largefiles: cache new largefiles for new heads when pulling

When the user pulls from a remote repository that is not his default repo, it
is quite likely that he will pull a new head.  This means that if he tries to
merge or rebase with the other head, he will run into a problem becuase
largefiles has no way of tracking where the remote repository for this other
head is, so it cannot download the largefiles from this other remote repository.
It will attempt to download them from its default remote repository, which will
not yet contain the largefiles.

This patch solves this problem by caching any new largefiles for all heads
directly into the system cache at the time of the pull, so they are available
later.

This behavior is actually more in line with Mercurial's distributed nature,
because pulling already implies we have a connection to the remote server, but
merging or rebasing does not.

hgext/largefiles/overrides.py

@@ -1,942 +1,955 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):
     # Mark the repo as being in the middle of a merge, so that
     # updatelfiles() will know that it needs to trust the standins in
     # the working copy, not in the standins in the current node
     repo._ismerging = True
     try:
         result = orig(repo, node, force, remind)
         lfcommands.updatelfiles(repo.ui, repo)
     finally:
         repo._ismerging = False
     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)
+        # If we do not have the new largefiles for any new heads we pulled, we
+        # will run into a problem later if we try to merge or rebase with one of
+        # these heads, so cache the largefiles now direclty into the system
+        # cache.
+        ui.status(_("caching new largefiles\n"))
+        numcached = 0
+        branches = repo.branchmap()
+        for branch in branches:
+            heads = repo.branchheads(branch)
+            for head in heads:
+                (cached, missing) = lfcommands.cachelfiles(ui, repo, head)
+                numcached += len(cached)
+        ui.status(_("%d largefiles cached\n" % numcached))
     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())
             if path is None:
                 raise util.Abort(
                     _('largefile %s not found in repo store or system cache')
                     % lfutil.splitstandin(f))
             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)
     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

tests/test-largefiles-cache.t

@@ -1,70 +1,72 b''
 Create user cache directory
 
   $ USERCACHE=`pwd`/cache; export USERCACHE
   $ cat <<EOF >> ${HGRCPATH}
   > [extensions]
   > hgext.largefiles=
   > [largefiles]
   > usercache=${USERCACHE}
   > EOF
   $ mkdir -p ${USERCACHE}
 
 Create source repo, and commit adding largefile.
 
   $ hg init src
   $ cd src
   $ echo large > large
   $ hg add --large large
   $ hg commit -m 'add largefile'
   $ cd ..
 
 Discard all cached largefiles in USERCACHE
 
   $ rm -rf ${USERCACHE}
 
 Create mirror repo, and pull from source without largefile:
 "pull" is used instead of "clone" for suppression of (1) updating to
 tip (= cahcing largefile from source repo), and (2) recording source
 repo as "default" path in .hg/hgrc.
 
   $ hg init mirror
   $ cd mirror
   $ hg pull ../src
   pulling from ../src
   requesting all changes
   adding changesets
   adding manifests
   adding file changes
   added 1 changesets with 1 changes to 1 files
   (run 'hg update' to get a working copy)
+  caching new largefiles
+  0 largefiles cached
 
 Update working directory to "tip", which requires largefile("large"),
 but there is no cache file for it.  So, hg must treat it as
 "missing"(!) file.
 
   $ hg update
   1 files updated, 0 files merged, 0 files removed, 0 files unresolved
   getting changed largefiles
   large: Can't get file locally
   (no default or default-push path set in hgrc)
   0 largefiles updated, 0 removed
   $ hg status
   ! large
 
 Update working directory to null: this cleanup .hg/largefiles/dirstate
 
   $ hg update null
   0 files updated, 0 files merged, 1 files removed, 0 files unresolved
   getting changed largefiles
   0 largefiles updated, 0 removed
 
 Update working directory to tip, again.
 
   $ hg update
   1 files updated, 0 files merged, 0 files removed, 0 files unresolved
   getting changed largefiles
   large: Can't get file locally
   (no default or default-push path set in hgrc)
   0 largefiles updated, 0 removed
   $ hg status
   ! large

tests/test-largefiles.t

@@ -1,923 +1,936 b''
   $ "$TESTDIR/hghave" symlink unix-permissions serve || exit 80
   $ USERCACHE=`pwd`/cache; export USERCACHE
   $ mkdir -p ${USERCACHE}
   $ cat >> $HGRCPATH <<EOF
   > [extensions]
   > largefiles=
   > purge=
   > rebase=
   > transplant=
   > [phases]
   > publish=False
   > [largefiles]
   > minsize=2
   > patterns=glob:**.dat
   > usercache=${USERCACHE}
   > [hooks]
   > precommit=echo "Invoking status precommit hook"; hg status
   > EOF
 
 Create the repo with a couple of revisions of both large and normal
 files, testing that status correctly shows largefiles and that summary output
 is correct.
 
   $ hg init a
   $ cd a
   $ mkdir sub
   $ echo normal1 > normal1
   $ echo normal2 > sub/normal2
   $ echo large1 > large1
   $ echo large2 > sub/large2
   $ hg add normal1 sub/normal2
   $ hg add --large large1 sub/large2
   $ hg commit -m "add files"
   Invoking status precommit hook
   A large1
   A normal1
   A sub/large2
   A sub/normal2
   $ echo normal11 > normal1
   $ echo normal22 > sub/normal2
   $ echo large11 > large1
   $ echo large22 > sub/large2
   $ hg commit -m "edit files"
   Invoking status precommit hook
   M large1
   M normal1
   M sub/large2
   M sub/normal2
   $ hg sum --large
   parent: 1:ce8896473775 tip
    edit files
   branch: default
   commit: (clean)
   update: (current)
   largefiles: No remote repo
 
 Commit preserved largefile contents.
 
   $ cat normal1
   normal11
   $ cat large1
   large11
   $ cat sub/normal2
   normal22
   $ cat sub/large2
   large22
 
 Remove both largefiles and normal files.
  
   $ hg remove normal1 large1
   $ hg commit -m "remove files"
   Invoking status precommit hook
   R large1
   R normal1
   $ ls
   sub
   $ echo "testlargefile" > large1-test
   $ hg add --large large1-test
   $ hg st
   A large1-test
   $ hg rm large1-test
   not removing large1-test: file has been marked for add (use forget to undo)
   $ hg st
   A large1-test
   $ hg forget large1-test
   $ hg st
   ? large1-test
   $ rm large1-test
 
 Copy both largefiles and normal files (testing that status output is correct).
 
   $ hg cp sub/normal2 normal1
   $ hg cp sub/large2 large1
   $ hg commit -m "copy files"
   Invoking status precommit hook
   A large1
   A normal1
   $ cat normal1
   normal22
   $ cat large1
   large22
 
 Test moving largefiles and verify that normal files are also unaffected.
 
   $ hg mv normal1 normal3
   $ hg mv large1 large3
   $ hg mv sub/normal2 sub/normal4
   $ hg mv sub/large2 sub/large4
   $ hg commit -m "move files"
   Invoking status precommit hook
   A large3
   A normal3
   A sub/large4
   A sub/normal4
   R large1
   R normal1
   R sub/large2
   R sub/normal2
   $ cat normal3
   normal22
   $ cat large3
   large22
   $ cat sub/normal4
   normal22
   $ cat sub/large4
   large22
 
 Test archiving the various revisions.  These hit corner cases known with
 archiving.
 
   $ hg archive -r 0 ../archive0
   $ hg archive -r 1 ../archive1
   $ hg archive -r 2 ../archive2
   $ hg archive -r 3 ../archive3
   $ hg archive -r 4 ../archive4
   $ cd ../archive0
   $ cat normal1 
   normal1
   $ cat large1
   large1
   $ cat sub/normal2
   normal2
   $ cat sub/large2
   large2
   $ cd ../archive1
   $ cat normal1
   normal11
   $ cat large1
   large11
   $ cat sub/normal2
   normal22
   $ cat sub/large2
   large22
   $ cd ../archive2
   $ ls
   sub
   $ cat sub/normal2
   normal22
   $ cat sub/large2
   large22
   $ cd ../archive3
   $ cat normal1
   normal22
   $ cat large1
   large22
   $ cat sub/normal2
   normal22
   $ cat sub/large2
   large22
   $ cd ../archive4
   $ cat normal3
   normal22
   $ cat large3
   large22
   $ cat sub/normal4
   normal22
   $ cat sub/large4
   large22
 
 Commit corner case: specify files to commit.
 
   $ cd ../a
   $ echo normal3 > normal3
   $ echo large3 > large3
   $ echo normal4 > sub/normal4
   $ echo large4 > sub/large4
   $ hg commit normal3 large3 sub/normal4 sub/large4 -m "edit files again"
   Invoking status precommit hook
   M large3
   M normal3
   M sub/large4
   M sub/normal4
   $ cat normal3
   normal3
   $ cat large3
   large3
   $ cat sub/normal4
   normal4
   $ cat sub/large4
   large4
 
 One more commit corner case: commit from a subdirectory.
 
   $ cd ../a
   $ echo normal33 > normal3
   $ echo large33 > large3
   $ echo normal44 > sub/normal4
   $ echo large44 > sub/large4
   $ cd sub
   $ hg commit -m "edit files yet again"
   Invoking status precommit hook
   M large3
   M normal3
   M sub/large4
   M sub/normal4
   $ cat ../normal3
   normal33
   $ cat ../large3
   large33
   $ cat normal4
   normal44
   $ cat large4
   large44
 
 Committing standins is not allowed.
 
   $ cd ..
   $ echo large3 > large3
   $ hg commit .hglf/large3 -m "try to commit standin"
   abort: file ".hglf/large3" is a largefile standin
   (commit the largefile itself instead)
   [255]
 
 Corner cases for adding largefiles.
 
   $ echo large5 > large5
   $ hg add --large large5
   $ hg add --large large5
   large5 already a largefile
   $ mkdir sub2
   $ echo large6 > sub2/large6
   $ echo large7 > sub2/large7
   $ hg add --large sub2
   adding sub2/large6 as a largefile (glob)
   adding sub2/large7 as a largefile (glob)
   $ hg st
   M large3
   A large5
   A sub2/large6
   A sub2/large7
 
 Config settings (pattern **.dat, minsize 2 MB) are respected.
 
   $ echo testdata > test.dat
   $ dd bs=1k count=2k if=/dev/zero of=reallylarge > /dev/null 2> /dev/null
   $ hg add
   adding reallylarge as a largefile
   adding test.dat as a largefile
 
 Test that minsize and --lfsize handle float values;
 also tests that --lfsize overrides largefiles.minsize.
 (0.250 MB = 256 kB = 262144 B)
 
   $ dd if=/dev/zero of=ratherlarge bs=1024 count=256 > /dev/null 2> /dev/null
   $ dd if=/dev/zero of=medium bs=1024 count=128 > /dev/null 2> /dev/null
   $ hg --config largefiles.minsize=.25 add
   adding ratherlarge as a largefile
   adding medium
   $ hg forget medium
   $ hg --config largefiles.minsize=.25 add --lfsize=.125
   adding medium as a largefile
   $ dd if=/dev/zero of=notlarge bs=1024 count=127 > /dev/null 2> /dev/null
   $ hg --config largefiles.minsize=.25 add --lfsize=.125
   adding notlarge
   $ hg forget notlarge
 
 Test forget on largefiles.
 
   $ hg forget large3 large5 test.dat reallylarge ratherlarge medium
   $ hg commit -m "add/edit more largefiles"
   Invoking status precommit hook
   A sub2/large6
   A sub2/large7
   R large3
   ? large5
   ? medium
   ? notlarge
   ? ratherlarge
   ? reallylarge
   ? test.dat
   $ hg st
   ? large3
   ? large5
   ? medium
   ? notlarge
   ? ratherlarge
   ? reallylarge
   ? test.dat
 
 Purge with largefiles: verify that largefiles are still in the working
 dir after a purge.
 
   $ hg purge --all
   $ cat sub/large4
   large44
   $ cat sub2/large6
   large6
   $ cat sub2/large7
   large7
 
 Test addremove: verify that files that should be added as largfiles are added as
 such and that already-existing largfiles are not added as normal files by
 accident.
 
   $ rm normal3
   $ rm sub/large4
   $ echo "testing addremove with patterns" > testaddremove.dat
   $ echo "normaladdremove" > normaladdremove
   $ hg addremove
   removing sub/large4
   adding testaddremove.dat as a largefile
   removing normal3
   adding normaladdremove
 
 Clone a largefiles repo.
 
   $ hg clone . ../b
   updating to branch default
   5 files updated, 0 files merged, 0 files removed, 0 files unresolved
   getting changed largefiles
   3 largefiles updated, 0 removed
   $ cd ../b
   $ hg log --template '{rev}:{node|short}  {desc|firstline}\n'
   7:daea875e9014  add/edit more largefiles
   6:4355d653f84f  edit files yet again
   5:9d5af5072dbd  edit files again
   4:74c02385b94c  move files
   3:9e8fbc4bce62  copy files
   2:51a0ae4d5864  remove files
   1:ce8896473775  edit files
   0:30d30fe6a5be  add files
   $ cat normal3
   normal33
   $ cat sub/normal4
   normal44
   $ cat sub/large4
   large44
   $ cat sub2/large6
   large6
   $ cat sub2/large7
   large7
   $ cd ..
   $ hg clone a -r 3 c
   adding changesets
   adding manifests
   adding file changes
   added 4 changesets with 10 changes to 4 files
   updating to branch default
   4 files updated, 0 files merged, 0 files removed, 0 files unresolved
   getting changed largefiles
   2 largefiles updated, 0 removed
   $ cd c
   $ hg log --template '{rev}:{node|short}  {desc|firstline}\n'
   3:9e8fbc4bce62  copy files
   2:51a0ae4d5864  remove files
   1:ce8896473775  edit files
   0:30d30fe6a5be  add files
   $ cat normal1
   normal22
   $ cat large1
   large22
   $ cat sub/normal2
   normal22
   $ cat sub/large2
   large22
 
 Old revisions of a clone have correct largefiles content (this also
 tests update).
 
   $ hg update -r 1 
   2 files updated, 0 files merged, 0 files removed, 0 files unresolved
   getting changed largefiles
   1 largefiles updated, 0 removed
   $ cat large1
   large11
   $ cat sub/large2
   large22
 
 Rebasing between two repositories does not revert largefiles to old
 revisions (this was a very bad bug that took a lot of work to fix).
 
   $ cd ..
   $ hg clone a d
   updating to branch default
   5 files updated, 0 files merged, 0 files removed, 0 files unresolved
   getting changed largefiles
   3 largefiles updated, 0 removed
   $ cd b
   $ echo large4-modified > sub/large4
   $ echo normal3-modified > normal3
   $ hg commit -m "modify normal file and largefile in repo b"
   Invoking status precommit hook
   M normal3
   M sub/large4
   $ cd ../d
   $ echo large6-modified > sub2/large6
   $ echo normal4-modified > sub/normal4
   $ hg commit -m "modify normal file largefile in repo d"
   Invoking status precommit hook
   M sub/normal4
   M sub2/large6
   $ cd ..
   $ hg clone d e
   updating to branch default
   5 files updated, 0 files merged, 0 files removed, 0 files unresolved
   getting changed largefiles
   3 largefiles updated, 0 removed
   $ cd d
   $ hg pull --rebase ../b
   pulling from ../b
   searching for changes
   adding changesets
   adding manifests
   adding file changes
   added 1 changesets with 2 changes to 2 files (+1 heads)
   getting changed largefiles
   1 largefiles updated, 0 removed
   Invoking status precommit hook
   M sub/normal4
   M sub2/large6
   saved backup bundle to $TESTTMP/d/.hg/strip-backup/f574fb32bb45-backup.hg
   nothing to rebase
   $ hg log --template '{rev}:{node|short}  {desc|firstline}\n'
   9:598410d3eb9a  modify normal file largefile in repo d
   8:a381d2c8c80e  modify normal file and largefile in repo b
   7:daea875e9014  add/edit more largefiles
   6:4355d653f84f  edit files yet again
   5:9d5af5072dbd  edit files again
   4:74c02385b94c  move files
   3:9e8fbc4bce62  copy files
   2:51a0ae4d5864  remove files
   1:ce8896473775  edit files
   0:30d30fe6a5be  add files
   $ cat normal3
   normal3-modified
   $ cat sub/normal4
   normal4-modified
   $ cat sub/large4
   large4-modified
   $ cat sub2/large6
   large6-modified
   $ cat sub2/large7
   large7
   $ cd ../e
   $ hg pull ../b
   pulling from ../b
   searching for changes
   adding changesets
   adding manifests
   adding file changes
   added 1 changesets with 2 changes to 2 files (+1 heads)
   (run 'hg heads' to see heads, 'hg merge' to merge)
+  caching new largefiles
+  0 largefiles cached
   $ hg rebase
   getting changed largefiles
   1 largefiles updated, 0 removed
   Invoking status precommit hook
   M sub/normal4
   M sub2/large6
   saved backup bundle to $TESTTMP/e/.hg/strip-backup/f574fb32bb45-backup.hg
   $ hg log --template '{rev}:{node|short}  {desc|firstline}\n'
   9:598410d3eb9a  modify normal file largefile in repo d
   8:a381d2c8c80e  modify normal file and largefile in repo b
   7:daea875e9014  add/edit more largefiles
   6:4355d653f84f  edit files yet again
   5:9d5af5072dbd  edit files again
   4:74c02385b94c  move files
   3:9e8fbc4bce62  copy files
   2:51a0ae4d5864  remove files
   1:ce8896473775  edit files
   0:30d30fe6a5be  add files
   $ cat normal3
   normal3-modified
   $ cat sub/normal4
   normal4-modified
   $ cat sub/large4
   large4-modified
   $ cat sub2/large6
   large6-modified
   $ cat sub2/large7
   large7
 
 Rollback on largefiles.
 
   $ echo large4-modified-again > sub/large4 
   $ hg commit -m "Modify large4 again"
   Invoking status precommit hook
   M sub/large4
   $ hg rollback
   repository tip rolled back to revision 9 (undo commit)
   working directory now based on revision 9
   $ hg st
   M sub/large4
   $ hg log --template '{rev}:{node|short}  {desc|firstline}\n'
   9:598410d3eb9a  modify normal file largefile in repo d
   8:a381d2c8c80e  modify normal file and largefile in repo b
   7:daea875e9014  add/edit more largefiles
   6:4355d653f84f  edit files yet again
   5:9d5af5072dbd  edit files again
   4:74c02385b94c  move files
   3:9e8fbc4bce62  copy files
   2:51a0ae4d5864  remove files
   1:ce8896473775  edit files
   0:30d30fe6a5be  add files
   $ cat sub/large4
   large4-modified-again
 
 "update --check" refuses to update with uncommitted changes.
   $ hg update --check 8
   abort: uncommitted local changes
   [255]
 
 "update --clean" leaves correct largefiles in working copy.
 
   $ hg update --clean 
   0 files updated, 0 files merged, 0 files removed, 0 files unresolved
   getting changed largefiles
   1 largefiles updated, 0 removed
   $ cat normal3
   normal3-modified
   $ cat sub/normal4
   normal4-modified
   $ cat sub/large4
   large4-modified
   $ cat sub2/large6
   large6-modified
   $ cat sub2/large7
   large7
 
 Now "update check" is happy.
   $ hg update --check 8
   2 files updated, 0 files merged, 0 files removed, 0 files unresolved
   getting changed largefiles
   1 largefiles updated, 0 removed
   $ hg update --check
   2 files updated, 0 files merged, 0 files removed, 0 files unresolved
   getting changed largefiles
   1 largefiles updated, 0 removed
 
 Test removing empty largefiles directories on update
   $ test -d sub2 && echo "sub2 exists"
   sub2 exists
   $ hg update -q null
   $ test -d sub2 && echo "error: sub2 should not exist anymore"
   [1]
   $ hg update -q
 
 "revert" works on largefiles (and normal files too).
   $ echo hack3 >> normal3
   $ echo hack4 >> sub/normal4
   $ echo hack4 >> sub/large4
   $ hg rm sub2/large6
   $ echo new >> sub2/large8
   $ hg add --large sub2/large8
 # XXX we don't really want to report that we're reverting the standin;
 # that's just an implementation detail. But I don't see an obvious fix. ;-(
   $ hg revert sub
   reverting .hglf/sub/large4 (glob)
   reverting sub/normal4 (glob)
   $ hg status
   M normal3
   A sub2/large8
   R sub2/large6
   ? sub/large4.orig
   ? sub/normal4.orig
   $ cat sub/normal4
   normal4-modified
   $ cat sub/large4
   large4-modified
   $ hg revert -a --no-backup
   undeleting .hglf/sub2/large6 (glob)
   forgetting .hglf/sub2/large8 (glob)
   reverting normal3
   $ hg status
   ? sub/large4.orig
   ? sub/normal4.orig
   ? sub2/large8
   $ cat normal3
   normal3-modified
   $ cat sub2/large6
   large6-modified
   $ rm sub/*.orig sub2/large8
 
 revert some files to an older revision
   $ hg revert --no-backup -r 8 sub2
   reverting .hglf/sub2/large6 (glob)
   $ cat sub2/large6
   large6
   $ hg revert --no-backup sub2
   reverting .hglf/sub2/large6 (glob)
   $ hg status
 
 "verify --large" actually verifies largefiles
 
   $ hg verify --large
   checking changesets
   checking manifests
   crosschecking files in changesets and manifests
   checking files
   10 files, 10 changesets, 28 total revisions
   searching 1 changesets for largefiles
   verified existence of 3 revisions of 3 largefiles
 
-Merging does not revert to old versions of largefiles (this has also
-been very problematic).
+Merging does not revert to old versions of largefiles and also check
+that merging after having pulled from a non-default remote works
+correctly.
 
   $ cd ..
-  $ hg clone -r 7 e f
+  $ hg clone -r 7 e temp
   adding changesets
   adding manifests
   adding file changes
   added 8 changesets with 24 changes to 10 files
   updating to branch default
   5 files updated, 0 files merged, 0 files removed, 0 files unresolved
   getting changed largefiles
   3 largefiles updated, 0 removed
+  $ hg clone temp f
+  updating to branch default
+  5 files updated, 0 files merged, 0 files removed, 0 files unresolved
+  getting changed largefiles
+  3 largefiles updated, 0 removed
+# Delete the largefiles in the largefiles system cache so that we have an
+# opportunity to test that caching after a pull works.
+  $ rm ${USERCACHE}/*
   $ cd f
   $ echo "large4-merge-test" > sub/large4
   $ hg commit -m "Modify large4 to test merge"
   Invoking status precommit hook
   M sub/large4
   $ hg pull ../e
   pulling from ../e
   searching for changes
   adding changesets
   adding manifests
   adding file changes
   added 2 changesets with 4 changes to 4 files (+1 heads)
   (run 'hg heads' to see heads, 'hg merge' to merge)
+  caching new largefiles
+  2 largefiles cached
   $ hg merge
   merging sub/large4
   largefile sub/large4 has a merge conflict
   keep (l)ocal or take (o)ther? l
   3 files updated, 1 files merged, 0 files removed, 0 files unresolved
   (branch merge, don't forget to commit)
   getting changed largefiles
   1 largefiles updated, 0 removed
   $ hg commit -m "Merge repos e and f"
   Invoking status precommit hook
   M normal3
   M sub/normal4
   M sub2/large6
   $ cat normal3
   normal3-modified
   $ cat sub/normal4
   normal4-modified
   $ cat sub/large4
   large4-merge-test
   $ cat sub2/large6
   large6-modified
   $ cat sub2/large7
   large7
 
 Test status after merging with a branch that introduces a new largefile:
 
   $ echo large > large
   $ hg add --large large
   $ hg commit -m 'add largefile'
   Invoking status precommit hook
   A large
   $ hg update -q ".^"
   $ echo change >> normal3
   $ hg commit -m 'some change'
   Invoking status precommit hook
   M normal3
   created new head
   $ hg merge
   1 files updated, 0 files merged, 0 files removed, 0 files unresolved
   (branch merge, don't forget to commit)
   getting changed largefiles
   1 largefiles updated, 0 removed
   $ hg status
   M large
 
 Test that a normal file and a largefile with the same name and path cannot
 coexist.
 
   $ rm sub2/large7
   $ echo "largeasnormal" > sub2/large7
   $ hg add sub2/large7
   sub2/large7 already a largefile
 
 Test that transplanting a largefile change works correctly.
 
   $ cd ..
   $ hg clone -r 8 d g
   adding changesets
   adding manifests
   adding file changes
   added 9 changesets with 26 changes to 10 files
   updating to branch default
   5 files updated, 0 files merged, 0 files removed, 0 files unresolved
   getting changed largefiles
   3 largefiles updated, 0 removed
   $ cd g
   $ hg transplant -s ../d 598410d3eb9a 
   searching for changes
   searching for changes
   adding changesets
   adding manifests
   adding file changes
   added 1 changesets with 2 changes to 2 files
   getting changed largefiles
   1 largefiles updated, 0 removed
   $ hg log --template '{rev}:{node|short}  {desc|firstline}\n'
   9:598410d3eb9a  modify normal file largefile in repo d
   8:a381d2c8c80e  modify normal file and largefile in repo b
   7:daea875e9014  add/edit more largefiles
   6:4355d653f84f  edit files yet again
   5:9d5af5072dbd  edit files again
   4:74c02385b94c  move files
   3:9e8fbc4bce62  copy files
   2:51a0ae4d5864  remove files
   1:ce8896473775  edit files
   0:30d30fe6a5be  add files
   $ cat normal3
   normal3-modified
   $ cat sub/normal4
   normal4-modified
   $ cat sub/large4
   large4-modified
   $ cat sub2/large6
   large6-modified
   $ cat sub2/large7
   large7
 
 Test that renaming a largefile results in correct output for status
 
   $ hg rename sub/large4 large4-renamed
   $ hg commit -m "test rename output"
   Invoking status precommit hook
   A large4-renamed
   R sub/large4
   $ cat large4-renamed
   large4-modified
   $ cd sub2
   $ hg rename large6 large6-renamed
   $ hg st
   A sub2/large6-renamed
   R sub2/large6
   $ cd ../..
 
 vanilla clients not locked out from largefiles servers on vanilla repos
   $ mkdir r1
   $ cd r1
   $ hg init
   $ echo c1 > f1
   $ hg add f1
   $ hg commit -m "m1"
   Invoking status precommit hook
   A f1
   $ cd ..
   $ hg serve -R r1 -d -p $HGPORT --pid-file hg.pid
   $ cat hg.pid >> $DAEMON_PIDS
   $ hg --config extensions.largefiles=! clone http://localhost:$HGPORT r2
   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
 
 largefiles clients still work with vanilla servers
   $ hg --config extensions.largefiles=! serve -R r1 -d -p $HGPORT1 --pid-file hg.pid
   $ cat hg.pid >> $DAEMON_PIDS
   $ hg clone http://localhost:$HGPORT1 r3
   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
 
 vanilla clients locked out from largefiles http repos
   $ mkdir r4
   $ cd r4
   $ hg init
   $ echo c1 > f1
   $ hg add --large f1
   $ hg commit -m "m1"
   Invoking status precommit hook
   A f1
   $ cd ..
   $ hg serve -R r4 -d -p $HGPORT2 --pid-file hg.pid
   $ cat hg.pid >> $DAEMON_PIDS
   $ hg --config extensions.largefiles=! clone http://localhost:$HGPORT2 r5
   abort: remote error:
   
   This repository uses the largefiles extension.
   
   Please enable it in your Mercurial config file.
   [255]
 
 used all HGPORTs, kill all daemons
   $ "$TESTDIR/killdaemons.py"
 
 vanilla clients locked out from largefiles ssh repos
   $ hg --config extensions.largefiles=! clone -e "python $TESTDIR/dummyssh" ssh://user@dummy/r4 r5
   abort: remote error:
   
   This repository uses the largefiles extension.
   
   Please enable it in your Mercurial config file.
   [255]
 
 largefiles clients refuse to push largefiles repos to vanilla servers
   $ mkdir r6
   $ cd r6
   $ hg init
   $ echo c1 > f1
   $ hg add f1
   $ hg commit -m "m1"
   Invoking status precommit hook
   A f1
   $ cat >> .hg/hgrc <<!
   > [web]
   > push_ssl = false
   > allow_push = *
   > !
   $ cd ..
   $ hg clone r6 r7
   updating to branch default
   1 files updated, 0 files merged, 0 files removed, 0 files unresolved
   $ cd r7
   $ echo c2 > f2
   $ hg add --large f2
   $ hg commit -m "m2"
   Invoking status precommit hook
   A f2
   $ hg --config extensions.largefiles=! -R ../r6 serve -d -p $HGPORT --pid-file ../hg.pid
   $ cat ../hg.pid >> $DAEMON_PIDS
   $ hg push http://localhost:$HGPORT
   pushing to http://localhost:$HGPORT/
   searching for changes
   abort: http://localhost:$HGPORT/ does not appear to be a largefile store
   [255]
   $ cd ..
 
 putlfile errors are shown (issue3123)
 Corrupt the cached largefile in r7
   $ echo corruption > $USERCACHE/4cdac4d8b084d0b599525cf732437fb337d422a8
   $ hg init empty
   $ hg serve -R empty -d -p $HGPORT1 --pid-file hg.pid \
   >   --config 'web.allow_push=*' --config web.push_ssl=False
   $ cat hg.pid >> $DAEMON_PIDS
   $ hg push -R r7 http://localhost:$HGPORT1
   pushing to http://localhost:$HGPORT1/
   searching for changes
   remote: largefiles: failed to put 4cdac4d8b084d0b599525cf732437fb337d422a8 into store: largefile contents do not match hash
   abort: remotestore: could not put $TESTTMP/r7/.hg/largefiles/4cdac4d8b084d0b599525cf732437fb337d422a8 to remote store http://localhost:$HGPORT1/
   [255]
   $ rm -rf empty
 
 Clone a local repository owned by another user
 We have to simulate that here by setting $HOME and removing write permissions
   $ ORIGHOME="$HOME"
   $ mkdir alice
   $ HOME="`pwd`/alice"
   $ cd alice
   $ hg init pubrepo
   $ cd pubrepo
   $ dd if=/dev/urandom bs=1k count=11k > a-large-file 2> /dev/null
   $ hg add --large a-large-file
   $ hg commit -m "Add a large file"
   Invoking status precommit hook
   A a-large-file
   $ cd ..
   $ chmod -R a-w pubrepo
   $ cd ..
   $ mkdir bob
   $ HOME="`pwd`/bob"
   $ cd bob
   $ hg clone --pull ../alice/pubrepo pubrepo
   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
   1 largefiles updated, 0 removed
   $ cd ..
   $ chmod -R u+w alice/pubrepo
   $ HOME="$ORIGHOME"
 
 Symlink to a large largefile should behave the same as a symlink to a normal file
   $ hg init largesymlink
   $ cd largesymlink
   $ dd if=/dev/zero bs=1k count=10k of=largefile 2>/dev/null
   $ hg add --large largefile
   $ hg commit -m "commit a large file"
   Invoking status precommit hook
   A largefile
   $ ln -s largefile largelink
   $ hg add largelink
   $ hg commit -m "commit a large symlink"
   Invoking status precommit hook
   A largelink
   $ rm -f largelink
   $ hg up >/dev/null
   $ test -f largelink
   [1]
   $ test -L largelink
   [1]
   $ rm -f largelink # make next part of the test independent of the previous
   $ hg up -C >/dev/null
   $ test -f largelink
   $ test -L largelink
   $ cd ..