Fwd: Another issue with Evolve (with test case)

[Angel Ezquerra]

Hi,

I recently found a couple of issues while using evolve. I discussed
them briefly with Pierre-Yves and he suggested that I should send the
info to the mailing list.

In this particular case I was able to make a small test case that
fails every time I run it.

The following commands, executed on the windows cmd.exe or
powershell.exe result in a seemingly endless loop:

mkdir test-evolve
cd test-evolve
hg  init
echo hello > c.txt
hg  add c.txt
hg  commit -m "add c.txt"
echo world > c.txt
hg  commit -m "change c.txt"
hg  rename c.txt d.txt
hg  commit -m "rename c.txt into d.txt"
hg  update 1
hg  forget c.txt
hg  log -G --hidden
hg  amend -m "forget c.txt"
hg  evolve --all

I also recorded in "test case format", including the actual output of
the test case:

  $ mkdir test-evolve
  $ cd test-evolve
  $ hg  init
  $ echo hello > c.txt
  $ hg  add c.txt
  $ hg  commit -m "add c.txt"
  $ echo world > c.txt
  $ hg  commit -m "change c.txt"
  $ hg  rename c.txt d.txt
  $ hg  commit -m "rename c.txt into d.txt"
  $ hg  update 1
  1 files updated, 0 files merged, 1 files removed, 0 files unresolved
  $ hg  forget c.txt
  $ hg  log -G --hidden
  o  changeset:   2:bcbf9f96baac
  |  tag:         tip
  |  user:        Angel Ezquerra <angel.ezquerra at gmail.com>
  |  date:        Wed Mar 20 20:14:30 2013 +0100
  |  summary:     rename c.txt into d.txt
  |
  @  changeset:   1:1e0016a962d9
  |  user:        Angel Ezquerra <angel.ezquerra at gmail.com>
  |  date:        Wed Mar 20 18:45:37 2013 +0100
  |  summary:     change c.txt
  |
  o  changeset:   0:56be9ac2d9d4
     user:        Angel Ezquerra <angel.ezquerra at gmail.com>
     date:        Wed Mar 20 18:45:32 2013 +0100
     summary:     add c.txt
  $ hg  amend -m "forget c.txt"
  1 new unstable changesets
  $ hg  evolve --all
  move:[2] rename c.txt into d.txt
  atop:[4] forget c.txt
  note: possible conflict - c.txt was deleted and renamed to:
   d.txt
  4 [<changectx 460ea8f795f4>, <changectx bcbf9f96baac>]
  5
  move:[5] rename c.txt into d.txt
  atop:[5] rename c.txt into d.txt
  move:[6] rename c.txt into d.txt
  atop:[6] rename c.txt into d.txt
  move:[7] rename c.txt into d.txt
  atop:[7] rename c.txt into d.txt
  move:[8] rename c.txt into d.txt
  atop:[8] rename c.txt into d.txt
  move:[9] rename c.txt into d.txt
  atop:[9] rename c.txt into d.txt
  move:[10] rename c.txt into d.txt
  atop:[10] rename c.txt into d.txt
  move:[11] rename c.txt into d.txt
  atop:[11] rename c.txt into d.txt
  move:[12] rename c.txt into d.txt
  atop:[12] rename c.txt into d.txt
  move:[13] rename c.txt into d.txt
  atop:[13] rename c.txt into d.txt
  move:[14] rename c.txt into d.txt
  atop:[14] rename c.txt into d.txt
  move:[15] rename c.txt into d.txt
  atop:[15] rename c.txt into d.txt
  move:[16] rename c.txt into d.txt
  atop:[16] rename c.txt into d.txt
  move:[17] rename c.txt into d.txt
  atop:[17] rename c.txt into d.txt
  interrupted!
  Exception WindowsError: WindowsError(2, 'The system cannot find the
file specified') in <bound method transact
ion.__del__ of <mercurial.transaction.transaction object at
0x0000000002FF9710>> ignored


The test stops when I hit Ctrl-C. Note that the exception does not
always happen (I think it depends on when I interrupt the command).

BTW, the purpose of this test was to see what happens when you try to
evolve a revision in which a file was renamed when the user amends one
of its ancestors by removing the file that the obsoleted revision
renamed. The idea was to test what happens when you merge a deleted
and renamed file in this case.

Matt and I have have been discussing on the list the behavior of merge
when a file is deleted on one side and renamed on the other on the
list. I suspect that in the case of evolve the current behavior of
merge will yield surprising results (i.e. a deleted file may come
back...).

Cheers,

Angel