hgext/remotefilelog/shallowbundle.py

commit: improve the files field of changelog for merges

Currently, the files list of merge commits repeats all the deletions
(either actual deletions, or files that got renamed) that happened
between base and p2 of the merge. If p2 is the main branch, the list
can easily be much bigger than the change being merged.

This results in various problems worth improving:
- changelog is bigger than necessary
- `hg log directory` lists many unrelated merge commits, and `hg log
  -v -r commit` frequently fills multiple screens worth of files
- it possibly slows down adjustlinkrev, by forcing it to read more
  manifests, and that function can certainly be a bottleneck
- the server side of pulls can waste a lot of time simply opening the
  filelogs for pointless files (the constant factors for opening even
  a tiny filelog is apparently pretty bad)

So stop listing such files as described in the code.  Impacted merge
commits and their descendants get a different hash than they would
have without this. This doesn't seem problematic, except for
convert. The previous commit helped with that in the hg->hg case (but
if you do svn->hg twice from scratch, hashes can still change).

The rest of the description is numbers. I don't have much to report,
because recreating the files list of existing repositories is not
easy:
- debugupgradeformat and bundle/unbundle don't recreate the list
- export/import tends to choke quickly applying patches or on
  description that contain diffs,
- merge commits from the convert extension don't have the right files
  list for reasons orthogonal to the current commit
- replaying the merge with hg update/hg merge/hg revert --all/hg
  commit can end up failing in hg revert
- I wasn't sure that using debugsetparents + debugrebuilddirstate
  would really build the right thing

I measured commit time before and after this change, in a case with no
files filtered out, several files filtered out (no difference) and 5k
files filtered out (+1% time).

Recreating the 100 more recent merges in a private repo, the
concatenated uncompressed files lists goes from 1.12MB to
0.52MB. Excluding 3 merges that are not representative, then the size
goes from 570k to 15k.
I converted part of mozilla-central, and observed file list shrinking
quite a bit too, starting at the very first merge, 733641d9feaf, going
from 550 files to 10 files (although they have relatively few merges,
so they probably wouldn't care).

Differential Revision: https://phab.mercurial-scm.org/D6613

# shallowbundle.py - bundle10 implementation for use with shallow repositories
#
# Copyright 2013 Facebook, Inc.
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.
from __future__ import absolute_import

from mercurial.i18n import _
from mercurial.node import bin, hex, nullid
from mercurial import (
    bundlerepo,
    changegroup,
    error,
    match,
    mdiff,
    pycompat,
)
from . import (
    constants,
    remotefilelog,
    shallowutil,
)

NoFiles = 0
LocalFiles = 1
AllFiles = 2

def shallowgroup(cls, self, nodelist, rlog, lookup, units=None, reorder=None):
    if not isinstance(rlog, remotefilelog.remotefilelog):
        for c in super(cls, self).group(nodelist, rlog, lookup,
                                        units=units):
            yield c
        return

    if len(nodelist) == 0:
        yield self.close()
        return

    nodelist = shallowutil.sortnodes(nodelist, rlog.parents)

    # add the parent of the first rev
    p = rlog.parents(nodelist[0])[0]
    nodelist.insert(0, p)

    # build deltas
    for i in pycompat.xrange(len(nodelist) - 1):
        prev, curr = nodelist[i], nodelist[i + 1]
        linknode = lookup(curr)
        for c in self.nodechunk(rlog, curr, prev, linknode):
            yield c

    yield self.close()

class shallowcg1packer(changegroup.cgpacker):
    def generate(self, commonrevs, clnodes, fastpathlinkrev, source):
        if shallowutil.isenabled(self._repo):
            fastpathlinkrev = False

        return super(shallowcg1packer, self).generate(commonrevs, clnodes,
            fastpathlinkrev, source)

    def group(self, nodelist, rlog, lookup, units=None, reorder=None):
        return shallowgroup(shallowcg1packer, self, nodelist, rlog, lookup,
                            units=units)

    def generatefiles(self, changedfiles, *args):
        try:
            linknodes, commonrevs, source = args
        except ValueError:
            commonrevs, source, mfdicts, fastpathlinkrev, fnodes, clrevs = args
        if shallowutil.isenabled(self._repo):
            repo = self._repo
            if isinstance(repo, bundlerepo.bundlerepository):
                # If the bundle contains filelogs, we can't pull from it, since
                # bundlerepo is heavily tied to revlogs. Instead require that
                # the user use unbundle instead.
                # Force load the filelog data.
                bundlerepo.bundlerepository.file(repo, 'foo')
                if repo._cgfilespos:
                    raise error.Abort("cannot pull from full bundles",
                                      hint="use `hg unbundle` instead")
                return []
            filestosend = self.shouldaddfilegroups(source)
            if filestosend == NoFiles:
                changedfiles = list([f for f in changedfiles
                                     if not repo.shallowmatch(f)])

        return super(shallowcg1packer, self).generatefiles(
            changedfiles, *args)

    def shouldaddfilegroups(self, source):
        repo = self._repo
        if not shallowutil.isenabled(repo):
            return AllFiles

        if source == "push" or source == "bundle":
            return AllFiles

        caps = self._bundlecaps or []
        if source == "serve" or source == "pull":
            if constants.BUNDLE2_CAPABLITY in caps:
                return LocalFiles
            else:
                # Serving to a full repo requires us to serve everything
                repo.ui.warn(_("pulling from a shallow repo\n"))
                return AllFiles

        return NoFiles

    def prune(self, rlog, missing, commonrevs):
        if not isinstance(rlog, remotefilelog.remotefilelog):
            return super(shallowcg1packer, self).prune(rlog, missing,
                commonrevs)

        repo = self._repo
        results = []
        for fnode in missing:
            fctx = repo.filectx(rlog.filename, fileid=fnode)
            if fctx.linkrev() not in commonrevs:
                results.append(fnode)
        return results

    def nodechunk(self, revlog, node, prevnode, linknode):
        prefix = ''
        if prevnode == nullid:
            delta = revlog.revision(node, raw=True)
            prefix = mdiff.trivialdiffheader(len(delta))
        else:
            # Actually uses remotefilelog.revdiff which works on nodes, not revs
            delta = revlog.revdiff(prevnode, node)
        p1, p2 = revlog.parents(node)
        flags = revlog.flags(node)
        meta = self.builddeltaheader(node, p1, p2, prevnode, linknode, flags)
        meta += prefix
        l = len(meta) + len(delta)
        yield changegroup.chunkheader(l)
        yield meta
        yield delta

def makechangegroup(orig, repo, outgoing, version, source, *args, **kwargs):
    if not shallowutil.isenabled(repo):
        return orig(repo, outgoing, version, source, *args, **kwargs)

    original = repo.shallowmatch
    try:
        # if serving, only send files the clients has patterns for
        if source == 'serve':
            bundlecaps = kwargs.get(r'bundlecaps')
            includepattern = None
            excludepattern = None
            for cap in (bundlecaps or []):
                if cap.startswith("includepattern="):
                    raw = cap[len("includepattern="):]
                    if raw:
                        includepattern = raw.split('\0')
                elif cap.startswith("excludepattern="):
                    raw = cap[len("excludepattern="):]
                    if raw:
                        excludepattern = raw.split('\0')
            if includepattern or excludepattern:
                repo.shallowmatch = match.match(repo.root, '', None,
                    includepattern, excludepattern)
            else:
                repo.shallowmatch = match.always()
        return orig(repo, outgoing, version, source, *args, **kwargs)
    finally:
        repo.shallowmatch = original

def addchangegroupfiles(orig, repo, source, revmap, trp, expectedfiles, *args):
    if not shallowutil.isenabled(repo):
        return orig(repo, source, revmap, trp, expectedfiles, *args)

    newfiles = 0
    visited = set()
    revisiondatas = {}
    queue = []

    # Normal Mercurial processes each file one at a time, adding all
    # the new revisions for that file at once. In remotefilelog a file
    # revision may depend on a different file's revision (in the case
    # of a rename/copy), so we must lay all revisions down across all
    # files in topological order.

    # read all the file chunks but don't add them
    progress = repo.ui.makeprogress(_('files'), total=expectedfiles)
    while True:
        chunkdata = source.filelogheader()
        if not chunkdata:
            break
        f = chunkdata["filename"]
        repo.ui.debug("adding %s revisions\n" % f)
        progress.increment()

        if not repo.shallowmatch(f):
            fl = repo.file(f)
            deltas = source.deltaiter()
            fl.addgroup(deltas, revmap, trp)
            continue

        chain = None
        while True:
            # returns: (node, p1, p2, cs, deltabase, delta, flags) or None
            revisiondata = source.deltachunk(chain)
            if not revisiondata:
                break

            chain = revisiondata[0]

            revisiondatas[(f, chain)] = revisiondata
            queue.append((f, chain))

            if f not in visited:
                newfiles += 1
                visited.add(f)

        if chain is None:
            raise error.Abort(_("received file revlog group is empty"))

    processed = set()
    def available(f, node, depf, depnode):
        if depnode != nullid and (depf, depnode) not in processed:
            if not (depf, depnode) in revisiondatas:
                # It's not in the changegroup, assume it's already
                # in the repo
                return True
            # re-add self to queue
            queue.insert(0, (f, node))
            # add dependency in front
            queue.insert(0, (depf, depnode))
            return False
        return True

    skipcount = 0

    # Prefetch the non-bundled revisions that we will need
    prefetchfiles = []
    for f, node in queue:
        revisiondata = revisiondatas[(f, node)]
        # revisiondata: (node, p1, p2, cs, deltabase, delta, flags)
        dependents = [revisiondata[1], revisiondata[2], revisiondata[4]]

        for dependent in dependents:
            if dependent == nullid or (f, dependent) in revisiondatas:
                continue
            prefetchfiles.append((f, hex(dependent)))

    repo.fileservice.prefetch(prefetchfiles)

    # Apply the revisions in topological order such that a revision
    # is only written once it's deltabase and parents have been written.
    while queue:
        f, node = queue.pop(0)
        if (f, node) in processed:
            continue

        skipcount += 1
        if skipcount > len(queue) + 1:
            raise error.Abort(_("circular node dependency"))

        fl = repo.file(f)

        revisiondata = revisiondatas[(f, node)]
        # revisiondata: (node, p1, p2, cs, deltabase, delta, flags)
        node, p1, p2, linknode, deltabase, delta, flags = revisiondata

        if not available(f, node, f, deltabase):
            continue

        base = fl.revision(deltabase, raw=True)
        text = mdiff.patch(base, delta)
        if not isinstance(text, bytes):
            text = bytes(text)

        meta, text = shallowutil.parsemeta(text)
        if 'copy' in meta:
            copyfrom = meta['copy']
            copynode = bin(meta['copyrev'])
            if not available(f, node, copyfrom, copynode):
                continue

        for p in [p1, p2]:
            if p != nullid:
                if not available(f, node, f, p):
                    continue

        fl.add(text, meta, trp, linknode, p1, p2)
        processed.add((f, node))
        skipcount = 0

    progress.complete()

    return len(revisiondatas), newfiles