largefiles: better handling of merge of largefiles that not are available

Before, when merging revisions with missing largefiles, the missing largefiles
would be fetched as a part of the merge. If that failed (for example because
the main repository temporarily was unavailable), the largefile would be left
missing. However, the next commit would abort and (seemed to) fail when
markcommitted tried to mark the standin file as normal and thus had to hash the
largefile that didn't exist. (Actually, the commit would succeed but the
largefile update that follows right after the commit transaction would abort -
quite confusing.)

To fix that, make sure that synclfdirstate only marks files as normal if they
actually exist.

hgext/largefiles/lfutil.py

@@ -1,612 +1,613 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 platform
 import shutil
 import stat
 import copy
 
 from mercurial import dirstate, httpconnection, match as match_, util, scmutil
 from mercurial.i18n import _
 from mercurial import node, error
 
 shortname = '.hglf'
 shortnameslash = shortname + '/'
 longname = 'largefiles'
 
 
 # -- 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 error.Abort(_('largefiles: size must be number (not %s)\n')
                              % lfsize)
     if lfsize is None:
         raise error.Abort(_('minimum size for largefiles must be specified'))
     return lfsize
 
 def link(src, dest):
     util.makedirs(os.path.dirname(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 error.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):
     path, exists = findstorepath(repo, hash)
     if exists:
         repo.ui.note(_('found %s in store\n') % hash)
         return path
     elif inusercache(repo.ui, hash):
         repo.ui.note(_('found %s in system cache\n') % hash)
         path = storepath(repo, hash)
         link(usercachepath(repo.ui, hash), path)
         return path
     return None
 
 class largefilesdirstate(dirstate.dirstate):
     def __getitem__(self, key):
         return super(largefilesdirstate, self).__getitem__(unixpath(key))
     def normal(self, f):
         return super(largefilesdirstate, self).normal(unixpath(f))
     def remove(self, f):
         return super(largefilesdirstate, self).remove(unixpath(f))
     def add(self, f):
         return super(largefilesdirstate, self).add(unixpath(f))
     def drop(self, f):
         return super(largefilesdirstate, self).drop(unixpath(f))
     def forget(self, f):
         return super(largefilesdirstate, self).forget(unixpath(f))
     def normallookup(self, f):
         return super(largefilesdirstate, self).normallookup(unixpath(f))
     def _ignore(self, f):
         return False
 
 def openlfdirstate(ui, repo, create=True):
     '''
     Return a dirstate object that tracks largefiles: i.e. its root is
     the repo root, but it is saved in .hg/largefiles/dirstate.
     '''
     lfstoredir = repo.join(longname)
     opener = scmutil.opener(lfstoredir)
     lfdirstate = largefilesdirstate(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.
     if create and not os.path.exists(os.path.join(lfstoredir, 'dirstate')):
         matcher = getstandinmatcher(repo)
         standins = repo.dirstate.walk(matcher, [], False, False)
 
         if len(standins) > 0:
             util.makedirs(lfstoredir)
 
         for standin in standins:
             lfile = splitstandin(standin)
             lfdirstate.normallookup(lfile)
     return lfdirstate
 
 def lfdirstatestatus(lfdirstate, repo):
     wctx = repo['.']
     match = match_.always(repo.root, repo.getcwd())
     unsure, s = lfdirstate.status(match, [], False, False, False)
     modified, clean = s.modified, s.clean
     for lfile in unsure:
         try:
             fctx = wctx[standin(lfile)]
         except LookupError:
             fctx = None
         if not fctx or fctx.data().strip() != hashfile(repo.wjoin(lfile)):
             modified.append(lfile)
         else:
             clean.append(lfile)
             lfdirstate.normal(lfile)
     return s
 
 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, forcelocal=False):
     return os.path.exists(storepath(repo, hash, forcelocal))
 
 def storepath(repo, hash, forcelocal=False):
     if not forcelocal and repo.shared():
         return repo.vfs.reljoin(repo.sharedpath, longname, hash)
     return repo.join(longname, hash)
 
 def findstorepath(repo, hash):
     '''Search through the local store path(s) to find the file for the given
     hash.  If the file is not found, its path in the primary store is returned.
     The return value is a tuple of (path, exists(path)).
     '''
     # For shared repos, the primary store is in the share source.  But for
     # backward compatibility, force a lookup in the local store if it wasn't
     # found in the share source.
     path = storepath(repo, hash, False)
 
     if instore(repo, hash):
         return (path, True)
     elif repo.shared() and instore(repo, hash, True):
         return storepath(repo, hash, True)
 
     return (path, False)
 
 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, rev)
     if instore(repo, hash):
         return
     copytostoreabsolute(repo, repo.wjoin(file), hash)
 
 def copyalltostore(repo, node):
     '''Copy all largefiles in a given revision to the store'''
 
     ctx = repo[node]
     for filename in ctx.files():
         if isstandin(filename) and filename in ctx.manifest():
             realfile = splitstandin(filename)
             copytostore(repo, ctx.node(), realfile)
 
 
 def copytostoreabsolute(repo, file, hash):
     if inusercache(repo.ui, hash):
         link(usercachepath(repo.ui, hash), storepath(repo, hash))
     else:
         util.makedirs(os.path.dirname(storepath(repo, hash)))
         dst = util.atomictempfile(storepath(repo, hash),
                                   createmode=repo.store.createmode)
         for chunk in util.filechunkiter(open(file, 'rb')):
             dst.write(chunk)
         dst.close()
         linktousercache(repo, hash)
 
 def linktousercache(repo, hash):
     path = usercachepath(repo.ui, hash)
     if path:
         link(storepath(repo, hash), path)
 
 def getstandinmatcher(repo, rmatcher=None):
     '''Return a match object that applies rmatcher to the standin directory'''
     standindir = repo.wjoin(shortname)
 
     # no warnings about missing files or directories
     badfn = lambda f, msg: None
 
     if rmatcher and not rmatcher.always():
         pats = [os.path.join(standindir, pat) for pat in rmatcher.files()]
         if not pats:
             pats = [standindir]
         match = scmutil.match(repo[None], pats, badfn=badfn)
         # if pats is empty, it would incorrectly always match, so clear _always
         match._always = False
     else:
         # no patterns: relative to repo root
         match = scmutil.match(repo[None], [standindir], badfn=badfn)
     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)
     isstandin = smatcher.matchfn
     def composedmatchfn(f):
         return isstandin(f) and rmatcher.matchfn(splitstandin(f))
     smatcher.matchfn = composedmatchfn
 
     return smatcher
 
 def standin(filename):
     '''Return the repo-relative path to the standin for the specified big
     file.'''
     # Notes:
     # 1) Some callers want an absolute path, but for instance addlargefiles
     #    needs it repo-relative so it can be passed to repo[None].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 shortnameslash + util.pconvert(filename)
 
 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(shortnameslash)
 
 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 = util.pconvert(filename).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>'''
     repo.wwrite(standin, hash + '\n', executable and 'x' or '')
 
 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. Return the hash.'''
     hasher = util.sha1('')
     for data in instream:
         hasher.update(data)
         outfile.write(data)
     return hasher.hexdigest()
 
 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 util.filechunkiter(fd, 128 * 1024):
         hasher.update(data)
     fd.close()
     return hasher.hexdigest()
 
 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 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 util.pconvert(os.path.normpath(path))
 
 def islfilesrepo(repo):
     if ('largefiles' in repo.requirements and
             any(shortnameslash in f[0] for f in repo.store.datafiles())):
         return True
 
     return any(openlfdirstate(repo.ui, repo, False))
 
 class storeprotonotcapable(Exception):
     def __init__(self, storetypes):
         self.storetypes = storetypes
 
 def getstandinsstate(repo):
     standins = []
     matcher = getstandinmatcher(repo)
     for standin in repo.dirstate.walk(matcher, [], False, False):
         lfile = splitstandin(standin)
         try:
             hash = readstandin(repo, lfile)
         except IOError:
             hash = None
         standins.append((lfile, hash))
     return standins
 
 def synclfdirstate(repo, lfdirstate, lfile, normallookup):
     lfstandin = standin(lfile)
     if lfstandin in repo.dirstate:
         stat = repo.dirstate._map[lfstandin]
         state, mtime = stat[0], stat[3]
     else:
         state, mtime = '?', -1
     if state == 'n':
-        if normallookup or mtime < 0:
+        if (normallookup or mtime < 0 or
+            not os.path.exists(repo.wjoin(lfile))):
             # state 'n' doesn't ensure 'clean' in this case
             lfdirstate.normallookup(lfile)
         else:
             lfdirstate.normal(lfile)
     elif state == 'm':
         lfdirstate.normallookup(lfile)
     elif state == 'r':
         lfdirstate.remove(lfile)
     elif state == 'a':
         lfdirstate.add(lfile)
     elif state == '?':
         lfdirstate.drop(lfile)
 
 def markcommitted(orig, ctx, node):
     repo = ctx.repo()
 
     orig(node)
 
     # ATTENTION: "ctx.files()" may differ from "repo[node].files()"
     # because files coming from the 2nd parent are omitted in the latter.
     #
     # The former should be used to get targets of "synclfdirstate",
     # because such files:
     # - are marked as "a" by "patch.patch()" (e.g. via transplant), and
     # - have to be marked as "n" after commit, but
     # - aren't listed in "repo[node].files()"
 
     lfdirstate = openlfdirstate(repo.ui, repo)
     for f in ctx.files():
         if isstandin(f):
             lfile = splitstandin(f)
             synclfdirstate(repo, lfdirstate, lfile, False)
     lfdirstate.write()
 
     # As part of committing, copy all of the largefiles into the cache.
     copyalltostore(repo, node)
 
 def getlfilestoupdate(oldstandins, newstandins):
     changedstandins = set(oldstandins).symmetric_difference(set(newstandins))
     filelist = []
     for f in changedstandins:
         if f[0] not in filelist:
             filelist.append(f[0])
     return filelist
 
 def getlfilestoupload(repo, missing, addfunc):
     for i, n in enumerate(missing):
         repo.ui.progress(_('finding outgoing largefiles'), i,
             unit=_('revision'), total=len(missing))
         parents = [p for p in repo.changelog.parents(n) if p != node.nullid]
 
         oldlfstatus = repo.lfstatus
         repo.lfstatus = False
         try:
             ctx = repo[n]
         finally:
             repo.lfstatus = oldlfstatus
 
         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)
         for fn in files:
             if isstandin(fn) and fn in ctx:
                 addfunc(fn, ctx[fn].data().strip())
     repo.ui.progress(_('finding outgoing largefiles'), None)
 
 def updatestandinsbymatch(repo, match):
     '''Update standins in the working directory according to specified match
 
     This returns (possibly modified) ``match`` object to be used for
     subsequent commit process.
     '''
 
     ui = repo.ui
 
     # 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 match.always():
         # 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 = openlfdirstate(ui, repo)
         dirtymatch = match_.always(repo.root, repo.getcwd())
         unsure, s = lfdirstate.status(dirtymatch, [], False, False,
                                       False)
         modifiedfiles = unsure + s.modified + s.added + s.removed
         lfiles = listlfiles(repo)
         # this only loops through largefiles that exist (not
         # removed/renamed)
         for lfile in lfiles:
             if lfile in modifiedfiles:
                 if os.path.exists(
                         repo.wjoin(standin(lfile))):
                     # this handles the case where a rebase is being
                     # performed and the working copy is not updated
                     # yet.
                     if os.path.exists(repo.wjoin(lfile)):
                         updatestandin(repo,
                             standin(lfile))
 
         return match
 
     lfiles = listlfiles(repo)
     match._files = repo._subdirlfs(match.files(), lfiles)
 
     # Case 2: user calls commit with specified patterns: refresh
     # any matching big files.
     smatcher = composestandinmatcher(repo, match)
     standins = repo.dirstate.walk(smatcher, [], False, False)
 
     # No matching big files: get out of the way and pass control to
     # the usual commit() method.
     if not standins:
         return match
 
     # 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 = openlfdirstate(ui, repo)
     for fstandin in standins:
         lfile = splitstandin(fstandin)
         if lfdirstate[lfile] != 'r':
             updatestandin(repo, fstandin)
 
     # 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.
     match = copy.copy(match)
     origmatchfn = 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 = standin(f)
 
         # ignore known largefiles and standins
         if f in lfiles or fstandin in standins:
             continue
 
         actualfiles.append(f)
     match._files = actualfiles
 
     def matchfn(f):
         if origmatchfn(f):
             return f not in lfiles
         else:
             return f in standins
 
     match.matchfn = matchfn
 
     return match
 
 class automatedcommithook(object):
     '''Stateful hook to update standins at the 1st commit of resuming
 
     For efficiency, updating standins in the working directory should
     be avoided while automated committing (like rebase, transplant and
     so on), because they should be updated before committing.
 
     But the 1st commit of resuming automated committing (e.g. ``rebase
     --continue``) should update them, because largefiles may be
     modified manually.
     '''
     def __init__(self, resuming):
         self.resuming = resuming
 
     def __call__(self, repo, match):
         if self.resuming:
             self.resuming = False # avoids updating at subsequent commits
             return updatestandinsbymatch(repo, match)
         else:
             return match
 
 def getstatuswriter(ui, repo, forcibly=None):
     '''Return the function to write largefiles specific status out
 
     If ``forcibly`` is ``None``, this returns the last element of
     ``repo._lfstatuswriters`` as "default" writer function.
 
     Otherwise, this returns the function to always write out (or
     ignore if ``not forcibly``) status.
     '''
     if forcibly is None and util.safehasattr(repo, '_largefilesenabled'):
         return repo._lfstatuswriters[-1]
     else:
         if forcibly:
             return ui.status # forcibly WRITE OUT
         else:
             return lambda *msg, **opts: None # forcibly IGNORE

tests/test-largefiles-update.t

@@ -1,706 +1,728 b''
 This file focuses mainly on updating largefiles in the working
 directory (and ".hg/largefiles/dirstate")
 
   $ cat >> $HGRCPATH <<EOF
   > [ui]
   > merge = internal:fail
   > [extensions]
   > largefiles =
   > EOF
 
   $ hg init repo
   $ cd repo
 
   $ echo large1 > large1
   $ echo large2 > large2
   $ hg add --large large1 large2
   $ echo normal1 > normal1
   $ hg add normal1
   $ hg commit -m '#0'
   $ echo 'large1 in #1' > large1
   $ echo 'normal1 in #1' > normal1
   $ hg commit -m '#1'
   $ hg extdiff -r '.^' --config extensions.extdiff=
   diff -Npru repo.0d9d9b8dc9a3/.hglf/large1 repo/.hglf/large1
   --- repo.0d9d9b8dc9a3/.hglf/large1	* (glob)
   +++ repo/.hglf/large1	* (glob)
   @@ -1 +1 @@
   -4669e532d5b2c093a78eca010077e708a071bb64
   +58e24f733a964da346e2407a2bee99d9001184f5
   diff -Npru repo.0d9d9b8dc9a3/normal1 repo/normal1
   --- repo.0d9d9b8dc9a3/normal1	* (glob)
   +++ repo/normal1	* (glob)
   @@ -1 +1 @@
   -normal1
   +normal1 in #1
   [1]
   $ hg update -q -C 0
   $ echo 'large2 in #2' > large2
   $ hg commit -m '#2'
   created new head
 
 Test that update also updates the lfdirstate of 'unsure' largefiles after
 hashing them:
 
 The previous operations will usually have left us with largefiles with a mtime
 within the same second as the dirstate was written.
 The lfdirstate entries will thus have been written with an invalidated/unset
 mtime to make sure further changes within the same second is detected.
 We will however occasionally be "lucky" and get a tick between writing
 largefiles and writing dirstate so we get valid lfdirstate timestamps. The
 following verification is thus disabled but can be verified manually.
 
 #if false
   $ hg debugdirstate --large --nodate
   n 644          7 unset               large1
   n 644         13 unset               large2
 #endif
 
 Wait to make sure we get a tick so the mtime of the largefiles become valid.
 
   $ sleep 1
 
 A linear merge will update standins before performing the actual merge. It will
 do a lfdirstate status walk and find 'unset'/'unsure' files, hash them, and
 update the corresponding standins.
 Verify that it actually marks the clean files as clean in lfdirstate so
 we don't have to hash them again next time we update.
 
   $ hg up
   0 files updated, 0 files merged, 0 files removed, 0 files unresolved
   $ hg debugdirstate --large --nodate
   n 644          7 set                 large1
   n 644         13 set                 large2
 
 Test that lfdirstate keeps track of last modification of largefiles and
 prevents unnecessary hashing of content - also after linear/noop update
 
   $ sleep 1
   $ hg st
   $ hg debugdirstate --large --nodate
   n 644          7 set                 large1
   n 644         13 set                 large2
   $ hg up
   0 files updated, 0 files merged, 0 files removed, 0 files unresolved
   $ hg debugdirstate --large --nodate
   n 644          7 set                 large1
   n 644         13 set                 large2
 
 Test that "hg merge" updates largefiles from "other" correctly
 
 (getting largefiles from "other" normally)
 
   $ hg status -A large1
   C large1
   $ cat large1
   large1
   $ cat .hglf/large1
   4669e532d5b2c093a78eca010077e708a071bb64
   $ hg merge --config debug.dirstate.delaywrite=2
   getting changed largefiles
   1 largefiles updated, 0 removed
   2 files updated, 0 files merged, 0 files removed, 0 files unresolved
   (branch merge, don't forget to commit)
   $ hg status -A large1
   M large1
   $ cat large1
   large1 in #1
   $ cat .hglf/large1
   58e24f733a964da346e2407a2bee99d9001184f5
   $ hg diff -c 1 --nodates .hglf/large1 | grep '^[+-][0-9a-z]'
   -4669e532d5b2c093a78eca010077e708a071bb64
   +58e24f733a964da346e2407a2bee99d9001184f5
 
 (getting largefiles from "other" via conflict prompt)
 
   $ hg update -q -C 2
   $ echo 'large1 in #3' > large1
   $ echo 'normal1 in #3' > normal1
   $ hg commit -m '#3'
   $ cat .hglf/large1
   e5bb990443d6a92aaf7223813720f7566c9dd05b
   $ hg merge --config debug.dirstate.delaywrite=2 --config ui.interactive=True <<EOF
   > o
   > EOF
   largefile large1 has a merge conflict
   ancestor was 4669e532d5b2c093a78eca010077e708a071bb64
   keep (l)ocal e5bb990443d6a92aaf7223813720f7566c9dd05b or
   take (o)ther 58e24f733a964da346e2407a2bee99d9001184f5? o
   merging normal1
   warning: conflicts while merging normal1! (edit, then use 'hg resolve --mark')
   getting changed largefiles
   1 largefiles updated, 0 removed
   0 files updated, 1 files merged, 0 files removed, 1 files unresolved
   use 'hg resolve' to retry unresolved file merges or 'hg update -C .' to abandon
   [1]
   $ hg status -A large1
   M large1
   $ cat large1
   large1 in #1
   $ cat .hglf/large1
   58e24f733a964da346e2407a2bee99d9001184f5
 
+(merge non-existing largefiles from "other" via conflict prompt -
+make sure the following commit doesn't abort in a confusing way when trying to
+mark the non-existing file as normal in lfdirstate)
+
+  $ mv .hg/largefiles/58e24f733a964da346e2407a2bee99d9001184f5 .
+  $ hg update -q -C 3
+  $ hg merge --config largefiles.usercache=not --config debug.dirstate.delaywrite=2 --tool :local --config ui.interactive=True <<EOF
+  > o
+  > EOF
+  largefile large1 has a merge conflict
+  ancestor was 4669e532d5b2c093a78eca010077e708a071bb64
+  keep (l)ocal e5bb990443d6a92aaf7223813720f7566c9dd05b or
+  take (o)ther 58e24f733a964da346e2407a2bee99d9001184f5? o
+  getting changed largefiles
+  large1: largefile 58e24f733a964da346e2407a2bee99d9001184f5 not available from file:/*/$TESTTMP/repo (glob)
+  0 largefiles updated, 0 removed
+  0 files updated, 2 files merged, 0 files removed, 0 files unresolved
+  (branch merge, don't forget to commit)
+  $ hg commit -m '1-2-3 testing'
+  $ hg rollback -q
+  $ mv 58e24f733a964da346e2407a2bee99d9001184f5 .hg/largefiles/
+
 Test that "hg revert -r REV" updates largefiles from "REV" correctly
 
   $ hg update -q -C 3
   $ hg status -A large1
   C large1
   $ cat large1
   large1 in #3
   $ cat .hglf/large1
   e5bb990443d6a92aaf7223813720f7566c9dd05b
   $ hg diff -c 1 --nodates .hglf/large1 | grep '^[+-][0-9a-z]'
   -4669e532d5b2c093a78eca010077e708a071bb64
   +58e24f733a964da346e2407a2bee99d9001184f5
   $ hg revert --no-backup -r 1 --config debug.dirstate.delaywrite=2 large1
   $ hg status -A large1
   M large1
   $ cat large1
   large1 in #1
   $ cat .hglf/large1
   58e24f733a964da346e2407a2bee99d9001184f5
 
 Test that "hg rollback" restores status of largefiles correctly
 
   $ hg update -C -q
   $ hg remove large1
   $ test -f .hglf/large1
   [1]
   $ hg forget large2
   $ test -f .hglf/large2
   [1]
   $ echo largeX > largeX
   $ hg add --large largeX
   $ cat .hglf/largeX
   
   $ hg commit -m 'will be rollback-ed soon'
   $ echo largeY > largeY
   $ hg add --large largeY
 #if windows
   $ hg status -A large1
   large1: * (glob)
 #else
   $ hg status -A large1
   large1: No such file or directory
 #endif
   $ hg status -A large2
   ? large2
   $ hg status -A largeX
   C largeX
   $ hg status -A largeY
   A largeY
   $ hg rollback
   repository tip rolled back to revision 3 (undo commit)
   working directory now based on revision 3
   $ hg status -A large1
   R large1
   $ test -f .hglf/large1
   [1]
   $ hg status -A large2
   R large2
   $ test -f .hglf/large2
   [1]
   $ hg status -A largeX
   A largeX
   $ cat .hglf/largeX
   
   $ hg status -A largeY
   ? largeY
   $ test -f .hglf/largeY
   [1]
 
 Test that "hg rollback" restores standins correctly
 
   $ hg commit -m 'will be rollback-ed soon'
   $ hg update -q -C 2
   $ cat large1
   large1
   $ cat .hglf/large1
   4669e532d5b2c093a78eca010077e708a071bb64
   $ cat large2
   large2 in #2
   $ cat .hglf/large2
   3cfce6277e7668985707b6887ce56f9f62f6ccd9
 
   $ hg rollback -q -f
   $ cat large1
   large1
   $ cat .hglf/large1
   4669e532d5b2c093a78eca010077e708a071bb64
   $ cat large2
   large2 in #2
   $ cat .hglf/large2
   3cfce6277e7668985707b6887ce56f9f62f6ccd9
 
 (rollback the parent of the working directory, when the parent of it
 is not branch-tip)
 
   $ hg update -q -C 1
   $ cat .hglf/large1
   58e24f733a964da346e2407a2bee99d9001184f5
   $ cat .hglf/large2
   1deebade43c8c498a3c8daddac0244dc55d1331d
 
   $ echo normalX > normalX
   $ hg add normalX
   $ hg commit -m 'will be rollback-ed soon'
   $ hg rollback -q
 
   $ cat .hglf/large1
   58e24f733a964da346e2407a2bee99d9001184f5
   $ cat .hglf/large2
   1deebade43c8c498a3c8daddac0244dc55d1331d
 
 Test that "hg status" shows status of largefiles correctly just after
 automated commit like rebase/transplant
 
   $ cat >> .hg/hgrc <<EOF
   > [extensions]
   > rebase =
   > strip =
   > transplant =
   > EOF
   $ hg update -q -C 1
   $ hg remove large1
   $ echo largeX > largeX
   $ hg add --large largeX
   $ hg commit -m '#4'
 
   $ hg rebase -s 1 -d 2 --keep
   rebasing 1:72518492caa6 "#1"
   rebasing 4:07d6153b5c04 "#4" (tip)
 #if windows
   $ hg status -A large1
   large1: * (glob)
 #else
   $ hg status -A large1
   large1: No such file or directory
 #endif
   $ hg status -A largeX
   C largeX
   $ hg strip -q 5
 
   $ hg update -q -C 2
   $ hg transplant -q 1 4
 #if windows
   $ hg status -A large1
   large1: * (glob)
 #else
   $ hg status -A large1
   large1: No such file or directory
 #endif
   $ hg status -A largeX
   C largeX
   $ hg strip -q 5
 
   $ hg update -q -C 2
   $ hg transplant -q --merge 1 --merge 4
 #if windows
   $ hg status -A large1
   large1: * (glob)
 #else
   $ hg status -A large1
   large1: No such file or directory
 #endif
   $ hg status -A largeX
   C largeX
   $ hg strip -q 5
 
 Test that linear merge can detect modification (and conflict) correctly
 
 (linear merge without conflict)
 
   $ echo 'large2 for linear merge (no conflict)' > large2
   $ hg update 3 --config debug.dirstate.delaywrite=2
   getting changed largefiles
   1 largefiles updated, 0 removed
   2 files updated, 0 files merged, 0 files removed, 0 files unresolved
   $ hg status -A large2
   M large2
   $ cat large2
   large2 for linear merge (no conflict)
   $ cat .hglf/large2
   9c4bf8f1b33536d6e5f89447e10620cfe52ea710
 
 (linear merge with conflict, choosing "other")
 
   $ hg update -q -C 2
   $ echo 'large1 for linear merge (conflict)' > large1
   $ hg update 3 --config ui.interactive=True <<EOF
   > o
   > EOF
   largefile large1 has a merge conflict
   ancestor was 4669e532d5b2c093a78eca010077e708a071bb64
   keep (l)ocal ba94c2efe5b7c5e0af8d189295ce00553b0612b7 or
   take (o)ther e5bb990443d6a92aaf7223813720f7566c9dd05b? o
   getting changed largefiles
   1 largefiles updated, 0 removed
   1 files updated, 1 files merged, 0 files removed, 0 files unresolved
   $ hg status -A large1
   C large1
   $ cat large1
   large1 in #3
   $ cat .hglf/large1
   e5bb990443d6a92aaf7223813720f7566c9dd05b
 
 (linear merge with conflict, choosing "local")
 
   $ hg update -q -C 2
   $ echo 'large1 for linear merge (conflict)' > large1
   $ hg update 3 --config debug.dirstate.delaywrite=2
   largefile large1 has a merge conflict
   ancestor was 4669e532d5b2c093a78eca010077e708a071bb64
   keep (l)ocal ba94c2efe5b7c5e0af8d189295ce00553b0612b7 or
   take (o)ther e5bb990443d6a92aaf7223813720f7566c9dd05b? l
   1 files updated, 1 files merged, 0 files removed, 0 files unresolved
   $ hg status -A large1
   M large1
   $ cat large1
   large1 for linear merge (conflict)
   $ cat .hglf/large1
   ba94c2efe5b7c5e0af8d189295ce00553b0612b7
 
 Test a linear merge to a revision containing same-name normal file
 
   $ hg update -q -C 3
   $ hg remove large2
   $ echo 'large2 as normal file' > large2
   $ hg add large2
   $ echo 'large3 as normal file' > large3
   $ hg add large3
   $ hg commit -m '#5'
   $ hg manifest
   .hglf/large1
   large2
   large3
   normal1
 
 (modified largefile is already switched to normal)
 
   $ hg update -q -C 2
   $ echo 'modified large2 for linear merge' > large2
   $ hg update -q 5
   remote turned local largefile large2 into a normal file
   keep (l)argefile or use (n)ormal file? l
   $ hg debugdirstate --nodates | grep large2
   a   0         -1 unset               .hglf/large2
   r   0          0 set                 large2
   $ hg status -A large2
   A large2
   $ cat large2
   modified large2 for linear merge
 
 (added largefile is already committed as normal)
 
   $ hg update -q -C 2
   $ echo 'large3 as large file for linear merge' > large3
   $ hg add --large large3
   $ hg update -q 5
   remote turned local largefile large3 into a normal file
   keep (l)argefile or use (n)ormal file? l
   $ hg debugdirstate --nodates | grep large3
   a   0         -1 unset               .hglf/large3
   r   0          0 set                 large3
   $ hg status -A large3
   A large3
   $ cat large3
   large3 as large file for linear merge
   $ rm -f large3 .hglf/large3
 
 Test that the internal linear merging works correctly
 (both heads are stripped to keep pairing of revision number and commit log)
 
   $ hg update -q -C 2
   $ hg strip 3 4
   saved backup bundle to $TESTTMP/repo/.hg/strip-backup/9530e27857f7-2e7b195d-backup.hg (glob)
   $ mv .hg/strip-backup/9530e27857f7-2e7b195d-backup.hg $TESTTMP
 
 (internal linear merging at "hg pull --update")
 
   $ echo 'large1 for linear merge (conflict)' > large1
   $ echo 'large2 for linear merge (conflict with normal file)' > large2
   $ hg pull --update --config debug.dirstate.delaywrite=2 $TESTTMP/9530e27857f7-2e7b195d-backup.hg
   pulling from $TESTTMP/9530e27857f7-2e7b195d-backup.hg (glob)
   searching for changes
   adding changesets
   adding manifests
   adding file changes
   added 3 changesets with 5 changes to 5 files
   remote turned local largefile large2 into a normal file
   keep (l)argefile or use (n)ormal file? l
   largefile large1 has a merge conflict
   ancestor was 4669e532d5b2c093a78eca010077e708a071bb64
   keep (l)ocal ba94c2efe5b7c5e0af8d189295ce00553b0612b7 or
   take (o)ther e5bb990443d6a92aaf7223813720f7566c9dd05b? l
   2 files updated, 1 files merged, 0 files removed, 0 files unresolved
 
   $ hg status -A large1
   M large1
   $ cat large1
   large1 for linear merge (conflict)
   $ cat .hglf/large1
   ba94c2efe5b7c5e0af8d189295ce00553b0612b7
   $ hg status -A large2
   A large2
   $ cat large2
   large2 for linear merge (conflict with normal file)
   $ cat .hglf/large2
   d7591fe9be0f6227d90bddf3e4f52ff41fc1f544
 
 (internal linear merging at "hg unbundle --update")
 
   $ hg update -q -C 2
   $ hg rollback -q
 
   $ echo 'large1 for linear merge (conflict)' > large1
   $ echo 'large2 for linear merge (conflict with normal file)' > large2
   $ hg unbundle --update --config debug.dirstate.delaywrite=2 $TESTTMP/9530e27857f7-2e7b195d-backup.hg
   adding changesets
   adding manifests
   adding file changes
   added 3 changesets with 5 changes to 5 files
   remote turned local largefile large2 into a normal file
   keep (l)argefile or use (n)ormal file? l
   largefile large1 has a merge conflict
   ancestor was 4669e532d5b2c093a78eca010077e708a071bb64
   keep (l)ocal ba94c2efe5b7c5e0af8d189295ce00553b0612b7 or
   take (o)ther e5bb990443d6a92aaf7223813720f7566c9dd05b? l
   2 files updated, 1 files merged, 0 files removed, 0 files unresolved
 
   $ hg status -A large1
   M large1
   $ cat large1
   large1 for linear merge (conflict)
   $ cat .hglf/large1
   ba94c2efe5b7c5e0af8d189295ce00553b0612b7
   $ hg status -A large2
   A large2
   $ cat large2
   large2 for linear merge (conflict with normal file)
   $ cat .hglf/large2
   d7591fe9be0f6227d90bddf3e4f52ff41fc1f544
 
 (internal linear merging in subrepo at "hg update")
 
   $ cd ..
   $ hg init subparent
   $ cd subparent
 
   $ hg clone -q -u 2 ../repo sub
   $ cat > .hgsub <<EOF
   > sub = sub
   > EOF
   $ hg add .hgsub
   $ hg commit -m '#0@parent'
   $ cat .hgsubstate
   f74e50bd9e5594b7cf1e6c5cbab86ddd25f3ca2f sub
   $ hg -R sub update -q
   $ hg commit -m '#1@parent'
   $ cat .hgsubstate
   d65e59e952a9638e2ce863b41a420ca723dd3e8d sub
   $ hg update -q 0
 
   $ echo 'large1 for linear merge (conflict)' > sub/large1
   $ echo 'large2 for linear merge (conflict with normal file)' > sub/large2
   $ hg update --config ui.interactive=True --config debug.dirstate.delaywrite=2 <<EOF
   > m
   > r
   > l
   > l
   > EOF
    subrepository sub diverged (local revision: f74e50bd9e55, remote revision: d65e59e952a9)
   (M)erge, keep (l)ocal or keep (r)emote? m
    subrepository sources for sub differ (in checked out version)
   use (l)ocal source (f74e50bd9e55) or (r)emote source (d65e59e952a9)? r
   remote turned local largefile large2 into a normal file
   keep (l)argefile or use (n)ormal file? l
   largefile large1 has a merge conflict
   ancestor was 4669e532d5b2c093a78eca010077e708a071bb64
   keep (l)ocal ba94c2efe5b7c5e0af8d189295ce00553b0612b7 or
   take (o)ther e5bb990443d6a92aaf7223813720f7566c9dd05b? l
   2 files updated, 1 files merged, 0 files removed, 0 files unresolved
   0 files updated, 0 files merged, 0 files removed, 0 files unresolved
 
   $ hg -R sub status -A sub/large1
   M sub/large1
   $ cat sub/large1
   large1 for linear merge (conflict)
   $ cat sub/.hglf/large1
   ba94c2efe5b7c5e0af8d189295ce00553b0612b7
   $ hg -R sub status -A sub/large2
   A sub/large2
   $ cat sub/large2
   large2 for linear merge (conflict with normal file)
   $ cat sub/.hglf/large2
   d7591fe9be0f6227d90bddf3e4f52ff41fc1f544
 
   $ cd ..
   $ cd repo
 
 Test that rebase updates largefiles in the working directory even if
 it is aborted by conflict.
 
   $ hg update -q -C 3
   $ cat .hglf/large1
   e5bb990443d6a92aaf7223813720f7566c9dd05b
   $ cat large1
   large1 in #3
   $ hg rebase -s 1 -d 3 --keep --config ui.interactive=True <<EOF
   > o
   > EOF
   rebasing 1:72518492caa6 "#1"
   largefile large1 has a merge conflict
   ancestor was 4669e532d5b2c093a78eca010077e708a071bb64
   keep (l)ocal e5bb990443d6a92aaf7223813720f7566c9dd05b or
   take (o)ther 58e24f733a964da346e2407a2bee99d9001184f5? o
   merging normal1
   warning: conflicts while merging normal1! (edit, then use 'hg resolve --mark')
   unresolved conflicts (see hg resolve, then hg rebase --continue)
   [1]
   $ cat .hglf/large1
   58e24f733a964da346e2407a2bee99d9001184f5
   $ cat large1
   large1 in #1
 
 Test that rebase updates standins for manually modified largefiles at
 the 1st commit of resuming.
 
   $ echo "manually modified before 'hg rebase --continue'" > large1
   $ hg resolve -m normal1
   (no more unresolved files)
   $ hg rebase --continue --config ui.interactive=True <<EOF
   > c
   > EOF
   rebasing 1:72518492caa6 "#1"
   rebasing 4:07d6153b5c04 "#4"
   local changed .hglf/large1 which remote deleted
   use (c)hanged version or (d)elete? c
 
   $ hg diff -c "tip~1" --nodates .hglf/large1 | grep '^[+-][0-9a-z]'
   -e5bb990443d6a92aaf7223813720f7566c9dd05b
   +8a4f783556e7dea21139ca0466eafce954c75c13
   $ rm -f large1
   $ hg update -q -C tip
   $ cat large1
   manually modified before 'hg rebase --continue'
 
 Test that transplant updates largefiles, of which standins are safely
 changed, even if it is aborted by conflict of other.
 
   $ hg update -q -C 5
   $ cat .hglf/large1
   e5bb990443d6a92aaf7223813720f7566c9dd05b
   $ cat large1
   large1 in #3
   $ hg diff -c 4 .hglf/largeX | grep '^[+-][0-9a-z]'
   +fa44618ea25181aff4f48b70428294790cec9f61
   $ hg transplant 4
   applying 07d6153b5c04
   patching file .hglf/large1
   Hunk #1 FAILED at 0
   1 out of 1 hunks FAILED -- saving rejects to file .hglf/large1.rej
   patch failed to apply
   abort: fix up the merge and run hg transplant --continue
   [255]
   $ hg status -A large1
   C large1
   $ cat .hglf/large1
   e5bb990443d6a92aaf7223813720f7566c9dd05b
   $ cat large1
   large1 in #3
   $ hg status -A largeX
   A largeX
   $ cat .hglf/largeX
   fa44618ea25181aff4f48b70428294790cec9f61
   $ cat largeX
   largeX
 
 Test that transplant updates standins for manually modified largefiles
 at the 1st commit of resuming.
 
   $ echo "manually modified before 'hg transplant --continue'" > large1
   $ hg transplant --continue
   07d6153b5c04 transplanted as f1bf30eb88cc
   $ hg diff -c tip .hglf/large1 | grep '^[+-][0-9a-z]'
   -e5bb990443d6a92aaf7223813720f7566c9dd05b
   +6a4f36d4075fbe0f30ec1d26ca44e63c05903671
   $ rm -f large1
   $ hg update -q -C tip
   $ cat large1
   manually modified before 'hg transplant --continue'
 
 Test that "hg status" doesn't show removal of largefiles not managed
 in the target context.
 
   $ hg update -q -C 4
   $ hg remove largeX
   $ hg status -A largeX
   R largeX
   $ hg status -A --rev '.^1' largeX
 
 #if execbit
 
 Test that "hg status" against revisions other than parent notices exec
 bit changes of largefiles.
 
   $ hg update -q -C 4
 
 (the case that large2 doesn't have exec bit in the target context but
 in the working context)
 
   $ chmod +x large2
   $ hg status -A --rev 0 large2
   M large2
   $ hg commit -m 'chmod +x large2'
 
 (the case that large2 has exec bit in the target context but not in
 the working context)
 
   $ echo dummy > dummy
   $ hg add dummy
   $ hg commit -m 'revision for separation'
   $ chmod -x large2
   $ hg status -A --rev '.^1' large2
   M large2
 
 #else
 
 Test that "hg status" against revisions other than parent ignores exec
 bit correctly on the platform being unaware of it.
 
   $ hg update -q -C 4
 
   $ cat > exec-bit.patch <<EOF
   > # HG changeset patch
   > # User test
   > # Date 0 0
   > #      Thu Jan 01 00:00:00 1970 +0000
   > # Node ID be1b433a65b12b27b5519d92213e14f7e1769b90
   > # Parent  07d6153b5c04313efb75deec9ba577de7faeb727
   > chmod +x large2
   > 
   > diff --git a/.hglf/large2 b/.hglf/large2
   > old mode 100644
   > new mode 100755
   > EOF
   $ hg import --exact --bypass exec-bit.patch
   applying exec-bit.patch
   $ hg status -A --rev tip large2
   C large2
 
 #endif
 
   $ cd ..
 
 Test that "hg convert" avoids copying largefiles from the working
 directory into store, because "hg convert" doesn't update largefiles
 in the working directory (removing files under ".cache/largefiles"
 forces "hg convert" to copy corresponding largefiles)
 
   $ cat >> $HGRCPATH <<EOF
   > [extensions]
   > convert =
   > EOF
 
   $ rm $TESTTMP/.cache/largefiles/6a4f36d4075fbe0f30ec1d26ca44e63c05903671
   $ hg convert -q repo repo.converted