view tests/test-copies.t @ 51681:522b4d729e89

mmap: populate the mapping by default

Without pre-population, accessing all data through a mmap can result in many
pagefault, reducing performance significantly. If the mmap is prepopulated, the
performance can no longer get slower than a full read.
(See benchmark number below)

In some cases were very few data is read, prepopulating can be overkill and
slower than populating on access (through page fault). So that behavior can be
controlled when the caller can pre-determine the best behavior.
(See benchmark number below)

In addition, testing with populating in a secondary thread yield great result
combining the best of each approach. This might be implemented in later
changesets.

In all cases, using mmap has a great effect on memory usage when many processes
run in parallel on the same machine.

### Benchmarks

# What did I run

A couple of month back I ran a large benchmark campaign to assess the impact of
various approach for using mmap with the revlog (and other files), it
highlighted a few benchmarks that capture the impact of the changes well. So to
validate this change I checked the following:

- log command displaying various revisions
  (read the changelog index)
- log command displaying the patch of listed revisions
  (read the changelog index, the manifest index and a few files indexes)
- unbundling a few revisions
  (read and write changelog, manifest and few files indexes, and walk the graph
  to update some cache)
- pushing a few revisions
  (read and write changelog, manifest and few files indexes, walk the graph to
  update some cache, performs various accesses locally and remotely during
  discovery)

Benchmarks were run using the default module policy (c+py) and the rust one. No
significant difference were found between the two implementation, so we will
present result using the default policy (unless otherwise specified).

I ran them on a few repositories :

- mercurial: a "public changeset only" copy of mercurial from 2018-08-01 using
  zstd compression and sparse-revlog

- pypy: a copy of pypy from 2018-08-01 using zstd compression and sparse-revlog

- netbeans: a copy of netbeans from 2018-08-01 using zstd compression and
  sparse-revlog

- mozilla-try: a copy of mozilla-try from 2019-02-18 using zstd compression and
  sparse-revlog

- mozilla-try persistent-nodemap: Same as the above but with a persistent
  nodemap. Used for the log --patch benchmark only

# Results

For the smaller repositories (mercurial, pypy), the impact of mmap is almost
imperceptible, other cost dominating the operation. The impact of prepopulating
is undiscernible in the benchmark we ran.

For larger repositories the benchmark support explanation given above:

On netbeans, the log can be about 1% faster without repopulation (for a
difference < 100ms) but unbundle becomes a bit slower, even when small.

### data-env-vars.name                             = netbeans-2018-08-01-zstd-sparse-revlog
  # benchmark.name                                 = hg.command.unbundle
  # benchmark.variants.issue6528                   = disabled
  # benchmark.variants.reuse-external-delta-parent = yes
  # benchmark.variants.revs                        = any-1-extra-rev
  # benchmark.variants.source                      = unbundle
  # benchmark.variants.verbosity                   = quiet
with-populate: 0.240157
no-populate:   0.265087  (+10.38%, +0.02)
  # benchmark.variants.revs                        = any-100-extra-rev
with-populate: 1.459518
no-populate:   1.481290   (+1.49%, +0.02)
 ## benchmark.name                                 = hg.command.push
  # benchmark.variants.explicit-rev                = none
  # benchmark.variants.issue6528                   = disabled
  # benchmark.variants.protocol                    = ssh
  # benchmark.variants.reuse-external-delta-parent = yes
  # benchmark.variants.revs                        = any-1-extra-rev
with-populate: 0.771919
no-populate:   0.792025   (+2.60%, +0.02)
  # benchmark.variants.revs                        = any-100-extra-rev
with-populate: 1.459518
no-populate:   1.481290   (+1.49%, +0.02)


For mozilla-try, the "slow down" from pre-populate for small `hg log` is more
visible, but still small in absolute time. (using rust value for the persistent
nodemap value to be relevant).

### data-env-vars.name            = mozilla-try-2019-02-18-ds2-pnm
  # benchmark.name                = hg.command.log
  # bin-env-vars.hg.flavor        = rust
  # benchmark.variants.patch      = yes
  # benchmark.variants.limit-rev  = 1
with-populate: 0.237813
no-populate:   0.229452   (-3.52%, -0.01)
  # benchmark.variants.limit-rev  = 10
  # benchmark.variants.patch      = yes
with-populate: 1.213578
no-populate:   1.205189
### data-env-vars.name            = mozilla-try-2019-02-18-zstd-sparse-revlog
  # benchmark.variants.limit-rev  = 1000
  # benchmark.variants.patch      = no
  # benchmark.variants.rev        = tip
with-populate: 0.198607
no-populate:   0.195038   (-1.80%, -0.00)

However pre-populating provide a significant boost on more complex operations
like unbundle or push:

### data-env-vars.name                             = mozilla-try-2019-02-18-zstd-sparse-revlog
  # benchmark.name                                 = hg.command.push
  # benchmark.variants.explicit-rev                = none
  # benchmark.variants.issue6528                   = disabled
  # benchmark.variants.protocol                    = ssh
  # benchmark.variants.reuse-external-delta-parent = yes
  # benchmark.variants.revs                        = any-1-extra-rev
with-populate: 4.798632
no-populate:   4.953295   (+3.22%, +0.15)
  # benchmark.variants.revs                        = any-100-extra-rev
with-populate: 4.903618
no-populate:   5.014963   (+2.27%, +0.11)
 ## benchmark.name                                 = hg.command.unbundle
  # benchmark.variants.revs                        = any-1-extra-rev
with-populate: 1.423411
no-populate:   1.585365  (+11.38%, +0.16)
  # benchmark.variants.revs                        = any-100-extra-rev
with-populate: 1.537909
no-populate:   1.688489   (+9.79%, +0.15)

#testcases filelog compatibility changeset sidedata

  $ cat >> $HGRCPATH << EOF
  > [extensions]
  > rebase=
  > [alias]
  > l = log -G -T '{rev} {desc}\n{files}\n'
  > EOF

#if compatibility
  $ cat >> $HGRCPATH << EOF
  > [experimental]
  > copies.read-from = compatibility
  > EOF
#endif

#if changeset
  $ cat >> $HGRCPATH << EOF
  > [experimental]
  > copies.read-from = changeset-only
  > copies.write-to = changeset-only
  > EOF
#endif

#if sidedata
  $ cat >> $HGRCPATH << EOF
  > [format]
  > exp-use-copies-side-data-changeset = yes
  > EOF
#endif

  $ REPONUM=0
  $ newrepo() {
  >     cd $TESTTMP
  >     REPONUM=`expr $REPONUM + 1`
  >     hg init repo-$REPONUM
  >     cd repo-$REPONUM
  > }

Simple rename case
  $ newrepo
  $ echo x > x
  $ hg ci -Aqm 'add x'
  $ hg mv x y
  $ hg debugp1copies
  x -> y
  $ hg debugp2copies
  $ hg ci -m 'rename x to y'
  $ hg l
  @  1 rename x to y
  |  x y
  o  0 add x
     x
  $ hg debugp1copies -r 1
  x -> y
  $ hg debugpathcopies 0 1
  x -> y
  $ hg debugpathcopies 1 0
  y -> x
Test filtering copies by path. We do filtering by destination.
  $ hg debugpathcopies 0 1 x
  $ hg debugpathcopies 1 0 x
  y -> x
  $ hg debugpathcopies 0 1 y
  x -> y
  $ hg debugpathcopies 1 0 y

Copies not including commit changes
  $ newrepo
  $ echo x > x
  $ hg ci -Aqm 'add x'
  $ hg mv x y
  $ hg debugpathcopies . .
  $ hg debugpathcopies . 'wdir()'
  x -> y
  $ hg debugpathcopies 'wdir()' .
  y -> x

Copy a file onto another file
  $ newrepo
  $ echo x > x
  $ echo y > y
  $ hg ci -Aqm 'add x and y'
  $ hg cp -f x y
  $ hg debugp1copies
  x -> y
  $ hg debugp2copies
  $ hg ci -m 'copy x onto y'
  $ hg l
  @  1 copy x onto y
  |  y
  o  0 add x and y
     x y
  $ hg debugp1copies -r 1
  x -> y
  $ hg debugpathcopies 0 1
  x -> y (no-filelog !)
  $ hg debugpathcopies 0 1  --config devel.copy-tracing.trace-all-files=yes
  x -> y

Copy a file onto another file with same content. If metadata is stored in changeset, this does not
produce a new filelog entry. The changeset's "files" entry should still list the file.
  $ newrepo
  $ echo x > x
  $ echo x > x2
  $ hg ci -Aqm 'add x and x2 with same content'
  $ hg cp -f x x2
  $ hg ci -m 'copy x onto x2'
  $ hg l
  @  1 copy x onto x2
  |  x2
  o  0 add x and x2 with same content
     x x2
  $ hg debugp1copies -r 1
  x -> x2
  $ hg debugpathcopies 0 1
  x -> x2 (no-filelog !)
  $ hg debugpathcopies 0 1  --config devel.copy-tracing.trace-all-files=yes
  x -> x2

Rename file in a loop: x->y->z->x
  $ newrepo
  $ echo x > x
  $ hg ci -Aqm 'add x'
  $ hg mv x y
  $ hg debugp1copies
  x -> y
  $ hg debugp2copies
  $ hg ci -m 'rename x to y'
  $ hg mv y z
  $ hg ci -m 'rename y to z'
  $ hg mv z x
  $ hg ci -m 'rename z to x'
  $ hg l
  @  3 rename z to x
  |  x z
  o  2 rename y to z
  |  y z
  o  1 rename x to y
  |  x y
  o  0 add x
     x
  $ hg debugpathcopies 0 3

Copy x to z, then remove z, then copy x2 (same content as x) to z. With copy metadata in the
changeset, the two copies here will have the same filelog entry, so ctx['z'].introrev() might point
to the first commit that added the file. We should still report the copy as being from x2.
  $ newrepo
  $ echo x > x
  $ echo x > x2
  $ hg ci -Aqm 'add x and x2 with same content'
  $ hg cp x z
  $ hg ci -qm 'copy x to z'
  $ hg rm z
  $ hg ci -m 'remove z'
  $ hg cp x2 z
  $ hg ci -m 'copy x2 to z'
  $ hg l
  @  3 copy x2 to z
  |  z
  o  2 remove z
  |  z
  o  1 copy x to z
  |  z
  o  0 add x and x2 with same content
     x x2
  $ hg debugp1copies -r 3
  x2 -> z
  $ hg debugpathcopies 0 3
  x2 -> z

Create x and y, then rename them both to the same name, but on different sides of a fork
  $ newrepo
  $ echo x > x
  $ echo y > y
  $ hg ci -Aqm 'add x and y'
  $ hg mv x z
  $ hg ci -qm 'rename x to z'
  $ hg co -q 0
  $ hg mv y z
  $ hg ci -qm 'rename y to z'
  $ hg l
  @  2 rename y to z
  |  y z
  | o  1 rename x to z
  |/   x z
  o  0 add x and y
     x y
  $ hg debugpathcopies 1 2
  z -> x
  y -> z

Fork renames x to y on one side and removes x on the other
  $ newrepo
  $ echo x > x
  $ hg ci -Aqm 'add x'
  $ hg mv x y
  $ hg ci -m 'rename x to y'
  $ hg co -q 0
  $ hg rm x
  $ hg ci -m 'remove x'
  created new head
  $ hg l
  @  2 remove x
  |  x
  | o  1 rename x to y
  |/   x y
  o  0 add x
     x
  $ hg debugpathcopies 1 2

Merge rename from other branch
  $ newrepo
  $ echo x > x
  $ hg ci -Aqm 'add x'
  $ hg mv x y
  $ hg ci -m 'rename x to y'
  $ hg co -q 0
  $ echo z > z
  $ hg ci -Aqm 'add z'
  $ hg merge -q 1
  $ hg debugp1copies
  $ hg debugp2copies
  $ hg ci -m 'merge rename from p2'
  $ hg l
  @    3 merge rename from p2
  |\
  | o  2 add z
  | |  z
  o |  1 rename x to y
  |/   x y
  o  0 add x
     x
Perhaps we should indicate the rename here, but `hg status` is documented to be weird during
merges, so...
  $ hg debugp1copies -r 3
  $ hg debugp2copies -r 3
  $ hg debugpathcopies 0 3
  x -> y
  $ hg debugpathcopies 1 2
  y -> x
  $ hg debugpathcopies 1 3
  $ hg debugpathcopies 2 3
  x -> y

Copy file from either side in a merge
  $ newrepo
  $ echo x > x
  $ hg ci -Aqm 'add x'
  $ hg co -q null
  $ echo y > y
  $ hg ci -Aqm 'add y'
  $ hg merge -q 0
  $ hg cp y z
  $ hg debugp1copies
  y -> z
  $ hg debugp2copies
  $ hg ci -m 'copy file from p1 in merge'
  $ hg co -q 1
  $ hg merge -q 0
  $ hg cp x z
  $ hg debugp1copies
  $ hg debugp2copies
  x -> z
  $ hg ci -qm 'copy file from p2 in merge'
  $ hg l
  @    3 copy file from p2 in merge
  |\   z
  +---o  2 copy file from p1 in merge
  | |/   z
  | o  1 add y
  |    y
  o  0 add x
     x
  $ hg debugp1copies -r 2
  y -> z
  $ hg debugp2copies -r 2
  $ hg debugpathcopies 1 2
  y -> z
  $ hg debugpathcopies 0 2
  $ hg debugp1copies -r 3
  $ hg debugp2copies -r 3
  x -> z
  $ hg debugpathcopies 1 3
  $ hg debugpathcopies 0 3
  x -> z

Copy file that exists on both sides of the merge, same content on both sides
  $ newrepo
  $ echo x > x
  $ hg ci -Aqm 'add x on branch 1'
  $ hg co -q null
  $ echo x > x
  $ hg ci -Aqm 'add x on branch 2'
  $ hg merge -q 0
  $ hg cp x z
  $ hg debugp1copies
  x -> z
  $ hg debugp2copies
  $ hg ci -qm 'merge'
  $ hg l
  @    2 merge
  |\   z
  | o  1 add x on branch 2
  |    x
  o  0 add x on branch 1
     x
  $ hg debugp1copies -r 2
  x -> z
  $ hg debugp2copies -r 2
It's a little weird that it shows up on both sides
  $ hg debugpathcopies 1 2
  x -> z
  $ hg debugpathcopies 0 2
  x -> z (filelog !)

Copy file that exists on both sides of the merge, different content
  $ newrepo
  $ echo branch1 > x
  $ hg ci -Aqm 'add x on branch 1'
  $ hg co -q null
  $ echo branch2 > x
  $ hg ci -Aqm 'add x on branch 2'
  $ hg merge -q 0
  warning: conflicts while merging x! (edit, then use 'hg resolve --mark')
  [1]
  $ echo resolved > x
  $ hg resolve -m x
  (no more unresolved files)
  $ hg cp x z
  $ hg debugp1copies
  x -> z
  $ hg debugp2copies
  $ hg ci -qm 'merge'
  $ hg l
  @    2 merge
  |\   x z
  | o  1 add x on branch 2
  |    x
  o  0 add x on branch 1
     x
  $ hg debugp1copies -r 2
  x -> z (changeset !)
  x -> z (sidedata !)
  $ hg debugp2copies -r 2
  x -> z (no-changeset no-sidedata !)
  $ hg debugpathcopies 1 2
  x -> z (changeset !)
  x -> z (sidedata !)
  $ hg debugpathcopies 0 2
  x -> z (no-changeset no-sidedata !)

Copy x->y on one side of merge and copy x->z on the other side. Pathcopies from one parent
of the merge to the merge should include the copy from the other side.
  $ newrepo
  $ echo x > x
  $ hg ci -Aqm 'add x'
  $ hg cp x y
  $ hg ci -qm 'copy x to y'
  $ hg co -q 0
  $ hg cp x z
  $ hg ci -qm 'copy x to z'
  $ hg merge -q 1
  $ hg ci -m 'merge copy x->y and copy x->z'
  $ hg l
  @    3 merge copy x->y and copy x->z
  |\
  | o  2 copy x to z
  | |  z
  o |  1 copy x to y
  |/   y
  o  0 add x
     x
  $ hg debugp1copies -r 3
  $ hg debugp2copies -r 3
  $ hg debugpathcopies 2 3
  x -> y
  $ hg debugpathcopies 1 3
  x -> z

Copy x->y on two separate branches. Pathcopies from one branch to the other
should not report the copy.
  $ newrepo
  $ echo x > x
  $ hg ci -Aqm 'add x'
  $ hg cp x y
  $ hg ci -qm 'copy x to y'
  $ hg co -q 0
  $ hg graft 1 -q
  $ hg l
  @  2 copy x to y
  |  y
  | o  1 copy x to y
  |/   y
  o  0 add x
     x
  $ hg debugp1copies -r 1
  x -> y
  $ hg debugp1copies -r 2
  x -> y
  $ hg debugpathcopies 1 2
  $ hg debugpathcopies 2 1

Copy x to y on one side of merge, create y and rename to z on the other side.
  $ newrepo
  $ echo x > x
  $ hg ci -Aqm 'add x'
  $ hg cp x y
  $ hg ci -qm 'copy x to y'
  $ hg co -q 0
  $ echo y > y
  $ hg ci -Aqm 'add y'
  $ hg mv y z
  $ hg ci -m 'rename y to z'
  $ hg merge -q 1
  $ hg ci -m 'merge'
  $ hg l
  @    4 merge
  |\
  | o  3 rename y to z
  | |  y z
  | o  2 add y
  | |  y
  o |  1 copy x to y
  |/   y
  o  0 add x
     x
  $ hg debugp1copies -r 3
  y -> z
  $ hg debugp2copies -r 3
  $ hg debugpathcopies 2 3
  y -> z
  $ hg debugpathcopies 1 3
  y -> z (no-filelog !)

Create x and y, then rename x to z on one side of merge, and rename y to z and
modify z on the other side. When storing copies in the changeset, we don't
filter out copies whose target was created on the other side of the merge.
  $ newrepo
  $ echo x > x
  $ echo y > y
  $ hg ci -Aqm 'add x and y'
  $ hg mv x z
  $ hg ci -qm 'rename x to z'
  $ hg co -q 0
  $ hg mv y z
  $ hg ci -qm 'rename y to z'
  $ echo z >> z
  $ hg ci -m 'modify z'
  $ hg merge -q 1
  warning: conflicts while merging z! (edit, then use 'hg resolve --mark')
  [1]
  $ echo z > z
  $ hg resolve -qm z
  $ hg ci -m 'merge 1 into 3'
Try merging the other direction too
  $ hg co -q 1
  $ hg merge -q 3
  warning: conflicts while merging z! (edit, then use 'hg resolve --mark')
  [1]
  $ echo z > z
  $ hg resolve -qm z
  $ hg ci -m 'merge 3 into 1'
  created new head
  $ hg l
  @    5 merge 3 into 1
  |\   z
  +---o  4 merge 1 into 3
  | |/   z
  | o  3 modify z
  | |  z
  | o  2 rename y to z
  | |  y z
  o |  1 rename x to z
  |/   x z
  o  0 add x and y
     x y
  $ hg debugpathcopies 1 4
  y -> z (no-filelog !)
  $ hg debugpathcopies 2 4
  x -> z (no-filelog !)
  $ hg debugpathcopies 0 4
  x -> z (filelog !)
  y -> z (no-filelog !)
  $ hg debugpathcopies 1 5
  y -> z (no-filelog !)
  $ hg debugpathcopies 2 5
  x -> z (no-filelog !)
  $ hg debugpathcopies 0 5
  x -> z

Create x and y, then remove y and rename x to y on one side of merge, and
modify x on the other side. The modification to x from the second side
should be propagated to y.
  $ newrepo
  $ echo original > x
  $ hg add x
  $ echo unrelated > y
  $ hg add y
  $ hg commit -m 'add x and y'
  $ hg remove y
  $ hg commit -m 'remove y'
  $ hg rename x y
  $ hg commit -m 'rename x to y'
  $ hg checkout -q 0
  $ echo modified > x
  $ hg commit -m 'modify x'
  created new head
  $ hg l
  @  3 modify x
  |  x
  | o  2 rename x to y
  | |  x y
  | o  1 remove y
  |/   y
  o  0 add x and y
     x y
#if filelog
  $ hg merge 2
  file 'x' was deleted in other [merge rev] but was modified in local [working copy].
  You can use (c)hanged version, (d)elete, or leave (u)nresolved.
  What do you want to do? u
  1 files updated, 0 files merged, 0 files removed, 1 files unresolved
  use 'hg resolve' to retry unresolved file merges or 'hg merge --abort' to abandon
  [1]
This should ideally be "modified", but we will probably not be able to fix
that in the filelog case.
  $ cat y
  original
#else
  $ hg merge 2
  merging x and y to y
  0 files updated, 1 files merged, 0 files removed, 0 files unresolved
  (branch merge, don't forget to commit)
  $ cat y
  modified
#endif
Same as above, but in the opposite direction
#if filelog
  $ hg co -qC 2
  $ hg merge 3
  file 'x' was deleted in local [working copy] but was modified in other [merge rev].
  You can use (c)hanged version, leave (d)eleted, or leave (u)nresolved.
  What do you want to do? u
  0 files updated, 0 files merged, 0 files removed, 1 files unresolved
  use 'hg resolve' to retry unresolved file merges or 'hg merge --abort' to abandon
  [1]
BROKEN: should be "modified"
  $ cat y
  original
#else
  $ hg co -qC 2
  $ hg merge 3
  merging y and x to y
  0 files updated, 1 files merged, 0 files removed, 0 files unresolved
  (branch merge, don't forget to commit)
  $ cat y
  modified
#endif

Create x and y, then rename x to z on one side of merge, and rename y to z and
then delete z on the other side.
  $ newrepo
  $ echo x > x
  $ echo y > y
  $ hg ci -Aqm 'add x and y'
  $ hg mv x z
  $ hg ci -qm 'rename x to z'
  $ hg co -q 0
  $ hg mv y z
  $ hg ci -qm 'rename y to z'
  $ hg rm z
  $ hg ci -m 'delete z'
  $ hg merge -q 1
  $ echo z > z
  $ hg ci -m 'merge 1 into 3'
Try merging the other direction too
  $ hg co -q 1
  $ hg merge -q 3
  $ echo z > z
  $ hg ci -m 'merge 3 into 1'
  created new head
  $ hg l
  @    5 merge 3 into 1
  |\   z
  +---o  4 merge 1 into 3
  | |/   z
  | o  3 delete z
  | |  z
  | o  2 rename y to z
  | |  y z
  o |  1 rename x to z
  |/   x z
  o  0 add x and y
     x y
  $ hg debugpathcopies 1 4
  $ hg debugpathcopies 2 4
  x -> z (no-filelog !)
  $ hg debugpathcopies 0 4
  x -> z (no-changeset no-compatibility !)
  $ hg debugpathcopies 1 5
  $ hg debugpathcopies 2 5
  x -> z (no-filelog !)
  $ hg debugpathcopies 0 5
  x -> z


Test for a case in fullcopytracing algorithm where neither of the merging csets
is a descendant of the merge base. This test reflects that the algorithm
correctly finds the copies:

  $ cat >> $HGRCPATH << EOF
  > [experimental]
  > evolution.createmarkers=True
  > evolution.allowunstable=True
  > EOF

  $ newrepo
  $ echo a > a
  $ hg add a
  $ hg ci -m "added a"
  $ echo b > b
  $ hg add b
  $ hg ci -m "added b"

  $ hg mv b b1
  $ hg ci -m "rename b to b1"

  $ hg up ".^"
  1 files updated, 0 files merged, 1 files removed, 0 files unresolved
  $ echo d > d
  $ hg add d
  $ hg ci -m "added d"
  created new head

  $ echo baba >> b
  $ hg ci --amend -m "added d, modified b"

  $ hg l --hidden
  @  4 added d, modified b
  |  b d
  | x  3 added d
  |/   d
  | o  2 rename b to b1
  |/   b b1
  o  1 added b
  |  b
  o  0 added a
     a

Grafting revision 4 on top of revision 2, showing that it respect the rename:

  $ hg up 2 -q
  $ hg graft -r 4 --base 3 --hidden
  grafting 4:af28412ec03c "added d, modified b" (tip) (no-changeset !)
  grafting 4:6325ca0b7a1c "added d, modified b" (tip) (changeset !)
  merging b1 and b to b1

  $ hg l -l1 -p
  @  5 added d, modified b
  |  b1
  ~  diff -r 5a4825cc2926 -r 94a2f1a0e8e2 b1 (no-changeset !)
  ~  diff -r 0a0ed3b3251c -r d544fb655520 b1 (changeset !)
     --- a/b1	Thu Jan 01 00:00:00 1970 +0000
     +++ b/b1	Thu Jan 01 00:00:00 1970 +0000
     @@ -1,1 +1,2 @@
      b
     +baba
  
Test to make sure that fullcopytracing algorithm doesn't fail when neither of the
merging csets is a descendant of the base.
-------------------------------------------------------------------------------------------------

  $ newrepo
  $ echo a > a
  $ hg add a
  $ hg ci -m "added a"
  $ echo b > b
  $ hg add b
  $ hg ci -m "added b"

  $ echo foobar > willconflict
  $ hg add willconflict
  $ hg ci -m "added willconflict"
  $ echo c > c
  $ hg add c
  $ hg ci -m "added c"

  $ hg l
  @  3 added c
  |  c
  o  2 added willconflict
  |  willconflict
  o  1 added b
  |  b
  o  0 added a
     a

  $ hg up ".^^"
  0 files updated, 0 files merged, 2 files removed, 0 files unresolved
  $ echo d > d
  $ hg add d
  $ hg ci -m "added d"
  created new head

  $ echo barfoo > willconflict
  $ hg add willconflict
  $ hg ci --amend -m "added willconflict and d"

  $ hg l
  @  5 added willconflict and d
  |  d willconflict
  | o  3 added c
  | |  c
  | o  2 added willconflict
  |/   willconflict
  o  1 added b
  |  b
  o  0 added a
     a

  $ hg rebase -r . -d 2 -t :other
  rebasing 5:5018b1509e94 tip "added willconflict and d" (no-changeset !)
  rebasing 5:af8d273bf580 tip "added willconflict and d" (changeset !)

  $ hg up 3 -q
  $ hg l --hidden
  o  6 added willconflict and d
  |  d willconflict
  | x  5 added willconflict and d
  | |  d willconflict
  | | x  4 added d
  | |/   d
  +---@  3 added c
  | |    c
  o |  2 added willconflict
  |/   willconflict
  o  1 added b
  |  b
  o  0 added a
     a

Now if we trigger a merge between revision 3 and 6 using base revision 4,
neither of the merging csets will be a descendant of the base revision:

  $ hg graft -r 6 --base 4 --hidden -t :other
  grafting 6:99802e4f1e46 "added willconflict and d" (tip) (no-changeset !)
  grafting 6:b19f0df72728 "added willconflict and d" (tip) (changeset !)