largefiles: fix inappropriate locking (issue3182)

Don't lock/write on operations that should be readonly (status).
Always lock when writing the lfdirstate (rollback).
Don't write lfdirstate until after committing; state isn't actually changed
until the commit is complete.

hgext/largefiles/lfutil.py

@@ -1,462 +1,450 b''
 # Copyright 2009-2010 Gregory P. Ward
 # Copyright 2009-2010 Intelerad Medical Systems Incorporated
 # Copyright 2010-2011 Fog Creek Software
 # Copyright 2010-2011 Unity Technologies
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 '''largefiles utility code: must not import other modules in this package.'''
 
 import os
 import errno
 import platform
 import shutil
 import stat
 import tempfile
 
 from mercurial import dirstate, httpconnection, match as match_, util, scmutil
 from mercurial.i18n import _
 
 shortname = '.hglf'
 longname = 'largefiles'
 
 
 # -- Portability wrappers ----------------------------------------------
 
 def dirstate_walk(dirstate, matcher, unknown=False, ignored=False):
     return dirstate.walk(matcher, [], unknown, ignored)
 
 def repo_add(repo, list):
     add = repo[None].add
     return add(list)
 
 def repo_remove(repo, list, unlink=False):
     def remove(list, unlink):
         wlock = repo.wlock()
         try:
             if unlink:
                 for f in list:
                     try:
                         util.unlinkpath(repo.wjoin(f))
                     except OSError, inst:
                         if inst.errno != errno.ENOENT:
                             raise
             repo[None].forget(list)
         finally:
             wlock.release()
     return remove(list, unlink=unlink)
 
 def repo_forget(repo, list):
     forget = repo[None].forget
     return forget(list)
 
 def findoutgoing(repo, remote, force):
     from mercurial import discovery
     common, _anyinc, _heads = discovery.findcommonincoming(repo,
         remote, force=force)
     return repo.changelog.findmissing(common)
 
 # -- Private worker functions ------------------------------------------
 
 def getminsize(ui, assumelfiles, opt, default=10):
     lfsize = opt
     if not lfsize and assumelfiles:
         lfsize = ui.config(longname, 'minsize', default=default)
     if lfsize:
         try:
             lfsize = float(lfsize)
         except ValueError:
             raise util.Abort(_('largefiles: size must be number (not %s)\n')
                              % lfsize)
     if lfsize is None:
         raise util.Abort(_('minimum size for largefiles must be specified'))
     return lfsize
 
 def link(src, dest):
     try:
         util.oslink(src, dest)
     except OSError:
         # if hardlinks fail, fallback on atomic copy
         dst = util.atomictempfile(dest)
         for chunk in util.filechunkiter(open(src, 'rb')):
             dst.write(chunk)
         dst.close()
         os.chmod(dest, os.stat(src).st_mode)
 
 def usercachepath(ui, hash):
     path = ui.configpath(longname, 'usercache', None)
     if path:
         path = os.path.join(path, hash)
     else:
         if os.name == 'nt':
             appdata = os.getenv('LOCALAPPDATA', os.getenv('APPDATA'))
             if appdata:
                 path = os.path.join(appdata, longname, hash)
         elif platform.system() == 'Darwin':
             home = os.getenv('HOME')
             if home:
                 path = os.path.join(home, 'Library', 'Caches',
                                     longname, hash)
         elif os.name == 'posix':
             path = os.getenv('XDG_CACHE_HOME')
             if path:
                 path = os.path.join(path, longname, hash)
             else:
                 home = os.getenv('HOME')
                 if home:
                     path = os.path.join(home, '.cache', longname, hash)
         else:
             raise util.Abort(_('unknown operating system: %s\n') % os.name)
     return path
 
 def inusercache(ui, hash):
     path = usercachepath(ui, hash)
     return path and os.path.exists(path)
 
 def findfile(repo, hash):
     if instore(repo, hash):
         repo.ui.note(_('Found %s in store\n') % hash)
     elif inusercache(repo.ui, hash):
         repo.ui.note(_('Found %s in system cache\n') % hash)
         path = storepath(repo, hash)
         util.makedirs(os.path.dirname(path))
         link(usercachepath(repo.ui, hash), path)
     else:
         return None
     return storepath(repo, hash)
 
 class largefiles_dirstate(dirstate.dirstate):
     def __getitem__(self, key):
         return super(largefiles_dirstate, self).__getitem__(unixpath(key))
     def normal(self, f):
         return super(largefiles_dirstate, self).normal(unixpath(f))
     def remove(self, f):
         return super(largefiles_dirstate, self).remove(unixpath(f))
     def add(self, f):
         return super(largefiles_dirstate, self).add(unixpath(f))
     def drop(self, f):
         return super(largefiles_dirstate, self).drop(unixpath(f))
     def forget(self, f):
         return super(largefiles_dirstate, self).forget(unixpath(f))
     def normallookup(self, f):
         return super(largefiles_dirstate, self).normallookup(unixpath(f))
 
 def openlfdirstate(ui, repo):
     '''
     Return a dirstate object that tracks largefiles: i.e. its root is
     the repo root, but it is saved in .hg/largefiles/dirstate.
     '''
     admin = repo.join(longname)
     opener = scmutil.opener(admin)
     lfdirstate = largefiles_dirstate(opener, ui, repo.root,
                                      repo.dirstate._validate)
 
     # If the largefiles dirstate does not exist, populate and create
     # it. This ensures that we create it on the first meaningful
-    # largefiles operation in a new clone. It also gives us an easy
-    # way to forcibly rebuild largefiles state:
-    #   rm .hg/largefiles/dirstate && hg status
-    # Or even, if things are really messed up:
-    #   rm -rf .hg/largefiles && hg status
+    # largefiles operation in a new clone.
     if not os.path.exists(os.path.join(admin, 'dirstate')):
         util.makedirs(admin)
         matcher = getstandinmatcher(repo)
         for standin in dirstate_walk(repo.dirstate, matcher):
             lfile = splitstandin(standin)
             hash = readstandin(repo, lfile)
             lfdirstate.normallookup(lfile)
             try:
                 if hash == hashfile(repo.wjoin(lfile)):
                     lfdirstate.normal(lfile)
             except OSError, err:
                 if err.errno != errno.ENOENT:
                     raise
-
-        lfdirstate.write()
-
     return lfdirstate
 
 def lfdirstate_status(lfdirstate, repo, rev):
-    wlock = repo.wlock()
-    try:
-        match = match_.always(repo.root, repo.getcwd())
-        s = lfdirstate.status(match, [], False, False, False)
-        unsure, modified, added, removed, missing, unknown, ignored, clean = s
-        for lfile in unsure:
-            if repo[rev][standin(lfile)].data().strip() != \
-                    hashfile(repo.wjoin(lfile)):
-                modified.append(lfile)
-            else:
-                clean.append(lfile)
-                lfdirstate.normal(lfile)
-        lfdirstate.write()
-    finally:
-        wlock.release()
+    match = match_.always(repo.root, repo.getcwd())
+    s = lfdirstate.status(match, [], False, False, False)
+    unsure, modified, added, removed, missing, unknown, ignored, clean = s
+    for lfile in unsure:
+        if repo[rev][standin(lfile)].data().strip() != \
+                hashfile(repo.wjoin(lfile)):
+            modified.append(lfile)
+        else:
+            clean.append(lfile)
+            lfdirstate.normal(lfile)
     return (modified, added, removed, missing, unknown, ignored, clean)
 
 def listlfiles(repo, rev=None, matcher=None):
     '''return a list of largefiles in the working copy or the
     specified changeset'''
 
     if matcher is None:
         matcher = getstandinmatcher(repo)
 
     # ignore unknown files in working directory
     return [splitstandin(f)
             for f in repo[rev].walk(matcher)
             if rev is not None or repo.dirstate[f] != '?']
 
 def instore(repo, hash):
     return os.path.exists(storepath(repo, hash))
 
 def storepath(repo, hash):
     return repo.join(os.path.join(longname, hash))
 
 def copyfromcache(repo, hash, filename):
     '''Copy the specified largefile from the repo or system cache to
     filename in the repository. Return true on success or false if the
     file was not found in either cache (which should not happened:
     this is meant to be called only after ensuring that the needed
     largefile exists in the cache).'''
     path = findfile(repo, hash)
     if path is None:
         return False
     util.makedirs(os.path.dirname(repo.wjoin(filename)))
     # The write may fail before the file is fully written, but we
     # don't use atomic writes in the working copy.
     shutil.copy(path, repo.wjoin(filename))
     return True
 
 def copytostore(repo, rev, file, uploaded=False):
     hash = readstandin(repo, file)
     if instore(repo, hash):
         return
     copytostoreabsolute(repo, repo.wjoin(file), hash)
 
 def copytostoreabsolute(repo, file, hash):
     util.makedirs(os.path.dirname(storepath(repo, hash)))
     if inusercache(repo.ui, hash):
         link(usercachepath(repo.ui, hash), storepath(repo, hash))
     else:
         dst = util.atomictempfile(storepath(repo, hash))
         for chunk in util.filechunkiter(open(file, 'rb')):
             dst.write(chunk)
         dst.close()
         util.copymode(file, storepath(repo, hash))
         linktousercache(repo, hash)
 
 def linktousercache(repo, hash):
     path = usercachepath(repo.ui, hash)
     if path:
         util.makedirs(os.path.dirname(path))
         link(storepath(repo, hash), path)
 
 def getstandinmatcher(repo, pats=[], opts={}):
     '''Return a match object that applies pats to the standin directory'''
     standindir = repo.pathto(shortname)
     if pats:
         # patterns supplied: search standin directory relative to current dir
         cwd = repo.getcwd()
         if os.path.isabs(cwd):
             # cwd is an absolute path for hg -R <reponame>
             # work relative to the repository root in this case
             cwd = ''
         pats = [os.path.join(standindir, cwd, pat) for pat in pats]
     elif os.path.isdir(standindir):
         # no patterns: relative to repo root
         pats = [standindir]
     else:
         # no patterns and no standin dir: return matcher that matches nothing
         match = match_.match(repo.root, None, [], exact=True)
         match.matchfn = lambda f: False
         return match
     return getmatcher(repo, pats, opts, showbad=False)
 
 def getmatcher(repo, pats=[], opts={}, showbad=True):
     '''Wrapper around scmutil.match() that adds showbad: if false,
     neuter the match object's bad() method so it does not print any
     warnings about missing files or directories.'''
     match = scmutil.match(repo[None], pats, opts)
 
     if not showbad:
         match.bad = lambda f, msg: None
     return match
 
 def composestandinmatcher(repo, rmatcher):
     '''Return a matcher that accepts standins corresponding to the
     files accepted by rmatcher. Pass the list of files in the matcher
     as the paths specified by the user.'''
     smatcher = getstandinmatcher(repo, rmatcher.files())
     isstandin = smatcher.matchfn
     def composed_matchfn(f):
         return isstandin(f) and rmatcher.matchfn(splitstandin(f))
     smatcher.matchfn = composed_matchfn
 
     return smatcher
 
 def standin(filename):
     '''Return the repo-relative path to the standin for the specified big
     file.'''
     # Notes:
     # 1) Most callers want an absolute path, but _create_standin() needs
     #    it repo-relative so lfadd() can pass it to repo_add().  So leave
     #    it up to the caller to use repo.wjoin() to get an absolute path.
     # 2) Join with '/' because that's what dirstate always uses, even on
     #    Windows. Change existing separator to '/' first in case we are
     #    passed filenames from an external source (like the command line).
     return shortname + '/' + filename.replace(os.sep, '/')
 
 def isstandin(filename):
     '''Return true if filename is a big file standin. filename must be
     in Mercurial's internal form (slash-separated).'''
     return filename.startswith(shortname + '/')
 
 def splitstandin(filename):
     # Split on / because that's what dirstate always uses, even on Windows.
     # Change local separator to / first just in case we are passed filenames
     # from an external source (like the command line).
     bits = filename.replace(os.sep, '/').split('/', 1)
     if len(bits) == 2 and bits[0] == shortname:
         return bits[1]
     else:
         return None
 
 def updatestandin(repo, standin):
     file = repo.wjoin(splitstandin(standin))
     if os.path.exists(file):
         hash = hashfile(file)
         executable = getexecutable(file)
         writestandin(repo, standin, hash, executable)
 
 def readstandin(repo, filename, node=None):
     '''read hex hash from standin for filename at given node, or working
     directory if no node is given'''
     return repo[node][standin(filename)].data().strip()
 
 def writestandin(repo, standin, hash, executable):
     '''write hash to <repo.root>/<standin>'''
     writehash(hash, repo.wjoin(standin), executable)
 
 def copyandhash(instream, outfile):
     '''Read bytes from instream (iterable) and write them to outfile,
     computing the SHA-1 hash of the data along the way.  Close outfile
     when done and return the binary hash.'''
     hasher = util.sha1('')
     for data in instream:
         hasher.update(data)
         outfile.write(data)
 
     # Blecch: closing a file that somebody else opened is rude and
     # wrong. But it's so darn convenient and practical! After all,
     # outfile was opened just to copy and hash.
     outfile.close()
 
     return hasher.digest()
 
 def hashrepofile(repo, file):
     return hashfile(repo.wjoin(file))
 
 def hashfile(file):
     if not os.path.exists(file):
         return ''
     hasher = util.sha1('')
     fd = open(file, 'rb')
     for data in blockstream(fd):
         hasher.update(data)
     fd.close()
     return hasher.hexdigest()
 
 class limitreader(object):
     def __init__(self, f, limit):
         self.f = f
         self.limit = limit
 
     def read(self, length):
         if self.limit == 0:
             return ''
         length = length > self.limit and self.limit or length
         self.limit -= length
         return self.f.read(length)
 
     def close(self):
         pass
 
 def blockstream(infile, blocksize=128 * 1024):
     """Generator that yields blocks of data from infile and closes infile."""
     while True:
         data = infile.read(blocksize)
         if not data:
             break
         yield data
     # same blecch as copyandhash() above
     infile.close()
 
 def readhash(filename):
     rfile = open(filename, 'rb')
     hash = rfile.read(40)
     rfile.close()
     if len(hash) < 40:
         raise util.Abort(_('bad hash in \'%s\' (only %d bytes long)')
                          % (filename, len(hash)))
     return hash
 
 def writehash(hash, filename, executable):
     util.makedirs(os.path.dirname(filename))
     util.writefile(filename, hash + '\n')
     os.chmod(filename, getmode(executable))
 
 def getexecutable(filename):
     mode = os.stat(filename).st_mode
     return ((mode & stat.S_IXUSR) and
             (mode & stat.S_IXGRP) and
             (mode & stat.S_IXOTH))
 
 def getmode(executable):
     if executable:
         return 0755
     else:
         return 0644
 
 def urljoin(first, second, *arg):
     def join(left, right):
         if not left.endswith('/'):
             left += '/'
         if right.startswith('/'):
             right = right[1:]
         return left + right
 
     url = join(first, second)
     for a in arg:
         url = join(url, a)
     return url
 
 def hexsha1(data):
     """hexsha1 returns the hex-encoded sha1 sum of the data in the file-like
     object data"""
     h = util.sha1()
     for chunk in util.filechunkiter(data):
         h.update(chunk)
     return h.hexdigest()
 
 def httpsendfile(ui, filename):
     return httpconnection.httpsendfile(ui, filename, 'rb')
 
 def unixpath(path):
     '''Return a version of path normalized for use with the lfdirstate.'''
     return os.path.normpath(path).replace(os.sep, '/')
 
 def islfilesrepo(repo):
     return ('largefiles' in repo.requirements and
             util.any(shortname + '/' in f[0] for f in repo.store.datafiles()))
 
 def mkstemp(repo, prefix):
     '''Returns a file descriptor and a filename corresponding to a temporary
     file in the repo's largefiles store.'''
     path = repo.join(longname)
     util.makedirs(path)
     return tempfile.mkstemp(prefix=prefix, dir=path)
 
 class storeprotonotcapable(Exception):
     def __init__(self, storetypes):
         self.storetypes = storetypes

hgext/largefiles/overrides.py

@@ -1,927 +1,931 b''
 # Copyright 2009-2010 Gregory P. Ward
 # Copyright 2009-2010 Intelerad Medical Systems Incorporated
 # Copyright 2010-2011 Fog Creek Software
 # Copyright 2010-2011 Unity Technologies
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 '''Overridden Mercurial commands and functions for the largefiles extension'''
 
 import os
 import copy
 
 from mercurial import hg, commands, util, cmdutil, scmutil, match as match_, \
     node, archival, error, merge
 from mercurial.i18n import _
 from mercurial.node import hex
 from hgext import rebase
 
 import lfutil
 import lfcommands
 
 # -- Utility functions: commonly/repeatedly needed functionality ---------------
 
 def installnormalfilesmatchfn(manifest):
     '''overrides scmutil.match so that the matcher it returns will ignore all
     largefiles'''
     oldmatch = None # for the closure
     def override_match(ctx, pats=[], opts={}, globbed=False,
             default='relpath'):
         match = oldmatch(ctx, pats, opts, globbed, default)
         m = copy.copy(match)
         notlfile = lambda f: not (lfutil.isstandin(f) or lfutil.standin(f) in
                 manifest)
         m._files = filter(notlfile, m._files)
         m._fmap = set(m._files)
         orig_matchfn = m.matchfn
         m.matchfn = lambda f: notlfile(f) and orig_matchfn(f) or None
         return m
     oldmatch = installmatchfn(override_match)
 
 def installmatchfn(f):
     oldmatch = scmutil.match
     setattr(f, 'oldmatch', oldmatch)
     scmutil.match = f
     return oldmatch
 
 def restorematchfn():
     '''restores scmutil.match to what it was before installnormalfilesmatchfn
     was called.  no-op if scmutil.match is its original function.
 
     Note that n calls to installnormalfilesmatchfn will require n calls to
     restore matchfn to reverse'''
     scmutil.match = getattr(scmutil.match, 'oldmatch', scmutil.match)
 
 def add_largefiles(ui, repo, *pats, **opts):
     large = opts.pop('large', None)
     lfsize = lfutil.getminsize(
         ui, lfutil.islfilesrepo(repo), opts.pop('lfsize', None))
 
     lfmatcher = None
     if lfutil.islfilesrepo(repo):
         lfpats = ui.configlist(lfutil.longname, 'patterns', default=[])
         if lfpats:
             lfmatcher = match_.match(repo.root, '', list(lfpats))
 
     lfnames = []
     m = scmutil.match(repo[None], pats, opts)
     m.bad = lambda x, y: None
     wctx = repo[None]
     for f in repo.walk(m):
         exact = m.exact(f)
         lfile = lfutil.standin(f) in wctx
         nfile = f in wctx
         exists = lfile or nfile
 
         # Don't warn the user when they attempt to add a normal tracked file.
         # The normal add code will do that for us.
         if exact and exists:
             if lfile:
                 ui.warn(_('%s already a largefile\n') % f)
             continue
 
         if exact or not exists:
             abovemin = (lfsize and
                         os.lstat(repo.wjoin(f)).st_size >= lfsize * 1024 * 1024)
             if large or abovemin or (lfmatcher and lfmatcher(f)):
                 lfnames.append(f)
                 if ui.verbose or not exact:
                     ui.status(_('adding %s as a largefile\n') % m.rel(f))
 
     bad = []
     standins = []
 
     # Need to lock, otherwise there could be a race condition between
     # when standins are created and added to the repo.
     wlock = repo.wlock()
     try:
         if not opts.get('dry_run'):
             lfdirstate = lfutil.openlfdirstate(ui, repo)
             for f in lfnames:
                 standinname = lfutil.standin(f)
                 lfutil.writestandin(repo, standinname, hash='',
                     executable=lfutil.getexecutable(repo.wjoin(f)))
                 standins.append(standinname)
                 if lfdirstate[f] == 'r':
                     lfdirstate.normallookup(f)
                 else:
                     lfdirstate.add(f)
             lfdirstate.write()
             bad += [lfutil.splitstandin(f)
                     for f in lfutil.repo_add(repo, standins)
                     if f in m.files()]
     finally:
         wlock.release()
     return bad
 
 def remove_largefiles(ui, repo, *pats, **opts):
     after = opts.get('after')
     if not pats and not after:
         raise util.Abort(_('no files specified'))
     m = scmutil.match(repo[None], pats, opts)
     try:
         repo.lfstatus = True
         s = repo.status(match=m, clean=True)
     finally:
         repo.lfstatus = False
     manifest = repo[None].manifest()
     modified, added, deleted, clean = [[f for f in list
                                         if lfutil.standin(f) in manifest]
                                        for list in [s[0], s[1], s[3], s[6]]]
 
     def warn(files, reason):
         for f in files:
             ui.warn(_('not removing %s: %s (use forget to undo)\n')
                     % (m.rel(f), reason))
 
     if after:
         remove, forget = deleted, []
         warn(modified + added + clean, _('file still exists'))
     else:
         remove, forget = deleted + clean, []
         warn(modified, _('file is modified'))
         warn(added, _('file has been marked for add'))
 
     for f in sorted(remove + forget):
         if ui.verbose or not m.exact(f):
             ui.status(_('removing %s\n') % m.rel(f))
 
     # Need to lock because standin files are deleted then removed from the
     # repository and we could race inbetween.
     wlock = repo.wlock()
     try:
         lfdirstate = lfutil.openlfdirstate(ui, repo)
         for f in remove:
             if not after:
                 # If this is being called by addremove, notify the user that we
                 # are removing the file.
                 if getattr(repo, "_isaddremove", False):
                     ui.status(_('removing %s\n' % f))
                 if os.path.exists(repo.wjoin(f)):
                     os.unlink(repo.wjoin(f))
                 currentdir = os.path.split(f)[0]
                 while currentdir and not os.listdir(repo.wjoin(currentdir)):
                     os.rmdir(repo.wjoin(currentdir))
                     currentdir = os.path.split(currentdir)[0]
             lfdirstate.remove(f)
         lfdirstate.write()
         forget = [lfutil.standin(f) for f in forget]
         remove = [lfutil.standin(f) for f in remove]
         lfutil.repo_forget(repo, forget)
         # If this is being called by addremove, let the original addremove
         # function handle this.
         if not getattr(repo, "_isaddremove", False):
             lfutil.repo_remove(repo, remove, unlink=True)
     finally:
         wlock.release()
 
 # -- Wrappers: modify existing commands --------------------------------
 
 # Add works by going through the files that the user wanted to add and
 # checking if they should be added as largefiles. Then it makes a new
 # matcher which matches only the normal files and runs the original
 # version of add.
 def override_add(orig, ui, repo, *pats, **opts):
     bad = add_largefiles(ui, repo, *pats, **opts)
     installnormalfilesmatchfn(repo[None].manifest())
     result = orig(ui, repo, *pats, **opts)
     restorematchfn()
 
     return (result == 1 or bad) and 1 or 0
 
 def override_remove(orig, ui, repo, *pats, **opts):
     installnormalfilesmatchfn(repo[None].manifest())
     orig(ui, repo, *pats, **opts)
     restorematchfn()
     remove_largefiles(ui, repo, *pats, **opts)
 
 def override_status(orig, ui, repo, *pats, **opts):
     try:
         repo.lfstatus = True
         return orig(ui, repo, *pats, **opts)
     finally:
         repo.lfstatus = False
 
 def override_log(orig, ui, repo, *pats, **opts):
     try:
         repo.lfstatus = True
         orig(ui, repo, *pats, **opts)
     finally:
         repo.lfstatus = False
 
 def override_verify(orig, ui, repo, *pats, **opts):
     large = opts.pop('large', False)
     all = opts.pop('lfa', False)
     contents = opts.pop('lfc', False)
 
     result = orig(ui, repo, *pats, **opts)
     if large:
         result = result or lfcommands.verifylfiles(ui, repo, all, contents)
     return result
 
 # Override needs to refresh standins so that update's normal merge
 # will go through properly. Then the other update hook (overriding repo.update)
 # will get the new files. Filemerge is also overriden so that the merge
 # will merge standins correctly.
 def override_update(orig, ui, repo, *pats, **opts):
     lfdirstate = lfutil.openlfdirstate(ui, repo)
     s = lfdirstate.status(match_.always(repo.root, repo.getcwd()), [], False,
         False, False)
     (unsure, modified, added, removed, missing, unknown, ignored, clean) = s
 
     # Need to lock between the standins getting updated and their
     # largefiles getting updated
     wlock = repo.wlock()
     try:
         if opts['check']:
             mod = len(modified) > 0
             for lfile in unsure:
                 standin = lfutil.standin(lfile)
                 if repo['.'][standin].data().strip() != \
                         lfutil.hashfile(repo.wjoin(lfile)):
                     mod = True
                 else:
                     lfdirstate.normal(lfile)
             lfdirstate.write()
             if mod:
                 raise util.Abort(_('uncommitted local changes'))
         # XXX handle removed differently
         if not opts['clean']:
             for lfile in unsure + modified + added:
                 lfutil.updatestandin(repo, lfutil.standin(lfile))
     finally:
         wlock.release()
     return orig(ui, repo, *pats, **opts)
 
 # Before starting the manifest merge, merge.updates will call
 # _checkunknown to check if there are any files in the merged-in
 # changeset that collide with unknown files in the working copy.
 #
 # The largefiles are seen as unknown, so this prevents us from merging
 # in a file 'foo' if we already have a largefile with the same name.
 #
 # The overridden function filters the unknown files by removing any
 # largefiles. This makes the merge proceed and we can then handle this
 # case further in the overridden manifestmerge function below.
 def override_checkunknown(origfn, wctx, mctx, folding):
     origunknown = wctx.unknown()
     wctx._unknown = filter(lambda f: lfutil.standin(f) not in wctx, origunknown)
     try:
         return origfn(wctx, mctx, folding)
     finally:
         wctx._unknown = origunknown
 
 # The manifest merge handles conflicts on the manifest level. We want
 # to handle changes in largefile-ness of files at this level too.
 #
 # The strategy is to run the original manifestmerge and then process
 # the action list it outputs. There are two cases we need to deal with:
 #
 # 1. Normal file in p1, largefile in p2. Here the largefile is
 #    detected via its standin file, which will enter the working copy
 #    with a "get" action. It is not "merge" since the standin is all
 #    Mercurial is concerned with at this level -- the link to the
 #    existing normal file is not relevant here.
 #
 # 2. Largefile in p1, normal file in p2. Here we get a "merge" action
 #    since the largefile will be present in the working copy and
 #    different from the normal file in p2. Mercurial therefore
 #    triggers a merge action.
 #
 # In both cases, we prompt the user and emit new actions to either
 # remove the standin (if the normal file was kept) or to remove the
 # normal file and get the standin (if the largefile was kept). The
 # default prompt answer is to use the largefile version since it was
 # presumably changed on purpose.
 #
 # Finally, the merge.applyupdates function will then take care of
 # writing the files into the working copy and lfcommands.updatelfiles
 # will update the largefiles.
 def override_manifestmerge(origfn, repo, p1, p2, pa, overwrite, partial):
     actions = origfn(repo, p1, p2, pa, overwrite, partial)
     processed = []
 
     for action in actions:
         if overwrite:
             processed.append(action)
             continue
         f, m = action[:2]
 
         choices = (_('&Largefile'), _('&Normal file'))
         if m == "g" and lfutil.splitstandin(f) in p1 and f in p2:
             # Case 1: normal file in the working copy, largefile in
             # the second parent
             lfile = lfutil.splitstandin(f)
             standin = f
             msg = _('%s has been turned into a largefile\n'
                     'use (l)argefile or keep as (n)ormal file?') % lfile
             if repo.ui.promptchoice(msg, choices, 0) == 0:
                 processed.append((lfile, "r"))
                 processed.append((standin, "g", p2.flags(standin)))
             else:
                 processed.append((standin, "r"))
         elif m == "m" and lfutil.standin(f) in p1 and f in p2:
             # Case 2: largefile in the working copy, normal file in
             # the second parent
             standin = lfutil.standin(f)
             lfile = f
             msg = _('%s has been turned into a normal file\n'
                     'keep as (l)argefile or use (n)ormal file?') % lfile
             if repo.ui.promptchoice(msg, choices, 0) == 0:
                 processed.append((lfile, "r"))
             else:
                 processed.append((standin, "r"))
                 processed.append((lfile, "g", p2.flags(lfile)))
         else:
             processed.append(action)
 
     return processed
 
 # Override filemerge to prompt the user about how they wish to merge
 # largefiles. This will handle identical edits, and copy/rename +
 # edit without prompting the user.
 def override_filemerge(origfn, repo, mynode, orig, fcd, fco, fca):
     # Use better variable names here. Because this is a wrapper we cannot
     # change the variable names in the function declaration.
     fcdest, fcother, fcancestor = fcd, fco, fca
     if not lfutil.isstandin(orig):
         return origfn(repo, mynode, orig, fcdest, fcother, fcancestor)
     else:
         if not fcother.cmp(fcdest): # files identical?
             return None
 
         # backwards, use working dir parent as ancestor
         if fcancestor == fcother:
             fcancestor = fcdest.parents()[0]
 
         if orig != fcother.path():
             repo.ui.status(_('merging %s and %s to %s\n')
                            % (lfutil.splitstandin(orig),
                               lfutil.splitstandin(fcother.path()),
                               lfutil.splitstandin(fcdest.path())))
         else:
             repo.ui.status(_('merging %s\n')
                            % lfutil.splitstandin(fcdest.path()))
 
         if fcancestor.path() != fcother.path() and fcother.data() == \
                 fcancestor.data():
             return 0
         if fcancestor.path() != fcdest.path() and fcdest.data() == \
                 fcancestor.data():
             repo.wwrite(fcdest.path(), fcother.data(), fcother.flags())
             return 0
 
         if repo.ui.promptchoice(_('largefile %s has a merge conflict\n'
                              'keep (l)ocal or take (o)ther?') %
                              lfutil.splitstandin(orig),
                              (_('&Local'), _('&Other')), 0) == 0:
             return 0
         else:
             repo.wwrite(fcdest.path(), fcother.data(), fcother.flags())
             return 0
 
 # Copy first changes the matchers to match standins instead of
 # largefiles.  Then it overrides util.copyfile in that function it
 # checks if the destination largefile already exists. It also keeps a
 # list of copied files so that the largefiles can be copied and the
 # dirstate updated.
 def override_copy(orig, ui, repo, pats, opts, rename=False):
     # doesn't remove largefile on rename
     if len(pats) < 2:
         # this isn't legal, let the original function deal with it
         return orig(ui, repo, pats, opts, rename)
 
     def makestandin(relpath):
         path = scmutil.canonpath(repo.root, repo.getcwd(), relpath)
         return os.path.join(repo.wjoin(lfutil.standin(path)))
 
     fullpats = scmutil.expandpats(pats)
     dest = fullpats[-1]
 
     if os.path.isdir(dest):
         if not os.path.isdir(makestandin(dest)):
             os.makedirs(makestandin(dest))
     # This could copy both lfiles and normal files in one command,
     # but we don't want to do that. First replace their matcher to
     # only match normal files and run it, then replace it to just
     # match largefiles and run it again.
     nonormalfiles = False
     nolfiles = False
     try:
         try:
             installnormalfilesmatchfn(repo[None].manifest())
             result = orig(ui, repo, pats, opts, rename)
         except util.Abort, e:
             if str(e) != 'no files to copy':
                 raise e
             else:
                 nonormalfiles = True
             result = 0
     finally:
         restorematchfn()
 
     # The first rename can cause our current working directory to be removed.
     # In that case there is nothing left to copy/rename so just quit.
     try:
         repo.getcwd()
     except OSError:
         return result
 
     try:
         try:
             # When we call orig below it creates the standins but we don't add them
             # to the dir state until later so lock during that time.
             wlock = repo.wlock()
 
             manifest = repo[None].manifest()
             oldmatch = None # for the closure
             def override_match(ctx, pats=[], opts={}, globbed=False,
                     default='relpath'):
                 newpats = []
                 # The patterns were previously mangled to add the standin
                 # directory; we need to remove that now
                 for pat in pats:
                     if match_.patkind(pat) is None and lfutil.shortname in pat:
                         newpats.append(pat.replace(lfutil.shortname, ''))
                     else:
                         newpats.append(pat)
                 match = oldmatch(ctx, newpats, opts, globbed, default)
                 m = copy.copy(match)
                 lfile = lambda f: lfutil.standin(f) in manifest
                 m._files = [lfutil.standin(f) for f in m._files if lfile(f)]
                 m._fmap = set(m._files)
                 orig_matchfn = m.matchfn
                 m.matchfn = lambda f: (lfutil.isstandin(f) and
                                     lfile(lfutil.splitstandin(f)) and
                                     orig_matchfn(lfutil.splitstandin(f)) or
                                     None)
                 return m
             oldmatch = installmatchfn(override_match)
             listpats = []
             for pat in pats:
                 if match_.patkind(pat) is not None:
                     listpats.append(pat)
                 else:
                     listpats.append(makestandin(pat))
 
             try:
                 origcopyfile = util.copyfile
                 copiedfiles = []
                 def override_copyfile(src, dest):
                     if (lfutil.shortname in src and
                         dest.startswith(repo.wjoin(lfutil.shortname))):
                         destlfile = dest.replace(lfutil.shortname, '')
                         if not opts['force'] and os.path.exists(destlfile):
                             raise IOError('',
                                 _('destination largefile already exists'))
                     copiedfiles.append((src, dest))
                     origcopyfile(src, dest)
 
                 util.copyfile = override_copyfile
                 result += orig(ui, repo, listpats, opts, rename)
             finally:
                 util.copyfile = origcopyfile
 
             lfdirstate = lfutil.openlfdirstate(ui, repo)
             for (src, dest) in copiedfiles:
                 if (lfutil.shortname in src and
                     dest.startswith(repo.wjoin(lfutil.shortname))):
                     srclfile = src.replace(repo.wjoin(lfutil.standin('')), '')
                     destlfile = dest.replace(repo.wjoin(lfutil.standin('')), '')
                     destlfiledir = os.path.dirname(destlfile) or '.'
                     if not os.path.isdir(destlfiledir):
                         os.makedirs(destlfiledir)
                     if rename:
                         os.rename(repo.wjoin(srclfile), repo.wjoin(destlfile))
                         lfdirstate.remove(srclfile)
                     else:
                         util.copyfile(srclfile, destlfile)
                     lfdirstate.add(destlfile)
             lfdirstate.write()
         except util.Abort, e:
             if str(e) != 'no files to copy':
                 raise e
             else:
                 nolfiles = True
     finally:
         restorematchfn()
         wlock.release()
 
     if nolfiles and nonormalfiles:
         raise util.Abort(_('no files to copy'))
 
     return result
 
 # When the user calls revert, we have to be careful to not revert any
 # changes to other largefiles accidentally. This means we have to keep
 # track of the largefiles that are being reverted so we only pull down
 # the necessary largefiles.
 #
 # Standins are only updated (to match the hash of largefiles) before
 # commits. Update the standins then run the original revert, changing
 # the matcher to hit standins instead of largefiles. Based on the
 # resulting standins update the largefiles. Then return the standins
 # to their proper state
 def override_revert(orig, ui, repo, *pats, **opts):
     # Because we put the standins in a bad state (by updating them)
     # and then return them to a correct state we need to lock to
     # prevent others from changing them in their incorrect state.
     wlock = repo.wlock()
     try:
         lfdirstate = lfutil.openlfdirstate(ui, repo)
         (modified, added, removed, missing, unknown, ignored, clean) = \
             lfutil.lfdirstate_status(lfdirstate, repo, repo['.'].rev())
         for lfile in modified:
             lfutil.updatestandin(repo, lfutil.standin(lfile))
 
         try:
             ctx = repo[opts.get('rev')]
             oldmatch = None # for the closure
             def override_match(ctx, pats=[], opts={}, globbed=False,
                     default='relpath'):
                 match = oldmatch(ctx, pats, opts, globbed, default)
                 m = copy.copy(match)
                 def tostandin(f):
                     if lfutil.standin(f) in ctx or lfutil.standin(f) in ctx:
                         return lfutil.standin(f)
                     elif lfutil.standin(f) in repo[None]:
                         return None
                     return f
                 m._files = [tostandin(f) for f in m._files]
                 m._files = [f for f in m._files if f is not None]
                 m._fmap = set(m._files)
                 orig_matchfn = m.matchfn
                 def matchfn(f):
                     if lfutil.isstandin(f):
                         # We need to keep track of what largefiles are being
                         # matched so we know which ones to update later --
                         # otherwise we accidentally revert changes to other
                         # largefiles. This is repo-specific, so duckpunch the
                         # repo object to keep the list of largefiles for us
                         # later.
                         if orig_matchfn(lfutil.splitstandin(f)) and \
                                 (f in repo[None] or f in ctx):
                             lfileslist = getattr(repo, '_lfilestoupdate', [])
                             lfileslist.append(lfutil.splitstandin(f))
                             repo._lfilestoupdate = lfileslist
                             return True
                         else:
                             return False
                     return orig_matchfn(f)
                 m.matchfn = matchfn
                 return m
             oldmatch = installmatchfn(override_match)
             scmutil.match
             matches = override_match(repo[None], pats, opts)
             orig(ui, repo, *pats, **opts)
         finally:
             restorematchfn()
         lfileslist = getattr(repo, '_lfilestoupdate', [])
         lfcommands.updatelfiles(ui, repo, filelist=lfileslist,
                                 printmessage=False)
 
         # empty out the largefiles list so we start fresh next time
         repo._lfilestoupdate = []
         for lfile in modified:
             if lfile in lfileslist:
                 if os.path.exists(repo.wjoin(lfutil.standin(lfile))) and lfile\
                         in repo['.']:
                     lfutil.writestandin(repo, lfutil.standin(lfile),
                         repo['.'][lfile].data().strip(),
                         'x' in repo['.'][lfile].flags())
         lfdirstate = lfutil.openlfdirstate(ui, repo)
         for lfile in added:
             standin = lfutil.standin(lfile)
             if standin not in ctx and (standin in matches or opts.get('all')):
                 if lfile in lfdirstate:
                     lfdirstate.drop(lfile)
                 util.unlinkpath(repo.wjoin(standin))
         lfdirstate.write()
     finally:
         wlock.release()
 
 def hg_update(orig, repo, node):
     result = orig(repo, node)
     lfcommands.updatelfiles(repo.ui, repo)
     return result
 
 def hg_clean(orig, repo, node, show_stats=True):
     result = orig(repo, node, show_stats)
     lfcommands.updatelfiles(repo.ui, repo)
     return result
 
 def hg_merge(orig, repo, node, force=None, remind=True):
     result = orig(repo, node, force, remind)
     lfcommands.updatelfiles(repo.ui, repo)
     return result
 
 # When we rebase a repository with remotely changed largefiles, we need to
 # take some extra care so that the largefiles are correctly updated in the
 # working copy
 def override_pull(orig, ui, repo, source=None, **opts):
     if opts.get('rebase', False):
         repo._isrebasing = True
         try:
             if opts.get('update'):
                  del opts['update']
                  ui.debug('--update and --rebase are not compatible, ignoring '
                           'the update flag\n')
             del opts['rebase']
             cmdutil.bailifchanged(repo)
             revsprepull = len(repo)
             origpostincoming = commands.postincoming
             def _dummy(*args, **kwargs):
                 pass
             commands.postincoming = _dummy
             repo.lfpullsource = source
             if not source:
                 source = 'default'
             try:
                 result = commands.pull(ui, repo, source, **opts)
             finally:
                 commands.postincoming = origpostincoming
             revspostpull = len(repo)
             if revspostpull > revsprepull:
                 result = result or rebase.rebase(ui, repo)
         finally:
             repo._isrebasing = False
     else:
         repo.lfpullsource = source
         if not source:
             source = 'default'
         result = orig(ui, repo, source, **opts)
     return result
 
 def override_rebase(orig, ui, repo, **opts):
     repo._isrebasing = True
     try:
         orig(ui, repo, **opts)
     finally:
         repo._isrebasing = False
 
 def override_archive(orig, repo, dest, node, kind, decode=True, matchfn=None,
             prefix=None, mtime=None, subrepos=None):
     # No need to lock because we are only reading history and
     # largefile caches, neither of which are modified.
     lfcommands.cachelfiles(repo.ui, repo, node)
 
     if kind not in archival.archivers:
         raise util.Abort(_("unknown archive type '%s'") % kind)
 
     ctx = repo[node]
 
     if kind == 'files':
         if prefix:
             raise util.Abort(
                 _('cannot give prefix when archiving to files'))
     else:
         prefix = archival.tidyprefix(dest, kind, prefix)
 
     def write(name, mode, islink, getdata):
         if matchfn and not matchfn(name):
             return
         data = getdata()
         if decode:
             data = repo.wwritedata(name, data)
         archiver.addfile(prefix + name, mode, islink, data)
 
     archiver = archival.archivers[kind](dest, mtime or ctx.date()[0])
 
     if repo.ui.configbool("ui", "archivemeta", True):
         def metadata():
             base = 'repo: %s\nnode: %s\nbranch: %s\n' % (
                 hex(repo.changelog.node(0)), hex(node), ctx.branch())
 
             tags = ''.join('tag: %s\n' % t for t in ctx.tags()
                            if repo.tagtype(t) == 'global')
             if not tags:
                 repo.ui.pushbuffer()
                 opts = {'template': '{latesttag}\n{latesttagdistance}',
                         'style': '', 'patch': None, 'git': None}
                 cmdutil.show_changeset(repo.ui, repo, opts).show(ctx)
                 ltags, dist = repo.ui.popbuffer().split('\n')
                 tags = ''.join('latesttag: %s\n' % t for t in ltags.split(':'))
                 tags += 'latesttagdistance: %s\n' % dist
 
             return base + tags
 
         write('.hg_archival.txt', 0644, False, metadata)
 
     for f in ctx:
         ff = ctx.flags(f)
         getdata = ctx[f].data
         if lfutil.isstandin(f):
             path = lfutil.findfile(repo, getdata().strip())
             f = lfutil.splitstandin(f)
 
             def getdatafn():
                 fd = None
                 try:
                     fd = open(path, 'rb')
                     return fd.read()
                 finally:
                     if fd:
                         fd.close()
 
             getdata = getdatafn
         write(f, 'x' in ff and 0755 or 0644, 'l' in ff, getdata)
 
     if subrepos:
         for subpath in ctx.substate:
             sub = ctx.sub(subpath)
             sub.archive(repo.ui, archiver, prefix)
 
     archiver.done()
 
 # If a largefile is modified, the change is not reflected in its
 # standin until a commit. cmdutil.bailifchanged() raises an exception
 # if the repo has uncommitted changes. Wrap it to also check if
 # largefiles were changed. This is used by bisect and backout.
 def override_bailifchanged(orig, repo):
     orig(repo)
     repo.lfstatus = True
     modified, added, removed, deleted = repo.status()[:4]
     repo.lfstatus = False
     if modified or added or removed or deleted:
         raise util.Abort(_('outstanding uncommitted changes'))
 
 # Fetch doesn't use cmdutil.bail_if_changed so override it to add the check
 def override_fetch(orig, ui, repo, *pats, **opts):
     repo.lfstatus = True
     modified, added, removed, deleted = repo.status()[:4]
     repo.lfstatus = False
     if modified or added or removed or deleted:
         raise util.Abort(_('outstanding uncommitted changes'))
     return orig(ui, repo, *pats, **opts)
 
 def override_forget(orig, ui, repo, *pats, **opts):
     installnormalfilesmatchfn(repo[None].manifest())
     orig(ui, repo, *pats, **opts)
     restorematchfn()
     m = scmutil.match(repo[None], pats, opts)
 
     try:
         repo.lfstatus = True
         s = repo.status(match=m, clean=True)
     finally:
         repo.lfstatus = False
     forget = sorted(s[0] + s[1] + s[3] + s[6])
     forget = [f for f in forget if lfutil.standin(f) in repo[None].manifest()]
 
     for f in forget:
         if lfutil.standin(f) not in repo.dirstate and not \
                 os.path.isdir(m.rel(lfutil.standin(f))):
             ui.warn(_('not removing %s: file is already untracked\n')
                     % m.rel(f))
 
     for f in forget:
         if ui.verbose or not m.exact(f):
             ui.status(_('removing %s\n') % m.rel(f))
 
     # Need to lock because standin files are deleted then removed from the
     # repository and we could race inbetween.
     wlock = repo.wlock()
     try:
         lfdirstate = lfutil.openlfdirstate(ui, repo)
         for f in forget:
             if lfdirstate[f] == 'a':
                 lfdirstate.drop(f)
             else:
                 lfdirstate.remove(f)
         lfdirstate.write()
         lfutil.repo_remove(repo, [lfutil.standin(f) for f in forget],
             unlink=True)
     finally:
         wlock.release()
 
 def getoutgoinglfiles(ui, repo, dest=None, **opts):
     dest = ui.expandpath(dest or 'default-push', dest or 'default')
     dest, branches = hg.parseurl(dest, opts.get('branch'))
     revs, checkout = hg.addbranchrevs(repo, repo, branches, opts.get('rev'))
     if revs:
         revs = [repo.lookup(rev) for rev in revs]
 
     remoteui = hg.remoteui
 
     try:
         remote = hg.repository(remoteui(repo, opts), dest)
     except error.RepoError:
         return None
     o = lfutil.findoutgoing(repo, remote, False)
     if not o:
         return None
     o = repo.changelog.nodesbetween(o, revs)[0]
     if opts.get('newest_first'):
         o.reverse()
 
     toupload = set()
     for n in o:
         parents = [p for p in repo.changelog.parents(n) if p != node.nullid]
         ctx = repo[n]
         files = set(ctx.files())
         if len(parents) == 2:
             mc = ctx.manifest()
             mp1 = ctx.parents()[0].manifest()
             mp2 = ctx.parents()[1].manifest()
             for f in mp1:
                 if f not in mc:
                         files.add(f)
             for f in mp2:
                 if f not in mc:
                     files.add(f)
             for f in mc:
                 if mc[f] != mp1.get(f, None) or mc[f] != mp2.get(f, None):
                     files.add(f)
         toupload = toupload.union(
             set([f for f in files if lfutil.isstandin(f) and f in ctx]))
     return toupload
 
 def override_outgoing(orig, ui, repo, dest=None, **opts):
     orig(ui, repo, dest, **opts)
 
     if opts.pop('large', None):
         toupload = getoutgoinglfiles(ui, repo, dest, **opts)
         if toupload is None:
             ui.status(_('largefiles: No remote repo\n'))
         else:
             ui.status(_('largefiles to upload:\n'))
             for file in toupload:
                 ui.status(lfutil.splitstandin(file) + '\n')
             ui.status('\n')
 
 def override_summary(orig, ui, repo, *pats, **opts):
     try:
         repo.lfstatus = True
         orig(ui, repo, *pats, **opts)
     finally:
         repo.lfstatus = False
 
     if opts.pop('large', None):
         toupload = getoutgoinglfiles(ui, repo, None, **opts)
         if toupload is None:
             ui.status(_('largefiles: No remote repo\n'))
         else:
             ui.status(_('largefiles: %d to upload\n') % len(toupload))
 
 def override_addremove(orig, ui, repo, *pats, **opts):
     # Get the list of missing largefiles so we can remove them
     lfdirstate = lfutil.openlfdirstate(ui, repo)
     s = lfdirstate.status(match_.always(repo.root, repo.getcwd()), [], False,
         False, False)
     (unsure, modified, added, removed, missing, unknown, ignored, clean) = s
 
     # Call into the normal remove code, but the removing of the standin, we want
     # to have handled by original addremove.  Monkey patching here makes sure
     # we don't remove the standin in the largefiles code, preventing a very
     # confused state later.
     repo._isaddremove = True
     remove_largefiles(ui, repo, *missing, **opts)
     repo._isaddremove = False
     # Call into the normal add code, and any files that *should* be added as
     # largefiles will be
     add_largefiles(ui, repo, *pats, **opts)
     # Now that we've handled largefiles, hand off to the original addremove
     # function to take care of the rest.  Make sure it doesn't do anything with
     # largefiles by installing a matcher that will ignore them.
     installnormalfilesmatchfn(repo[None].manifest())
     result = orig(ui, repo, *pats, **opts)
     restorematchfn()
     return result
 
 # Calling purge with --all will cause the largefiles to be deleted.
 # Override repo.status to prevent this from happening.
 def override_purge(orig, ui, repo, *dirs, **opts):
     oldstatus = repo.status
     def override_status(node1='.', node2=None, match=None, ignored=False,
                         clean=False, unknown=False, listsubrepos=False):
         r = oldstatus(node1, node2, match, ignored, clean, unknown,
                       listsubrepos)
         lfdirstate = lfutil.openlfdirstate(ui, repo)
         modified, added, removed, deleted, unknown, ignored, clean = r
         unknown = [f for f in unknown if lfdirstate[f] == '?']
         ignored = [f for f in ignored if lfdirstate[f] == '?']
         return modified, added, removed, deleted, unknown, ignored, clean
     repo.status = override_status
     orig(ui, repo, *dirs, **opts)
     repo.status = oldstatus
 
 def override_rollback(orig, ui, repo, **opts):
     result = orig(ui, repo, **opts)
     merge.update(repo, node=None, branchmerge=False, force=True,
         partial=lfutil.isstandin)
-    lfdirstate = lfutil.openlfdirstate(ui, repo)
-    lfiles = lfutil.listlfiles(repo)
-    oldlfiles = lfutil.listlfiles(repo, repo[None].parents()[0].rev())
-    for file in lfiles:
-        if file in oldlfiles:
-            lfdirstate.normallookup(file)
-        else:
-            lfdirstate.add(file)
-    lfdirstate.write()
+    wlock = repo.wlock()
+    try:
+        lfdirstate = lfutil.openlfdirstate(ui, repo)
+        lfiles = lfutil.listlfiles(repo)
+        oldlfiles = lfutil.listlfiles(repo, repo[None].parents()[0].rev())
+        for file in lfiles:
+            if file in oldlfiles:
+                lfdirstate.normallookup(file)
+            else:
+                lfdirstate.add(file)
+        lfdirstate.write()
+    finally:
+        wlock.release()
     return result
 
 def override_transplant(orig, ui, repo, *revs, **opts):
     result = orig(ui, repo, *revs, **opts)
     lfcommands.updatelfiles(repo.ui, repo)
     return result

hgext/largefiles/reposetup.py

@@ -1,450 +1,451 b''
 # Copyright 2009-2010 Gregory P. Ward
 # Copyright 2009-2010 Intelerad Medical Systems Incorporated
 # Copyright 2010-2011 Fog Creek Software
 # Copyright 2010-2011 Unity Technologies
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 '''setup for largefiles repositories: reposetup'''
 import copy
 import types
 import os
 
 from mercurial import context, error, manifest, match as match_, util
 from mercurial import node as node_
 from mercurial.i18n import _
 
 import lfcommands
 import proto
 import lfutil
 
 def reposetup(ui, repo):
     # wire repositories should be given new wireproto functions but not the
     # other largefiles modifications
     if not repo.local():
         return proto.wirereposetup(ui, repo)
 
     for name in ('status', 'commitctx', 'commit', 'push'):
         method = getattr(repo, name)
         if (isinstance(method, types.FunctionType) and
             method.func_name == 'wrap'):
             ui.warn(_('largefiles: repo method %r appears to have already been'
                     ' wrapped by another extension: '
                     'largefiles may behave incorrectly\n')
                     % name)
 
     class lfiles_repo(repo.__class__):
         lfstatus = False
         def status_nolfiles(self, *args, **kwargs):
             return super(lfiles_repo, self).status(*args, **kwargs)
 
         # When lfstatus is set, return a context that gives the names
         # of largefiles instead of their corresponding standins and
         # identifies the largefiles as always binary, regardless of
         # their actual contents.
         def __getitem__(self, changeid):
             ctx = super(lfiles_repo, self).__getitem__(changeid)
             if self.lfstatus:
                 class lfiles_manifestdict(manifest.manifestdict):
                     def __contains__(self, filename):
                         if super(lfiles_manifestdict,
                                 self).__contains__(filename):
                             return True
                         return super(lfiles_manifestdict,
                             self).__contains__(lfutil.standin(filename))
                 class lfiles_ctx(ctx.__class__):
                     def files(self):
                         filenames = super(lfiles_ctx, self).files()
                         return [lfutil.splitstandin(f) or f for f in filenames]
                     def manifest(self):
                         man1 = super(lfiles_ctx, self).manifest()
                         man1.__class__ = lfiles_manifestdict
                         return man1
                     def filectx(self, path, fileid=None, filelog=None):
                         try:
                             result = super(lfiles_ctx, self).filectx(path,
                                 fileid, filelog)
                         except error.LookupError:
                             # Adding a null character will cause Mercurial to
                             # identify this as a binary file.
                             result = super(lfiles_ctx, self).filectx(
                                 lfutil.standin(path), fileid, filelog)
                             olddata = result.data
                             result.data = lambda: olddata() + '\0'
                         return result
                 ctx.__class__ = lfiles_ctx
             return ctx
 
         # Figure out the status of big files and insert them into the
         # appropriate list in the result. Also removes standin files
         # from the listing. Revert to the original status if
         # self.lfstatus is False.
         def status(self, node1='.', node2=None, match=None, ignored=False,
                 clean=False, unknown=False, listsubrepos=False):
             listignored, listclean, listunknown = ignored, clean, unknown
             if not self.lfstatus:
                 return super(lfiles_repo, self).status(node1, node2, match,
                     listignored, listclean, listunknown, listsubrepos)
             else:
                 # some calls in this function rely on the old version of status
                 self.lfstatus = False
                 if isinstance(node1, context.changectx):
                     ctx1 = node1
                 else:
                     ctx1 = repo[node1]
                 if isinstance(node2, context.changectx):
                     ctx2 = node2
                 else:
                     ctx2 = repo[node2]
                 working = ctx2.rev() is None
                 parentworking = working and ctx1 == self['.']
 
                 def inctx(file, ctx):
                     try:
                         if ctx.rev() is None:
                             return file in ctx.manifest()
                         ctx[file]
                         return True
                     except KeyError:
                         return False
 
                 if match is None:
                     match = match_.always(self.root, self.getcwd())
 
                 # First check if there were files specified on the
                 # command line.  If there were, and none of them were
                 # largefiles, we should just bail here and let super
                 # handle it -- thus gaining a big performance boost.
                 lfdirstate = lfutil.openlfdirstate(ui, self)
                 if match.files() and not match.anypats():
                     matchedfiles = [f for f in match.files() if f in lfdirstate]
                     if not matchedfiles:
                         return super(lfiles_repo, self).status(node1, node2,
                                 match, listignored, listclean,
                                 listunknown, listsubrepos)
 
                 # Create a copy of match that matches standins instead
                 # of largefiles.
                 def tostandin(file):
                     if inctx(lfutil.standin(file), ctx2):
                         return lfutil.standin(file)
                     return file
 
                 # Create a function that we can use to override what is
                 # normally the ignore matcher.  We've already checked
                 # for ignored files on the first dirstate walk, and
                 # unecessarily re-checking here causes a huge performance
                 # hit because lfdirstate only knows about largefiles
                 def _ignoreoverride(self):
                     return False
 
                 m = copy.copy(match)
                 m._files = [tostandin(f) for f in m._files]
 
                 # Get ignored files here even if we weren't asked for them; we
                 # must use the result here for filtering later
                 result = super(lfiles_repo, self).status(node1, node2, m,
                     True, clean, unknown, listsubrepos)
                 if working:
-                    # hold the wlock while we read largefiles and
-                    # update the lfdirstate
-                    wlock = repo.wlock()
                     try:
                         # Any non-largefiles that were explicitly listed must be
                         # taken out or lfdirstate.status will report an error.
                         # The status of these files was already computed using
                         # super's status.
                         # Override lfdirstate's ignore matcher to not do
                         # anything
                         orig_ignore = lfdirstate._ignore
                         lfdirstate._ignore = _ignoreoverride
 
                         match._files = [f for f in match._files if f in
                             lfdirstate]
                         # Don't waste time getting the ignored and unknown
                         # files again; we already have them
                         s = lfdirstate.status(match, [], False,
                                 listclean, False)
                         (unsure, modified, added, removed, missing, unknown,
                                 ignored, clean) = s
                         # Replace the list of ignored and unknown files with
                         # the previously caclulated lists, and strip out the
                         # largefiles
                         lfiles = set(lfdirstate._map)
                         ignored = set(result[5]).difference(lfiles)
                         unknown = set(result[4]).difference(lfiles)
                         if parentworking:
                             for lfile in unsure:
                                 standin = lfutil.standin(lfile)
                                 if standin not in ctx1:
                                     # from second parent
                                     modified.append(lfile)
                                 elif ctx1[standin].data().strip() \
                                         != lfutil.hashfile(self.wjoin(lfile)):
                                     modified.append(lfile)
                                 else:
                                     clean.append(lfile)
                                     lfdirstate.normal(lfile)
-                            lfdirstate.write()
                         else:
                             tocheck = unsure + modified + added + clean
                             modified, added, clean = [], [], []
 
                             for lfile in tocheck:
                                 standin = lfutil.standin(lfile)
                                 if inctx(standin, ctx1):
                                     if ctx1[standin].data().strip() != \
                                             lfutil.hashfile(self.wjoin(lfile)):
                                         modified.append(lfile)
                                     else:
                                         clean.append(lfile)
                                 else:
                                     added.append(lfile)
+                    finally:
                         # Replace the original ignore function
                         lfdirstate._ignore = orig_ignore
-                    finally:
-                        wlock.release()
 
                     for standin in ctx1.manifest():
                         if not lfutil.isstandin(standin):
                             continue
                         lfile = lfutil.splitstandin(standin)
                         if not match(lfile):
                             continue
                         if lfile not in lfdirstate:
                             removed.append(lfile)
 
                     # Filter result lists
                     result = list(result)
 
                     # Largefiles are not really removed when they're
                     # still in the normal dirstate. Likewise, normal
                     # files are not really removed if it's still in
                     # lfdirstate. This happens in merges where files
                     # change type.
                     removed = [f for f in removed if f not in repo.dirstate]
                     result[2] = [f for f in result[2] if f not in lfdirstate]
 
                     # Unknown files
                     unknown = set(unknown).difference(ignored)
                     result[4] = [f for f in unknown
                                  if (repo.dirstate[f] == '?' and
                                      not lfutil.isstandin(f))]
                     # Ignored files were calculated earlier by the dirstate,
                     # and we already stripped out the largefiles from the list
                     result[5] = ignored
                     # combine normal files and largefiles
                     normals = [[fn for fn in filelist
                                 if not lfutil.isstandin(fn)]
                                for filelist in result]
                     lfiles = (modified, added, removed, missing, [], [], clean)
                     result = [sorted(list1 + list2)
                               for (list1, list2) in zip(normals, lfiles)]
                 else:
                     def toname(f):
                         if lfutil.isstandin(f):
                             return lfutil.splitstandin(f)
                         return f
                     result = [[toname(f) for f in items] for items in result]
 
                 if not listunknown:
                     result[4] = []
                 if not listignored:
                     result[5] = []
                 if not listclean:
                     result[6] = []
                 self.lfstatus = True
                 return result
 
         # As part of committing, copy all of the largefiles into the
         # cache.
         def commitctx(self, *args, **kwargs):
             node = super(lfiles_repo, self).commitctx(*args, **kwargs)
             ctx = self[node]
             for filename in ctx.files():
                 if lfutil.isstandin(filename) and filename in ctx.manifest():
                     realfile = lfutil.splitstandin(filename)
                     lfutil.copytostore(self, ctx.node(), realfile)
 
             return node
 
         # Before commit, largefile standins have not had their
         # contents updated to reflect the hash of their largefile.
         # Do that here.
         def commit(self, text="", user=None, date=None, match=None,
                 force=False, editor=False, extra={}):
             orig = super(lfiles_repo, self).commit
 
             wlock = repo.wlock()
             try:
                 # Case 0: Rebase
                 # We have to take the time to pull down the new largefiles now.
                 # Otherwise if we are rebasing, any largefiles that were
                 # modified in the destination changesets get overwritten, either
                 # by the rebase or in the first commit after the rebase.
                 # updatelfiles will update the dirstate to mark any pulled
                 # largefiles as modified
                 if getattr(repo, "_isrebasing", False):
                     lfcommands.updatelfiles(repo.ui, repo)
                     result = orig(text=text, user=user, date=date, match=match,
                                     force=force, editor=editor, extra=extra)
                     return result
                 # Case 1: user calls commit with no specific files or
                 # include/exclude patterns: refresh and commit all files that
                 # are "dirty".
                 if ((match is None) or
                     (not match.anypats() and not match.files())):
                     # Spend a bit of time here to get a list of files we know
                     # are modified so we can compare only against those.
                     # It can cost a lot of time (several seconds)
                     # otherwise to update all standins if the largefiles are
                     # large.
                     lfdirstate = lfutil.openlfdirstate(ui, self)
                     dirtymatch = match_.always(repo.root, repo.getcwd())
                     s = lfdirstate.status(dirtymatch, [], False, False, False)
                     modifiedfiles = []
                     for i in s:
                         modifiedfiles.extend(i)
                     lfiles = lfutil.listlfiles(self)
                     # this only loops through largefiles that exist (not
                     # removed/renamed)
                     for lfile in lfiles:
                         if lfile in modifiedfiles:
                             if os.path.exists(self.wjoin(lfutil.standin(lfile))):
                                 # this handles the case where a rebase is being
                                 # performed and the working copy is not updated
                                 # yet.
                                 if os.path.exists(self.wjoin(lfile)):
                                     lfutil.updatestandin(self,
                                         lfutil.standin(lfile))
                                     lfdirstate.normal(lfile)
                     for lfile in lfdirstate:
                         if lfile in modifiedfiles:
                             if not os.path.exists(
                                     repo.wjoin(lfutil.standin(lfile))):
                                 lfdirstate.drop(lfile)
+
+                    result = orig(text=text, user=user, date=date, match=match,
+                                    force=force, editor=editor, extra=extra)
+                    # This needs to be after commit; otherwise precommit hooks
+                    # get the wrong status
                     lfdirstate.write()
-
-                    return orig(text=text, user=user, date=date, match=match,
-                                    force=force, editor=editor, extra=extra)
+                    return result
 
                 for f in match.files():
                     if lfutil.isstandin(f):
                         raise util.Abort(
                             _('file "%s" is a largefile standin') % f,
                             hint=('commit the largefile itself instead'))
 
                 # Case 2: user calls commit with specified patterns: refresh
                 # any matching big files.
                 smatcher = lfutil.composestandinmatcher(self, match)
                 standins = lfutil.dirstate_walk(self.dirstate, smatcher)
 
                 # No matching big files: get out of the way and pass control to
                 # the usual commit() method.
                 if not standins:
                     return orig(text=text, user=user, date=date, match=match,
                                     force=force, editor=editor, extra=extra)
 
                 # Refresh all matching big files.  It's possible that the
                 # commit will end up failing, in which case the big files will
                 # stay refreshed.  No harm done: the user modified them and
                 # asked to commit them, so sooner or later we're going to
                 # refresh the standins.  Might as well leave them refreshed.
                 lfdirstate = lfutil.openlfdirstate(ui, self)
                 for standin in standins:
                     lfile = lfutil.splitstandin(standin)
                     if lfdirstate[lfile] <> 'r':
                         lfutil.updatestandin(self, standin)
                         lfdirstate.normal(lfile)
                     else:
                         lfdirstate.drop(lfile)
-                lfdirstate.write()
 
                 # Cook up a new matcher that only matches regular files or
                 # standins corresponding to the big files requested by the
                 # user.  Have to modify _files to prevent commit() from
                 # complaining "not tracked" for big files.
                 lfiles = lfutil.listlfiles(repo)
                 match = copy.copy(match)
                 orig_matchfn = match.matchfn
 
                 # Check both the list of largefiles and the list of
                 # standins because if a largefile was removed, it
                 # won't be in the list of largefiles at this point
                 match._files += sorted(standins)
 
                 actualfiles = []
                 for f in match._files:
                     fstandin = lfutil.standin(f)
 
                     # ignore known largefiles and standins
                     if f in lfiles or fstandin in standins:
                         continue
 
                     # append directory separator to avoid collisions
                     if not fstandin.endswith(os.sep):
                         fstandin += os.sep
 
                     # prevalidate matching standin directories
                     if util.any(st for st in match._files
                                    if st.startswith(fstandin)):
                         continue
                     actualfiles.append(f)
                 match._files = actualfiles
 
                 def matchfn(f):
                     if orig_matchfn(f):
                         return f not in lfiles
                     else:
                         return f in standins
 
                 match.matchfn = matchfn
-                return orig(text=text, user=user, date=date, match=match,
+                result = orig(text=text, user=user, date=date, match=match,
                                 force=force, editor=editor, extra=extra)
+                # This needs to be after commit; otherwise precommit hooks
+                # get the wrong status
+                lfdirstate.write()
+                return result
             finally:
                 wlock.release()
 
         def push(self, remote, force=False, revs=None, newbranch=False):
             o = lfutil.findoutgoing(repo, remote, force)
             if o:
                 toupload = set()
                 o = repo.changelog.nodesbetween(o, revs)[0]
                 for n in o:
                     parents = [p for p in repo.changelog.parents(n)
                                if p != node_.nullid]
                     ctx = repo[n]
                     files = set(ctx.files())
                     if len(parents) == 2:
                         mc = ctx.manifest()
                         mp1 = ctx.parents()[0].manifest()
                         mp2 = ctx.parents()[1].manifest()
                         for f in mp1:
                             if f not in mc:
                                 files.add(f)
                         for f in mp2:
                             if f not in mc:
                                 files.add(f)
                         for f in mc:
                             if mc[f] != mp1.get(f, None) or mc[f] != mp2.get(f,
                                     None):
                                 files.add(f)
 
                     toupload = toupload.union(
                         set([ctx[f].data().strip()
                              for f in files
                              if lfutil.isstandin(f) and f in ctx]))
                 lfcommands.uploadlfiles(ui, self, remote, toupload)
             return super(lfiles_repo, self).push(remote, force, revs,
                 newbranch)
 
     repo.__class__ = lfiles_repo
 
     def checkrequireslfiles(ui, repo, **kwargs):
         if 'largefiles' not in repo.requirements and util.any(
                 lfutil.shortname+'/' in f[0] for f in repo.store.datafiles()):
             repo.requirements.add('largefiles')
             repo._writerequirements()
 
     ui.setconfig('hooks', 'changegroup.lfiles', checkrequireslfiles)
     ui.setconfig('hooks', 'commit.lfiles', checkrequireslfiles)

tests/test-largefiles-small-disk.t

@@ -1,68 +1,67 b''
 Test how largefiles abort in case the disk runs full
 
   $ cat > criple.py <<EOF
   > import os, errno, shutil
   > from mercurial import util
   > #
   > # this makes the original largefiles code abort:
   > def copyfileobj(fsrc, fdst, length=16*1024):
   >     fdst.write(fsrc.read(4))
   >     raise IOError(errno.ENOSPC, os.strerror(errno.ENOSPC))
   > shutil.copyfileobj = copyfileobj
   > #
   > # this makes the rewritten code abort:
   > def filechunkiter(f, size=65536, limit=None):
   >     yield f.read(4)
   >     raise IOError(errno.ENOSPC, os.strerror(errno.ENOSPC))
   > util.filechunkiter = filechunkiter
   > #
   > def oslink(src, dest):
   >     raise OSError("no hardlinks, try copying instead")
   > util.oslink = oslink
   > EOF
 
   $ echo "[extensions]" >> $HGRCPATH
   $ echo "largefiles =" >> $HGRCPATH
 
   $ hg init alice
   $ cd alice
   $ echo "this is a very big file" > big
   $ hg add --large big
   $ hg commit --config extensions.criple=$TESTTMP/criple.py -m big
   abort: No space left on device
   [255]
 
 The largefile is not created in .hg/largefiles:
 
   $ ls .hg/largefiles
   dirstate
 
 The user cache is not even created:
 
   >>> import os; os.path.exists("$HOME/.cache/largefiles/")
   False
 
 Make the commit with space on the device:
 
   $ hg commit -m big
 
 Now make a clone with a full disk, and make sure lfutil.link function
 makes copies instead of hardlinks:
 
   $ cd ..
   $ hg --config extensions.criple=$TESTTMP/criple.py clone --pull alice bob
   requesting all changes
   adding changesets
   adding manifests
   adding file changes
   added 1 changesets with 1 changes to 1 files
   updating to branch default
   1 files updated, 0 files merged, 0 files removed, 0 files unresolved
   getting changed largefiles
   abort: No space left on device
   [255]
 
 The largefile is not created in .hg/largefiles:
 
   $ ls bob/.hg/largefiles
-  dirstate