copies: iterate over children directly (instead of parents)

Before this change we would gather all parent → child edges and iterate on
all parent, gathering copy information for children and aggregating them from
there.

They are not strict requirement for edges to be processed in that specific
order. We could also simply iterate over all "children" revision and aggregate
data from both parents at the same time. This patch does that.

It make various things simpler:

 * since both parents are processed at the same time, we no longer need to
   cache data for merge (see next changeset for details),

 * we no longer need nested loop to process data,

 * we no longer need to store partial merge data for a rev from distinct loop
   interaction to another when processing merges,

 * we no longer need to build a full parent -> children mapping (we only rely on
   a simpler "parent -> number of children" map (for memory efficiency),

 * the data access pattern is now simpler (from lower revisions to higher
   revisions) and entirely predicable. That predictability open the way to
   prefetching and parallel processing.

So that new iterations order requires simpler code and open the way to
interesting optimisation.

The effect on performance is quite good. In the worse case, we don't see any
significant negative impact. And in the best case, the reduction of roundtrip
to Python provide us with a significant speed. Some example below:

Repo            Case                                  Source-Rev   Dest-Rev      # of revisions  old time    new time      Difference    Factor    time per rev
---------------------------------------------------------------------------------------------------------------------------------------------------------------
mozilla-try     x00000_revs_x00000_added_0_copies     dc8a3ca7010e d16fde900c9c :  34414 revs,   0.962867 s,   0.502584 s,  -0.460283 s, × 0.5220,    14 µs/rev
mozilla-try     x0000_revs_xx000_added_x_copies       156f6e2674f2 4d0f2c178e66 :   8598 revs,   0.110717 s,   0.076323 s,  -0.034394 s, × 0.6894,     8 µs/rev



# full comparison between the previous changeset and this one

Repo            Case                                  Source-Rev   Dest-Rev      # of revisions  old time    new time      Difference    Factor    time per rev
---------------------------------------------------------------------------------------------------------------------------------------------------------------
mercurial       x_revs_x_added_0_copies               ad6b123de1c7 39cfcef4f463 :      1 revs,   0.000048 s,   0.000041 s,  -0.000007 s, × 0.8542,    41 µs/rev
mercurial       x_revs_x_added_x_copies               2b1c78674230 0c1d10351869 :      6 revs,   0.000153 s,   0.000102 s,  -0.000051 s, × 0.6667,    17 µs/rev
mercurial       x000_revs_x000_added_x_copies         81f8ff2a9bf2 dd3267698d84 :   1032 revs,   0.004209 s,   0.004254 s,  +0.000045 s, × 1.0107,     4 µs/rev
pypy            x_revs_x_added_0_copies               aed021ee8ae8 099ed31b181b :      9 revs,   0.000203 s,   0.000282 s,  +0.000079 s, × 1.3892,    31 µs/rev
pypy            x_revs_x000_added_0_copies            4aa4e1f8e19a 359343b9ac0e :      1 revs,   0.000059 s,   0.000048 s,  -0.000011 s, × 0.8136,    48 µs/rev
pypy            x_revs_x_added_x_copies               ac52eb7bbbb0 72e022663155 :      7 revs,   0.000194 s,   0.000211 s,  +0.000017 s, × 1.0876,    30 µs/rev
pypy            x_revs_x00_added_x_copies             c3b14617fbd7 ace7255d9a26 :      1 revs,   0.000380 s,   0.000375 s,  -0.000005 s, × 0.9868,   375 µs/rev
pypy            x_revs_x000_added_x000_copies         df6f7a526b60 a83dc6a2d56f :      6 revs,   0.010588 s,   0.010574 s,  -0.000014 s, × 0.9987,  1762 µs/rev
pypy            x000_revs_xx00_added_0_copies         89a76aede314 2f22446ff07e :   4785 revs,   0.048961 s,   0.049974 s,  +0.001013 s, × 1.0207,    10 µs/rev
pypy            x000_revs_x000_added_x_copies         8a3b5bfd266e 2c68e87c3efe :   6780 revs,   0.083612 s,   0.084300 s,  +0.000688 s, × 1.0082,    12 µs/rev
pypy            x000_revs_x000_added_x000_copies      89a76aede314 7b3dda341c84 :   5441 revs,   0.058579 s,   0.060128 s,  +0.001549 s, × 1.0264,    11 µs/rev
pypy            x0000_revs_x_added_0_copies           d1defd0dc478 c9cb1334cc78 :  43645 revs,   0.736783 s,   0.686542 s,  -0.050241 s, × 0.9318,    15 µs/rev
pypy            x0000_revs_xx000_added_0_copies       bf2c629d0071 4ffed77c095c :      2 revs,   0.022050 s,   0.009277 s,  -0.012773 s, × 0.4207,  4638 µs/rev
pypy            x0000_revs_xx000_added_x000_copies    08ea3258278e d9fa043f30c0 :  11316 revs,   0.120800 s,   0.114733 s,  -0.006067 s, × 0.9498,    10 µs/rev
netbeans        x_revs_x_added_0_copies               fb0955ffcbcd a01e9239f9e7 :      2 revs,   0.000140 s,   0.000081 s,  -0.000059 s, × 0.5786,    40 µs/rev
netbeans        x_revs_x000_added_0_copies            6f360122949f 20eb231cc7d0 :      2 revs,   0.000114 s,   0.000107 s,  -0.000007 s, × 0.9386,    53 µs/rev
netbeans        x_revs_x_added_x_copies               1ada3faf6fb6 5a39d12eecf4 :      3 revs,   0.000224 s,   0.000173 s,  -0.000051 s, × 0.7723,    57 µs/rev
netbeans        x_revs_x00_added_x_copies             35be93ba1e2c 9eec5e90c05f :      9 revs,   0.000723 s,   0.000698 s,  -0.000025 s, × 0.9654,    77 µs/rev
netbeans        x000_revs_xx00_added_0_copies         eac3045b4fdd 51d4ae7f1290 :   1421 revs,   0.009665 s,   0.009248 s,  -0.000417 s, × 0.9569,     6 µs/rev
netbeans        x000_revs_x000_added_x_copies         e2063d266acd 6081d72689dc :   1533 revs,   0.014820 s,   0.015446 s,  +0.000626 s, × 1.0422,    10 µs/rev
netbeans        x000_revs_x000_added_x000_copies      ff453e9fee32 411350406ec2 :   5750 revs,   0.076049 s,   0.074373 s,  -0.001676 s, × 0.9780,    12 µs/rev
netbeans        x0000_revs_xx000_added_x000_copies    588c2d1ced70 1aad62e59ddd :  66949 revs,   0.683603 s,   0.639870 s,  -0.043733 s, × 0.9360,     9 µs/rev
mozilla-central x_revs_x_added_0_copies               3697f962bb7b 7015fcdd43a2 :      2 revs,   0.000161 s,   0.000088 s,  -0.000073 s, × 0.5466,    44 µs/rev
mozilla-central x_revs_x000_added_0_copies            dd390860c6c9 40d0c5bed75d :      8 revs,   0.000234 s,   0.000199 s,  -0.000035 s, × 0.8504,    24 µs/rev
mozilla-central x_revs_x_added_x_copies               8d198483ae3b 14207ffc2b2f :      9 revs,   0.000247 s,   0.000171 s,  -0.000076 s, × 0.6923,    19 µs/rev
mozilla-central x_revs_x00_added_x_copies             98cbc58cc6bc 446a150332c3 :      7 revs,   0.000630 s,   0.000592 s,  -0.000038 s, × 0.9397,    84 µs/rev
mozilla-central x_revs_x000_added_x000_copies         3c684b4b8f68 0a5e72d1b479 :      3 revs,   0.003286 s,   0.003151 s,  -0.000135 s, × 0.9589,  1050 µs/rev
mozilla-central x_revs_x0000_added_x0000_copies       effb563bb7e5 c07a39dc4e80 :      6 revs,   0.062441 s,   0.061612 s,  -0.000829 s, × 0.9867, 10268 µs/rev
mozilla-central x000_revs_xx00_added_0_copies         6100d773079a 04a55431795e :   1593 revs,   0.005423 s,   0.005381 s,  -0.000042 s, × 0.9923,     3 µs/rev
mozilla-central x000_revs_x000_added_x_copies         9f17a6fc04f9 2d37b966abed :     41 revs,   0.005919 s,   0.003742 s,  -0.002177 s, × 0.6322,    91 µs/rev
mozilla-central x000_revs_x000_added_x000_copies      7c97034feb78 4407bd0c6330 :   7839 revs,   0.062597 s,   0.061983 s,  -0.000614 s, × 0.9902,     7 µs/rev
mozilla-central x0000_revs_xx000_added_0_copies       9eec5917337d 67118cc6dcad :    615 revs,   0.043551 s,   0.019861 s,  -0.023690 s, × 0.4560,    32 µs/rev
mozilla-central x0000_revs_xx000_added_x000_copies    f78c615a656c 96a38b690156 :  30263 revs,   0.192475 s,   0.188101 s,  -0.004374 s, × 0.9773,     6 µs/rev
mozilla-central x00000_revs_x0000_added_x0000_copies  6832ae71433c 4c222a1d9a00 : 153721 revs,   1.955575 s,   1.806696 s,  -0.148879 s, × 0.9239,    11 µs/rev
mozilla-central x00000_revs_x00000_added_x000_copies  76caed42cf7c 1daa622bbe42 : 204976 revs,   2.886501 s,   2.682987 s,  -0.203514 s, × 0.9295,    13 µs/rev
mozilla-try     x_revs_x_added_0_copies               aaf6dde0deb8 9790f499805a :      2 revs,   0.001181 s,   0.000852 s,  -0.000329 s, × 0.7214,   426 µs/rev
mozilla-try     x_revs_x000_added_0_copies            d8d0222927b4 5bb8ce8c7450 :      2 revs,   0.001189 s,   0.000859 s,  -0.000330 s, × 0.7225,   429 µs/rev
mozilla-try     x_revs_x_added_x_copies               092fcca11bdb 936255a0384a :      4 revs,   0.000563 s,   0.000150 s,  -0.000413 s, × 0.2664,    37 µs/rev
mozilla-try     x_revs_x00_added_x_copies             b53d2fadbdb5 017afae788ec :      2 revs,   0.001548 s,   0.001158 s,  -0.000390 s, × 0.7481,   579 µs/rev
mozilla-try     x_revs_x000_added_x000_copies         20408ad61ce5 6f0ee96e21ad :      1 revs,   0.027782 s,   0.027240 s,  -0.000542 s, × 0.9805, 27240 µs/rev
mozilla-try     x_revs_x0000_added_x0000_copies       effb563bb7e5 c07a39dc4e80 :      6 revs,   0.062781 s,   0.062824 s,  +0.000043 s, × 1.0007, 10470 µs/rev
mozilla-try     x000_revs_xx00_added_0_copies         6100d773079a 04a55431795e :   1593 revs,   0.005778 s,   0.005463 s,  -0.000315 s, × 0.9455,     3 µs/rev
mozilla-try     x000_revs_x000_added_x_copies         9f17a6fc04f9 2d37b966abed :     41 revs,   0.006192 s,   0.004238 s,  -0.001954 s, × 0.6844,   103 µs/rev
mozilla-try     x000_revs_x000_added_x000_copies      1346fd0130e4 4c65cbdabc1f :   6657 revs,   0.065391 s,   0.064113 s,  -0.001278 s, × 0.9805,     9 µs/rev
mozilla-try     x0000_revs_x_added_0_copies           63519bfd42ee a36a2a865d92 :  40314 revs,   0.317216 s,   0.294063 s,  -0.023153 s, × 0.9270,     7 µs/rev
mozilla-try     x0000_revs_x_added_x_copies           9fe69ff0762d bcabf2a78927 :  38690 revs,   0.303119 s,   0.281493 s,  -0.021626 s, × 0.9287,     7 µs/rev
mozilla-try     x0000_revs_xx000_added_x_copies       156f6e2674f2 4d0f2c178e66 :   8598 revs,   0.110717 s,   0.076323 s,  -0.034394 s, × 0.6894,     8 µs/rev
mozilla-try     x0000_revs_xx000_added_0_copies       9eec5917337d 67118cc6dcad :    615 revs,   0.045739 s,   0.020390 s,  -0.025349 s, × 0.4458,    33 µs/rev
mozilla-try     x0000_revs_xx000_added_x000_copies    89294cd501d9 7ccb2fc7ccb5 :  97052 revs,   3.098021 s,   3.023879 s,  -0.074142 s, × 0.9761,    31 µs/rev
mozilla-try     x0000_revs_x0000_added_x0000_copies   e928c65095ed e951f4ad123a :  52031 revs,   0.771480 s,   0.735549 s,  -0.035931 s, × 0.9534,    14 µs/rev
mozilla-try     x00000_revs_x_added_0_copies          6a320851d377 1ebb79acd503 : 363753 revs,  18.813422 s,  18.568900 s,  -0.244522 s, × 0.9870,    51 µs/rev
mozilla-try     x00000_revs_x00000_added_0_copies     dc8a3ca7010e d16fde900c9c :  34414 revs,   0.962867 s,   0.502584 s,  -0.460283 s, × 0.5220,    14 µs/rev
mozilla-try     x00000_revs_x_added_x_copies          5173c4b6f97c 95d83ee7242d : 362229 revs,  18.684923 s,  18.356645 s,  -0.328278 s, × 0.9824,    50 µs/rev
mozilla-try     x00000_revs_x000_added_x_copies       9126823d0e9c ca82787bb23c : 359344 revs,  18.296305 s,  18.250393 s,  -0.045912 s, × 0.9975,    50 µs/rev
mozilla-try     x00000_revs_x0000_added_x0000_copies  8d3fafa80d4b eb884023b810 : 192665 revs,   3.061887 s,   2.792459 s,  -0.269428 s, × 0.9120,    14 µs/rev
mozilla-try     x00000_revs_x00000_added_x0000_copies 1b661134e2ca 1ae03d022d6d : 228985 revs, 103.869641 s, 107.697264 s,  +3.827623 s, × 1.0369,   470 µs/rev
mozilla-try     x00000_revs_x00000_added_x000_copies  9b2a99adc05e 8e29777b48e6 : 382065 revs,  64.262957 s,  63.961040 s,  -0.301917 s, × 0.9953,   167 µs/rev

Differential Revision: https://phab.mercurial-scm.org/D9422

mercurial/copies.py

@@ -1,1210 +1,1263
 # coding: utf8
 # copies.py - copy detection for Mercurial
 #
 # Copyright 2008 Matt Mackall <[e-mail]>
 #
 # 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
 
 import collections
 import os
 
 from .i18n import _
 from .node import (
     nullid,
     nullrev,
 )
 
 from . import (
     match as matchmod,
     pathutil,
     policy,
     pycompat,
     util,
 )
 
 
 from .utils import stringutil
 
 from .revlogutils import (
     flagutil,
     sidedata as sidedatamod,
 )
 
 rustmod = policy.importrust("copy_tracing")
 
 
 def _filter(src, dst, t):
     """filters out invalid copies after chaining"""
 
     # When _chain()'ing copies in 'a' (from 'src' via some other commit 'mid')
     # with copies in 'b' (from 'mid' to 'dst'), we can get the different cases
     # in the following table (not including trivial cases). For example, case 2
     # is where a file existed in 'src' and remained under that name in 'mid' and
     # then was renamed between 'mid' and 'dst'.
     #
     # case src mid dst result
     #   1   x   y   -    -
     #   2   x   y   y   x->y
     #   3   x   y   x    -
     #   4   x   y   z   x->z
     #   5   -   x   y    -
     #   6   x   x   y   x->y
     #
     # _chain() takes care of chaining the copies in 'a' and 'b', but it
     # cannot tell the difference between cases 1 and 2, between 3 and 4, or
     # between 5 and 6, so it includes all cases in its result.
     # Cases 1, 3, and 5 are then removed by _filter().
 
     for k, v in list(t.items()):
         # remove copies from files that didn't exist
         if v not in src:
             del t[k]
         # remove criss-crossed copies
         elif k in src and v in dst:
             del t[k]
         # remove copies to files that were then removed
         elif k not in dst:
             del t[k]
 
 
 def _chain(prefix, suffix):
     """chain two sets of copies 'prefix' and 'suffix'"""
     result = prefix.copy()
     for key, value in pycompat.iteritems(suffix):
         result[key] = prefix.get(value, value)
     return result
 
 
 def _tracefile(fctx, am, basemf):
     """return file context that is the ancestor of fctx present in ancestor
     manifest am
 
     Note: we used to try and stop after a given limit, however checking if that
     limit is reached turned out to be very expensive. we are better off
     disabling that feature."""
 
     for f in fctx.ancestors():
         path = f.path()
         if am.get(path, None) == f.filenode():
             return path
         if basemf and basemf.get(path, None) == f.filenode():
             return path
 
 
 def _dirstatecopies(repo, match=None):
     ds = repo.dirstate
     c = ds.copies().copy()
     for k in list(c):
         if ds[k] not in b'anm' or (match and not match(k)):
             del c[k]
     return c
 
 
 def _computeforwardmissing(a, b, match=None):
     """Computes which files are in b but not a.
     This is its own function so extensions can easily wrap this call to see what
     files _forwardcopies is about to process.
     """
     ma = a.manifest()
     mb = b.manifest()
     return mb.filesnotin(ma, match=match)
 
 
 def usechangesetcentricalgo(repo):
     """Checks if we should use changeset-centric copy algorithms"""
     if repo.filecopiesmode == b'changeset-sidedata':
         return True
     readfrom = repo.ui.config(b'experimental', b'copies.read-from')
     changesetsource = (b'changeset-only', b'compatibility')
     return readfrom in changesetsource
 
 
 def _committedforwardcopies(a, b, base, match):
     """Like _forwardcopies(), but b.rev() cannot be None (working copy)"""
     # files might have to be traced back to the fctx parent of the last
     # one-side-only changeset, but not further back than that
     repo = a._repo
 
     if usechangesetcentricalgo(repo):
         return _changesetforwardcopies(a, b, match)
 
     debug = repo.ui.debugflag and repo.ui.configbool(b'devel', b'debug.copies')
     dbg = repo.ui.debug
     if debug:
         dbg(b'debug.copies:    looking into rename from %s to %s\n' % (a, b))
     am = a.manifest()
     basemf = None if base is None else base.manifest()
 
     # find where new files came from
     # we currently don't try to find where old files went, too expensive
     # this means we can miss a case like 'hg rm b; hg cp a b'
     cm = {}
 
     # Computing the forward missing is quite expensive on large manifests, since
     # it compares the entire manifests. We can optimize it in the common use
     # case of computing what copies are in a commit versus its parent (like
     # during a rebase or histedit). Note, we exclude merge commits from this
     # optimization, since the ctx.files() for a merge commit is not correct for
     # this comparison.
     forwardmissingmatch = match
     if b.p1() == a and b.p2().node() == nullid:
         filesmatcher = matchmod.exact(b.files())
         forwardmissingmatch = matchmod.intersectmatchers(match, filesmatcher)
     missing = _computeforwardmissing(a, b, match=forwardmissingmatch)
 
     ancestrycontext = a._repo.changelog.ancestors([b.rev()], inclusive=True)
 
     if debug:
         dbg(b'debug.copies:      missing files to search: %d\n' % len(missing))
 
     for f in sorted(missing):
         if debug:
             dbg(b'debug.copies:        tracing file: %s\n' % f)
         fctx = b[f]
         fctx._ancestrycontext = ancestrycontext
 
         if debug:
             start = util.timer()
         opath = _tracefile(fctx, am, basemf)
         if opath:
             if debug:
                 dbg(b'debug.copies:          rename of: %s\n' % opath)
             cm[f] = opath
         if debug:
             dbg(
                 b'debug.copies:          time: %f seconds\n'
                 % (util.timer() - start)
             )
     return cm
 
 
 def _revinfo_getter(repo, match):
     """returns a function that returns the following data given a <rev>"
 
     * p1: revision number of first parent
     * p2: revision number of first parent
     * changes: a ChangingFiles object
     """
     cl = repo.changelog
     parents = cl.parentrevs
     flags = cl.flags
 
     HASCOPIESINFO = flagutil.REVIDX_HASCOPIESINFO
 
     changelogrevision = cl.changelogrevision
 
     # A small cache to avoid doing the work twice for merges
     #
     # In the vast majority of cases, if we ask information for a revision
     # about 1 parent, we'll later ask it for the other. So it make sense to
     # keep the information around when reaching the first parent of a merge
     # and dropping it after it was provided for the second parents.
     #
     # It exists cases were only one parent of the merge will be walked. It
     # happens when the "destination" the copy tracing is descendant from a
     # new root, not common with the "source". In that case, we will only walk
     # through merge parents that are descendant of changesets common
     # between "source" and "destination".
     #
     # With the current case implementation if such changesets have a copy
     # information, we'll keep them in memory until the end of
     # _changesetforwardcopies. We don't expect the case to be frequent
     # enough to matters.
     #
     # In addition, it would be possible to reach pathological case, were
     # many first parent are met before any second parent is reached. In
     # that case the cache could grow. If this even become an issue one can
     # safely introduce a maximum cache size. This would trade extra CPU/IO
     # time to save memory.
     merge_caches = {}
 
     alwaysmatch = match.always()
 
     if rustmod is not None and alwaysmatch:
 
         def revinfo(rev):
             p1, p2 = parents(rev)
             value = None
             e = merge_caches.pop(rev, None)
             if e is not None:
                 return e
             if flags(rev) & HASCOPIESINFO:
                 raw = changelogrevision(rev)._sidedata.get(sidedatamod.SD_FILES)
             else:
                 raw = None
             value = (p1, p2, raw)
             if p1 != nullrev and p2 != nullrev:
                 # XXX some case we over cache, IGNORE
                 merge_caches[rev] = value
             return value
 
     else:
 
         def revinfo(rev):
             p1, p2 = parents(rev)
             value = None
             e = merge_caches.pop(rev, None)
             if e is not None:
                 return e
             changes = None
             if flags(rev) & HASCOPIESINFO:
                 changes = changelogrevision(rev).changes
             value = (p1, p2, changes)
             if p1 != nullrev and p2 != nullrev:
                 # XXX some case we over cache, IGNORE
                 merge_caches[rev] = value
             return value
 
     return revinfo
 
 
 def cached_is_ancestor(is_ancestor):
     """return a cached version of is_ancestor"""
     cache = {}
 
     def _is_ancestor(anc, desc):
         if anc > desc:
             return False
         elif anc == desc:
             return True
         key = (anc, desc)
         ret = cache.get(key)
         if ret is None:
             ret = cache[key] = is_ancestor(anc, desc)
         return ret
 
     return _is_ancestor
 
 
 def _changesetforwardcopies(a, b, match):
     if a.rev() in (nullrev, b.rev()):
         return {}
 
     repo = a.repo().unfiltered()
     children = {}
 
     cl = repo.changelog
     isancestor = cl.isancestorrev
-    missingrevs = cl.findmissingrevs(common=[a.rev()], heads=[b.rev()])
-    mrset = set(missingrevs)
+
+    # To track rename from "A" to B, we need to gather all parent → children
+    # edges that are contains in `::B` but not in `::A`.
+    #
+    #
+    # To do so, we need to gather all revisions exclusive¹ to "B" (ie¹: `::b -
+    # ::a`) and also all the "roots point", ie the parents of the exclusive set
+    # that belong to ::a. These are exactly all the revisions needed to express
+    # the parent → children we need to combine.
+    #
+    # [1] actually, we need to gather all the edges within `(::a)::b`, ie:
+    # excluding paths that leads to roots that are not ancestors of `a`. We
+    # keep this out of the explanation because it is hard enough without this special case..
+
+    parents = cl._uncheckedparentrevs
+    graph_roots = (nullrev, nullrev)
+
+    ancestors = cl.ancestors([a.rev()], inclusive=True)
+    revs = cl.findmissingrevs(common=[a.rev()], heads=[b.rev()])
     roots = set()
-    for r in missingrevs:
-        for p in cl.parentrevs(r):
-            if p == nullrev:
-                continue
-            if p not in children:
-                children[p] = [r]
-            else:
-                children[p].append(r)
-            if p not in mrset:
-                roots.add(p)
+    has_graph_roots = False
+
+    # iterate over `only(B, A)`
+    for r in revs:
+        ps = parents(r)
+        if ps == graph_roots:
+            has_graph_roots = True
+        else:
+            p1, p2 = ps
+
+            # find all the "root points" (see larger comment above)
+            if p1 != nullrev and p1 in ancestors:
+                roots.add(p1)
+            if p2 != nullrev and p2 in ancestors:
+                roots.add(p2)
     if not roots:
         # no common revision to track copies from
         return {}
-    min_root = min(roots)
-
-    from_head = set(
-        cl.reachableroots(min_root, [b.rev()], list(roots), includepath=True)
-    )
-
-    iterrevs = set(from_head)
-    iterrevs &= mrset
-    iterrevs.update(roots)
-    iterrevs.remove(b.rev())
-    revs = sorted(iterrevs)
+    if has_graph_roots:
+        # this deal with the special case mentionned in the [1] footnotes. We
+        # must filter out revisions that leads to non-common graphroots.
+        roots = list(roots)
+        m = min(roots)
+        h = [b.rev()]
+        roots_to_head = cl.reachableroots(m, h, roots, includepath=True)
+        roots_to_head = set(roots_to_head)
+        revs = [r for r in revs if r in roots_to_head]
 
     if repo.filecopiesmode == b'changeset-sidedata':
+        # When using side-data, we will process the edges "from" the children.
+        # We iterate over the childre, gathering previous collected data for
+        # the parents. Do know when the parents data is no longer necessary, we
+        # keep a counter of how many children each revision has.
+        #
+        # An interresting property of `children_count` is that it only contains
+        # revision that will be relevant for a edge of the graph. So if a
+        # children has parent not in `children_count`, that edges should not be
+        # processed.
+        children_count = dict((r, 0) for r in roots)
+        for r in revs:
+            for p in cl.parentrevs(r):
+                if p == nullrev:
+                    continue
+                children_count[r] = 0
+                if p in children_count:
+                    children_count[p] += 1
         revinfo = _revinfo_getter(repo, match)
         return _combine_changeset_copies(
-            revs, children, b.rev(), revinfo, match, isancestor
+            revs, children_count, b.rev(), revinfo, match, isancestor
         )
     else:
+        # When not using side-data, we will process the edges "from" the parent.
+        # so we need a full mapping of the parent -> children relation.
+        children = dict((r, []) for r in roots)
+        for r in revs:
+            for p in cl.parentrevs(r):
+                if p == nullrev:
+                    continue
+                children[r] = []
+                if p in children:
+                    children[p].append(r)
+        x = revs.pop()
+        assert x == b.rev()
+        revs.extend(roots)
+        revs.sort()
+
         revinfo = _revinfo_getter_extra(repo)
         return _combine_changeset_copies_extra(
             revs, children, b.rev(), revinfo, match, isancestor
         )
 
 
 def _combine_changeset_copies(
-    revs, children, targetrev, revinfo, match, isancestor
+    revs, children_count, targetrev, revinfo, match, isancestor
 ):
     """combine the copies information for each item of iterrevs
 
     revs: sorted iterable of revision to visit
-    children: a {parent: [children]} mapping.
+    children_count: a {parent: <number-of-relevant-children>} mapping.
     targetrev: the final copies destination revision (not in iterrevs)
     revinfo(rev): a function that return (p1, p2, p1copies, p2copies, removed)
     match: a matcher
 
     It returns the aggregated copies information for `targetrev`.
     """
 
     alwaysmatch = match.always()
 
     if rustmod is not None and alwaysmatch:
         return rustmod.combine_changeset_copies(
-            list(revs), children, targetrev, revinfo, isancestor
+            list(revs), children_count, targetrev, revinfo, isancestor
         )
 
     isancestor = cached_is_ancestor(isancestor)
 
     all_copies = {}
-    # iterate over all the "parent" side of copy tracing "edge"
-    for r in revs:
-        # fetch potential previously computed data for that parent
-        copies = all_copies.pop(r, None)
-        if copies is None:
-            # this is a root
-            copies = {}
+    # iterate over all the "children" side of copy tracing "edge"
+    for current_rev in revs:
+        p1, p2, changes = revinfo(current_rev)
+        current_copies = None
 
-        # iterate over all known children to chain the existing data with the
+        # iterate over all parents to chain the existing data with the
         # data from the parent → child edge.
-        for i, c in enumerate(children[r]):
-            p1, p2, changes = revinfo(c)
-            childcopies = {}
+        for parent, parent_rev in ((1, p1), (2, p2)):
+            if parent_rev == nullrev:
+                continue
+            remaining_children = children_count.get(parent_rev)
+            if remaining_children is None:
+                continue
+            remaining_children -= 1
+            children_count[parent_rev] = remaining_children
+            if remaining_children:
+                copies = all_copies.get(parent_rev, None)
+            else:
+                copies = all_copies.pop(parent_rev, None)
 
-            # select the right parent → child edge
-            if r == p1:
-                parent = 1
-                if changes is not None:
+            if copies is None:
+                # this is a root
+                copies = {}
+
+            newcopies = copies
+            # chain the data in the edge with the existing data
+            if changes is not None:
+                childcopies = {}
+                if parent == 1:
                     childcopies = changes.copied_from_p1
-            else:
-                assert r == p2
-                parent = 2
-                if changes is not None:
+                elif parent == 2:
                     childcopies = changes.copied_from_p2
-            if not alwaysmatch:
-                childcopies = {
-                    dst: src for dst, src in childcopies.items() if match(dst)
-                }
 
-            # chain the data in the edge with the existing data
-            newcopies = copies
-            if childcopies:
-                newcopies = copies.copy()
-                for dest, source in pycompat.iteritems(childcopies):
-                    prev = copies.get(source)
-                    if prev is not None and prev[1] is not None:
-                        source = prev[1]
-                    newcopies[dest] = (c, source)
-                assert newcopies is not copies
-            if changes is not None and changes.removed:
-                if newcopies is copies:
+                if not alwaysmatch:
+                    childcopies = {
+                        dst: src
+                        for dst, src in childcopies.items()
+                        if match(dst)
+                    }
+                if childcopies:
                     newcopies = copies.copy()
-                for f in changes.removed:
-                    if f in newcopies:
-                        if newcopies is copies:
-                            # copy on write to avoid affecting potential other
-                            # branches.  when there are no other branches, this
-                            # could be avoided.
-                            newcopies = copies.copy()
-                        newcopies[f] = (c, None)
+                    for dest, source in pycompat.iteritems(childcopies):
+                        prev = copies.get(source)
+                        if prev is not None and prev[1] is not None:
+                            source = prev[1]
+                        newcopies[dest] = (current_rev, source)
+                    assert newcopies is not copies
+                if changes.removed:
+                    if newcopies is copies:
+                        newcopies = copies.copy()
+                    for f in changes.removed:
+                        if f in newcopies:
+                            if newcopies is copies:
+                                # copy on write to avoid affecting potential other
+                                # branches.  when there are no other branches, this
+                                # could be avoided.
+                                newcopies = copies.copy()
+                            newcopies[f] = (current_rev, None)
 
             # check potential need to combine the data from another parent (for
             # that child). See comment below for details.
-            othercopies = all_copies.get(c)
-            if othercopies is None:
-                all_copies[c] = newcopies
-            elif newcopies is othercopies:
+            if current_copies is None:
+                current_copies = newcopies
+            elif current_copies is newcopies:
                 # nothing to merge:
                 pass
             else:
                 # we are the second parent to work on c, we need to merge our
                 # work with the other.
                 #
                 # In case of conflict, parent 1 take precedence over parent 2.
                 # This is an arbitrary choice made anew when implementing
                 # changeset based copies. It was made without regards with
                 # potential filelog related behavior.
-                if parent == 1:
-                    if newcopies is copies:
-                        newcopies = copies.copy()
-                    minor, major = othercopies, newcopies
-                else:
-                    # we do not know if the other dict is a copy or not, so we
-                    # need to blindly copy it. Future change should make this
-                    # unnecessary.
-                    minor, major = newcopies, othercopies.copy()
-                copies = _merge_copies_dict(minor, major, isancestor, changes)
-                all_copies[c] = copies
+                assert parent == 2
+                current_copies = _merge_copies_dict(
+                    newcopies, current_copies, isancestor, changes
+                )
+        all_copies[current_rev] = current_copies
 
     # filter out internal details and return a {dest: source mapping}
     final_copies = {}
     for dest, (tt, source) in all_copies[targetrev].items():
         if source is not None:
             final_copies[dest] = source
     return final_copies
 
 
 def _merge_copies_dict(minor, major, isancestor, changes):
     """merge two copies-mapping together, minor and major
 
     In case of conflict, value from "major" will be picked.
 
     - `isancestors(low_rev, high_rev)`: callable return True if `low_rev` is an
                                         ancestors of `high_rev`,
 
     - `ismerged(path)`: callable return True if `path` have been merged in the
                         current revision,
 
     return the resulting dict (in practice, the "minor" object, updated)
     """
     for dest, value in major.items():
         other = minor.get(dest)
         if other is None:
             minor[dest] = value
         else:
             new_tt = value[0]
             other_tt = other[0]
             if value[1] == other[1]:
                 continue
             # content from "major" wins, unless it is older
             # than the branch point or there is a merge
             if new_tt == other_tt:
                 minor[dest] = value
             elif (
                 changes is not None
                 and value[1] is None
                 and dest in changes.salvaged
             ):
                 pass
             elif (
                 changes is not None
                 and other[1] is None
                 and dest in changes.salvaged
             ):
                 minor[dest] = value
             elif changes is not None and dest in changes.merged:
                 minor[dest] = value
             elif not isancestor(new_tt, other_tt):
                 if value[1] is not None:
                     minor[dest] = value
                 elif isancestor(other_tt, new_tt):
                     minor[dest] = value
     return minor
 
 
 def _revinfo_getter_extra(repo):
     """return a function that return multiple data given a <rev>"i
 
     * p1: revision number of first parent
     * p2: revision number of first parent
     * p1copies: mapping of copies from p1
     * p2copies: mapping of copies from p2
     * removed: a list of removed files
     * ismerged: a callback to know if file was merged in that revision
     """
     cl = repo.changelog
     parents = cl.parentrevs
 
     def get_ismerged(rev):
         ctx = repo[rev]
 
         def ismerged(path):
             if path not in ctx.files():
                 return False
             fctx = ctx[path]
             parents = fctx._filelog.parents(fctx._filenode)
             nb_parents = 0
             for n in parents:
                 if n != nullid:
                     nb_parents += 1
             return nb_parents >= 2
 
         return ismerged
 
     def revinfo(rev):
         p1, p2 = parents(rev)
         ctx = repo[rev]
         p1copies, p2copies = ctx._copies
         removed = ctx.filesremoved()
         return p1, p2, p1copies, p2copies, removed, get_ismerged(rev)
 
     return revinfo
 
 
 def _combine_changeset_copies_extra(
     revs, children, targetrev, revinfo, match, isancestor
 ):
     """version of `_combine_changeset_copies` that works with the Google
     specific "extra" based storage for copy information"""
     all_copies = {}
     alwaysmatch = match.always()
     for r in revs:
         copies = all_copies.pop(r, None)
         if copies is None:
             # this is a root
             copies = {}
         for i, c in enumerate(children[r]):
             p1, p2, p1copies, p2copies, removed, ismerged = revinfo(c)
             if r == p1:
                 parent = 1
                 childcopies = p1copies
             else:
                 assert r == p2
                 parent = 2
                 childcopies = p2copies
             if not alwaysmatch:
                 childcopies = {
                     dst: src for dst, src in childcopies.items() if match(dst)
                 }
             newcopies = copies
             if childcopies:
                 newcopies = copies.copy()
                 for dest, source in pycompat.iteritems(childcopies):
                     prev = copies.get(source)
                     if prev is not None and prev[1] is not None:
                         source = prev[1]
                     newcopies[dest] = (c, source)
                 assert newcopies is not copies
             for f in removed:
                 if f in newcopies:
                     if newcopies is copies:
                         # copy on write to avoid affecting potential other
                         # branches.  when there are no other branches, this
                         # could be avoided.
                         newcopies = copies.copy()
                     newcopies[f] = (c, None)
             othercopies = all_copies.get(c)
             if othercopies is None:
                 all_copies[c] = newcopies
             else:
                 # we are the second parent to work on c, we need to merge our
                 # work with the other.
                 #
                 # In case of conflict, parent 1 take precedence over parent 2.
                 # This is an arbitrary choice made anew when implementing
                 # changeset based copies. It was made without regards with
                 # potential filelog related behavior.
                 if parent == 1:
                     _merge_copies_dict_extra(
                         othercopies, newcopies, isancestor, ismerged
                     )
                 else:
                     _merge_copies_dict_extra(
                         newcopies, othercopies, isancestor, ismerged
                     )
                     all_copies[c] = newcopies
 
     final_copies = {}
     for dest, (tt, source) in all_copies[targetrev].items():
         if source is not None:
             final_copies[dest] = source
     return final_copies
 
 
 def _merge_copies_dict_extra(minor, major, isancestor, ismerged):
     """version of `_merge_copies_dict` that works with the Google
     specific "extra" based storage for copy information"""
     for dest, value in major.items():
         other = minor.get(dest)
         if other is None:
             minor[dest] = value
         else:
             new_tt = value[0]
             other_tt = other[0]
             if value[1] == other[1]:
                 continue
             # content from "major" wins, unless it is older
             # than the branch point or there is a merge
             if (
                 new_tt == other_tt
                 or not isancestor(new_tt, other_tt)
                 or ismerged(dest)
             ):
                 minor[dest] = value
 
 
 def _forwardcopies(a, b, base=None, match=None):
     """find {dst@b: src@a} copy mapping where a is an ancestor of b"""
 
     if base is None:
         base = a
     match = a.repo().narrowmatch(match)
     # check for working copy
     if b.rev() is None:
         cm = _committedforwardcopies(a, b.p1(), base, match)
         # combine copies from dirstate if necessary
         copies = _chain(cm, _dirstatecopies(b._repo, match))
     else:
         copies = _committedforwardcopies(a, b, base, match)
     return copies
 
 
 def _backwardrenames(a, b, match):
     if a._repo.ui.config(b'experimental', b'copytrace') == b'off':
         return {}
 
     # Even though we're not taking copies into account, 1:n rename situations
     # can still exist (e.g. hg cp a b; hg mv a c). In those cases we
     # arbitrarily pick one of the renames.
     # We don't want to pass in "match" here, since that would filter
     # the destination by it. Since we're reversing the copies, we want
     # to filter the source instead.
     f = _forwardcopies(b, a)
     r = {}
     for k, v in sorted(pycompat.iteritems(f)):
         if match and not match(v):
             continue
         # remove copies
         if v in a:
             continue
         r[v] = k
     return r
 
 
 def pathcopies(x, y, match=None):
     """find {dst@y: src@x} copy mapping for directed compare"""
     repo = x._repo
     debug = repo.ui.debugflag and repo.ui.configbool(b'devel', b'debug.copies')
     if debug:
         repo.ui.debug(
             b'debug.copies: searching copies from %s to %s\n' % (x, y)
         )
     if x == y or not x or not y:
         return {}
     if y.rev() is None and x == y.p1():
         if debug:
             repo.ui.debug(b'debug.copies: search mode: dirstate\n')
         # short-circuit to avoid issues with merge states
         return _dirstatecopies(repo, match)
     a = y.ancestor(x)
     if a == x:
         if debug:
             repo.ui.debug(b'debug.copies: search mode: forward\n')
         copies = _forwardcopies(x, y, match=match)
     elif a == y:
         if debug:
             repo.ui.debug(b'debug.copies: search mode: backward\n')
         copies = _backwardrenames(x, y, match=match)
     else:
         if debug:
             repo.ui.debug(b'debug.copies: search mode: combined\n')
         base = None
         if a.rev() != nullrev:
             base = x
         copies = _chain(
             _backwardrenames(x, a, match=match),
             _forwardcopies(a, y, base, match=match),
         )
     _filter(x, y, copies)
     return copies
 
 
 def mergecopies(repo, c1, c2, base):
     """
     Finds moves and copies between context c1 and c2 that are relevant for
     merging. 'base' will be used as the merge base.
 
     Copytracing is used in commands like rebase, merge, unshelve, etc to merge
     files that were moved/ copied in one merge parent and modified in another.
     For example:
 
     o          ---> 4 another commit
     |
     |   o      ---> 3 commit that modifies a.txt
     |  /
     o /        ---> 2 commit that moves a.txt to b.txt
     |/
     o          ---> 1 merge base
 
     If we try to rebase revision 3 on revision 4, since there is no a.txt in
     revision 4, and if user have copytrace disabled, we prints the following
     message:
 
     ```other changed <file> which local deleted```
 
     Returns a tuple where:
 
     "branch_copies" an instance of branch_copies.
 
     "diverge" is a mapping of source name -> list of destination names
     for divergent renames.
 
     This function calls different copytracing algorithms based on config.
     """
     # avoid silly behavior for update from empty dir
     if not c1 or not c2 or c1 == c2:
         return branch_copies(), branch_copies(), {}
 
     narrowmatch = c1.repo().narrowmatch()
 
     # avoid silly behavior for parent -> working dir
     if c2.node() is None and c1.node() == repo.dirstate.p1():
         return (
             branch_copies(_dirstatecopies(repo, narrowmatch)),
             branch_copies(),
             {},
         )
 
     copytracing = repo.ui.config(b'experimental', b'copytrace')
     if stringutil.parsebool(copytracing) is False:
         # stringutil.parsebool() returns None when it is unable to parse the
         # value, so we should rely on making sure copytracing is on such cases
         return branch_copies(), branch_copies(), {}
 
     if usechangesetcentricalgo(repo):
         # The heuristics don't make sense when we need changeset-centric algos
         return _fullcopytracing(repo, c1, c2, base)
 
     # Copy trace disabling is explicitly below the node == p1 logic above
     # because the logic above is required for a simple copy to be kept across a
     # rebase.
     if copytracing == b'heuristics':
         # Do full copytracing if only non-public revisions are involved as
         # that will be fast enough and will also cover the copies which could
         # be missed by heuristics
         if _isfullcopytraceable(repo, c1, base):
             return _fullcopytracing(repo, c1, c2, base)
         return _heuristicscopytracing(repo, c1, c2, base)
     else:
         return _fullcopytracing(repo, c1, c2, base)
 
 
 def _isfullcopytraceable(repo, c1, base):
     """Checks that if base, source and destination are all no-public branches,
     if yes let's use the full copytrace algorithm for increased capabilities
     since it will be fast enough.
 
     `experimental.copytrace.sourcecommitlimit` can be used to set a limit for
     number of changesets from c1 to base such that if number of changesets are
     more than the limit, full copytracing algorithm won't be used.
     """
     if c1.rev() is None:
         c1 = c1.p1()
     if c1.mutable() and base.mutable():
         sourcecommitlimit = repo.ui.configint(
             b'experimental', b'copytrace.sourcecommitlimit'
         )
         commits = len(repo.revs(b'%d::%d', base.rev(), c1.rev()))
         return commits < sourcecommitlimit
     return False
 
 
 def _checksinglesidecopies(
     src, dsts1, m1, m2, mb, c2, base, copy, renamedelete
 ):
     if src not in m2:
         # deleted on side 2
         if src not in m1:
             # renamed on side 1, deleted on side 2
             renamedelete[src] = dsts1
     elif src not in mb:
         # Work around the "short-circuit to avoid issues with merge states"
         # thing in pathcopies(): pathcopies(x, y) can return a copy where the
         # destination doesn't exist in y.
         pass
     elif mb[src] != m2[src] and not _related(c2[src], base[src]):
         return
     elif mb[src] != m2[src] or mb.flags(src) != m2.flags(src):
         # modified on side 2
         for dst in dsts1:
             copy[dst] = src
 
 
 class branch_copies(object):
     """Information about copies made on one side of a merge/graft.
 
     "copy" is a mapping from destination name -> source name,
     where source is in c1 and destination is in c2 or vice-versa.
 
     "movewithdir" is a mapping from source name -> destination name,
     where the file at source present in one context but not the other
     needs to be moved to destination by the merge process, because the
     other context moved the directory it is in.
 
     "renamedelete" is a mapping of source name -> list of destination
     names for files deleted in c1 that were renamed in c2 or vice-versa.
 
     "dirmove" is a mapping of detected source dir -> destination dir renames.
     This is needed for handling changes to new files previously grafted into
     renamed directories.
     """
 
     def __init__(
         self, copy=None, renamedelete=None, dirmove=None, movewithdir=None
     ):
         self.copy = {} if copy is None else copy
         self.renamedelete = {} if renamedelete is None else renamedelete
         self.dirmove = {} if dirmove is None else dirmove
         self.movewithdir = {} if movewithdir is None else movewithdir
 
     def __repr__(self):
         return '<branch_copies\n  copy=%r\n  renamedelete=%r\n  dirmove=%r\n  movewithdir=%r\n>' % (
             self.copy,
             self.renamedelete,
             self.dirmove,
             self.movewithdir,
         )
 
 
 def _fullcopytracing(repo, c1, c2, base):
     """The full copytracing algorithm which finds all the new files that were
     added from merge base up to the top commit and for each file it checks if
     this file was copied from another file.
 
     This is pretty slow when a lot of changesets are involved but will track all
     the copies.
     """
     m1 = c1.manifest()
     m2 = c2.manifest()
     mb = base.manifest()
 
     copies1 = pathcopies(base, c1)
     copies2 = pathcopies(base, c2)
 
     if not (copies1 or copies2):
         return branch_copies(), branch_copies(), {}
 
     inversecopies1 = {}
     inversecopies2 = {}
     for dst, src in copies1.items():
         inversecopies1.setdefault(src, []).append(dst)
     for dst, src in copies2.items():
         inversecopies2.setdefault(src, []).append(dst)
 
     copy1 = {}
     copy2 = {}
     diverge = {}
     renamedelete1 = {}
     renamedelete2 = {}
     allsources = set(inversecopies1) | set(inversecopies2)
     for src in allsources:
         dsts1 = inversecopies1.get(src)
         dsts2 = inversecopies2.get(src)
         if dsts1 and dsts2:
             # copied/renamed on both sides
             if src not in m1 and src not in m2:
                 # renamed on both sides
                 dsts1 = set(dsts1)
                 dsts2 = set(dsts2)
                 # If there's some overlap in the rename destinations, we
                 # consider it not divergent. For example, if side 1 copies 'a'
                 # to 'b' and 'c' and deletes 'a', and side 2 copies 'a' to 'c'
                 # and 'd' and deletes 'a'.
                 if dsts1 & dsts2:
                     for dst in dsts1 & dsts2:
                         copy1[dst] = src
                         copy2[dst] = src
                 else:
                     diverge[src] = sorted(dsts1 | dsts2)
             elif src in m1 and src in m2:
                 # copied on both sides
                 dsts1 = set(dsts1)
                 dsts2 = set(dsts2)
                 for dst in dsts1 & dsts2:
                     copy1[dst] = src
                     copy2[dst] = src
             # TODO: Handle cases where it was renamed on one side and copied
             # on the other side
         elif dsts1:
             # copied/renamed only on side 1
             _checksinglesidecopies(
                 src, dsts1, m1, m2, mb, c2, base, copy1, renamedelete1
             )
         elif dsts2:
             # copied/renamed only on side 2
             _checksinglesidecopies(
                 src, dsts2, m2, m1, mb, c1, base, copy2, renamedelete2
             )
 
     # find interesting file sets from manifests
     cache = []
 
     def _get_addedfiles(idx):
         if not cache:
             addedinm1 = m1.filesnotin(mb, repo.narrowmatch())
             addedinm2 = m2.filesnotin(mb, repo.narrowmatch())
             u1 = sorted(addedinm1 - addedinm2)
             u2 = sorted(addedinm2 - addedinm1)
             cache.extend((u1, u2))
         return cache[idx]
 
     u1fn = lambda: _get_addedfiles(0)
     u2fn = lambda: _get_addedfiles(1)
     if repo.ui.debugflag:
         u1 = u1fn()
         u2 = u2fn()
 
         header = b"  unmatched files in %s"
         if u1:
             repo.ui.debug(
                 b"%s:\n   %s\n" % (header % b'local', b"\n   ".join(u1))
             )
         if u2:
             repo.ui.debug(
                 b"%s:\n   %s\n" % (header % b'other', b"\n   ".join(u2))
             )
 
         renamedeleteset = set()
         divergeset = set()
         for dsts in diverge.values():
             divergeset.update(dsts)
         for dsts in renamedelete1.values():
             renamedeleteset.update(dsts)
         for dsts in renamedelete2.values():
             renamedeleteset.update(dsts)
 
         repo.ui.debug(
             b"  all copies found (* = to merge, ! = divergent, "
             b"% = renamed and deleted):\n"
         )
         for side, copies in ((b"local", copies1), (b"remote", copies2)):
             if not copies:
                 continue
             repo.ui.debug(b"   on %s side:\n" % side)
             for f in sorted(copies):
                 note = b""
                 if f in copy1 or f in copy2:
                     note += b"*"
                 if f in divergeset:
                     note += b"!"
                 if f in renamedeleteset:
                     note += b"%"
                 repo.ui.debug(
                     b"    src: '%s' -> dst: '%s' %s\n" % (copies[f], f, note)
                 )
         del renamedeleteset
         del divergeset
 
     repo.ui.debug(b"  checking for directory renames\n")
 
     dirmove1, movewithdir2 = _dir_renames(repo, c1, copy1, copies1, u2fn)
     dirmove2, movewithdir1 = _dir_renames(repo, c2, copy2, copies2, u1fn)
 
     branch_copies1 = branch_copies(copy1, renamedelete1, dirmove1, movewithdir1)
     branch_copies2 = branch_copies(copy2, renamedelete2, dirmove2, movewithdir2)
 
     return branch_copies1, branch_copies2, diverge
 
 
 def _dir_renames(repo, ctx, copy, fullcopy, addedfilesfn):
     """Finds moved directories and files that should move with them.
 
     ctx: the context for one of the sides
     copy: files copied on the same side (as ctx)
     fullcopy: files copied on the same side (as ctx), including those that
               merge.manifestmerge() won't care about
     addedfilesfn: function returning added files on the other side (compared to
                   ctx)
     """
     # generate a directory move map
     invalid = set()
     dirmove = {}
 
     # examine each file copy for a potential directory move, which is
     # when all the files in a directory are moved to a new directory
     for dst, src in pycompat.iteritems(fullcopy):
         dsrc, ddst = pathutil.dirname(src), pathutil.dirname(dst)
         if dsrc in invalid:
             # already seen to be uninteresting
             continue
         elif ctx.hasdir(dsrc) and ctx.hasdir(ddst):
             # directory wasn't entirely moved locally
             invalid.add(dsrc)
         elif dsrc in dirmove and dirmove[dsrc] != ddst:
             # files from the same directory moved to two different places
             invalid.add(dsrc)
         else:
             # looks good so far
             dirmove[dsrc] = ddst
 
     for i in invalid:
         if i in dirmove:
             del dirmove[i]
     del invalid
 
     if not dirmove:
         return {}, {}
 
     dirmove = {k + b"/": v + b"/" for k, v in pycompat.iteritems(dirmove)}
 
     for d in dirmove:
         repo.ui.debug(
             b"   discovered dir src: '%s' -> dst: '%s'\n" % (d, dirmove[d])
         )
 
     movewithdir = {}
     # check unaccounted nonoverlapping files against directory moves
     for f in addedfilesfn():
         if f not in fullcopy:
             for d in dirmove:
                 if f.startswith(d):
                     # new file added in a directory that was moved, move it
                     df = dirmove[d] + f[len(d) :]
                     if df not in copy:
                         movewithdir[f] = df
                         repo.ui.debug(
                             b"   pending file src: '%s' -> dst: '%s'\n"
                             % (f, df)
                         )
                     break
 
     return dirmove, movewithdir
 
 
 def _heuristicscopytracing(repo, c1, c2, base):
     """Fast copytracing using filename heuristics
 
     Assumes that moves or renames are of following two types:
 
     1) Inside a directory only (same directory name but different filenames)
     2) Move from one directory to another
                     (same filenames but different directory names)
 
     Works only when there are no merge commits in the "source branch".
     Source branch is commits from base up to c2 not including base.
 
     If merge is involved it fallbacks to _fullcopytracing().
 
     Can be used by setting the following config:
 
         [experimental]
         copytrace = heuristics
 
     In some cases the copy/move candidates found by heuristics can be very large
     in number and that will make the algorithm slow. The number of possible
     candidates to check can be limited by using the config
     `experimental.copytrace.movecandidateslimit` which defaults to 100.
     """
 
     if c1.rev() is None:
         c1 = c1.p1()
     if c2.rev() is None:
         c2 = c2.p1()
 
     changedfiles = set()
     m1 = c1.manifest()
     if not repo.revs(b'%d::%d', base.rev(), c2.rev()):
         # If base is not in c2 branch, we switch to fullcopytracing
         repo.ui.debug(
             b"switching to full copytracing as base is not "
             b"an ancestor of c2\n"
         )
         return _fullcopytracing(repo, c1, c2, base)
 
     ctx = c2
     while ctx != base:
         if len(ctx.parents()) == 2:
             # To keep things simple let's not handle merges
             repo.ui.debug(b"switching to full copytracing because of merges\n")
             return _fullcopytracing(repo, c1, c2, base)
         changedfiles.update(ctx.files())
         ctx = ctx.p1()
 
     copies2 = {}
     cp = _forwardcopies(base, c2)
     for dst, src in pycompat.iteritems(cp):
         if src in m1:
             copies2[dst] = src
 
     # file is missing if it isn't present in the destination, but is present in
     # the base and present in the source.
     # Presence in the base is important to exclude added files, presence in the
     # source is important to exclude removed files.
     filt = lambda f: f not in m1 and f in base and f in c2
     missingfiles = [f for f in changedfiles if filt(f)]
 
     copies1 = {}
     if missingfiles:
         basenametofilename = collections.defaultdict(list)
         dirnametofilename = collections.defaultdict(list)
 
         for f in m1.filesnotin(base.manifest()):
             basename = os.path.basename(f)
             dirname = os.path.dirname(f)
             basenametofilename[basename].append(f)
             dirnametofilename[dirname].append(f)
 
         for f in missingfiles:
             basename = os.path.basename(f)
             dirname = os.path.dirname(f)
             samebasename = basenametofilename[basename]
             samedirname = dirnametofilename[dirname]
             movecandidates = samebasename + samedirname
             # f is guaranteed to be present in c2, that's why
             # c2.filectx(f) won't fail
             f2 = c2.filectx(f)
             # we can have a lot of candidates which can slow down the heuristics
             # config value to limit the number of candidates moves to check
             maxcandidates = repo.ui.configint(
                 b'experimental', b'copytrace.movecandidateslimit'
             )
 
             if len(movecandidates) > maxcandidates:
                 repo.ui.status(
                     _(
                         b"skipping copytracing for '%s', more "
                         b"candidates than the limit: %d\n"
                     )
                     % (f, len(movecandidates))
                 )
                 continue
 
             for candidate in movecandidates:
                 f1 = c1.filectx(candidate)
                 if _related(f1, f2):
                     # if there are a few related copies then we'll merge
                     # changes into all of them. This matches the behaviour
                     # of upstream copytracing
                     copies1[candidate] = f
 
     return branch_copies(copies1), branch_copies(copies2), {}
 
 
 def _related(f1, f2):
     """return True if f1 and f2 filectx have a common ancestor
 
     Walk back to common ancestor to see if the two files originate
     from the same file. Since workingfilectx's rev() is None it messes
     up the integer comparison logic, hence the pre-step check for
     None (f1 and f2 can only be workingfilectx's initially).
     """
 
     if f1 == f2:
         return True  # a match
 
     g1, g2 = f1.ancestors(), f2.ancestors()
     try:
         f1r, f2r = f1.linkrev(), f2.linkrev()
 
         if f1r is None:
             f1 = next(g1)
         if f2r is None:
             f2 = next(g2)
 
         while True:
             f1r, f2r = f1.linkrev(), f2.linkrev()
             if f1r > f2r:
                 f1 = next(g1)
             elif f2r > f1r:
                 f2 = next(g2)
             else:  # f1 and f2 point to files in the same linkrev
                 return f1 == f2  # true if they point to the same file
     except StopIteration:
         return False
 
 
 def graftcopies(wctx, ctx, base):
     """reproduce copies between base and ctx in the wctx
 
     Unlike mergecopies(), this function will only consider copies between base
     and ctx; it will ignore copies between base and wctx. Also unlike
     mergecopies(), this function will apply copies to the working copy (instead
     of just returning information about the copies). That makes it cheaper
     (especially in the common case of base==ctx.p1()) and useful also when
     experimental.copytrace=off.
 
     merge.update() will have already marked most copies, but it will only
     mark copies if it thinks the source files are related (see
     merge._related()). It will also not mark copies if the file wasn't modified
     on the local side. This function adds the copies that were "missed"
     by merge.update().
     """
     new_copies = pathcopies(base, ctx)
     _filter(wctx.p1(), wctx, new_copies)
     for dst, src in pycompat.iteritems(new_copies):
         wctx[dst].markcopied(src)

rust/hg-core/src/copy_tracing.rs

@@ -1,665 +1,704
 use crate::utils::hg_path::HgPath;
 use crate::utils::hg_path::HgPathBuf;
 use crate::Revision;
+use crate::NULL_REVISION;
 
 use im_rc::ordmap::DiffItem;
 use im_rc::ordmap::OrdMap;
 
 use std::cmp::Ordering;
 use std::collections::HashMap;
 use std::convert::TryInto;
 
 pub type PathCopies = HashMap<HgPathBuf, HgPathBuf>;
 
 type PathToken = HgPathBuf;
 
 #[derive(Clone, Debug, PartialEq)]
 struct TimeStampedPathCopy {
     /// revision at which the copy information was added
     rev: Revision,
     /// the copy source, (Set to None in case of deletion of the associated
     /// key)
     path: Option<PathToken>,
 }
 
 /// maps CopyDestination to Copy Source (+ a "timestamp" for the operation)
 type TimeStampedPathCopies = OrdMap<PathToken, TimeStampedPathCopy>;
 
 /// hold parent 1, parent 2 and relevant files actions.
 pub type RevInfo<'a> = (Revision, Revision, ChangedFiles<'a>);
 
 /// represent the files affected by a changesets
 ///
 /// This hold a subset of mercurial.metadata.ChangingFiles as we do not need
 /// all the data categories tracked by it.
 /// This hold a subset of mercurial.metadata.ChangingFiles as we do not need
 /// all the data categories tracked by it.
 pub struct ChangedFiles<'a> {
     nb_items: u32,
     index: &'a [u8],
     data: &'a [u8],
 }
 
 /// Represent active changes that affect the copy tracing.
 enum Action<'a> {
     /// The parent ? children edge is removing a file
     ///
     /// (actually, this could be the edge from the other parent, but it does
     /// not matters)
     Removed(&'a HgPath),
     /// The parent ? children edge introduce copy information between (dest,
     /// source)
     Copied(&'a HgPath, &'a HgPath),
 }
 
 /// This express the possible "special" case we can get in a merge
 ///
 /// See mercurial/metadata.py for details on these values.
 #[derive(PartialEq)]
 enum MergeCase {
     /// Merged: file had history on both side that needed to be merged
     Merged,
     /// Salvaged: file was candidate for deletion, but survived the merge
     Salvaged,
     /// Normal: Not one of the two cases above
     Normal,
 }
 
 type FileChange<'a> = (u8, &'a HgPath, &'a HgPath);
 
 const EMPTY: &[u8] = b"";
 const COPY_MASK: u8 = 3;
 const P1_COPY: u8 = 2;
 const P2_COPY: u8 = 3;
 const ACTION_MASK: u8 = 28;
 const REMOVED: u8 = 12;
 const MERGED: u8 = 8;
 const SALVAGED: u8 = 16;
 
 impl<'a> ChangedFiles<'a> {
     const INDEX_START: usize = 4;
     const ENTRY_SIZE: u32 = 9;
     const FILENAME_START: u32 = 1;
     const COPY_SOURCE_START: u32 = 5;
 
     pub fn new(data: &'a [u8]) -> Self {
         assert!(
             data.len() >= 4,
             "data size ({}) is too small to contain the header (4)",
             data.len()
         );
         let nb_items_raw: [u8; 4] = (&data[0..=3])
             .try_into()
             .expect("failed to turn 4 bytes into 4 bytes");
         let nb_items = u32::from_be_bytes(nb_items_raw);
 
         let index_size = (nb_items * Self::ENTRY_SIZE) as usize;
         let index_end = Self::INDEX_START + index_size;
 
         assert!(
             data.len() >= index_end,
             "data size ({}) is too small to fit the index_data ({})",
             data.len(),
             index_end
         );
 
         let ret = ChangedFiles {
             nb_items,
             index: &data[Self::INDEX_START..index_end],
             data: &data[index_end..],
         };
         let max_data = ret.filename_end(nb_items - 1) as usize;
         assert!(
             ret.data.len() >= max_data,
             "data size ({}) is too small to fit all data ({})",
             data.len(),
             index_end + max_data
         );
         ret
     }
 
     pub fn new_empty() -> Self {
         ChangedFiles {
             nb_items: 0,
             index: EMPTY,
             data: EMPTY,
         }
     }
 
     /// internal function to return an individual entry at a given index
     fn entry(&'a self, idx: u32) -> FileChange<'a> {
         if idx >= self.nb_items {
             panic!(
                 "index for entry is higher that the number of file {} >= {}",
                 idx, self.nb_items
             )
         }
         let flags = self.flags(idx);
         let filename = self.filename(idx);
         let copy_idx = self.copy_idx(idx);
         let copy_source = self.filename(copy_idx);
         (flags, filename, copy_source)
     }
 
     /// internal function to return the filename of the entry at a given index
     fn filename(&self, idx: u32) -> &HgPath {
         let filename_start;
         if idx == 0 {
             filename_start = 0;
         } else {
             filename_start = self.filename_end(idx - 1)
         }
         let filename_end = self.filename_end(idx);
         let filename_start = filename_start as usize;
         let filename_end = filename_end as usize;
         HgPath::new(&self.data[filename_start..filename_end])
     }
 
     /// internal function to return the flag field of the entry at a given
     /// index
     fn flags(&self, idx: u32) -> u8 {
         let idx = idx as usize;
         self.index[idx * (Self::ENTRY_SIZE as usize)]
     }
 
     /// internal function to return the end of a filename part at a given index
     fn filename_end(&self, idx: u32) -> u32 {
         let start = (idx * Self::ENTRY_SIZE) + Self::FILENAME_START;
         let end = (idx * Self::ENTRY_SIZE) + Self::COPY_SOURCE_START;
         let start = start as usize;
         let end = end as usize;
         let raw = (&self.index[start..end])
             .try_into()
             .expect("failed to turn 4 bytes into 4 bytes");
         u32::from_be_bytes(raw)
     }
 
     /// internal function to return index of the copy source of the entry at a
     /// given index
     fn copy_idx(&self, idx: u32) -> u32 {
         let start = (idx * Self::ENTRY_SIZE) + Self::COPY_SOURCE_START;
         let end = (idx + 1) * Self::ENTRY_SIZE;
         let start = start as usize;
         let end = end as usize;
         let raw = (&self.index[start..end])
             .try_into()
             .expect("failed to turn 4 bytes into 4 bytes");
         u32::from_be_bytes(raw)
     }
 
     /// Return an iterator over all the `Action` in this instance.
     fn iter_actions(&self, parent: Parent) -> ActionsIterator {
         ActionsIterator {
             changes: &self,
             parent: parent,
             current: 0,
         }
     }
 
     /// return the MergeCase value associated with a filename
     fn get_merge_case(&self, path: &HgPath) -> MergeCase {
         if self.nb_items == 0 {
             return MergeCase::Normal;
         }
         let mut low_part = 0;
         let mut high_part = self.nb_items;
 
         while low_part < high_part {
             let cursor = (low_part + high_part - 1) / 2;
             let (flags, filename, _source) = self.entry(cursor);
             match path.cmp(filename) {
                 Ordering::Less => low_part = cursor + 1,
                 Ordering::Greater => high_part = cursor,
                 Ordering::Equal => {
                     return match flags & ACTION_MASK {
                         MERGED => MergeCase::Merged,
                         SALVAGED => MergeCase::Salvaged,
                         _ => MergeCase::Normal,
                     };
                 }
             }
         }
         MergeCase::Normal
     }
 }
 
 /// A struct responsible for answering "is X ancestors of Y" quickly
 ///
 /// The structure will delegate ancestors call to a callback, and cache the
 /// result.
 #[derive(Debug)]
 struct AncestorOracle<'a, A: Fn(Revision, Revision) -> bool> {
     inner: &'a A,
     pairs: HashMap<(Revision, Revision), bool>,
 }
 
 impl<'a, A: Fn(Revision, Revision) -> bool> AncestorOracle<'a, A> {
     fn new(func: &'a A) -> Self {
         Self {
             inner: func,
             pairs: HashMap::default(),
         }
     }
 
     /// returns `true` if `anc` is an ancestors of `desc`, `false` otherwise
     fn is_ancestor(&mut self, anc: Revision, desc: Revision) -> bool {
         if anc > desc {
             false
         } else if anc == desc {
             true
         } else {
             if let Some(b) = self.pairs.get(&(anc, desc)) {
                 *b
             } else {
                 let b = (self.inner)(anc, desc);
                 self.pairs.insert((anc, desc), b);
                 b
             }
         }
     }
 }
 
 struct ActionsIterator<'a> {
     changes: &'a ChangedFiles<'a>,
     parent: Parent,
     current: u32,
 }
 
 impl<'a> Iterator for ActionsIterator<'a> {
     type Item = Action<'a>;
 
     fn next(&mut self) -> Option<Action<'a>> {
         let copy_flag = match self.parent {
             Parent::FirstParent => P1_COPY,
             Parent::SecondParent => P2_COPY,
         };
         while self.current < self.changes.nb_items {
             let (flags, file, source) = self.changes.entry(self.current);
             self.current += 1;
             if (flags & ACTION_MASK) == REMOVED {
                 return Some(Action::Removed(file));
             }
             let copy = flags & COPY_MASK;
             if copy == copy_flag {
                 return Some(Action::Copied(file, source));
             }
         }
         return None;
     }
 }
 
 /// A small struct whose purpose is to ensure lifetime of bytes referenced in
 /// ChangedFiles
 ///
 /// It is passed to the RevInfoMaker callback who can assign any necessary
 /// content to the `data` attribute. The copy tracing code is responsible for
 /// keeping the DataHolder alive at least as long as the ChangedFiles object.
 pub struct DataHolder<D> {
     /// RevInfoMaker callback should assign data referenced by the
     /// ChangedFiles struct it return to this attribute. The DataHolder
     /// lifetime will be at least as long as the ChangedFiles one.
     pub data: Option<D>,
 }
 
 pub type RevInfoMaker<'a, D> =
     Box<dyn for<'r> Fn(Revision, &'r mut DataHolder<D>) -> RevInfo<'r> + 'a>;
 
 /// enum used to carry information about the parent → child currently processed
 #[derive(Copy, Clone, Debug)]
 enum Parent {
     /// The `p1(x) → x` edge
     FirstParent,
     /// The `p2(x) → x` edge
     SecondParent,
 }
 
 /// Same as mercurial.copies._combine_changeset_copies, but in Rust.
 ///
 /// Arguments are:
 ///
 /// revs: all revisions to be considered
 /// children: a {parent ? [childrens]} mapping
 /// target_rev: the final revision we are combining copies to
 /// rev_info(rev): callback to get revision information:
 ///   * first parent
 ///   * second parent
 ///   * ChangedFiles
 /// isancestors(low_rev, high_rev): callback to check if a revision is an
 ///                                 ancestor of another
 pub fn combine_changeset_copies<A: Fn(Revision, Revision) -> bool, D>(
     revs: Vec<Revision>,
-    children: HashMap<Revision, Vec<Revision>>,
+    mut children_count: HashMap<Revision, usize>,
     target_rev: Revision,
     rev_info: RevInfoMaker<D>,
     is_ancestor: &A,
 ) -> PathCopies {
     let mut all_copies = HashMap::new();
     let mut oracle = AncestorOracle::new(is_ancestor);
 
     for rev in revs {
-        // Retrieve data computed in a previous iteration
-        let copies = all_copies.remove(&rev);
-        let copies = match copies {
-            Some(c) => c,
-            None => TimeStampedPathCopies::default(), // root of the walked set
-        };
-
-        let current_children = match children.get(&rev) {
-            Some(c) => c,
-            None => panic!("inconsistent `revs` and `children`"),
-        };
-
-        for child in current_children {
-            // We will chain the copies information accumulated for `rev` with
-            // the individual copies information for each of its children.
-            // Creating a new PathCopies for each `rev` → `children` vertex.
-            let mut d: DataHolder<D> = DataHolder { data: None };
-            let (p1, p2, changes) = rev_info(*child, &mut d);
+        let mut d: DataHolder<D> = DataHolder { data: None };
+        let (p1, p2, changes) = rev_info(rev, &mut d);
 
-            let parent = if rev == p1 {
-                Parent::FirstParent
-            } else {
-                assert_eq!(rev, p2);
-                Parent::SecondParent
-            };
-            let new_copies =
-                add_from_changes(&copies, &changes, parent, *child);
+        // We will chain the copies information accumulated for the parent with
+        // the individual copies information the curent revision.  Creating a
+        // new TimeStampedPath for each `rev` → `children` vertex.
+        let mut copies: Option<TimeStampedPathCopies> = None;
+        if p1 != NULL_REVISION {
+            // Retrieve data computed in a previous iteration
+            let parent_copies = get_and_clean_parent_copies(
+                &mut all_copies,
+                &mut children_count,
+                p1,
+            );
+            if let Some(parent_copies) = parent_copies {
+                // combine it with data for that revision
+                let vertex_copies = add_from_changes(
+                    &parent_copies,
+                    &changes,
+                    Parent::FirstParent,
+                    rev,
+                );
+                // keep that data around for potential later combination
+                copies = Some(vertex_copies);
+            }
+        }
+        if p2 != NULL_REVISION {
+            // Retrieve data computed in a previous iteration
+            let parent_copies = get_and_clean_parent_copies(
+                &mut all_copies,
+                &mut children_count,
+                p2,
+            );
+            if let Some(parent_copies) = parent_copies {
+                // combine it with data for that revision
+                let vertex_copies = add_from_changes(
+                    &parent_copies,
+                    &changes,
+                    Parent::SecondParent,
+                    rev,
+                );
 
-            // Merge has two parents needs to combines their copy information.
-            //
-            // If the vertex from the other parent was already processed, we
-            // will have a value for the child ready to be used. We need to
-            // grab it and combine it with the one we already
-            // computed. If not we can simply store the newly
-            // computed data. The processing happening at
-            // the time of the second parent will take care of combining the
-            // two TimeStampedPathCopies instance.
-            match all_copies.remove(child) {
-                None => {
-                    all_copies.insert(child, new_copies);
-                }
-                Some(other_copies) => {
-                    let (minor, major) = match parent {
-                        Parent::FirstParent => (other_copies, new_copies),
-                        Parent::SecondParent => (new_copies, other_copies),
-                    };
-                    let merged_copies =
-                        merge_copies_dict(minor, major, &changes, &mut oracle);
-                    all_copies.insert(child, merged_copies);
-                }
-            };
+                copies = match copies {
+                    None => Some(vertex_copies),
+                    // Merge has two parents needs to combines their copy
+                    // information.
+                    //
+                    // If we got data from both parents, We need to combine
+                    // them.
+                    Some(copies) => Some(merge_copies_dict(
+                        vertex_copies,
+                        copies,
+                        &changes,
+                        &mut oracle,
+                    )),
+                };
+            }
+        }
+        match copies {
+            Some(copies) => {
+                all_copies.insert(rev, copies);
+            }
+            _ => {}
         }
     }
 
     // Drop internal information (like the timestamp) and return the final
     // mapping.
     let tt_result = all_copies
         .remove(&target_rev)
         .expect("target revision was not processed");
     let mut result = PathCopies::default();
     for (dest, tt_source) in tt_result {
         if let Some(path) = tt_source.path {
             result.insert(dest, path);
         }
     }
     result
 }
 
+/// fetch previous computed information
+///
+/// If no other children are expected to need this information, we drop it from
+/// the cache.
+///
+/// If parent is not part of the set we are expected to walk, return None.
+fn get_and_clean_parent_copies(
+    all_copies: &mut HashMap<Revision, TimeStampedPathCopies>,
+    children_count: &mut HashMap<Revision, usize>,
+    parent_rev: Revision,
+) -> Option<TimeStampedPathCopies> {
+    let count = children_count.get_mut(&parent_rev)?;
+    *count -= 1;
+    if *count == 0 {
+        match all_copies.remove(&parent_rev) {
+            Some(c) => Some(c),
+            None => Some(TimeStampedPathCopies::default()),
+        }
+    } else {
+        match all_copies.get(&parent_rev) {
+            Some(c) => Some(c.clone()),
+            None => Some(TimeStampedPathCopies::default()),
+        }
+    }
+}
+
 /// Combine ChangedFiles with some existing PathCopies information and return
 /// the result
 fn add_from_changes(
     base_copies: &TimeStampedPathCopies,
     changes: &ChangedFiles,
     parent: Parent,
     current_rev: Revision,
 ) -> TimeStampedPathCopies {
     let mut copies = base_copies.clone();
     for action in changes.iter_actions(parent) {
         match action {
             Action::Copied(dest, source) => {
                 let entry;
                 if let Some(v) = base_copies.get(source) {
                     entry = match &v.path {
                         Some(path) => Some((*(path)).to_owned()),
                         None => Some(source.to_owned()),
                     }
                 } else {
                     entry = Some(source.to_owned());
                 }
                 // Each new entry is introduced by the children, we
                 // record this information as we will need it to take
                 // the right decision when merging conflicting copy
                 // information. See merge_copies_dict for details.
                 let ttpc = TimeStampedPathCopy {
                     rev: current_rev,
                     path: entry,
                 };
                 copies.insert(dest.to_owned(), ttpc);
             }
             Action::Removed(f) => {
                 // We must drop copy information for removed file.
                 //
                 // We need to explicitly record them as dropped to
                 // propagate this information when merging two
                 // TimeStampedPathCopies object.
                 if copies.contains_key(f.as_ref()) {
                     let ttpc = TimeStampedPathCopy {
                         rev: current_rev,
                         path: None,
                     };
                     copies.insert(f.to_owned(), ttpc);
                 }
             }
         }
     }
     copies
 }
 
 /// merge two copies-mapping together, minor and major
 ///
 /// In case of conflict, value from "major" will be picked, unless in some
 /// cases. See inline documentation for details.
 fn merge_copies_dict<A: Fn(Revision, Revision) -> bool>(
     mut minor: TimeStampedPathCopies,
     mut major: TimeStampedPathCopies,
     changes: &ChangedFiles,
     oracle: &mut AncestorOracle<A>,
 ) -> TimeStampedPathCopies {
     // This closure exist as temporary help while multiple developper are
     // actively working on this code. Feel free to re-inline it once this
     // code is more settled.
     let mut cmp_value =
         |dest: &PathToken,
          src_minor: &TimeStampedPathCopy,
          src_major: &TimeStampedPathCopy| {
             compare_value(changes, oracle, dest, src_minor, src_major)
         };
     if minor.is_empty() {
         major
     } else if major.is_empty() {
         minor
     } else if minor.len() * 2 < major.len() {
         // Lets says we are merging two TimeStampedPathCopies instance A and B.
         //
         // If A contains N items, the merge result will never contains more
         // than N values differents than the one in A
         //
         // If B contains M items, with M > N, the merge result will always
         // result in a minimum of M - N value differents than the on in
         // A
         //
         // As a result, if N < (M-N), we know that simply iterating over A will
         // yield less difference than iterating over the difference
         // between A and B.
         //
         // This help performance a lot in case were a tiny
         // TimeStampedPathCopies is merged with a much larger one.
         for (dest, src_minor) in minor {
             let src_major = major.get(&dest);
             match src_major {
                 None => major.insert(dest, src_minor),
                 Some(src_major) => {
                     match cmp_value(&dest, &src_minor, src_major) {
                         MergePick::Any | MergePick::Major => None,
                         MergePick::Minor => major.insert(dest, src_minor),
                     }
                 }
             };
         }
         major
     } else if major.len() * 2 < minor.len() {
         // This use the same rational than the previous block.
         // (Check previous block documentation for details.)
         for (dest, src_major) in major {
             let src_minor = minor.get(&dest);
             match src_minor {
                 None => minor.insert(dest, src_major),
                 Some(src_minor) => {
                     match cmp_value(&dest, src_minor, &src_major) {
                         MergePick::Any | MergePick::Minor => None,
                         MergePick::Major => minor.insert(dest, src_major),
                     }
                 }
             };
         }
         minor
     } else {
         let mut override_minor = Vec::new();
         let mut override_major = Vec::new();
 
         let mut to_major = |k: &PathToken, v: &TimeStampedPathCopy| {
             override_major.push((k.clone(), v.clone()))
         };
         let mut to_minor = |k: &PathToken, v: &TimeStampedPathCopy| {
             override_minor.push((k.clone(), v.clone()))
         };
 
         // The diff function leverage detection of the identical subpart if
         // minor and major has some common ancestors. This make it very
         // fast is most case.
         //
         // In case where the two map are vastly different in size, the current
         // approach is still slowish because the iteration will iterate over
         // all the "exclusive" content of the larger on. This situation can be
         // frequent when the subgraph of revision we are processing has a lot
         // of roots. Each roots adding they own fully new map to the mix (and
         // likely a small map, if the path from the root to the "main path" is
         // small.
         //
         // We could do better by detecting such situation and processing them
         // differently.
         for d in minor.diff(&major) {
             match d {
                 DiffItem::Add(k, v) => to_minor(k, v),
                 DiffItem::Remove(k, v) => to_major(k, v),
                 DiffItem::Update { old, new } => {
                     let (dest, src_major) = new;
                     let (_, src_minor) = old;
                     match cmp_value(dest, src_minor, src_major) {
                         MergePick::Major => to_minor(dest, src_major),
                         MergePick::Minor => to_major(dest, src_minor),
                         // If the two entry are identical, no need to do
                         // anything (but diff should not have yield them)
                         MergePick::Any => unreachable!(),
                     }
                 }
             };
         }
 
         let updates;
         let mut result;
         if override_major.is_empty() {
             result = major
         } else if override_minor.is_empty() {
             result = minor
         } else {
             if override_minor.len() < override_major.len() {
                 updates = override_minor;
                 result = minor;
             } else {
                 updates = override_major;
                 result = major;
             }
             for (k, v) in updates {
                 result.insert(k, v);
             }
         }
         result
     }
 }
 
 /// represent the side that should prevail when merging two
 /// TimeStampedPathCopies
 enum MergePick {
     /// The "major" (p1) side prevails
     Major,
     /// The "minor" (p2) side prevails
     Minor,
     /// Any side could be used (because they are the same)
     Any,
 }
 
 /// decide which side prevails in case of conflicting values
 #[allow(clippy::if_same_then_else)]
 fn compare_value<A: Fn(Revision, Revision) -> bool>(
     changes: &ChangedFiles,
     oracle: &mut AncestorOracle<A>,
     dest: &PathToken,
     src_minor: &TimeStampedPathCopy,
     src_major: &TimeStampedPathCopy,
 ) -> MergePick {
     if src_major.path == src_minor.path {
         // we have the same value, but from other source;
         if src_major.rev == src_minor.rev {
             // If the two entry are identical, they are both valid
             MergePick::Any
         } else if oracle.is_ancestor(src_major.rev, src_minor.rev) {
             MergePick::Minor
         } else {
             MergePick::Major
         }
     } else if src_major.rev == src_minor.rev {
         // We cannot get copy information for both p1 and p2 in the
         // same rev. So this is the same value.
         unreachable!(
             "conflict information from p1 and p2 in the same revision"
         );
     } else {
         let action = changes.get_merge_case(&dest);
         if src_major.path.is_none() && action == MergeCase::Salvaged {
             // If the file is "deleted" in the major side but was
             // salvaged by the merge, we keep the minor side alive
             MergePick::Minor
         } else if src_minor.path.is_none() && action == MergeCase::Salvaged {
             // If the file is "deleted" in the minor side but was
             // salvaged by the merge, unconditionnaly preserve the
             // major side.
             MergePick::Major
         } else if action == MergeCase::Merged {
             // If the file was actively merged, copy information
             // from each side might conflict.  The major side will
             // win such conflict.
             MergePick::Major
         } else if oracle.is_ancestor(src_major.rev, src_minor.rev) {
             // If the minor side is strictly newer than the major
             // side, it should be kept.
             MergePick::Minor
         } else if src_major.path.is_some() {
             // without any special case, the "major" value win
             // other the "minor" one.
             MergePick::Major
         } else if oracle.is_ancestor(src_minor.rev, src_major.rev) {
             // the "major" rev is a direct ancestors of "minor",
             // any different value should
             // overwrite
             MergePick::Major
         } else {
             // major version is None (so the file was deleted on
             // that branch) and that branch is independant (neither
             // minor nor major is an ancestors of the other one.)
             // We preserve the new
             // information about the new file.
             MergePick::Minor
         }
     }
 }

rust/hg-cpython/src/copy_tracing.rs

@@ -1,153 +1,148
 use cpython::ObjectProtocol;
 use cpython::PyBool;
 use cpython::PyBytes;
 use cpython::PyDict;
 use cpython::PyList;
 use cpython::PyModule;
 use cpython::PyObject;
 use cpython::PyResult;
 use cpython::PyTuple;
 use cpython::Python;
 
 use hg::copy_tracing::combine_changeset_copies;
 use hg::copy_tracing::ChangedFiles;
 use hg::copy_tracing::DataHolder;
 use hg::copy_tracing::RevInfo;
 use hg::copy_tracing::RevInfoMaker;
 use hg::Revision;
 
 /// Combines copies information contained into revision `revs` to build a copy
 /// map.
 ///
 /// See mercurial/copies.py for details
 pub fn combine_changeset_copies_wrapper(
     py: Python,
     revs: PyList,
-    children: PyDict,
+    children_count: PyDict,
     target_rev: Revision,
     rev_info: PyObject,
     is_ancestor: PyObject,
 ) -> PyResult<PyDict> {
     let revs: PyResult<_> =
         revs.iter(py).map(|r| Ok(r.extract(py)?)).collect();
 
     // Wrap the `is_ancestor` python callback as a Rust closure
     //
     // No errors are expected from the Python side, and they will should only
     // happens in case of programing error or severe data corruption. Such
     // errors will raise panic and the rust-cpython harness will turn them into
     // Python exception.
     let is_ancestor_wrap = |anc: Revision, desc: Revision| -> bool {
         is_ancestor
             .call(py, (anc, desc), None)
             .expect(
                 "rust-copy-tracing: python call  to `is_ancestor` \
                 failed",
             )
             .cast_into::<PyBool>(py)
             .expect(
                 "rust-copy-tracing: python call  to `is_ancestor` \
                 returned unexpected non-Bool value",
             )
             .is_true()
     };
 
     // Wrap the `rev_info_maker` python callback as a Rust closure
     //
     // No errors are expected from the Python side, and they will should only
     // happens in case of programing error or severe data corruption. Such
     // errors will raise panic and the rust-cpython harness will turn them into
     // Python exception.
     let rev_info_maker: RevInfoMaker<PyBytes> =
         Box::new(|rev: Revision, d: &mut DataHolder<PyBytes>| -> RevInfo {
             let res: PyTuple = rev_info
                 .call(py, (rev,), None)
                 .expect("rust-copy-tracing: python call to `rev_info` failed")
                 .cast_into(py)
                 .expect(
                     "rust-copy_tracing: python call to `rev_info` returned \
                     unexpected non-Tuple value",
                 );
             let p1 = res.get_item(py, 0).extract(py).expect(
                 "rust-copy-tracing: rev_info return is invalid, first item \
                 is a not a revision",
             );
             let p2 = res.get_item(py, 1).extract(py).expect(
                 "rust-copy-tracing: rev_info return is invalid, first item \
                 is a not a revision",
             );
 
             let files = match res.get_item(py, 2).extract::<PyBytes>(py) {
                 Ok(raw) => {
                     // Give responsability for the raw bytes lifetime to
                     // hg-core
                     d.data = Some(raw);
                     let addrs = d.data.as_ref().expect(
                         "rust-copy-tracing: failed to get a reference to the \
                         raw bytes for copy data").data(py);
                     ChangedFiles::new(addrs)
                 }
                 // value was presumably None, meaning they was no copy data.
                 Err(_) => ChangedFiles::new_empty(),
             };
 
             (p1, p2, files)
         });
-    let children: PyResult<_> = children
+    let children_count: PyResult<_> = children_count
         .items(py)
         .iter()
-        .map(|(k, v)| {
-            let v: &PyList = v.cast_as(py)?;
-            let v: PyResult<_> =
-                v.iter(py).map(|child| Ok(child.extract(py)?)).collect();
-            Ok((k.extract(py)?, v?))
-        })
+        .map(|(k, v)| Ok((k.extract(py)?, v.extract(py)?)))
         .collect();
 
     let res = combine_changeset_copies(
         revs?,
-        children?,
+        children_count?,
         target_rev,
         rev_info_maker,
         &is_ancestor_wrap,
     );
     let out = PyDict::new(py);
     for (dest, source) in res.into_iter() {
         out.set_item(
             py,
             PyBytes::new(py, &dest.into_vec()),
             PyBytes::new(py, &source.into_vec()),
         )?;
     }
     Ok(out)
 }
 
 /// Create the module, with `__package__` given from parent
 pub fn init_module(py: Python, package: &str) -> PyResult<PyModule> {
     let dotted_name = &format!("{}.copy_tracing", package);
     let m = PyModule::new(py, dotted_name)?;
 
     m.add(py, "__package__", package)?;
     m.add(py, "__doc__", "Copy tracing - Rust implementation")?;
 
     m.add(
         py,
         "combine_changeset_copies",
         py_fn!(
             py,
             combine_changeset_copies_wrapper(
                 revs: PyList,
                 children: PyDict,
                 target_rev: Revision,
                 rev_info: PyObject,
                 is_ancestor: PyObject
             )
         ),
     )?;
 
     let sys = PyModule::import(py, "sys")?;
     let sys_modules: PyDict = sys.get(py, "modules")?.extract(py)?;
     sys_modules.set_item(py, dotted_name, &m)?;
 
     Ok(m)
 }