
         # hold sets of revision to be filtered
         # should be cleared when something might have changed the filter value:
         # - new changesets,
         # - phase change,
         # - new obsolescence marker,
         # - working directory parent change,
         # - bookmark changes
         self.filteredrevcache = {}
 
         # generic mapping between names and nodes
         self.names = namespaces.namespaces()
 
     def close(self):
         self._writecaches()
 
     def _writecaches(self):
         if self._revbranchcache:
             self._revbranchcache.write()
 
     def _restrictcapabilities(self, caps):
         if self.ui.configbool('experimental', 'bundle2-advertise', True):
             caps = set(caps)
             capsblob = bundle2.encodecaps(bundle2.getrepocaps(self))
             caps.add('bundle2=' + urllib.quote(capsblob))
         return caps
 
     def _applyopenerreqs(self):
         self.svfs.options = dict((r, 1) for r in self.requirements
                                            if r in self.openerreqs)
         # experimental config: format.chunkcachesize
         chunkcachesize = self.ui.configint('format', 'chunkcachesize')
         if chunkcachesize is not None:
             self.svfs.options['chunkcachesize'] = chunkcachesize
         # experimental config: format.maxchainlen
         maxchainlen = self.ui.configint('format', 'maxchainlen')
         if maxchainlen is not None:
             self.svfs.options['maxchainlen'] = maxchainlen
         # experimental config: format.manifestcachesize
         manifestcachesize = self.ui.configint('format', 'manifestcachesize')
         if manifestcachesize is not None:
             self.svfs.options['manifestcachesize'] = manifestcachesize
+        # experimental config: format.aggressivemergedeltas
+        aggressivemergedeltas = self.ui.configbool('format',
+            'aggressivemergedeltas', False)
+        self.svfs.options['aggressivemergedeltas'] = aggressivemergedeltas
 
     def _writerequirements(self):
         scmutil.writerequires(self.vfs, self.requirements)
 
     def _checknested(self, path):
         """Determine if path is a legal nested repository."""
         if not path.startswith(self.root):
             return False
         subpath = path[len(self.root) + 1:]
         normsubpath = util.pconvert(subpath)
 
         # XXX: Checking against the current working copy is wrong in
         # the sense that it can reject things like
         #
         #   $ hg cat -r 10 sub/x.txt
         #
         # if sub/ is no longer a subrepository in the working copy
         # parent revision.
         #
         # However, it can of course also allow things that would have
         # been rejected before, such as the above cat command if sub/
         # is a subrepository now, but was a normal directory before.
         # The old path auditor would have rejected by mistake since it
         # panics when it sees sub/.hg/.
         #
         # All in all, checking against the working copy seems sensible
         # since we want to prevent access to nested repositories on
         # the filesystem *now*.
         ctx = self[None]
         parts = util.splitpath(subpath)
         while parts:
             prefix = '/'.join(parts)
             if prefix in ctx.substate:
                 if prefix == normsubpath:
                     return True
                 else:
                     sub = ctx.sub(prefix)
                     return sub.checknested(subpath[len(prefix) + 1:])
             else:
                 parts.pop()
         return False
 
     def peer(self):
         return localpeer(self) # not cached to avoid reference cycle
 
     def unfiltered(self):
         """Return unfiltered version of the repository
 
         Intended to be overwritten by filtered repo."""
         return self
 
     def filtered(self, name):
         """Return a filtered version of a repository"""
         # build a new class with the mixin and the current class
         # (possibly subclass of the repo)
         class proxycls(repoview.repoview, self.unfiltered().__class__):
             pass
         return proxycls(self, name)
 
     @repofilecache('bookmarks')
     def _bookmarks(self):
         return bookmarks.bmstore(self)
 
     @repofilecache('bookmarks.current')
     def _activebookmark(self):
         return bookmarks.readactive(self)
 
     def bookmarkheads(self, bookmark):
         name = bookmark.split('@', 1)[0]
         heads = []
         for mark, n in self._bookmarks.iteritems():
             if mark.split('@', 1)[0] == name:
                 heads.append(n)
         return heads
 
     @storecache('phaseroots')
     def _phasecache(self):
         return phases.phasecache(self, self._phasedefaults)
 
     @storecache('obsstore')
     def obsstore(self):
         # read default format for new obsstore.
         # developer config: format.obsstore-version
         defaultformat = self.ui.configint('format', 'obsstore-version', None)
         # rely on obsstore class default when possible.
         kwargs = {}
         if defaultformat is not None:
             kwargs['defaultformat'] = defaultformat
         readonly = not obsolete.isenabled(self, obsolete.createmarkersopt)
         store = obsolete.obsstore(self.svfs, readonly=readonly,
                                   **kwargs)
         if store and readonly:
             self.ui.warn(
                 _('obsolete feature not enabled but %i markers found!\n')
                 % len(list(store)))
         return store
 
     @storecache('00changelog.i')
     def changelog(self):
         c = changelog.changelog(self.svfs)
         if 'HG_PENDING' in os.environ:
             p = os.environ['HG_PENDING']
             if p.startswith(self.root):
                 c.readpending('00changelog.i.a')
         return c
 
     @storecache('00manifest.i')
     def manifest(self):
         return manifest.manifest(self.svfs)
 
     def dirlog(self, dir):
         return self.manifest.dirlog(dir)
 
     @repofilecache('dirstate')
     def dirstate(self):
         warned = [0]
         def validate(node):
             try:
                 self.changelog.rev(node)
                 return node
             except error.LookupError:
                 if not warned[0]:
                     warned[0] = True
                     self.ui.warn(_("warning: ignoring unknown"
                                    " working parent %s!\n") % short(node))
                 return nullid
 
         return dirstate.dirstate(self.vfs, self.ui, self.root, validate)
 
     def __getitem__(self, changeid):
         if changeid is None or changeid == wdirrev:
             return context.workingctx(self)
         if isinstance(changeid, slice):
             return [context.changectx(self, i)
                     for i in xrange(*changeid.indices(len(self)))
                     if i not in self.changelog.filteredrevs]
         return context.changectx(self, changeid)
 
     def __contains__(self, changeid):
         try:
             self[changeid]
             return True
         except error.RepoLookupError:
             return False
 
     def __nonzero__(self):
         return True
 
     def __len__(self):
         return len(self.changelog)
 
     def __iter__(self):
         return iter(self.changelog)
 
     def revs(self, expr, *args):
         '''Return a list of revisions matching the given revset'''
         expr = revset.formatspec(expr, *args)
         m = revset.match(None, expr)
         return m(self)
 
     def set(self, expr, *args):
         '''
         Yield a context for each matching revision, after doing arg
         replacement via revset.formatspec
         '''
         for r in self.revs(expr, *args):
             yield self[r]
 
     def url(self):
         return 'file:' + self.root
 
     def hook(self, name, throw=False, **args):
         """Call a hook, passing this repo instance.
 
         This a convenience method to aid invoking hooks. Extensions likely
         won't call this unless they have registered a custom hook or are
         replacing code that is expected to call a hook.
         """
         return hook.hook(self.ui, self, name, throw, **args)
 
     @unfilteredmethod
     def _tag(self, names, node, message, local, user, date, extra={},
              editor=False):
         if isinstance(names, str):
             names = (names,)
 
         branches = self.branchmap()
         for name in names:
             self.hook('pretag', throw=True, node=hex(node), tag=name,
                       local=local)
             if name in branches:
                 self.ui.warn(_("warning: tag %s conflicts with existing"
                 " branch name\n") % name)
 
         def writetags(fp, names, munge, prevtags):
             fp.seek(0, 2)
             if prevtags and prevtags[-1] != '\n':
                 fp.write('\n')
             for name in names:
                 if munge:
                     m = munge(name)
                 else:
                     m = name
 
                 if (self._tagscache.tagtypes and
                     name in self._tagscache.tagtypes):
                     old = self.tags().get(name, nullid)
                     fp.write('%s %s\n' % (hex(old), m))
                 fp.write('%s %s\n' % (hex(node), m))
             fp.close()
 
         prevtags = ''
         if local:
             try:
                 fp = self.vfs('localtags', 'r+')
             except IOError:
                 fp = self.vfs('localtags', 'a')
             else:
                 prevtags = fp.read()
 
             # local tags are stored in the current charset
             writetags(fp, names, None, prevtags)
             for name in names:
                 self.hook('tag', node=hex(node), tag=name, local=local)
             return
 
         try:
             fp = self.wfile('.hgtags', 'rb+')
         except IOError as e:
             if e.errno != errno.ENOENT:
                 raise
             fp = self.wfile('.hgtags', 'ab')
         else:
             prevtags = fp.read()
 
         # committed tags are stored in UTF-8
         writetags(fp, names, encoding.fromlocal, prevtags)
 
         fp.close()
 
         self.invalidatecaches()
 
         if '.hgtags' not in self.dirstate:
             self[None].add(['.hgtags'])
 
         m = matchmod.exact(self.root, '', ['.hgtags'])
         tagnode = self.commit(message, user, date, extra=extra, match=m,
                               editor=editor)
 
         for name in names:
             self.hook('tag', node=hex(node), tag=name, local=local)
 
         return tagnode
 
     def tag(self, names, node, message, local, user, date, editor=False):
         '''tag a revision with one or more symbolic names.
 
         names is a list of strings or, when adding a single tag, names may be a
         string.
 
         if local is True, the tags are stored in a per-repository file.
         otherwise, they are stored in the .hgtags file, and a new
         changeset is committed with the change.
 
         keyword arguments:
 
         local: whether to store tags in non-version-controlled file
         (default False)
 
         message: commit message to use if committing
 
         user: name of user to use if committing
 
         date: date tuple to use if committing'''
 
         if not local:
             m = matchmod.exact(self.root, '', ['.hgtags'])
             if any(self.status(match=m, unknown=True, ignored=True)):
                 raise util.Abort(_('working copy of .hgtags is changed'),
                                  hint=_('please commit .hgtags manually'))
 
         self.tags() # instantiate the cache
         self._tag(names, node, message, local, user, date, editor=editor)
 
     @filteredpropertycache
     def _tagscache(self):
         '''Returns a tagscache object that contains various tags related
         caches.'''
 
         # This simplifies its cache management by having one decorated
         # function (this one) and the rest simply fetch things from it.
         class tagscache(object):
             def __init__(self):
                 # These two define the set of tags for this repository. tags
                 # maps tag name to node; tagtypes maps tag name to 'global' or
                 # 'local'. (Global tags are defined by .hgtags across all
                 # heads, and local tags are defined in .hg/localtags.)
                 # They constitute the in-memory cache of tags.
                 self.tags = self.tagtypes = None
 
                 self.nodetagscache = self.tagslist = None
 
         cache = tagscache()
         cache.tags, cache.tagtypes = self._findtags()
 
         return cache
 
     def tags(self):
         '''return a mapping of tag to node'''
         t = {}
         if self.changelog.filteredrevs:
             tags, tt = self._findtags()
         else:
             tags = self._tagscache.tags
         for k, v in tags.iteritems():
             try:
                 # ignore tags to unknown nodes
                 self.changelog.rev(v)
                 t[k] = v
             except (error.LookupError, ValueError):
                 pass
         return t
 
     def _findtags(self):
         '''Do the hard work of finding tags.  Return a pair of dicts
         (tags, tagtypes) where tags maps tag name to node, and tagtypes
         maps tag name to a string like \'global\' or \'local\'.
         Subclasses or extensions are free to add their own tags, but
         should be aware that the returned dicts will be retained for the
         duration of the localrepo object.'''
 
         # XXX what tagtype should subclasses/extensions use?  Currently
         # mq and bookmarks add tags, but do not set the tagtype at all.
         # Should each extension invent its own tag type?  Should there
         # be one tagtype for all such "virtual" tags?  Or is the status
         # quo fine?
 
         alltags = {}                    # map tag name to (node, hist)
         tagtypes = {}
 
         tagsmod.findglobaltags(self.ui, self, alltags, tagtypes)
         tagsmod.readlocaltags(self.ui, self, alltags, tagtypes)
 
         # Build the return dicts.  Have to re-encode tag names because
         # the tags module always uses UTF-8 (in order not to lose info
         # writing to the cache), but the rest of Mercurial wants them in
         # local encoding.
         tags = {}
         for (name, (node, hist)) in alltags.iteritems():
             if node != nullid:
                 tags[encoding.tolocal(name)] = node
         tags['tip'] = self.changelog.tip()
         tagtypes = dict([(encoding.tolocal(name), value)
                          for (name, value) in tagtypes.iteritems()])
         return (tags, tagtypes)
 
     def tagtype(self, tagname):
         '''
         return the type of the given tag. result can be:
 
         'local'  : a local tag
         'global' : a global tag
         None     : tag does not exist
         '''
 
         return self._tagscache.tagtypes.get(tagname)
 
     def tagslist(self):
         '''return a list of tags ordered by revision'''
         if not self._tagscache.tagslist:
             l = []
             for t, n in self.tags().iteritems():
                 l.append((self.changelog.rev(n), t, n))
             self._tagscache.tagslist = [(t, n) for r, t, n in sorted(l)]
 
         return self._tagscache.tagslist
 
     def nodetags(self, node):
         '''return the tags associated with a node'''
         if not self._tagscache.nodetagscache:
             nodetagscache = {}
             for t, n in self._tagscache.tags.iteritems():
                 nodetagscache.setdefault(n, []).append(t)
             for tags in nodetagscache.itervalues():
                 tags.sort()
             self._tagscache.nodetagscache = nodetagscache
         return self._tagscache.nodetagscache.get(node, [])
 
     def nodebookmarks(self, node):
         marks = []
         for bookmark, n in self._bookmarks.iteritems():
             if n == node:
                 marks.append(bookmark)
         return sorted(marks)
 
     def branchmap(self):
         '''returns a dictionary {branch: [branchheads]} with branchheads
         ordered by increasing revision number'''
         branchmap.updatecache(self)
         return self._branchcaches[self.filtername]
 
     @unfilteredmethod
     def revbranchcache(self):
         if not self._revbranchcache:
             self._revbranchcache = branchmap.revbranchcache(self.unfiltered())
         return self._revbranchcache
 
     def branchtip(self, branch, ignoremissing=False):
         '''return the tip node for a given branch
 
         If ignoremissing is True, then this method will not raise an error.
         This is helpful for callers that only expect None for a missing branch
         (e.g. namespace).
 
         '''
         try:
             return self.branchmap().branchtip(branch)
         except KeyError:
             if not ignoremissing:
                 raise error.RepoLookupError(_("unknown branch '%s'") % branch)
             else:
                 pass
 
     def lookup(self, key):
         return self[key].node()
 
     def lookupbranch(self, key, remote=None):
         repo = remote or self
         if key in repo.branchmap():
             return key
 
         repo = (remote and remote.local()) and remote or self
         return repo[key].branch()
 
     def known(self, nodes):
         nm = self.changelog.nodemap
         pc = self._phasecache
         result = []
         for n in nodes:
             r = nm.get(n)
             resp = not (r is None or pc.phase(self, r) >= phases.secret)
             result.append(resp)
         return result
 
     def local(self):
         return self
 
     def publishing(self):
         # it's safe (and desirable) to trust the publish flag unconditionally
         # so that we don't finalize changes shared between users via ssh or nfs
         return self.ui.configbool('phases', 'publish', True, untrusted=True)
 
     def cancopy(self):
         # so statichttprepo's override of local() works
         if not self.local():
             return False
         if not self.publishing():
             return True
         # if publishing we can't copy if there is filtered content
         return not self.filtered('visible').changelog.filteredrevs
 
     def shared(self):
         '''the type of shared repository (None if not shared)'''
         if self.sharedpath != self.path:
             return 'store'
         return None
 
     def join(self, f, *insidef):
         return self.vfs.join(os.path.join(f, *insidef))
 
     def wjoin(self, f, *insidef):
         return self.vfs.reljoin(self.root, f, *insidef)
 
     def file(self, f):
         if f[0] == '/':
             f = f[1:]
         return filelog.filelog(self.svfs, f)
 
     def changectx(self, changeid):
         return self[changeid]
 
     def parents(self, changeid=None):
         '''get list of changectxs for parents of changeid'''
         return self[changeid].parents()
 
     def setparents(self, p1, p2=nullid):
         self.dirstate.beginparentchange()
         copies = self.dirstate.setparents(p1, p2)
         pctx = self[p1]
         if copies:
             # Adjust copy records, the dirstate cannot do it, it
             # requires access to parents manifests. Preserve them
             # only for entries added to first parent.
             for f in copies:
                 if f not in pctx and copies[f] in pctx:
                     self.dirstate.copy(copies[f], f)
         if p2 == nullid:
             for f, s in sorted(self.dirstate.copies().items()):
                 if f not in pctx and s not in pctx:
                     self.dirstate.copy(None, f)
         self.dirstate.endparentchange()
 
     def filectx(self, path, changeid=None, fileid=None):
         """changeid can be a changeset revision, node, or tag.
            fileid can be a file revision or node."""
         return context.filectx(self, path, changeid, fileid)
 
     def getcwd(self):
         return self.dirstate.getcwd()
 
     def pathto(self, f, cwd=None):
         return self.dirstate.pathto(f, cwd)
 
     def wfile(self, f, mode='r'):
         return self.wvfs(f, mode)
 
     def _link(self, f):
         return self.wvfs.islink(f)
 
     def _loadfilter(self, filter):
         if filter not in self.filterpats:
             l = []
             for pat, cmd in self.ui.configitems(filter):
                 if cmd == '!':
                     continue
                 mf = matchmod.match(self.root, '', [pat])
                 fn = None
                 params = cmd
                 for name, filterfn in self._datafilters.iteritems():
                     if cmd.startswith(name):
                         fn = filterfn
                         params = cmd[len(name):].lstrip()
                         break
                 if not fn:
                     fn = lambda s, c, **kwargs: util.filter(s, c)
                 # Wrap old filters not supporting keyword arguments
                 if not inspect.getargspec(fn)[2]:
                     oldfn = fn
                     fn = lambda s, c, **kwargs: oldfn(s, c)
                 l.append((mf, fn, params))
             self.filterpats[filter] = l
         return self.filterpats[filter]
 
     def _filter(self, filterpats, filename, data):
         for mf, fn, cmd in filterpats:
             if mf(filename):
                 self.ui.debug("filtering %s through %s\n" % (filename, cmd))
                 data = fn(data, cmd, ui=self.ui, repo=self, filename=filename)
                 break
 
         return data
 
     @unfilteredpropertycache
     def _encodefilterpats(self):
         return self._loadfilter('encode')
 
     @unfilteredpropertycache
     def _decodefilterpats(self):
         return self._loadfilter('decode')
 
     def adddatafilter(self, name, filter):
         self._datafilters[name] = filter
 
     def wread(self, filename):
         if self._link(filename):
             data = self.wvfs.readlink(filename)
         else:
             data = self.wvfs.read(filename)
         return self._filter(self._encodefilterpats, filename, data)
 
     def wwrite(self, filename, data, flags):
         """write ``data`` into ``filename`` in the working directory
 
         This returns length of written (maybe decoded) data.
         """
         data = self._filter(self._decodefilterpats, filename, data)
         if 'l' in flags:
             self.wvfs.symlink(data, filename)
         else:
             self.wvfs.write(filename, data)
             if 'x' in flags:
                 self.wvfs.setflags(filename, False, True)
         return len(data)
 
     def wwritedata(self, filename, data):
         return self._filter(self._decodefilterpats, filename, data)
 
     def currenttransaction(self):
         """return the current transaction or None if non exists"""
         if self._transref:
             tr = self._transref()
         else:
             tr = None
 
         if tr and tr.running():
             return tr
         return None
 
     def transaction(self, desc, report=None):
         if (self.ui.configbool('devel', 'all-warnings')
                 or self.ui.configbool('devel', 'check-locks')):
             l = self._lockref and self._lockref()
             if l is None or not l.held:
                 self.ui.develwarn('transaction with no lock')
         tr = self.currenttransaction()
         if tr is not None:
             return tr.nest()
 
         # abort here if the journal already exists
         if self.svfs.exists("journal"):
             raise error.RepoError(
                 _("abandoned transaction found"),
                 hint=_("run 'hg recover' to clean up transaction"))
 
         # make journal.dirstate contain in-memory changes at this point
         self.dirstate.write()
 
         idbase = "%.40f#%f" % (random.random(), time.time())
         txnid = 'TXN:' + util.sha1(idbase).hexdigest()
         self.hook('pretxnopen', throw=True, txnname=desc, txnid=txnid)
 
         self._writejournal(desc)
         renames = [(vfs, x, undoname(x)) for vfs, x in self._journalfiles()]
         if report:
             rp = report
         else:
             rp = self.ui.warn
         vfsmap = {'plain': self.vfs} # root of .hg/
         # we must avoid cyclic reference between repo and transaction.
         reporef = weakref.ref(self)
         def validate(tr):
             """will run pre-closing hooks"""
             pending = lambda: tr.writepending() and self.root or ""
             reporef().hook('pretxnclose', throw=True, pending=pending,
                            txnname=desc, **tr.hookargs)
 
         tr = transaction.transaction(rp, self.svfs, vfsmap,
                                      "journal",
                                      "undo",
                                      aftertrans(renames),
                                      self.store.createmode,
                                      validator=validate)
 
         tr.hookargs['txnid'] = txnid
         # note: writing the fncache only during finalize mean that the file is
         # outdated when running hooks. As fncache is used for streaming clone,
         # this is not expected to break anything that happen during the hooks.
         tr.addfinalize('flush-fncache', self.store.write)
         def txnclosehook(tr2):
             """To be run if transaction is successful, will schedule a hook run
             """
             def hook():
                 reporef().hook('txnclose', throw=False, txnname=desc,
                                **tr2.hookargs)
             reporef()._afterlock(hook)
         tr.addfinalize('txnclose-hook', txnclosehook)
         def txnaborthook(tr2):
             """To be run if transaction is aborted
             """
             reporef().hook('txnabort', throw=False, txnname=desc,
                            **tr2.hookargs)
         tr.addabort('txnabort-hook', txnaborthook)
         self._transref = weakref.ref(tr)
         return tr
 
     def _journalfiles(self):
         return ((self.svfs, 'journal'),
                 (self.vfs, 'journal.dirstate'),
                 (self.vfs, 'journal.branch'),
                 (self.vfs, 'journal.desc'),
                 (self.vfs, 'journal.bookmarks'),
                 (self.svfs, 'journal.phaseroots'))
 
     def undofiles(self):
         return [(vfs, undoname(x)) for vfs, x in self._journalfiles()]
 
     def _writejournal(self, desc):
         self.vfs.write("journal.dirstate",
                           self.vfs.tryread("dirstate"))
         self.vfs.write("journal.branch",
                           encoding.fromlocal(self.dirstate.branch()))
         self.vfs.write("journal.desc",
                           "%d\n%s\n" % (len(self), desc))
         self.vfs.write("journal.bookmarks",
                           self.vfs.tryread("bookmarks"))
         self.svfs.write("journal.phaseroots",
                            self.svfs.tryread("phaseroots"))
 
     def recover(self):
         lock = self.lock()
         try:
             if self.svfs.exists("journal"):
                 self.ui.status(_("rolling back interrupted transaction\n"))
                 vfsmap = {'': self.svfs,
                           'plain': self.vfs,}
                 transaction.rollback(self.svfs, vfsmap, "journal",
                                      self.ui.warn)
                 self.invalidate()
                 return True
             else:
                 self.ui.warn(_("no interrupted transaction available\n"))
                 return False
         finally:
             lock.release()
 
     def rollback(self, dryrun=False, force=False):
         wlock = lock = None
         try:
             wlock = self.wlock()
             lock = self.lock()
             if self.svfs.exists("undo"):
                 return self._rollback(dryrun, force)
             else:
                 self.ui.warn(_("no rollback information available\n"))
                 return 1
         finally:
             release(lock, wlock)
 
     @unfilteredmethod # Until we get smarter cache management
     def _rollback(self, dryrun, force):
         ui = self.ui
         try:
             args = self.vfs.read('undo.desc').splitlines()
             (oldlen, desc, detail) = (int(args[0]), args[1], None)
             if len(args) >= 3:
                 detail = args[2]
             oldtip = oldlen - 1
 
             if detail and ui.verbose:
                 msg = (_('repository tip rolled back to revision %s'
                          ' (undo %s: %s)\n')
                        % (oldtip, desc, detail))
             else:
                 msg = (_('repository tip rolled back to revision %s'
                          ' (undo %s)\n')
                        % (oldtip, desc))
         except IOError:
             msg = _('rolling back unknown transaction\n')
             desc = None
 
         if not force and self['.'] != self['tip'] and desc == 'commit':
             raise util.Abort(
                 _('rollback of last commit while not checked out '
                   'may lose data'), hint=_('use -f to force'))
 
         ui.status(msg)
         if dryrun:
             return 0
 
         parents = self.dirstate.parents()
         self.destroying()
         vfsmap = {'plain': self.vfs, '': self.svfs}
         transaction.rollback(self.svfs, vfsmap, 'undo', ui.warn)
         if self.vfs.exists('undo.bookmarks'):
             self.vfs.rename('undo.bookmarks', 'bookmarks')
         if self.svfs.exists('undo.phaseroots'):
             self.svfs.rename('undo.phaseroots', 'phaseroots')
         self.invalidate()
 
         parentgone = (parents[0] not in self.changelog.nodemap or
                       parents[1] not in self.changelog.nodemap)
         if parentgone:
             self.vfs.rename('undo.dirstate', 'dirstate')
             try:
                 branch = self.vfs.read('undo.branch')
                 self.dirstate.setbranch(encoding.tolocal(branch))
             except IOError:
                 ui.warn(_('named branch could not be reset: '
                           'current branch is still \'%s\'\n')
                         % self.dirstate.branch())
 
             self.dirstate.invalidate()
             parents = tuple([p.rev() for p in self.parents()])
             if len(parents) > 1:
                 ui.status(_('working directory now based on '
                             'revisions %d and %d\n') % parents)
             else:
                 ui.status(_('working directory now based on '
                             'revision %d\n') % parents)
             ms = mergemod.mergestate(self)
             ms.reset(self['.'].node())
 
         # TODO: if we know which new heads may result from this rollback, pass
         # them to destroy(), which will prevent the branchhead cache from being
         # invalidated.
         self.destroyed()
         return 0
 
     def invalidatecaches(self):
 
         if '_tagscache' in vars(self):
             # can't use delattr on proxy
             del self.__dict__['_tagscache']
 
         self.unfiltered()._branchcaches.clear()
         self.invalidatevolatilesets()
 
     def invalidatevolatilesets(self):
         self.filteredrevcache.clear()
         obsolete.clearobscaches(self)
 
     def invalidatedirstate(self):
         '''Invalidates the dirstate, causing the next call to dirstate
         to check if it was modified since the last time it was read,
         rereading it if it has.
 
         This is different to dirstate.invalidate() that it doesn't always
         rereads the dirstate. Use dirstate.invalidate() if you want to
         explicitly read the dirstate again (i.e. restoring it to a previous
         known good state).'''
         if hasunfilteredcache(self, 'dirstate'):
             for k in self.dirstate._filecache:
                 try:
                     delattr(self.dirstate, k)
                 except AttributeError:
                     pass
             delattr(self.unfiltered(), 'dirstate')
 
     def invalidate(self):
         unfiltered = self.unfiltered() # all file caches are stored unfiltered
         for k in self._filecache:
             # dirstate is invalidated separately in invalidatedirstate()
             if k == 'dirstate':
                 continue
 
             try:
                 delattr(unfiltered, k)
             except AttributeError:
                 pass
         self.invalidatecaches()
         self.store.invalidatecaches()
 
     def invalidateall(self):
         '''Fully invalidates both store and non-store parts, causing the
         subsequent operation to reread any outside changes.'''
         # extension should hook this to invalidate its caches
         self.invalidate()
         self.invalidatedirstate()
 
     def _lock(self, vfs, lockname, wait, releasefn, acquirefn, desc):
         try:
             l = lockmod.lock(vfs, lockname, 0, releasefn, desc=desc)
         except error.LockHeld as inst:
             if not wait:
                 raise
             self.ui.warn(_("waiting for lock on %s held by %r\n") %
                          (desc, inst.locker))
             # default to 600 seconds timeout
             l = lockmod.lock(vfs, lockname,
                              int(self.ui.config("ui", "timeout", "600")),
                              releasefn, desc=desc)
             self.ui.warn(_("got lock after %s seconds\n") % l.delay)
         if acquirefn:
             acquirefn()
         return l
 
     def _afterlock(self, callback):
         """add a callback to be run when the repository is fully unlocked
 
         The callback will be executed when the outermost lock is released
         (with wlock being higher level than 'lock')."""
         for ref in (self._wlockref, self._lockref):
             l = ref and ref()
             if l and l.held:
                 l.postrelease.append(callback)
                 break
         else: # no lock have been found.
             callback()
 
     def lock(self, wait=True):
         '''Lock the repository store (.hg/store) and return a weak reference
         to the lock. Use this before modifying the store (e.g. committing or
         stripping). If you are opening a transaction, get a lock as well.)
 
         If both 'lock' and 'wlock' must be acquired, ensure you always acquires
         'wlock' first to avoid a dead-lock hazard.'''
         l = self._lockref and self._lockref()
         if l is not None and l.held:
             l.lock()
             return l
 
         def unlock():
             for k, ce in self._filecache.items():
                 if k == 'dirstate' or k not in self.__dict__:
                     continue
                 ce.refresh()
 
         l = self._lock(self.svfs, "lock", wait, unlock,
                        self.invalidate, _('repository %s') % self.origroot)
         self._lockref = weakref.ref(l)
         return l
 
     def wlock(self, wait=True):
         '''Lock the non-store parts of the repository (everything under
         .hg except .hg/store) and return a weak reference to the lock.
 
         Use this before modifying files in .hg.
 
         If both 'lock' and 'wlock' must be acquired, ensure you always acquires
         'wlock' first to avoid a dead-lock hazard.'''
         l = self._wlockref and self._wlockref()
         if l is not None and l.held:
             l.lock()
             return l
 
         # We do not need to check for non-waiting lock aquisition.  Such
         # acquisition would not cause dead-lock as they would just fail.
         if wait and (self.ui.configbool('devel', 'all-warnings')
                      or self.ui.configbool('devel', 'check-locks')):
             l = self._lockref and self._lockref()
             if l is not None and l.held:
                 self.ui.develwarn('"wlock" acquired after "lock"')
 
         def unlock():
             if self.dirstate.pendingparentchange():
                 self.dirstate.invalidate()
             else:
                 self.dirstate.write()
 
             self._filecache['dirstate'].refresh()
 
         l = self._lock(self.vfs, "wlock", wait, unlock,
                        self.invalidatedirstate, _('working directory of %s') %
                        self.origroot)
         self._wlockref = weakref.ref(l)
         return l
 
     def _filecommit(self, fctx, manifest1, manifest2, linkrev, tr, changelist):
         """
         commit an individual file as part of a larger transaction
         """
 
         fname = fctx.path()
         fparent1 = manifest1.get(fname, nullid)
         fparent2 = manifest2.get(fname, nullid)
         if isinstance(fctx, context.filectx):
             node = fctx.filenode()
             if node in [fparent1, fparent2]:
                 self.ui.debug('reusing %s filelog entry\n' % fname)
                 return node
 
         flog = self.file(fname)
         meta = {}
         copy = fctx.renamed()
         if copy and copy[0] != fname:
             # Mark the new revision of this file as a copy of another
             # file.  This copy data will effectively act as a parent
             # of this new revision.  If this is a merge, the first
             # parent will be the nullid (meaning "look up the copy data")
             # and the second one will be the other parent.  For example:
             #
             # 0 --- 1 --- 3   rev1 changes file foo
             #   \       /     rev2 renames foo to bar and changes it
             #    \- 2 -/      rev3 should have bar with all changes and
             #                      should record that bar descends from
             #                      bar in rev2 and foo in rev1
             #
             # this allows this merge to succeed:
             #
             # 0 --- 1 --- 3   rev4 reverts the content change from rev2
             #   \       /     merging rev3 and rev4 should use bar@rev2
             #    \- 2 --- 4        as the merge base
             #
 
             cfname = copy[0]
             crev = manifest1.get(cfname)
             newfparent = fparent2
 
             if manifest2: # branch merge
                 if fparent2 == nullid or crev is None: # copied on remote side
                     if cfname in manifest2:
                         crev = manifest2[cfname]
                         newfparent = fparent1
 
             # Here, we used to search backwards through history to try to find
             # where the file copy came from if the source of a copy was not in
             # the parent directory. However, this doesn't actually make sense to
             # do (what does a copy from something not in your working copy even
             # mean?) and it causes bugs (eg, issue4476). Instead, we will warn
             # the user that copy information was dropped, so if they didn't
             # expect this outcome it can be fixed, but this is the correct
             # behavior in this circumstance.
 
             if crev:
                 self.ui.debug(" %s: copy %s:%s\n" % (fname, cfname, hex(crev)))
                 meta["copy"] = cfname
                 meta["copyrev"] = hex(crev)
                 fparent1, fparent2 = nullid, newfparent
             else:
                 self.ui.warn(_("warning: can't find ancestor for '%s' "
                                "copied from '%s'!\n") % (fname, cfname))
 
         elif fparent1 == nullid:
             fparent1, fparent2 = fparent2, nullid
         elif fparent2 != nullid:
             # is one parent an ancestor of the other?
             fparentancestors = flog.commonancestorsheads(fparent1, fparent2)
             if fparent1 in fparentancestors:
                 fparent1, fparent2 = fparent2, nullid
             elif fparent2 in fparentancestors:
                 fparent2 = nullid
 
         # is the file changed?
         text = fctx.data()
         if fparent2 != nullid or flog.cmp(fparent1, text) or meta:
             changelist.append(fname)
             return flog.add(text, meta, tr, linkrev, fparent1, fparent2)
         # are just the flags changed during merge?
         elif fname in manifest1 and manifest1.flags(fname) != fctx.flags():
             changelist.append(fname)
 
         return fparent1
 
     @unfilteredmethod
     def commit(self, text="", user=None, date=None, match=None, force=False,
                editor=False, extra={}):
         """Add a new revision to current repository.
 
         Revision information is gathered from the working directory,
         match can be used to filter the committed files. If editor is
         supplied, it is called to get a commit message.
         """
 
         def fail(f, msg):
             raise util.Abort('%s: %s' % (f, msg))
 
         if not match:
             match = matchmod.always(self.root, '')
 
         if not force:
             vdirs = []
             match.explicitdir = vdirs.append
             match.bad = fail
 
         wlock = self.wlock()
         try:
             wctx = self[None]
             merge = len(wctx.parents()) > 1
 
             if not force and merge and match.ispartial():
                 raise util.Abort(_('cannot partially commit a merge '
                                    '(do not specify files or patterns)'))
 
             status = self.status(match=match, clean=force)
             if force:
                 status.modified.extend(status.clean) # mq may commit clean files
 
             # check subrepos
             subs = []
             commitsubs = set()
             newstate = wctx.substate.copy()
             # only manage subrepos and .hgsubstate if .hgsub is present
             if '.hgsub' in wctx:
                 # we'll decide whether to track this ourselves, thanks
                 for c in status.modified, status.added, status.removed:
                     if '.hgsubstate' in c:
                         c.remove('.hgsubstate')
 
                 # compare current state to last committed state
                 # build new substate based on last committed state
                 oldstate = wctx.p1().substate
                 for s in sorted(newstate.keys()):
                     if not match(s):
                         # ignore working copy, use old state if present
                         if s in oldstate:
                             newstate[s] = oldstate[s]
                             continue
                         if not force:
                             raise util.Abort(
                                 _("commit with new subrepo %s excluded") % s)
                     dirtyreason = wctx.sub(s).dirtyreason(True)
                     if dirtyreason:
                         if not self.ui.configbool('ui', 'commitsubrepos'):
                             raise util.Abort(dirtyreason,
                                 hint=_("use --subrepos for recursive commit"))
                         subs.append(s)
                         commitsubs.add(s)
                     else:
                         bs = wctx.sub(s).basestate()
                         newstate[s] = (newstate[s][0], bs, newstate[s][2])
                         if oldstate.get(s, (None, None, None))[1] != bs:
                             subs.append(s)
 
                 # check for removed subrepos
                 for p in wctx.parents():
                     r = [s for s in p.substate if s not in newstate]
                     subs += [s for s in r if match(s)]
                 if subs:
                     if (not match('.hgsub') and
                         '.hgsub' in (wctx.modified() + wctx.added())):
                         raise util.Abort(
                             _("can't commit subrepos without .hgsub"))
                     status.modified.insert(0, '.hgsubstate')
 
             elif '.hgsub' in status.removed:
                 # clean up .hgsubstate when .hgsub is removed
                 if ('.hgsubstate' in wctx and
                     '.hgsubstate' not in (status.modified + status.added +
                                           status.removed)):
                     status.removed.insert(0, '.hgsubstate')
 
             # make sure all explicit patterns are matched
             if not force and (match.isexact() or match.prefix()):
                 matched = set(status.modified + status.added + status.removed)
 
                 for f in match.files():
                     f = self.dirstate.normalize(f)
                     if f == '.' or f in matched or f in wctx.substate:
                         continue
                     if f in status.deleted:
                         fail(f, _('file not found!'))
                     if f in vdirs: # visited directory
                         d = f + '/'
                         for mf in matched:
                             if mf.startswith(d):
                                 break
                         else:
                             fail(f, _("no match under directory!"))
                     elif f not in self.dirstate:
                         fail(f, _("file not tracked!"))
 
             cctx = context.workingcommitctx(self, status,
                                             text, user, date, extra)
 
             # internal config: ui.allowemptycommit
             allowemptycommit = (wctx.branch() != wctx.p1().branch()
                                 or extra.get('close') or merge or cctx.files()
                                 or self.ui.configbool('ui', 'allowemptycommit'))
             if not allowemptycommit:
                 return None
 
             if merge and cctx.deleted():
                 raise util.Abort(_("cannot commit merge with missing files"))
 
             ms = mergemod.mergestate(self)
             for f in status.modified:
                 if f in ms and ms[f] == 'u':
                     raise util.Abort(_('unresolved merge conflicts '
                                        '(see "hg help resolve")'))
 
             if editor:
                 cctx._text = editor(self, cctx, subs)
             edited = (text != cctx._text)
 
             # Save commit message in case this transaction gets rolled back
             # (e.g. by a pretxncommit hook).  Leave the content alone on
             # the assumption that the user will use the same editor again.
             msgfn = self.savecommitmessage(cctx._text)
 
             # commit subs and write new state
             if subs:
                 for s in sorted(commitsubs):
                     sub = wctx.sub(s)
                     self.ui.status(_('committing subrepository %s\n') %
                         subrepo.subrelpath(sub))
                     sr = sub.commit(cctx._text, user, date)
                     newstate[s] = (newstate[s][0], sr)
                 subrepo.writestate(self, newstate)
 
             p1, p2 = self.dirstate.parents()
             hookp1, hookp2 = hex(p1), (p2 != nullid and hex(p2) or '')
             try:
                 self.hook("precommit", throw=True, parent1=hookp1,
                           parent2=hookp2)
                 ret = self.commitctx(cctx, True)
             except: # re-raises
                 if edited:
                     self.ui.write(
                         _('note: commit message saved in %s\n') % msgfn)
                 raise
 
             # update bookmarks, dirstate and mergestate
             bookmarks.update(self, [p1, p2], ret)
             cctx.markcommitted(ret)
             ms.reset()
         finally:
             wlock.release()
 
         def commithook(node=hex(ret), parent1=hookp1, parent2=hookp2):
             # hack for command that use a temporary commit (eg: histedit)
             # temporary commit got stripped before hook release
             if self.changelog.hasnode(ret):
                 self.hook("commit", node=node, parent1=parent1,
                           parent2=parent2)
         self._afterlock(commithook)
         return ret
 
     @unfilteredmethod
     def commitctx(self, ctx, error=False):
         """Add a new revision to current repository.
         Revision information is passed via the context argument.
         """
 
         tr = None
         p1, p2 = ctx.p1(), ctx.p2()
         user = ctx.user()
 
         lock = self.lock()
         try:
             tr = self.transaction("commit")
             trp = weakref.proxy(tr)
 
             if ctx.files():
                 m1 = p1.manifest()
                 m2 = p2.manifest()
                 m = m1.copy()
 
                 # check in files
                 added = []
                 changed = []
                 removed = list(ctx.removed())
                 linkrev = len(self)
                 self.ui.note(_("committing files:\n"))
                 for f in sorted(ctx.modified() + ctx.added()):
                     self.ui.note(f + "\n")
                     try:
                         fctx = ctx[f]
                         if fctx is None:
                             removed.append(f)
                         else:
                             added.append(f)
                             m[f] = self._filecommit(fctx, m1, m2, linkrev,
                                                     trp, changed)
                             m.setflag(f, fctx.flags())
                     except OSError as inst:
                         self.ui.warn(_("trouble committing %s!\n") % f)
                         raise
                     except IOError as inst:
                         errcode = getattr(inst, 'errno', errno.ENOENT)
                         if error or errcode and errcode != errno.ENOENT:
                             self.ui.warn(_("trouble committing %s!\n") % f)
                         raise
 
                 # update manifest
                 self.ui.note(_("committing manifest\n"))
                 removed = [f for f in sorted(removed) if f in m1 or f in m2]
                 drop = [f for f in removed if f in m]
                 for f in drop:
                     del m[f]
                 mn = self.manifest.add(m, trp, linkrev,
                                        p1.manifestnode(), p2.manifestnode(),
                                        added, drop)
                 files = changed + removed
             else:
                 mn = p1.manifestnode()
                 files = []
 
             # update changelog
             self.ui.note(_("committing changelog\n"))
             self.changelog.delayupdate(tr)
             n = self.changelog.add(mn, files, ctx.description(),
                                    trp, p1.node(), p2.node(),
                                    user, ctx.date(), ctx.extra().copy())
             p = lambda: tr.writepending() and self.root or ""
             xp1, xp2 = p1.hex(), p2 and p2.hex() or ''
             self.hook('pretxncommit', throw=True, node=hex(n), parent1=xp1,
                       parent2=xp2, pending=p)
             # set the new commit is proper phase
             targetphase = subrepo.newcommitphase(self.ui, ctx)
             if targetphase:
                 # retract boundary do not alter parent changeset.
                 # if a parent have higher the resulting phase will
                 # be compliant anyway
                 #
                 # if minimal phase was 0 we don't need to retract anything
                 phases.retractboundary(self, tr, targetphase, [n])
             tr.close()
             branchmap.updatecache(self.filtered('served'))
             return n
         finally:
             if tr:
                 tr.release()
             lock.release()
 
     @unfilteredmethod
     def destroying(self):
         '''Inform the repository that nodes are about to be destroyed.
         Intended for use by strip and rollback, so there's a common
         place for anything that has to be done before destroying history.
 
         This is mostly useful for saving state that is in memory and waiting
         to be flushed when the current lock is released. Because a call to
         destroyed is imminent, the repo will be invalidated causing those
         changes to stay in memory (waiting for the next unlock), or vanish
         completely.
         '''
         # When using the same lock to commit and strip, the phasecache is left
         # dirty after committing. Then when we strip, the repo is invalidated,
         # causing those changes to disappear.
         if '_phasecache' in vars(self):
             self._phasecache.write()
 
     @unfilteredmethod
     def destroyed(self):
         '''Inform the repository that nodes have been destroyed.
         Intended for use by strip and rollback, so there's a common
         place for anything that has to be done after destroying history.
         '''
         # When one tries to:
         # 1) destroy nodes thus calling this method (e.g. strip)
         # 2) use phasecache somewhere (e.g. commit)
         #
         # then 2) will fail because the phasecache contains nodes that were
         # removed. We can either remove phasecache from the filecache,
         # causing it to reload next time it is accessed, or simply filter
         # the removed nodes now and write the updated cache.
         self._phasecache.filterunknown(self)
         self._phasecache.write()
 
         # update the 'served' branch cache to help read only server process
         # Thanks to branchcache collaboration this is done from the nearest
         # filtered subset and it is expected to be fast.
         branchmap.updatecache(self.filtered('served'))
 
         # Ensure the persistent tag cache is updated.  Doing it now
         # means that the tag cache only has to worry about destroyed
         # heads immediately after a strip/rollback.  That in turn
         # guarantees that "cachetip == currenttip" (comparing both rev
         # and node) always means no nodes have been added or destroyed.
 
         # XXX this is suboptimal when qrefresh'ing: we strip the current
         # head, refresh the tag cache, then immediately add a new head.
         # But I think doing it this way is necessary for the "instant
         # tag cache retrieval" case to work.
         self.invalidate()
 
     def walk(self, match, node=None):
         '''
         walk recursively through the directory tree or a given
         changeset, finding all files matched by the match
         function
         '''
         return self[node].walk(match)
 
     def status(self, node1='.', node2=None, match=None,
                ignored=False, clean=False, unknown=False,
                listsubrepos=False):
         '''a convenience method that calls node1.status(node2)'''
         return self[node1].status(node2, match, ignored, clean, unknown,
                                   listsubrepos)
 
     def heads(self, start=None):
         heads = self.changelog.heads(start)
         # sort the output in rev descending order
         return sorted(heads, key=self.changelog.rev, reverse=True)
 
     def branchheads(self, branch=None, start=None, closed=False):
         '''return a (possibly filtered) list of heads for the given branch
 
         Heads are returned in topological order, from newest to oldest.
         If branch is None, use the dirstate branch.
         If start is not None, return only heads reachable from start.
         If closed is True, return heads that are marked as closed as well.
         '''
         if branch is None:
             branch = self[None].branch()
         branches = self.branchmap()
         if branch not in branches:
             return []
         # the cache returns heads ordered lowest to highest
         bheads = list(reversed(branches.branchheads(branch, closed=closed)))
         if start is not None:
             # filter out the heads that cannot be reached from startrev
             fbheads = set(self.changelog.nodesbetween([start], bheads)[2])
             bheads = [h for h in bheads if h in fbheads]
         return bheads
 
     def branches(self, nodes):
         if not nodes:
             nodes = [self.changelog.tip()]
         b = []
         for n in nodes:
             t = n
             while True:
                 p = self.changelog.parents(n)
                 if p[1] != nullid or p[0] == nullid:
                     b.append((t, n, p[0], p[1]))
                     break
                 n = p[0]
         return b
 
     def between(self, pairs):
         r = []
 
         for top, bottom in pairs:
             n, l, i = top, [], 0
             f = 1
 
             while n != bottom and n != nullid:
                 p = self.changelog.parents(n)[0]
                 if i == f:
                     l.append(n)
                     f = f * 2
                 n = p
                 i += 1
 
             r.append(l)
 
         return r
 
     def checkpush(self, pushop):
         """Extensions can override this function if additional checks have
         to be performed before pushing, or call it if they override push
         command.
         """
         pass
 
     @unfilteredpropertycache
     def prepushoutgoinghooks(self):
         """Return util.hooks consists of "(repo, remote, outgoing)"
         functions, which are called before pushing changesets.
         """
         return util.hooks()
 
     def stream_in(self, remote, remotereqs):
         # Save remote branchmap. We will use it later
         # to speed up branchcache creation
         rbranchmap = None
         if remote.capable("branchmap"):
             rbranchmap = remote.branchmap()
 
         fp = remote.stream_out()
         l = fp.readline()
         try:
             resp = int(l)
         except ValueError:
             raise error.ResponseError(
                 _('unexpected response from remote server:'), l)
         if resp == 1:
             raise util.Abort(_('operation forbidden by server'))
         elif resp == 2:
             raise util.Abort(_('locking the remote repository failed'))
         elif resp != 0:
             raise util.Abort(_('the server sent an unknown error code'))
 
         self.applystreamclone(remotereqs, rbranchmap, fp)
         return len(self.heads()) + 1
 
     def applystreamclone(self, remotereqs, remotebranchmap, fp):
         """Apply stream clone data to this repository.
 
         "remotereqs" is a set of requirements to handle the incoming data.
         "remotebranchmap" is the result of a branchmap lookup on the remote. It
         can be None.
         "fp" is a file object containing the raw stream data, suitable for
         feeding into exchange.consumestreamclone.
         """
         lock = self.lock()
         try:
             exchange.consumestreamclone(self, fp)
 
             # new requirements = old non-format requirements +
             #                    new format-related remote requirements
             # requirements from the streamed-in repository
             self.requirements = remotereqs | (
                     self.requirements - self.supportedformats)
             self._applyopenerreqs()
             self._writerequirements()
 
             if remotebranchmap:
                 rbheads = []
                 closed = []
                 for bheads in remotebranchmap.itervalues():
                     rbheads.extend(bheads)
                     for h in bheads:
                         r = self.changelog.rev(h)
                         b, c = self.changelog.branchinfo(r)
                         if c:
                             closed.append(h)
 
                 if rbheads:
                     rtiprev = max((int(self.changelog.rev(node))
                             for node in rbheads))
                     cache = branchmap.branchcache(remotebranchmap,
                                                   self[rtiprev].node(),
                                                   rtiprev,
                                                   closednodes=closed)
                     # Try to stick it as low as possible
                     # filter above served are unlikely to be fetch from a clone
                     for candidate in ('base', 'immutable', 'served'):
                         rview = self.filtered(candidate)
                         if cache.validfor(rview):
                             self._branchcaches[candidate] = cache
                             cache.write(rview)
                             break
             self.invalidate()
         finally:
             lock.release()
 
     def clone(self, remote, heads=[], stream=None):
         '''clone remote repository.
 
         keyword arguments:
         heads: list of revs to clone (forces use of pull)
         stream: use streaming clone if possible'''
 
         # now, all clients that can request uncompressed clones can
         # read repo formats supported by all servers that can serve
         # them.
 
         # if revlog format changes, client will have to check version
         # and format flags on "stream" capability, and use
         # uncompressed only if compatible.
 
         if stream is None:
             # if the server explicitly prefers to stream (for fast LANs)
             stream = remote.capable('stream-preferred')
 
         if stream and not heads:
             # 'stream' means remote revlog format is revlogv1 only
             if remote.capable('stream'):
                 self.stream_in(remote, set(('revlogv1',)))
             else:
                 # otherwise, 'streamreqs' contains the remote revlog format
                 streamreqs = remote.capable('streamreqs')
                 if streamreqs:
                     streamreqs = set(streamreqs.split(','))
                     # if we support it, stream in and adjust our requirements
                     if not streamreqs - self.supportedformats:
                         self.stream_in(remote, streamreqs)
 
         # internal config: ui.quietbookmarkmove
         quiet = self.ui.backupconfig('ui', 'quietbookmarkmove')
         try:
             self.ui.setconfig('ui', 'quietbookmarkmove', True, 'clone')
             ret = exchange.pull(self, remote, heads).cgresult
         finally:
             self.ui.restoreconfig(quiet)
         return ret
 
     def pushkey(self, namespace, key, old, new):
         try:
             tr = self.currenttransaction()
             hookargs = {}
             if tr is not None:
                 hookargs.update(tr.hookargs)
                 pending = lambda: tr.writepending() and self.root or ""
                 hookargs['pending'] = pending
             hookargs['namespace'] = namespace
             hookargs['key'] = key
             hookargs['old'] = old
             hookargs['new'] = new
             self.hook('prepushkey', throw=True, **hookargs)
         except error.HookAbort as exc:
             self.ui.write_err(_("pushkey-abort: %s\n") % exc)
             if exc.hint:
                 self.ui.write_err(_("(%s)\n") % exc.hint)
             return False
         self.ui.debug('pushing key for "%s:%s"\n' % (namespace, key))
         ret = pushkey.push(self, namespace, key, old, new)
         def runhook():
             self.hook('pushkey', namespace=namespace, key=key, old=old, new=new,
                       ret=ret)
         self._afterlock(runhook)
         return ret
 
     def listkeys(self, namespace):
         self.hook('prelistkeys', throw=True, namespace=namespace)
         self.ui.debug('listing keys for "%s"\n' % namespace)
         values = pushkey.list(self, namespace)
         self.hook('listkeys', namespace=namespace, values=values)
         return values
 
     def debugwireargs(self, one, two, three=None, four=None, five=None):
         '''used to test argument passing over the wire'''
         return "%s %s %s %s %s" % (one, two, three, four, five)
 
     def savecommitmessage(self, text):
         fp = self.vfs('last-message.txt', 'wb')
         try:
             fp.write(text)
         finally:
             fp.close()
         return self.pathto(fp.name[len(self.root) + 1:])
 
 # used to avoid circular references so destructors work
 def aftertrans(files):
     renamefiles = [tuple(t) for t in files]
     def a():
         for vfs, src, dest in renamefiles:
             try:
                 vfs.rename(src, dest)
             except OSError: # journal file does not yet exist
                 pass
     return a
 
 def undoname(fn):
     base, name = os.path.split(fn)
     assert name.startswith('journal')
     return os.path.join(base, name.replace('journal', 'undo', 1))
 
 def instance(ui, path, create):
     return localrepository(ui, util.urllocalpath(path), create)
 
 def islocal(path):
     return True

mercurial/revlog.py

@@ -1,1629 +1,1651
 # revlog.py - storage back-end for mercurial
 #
 # Copyright 2005-2007 Matt Mackall <[e-mail]>
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 """Storage back-end for Mercurial.
 
 This provides efficient delta storage with O(1) retrieve and append
 and O(changes) merge between branches.
 """
 
 # import stuff from node for others to import from revlog
 import collections
 from node import bin, hex, nullid, nullrev
 from i18n import _
 import ancestor, mdiff, parsers, error, util, templatefilters
 import struct, zlib, errno
 
 _pack = struct.pack
 _unpack = struct.unpack
 _compress = zlib.compress
 _decompress = zlib.decompress
 _sha = util.sha1
 
 # revlog header flags
 REVLOGV0 = 0
 REVLOGNG = 1
 REVLOGNGINLINEDATA = (1 << 16)
 REVLOGGENERALDELTA = (1 << 17)
 REVLOG_DEFAULT_FLAGS = REVLOGNGINLINEDATA
 REVLOG_DEFAULT_FORMAT = REVLOGNG
 REVLOG_DEFAULT_VERSION = REVLOG_DEFAULT_FORMAT | REVLOG_DEFAULT_FLAGS
 REVLOGNG_FLAGS = REVLOGNGINLINEDATA | REVLOGGENERALDELTA
 
 # revlog index flags
 REVIDX_ISCENSORED = (1 << 15) # revision has censor metadata, must be verified
 REVIDX_DEFAULT_FLAGS = 0
 REVIDX_KNOWN_FLAGS = REVIDX_ISCENSORED
 
 # max size of revlog with inline data
 _maxinline = 131072
 _chunksize = 1048576
 
 RevlogError = error.RevlogError
 LookupError = error.LookupError
 CensoredNodeError = error.CensoredNodeError
 
 def getoffset(q):
     return int(q >> 16)
 
 def gettype(q):
     return int(q & 0xFFFF)
 
 def offset_type(offset, type):
     return long(long(offset) << 16 | type)
 
 _nullhash = _sha(nullid)
 
 def hash(text, p1, p2):
     """generate a hash from the given text and its parent hashes
 
     This hash combines both the current file contents and its history
     in a manner that makes it easy to distinguish nodes with the same
     content in the revision graph.
     """
     # As of now, if one of the parent node is null, p2 is null
     if p2 == nullid:
         # deep copy of a hash is faster than creating one
         s = _nullhash.copy()
         s.update(p1)
     else:
         # none of the parent nodes are nullid
         l = [p1, p2]
         l.sort()
         s = _sha(l[0])
         s.update(l[1])
     s.update(text)
     return s.digest()
 
 def decompress(bin):
     """ decompress the given input """
     if not bin:
         return bin
     t = bin[0]
     if t == '\0':
         return bin
     if t == 'x':
         try:
             return _decompress(bin)
         except zlib.error as e:
             raise RevlogError(_("revlog decompress error: %s") % str(e))
     if t == 'u':
         return bin[1:]
     raise RevlogError(_("unknown compression type %r") % t)
 
 # index v0:
 #  4 bytes: offset
 #  4 bytes: compressed length
 #  4 bytes: base rev
 #  4 bytes: link rev
 # 20 bytes: parent 1 nodeid
 # 20 bytes: parent 2 nodeid
 # 20 bytes: nodeid
 indexformatv0 = ">4l20s20s20s"
 
 class revlogoldio(object):
     def __init__(self):
         self.size = struct.calcsize(indexformatv0)
 
     def parseindex(self, data, inline):
         s = self.size
         index = []
         nodemap =  {nullid: nullrev}
         n = off = 0
         l = len(data)
         while off + s <= l:
             cur = data[off:off + s]
             off += s
             e = _unpack(indexformatv0, cur)
             # transform to revlogv1 format
             e2 = (offset_type(e[0], 0), e[1], -1, e[2], e[3],
                   nodemap.get(e[4], nullrev), nodemap.get(e[5], nullrev), e[6])
             index.append(e2)
             nodemap[e[6]] = n
             n += 1
 
         # add the magic null revision at -1
         index.append((0, 0, 0, -1, -1, -1, -1, nullid))
 
         return index, nodemap, None
 
     def packentry(self, entry, node, version, rev):
         if gettype(entry[0]):
             raise RevlogError(_("index entry flags need RevlogNG"))
         e2 = (getoffset(entry[0]), entry[1], entry[3], entry[4],
               node(entry[5]), node(entry[6]), entry[7])
         return _pack(indexformatv0, *e2)
 
 # index ng:
 #  6 bytes: offset
 #  2 bytes: flags
 #  4 bytes: compressed length
 #  4 bytes: uncompressed length
 #  4 bytes: base rev
 #  4 bytes: link rev
 #  4 bytes: parent 1 rev
 #  4 bytes: parent 2 rev
 # 32 bytes: nodeid
 indexformatng = ">Qiiiiii20s12x"
 versionformat = ">I"
 
 # corresponds to uncompressed length of indexformatng (2 gigs, 4-byte
 # signed integer)
 _maxentrysize = 0x7fffffff
 
 class revlogio(object):
     def __init__(self):
         self.size = struct.calcsize(indexformatng)
 
     def parseindex(self, data, inline):
         # call the C implementation to parse the index data
         index, cache = parsers.parse_index2(data, inline)
         return index, getattr(index, 'nodemap', None), cache
 
     def packentry(self, entry, node, version, rev):
         p = _pack(indexformatng, *entry)
         if rev == 0:
             p = _pack(versionformat, version) + p[4:]
         return p
 
 class revlog(object):
     """
     the underlying revision storage object
 
     A revlog consists of two parts, an index and the revision data.
 
     The index is a file with a fixed record size containing
     information on each revision, including its nodeid (hash), the
     nodeids of its parents, the position and offset of its data within
     the data file, and the revision it's based on. Finally, each entry
     contains a linkrev entry that can serve as a pointer to external
     data.
 
     The revision data itself is a linear collection of data chunks.
     Each chunk represents a revision and is usually represented as a
     delta against the previous chunk. To bound lookup time, runs of
     deltas are limited to about 2 times the length of the original
     version data. This makes retrieval of a version proportional to
     its size, or O(1) relative to the number of revisions.
 
     Both pieces of the revlog are written to in an append-only
     fashion, which means we never need to rewrite a file to insert or
     remove data, and can use some simple techniques to avoid the need
     for locking while reading.
     """
     def __init__(self, opener, indexfile):
         """
         create a revlog object
 
         opener is a function that abstracts the file opening operation
         and can be used to implement COW semantics or the like.
         """
         self.indexfile = indexfile
         self.datafile = indexfile[:-2] + ".d"
         self.opener = opener
         self._cache = None
         self._basecache = None
         self._chunkcache = (0, '')
         self._chunkcachesize = 65536
         self._maxchainlen = None
+        self._aggressivemergedeltas = False
         self.index = []
         self._pcache = {}
         self._nodecache = {nullid: nullrev}
         self._nodepos = None
 
         v = REVLOG_DEFAULT_VERSION
         opts = getattr(opener, 'options', None)
         if opts is not None:
             if 'revlogv1' in opts:
                 if 'generaldelta' in opts:
                     v |= REVLOGGENERALDELTA
             else:
                 v = 0
             if 'chunkcachesize' in opts:
                 self._chunkcachesize = opts['chunkcachesize']
             if 'maxchainlen' in opts:
                 self._maxchainlen = opts['maxchainlen']
+            if 'aggressivemergedeltas' in opts:
+                self._aggressivemergedeltas = opts['aggressivemergedeltas']
 
         if self._chunkcachesize <= 0:
             raise RevlogError(_('revlog chunk cache size %r is not greater '
                                 'than 0') % self._chunkcachesize)
         elif self._chunkcachesize & (self._chunkcachesize - 1):
             raise RevlogError(_('revlog chunk cache size %r is not a power '
                                 'of 2') % self._chunkcachesize)
 
         i = ''
         self._initempty = True
         try:
             f = self.opener(self.indexfile)
             i = f.read()
             f.close()
             if len(i) > 0:
                 v = struct.unpack(versionformat, i[:4])[0]
                 self._initempty = False
         except IOError as inst:
             if inst.errno != errno.ENOENT:
                 raise
 
         self.version = v
         self._inline = v & REVLOGNGINLINEDATA
         self._generaldelta = v & REVLOGGENERALDELTA
         flags = v & ~0xFFFF
         fmt = v & 0xFFFF
         if fmt == REVLOGV0 and flags:
             raise RevlogError(_("index %s unknown flags %#04x for format v0")
                               % (self.indexfile, flags >> 16))
         elif fmt == REVLOGNG and flags & ~REVLOGNG_FLAGS:
             raise RevlogError(_("index %s unknown flags %#04x for revlogng")
                               % (self.indexfile, flags >> 16))
         elif fmt > REVLOGNG:
             raise RevlogError(_("index %s unknown format %d")
                               % (self.indexfile, fmt))
 
         self._io = revlogio()
         if self.version == REVLOGV0:
             self._io = revlogoldio()
         try:
             d = self._io.parseindex(i, self._inline)
         except (ValueError, IndexError):
             raise RevlogError(_("index %s is corrupted") % (self.indexfile))
         self.index, nodemap, self._chunkcache = d
         if nodemap is not None:
             self.nodemap = self._nodecache = nodemap
         if not self._chunkcache:
             self._chunkclear()
         # revnum -> (chain-length, sum-delta-length)
         self._chaininfocache = {}
 
     def tip(self):
         return self.node(len(self.index) - 2)
     def __contains__(self, rev):
         return 0 <= rev < len(self)
     def __len__(self):
         return len(self.index) - 1
     def __iter__(self):
         return iter(xrange(len(self)))
     def revs(self, start=0, stop=None):
         """iterate over all rev in this revlog (from start to stop)"""
         step = 1
         if stop is not None:
             if start > stop:
                 step = -1
             stop += step
         else:
             stop = len(self)
         return xrange(start, stop, step)
 
     @util.propertycache
     def nodemap(self):
         self.rev(self.node(0))
         return self._nodecache
 
     def hasnode(self, node):
         try:
             self.rev(node)
             return True
         except KeyError:
             return False
 
     def clearcaches(self):
         try:
             self._nodecache.clearcaches()
         except AttributeError:
             self._nodecache = {nullid: nullrev}
             self._nodepos = None
 
     def rev(self, node):
         try:
             return self._nodecache[node]
         except TypeError:
             raise
         except RevlogError:
             # parsers.c radix tree lookup failed
             raise LookupError(node, self.indexfile, _('no node'))
         except KeyError:
             # pure python cache lookup failed
             n = self._nodecache
             i = self.index
             p = self._nodepos
             if p is None:
                 p = len(i) - 2
             for r in xrange(p, -1, -1):
                 v = i[r][7]
                 n[v] = r
                 if v == node:
                     self._nodepos = r - 1
                     return r
             raise LookupError(node, self.indexfile, _('no node'))
 
     def node(self, rev):
         return self.index[rev][7]
     def linkrev(self, rev):
         return self.index[rev][4]
     def parents(self, node):
         i = self.index
         d = i[self.rev(node)]
         return i[d[5]][7], i[d[6]][7] # map revisions to nodes inline
     def parentrevs(self, rev):
         return self.index[rev][5:7]
     def start(self, rev):
         return int(self.index[rev][0] >> 16)
     def end(self, rev):
         return self.start(rev) + self.length(rev)
     def length(self, rev):
         return self.index[rev][1]
     def chainbase(self, rev):
         index = self.index
         base = index[rev][3]
         while base != rev:
             rev = base
             base = index[rev][3]
         return base
     def chainlen(self, rev):
         return self._chaininfo(rev)[0]
 
     def _chaininfo(self, rev):
         chaininfocache = self._chaininfocache
         if rev in chaininfocache:
             return chaininfocache[rev]
         index = self.index
         generaldelta = self._generaldelta
         iterrev = rev
         e = index[iterrev]
         clen = 0
         compresseddeltalen = 0
         while iterrev != e[3]:
             clen += 1
             compresseddeltalen += e[1]
             if generaldelta:
                 iterrev = e[3]
             else:
                 iterrev -= 1
             if iterrev in chaininfocache:
                 t = chaininfocache[iterrev]
                 clen += t[0]
                 compresseddeltalen += t[1]
                 break
             e = index[iterrev]
         else:
             # Add text length of base since decompressing that also takes
             # work. For cache hits the length is already included.
             compresseddeltalen += e[1]
         r = (clen, compresseddeltalen)
         chaininfocache[rev] = r
         return r
 
     def flags(self, rev):
         return self.index[rev][0] & 0xFFFF
     def rawsize(self, rev):
         """return the length of the uncompressed text for a given revision"""
         l = self.index[rev][2]
         if l >= 0:
             return l
 
         t = self.revision(self.node(rev))
         return len(t)
     size = rawsize
 
     def ancestors(self, revs, stoprev=0, inclusive=False):
         """Generate the ancestors of 'revs' in reverse topological order.
         Does not generate revs lower than stoprev.
 
         See the documentation for ancestor.lazyancestors for more details."""
 
         return ancestor.lazyancestors(self.parentrevs, revs, stoprev=stoprev,
                                       inclusive=inclusive)
 
     def descendants(self, revs):
         """Generate the descendants of 'revs' in revision order.
 
         Yield a sequence of revision numbers starting with a child of
         some rev in revs, i.e., each revision is *not* considered a
         descendant of itself.  Results are ordered by revision number (a
         topological sort)."""
         first = min(revs)
         if first == nullrev:
             for i in self:
                 yield i
             return
 
         seen = set(revs)
         for i in self.revs(start=first + 1):
             for x in self.parentrevs(i):
                 if x != nullrev and x in seen:
                     seen.add(i)
                     yield i
                     break
 
     def findcommonmissing(self, common=None, heads=None):
         """Return a tuple of the ancestors of common and the ancestors of heads
         that are not ancestors of common. In revset terminology, we return the
         tuple:
 
           ::common, (::heads) - (::common)
 
         The list is sorted by revision number, meaning it is
         topologically sorted.
 
         'heads' and 'common' are both lists of node IDs.  If heads is
         not supplied, uses all of the revlog's heads.  If common is not
         supplied, uses nullid."""
         if common is None:
             common = [nullid]
         if heads is None:
             heads = self.heads()
 
         common = [self.rev(n) for n in common]
         heads = [self.rev(n) for n in heads]
 
         # we want the ancestors, but inclusive
         class lazyset(object):
             def __init__(self, lazyvalues):
                 self.addedvalues = set()
                 self.lazyvalues = lazyvalues
 
             def __contains__(self, value):
                 return value in self.addedvalues or value in self.lazyvalues
 
             def __iter__(self):
                 added = self.addedvalues
                 for r in added:
                     yield r
                 for r in self.lazyvalues:
                     if not r in added:
                         yield r
 
             def add(self, value):
                 self.addedvalues.add(value)
 
             def update(self, values):
                 self.addedvalues.update(values)
 
         has = lazyset(self.ancestors(common))
         has.add(nullrev)
         has.update(common)
 
         # take all ancestors from heads that aren't in has
         missing = set()
         visit = collections.deque(r for r in heads if r not in has)
         while visit:
             r = visit.popleft()
             if r in missing:
                 continue
             else:
                 missing.add(r)
                 for p in self.parentrevs(r):
                     if p not in has:
                         visit.append(p)
         missing = list(missing)
         missing.sort()
         return has, [self.node(r) for r in missing]
 
     def incrementalmissingrevs(self, common=None):
         """Return an object that can be used to incrementally compute the
         revision numbers of the ancestors of arbitrary sets that are not
         ancestors of common. This is an ancestor.incrementalmissingancestors
         object.
 
         'common' is a list of revision numbers. If common is not supplied, uses
         nullrev.
         """
         if common is None:
             common = [nullrev]
 
         return ancestor.incrementalmissingancestors(self.parentrevs, common)
 
     def findmissingrevs(self, common=None, heads=None):
         """Return the revision numbers of the ancestors of heads that
         are not ancestors of common.
 
         More specifically, return a list of revision numbers corresponding to
         nodes N such that every N satisfies the following constraints:
 
           1. N is an ancestor of some node in 'heads'
           2. N is not an ancestor of any node in 'common'
 
         The list is sorted by revision number, meaning it is
         topologically sorted.
 
         'heads' and 'common' are both lists of revision numbers.  If heads is
         not supplied, uses all of the revlog's heads.  If common is not
         supplied, uses nullid."""
         if common is None:
             common = [nullrev]
         if heads is None:
             heads = self.headrevs()
 
         inc = self.incrementalmissingrevs(common=common)
         return inc.missingancestors(heads)
 
     def findmissing(self, common=None, heads=None):
         """Return the ancestors of heads that are not ancestors of common.
 
         More specifically, return a list of nodes N such that every N
         satisfies the following constraints:
 
           1. N is an ancestor of some node in 'heads'
           2. N is not an ancestor of any node in 'common'
 
         The list is sorted by revision number, meaning it is
         topologically sorted.
 
         'heads' and 'common' are both lists of node IDs.  If heads is
         not supplied, uses all of the revlog's heads.  If common is not
         supplied, uses nullid."""
         if common is None:
             common = [nullid]
         if heads is None:
             heads = self.heads()
 
         common = [self.rev(n) for n in common]
         heads = [self.rev(n) for n in heads]
 
         inc = self.incrementalmissingrevs(common=common)
         return [self.node(r) for r in inc.missingancestors(heads)]
 
     def nodesbetween(self, roots=None, heads=None):
         """Return a topological path from 'roots' to 'heads'.
 
         Return a tuple (nodes, outroots, outheads) where 'nodes' is a
         topologically sorted list of all nodes N that satisfy both of
         these constraints:
 
           1. N is a descendant of some node in 'roots'
           2. N is an ancestor of some node in 'heads'
 
         Every node is considered to be both a descendant and an ancestor
         of itself, so every reachable node in 'roots' and 'heads' will be
         included in 'nodes'.
 
         'outroots' is the list of reachable nodes in 'roots', i.e., the
         subset of 'roots' that is returned in 'nodes'.  Likewise,
         'outheads' is the subset of 'heads' that is also in 'nodes'.
 
         'roots' and 'heads' are both lists of node IDs.  If 'roots' is
         unspecified, uses nullid as the only root.  If 'heads' is
         unspecified, uses list of all of the revlog's heads."""
         nonodes = ([], [], [])
         if roots is not None:
             roots = list(roots)
             if not roots:
                 return nonodes
             lowestrev = min([self.rev(n) for n in roots])
         else:
             roots = [nullid] # Everybody's a descendant of nullid
             lowestrev = nullrev
         if (lowestrev == nullrev) and (heads is None):
             # We want _all_ the nodes!
             return ([self.node(r) for r in self], [nullid], list(self.heads()))
         if heads is None:
             # All nodes are ancestors, so the latest ancestor is the last
             # node.
             highestrev = len(self) - 1
             # Set ancestors to None to signal that every node is an ancestor.
             ancestors = None
             # Set heads to an empty dictionary for later discovery of heads
             heads = {}
         else:
             heads = list(heads)
             if not heads:
                 return nonodes
             ancestors = set()
             # Turn heads into a dictionary so we can remove 'fake' heads.
             # Also, later we will be using it to filter out the heads we can't
             # find from roots.
             heads = dict.fromkeys(heads, False)
             # Start at the top and keep marking parents until we're done.
             nodestotag = set(heads)
             # Remember where the top was so we can use it as a limit later.
             highestrev = max([self.rev(n) for n in nodestotag])
             while nodestotag:
                 # grab a node to tag
                 n = nodestotag.pop()
                 # Never tag nullid
                 if n == nullid:
                     continue
                 # A node's revision number represents its place in a
                 # topologically sorted list of nodes.
                 r = self.rev(n)
                 if r >= lowestrev:
                     if n not in ancestors:
                         # If we are possibly a descendant of one of the roots
                         # and we haven't already been marked as an ancestor
                         ancestors.add(n) # Mark as ancestor
                         # Add non-nullid parents to list of nodes to tag.
                         nodestotag.update([p for p in self.parents(n) if
                                            p != nullid])
                     elif n in heads: # We've seen it before, is it a fake head?
                         # So it is, real heads should not be the ancestors of
                         # any other heads.
                         heads.pop(n)
             if not ancestors:
                 return nonodes
             # Now that we have our set of ancestors, we want to remove any
             # roots that are not ancestors.
 
             # If one of the roots was nullid, everything is included anyway.
             if lowestrev > nullrev:
                 # But, since we weren't, let's recompute the lowest rev to not
                 # include roots that aren't ancestors.
 
                 # Filter out roots that aren't ancestors of heads
                 roots = [n for n in roots if n in ancestors]
                 # Recompute the lowest revision
                 if roots:
                     lowestrev = min([self.rev(n) for n in roots])
                 else:
                     # No more roots?  Return empty list
                     return nonodes
             else:
                 # We are descending from nullid, and don't need to care about
                 # any other roots.
                 lowestrev = nullrev
                 roots = [nullid]
         # Transform our roots list into a set.
         descendants = set(roots)
         # Also, keep the original roots so we can filter out roots that aren't
         # 'real' roots (i.e. are descended from other roots).
         roots = descendants.copy()
         # Our topologically sorted list of output nodes.
         orderedout = []
         # Don't start at nullid since we don't want nullid in our output list,
         # and if nullid shows up in descendants, empty parents will look like
         # they're descendants.
         for r in self.revs(start=max(lowestrev, 0), stop=highestrev + 1):
             n = self.node(r)
             isdescendant = False
             if lowestrev == nullrev:  # Everybody is a descendant of nullid
                 isdescendant = True
             elif n in descendants:
                 # n is already a descendant
                 isdescendant = True
                 # This check only needs to be done here because all the roots
                 # will start being marked is descendants before the loop.
                 if n in roots:
                     # If n was a root, check if it's a 'real' root.
                     p = tuple(self.parents(n))
                     # If any of its parents are descendants, it's not a root.
                     if (p[0] in descendants) or (p[1] in descendants):
                         roots.remove(n)
             else:
                 p = tuple(self.parents(n))
                 # A node is a descendant if either of its parents are
                 # descendants.  (We seeded the dependents list with the roots
                 # up there, remember?)
                 if (p[0] in descendants) or (p[1] in descendants):
                     descendants.add(n)
                     isdescendant = True
             if isdescendant and ((ancestors is None) or (n in ancestors)):
                 # Only include nodes that are both descendants and ancestors.
                 orderedout.append(n)
                 if (ancestors is not None) and (n in heads):
                     # We're trying to figure out which heads are reachable
                     # from roots.
                     # Mark this head as having been reached
                     heads[n] = True
                 elif ancestors is None:
                     # Otherwise, we're trying to discover the heads.
                     # Assume this is a head because if it isn't, the next step
                     # will eventually remove it.
                     heads[n] = True
                     # But, obviously its parents aren't.
                     for p in self.parents(n):
                         heads.pop(p, None)
         heads = [n for n, flag in heads.iteritems() if flag]
         roots = list(roots)
         assert orderedout
         assert roots
         assert heads
         return (orderedout, roots, heads)
 
     def headrevs(self):
         try:
             return self.index.headrevs()
         except AttributeError:
             return self._headrevs()
 
     def computephases(self, roots):
         return self.index.computephasesmapsets(roots)
 
     def _headrevs(self):
         count = len(self)
         if not count:
             return [nullrev]
         # we won't iter over filtered rev so nobody is a head at start
         ishead = [0] * (count + 1)
         index = self.index
         for r in self:
             ishead[r] = 1  # I may be an head
             e = index[r]
             ishead[e[5]] = ishead[e[6]] = 0  # my parent are not
         return [r for r, val in enumerate(ishead) if val]
 
     def heads(self, start=None, stop=None):
         """return the list of all nodes that have no children
 
         if start is specified, only heads that are descendants of
         start will be returned
         if stop is specified, it will consider all the revs from stop
         as if they had no children
         """
         if start is None and stop is None:
             if not len(self):
                 return [nullid]
             return [self.node(r) for r in self.headrevs()]
 
         if start is None:
             start = nullid
         if stop is None:
             stop = []
         stoprevs = set([self.rev(n) for n in stop])
         startrev = self.rev(start)
         reachable = set((startrev,))
         heads = set((startrev,))
 
         parentrevs = self.parentrevs
         for r in self.revs(start=startrev + 1):
             for p in parentrevs(r):
                 if p in reachable:
                     if r not in stoprevs:
                         reachable.add(r)
                     heads.add(r)
                 if p in heads and p not in stoprevs:
                     heads.remove(p)
 
         return [self.node(r) for r in heads]
 
     def children(self, node):
         """find the children of a given node"""
         c = []
         p = self.rev(node)
         for r in self.revs(start=p + 1):
             prevs = [pr for pr in self.parentrevs(r) if pr != nullrev]
             if prevs:
                 for pr in prevs:
                     if pr == p:
                         c.append(self.node(r))
             elif p == nullrev:
                 c.append(self.node(r))
         return c
 
     def descendant(self, start, end):
         if start == nullrev:
             return True
         for i in self.descendants([start]):
             if i == end:
                 return True
             elif i > end:
                 break
         return False
 
     def commonancestorsheads(self, a, b):
         """calculate all the heads of the common ancestors of nodes a and b"""
         a, b = self.rev(a), self.rev(b)
         try:
             ancs = self.index.commonancestorsheads(a, b)
         except (AttributeError, OverflowError): # C implementation failed
             ancs = ancestor.commonancestorsheads(self.parentrevs, a, b)
         return map(self.node, ancs)
 
     def isancestor(self, a, b):
         """return True if node a is an ancestor of node b
 
         The implementation of this is trivial but the use of
         commonancestorsheads is not."""
         return a in self.commonancestorsheads(a, b)
 
     def ancestor(self, a, b):
         """calculate the "best" common ancestor of nodes a and b"""
 
         a, b = self.rev(a), self.rev(b)
         try:
             ancs = self.index.ancestors(a, b)
         except (AttributeError, OverflowError):
             ancs = ancestor.ancestors(self.parentrevs, a, b)
         if ancs:
             # choose a consistent winner when there's a tie
             return min(map(self.node, ancs))
         return nullid
 
     def _match(self, id):
         if isinstance(id, int):
             # rev
             return self.node(id)
         if len(id) == 20:
             # possibly a binary node
             # odds of a binary node being all hex in ASCII are 1 in 10**25
             try:
                 node = id
                 self.rev(node) # quick search the index
                 return node
             except LookupError:
                 pass # may be partial hex id
         try:
             # str(rev)
             rev = int(id)
             if str(rev) != id:
                 raise ValueError
             if rev < 0:
                 rev = len(self) + rev
             if rev < 0 or rev >= len(self):
                 raise ValueError
             return self.node(rev)
         except (ValueError, OverflowError):
             pass
         if len(id) == 40:
             try:
                 # a full hex nodeid?
                 node = bin(id)
                 self.rev(node)
                 return node
             except (TypeError, LookupError):
                 pass
 
     def _partialmatch(self, id):
         try:
             n = self.index.partialmatch(id)
             if n and self.hasnode(n):
                 return n
             return None
         except RevlogError:
             # parsers.c radix tree lookup gave multiple matches
             # fall through to slow path that filters hidden revisions
             pass
         except (AttributeError, ValueError):
             # we are pure python, or key was too short to search radix tree
             pass
 
         if id in self._pcache:
             return self._pcache[id]
 
         if len(id) < 40:
             try:
                 # hex(node)[:...]
                 l = len(id) // 2  # grab an even number of digits
                 prefix = bin(id[:l * 2])
                 nl = [e[7] for e in self.index if e[7].startswith(prefix)]
                 nl = [n for n in nl if hex(n).startswith(id) and
                       self.hasnode(n)]
                 if len(nl) > 0:
                     if len(nl) == 1:
                         self._pcache[id] = nl[0]
                         return nl[0]
                     raise LookupError(id, self.indexfile,
                                       _('ambiguous identifier'))
                 return None
             except TypeError:
                 pass
 
     def lookup(self, id):
         """locate a node based on:
             - revision number or str(revision number)
             - nodeid or subset of hex nodeid
         """
         n = self._match(id)
         if n is not None:
             return n
         n = self._partialmatch(id)
         if n:
             return n
 
         raise LookupError(id, self.indexfile, _('no match found'))
 
     def cmp(self, node, text):
         """compare text with a given file revision
 
         returns True if text is different than what is stored.
         """
         p1, p2 = self.parents(node)
         return hash(text, p1, p2) != node
 
     def _addchunk(self, offset, data):
         o, d = self._chunkcache
         # try to add to existing cache
         if o + len(d) == offset and len(d) + len(data) < _chunksize:
             self._chunkcache = o, d + data
         else:
             self._chunkcache = offset, data
 
     def _loadchunk(self, offset, length):
         if self._inline:
             df = self.opener(self.indexfile)
         else:
             df = self.opener(self.datafile)
 
         # Cache data both forward and backward around the requested
         # data, in a fixed size window. This helps speed up operations
         # involving reading the revlog backwards.
         cachesize = self._chunkcachesize
         realoffset = offset & ~(cachesize - 1)
         reallength = (((offset + length + cachesize) & ~(cachesize - 1))
                       - realoffset)
         df.seek(realoffset)
         d = df.read(reallength)
         df.close()
         self._addchunk(realoffset, d)
         if offset != realoffset or reallength != length:
             return util.buffer(d, offset - realoffset, length)
         return d
 
     def _getchunk(self, offset, length):
         o, d = self._chunkcache
         l = len(d)
 
         # is it in the cache?
         cachestart = offset - o
         cacheend = cachestart + length
         if cachestart >= 0 and cacheend <= l:
             if cachestart == 0 and cacheend == l:
                 return d # avoid a copy
             return util.buffer(d, cachestart, cacheend - cachestart)
 
         return self._loadchunk(offset, length)
 
     def _chunkraw(self, startrev, endrev):
         start = self.start(startrev)
         end = self.end(endrev)
         if self._inline:
             start += (startrev + 1) * self._io.size
             end += (endrev + 1) * self._io.size
         length = end - start
         return self._getchunk(start, length)
 
     def _chunk(self, rev):
         return decompress(self._chunkraw(rev, rev))
 
     def _chunks(self, revs):
         '''faster version of [self._chunk(rev) for rev in revs]
 
         Assumes that revs is in ascending order.'''
         if not revs:
             return []
         start = self.start
         length = self.length
         inline = self._inline
         iosize = self._io.size
         buffer = util.buffer
 
         l = []
         ladd = l.append
 
         # preload the cache
         try:
             while True:
                 # ensure that the cache doesn't change out from under us
                 _cache = self._chunkcache
                 self._chunkraw(revs[0], revs[-1])
                 if _cache == self._chunkcache:
                     break
             offset, data = _cache
         except OverflowError:
             # issue4215 - we can't cache a run of chunks greater than
             # 2G on Windows
             return [self._chunk(rev) for rev in revs]
 
         for rev in revs:
             chunkstart = start(rev)
             if inline:
                 chunkstart += (rev + 1) * iosize
             chunklength = length(rev)
             ladd(decompress(buffer(data, chunkstart - offset, chunklength)))
 
         return l
 
     def _chunkclear(self):
         self._chunkcache = (0, '')
 
     def deltaparent(self, rev):
         """return deltaparent of the given revision"""
         base = self.index[rev][3]
         if base == rev:
             return nullrev
         elif self._generaldelta:
             return base
         else:
             return rev - 1
 
     def revdiff(self, rev1, rev2):
         """return or calculate a delta between two revisions"""
         if rev1 != nullrev and self.deltaparent(rev2) == rev1:
             return str(self._chunk(rev2))
 
         return mdiff.textdiff(self.revision(rev1),
                               self.revision(rev2))
 
     def revision(self, nodeorrev):
         """return an uncompressed revision of a given node or revision
         number.
         """
         if isinstance(nodeorrev, int):
             rev = nodeorrev
             node = self.node(rev)
         else:
             node = nodeorrev
             rev = None
 
         _cache = self._cache # grab local copy of cache to avoid thread race
         cachedrev = None
         if node == nullid:
             return ""
         if _cache:
             if _cache[0] == node:
                 return _cache[2]
             cachedrev = _cache[1]
 
         # look up what we need to read
         text = None
         if rev is None:
             rev = self.rev(node)
 
         # check rev flags
         if self.flags(rev) & ~REVIDX_KNOWN_FLAGS:
             raise RevlogError(_('incompatible revision flag %x') %
                               (self.flags(rev) & ~REVIDX_KNOWN_FLAGS))
 
         # build delta chain
         chain = []
         index = self.index # for performance
         generaldelta = self._generaldelta
         iterrev = rev
         e = index[iterrev]
         while iterrev != e[3] and iterrev != cachedrev:
             chain.append(iterrev)
             if generaldelta:
                 iterrev = e[3]
             else:
                 iterrev -= 1
             e = index[iterrev]
 
         if iterrev == cachedrev:
             # cache hit
             text = _cache[2]
         else:
             chain.append(iterrev)
         chain.reverse()
 
         # drop cache to save memory
         self._cache = None
 
         bins = self._chunks(chain)
         if text is None:
             text = str(bins[0])
             bins = bins[1:]
 
         text = mdiff.patches(text, bins)
 
         text = self._checkhash(text, node, rev)
 
         self._cache = (node, rev, text)
         return text
 
     def hash(self, text, p1, p2):
         """Compute a node hash.
 
         Available as a function so that subclasses can replace the hash
         as needed.
         """
         return hash(text, p1, p2)
 
     def _checkhash(self, text, node, rev):
         p1, p2 = self.parents(node)
         self.checkhash(text, p1, p2, node, rev)
         return text
 
     def checkhash(self, text, p1, p2, node, rev=None):
         if node != self.hash(text, p1, p2):
             revornode = rev
             if revornode is None:
                 revornode = templatefilters.short(hex(node))
             raise RevlogError(_("integrity check failed on %s:%s")
                 % (self.indexfile, revornode))
 
     def checkinlinesize(self, tr, fp=None):
         if not self._inline or (self.start(-2) + self.length(-2)) < _maxinline:
             return
 
         trinfo = tr.find(self.indexfile)
         if trinfo is None:
             raise RevlogError(_("%s not found in the transaction")
                               % self.indexfile)
 
         trindex = trinfo[2]
         if trindex is not None:
             dataoff = self.start(trindex)
         else:
             # revlog was stripped at start of transaction, use all leftover data
             trindex = len(self) - 1
             dataoff = self.end(-2)
 
         tr.add(self.datafile, dataoff)
 
         if fp:
             fp.flush()
             fp.close()
 
         df = self.opener(self.datafile, 'w')
         try:
             for r in self:
                 df.write(self._chunkraw(r, r))
         finally:
             df.close()
 
         fp = self.opener(self.indexfile, 'w', atomictemp=True)
         self.version &= ~(REVLOGNGINLINEDATA)
         self._inline = False
         for i in self:
             e = self._io.packentry(self.index[i], self.node, self.version, i)
             fp.write(e)
 
         # if we don't call close, the temp file will never replace the
         # real index
         fp.close()
 
         tr.replace(self.indexfile, trindex * self._io.size)
         self._chunkclear()
 
     def addrevision(self, text, transaction, link, p1, p2, cachedelta=None,
                     node=None):
         """add a revision to the log
 
         text - the revision data to add
         transaction - the transaction object used for rollback
         link - the linkrev data to add
         p1, p2 - the parent nodeids of the revision
         cachedelta - an optional precomputed delta
         node - nodeid of revision; typically node is not specified, and it is
             computed by default as hash(text, p1, p2), however subclasses might
             use different hashing method (and override checkhash() in such case)
         """
         if link == nullrev:
             raise RevlogError(_("attempted to add linkrev -1 to %s")
                               % self.indexfile)
 
         if len(text) > _maxentrysize:
             raise RevlogError(
                 _("%s: size of %d bytes exceeds maximum revlog storage of 2GiB")
                 % (self.indexfile, len(text)))
 
         node = node or self.hash(text, p1, p2)
         if node in self.nodemap:
             return node
 
         dfh = None
         if not self._inline:
             dfh = self.opener(self.datafile, "a")
         ifh = self.opener(self.indexfile, "a+")
         try:
             return self._addrevision(node, text, transaction, link, p1, p2,
                                      REVIDX_DEFAULT_FLAGS, cachedelta, ifh, dfh)
         finally:
             if dfh:
                 dfh.close()
             ifh.close()
 
     def compress(self, text):
         """ generate a possibly-compressed representation of text """
         if not text:
             return ("", text)
         l = len(text)
         bin = None
         if l < 44:
             pass
         elif l > 1000000:
             # zlib makes an internal copy, thus doubling memory usage for
             # large files, so lets do this in pieces
             z = zlib.compressobj()
             p = []
             pos = 0
             while pos < l:
                 pos2 = pos + 2**20
                 p.append(z.compress(text[pos:pos2]))
                 pos = pos2
             p.append(z.flush())
             if sum(map(len, p)) < l:
                 bin = "".join(p)
         else:
             bin = _compress(text)
         if bin is None or len(bin) > l:
             if text[0] == '\0':
                 return ("", text)
             return ('u', text)
         return ("", bin)
 
     def _isgooddelta(self, d, textlen):
         """Returns True if the given delta is good. Good means that it is within
         the disk span, disk size, and chain length bounds that we know to be
         performant."""
         if d is None:
             return False
 
         # - 'dist' is the distance from the base revision -- bounding it limits
         #   the amount of I/O we need to do.
         # - 'compresseddeltalen' is the sum of the total size of deltas we need
         #   to apply -- bounding it limits the amount of CPU we consume.
         dist, l, data, base, chainbase, chainlen, compresseddeltalen = d
         if (dist > textlen * 4 or l > textlen or
             compresseddeltalen > textlen * 2 or
             (self._maxchainlen and chainlen > self._maxchainlen)):
             return False
 
         return True
 
     def _addrevision(self, node, text, transaction, link, p1, p2, flags,
                      cachedelta, ifh, dfh):
         """internal function to add revisions to the log
 
         see addrevision for argument descriptions.
         invariants:
         - text is optional (can be None); if not set, cachedelta must be set.
           if both are set, they must correspond to each other.
         """
         btext = [text]
         def buildtext():
             if btext[0] is not None:
                 return btext[0]
             # flush any pending writes here so we can read it in revision
             if dfh:
                 dfh.flush()
             ifh.flush()
             baserev = cachedelta[0]
             delta = cachedelta[1]
             # special case deltas which replace entire base; no need to decode
             # base revision. this neatly avoids censored bases, which throw when
             # they're decoded.
             hlen = struct.calcsize(">lll")
             if delta[:hlen] == mdiff.replacediffheader(self.rawsize(baserev),
                                                        len(delta) - hlen):
                 btext[0] = delta[hlen:]
             else:
                 basetext = self.revision(self.node(baserev))
                 btext[0] = mdiff.patch(basetext, delta)
             try:
                 self.checkhash(btext[0], p1, p2, node)
                 if flags & REVIDX_ISCENSORED:
                     raise RevlogError(_('node %s is not censored') % node)
             except CensoredNodeError:
                 # must pass the censored index flag to add censored revisions
                 if not flags & REVIDX_ISCENSORED:
                     raise
             return btext[0]
 
         def builddelta(rev):
             # can we use the cached delta?
             if cachedelta and cachedelta[0] == rev:
                 delta = cachedelta[1]
             else:
                 t = buildtext()
                 if self.iscensored(rev):
                     # deltas based on a censored revision must replace the
                     # full content in one patch, so delta works everywhere
                     header = mdiff.replacediffheader(self.rawsize(rev), len(t))
                     delta = header + t
                 else:
                     ptext = self.revision(self.node(rev))
                     delta = mdiff.textdiff(ptext, t)
             data = self.compress(delta)
             l = len(data[1]) + len(data[0])
             if basecache[0] == rev:
                 chainbase = basecache[1]
             else:
                 chainbase = self.chainbase(rev)
             dist = l + offset - self.start(chainbase)
             if self._generaldelta:
                 base = rev
             else:
                 base = chainbase
             chainlen, compresseddeltalen = self._chaininfo(rev)
             chainlen += 1
             compresseddeltalen += l
             return dist, l, data, base, chainbase, chainlen, compresseddeltalen
 
         curr = len(self)
         prev = curr - 1
         base = chainbase = curr
         chainlen = None
         offset = self.end(prev)
         d = None
         if self._basecache is None:
             self._basecache = (prev, self.chainbase(prev))
         basecache = self._basecache
         p1r, p2r = self.rev(p1), self.rev(p2)
 
         # full versions are inserted when the needed deltas
         # become comparable to the uncompressed text
         if text is None:
             textlen = mdiff.patchedsize(self.rawsize(cachedelta[0]),
                                         cachedelta[1])
         else:
             textlen = len(text)
 
         # should we try to build a delta?
         if prev != nullrev:
             if self._generaldelta:
+                if p2r != nullrev and self._aggressivemergedeltas:
+                    d = builddelta(p1r)
+                    d2 = builddelta(p2r)
+                    p1good = self._isgooddelta(d, textlen)
+                    p2good = self._isgooddelta(d2, textlen)
+                    if p1good and p2good:
+                        # If both are good deltas, choose the smallest
+                        if d2[1] < d[1]:
+                            d = d2
+                    elif p2good:
+                        # If only p2 is good, use it
+                        d = d2
+                    elif p1good:
+                        pass
+                    else:
+                        # Neither is good, try against prev to hopefully save us
+                        # a fulltext.
+                        d = builddelta(prev)
+                else:
                 # Pick whichever parent is closer to us (to minimize the
                 # chance of having to build a fulltext). Since
                 # nullrev == -1, any non-merge commit will always pick p1r.
                 drev = p2r if p2r > p1r else p1r
                 d = builddelta(drev)
                 # If the chosen delta will result in us making a full text,
                 # give it one last try against prev.
                 if drev != prev and not self._isgooddelta(d, textlen):
                     d = builddelta(prev)
             else:
                 d = builddelta(prev)
             dist, l, data, base, chainbase, chainlen, compresseddeltalen = d
 
         if not self._isgooddelta(d, textlen):
             text = buildtext()
             data = self.compress(text)
             l = len(data[1]) + len(data[0])
             base = chainbase = curr
 
         e = (offset_type(offset, flags), l, textlen,
              base, link, p1r, p2r, node)
         self.index.insert(-1, e)
         self.nodemap[node] = curr
 
         entry = self._io.packentry(e, self.node, self.version, curr)
         self._writeentry(transaction, ifh, dfh, entry, data, link, offset)
 
         if type(text) == str: # only accept immutable objects
             self._cache = (node, curr, text)
         self._basecache = (curr, chainbase)
         return node
 
     def _writeentry(self, transaction, ifh, dfh, entry, data, link, offset):
         curr = len(self) - 1
         if not self._inline:
             transaction.add(self.datafile, offset)
             transaction.add(self.indexfile, curr * len(entry))
             if data[0]:
                 dfh.write(data[0])
             dfh.write(data[1])
             dfh.flush()
             ifh.write(entry)
         else:
             offset += curr * self._io.size
             transaction.add(self.indexfile, offset, curr)
             ifh.write(entry)
             ifh.write(data[0])
             ifh.write(data[1])
             self.checkinlinesize(transaction, ifh)
 
     def addgroup(self, bundle, linkmapper, transaction, addrevisioncb=None):
         """
         add a delta group
 
         given a set of deltas, add them to the revision log. the
         first delta is against its parent, which should be in our
         log, the rest are against the previous delta.
 
         If ``addrevisioncb`` is defined, it will be called with arguments of
         this revlog and the node that was added.
         """
 
         # track the base of the current delta log
         content = []
         node = None
 
         r = len(self)
         end = 0
         if r:
             end = self.end(r - 1)
         ifh = self.opener(self.indexfile, "a+")
         isize = r * self._io.size
         if self._inline:
             transaction.add(self.indexfile, end + isize, r)
             dfh = None
         else:
             transaction.add(self.indexfile, isize, r)
             transaction.add(self.datafile, end)
             dfh = self.opener(self.datafile, "a")
         def flush():
             if dfh:
                 dfh.flush()
             ifh.flush()
         try:
             # loop through our set of deltas
             chain = None
             while True:
                 chunkdata = bundle.deltachunk(chain)
                 if not chunkdata:
                     break
                 node = chunkdata['node']
                 p1 = chunkdata['p1']
                 p2 = chunkdata['p2']
                 cs = chunkdata['cs']
                 deltabase = chunkdata['deltabase']
                 delta = chunkdata['delta']
 
                 content.append(node)
 
                 link = linkmapper(cs)
                 if node in self.nodemap:
                     # this can happen if two branches make the same change
                     chain = node
                     continue
 
                 for p in (p1, p2):
                     if p not in self.nodemap:
                         raise LookupError(p, self.indexfile,
                                           _('unknown parent'))
 
                 if deltabase not in self.nodemap:
                     raise LookupError(deltabase, self.indexfile,
                                       _('unknown delta base'))
 
                 baserev = self.rev(deltabase)
 
                 if baserev != nullrev and self.iscensored(baserev):
                     # if base is censored, delta must be full replacement in a
                     # single patch operation
                     hlen = struct.calcsize(">lll")
                     oldlen = self.rawsize(baserev)
                     newlen = len(delta) - hlen
                     if delta[:hlen] != mdiff.replacediffheader(oldlen, newlen):
                         raise error.CensoredBaseError(self.indexfile,
                                                       self.node(baserev))
 
                 flags = REVIDX_DEFAULT_FLAGS
                 if self._peek_iscensored(baserev, delta, flush):
                     flags |= REVIDX_ISCENSORED
 
                 chain = self._addrevision(node, None, transaction, link,
                                           p1, p2, flags, (baserev, delta),
                                           ifh, dfh)
 
                 if addrevisioncb:
                     # Data for added revision can't be read unless flushed
                     # because _loadchunk always opensa new file handle and
                     # there is no guarantee data was actually written yet.
                     flush()
                     addrevisioncb(self, chain)
 
                 if not dfh and not self._inline:
                     # addrevision switched from inline to conventional
                     # reopen the index
                     ifh.close()
                     dfh = self.opener(self.datafile, "a")
                     ifh = self.opener(self.indexfile, "a")
         finally:
             if dfh:
                 dfh.close()
             ifh.close()
 
         return content
 
     def iscensored(self, rev):
         """Check if a file revision is censored."""
         return False
 
     def _peek_iscensored(self, baserev, delta, flush):
         """Quickly check if a delta produces a censored revision."""
         return False
 
     def getstrippoint(self, minlink):
         """find the minimum rev that must be stripped to strip the linkrev
 
         Returns a tuple containing the minimum rev and a set of all revs that
         have linkrevs that will be broken by this strip.
         """
         brokenrevs = set()
         strippoint = len(self)
 
         heads = {}
         futurelargelinkrevs = set()
         for head in self.headrevs():
             headlinkrev = self.linkrev(head)
             heads[head] = headlinkrev
             if headlinkrev >= minlink:
                 futurelargelinkrevs.add(headlinkrev)
 
         # This algorithm involves walking down the rev graph, starting at the
         # heads. Since the revs are topologically sorted according to linkrev,
         # once all head linkrevs are below the minlink, we know there are
         # no more revs that could have a linkrev greater than minlink.
         # So we can stop walking.
         while futurelargelinkrevs:
             strippoint -= 1
             linkrev = heads.pop(strippoint)
 
             if linkrev < minlink:
                 brokenrevs.add(strippoint)
             else:
                 futurelargelinkrevs.remove(linkrev)
 
             for p in self.parentrevs(strippoint):
                 if p != nullrev:
                     plinkrev = self.linkrev(p)
                     heads[p] = plinkrev
                     if plinkrev >= minlink:
                         futurelargelinkrevs.add(plinkrev)
 
         return strippoint, brokenrevs
 
     def strip(self, minlink, transaction):
         """truncate the revlog on the first revision with a linkrev >= minlink
 
         This function is called when we're stripping revision minlink and
         its descendants from the repository.
 
         We have to remove all revisions with linkrev >= minlink, because
         the equivalent changelog revisions will be renumbered after the
         strip.
 
         So we truncate the revlog on the first of these revisions, and
         trust that the caller has saved the revisions that shouldn't be
         removed and that it'll re-add them after this truncation.
         """
         if len(self) == 0:
             return
 
         rev, _ = self.getstrippoint(minlink)
         if rev == len(self):
             return
 
         # first truncate the files on disk
         end = self.start(rev)
         if not self._inline:
             transaction.add(self.datafile, end)
             end = rev * self._io.size
         else:
             end += rev * self._io.size
 
         transaction.add(self.indexfile, end)
 
         # then reset internal state in memory to forget those revisions
         self._cache = None
         self._chaininfocache = {}
         self._chunkclear()
         for x in xrange(rev, len(self)):
             del self.nodemap[self.node(x)]
 
         del self.index[rev:-1]
 
     def checksize(self):
         expected = 0
         if len(self):
             expected = max(0, self.end(len(self) - 1))
 
         try:
             f = self.opener(self.datafile)
             f.seek(0, 2)
             actual = f.tell()
             f.close()
             dd = actual - expected
         except IOError as inst:
             if inst.errno != errno.ENOENT:
                 raise
             dd = 0
 
         try:
             f = self.opener(self.indexfile)
             f.seek(0, 2)
             actual = f.tell()
             f.close()
             s = self._io.size
             i = max(0, actual // s)
             di = actual - (i * s)
             if self._inline:
                 databytes = 0
                 for r in self:
                     databytes += max(0, self.length(r))
                 dd = 0
                 di = actual - len(self) * s - databytes
         except IOError as inst:
             if inst.errno != errno.ENOENT:
                 raise
             di = 0
 
         return (dd, di)
 
     def files(self):
         res = [self.indexfile]
         if not self._inline:
             res.append(self.datafile)
         return res

tests/test-generaldelta.t

@@ -1,71 +1,105
 Check whether size of generaldelta revlog is not bigger than its
 regular equivalent. Test would fail if generaldelta was naive
 implementation of parentdelta: third manifest revision would be fully
 inserted due to big distance from its paren revision (zero).
 
   $ hg init repo
   $ cd repo
   $ echo foo > foo
   $ echo bar > bar
   $ hg commit -q -Am boo
   $ hg clone --pull . ../gdrepo -q --config format.generaldelta=yes
   $ for r in 1 2 3; do
   >   echo $r > foo
   >   hg commit -q -m $r
   >   hg up -q -r 0
   >   hg pull . -q -r $r -R ../gdrepo
   > done
 
   $ cd ..
   >>> import os
   >>> regsize = os.stat("repo/.hg/store/00manifest.i").st_size
   >>> gdsize = os.stat("gdrepo/.hg/store/00manifest.i").st_size
   >>> if regsize < gdsize:
   ...     print 'generaldata increased size of manifest'
 
 Verify rev reordering doesnt create invalid bundles (issue4462)
 This requires a commit tree that when pulled will reorder manifest revs such
 that the second manifest to create a file rev will be ordered before the first
 manifest to create that file rev. We also need to do a partial pull to ensure
 reordering happens. At the end we verify the linkrev points at the earliest
 commit.
 
   $ hg init server --config format.generaldelta=True
   $ cd server
   $ touch a
   $ hg commit -Aqm a
   $ echo x > x
   $ echo y > y
   $ hg commit -Aqm xy
   $ hg up -q '.^'
   $ echo x > x
   $ echo z > z
   $ hg commit -Aqm xz
   $ hg up -q 1
   $ echo b > b
   $ hg commit -Aqm b
   $ hg merge -q 2
   $ hg commit -Aqm merge
   $ echo c > c
   $ hg commit -Aqm c
   $ hg log -G -T '{rev} {shortest(node)} {desc}'
   @  5 ebb8 c
   |
   o    4 baf7 merge
   |\
   | o  3 a129 b
   | |
   o |  2 958c xz
   | |
   | o  1 f00c xy
   |/
   o  0 3903 a
   
   $ cd ..
   $ hg init client
   $ cd client
   $ hg pull -q ../server -r 4
   $ hg debugindex x
      rev    offset  length   base linkrev nodeid       p1           p2
        0         0       3      0       1 1406e7411862 000000000000 000000000000
 
+  $ cd ..
+
+Test format.aggressivemergedeltas
+
+  $ hg init --config format.generaldelta=1 aggressive
+  $ cd aggressive
+  $ touch a b c d e
+  $ hg commit -Aqm side1
+  $ hg up -q null
+  $ touch x y
+  $ hg commit -Aqm side2
+
+- Verify non-aggressive merge uses p1 (commit 1) as delta parent
+  $ hg merge -q 0
+  $ hg commit -q -m merge
+  $ hg debugindex -m
+     rev    offset  length  delta linkrev nodeid       p1           p2
+       0         0      59     -1       0 8dde941edb6e 000000000000 000000000000
+       1        59      59     -1       1 315c023f341d 000000000000 000000000000
+       2       118      65      1       2 2ab389a983eb 315c023f341d 8dde941edb6e
+
+  $ hg strip -q -r . --config extensions.strip=
+
+- Verify aggressive merge uses p2 (commit 0) as delta parent
+  $ hg up -q -C 1
+  $ hg merge -q 0
+  $ hg commit -q -m merge --config format.aggressivemergedeltas=True
+  $ hg debugindex -m
+     rev    offset  length  delta linkrev nodeid       p1           p2
+       0         0      59     -1       0 8dde941edb6e 000000000000 000000000000
+       1        59      59     -1       1 315c023f341d 000000000000 000000000000
+       2       118      62      0       2 2ab389a983eb 315c023f341d 8dde941edb6e
+
+  $ cd ..