hgext/largefiles/overrides.py

phases: rework the logic of _pushdiscoveryphase to bound complexity

This rework the various graph traversal in _pushdiscoveryphase to keep the
complexity in check.

This is done though a couple of things:

- first, limiting the space we have to explore, for example, if we are not in
  publishing push, we don't need to consider remote draft roots that are also
  draft locally, as there is nothing to be moved there.

- avoid unbounded descendant computation, and use the faster "rev between"
  computation.

This provide a massive boost to performance when exchanging with repository with
a massive amount of draft, like mozilla-try:

### data-env-vars.name                             = mozilla-try-2023-03-22-zstd-sparse-revlog
  # benchmark.name                                 = hg.command.push
  # bin-env-vars.hg.flavor                         = default
  # bin-env-vars.hg.py-re2-module                  = default
  # benchmark.variants.explicit-rev                = all-out-heads
  # benchmark.variants.issue6528                   = disabled
  # benchmark.variants.protocol                    = ssh
  # benchmark.variants.reuse-external-delta-parent = default

 ## benchmark.variants.revs                        = any-1-extra-rev
before: 20.346590 seconds
after:  11.232059 seconds (-38.15%, -7.48 seconds)
 ## benchmark.variants.revs                        = any-100-extra-rev
before: 24.752051 seconds
after:  15.367412 seconds (-37.91%, -9.38 seconds)


After this changes, the push operation is still quite too slow. Some of this
can be attributed to general phases slowness (reading all the roots from disk
for example) and other know slowness (not using persistent-nodemap, branchmap,
tags, etc. We are also working on them, but with this series, phase discovery
during push no longer showing up in profile and this is a pretty nice and bit
low-hanging fruit out of the way.

### (same case as the above)
  # benchmark.variants.revs                        = any-1-extra-rev
pre-%ln-change: 44.235070
this-changeset: 11.232059 seconds (-74.61%, -33.00 seconds)
  # benchmark.variants.revs                        = any-100-extra-rev
pre-%ln-change: 49.234697
this-changeset: 15.367412 seconds (-68.79%, -33.87 seconds)


Note that with this change, the `hg push` performance is now much closer to the
`hg pull` performance, even it still lagging behind a bit. (and the overall
performance are still too slow).

### data-env-vars.name                           = mozilla-try-2023-03-22-ds2-pnm
  # benchmark.variants.explicit-rev              = all-out-heads
  # benchmark.variants.issue6528                 = disabled
  # benchmark.variants.protocol                  = ssh
  # benchmark.variants.pulled-delta-reuse-policy = default
  # bin-env-vars.hg.flavor                       = rust
 ## benchmark.variants.revs                      = any-1-extra-rev
hg.command.pull: 6.517450
hg.command.push: 11.219888
 ## benchmark.variants.revs                      = any-100-extra-rev
hg.command.pull: 10.160991
hg.command.push: 14.251107

### data-env-vars.name                           = mozilla-try-2023-03-22-zstd-sparse-revlog
  # bin-env-vars.hg.py-re2-module                = default
  # benchmark.variants.explicit-rev              = all-out-heads
  # benchmark.variants.issue6528                 = disabled
  # benchmark.variants.protocol                  = ssh
  # benchmark.variants.pulled-delta-reuse-policy = default
 ## bin-env-vars.hg.flavor                       = default
 ## benchmark.variants.revs                      = any-1-extra-rev
hg.command.pull: 8.577772
hg.command.push: 11.232059
 ## bin-env-vars.hg.flavor                       = default
 ## benchmark.variants.revs                      = any-100-extra-rev
hg.command.pull: 13.152976
hg.command.push: 15.367412
 ## bin-env-vars.hg.flavor                       = rust
 ## benchmark.variants.revs                      = any-1-extra-rev
hg.command.pull: 8.731982
hg.command.push: 11.178751
 ## bin-env-vars.hg.flavor                       = rust
 ## benchmark.variants.revs                      = any-100-extra-rev
hg.command.pull: 13.184236
hg.command.push: 15.620843

# 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 contextlib
import copy
import os

from mercurial.i18n import _

from mercurial.pycompat import open

from mercurial.hgweb import webcommands

from mercurial import (
    archival,
    cmdutil,
    copies as copiesmod,
    dirstate,
    error,
    exchange,
    extensions,
    exthelper,
    filemerge,
    hg,
    logcmdutil,
    match as matchmod,
    merge,
    mergestate as mergestatemod,
    pathutil,
    pycompat,
    scmutil,
    smartset,
    subrepo,
    url as urlmod,
    util,
)

from mercurial.upgrade_utils import (
    actions as upgrade_actions,
)

from . import (
    lfcommands,
    lfutil,
    storefactory,
)

ACTION_ADD = mergestatemod.ACTION_ADD
ACTION_DELETED_CHANGED = mergestatemod.ACTION_DELETED_CHANGED
ACTION_GET = mergestatemod.ACTION_GET
ACTION_KEEP = mergestatemod.ACTION_KEEP
ACTION_REMOVE = mergestatemod.ACTION_REMOVE

eh = exthelper.exthelper()

lfstatus = lfutil.lfstatus

MERGE_ACTION_LARGEFILE_MARK_REMOVED = mergestatemod.MergeAction('lfmr')

# -- Utility functions: commonly/repeatedly needed functionality ---------------


def composelargefilematcher(match, manifest):
    """create a matcher that matches only the largefiles in the original
    matcher"""
    m = copy.copy(match)
    lfile = lambda f: lfutil.standin(f) in manifest
    m._files = [lf for lf in m._files if lfile(lf)]
    m._fileset = set(m._files)
    m.always = lambda: False
    origmatchfn = m.matchfn
    m.matchfn = lambda f: lfile(f) and origmatchfn(f)
    return m


def composenormalfilematcher(match, manifest, exclude=None):
    excluded = set()
    if exclude is not None:
        excluded.update(exclude)

    m = copy.copy(match)
    notlfile = lambda f: not (
        lfutil.isstandin(f) or lfutil.standin(f) in manifest or f in excluded
    )
    m._files = [lf for lf in m._files if notlfile(lf)]
    m._fileset = set(m._files)
    m.always = lambda: False
    origmatchfn = m.matchfn
    m.matchfn = lambda f: notlfile(f) and origmatchfn(f)
    return m


def addlargefiles(ui, repo, isaddremove, matcher, uipathfn, **opts):
    large = opts.get('large')
    lfsize = lfutil.getminsize(
        ui, lfutil.islfilesrepo(repo), opts.get('lfsize')
    )

    lfmatcher = None
    if lfutil.islfilesrepo(repo):
        lfpats = ui.configlist(lfutil.longname, b'patterns')
        if lfpats:
            lfmatcher = matchmod.match(repo.root, b'', list(lfpats))

    lfnames = []
    m = matcher

    wctx = repo[None]
    for f in wctx.walk(matchmod.badmatch(m, lambda x, y: None)):
        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(_(b'%s already a largefile\n') % uipathfn(f))
            continue

        if (exact or not exists) and not lfutil.isstandin(f):
            # In case the file was removed previously, but not committed
            # (issue3507)
            if not repo.wvfs.exists(f):
                continue

            abovemin = (
                lfsize and repo.wvfs.lstat(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(_(b'adding %s as a largefile\n') % uipathfn(f))

    bad = []

    # Need to lock, otherwise there could be a race condition between
    # when standins are created and added to the repo.
    with repo.wlock():
        if not opts.get('dry_run'):
            standins = []
            lfdirstate = lfutil.openlfdirstate(ui, repo)
            for f in lfnames:
                standinname = lfutil.standin(f)
                lfutil.writestandin(
                    repo,
                    standinname,
                    hash=b'',
                    executable=lfutil.getexecutable(repo.wjoin(f)),
                )
                standins.append(standinname)
                lfdirstate.set_tracked(f)
            lfdirstate.write(repo.currenttransaction())
            bad += [
                lfutil.splitstandin(f)
                for f in repo[None].add(standins)
                if f in m.files()
            ]

        added = [f for f in lfnames if f not in bad]
    return added, bad


def removelargefiles(ui, repo, isaddremove, matcher, uipathfn, dryrun, **opts):
    after = opts.get('after')
    m = composelargefilematcher(matcher, repo[None].manifest())
    with lfstatus(repo):
        s = repo.status(match=m, clean=not isaddremove)
    manifest = repo[None].manifest()
    modified, added, deleted, clean = [
        [f for f in list if lfutil.standin(f) in manifest]
        for list in (s.modified, s.added, s.deleted, s.clean)
    ]

    def warn(files, msg):
        for f in files:
            ui.warn(msg % uipathfn(f))
        return int(len(files) > 0)

    if after:
        remove = deleted
        result = warn(
            modified + added + clean, _(b'not removing %s: file still exists\n')
        )
    else:
        remove = deleted + clean
        result = warn(
            modified,
            _(
                b'not removing %s: file is modified (use -f'
                b' to force removal)\n'
            ),
        )
        result = (
            warn(
                added,
                _(
                    b'not removing %s: file has been marked for add'
                    b' (use forget to undo)\n'
                ),
            )
            or result
        )

    # Need to lock because standin files are deleted then removed from the
    # repository and we could race in-between.
    with repo.wlock():
        lfdirstate = lfutil.openlfdirstate(ui, repo)
        for f in sorted(remove):
            if ui.verbose or not m.exact(f):
                ui.status(_(b'removing %s\n') % uipathfn(f))

            if not dryrun:
                if not after:
                    repo.wvfs.unlinkpath(f, ignoremissing=True)

        if dryrun:
            return result

        remove = [lfutil.standin(f) for f in remove]
        # If this is being called by addremove, let the original addremove
        # function handle this.
        if not isaddremove:
            for f in remove:
                repo.wvfs.unlinkpath(f, ignoremissing=True)
        repo[None].forget(remove)

        for f in remove:
            lfdirstate.set_untracked(lfutil.splitstandin(f))

        lfdirstate.write(repo.currenttransaction())

    return result


# For overriding mercurial.hgweb.webcommands so that largefiles will
# appear at their right place in the manifests.
@eh.wrapfunction(webcommands, 'decodepath')
def decodepath(orig, path):
    return lfutil.splitstandin(path) or path


# -- Wrappers: modify existing commands --------------------------------


@eh.wrapcommand(
    b'add',
    opts=[
        (b'', b'large', None, _(b'add as largefile')),
        (b'', b'normal', None, _(b'add as normal file')),
        (
            b'',
            b'lfsize',
            b'',
            _(
                b'add all files above this size (in megabytes) '
                b'as largefiles (default: 10)'
            ),
        ),
    ],
)
def overrideadd(orig, ui, repo, *pats, **opts):
    if opts.get('normal') and opts.get('large'):
        raise error.Abort(_(b'--normal cannot be used with --large'))
    return orig(ui, repo, *pats, **opts)


@eh.wrapfunction(cmdutil, 'add')
def cmdutiladd(orig, ui, repo, matcher, prefix, uipathfn, explicitonly, **opts):
    # The --normal flag short circuits this override
    if opts.get('normal'):
        return orig(ui, repo, matcher, prefix, uipathfn, explicitonly, **opts)

    ladded, lbad = addlargefiles(ui, repo, False, matcher, uipathfn, **opts)
    normalmatcher = composenormalfilematcher(
        matcher, repo[None].manifest(), ladded
    )
    bad = orig(ui, repo, normalmatcher, prefix, uipathfn, explicitonly, **opts)

    bad.extend(f for f in lbad)
    return bad


@eh.wrapfunction(cmdutil, 'remove')
def cmdutilremove(
    orig, ui, repo, matcher, prefix, uipathfn, after, force, subrepos, dryrun
):
    normalmatcher = composenormalfilematcher(matcher, repo[None].manifest())
    result = orig(
        ui,
        repo,
        normalmatcher,
        prefix,
        uipathfn,
        after,
        force,
        subrepos,
        dryrun,
    )
    return (
        removelargefiles(
            ui, repo, False, matcher, uipathfn, dryrun, after=after, force=force
        )
        or result
    )


@eh.wrapfunction(dirstate.dirstate, '_changing')
@contextlib.contextmanager
def _changing(orig, self, repo, change_type):
    pre = sub_dirstate = getattr(self, '_sub_dirstate', None)
    try:
        lfd = getattr(self, '_large_file_dirstate', False)
        if sub_dirstate is None and not lfd:
            sub_dirstate = lfutil.openlfdirstate(repo.ui, repo)
            self._sub_dirstate = sub_dirstate
        if not lfd:
            assert self._sub_dirstate is not None
        with orig(self, repo, change_type):
            if sub_dirstate is None:
                yield
            else:
                with sub_dirstate._changing(repo, change_type):
                    yield
    finally:
        self._sub_dirstate = pre


@eh.wrapfunction(dirstate.dirstate, 'running_status')
@contextlib.contextmanager
def running_status(orig, self, repo):
    pre = sub_dirstate = getattr(self, '_sub_dirstate', None)
    try:
        lfd = getattr(self, '_large_file_dirstate', False)
        if sub_dirstate is None and not lfd:
            sub_dirstate = lfutil.openlfdirstate(repo.ui, repo)
            self._sub_dirstate = sub_dirstate
        if not lfd:
            assert self._sub_dirstate is not None
        with orig(self, repo):
            if sub_dirstate is None:
                yield
            else:
                with sub_dirstate.running_status(repo):
                    yield
    finally:
        self._sub_dirstate = pre


@eh.wrapfunction(subrepo.hgsubrepo, 'status')
def overridestatusfn(orig, repo, rev2, **opts):
    with lfstatus(repo._repo):
        return orig(repo, rev2, **opts)


@eh.wrapcommand(b'status')
def overridestatus(orig, ui, repo, *pats, **opts):
    with lfstatus(repo):
        return orig(ui, repo, *pats, **opts)


@eh.wrapfunction(subrepo.hgsubrepo, 'dirty')
def overridedirty(orig, repo, ignoreupdate=False, missing=False):
    with lfstatus(repo._repo):
        return orig(repo, ignoreupdate=ignoreupdate, missing=missing)


@eh.wrapcommand(b'log')
def overridelog(orig, ui, repo, *pats, **opts):
    def overridematchandpats(
        orig,
        ctx,
        pats=(),
        opts=None,
        globbed=False,
        default=b'relpath',
        badfn=None,
    ):
        """Matcher that merges root directory with .hglf, suitable for log.
        It is still possible to match .hglf directly.
        For any listed files run log on the standin too.
        matchfn tries both the given filename and with .hglf stripped.
        """
        if opts is None:
            opts = {}
        matchandpats = orig(ctx, pats, opts, globbed, default, badfn=badfn)
        m, p = copy.copy(matchandpats)

        if m.always():
            # We want to match everything anyway, so there's no benefit trying
            # to add standins.
            return matchandpats

        pats = set(p)

        def fixpats(pat, tostandin=lfutil.standin):
            if pat.startswith(b'set:'):
                return pat

            kindpat = matchmod._patsplit(pat, None)

            if kindpat[0] is not None:
                return kindpat[0] + b':' + tostandin(kindpat[1])
            return tostandin(kindpat[1])

        cwd = repo.getcwd()
        if cwd:
            hglf = lfutil.shortname
            back = util.pconvert(repo.pathto(hglf)[: -len(hglf)])

            def tostandin(f):
                # The file may already be a standin, so truncate the back
                # prefix and test before mangling it.  This avoids turning
                # 'glob:../.hglf/foo*' into 'glob:../.hglf/../.hglf/foo*'.
                if f.startswith(back) and lfutil.splitstandin(f[len(back) :]):
                    return f

                # An absolute path is from outside the repo, so truncate the
                # path to the root before building the standin.  Otherwise cwd
                # is somewhere in the repo, relative to root, and needs to be
                # prepended before building the standin.
                if os.path.isabs(cwd):
                    f = f[len(back) :]
                else:
                    f = cwd + b'/' + f
                return back + lfutil.standin(f)

        else:

            def tostandin(f):
                if lfutil.isstandin(f):
                    return f
                return lfutil.standin(f)

        pats.update(fixpats(f, tostandin) for f in p)

        for i in range(0, len(m._files)):
            # Don't add '.hglf' to m.files, since that is already covered by '.'
            if m._files[i] == b'.':
                continue
            standin = lfutil.standin(m._files[i])
            # If the "standin" is a directory, append instead of replace to
            # support naming a directory on the command line with only
            # largefiles.  The original directory is kept to support normal
            # files.
            if standin in ctx:
                m._files[i] = standin
            elif m._files[i] not in ctx and repo.wvfs.isdir(standin):
                m._files.append(standin)

        m._fileset = set(m._files)
        m.always = lambda: False
        origmatchfn = m.matchfn

        def lfmatchfn(f):
            lf = lfutil.splitstandin(f)
            if lf is not None and origmatchfn(lf):
                return True
            r = origmatchfn(f)
            return r

        m.matchfn = lfmatchfn

        ui.debug(b'updated patterns: %s\n' % b', '.join(sorted(pats)))
        return m, pats

    # For hg log --patch, the match object is used in two different senses:
    # (1) to determine what revisions should be printed out, and
    # (2) to determine what files to print out diffs for.
    # The magic matchandpats override should be used for case (1) but not for
    # case (2).
    oldmatchandpats = scmutil.matchandpats

    def overridemakefilematcher(orig, repo, pats, opts, badfn=None):
        wctx = repo[None]
        match, pats = oldmatchandpats(wctx, pats, opts, badfn=badfn)
        return lambda ctx: match

    wrappedmatchandpats = extensions.wrappedfunction(
        scmutil, 'matchandpats', overridematchandpats
    )
    wrappedmakefilematcher = extensions.wrappedfunction(
        logcmdutil, '_makenofollowfilematcher', overridemakefilematcher
    )
    with wrappedmatchandpats, wrappedmakefilematcher:
        return orig(ui, repo, *pats, **opts)


@eh.wrapcommand(
    b'verify',
    opts=[
        (
            b'',
            b'large',
            None,
            _(b'verify that all largefiles in current revision exists'),
        ),
        (
            b'',
            b'lfa',
            None,
            _(b'verify largefiles in all revisions, not just current'),
        ),
        (
            b'',
            b'lfc',
            None,
            _(b'verify local largefile contents, not just existence'),
        ),
    ],
)
def overrideverify(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 or all or contents:
        result = result or lfcommands.verifylfiles(ui, repo, all, contents)
    return result


@eh.wrapcommand(
    b'debugstate',
    opts=[(b'', b'large', None, _(b'display largefiles dirstate'))],
)
def overridedebugstate(orig, ui, repo, *pats, **opts):
    large = opts.pop('large', False)
    if large:

        class fakerepo:
            dirstate = lfutil.openlfdirstate(ui, repo)

        orig(ui, fakerepo, *pats, **opts)
    else:
        orig(ui, repo, *pats, **opts)


# Before starting the manifest merge, merge.updates will call
# _checkunknownfile 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 calculateupdates function below.
@eh.wrapfunction(merge, '_checkunknownfile')
def overridecheckunknownfile(
    origfn, dirstate, wvfs, dircache, wctx, mctx, f, f2=None
):
    if lfutil.standin(dirstate.normalize(f)) in wctx:
        return False
    return origfn(dirstate, wvfs, dircache, wctx, mctx, f, f2)


# 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 calculateupdates 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.
@eh.wrapfunction(merge, 'calculateupdates')
def overridecalculateupdates(
    origfn, repo, p1, p2, pas, branchmerge, force, acceptremote, *args, **kwargs
):
    overwrite = force and not branchmerge
    mresult = origfn(
        repo, p1, p2, pas, branchmerge, force, acceptremote, *args, **kwargs
    )

    if overwrite:
        return mresult

    # Convert to dictionary with filename as key and action as value.
    lfiles = set()
    for f in mresult.files():
        splitstandin = lfutil.splitstandin(f)
        if splitstandin is not None and splitstandin in p1:
            lfiles.add(splitstandin)
        elif lfutil.standin(f) in p1:
            lfiles.add(f)

    for lfile in sorted(lfiles):
        standin = lfutil.standin(lfile)
        (lm, largs, lmsg) = mresult.getfile(lfile, (None, None, None))
        (sm, sargs, smsg) = mresult.getfile(standin, (None, None, None))

        if sm in (ACTION_GET, ACTION_DELETED_CHANGED) and lm != ACTION_REMOVE:
            if sm == ACTION_DELETED_CHANGED:
                f1, f2, fa, move, anc = sargs
                sargs = (p2[f2].flags(), False)
            # Case 1: normal file in the working copy, largefile in
            # the second parent
            usermsg = (
                _(
                    b'remote turned local normal file %s into a largefile\n'
                    b'use (l)argefile or keep (n)ormal file?'
                    b'$$ &Largefile $$ &Normal file'
                )
                % lfile
            )
            if repo.ui.promptchoice(usermsg, 0) == 0:  # pick remote largefile
                mresult.addfile(
                    lfile, ACTION_REMOVE, None, b'replaced by standin'
                )
                mresult.addfile(standin, ACTION_GET, sargs, b'replaces standin')
            else:  # keep local normal file
                mresult.addfile(lfile, ACTION_KEEP, None, b'replaces standin')
                if branchmerge:
                    mresult.addfile(
                        standin,
                        ACTION_KEEP,
                        None,
                        b'replaced by non-standin',
                    )
                else:
                    mresult.addfile(
                        standin,
                        ACTION_REMOVE,
                        None,
                        b'replaced by non-standin',
                    )
        if lm in (ACTION_GET, ACTION_DELETED_CHANGED) and sm != ACTION_REMOVE:
            if lm == ACTION_DELETED_CHANGED:
                f1, f2, fa, move, anc = largs
                largs = (p2[f2].flags(), False)
            # Case 2: largefile in the working copy, normal file in
            # the second parent
            usermsg = (
                _(
                    b'remote turned local largefile %s into a normal file\n'
                    b'keep (l)argefile or use (n)ormal file?'
                    b'$$ &Largefile $$ &Normal file'
                )
                % lfile
            )
            if repo.ui.promptchoice(usermsg, 0) == 0:  # keep local largefile
                if branchmerge:
                    # largefile can be restored from standin safely
                    mresult.addfile(
                        lfile,
                        ACTION_KEEP,
                        None,
                        b'replaced by standin',
                    )
                    mresult.addfile(
                        standin, ACTION_KEEP, None, b'replaces standin'
                    )
                else:
                    # "lfile" should be marked as "removed" without
                    # removal of itself
                    mresult.addfile(
                        lfile,
                        MERGE_ACTION_LARGEFILE_MARK_REMOVED,
                        None,
                        b'forget non-standin largefile',
                    )

                    # linear-merge should treat this largefile as 're-added'
                    mresult.addfile(standin, ACTION_ADD, None, b'keep standin')
            else:  # pick remote normal file
                mresult.addfile(lfile, ACTION_GET, largs, b'replaces standin')
                mresult.addfile(
                    standin,
                    ACTION_REMOVE,
                    None,
                    b'replaced by non-standin',
                )

    return mresult


@eh.wrapfunction(mergestatemod, 'recordupdates')
def mergerecordupdates(orig, repo, actions, branchmerge, getfiledata):
    if MERGE_ACTION_LARGEFILE_MARK_REMOVED in actions:
        lfdirstate = lfutil.openlfdirstate(repo.ui, repo)
        for lfile, args, msg in actions[MERGE_ACTION_LARGEFILE_MARK_REMOVED]:
            # this should be executed before 'orig', to execute 'remove'
            # before all other actions
            repo.dirstate.update_file(lfile, p1_tracked=True, wc_tracked=False)
            # make sure lfile doesn't get synclfdirstate'd as normal
            lfdirstate.update_file(lfile, p1_tracked=False, wc_tracked=True)

    return orig(repo, actions, branchmerge, getfiledata)


# Override filemerge to prompt the user about how they wish to merge
# largefiles. This will handle identical edits without prompting the user.
@eh.wrapfunction(filemerge, 'filemerge')
def overridefilemerge(
    origfn, repo, wctx, mynode, orig, fcd, fco, fca, labels=None
):
    if not lfutil.isstandin(orig) or fcd.isabsent() or fco.isabsent():
        return origfn(repo, wctx, mynode, orig, fcd, fco, fca, labels=labels)

    ahash = lfutil.readasstandin(fca).lower()
    dhash = lfutil.readasstandin(fcd).lower()
    ohash = lfutil.readasstandin(fco).lower()
    if (
        ohash != ahash
        and ohash != dhash
        and (
            dhash == ahash
            or repo.ui.promptchoice(
                _(
                    b'largefile %s has a merge conflict\nancestor was %s\n'
                    b'you can keep (l)ocal %s or take (o)ther %s.\n'
                    b'what do you want to do?'
                    b'$$ &Local $$ &Other'
                )
                % (lfutil.splitstandin(orig), ahash, dhash, ohash),
                0,
            )
            == 1
        )
    ):
        repo.wwrite(fcd.path(), fco.data(), fco.flags())
    return 0, False


@eh.wrapfunction(copiesmod, 'pathcopies')
def copiespathcopies(orig, ctx1, ctx2, match=None):
    copies = orig(ctx1, ctx2, match=match)
    updated = {}

    for k, v in copies.items():
        updated[lfutil.splitstandin(k) or k] = lfutil.splitstandin(v) or v

    return updated


# 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.
@eh.wrapfunction(cmdutil, 'copy')
def overridecopy(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)

    # 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
    manifest = repo[None].manifest()

    def normalfilesmatchfn(
        orig,
        ctx,
        pats=(),
        opts=None,
        globbed=False,
        default=b'relpath',
        badfn=None,
    ):
        if opts is None:
            opts = {}
        match = orig(ctx, pats, opts, globbed, default, badfn=badfn)
        return composenormalfilematcher(match, manifest)

    with extensions.wrappedfunction(scmutil, 'match', normalfilesmatchfn):
        try:
            result = orig(ui, repo, pats, opts, rename)
        except error.Abort as e:
            if e.message != _(b'no files to copy'):
                raise e
            else:
                nonormalfiles = True
            result = 0

    # 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

    def makestandin(relpath):
        path = pathutil.canonpath(repo.root, repo.getcwd(), relpath)
        return repo.wvfs.join(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))

    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()

        def overridematch(
            orig,
            ctx,
            pats=(),
            opts=None,
            globbed=False,
            default=b'relpath',
            badfn=None,
        ):
            if opts is None:
                opts = {}
            newpats = []
            # The patterns were previously mangled to add the standin
            # directory; we need to remove that now
            for pat in pats:
                if matchmod.patkind(pat) is None and lfutil.shortname in pat:
                    newpats.append(pat.replace(lfutil.shortname, b''))
                else:
                    newpats.append(pat)
            match = orig(ctx, newpats, opts, globbed, default, badfn=badfn)
            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._fileset = set(m._files)
            origmatchfn = m.matchfn

            def matchfn(f):
                lfile = lfutil.splitstandin(f)
                return (
                    lfile is not None
                    and (f in manifest)
                    and origmatchfn(lfile)
                    or None
                )

            m.matchfn = matchfn
            return m

        listpats = []
        for pat in pats:
            if matchmod.patkind(pat) is not None:
                listpats.append(pat)
            else:
                listpats.append(makestandin(pat))

        copiedfiles = []

        def overridecopyfile(orig, src, dest, *args, **kwargs):
            if lfutil.shortname in src and dest.startswith(
                repo.wjoin(lfutil.shortname)
            ):
                destlfile = dest.replace(lfutil.shortname, b'')
                if not opts[b'force'] and os.path.exists(destlfile):
                    raise IOError(
                        b'', _(b'destination largefile already exists')
                    )
            copiedfiles.append((src, dest))
            orig(src, dest, *args, **kwargs)

        with extensions.wrappedfunction(util, 'copyfile', overridecopyfile):
            with extensions.wrappedfunction(scmutil, 'match', overridematch):
                result += orig(ui, repo, listpats, opts, rename)

        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(b'')), b'')
                destlfile = dest.replace(repo.wjoin(lfutil.standin(b'')), b'')
                destlfiledir = repo.wvfs.dirname(repo.wjoin(destlfile)) or b'.'
                if not os.path.isdir(destlfiledir):
                    os.makedirs(destlfiledir)
                if rename:
                    os.rename(repo.wjoin(srclfile), repo.wjoin(destlfile))

                    # The file is gone, but this deletes any empty parent
                    # directories as a side-effect.
                    repo.wvfs.unlinkpath(srclfile, ignoremissing=True)
                    lfdirstate.set_untracked(srclfile)
                else:
                    util.copyfile(repo.wjoin(srclfile), repo.wjoin(destlfile))

                lfdirstate.set_tracked(destlfile)
        lfdirstate.write(repo.currenttransaction())
    except error.Abort as e:
        if e.message != _(b'no files to copy'):
            raise e
        else:
            nolfiles = True
    finally:
        wlock.release()

    if nolfiles and nonormalfiles:
        raise error.Abort(_(b'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.
@eh.wrapfunction(cmdutil, 'revert')
def overriderevert(orig, ui, repo, ctx, *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.
    with repo.wlock(), repo.dirstate.running_status(repo):
        lfdirstate = lfutil.openlfdirstate(ui, repo)
        s = lfutil.lfdirstatestatus(lfdirstate, repo)
        lfdirstate.write(repo.currenttransaction())
        for lfile in s.modified:
            lfutil.updatestandin(repo, lfile, lfutil.standin(lfile))
        for lfile in s.deleted:
            fstandin = lfutil.standin(lfile)
            if repo.wvfs.exists(fstandin):
                repo.wvfs.unlink(fstandin)

        oldstandins = lfutil.getstandinsstate(repo)

        def overridematch(
            orig,
            mctx,
            pats=(),
            opts=None,
            globbed=False,
            default=b'relpath',
            badfn=None,
        ):
            if opts is None:
                opts = {}
            match = orig(mctx, pats, opts, globbed, default, badfn=badfn)
            m = copy.copy(match)

            # revert supports recursing into subrepos, and though largefiles
            # currently doesn't work correctly in that case, this match is
            # called, so the lfdirstate above may not be the correct one for
            # this invocation of match.
            lfdirstate = lfutil.openlfdirstate(
                mctx.repo().ui, mctx.repo(), False
            )

            wctx = repo[None]
            matchfiles = []
            for f in m._files:
                standin = lfutil.standin(f)
                if standin in ctx or standin in mctx:
                    matchfiles.append(standin)
                elif standin in wctx or lfdirstate.get_entry(f).removed:
                    continue
                else:
                    matchfiles.append(f)
            m._files = matchfiles
            m._fileset = set(m._files)
            origmatchfn = m.matchfn

            def matchfn(f):
                lfile = lfutil.splitstandin(f)
                if lfile is not None:
                    return origmatchfn(lfile) and (f in ctx or f in mctx)
                return origmatchfn(f)

            m.matchfn = matchfn
            return m

        with extensions.wrappedfunction(scmutil, 'match', overridematch):
            orig(ui, repo, ctx, *pats, **opts)

        newstandins = lfutil.getstandinsstate(repo)
        filelist = lfutil.getlfilestoupdate(oldstandins, newstandins)
        # lfdirstate should be 'normallookup'-ed for updated files,
        # because reverting doesn't touch dirstate for 'normal' files
        # when target revision is explicitly specified: in such case,
        # 'n' and valid timestamp in dirstate doesn't ensure 'clean'
        # of target (standin) file.
        lfcommands.updatelfiles(
            ui, repo, filelist, printmessage=False, normallookup=True
        )


# after pulling changesets, we need to take some extra care to get
# largefiles updated remotely
@eh.wrapcommand(
    b'pull',
    opts=[
        (
            b'',
            b'all-largefiles',
            None,
            _(b'download all pulled versions of largefiles (DEPRECATED)'),
        ),
        (
            b'',
            b'lfrev',
            [],
            _(b'download largefiles for these revisions'),
            _(b'REV'),
        ),
    ],
)
def overridepull(orig, ui, repo, source=None, **opts):
    revsprepull = len(repo)
    if not source:
        source = b'default'
    repo.lfpullsource = source
    result = orig(ui, repo, source, **opts)
    revspostpull = len(repo)
    lfrevs = opts.get('lfrev', [])
    if opts.get('all_largefiles'):
        lfrevs.append(b'pulled()')
    if lfrevs and revspostpull > revsprepull:
        numcached = 0
        repo.firstpulled = revsprepull  # for pulled() revset expression
        try:
            for rev in logcmdutil.revrange(repo, lfrevs):
                ui.note(_(b'pulling largefiles for revision %d\n') % rev)
                (cached, missing) = lfcommands.cachelfiles(ui, repo, rev)
                numcached += len(cached)
        finally:
            del repo.firstpulled
        ui.status(_(b"%d largefiles cached\n") % numcached)
    return result


@eh.wrapcommand(
    b'push',
    opts=[
        (
            b'',
            b'lfrev',
            [],
            _(b'upload largefiles for these revisions'),
            _(b'REV'),
        )
    ],
)
def overridepush(orig, ui, repo, *args, **kwargs):
    """Override push command and store --lfrev parameters in opargs"""
    lfrevs = kwargs.pop('lfrev', None)
    if lfrevs:
        opargs = kwargs.setdefault('opargs', {})
        opargs[b'lfrevs'] = logcmdutil.revrange(repo, lfrevs)
    return orig(ui, repo, *args, **kwargs)


@eh.wrapfunction(exchange, 'pushoperation')
def exchangepushoperation(orig, *args, **kwargs):
    """Override pushoperation constructor and store lfrevs parameter"""
    lfrevs = kwargs.pop('lfrevs', None)
    pushop = orig(*args, **kwargs)
    pushop.lfrevs = lfrevs
    return pushop


@eh.revsetpredicate(b'pulled()')
def pulledrevsetsymbol(repo, subset, x):
    """Changesets that just has been pulled.

    Only available with largefiles from pull --lfrev expressions.

    .. container:: verbose

      Some examples:

      - pull largefiles for all new changesets::

          hg pull -lfrev "pulled()"

      - pull largefiles for all new branch heads::

          hg pull -lfrev "head(pulled()) and not closed()"

    """

    try:
        firstpulled = repo.firstpulled
    except AttributeError:
        raise error.Abort(_(b"pulled() only available in --lfrev"))
    return smartset.baseset([r for r in subset if r >= firstpulled])


@eh.wrapcommand(
    b'clone',
    opts=[
        (
            b'',
            b'all-largefiles',
            None,
            _(b'download all versions of all largefiles'),
        )
    ],
)
def overrideclone(orig, ui, source, dest=None, **opts):
    d = dest
    if d is None:
        d = hg.defaultdest(source)
    if opts.get('all_largefiles') and not hg.islocal(d):
        raise error.Abort(
            _(b'--all-largefiles is incompatible with non-local destination %s')
            % d
        )

    return orig(ui, source, dest, **opts)


@eh.wrapfunction(hg, 'clone')
def hgclone(orig, ui, opts, *args, **kwargs):
    result = orig(ui, opts, *args, **kwargs)

    if result is not None:
        sourcerepo, destrepo = result
        repo = destrepo.local()

        # When cloning to a remote repo (like through SSH), no repo is available
        # from the peer.   Therefore the largefiles can't be downloaded and the
        # hgrc can't be updated.
        if not repo:
            return result

        # Caching is implicitly limited to 'rev' option, since the dest repo was
        # truncated at that point.  The user may expect a download count with
        # this option, so attempt whether or not this is a largefile repo.
        if opts.get(b'all_largefiles'):
            success, missing = lfcommands.downloadlfiles(ui, repo)

            if missing != 0:
                return None

    return result


@eh.wrapcommand(b'rebase', extension=b'rebase')
def overriderebasecmd(orig, ui, repo, **opts):
    if not hasattr(repo, '_largefilesenabled'):
        return orig(ui, repo, **opts)

    resuming = opts.get('continue')
    repo._lfcommithooks.append(lfutil.automatedcommithook(resuming))
    repo._lfstatuswriters.append(lambda *msg, **opts: None)
    try:
        with ui.configoverride(
            {(b'rebase', b'experimental.inmemory'): False}, b"largefiles"
        ):
            return orig(ui, repo, **opts)
    finally:
        repo._lfstatuswriters.pop()
        repo._lfcommithooks.pop()


@eh.extsetup
def overriderebase(ui):
    try:
        rebase = extensions.find(b'rebase')
    except KeyError:
        pass
    else:

        def _dorebase(orig, *args, **kwargs):
            kwargs['inmemory'] = False
            return orig(*args, **kwargs)

        extensions.wrapfunction(rebase, '_dorebase', _dorebase)


@eh.wrapcommand(b'archive')
def overridearchivecmd(orig, ui, repo, dest, **opts):
    with lfstatus(repo.unfiltered()):
        return orig(ui, repo.unfiltered(), dest, **opts)


@eh.wrapfunction(webcommands, 'archive')
def hgwebarchive(orig, web):
    with lfstatus(web.repo):
        return orig(web)


@eh.wrapfunction(archival, 'archive')
def overridearchive(
    orig,
    repo,
    dest,
    node,
    kind,
    decode=True,
    match=None,
    prefix=b'',
    mtime=None,
    subrepos=None,
):
    # For some reason setting repo.lfstatus in hgwebarchive only changes the
    # unfiltered repo's attr, so check that as well.
    if not repo.lfstatus and not repo.unfiltered().lfstatus:
        return orig(
            repo, dest, node, kind, decode, match, prefix, mtime, subrepos
        )

    # No need to lock because we are only reading history and
    # largefile caches, neither of which are modified.
    if node is not None:
        lfcommands.cachelfiles(repo.ui, repo, node)

    if kind not in archival.archivers:
        raise error.Abort(_(b"unknown archive type '%s'") % kind)

    ctx = repo[node]

    if kind == b'files':
        if prefix:
            raise error.Abort(_(b'cannot give prefix when archiving to files'))
    else:
        prefix = archival.tidyprefix(dest, kind, prefix)

    def write(name, mode, islink, getdata):
        if match and not match(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(b"ui", b"archivemeta"):
        write(
            b'.hg_archival.txt',
            0o644,
            False,
            lambda: archival.buildmetadata(ctx),
        )

    for f in ctx:
        ff = ctx.flags(f)
        getdata = ctx[f].data
        lfile = lfutil.splitstandin(f)
        if lfile is not None:
            if node is not None:
                path = lfutil.findfile(repo, getdata().strip())

                if path is None:
                    raise error.Abort(
                        _(
                            b'largefile %s not found in repo store or system cache'
                        )
                        % lfile
                    )
            else:
                path = lfile

            f = lfile

            getdata = lambda: util.readfile(path)
        write(f, b'x' in ff and 0o755 or 0o644, b'l' in ff, getdata)

    if subrepos:
        for subpath in sorted(ctx.substate):
            sub = ctx.workingsub(subpath)
            submatch = matchmod.subdirmatcher(subpath, match)
            subprefix = prefix + subpath + b'/'

            # TODO: Only hgsubrepo instances have `_repo`, so figure out how to
            # infer and possibly set lfstatus in hgsubrepoarchive.  That would
            # allow only hgsubrepos to set this, instead of the current scheme
            # where the parent sets this for the child.
            with (
                hasattr(sub, '_repo')
                and lfstatus(sub._repo)
                or util.nullcontextmanager()
            ):
                sub.archive(archiver, subprefix, submatch)

    archiver.done()


@eh.wrapfunction(subrepo.hgsubrepo, 'archive')
def hgsubrepoarchive(orig, repo, archiver, prefix, match=None, decode=True):
    lfenabled = hasattr(repo._repo, '_largefilesenabled')
    if not lfenabled or not repo._repo.lfstatus:
        return orig(repo, archiver, prefix, match, decode)

    repo._get(repo._state + (b'hg',))
    rev = repo._state[1]
    ctx = repo._repo[rev]

    if ctx.node() is not None:
        lfcommands.cachelfiles(repo.ui, repo._repo, ctx.node())

    def write(name, mode, islink, getdata):
        # At this point, the standin has been replaced with the largefile name,
        # so the normal matcher works here without the lfutil variants.
        if match and not match(f):
            return
        data = getdata()
        if decode:
            data = repo._repo.wwritedata(name, data)

        archiver.addfile(prefix + name, mode, islink, data)

    for f in ctx:
        ff = ctx.flags(f)
        getdata = ctx[f].data
        lfile = lfutil.splitstandin(f)
        if lfile is not None:
            if ctx.node() is not None:
                path = lfutil.findfile(repo._repo, getdata().strip())

                if path is None:
                    raise error.Abort(
                        _(
                            b'largefile %s not found in repo store or system cache'
                        )
                        % lfile
                    )
            else:
                path = lfile

            f = lfile

            getdata = lambda: util.readfile(os.path.join(prefix, path))

        write(f, b'x' in ff and 0o755 or 0o644, b'l' in ff, getdata)

    for subpath in sorted(ctx.substate):
        sub = ctx.workingsub(subpath)
        submatch = matchmod.subdirmatcher(subpath, match)
        subprefix = prefix + subpath + b'/'
        # TODO: Only hgsubrepo instances have `_repo`, so figure out how to
        # infer and possibly set lfstatus at the top of this function.  That
        # would allow only hgsubrepos to set this, instead of the current scheme
        # where the parent sets this for the child.
        with (
            hasattr(sub, '_repo')
            and lfstatus(sub._repo)
            or util.nullcontextmanager()
        ):
            sub.archive(archiver, subprefix, submatch, decode)


# 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, backout and fetch.
@eh.wrapfunction(cmdutil, 'bailifchanged')
def overridebailifchanged(orig, repo, *args, **kwargs):
    orig(repo, *args, **kwargs)
    with lfstatus(repo):
        s = repo.status()
    if s.modified or s.added or s.removed or s.deleted:
        raise error.Abort(_(b'uncommitted changes'))


@eh.wrapfunction(cmdutil, 'postcommitstatus')
def postcommitstatus(orig, repo, *args, **kwargs):
    with lfstatus(repo):
        return orig(repo, *args, **kwargs)


@eh.wrapfunction(cmdutil, 'forget')
def cmdutilforget(
    orig, ui, repo, match, prefix, uipathfn, explicitonly, dryrun, interactive
):
    normalmatcher = composenormalfilematcher(match, repo[None].manifest())
    bad, forgot = orig(
        ui,
        repo,
        normalmatcher,
        prefix,
        uipathfn,
        explicitonly,
        dryrun,
        interactive,
    )
    m = composelargefilematcher(match, repo[None].manifest())

    with lfstatus(repo):
        s = repo.status(match=m, clean=True)
    manifest = repo[None].manifest()
    forget = sorted(s.modified + s.added + s.deleted + s.clean)
    forget = [f for f in forget if lfutil.standin(f) in manifest]

    for f in forget:
        fstandin = lfutil.standin(f)
        if fstandin not in repo.dirstate and not repo.wvfs.isdir(fstandin):
            ui.warn(
                _(b'not removing %s: file is already untracked\n') % uipathfn(f)
            )
            bad.append(f)

    for f in forget:
        if ui.verbose or not m.exact(f):
            ui.status(_(b'removing %s\n') % uipathfn(f))

    # Need to lock because standin files are deleted then removed from the
    # repository and we could race in-between.
    with repo.wlock():
        lfdirstate = lfutil.openlfdirstate(ui, repo)
        for f in forget:
            lfdirstate.set_untracked(f)
        lfdirstate.write(repo.currenttransaction())
        standins = [lfutil.standin(f) for f in forget]
        for f in standins:
            repo.wvfs.unlinkpath(f, ignoremissing=True)
        rejected = repo[None].forget(standins)

    bad.extend(f for f in rejected if f in m.files())
    forgot.extend(f for f in forget if f not in rejected)
    return bad, forgot


def _getoutgoings(repo, other, missing, addfunc):
    """get pairs of filename and largefile hash in outgoing revisions
    in 'missing'.

    largefiles already existing on 'other' repository are ignored.

    'addfunc' is invoked with each unique pairs of filename and
    largefile hash value.
    """
    knowns = set()
    lfhashes = set()

    def dedup(fn, lfhash):
        k = (fn, lfhash)
        if k not in knowns:
            knowns.add(k)
            lfhashes.add(lfhash)

    lfutil.getlfilestoupload(repo, missing, dedup)
    if lfhashes:
        lfexists = storefactory.openstore(repo, other).exists(lfhashes)
        for fn, lfhash in knowns:
            if not lfexists[lfhash]:  # lfhash doesn't exist on "other"
                addfunc(fn, lfhash)


def outgoinghook(ui, repo, other, opts, missing):
    if opts.pop(b'large', None):
        lfhashes = set()
        if ui.debugflag:
            toupload = {}

            def addfunc(fn, lfhash):
                if fn not in toupload:
                    toupload[fn] = []  # pytype: disable=unsupported-operands
                toupload[fn].append(lfhash)
                lfhashes.add(lfhash)

            def showhashes(fn):
                for lfhash in sorted(toupload[fn]):
                    ui.debug(b'    %s\n' % lfhash)

        else:
            toupload = set()

            def addfunc(fn, lfhash):
                toupload.add(fn)
                lfhashes.add(lfhash)

            def showhashes(fn):
                pass

        _getoutgoings(repo, other, missing, addfunc)

        if not toupload:
            ui.status(_(b'largefiles: no files to upload\n'))
        else:
            ui.status(
                _(b'largefiles to upload (%d entities):\n') % (len(lfhashes))
            )
            for file in sorted(toupload):
                ui.status(lfutil.splitstandin(file) + b'\n')
                showhashes(file)
            ui.status(b'\n')


@eh.wrapcommand(
    b'outgoing', opts=[(b'', b'large', None, _(b'display outgoing largefiles'))]
)
def _outgoingcmd(orig, *args, **kwargs):
    # Nothing to do here other than add the extra help option- the hook above
    # processes it.
    return orig(*args, **kwargs)


def summaryremotehook(ui, repo, opts, changes):
    largeopt = opts.get(b'large', False)
    if changes is None:
        if largeopt:
            return (False, True)  # only outgoing check is needed
        else:
            return (False, False)
    elif largeopt:
        url, branch, peer, outgoing = changes[1]
        if peer is None:
            # i18n: column positioning for "hg summary"
            ui.status(_(b'largefiles: (no remote repo)\n'))
            return

        toupload = set()
        lfhashes = set()

        def addfunc(fn, lfhash):
            toupload.add(fn)
            lfhashes.add(lfhash)

        _getoutgoings(repo, peer, outgoing.missing, addfunc)

        if not toupload:
            # i18n: column positioning for "hg summary"
            ui.status(_(b'largefiles: (no files to upload)\n'))
        else:
            # i18n: column positioning for "hg summary"
            ui.status(
                _(b'largefiles: %d entities for %d files to upload\n')
                % (len(lfhashes), len(toupload))
            )


@eh.wrapcommand(
    b'summary', opts=[(b'', b'large', None, _(b'display outgoing largefiles'))]
)
def overridesummary(orig, ui, repo, *pats, **opts):
    with lfstatus(repo):
        orig(ui, repo, *pats, **opts)


@eh.wrapfunction(scmutil, 'addremove')
def scmutiladdremove(
    orig,
    repo,
    matcher,
    prefix,
    uipathfn,
    opts=None,
    open_tr=None,
):
    if opts is None:
        opts = {}
    if not lfutil.islfilesrepo(repo):
        return orig(repo, matcher, prefix, uipathfn, opts, open_tr=open_tr)

    # open the transaction and changing_files context
    if open_tr is not None:
        open_tr()

    # Get the list of missing largefiles so we can remove them
    with repo.dirstate.running_status(repo):
        lfdirstate = lfutil.openlfdirstate(repo.ui, repo)
        unsure, s, mtime_boundary = lfdirstate.status(
            matchmod.always(),
            subrepos=[],
            ignored=False,
            clean=False,
            unknown=False,
        )

    # 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.
    if s.deleted:
        m = copy.copy(matcher)

        # The m._files and m._map attributes are not changed to the deleted list
        # because that affects the m.exact() test, which in turn governs whether
        # or not the file name is printed, and how.  Simply limit the original
        # matches to those in the deleted status list.
        matchfn = m.matchfn
        m.matchfn = lambda f: f in s.deleted and matchfn(f)

        removelargefiles(
            repo.ui,
            repo,
            True,
            m,
            uipathfn,
            opts.get(b'dry_run'),
            **pycompat.strkwargs(opts)
        )
    # Call into the normal add code, and any files that *should* be added as
    # largefiles will be
    added, bad = addlargefiles(
        repo.ui, repo, True, matcher, uipathfn, **pycompat.strkwargs(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 passing a matcher that will ignore them.
    matcher = composenormalfilematcher(matcher, repo[None].manifest(), added)

    return orig(repo, matcher, prefix, uipathfn, opts, open_tr=open_tr)


# Calling purge with --all will cause the largefiles to be deleted.
# Override repo.status to prevent this from happening.
@eh.wrapcommand(b'purge')
def overridepurge(orig, ui, repo, *dirs, **opts):
    # XXX Monkey patching a repoview will not work. The assigned attribute will
    # be set on the unfiltered repo, but we will only lookup attributes in the
    # unfiltered repo if the lookup in the repoview object itself fails. As the
    # monkey patched method exists on the repoview class the lookup will not
    # fail. As a result, the original version will shadow the monkey patched
    # one, defeating the monkey patch.
    #
    # As a work around we use an unfiltered repo here. We should do something
    # cleaner instead.
    repo = repo.unfiltered()
    oldstatus = repo.status

    def overridestatus(
        node1=b'.',
        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)
        unknown = [
            f for f in r.unknown if not lfdirstate.get_entry(f).any_tracked
        ]
        ignored = [
            f for f in r.ignored if not lfdirstate.get_entry(f).any_tracked
        ]
        return scmutil.status(
            r.modified, r.added, r.removed, r.deleted, unknown, ignored, r.clean
        )

    repo.status = overridestatus
    orig(ui, repo, *dirs, **opts)
    repo.status = oldstatus


@eh.wrapcommand(b'rollback')
def overriderollback(orig, ui, repo, **opts):
    with repo.wlock():
        before = repo.dirstate.parents()
        orphans = {
            f
            for f in repo.dirstate
            if lfutil.isstandin(f) and not repo.dirstate.get_entry(f).removed
        }
        result = orig(ui, repo, **opts)
        after = repo.dirstate.parents()
        if before == after:
            return result  # no need to restore standins

        pctx = repo[b'.']
        for f in repo.dirstate:
            if lfutil.isstandin(f):
                orphans.discard(f)
                if repo.dirstate.get_entry(f).removed:
                    repo.wvfs.unlinkpath(f, ignoremissing=True)
                elif f in pctx:
                    fctx = pctx[f]
                    repo.wwrite(f, fctx.data(), fctx.flags())
                else:
                    # content of standin is not so important in 'a',
                    # 'm' or 'n' (coming from the 2nd parent) cases
                    lfutil.writestandin(repo, f, b'', False)
        for standin in orphans:
            repo.wvfs.unlinkpath(standin, ignoremissing=True)

    return result


@eh.wrapcommand(b'transplant', extension=b'transplant')
def overridetransplant(orig, ui, repo, *revs, **opts):
    resuming = opts.get('continue')
    repo._lfcommithooks.append(lfutil.automatedcommithook(resuming))
    repo._lfstatuswriters.append(lambda *msg, **opts: None)
    try:
        result = orig(ui, repo, *revs, **opts)
    finally:
        repo._lfstatuswriters.pop()
        repo._lfcommithooks.pop()
    return result


@eh.wrapcommand(b'cat')
def overridecat(orig, ui, repo, file1, *pats, **opts):
    ctx = logcmdutil.revsingle(repo, opts.get('rev'))
    err = 1
    notbad = set()
    m = scmutil.match(ctx, (file1,) + pats, pycompat.byteskwargs(opts))
    origmatchfn = m.matchfn

    def lfmatchfn(f):
        if origmatchfn(f):
            return True
        lf = lfutil.splitstandin(f)
        if lf is None:
            return False
        notbad.add(lf)
        return origmatchfn(lf)

    m.matchfn = lfmatchfn
    origbadfn = m.bad

    def lfbadfn(f, msg):
        if not f in notbad:
            origbadfn(f, msg)

    m.bad = lfbadfn

    origvisitdirfn = m.visitdir

    def lfvisitdirfn(dir):
        if dir == lfutil.shortname:
            return True
        ret = origvisitdirfn(dir)
        if ret:
            return ret
        lf = lfutil.splitstandin(dir)
        if lf is None:
            return False
        return origvisitdirfn(lf)

    m.visitdir = lfvisitdirfn

    for f in ctx.walk(m):
        with cmdutil.makefileobj(ctx, opts.get('output'), pathname=f) as fp:
            lf = lfutil.splitstandin(f)
            if lf is None or origmatchfn(f):
                # duplicating unreachable code from commands.cat
                data = ctx[f].data()
                if opts.get('decode'):
                    data = repo.wwritedata(f, data)
                fp.write(data)
            else:
                hash = lfutil.readasstandin(ctx[f])
                if not lfutil.inusercache(repo.ui, hash):
                    store = storefactory.openstore(repo)
                    success, missing = store.get([(lf, hash)])
                    if len(success) != 1:
                        raise error.Abort(
                            _(
                                b'largefile %s is not in cache and could not be '
                                b'downloaded'
                            )
                            % lf
                        )
                path = lfutil.usercachepath(repo.ui, hash)
                with open(path, b"rb") as fpin:
                    for chunk in util.filechunkiter(fpin):
                        fp.write(chunk)
        err = 0
    return err


@eh.wrapfunction(merge, '_update')
def mergeupdate(orig, repo, node, branchmerge, force, *args, **kwargs):
    matcher = kwargs.get('matcher', None)
    # note if this is a partial update
    partial = matcher and not matcher.always()
    with repo.wlock(), repo.dirstate.changing_parents(repo):
        # branch |       |         |
        #  merge | force | partial | action
        # -------+-------+---------+--------------
        #    x   |   x   |    x    | linear-merge
        #    o   |   x   |    x    | branch-merge
        #    x   |   o   |    x    | overwrite (as clean update)
        #    o   |   o   |    x    | force-branch-merge (*1)
        #    x   |   x   |    o    |   (*)
        #    o   |   x   |    o    |   (*)
        #    x   |   o   |    o    | overwrite (as revert)
        #    o   |   o   |    o    |   (*)
        #
        # (*) don't care
        # (*1) deprecated, but used internally (e.g: "rebase --collapse")
        with repo.dirstate.running_status(repo):
            lfdirstate = lfutil.openlfdirstate(repo.ui, repo)
            unsure, s, mtime_boundary = lfdirstate.status(
                matchmod.always(),
                subrepos=[],
                ignored=False,
                clean=True,
                unknown=False,
            )
        oldclean = set(s.clean)
        pctx = repo[b'.']
        dctx = repo[node]
        for lfile in unsure + s.modified:
            lfileabs = repo.wvfs.join(lfile)
            if not repo.wvfs.exists(lfileabs):
                continue
            lfhash = lfutil.hashfile(lfileabs)
            standin = lfutil.standin(lfile)
            lfutil.writestandin(
                repo, standin, lfhash, lfutil.getexecutable(lfileabs)
            )
            if standin in pctx and lfhash == lfutil.readasstandin(
                pctx[standin]
            ):
                oldclean.add(lfile)
        for lfile in s.added:
            fstandin = lfutil.standin(lfile)
            if fstandin not in dctx:
                # in this case, content of standin file is meaningless
                # (in dctx, lfile is unknown, or normal file)
                continue
            lfutil.updatestandin(repo, lfile, fstandin)
        # mark all clean largefiles as dirty, just in case the update gets
        # interrupted before largefiles and lfdirstate are synchronized
        for lfile in oldclean:
            entry = lfdirstate.get_entry(lfile)
            lfdirstate.hacky_extension_update_file(
                lfile,
                wc_tracked=entry.tracked,
                p1_tracked=entry.p1_tracked,
                p2_info=entry.p2_info,
                possibly_dirty=True,
            )
        lfdirstate.write(repo.currenttransaction())

        oldstandins = lfutil.getstandinsstate(repo)
        wc = kwargs.get('wc')
        if wc and wc.isinmemory():
            # largefiles is not a good candidate for in-memory merge (large
            # files, custom dirstate, matcher usage).
            raise error.ProgrammingError(
                b'largefiles is not compatible with in-memory merge'
            )
        result = orig(repo, node, branchmerge, force, *args, **kwargs)

        newstandins = lfutil.getstandinsstate(repo)
        filelist = lfutil.getlfilestoupdate(oldstandins, newstandins)

        # to avoid leaving all largefiles as dirty and thus rehash them, mark
        # all the ones that didn't change as clean
        for lfile in oldclean.difference(filelist):
            lfdirstate.update_file(lfile, p1_tracked=True, wc_tracked=True)

        if branchmerge or force or partial:
            filelist.extend(s.deleted + s.removed)

        lfcommands.updatelfiles(
            repo.ui, repo, filelist=filelist, normallookup=partial
        )

        return result


@eh.wrapfunction(scmutil, 'marktouched')
def scmutilmarktouched(orig, repo, files, *args, **kwargs):
    result = orig(repo, files, *args, **kwargs)

    filelist = []
    for f in files:
        lf = lfutil.splitstandin(f)
        if lf is not None:
            filelist.append(lf)
    if filelist:
        lfcommands.updatelfiles(
            repo.ui,
            repo,
            filelist=filelist,
            printmessage=False,
            normallookup=True,
        )

    return result


@eh.wrapfunction(upgrade_actions, 'preservedrequirements')
@eh.wrapfunction(upgrade_actions, 'supporteddestrequirements')
def upgraderequirements(orig, repo):
    reqs = orig(repo)
    if b'largefiles' in repo.requirements:
        reqs.add(b'largefiles')
    return reqs


_lfscheme = b'largefile://'


@eh.wrapfunction(urlmod, 'open')
def openlargefile(orig, ui, url_, data=None, **kwargs):
    if url_.startswith(_lfscheme):
        if data:
            msg = b"cannot use data on a 'largefile://' url"
            raise error.ProgrammingError(msg)
        lfid = url_[len(_lfscheme) :]
        return storefactory.getlfile(ui, lfid)
    else:
        return orig(ui, url_, data=data, **kwargs)