Crash when renaming a directory

The crash is a potential data loss case - the renamed directory had been forgotten about.

$ bzr st
modified:
  lib/canonical/launchpad/scripts/runlaunchpad.py
unknown:
  lib/lp/services/memcache/

This isn't actually 'renaming it back', it's another ordering issue. Naming to a lexically earlier name is what fails.

Example reproduction script (tested on bzr.dev and 1.18):

#!/bin/sh -x
bzr="/usr/local/bin/bzr --no-plugins --no-aliases"

${bzr} init A
(
 # Setup
 cd A ;
 ${bzr} mkdir a ;
 ${bzr} ci -m '1' ;

 # Add it
 ${bzr} mkdir a/b ;

 # Both work
 ${bzr} mv a/b a/c ;
 ${bzr} mv a/c a/d ;

 # This fails
 ${bzr} mv a/d a/a ;

 # Show that we lost it.
 ${bzr} stat
)

Note that this can also be made to fail by "mv a/d c" in that penultimate step.

Yes we can !

class TestFullerMd(script.TestCaseWithTransportAndScript):

    def test_bug_430672(self):
        self.run_script("""
$ bzr init A
# Setup
$ cd A
$ bzr mkdir a
$ bzr ci -m '1'

# Add it
$ bzr mkdir a/b

# Both work
$ bzr mv a/b a/c
$ bzr mv a/c a/d

# This fails
$ bzr mv a/d a/a

# Show that we lost it.
$ bzr stat
""")

Note that we don't support the ';' and that empty lines are ignored, which may lead to
subtle bugs if we want to match empty output lines...so we have to treat empty lines
as expected output and use either '$ ' or '#' lines to replace them.

Also, using that kind of test brings the added benefit that they can be re-run (unlike the shell version
which miss a leading 'rm -fr A') and the 'A' directory isn't needed anymore.

And don't get me wrong, that's a very similar shell script that started me to write such shell scripts to reproduce
bugs (in particular the ability to reuse them with various versions of bzr) and gave me the clearer idea
that the above could be implemented (so take credit :).

So it isn't entirely 'we lost it', the rename of the directory on disk succeeded but the rename in the dirstate failed.
If you 'mv a/a a/d' then it will show back up.
$ mv a/a/ a/d
$ bzr st
added:
  a/d/

If we didn't have implicit removals for 'missing' entries then 'bzr commit' would probably block on this.

Anyway, I agree it is a serious thing which should be fixed. It certainly is present in both bzr.dev and the 2.0 series.

Note that it doesn't matter if there are other files /dirs in the directory, but it only triggers on a newly added directory. If you do:

bzr mkdir a/a
bzr mv a/a/ a/d
bzr commit -m 2
bzr mv a/d a/a

it works just fine.

So it is a bug in how we handle entries that are 'added' and renaming them to an earlier path.

I will try your way on workstartion Monday.
It's not big bug for me but if my report can help to make bazaar
better - why not.

On Fri, Sep 25, 2009 at 6:07 PM, John A Meinel <email address hidden> wrote:
> Note that it doesn't matter if there are other files /dirs in the
> directory, but it only triggers on a newly added directory. If you do:
>
> bzr mkdir a/a
> bzr mv a/a/ a/d
> bzr commit -m 2
> bzr mv a/d a/a
>
> it works just fine.
>
> So it is a bug in how we handle entries that are 'added' and renaming
> them to an earlier path.
>
> --
> Crash when renaming a directory
> https://bugs.launchpad.net/bugs/430672
> You received this bug notification because you are a direct subscriber
> of a duplicate bug.
>
> Status in Bazaar Version Control System: Confirmed
> Status in Bazaar 2.0 series: Confirmed
>
> Bug description:
> I renamed a directory and attempted to rename it back, triggering a crash.
>
> $ bzr st
> added:
>  lib/lp/services/memcached/
>  lib/lp/services/memcached/__init__.py
>  lib/lp/services/memcached/interfaces.py
> modified:
>  lib/canonical/launchpad/scripts/runlaunchpad.py
>
> $ bzr mv lib/lp/services/memcached lib/lp/services/memcache
> bzr: ERROR: exceptions.IndexError: list index out of range
>
> *** Bazaar has encountered an internal error.  This probably indicates a
>    bug in Bazaar.  You can help us fix it by filing a bug report at
>        https://bugs.launchpad.net/bzr/+filebug
>    attaching the crash file
>        /home/stub/.cache/crash/bzr-20090916114005-29872.crash
>    and including a description of the problem.
>
>    The crash file is plain text and you can inspect or edit it to remove
>    private information.
>
> $ bzr --version
> Bazaar (bzr) 2.0rc2
>  Python interpreter: /usr/bin/python 2.6.2
>  Python standard library: /usr/lib/python2.6
>  Platform: Linux-2.6.28-15-generic-x86_64-with-Ubuntu-9.04-jaunty
>  bzrlib: /usr/lib/python2.6/dist-packages/bzrlib
>  Bazaar configuration: /home/stub/.bazaar
>  Bazaar log file: /home/stub/.bzr.log
>
> Copyright 2005, 2006, 2007, 2008, 2009 Canonical Ltd.
> http://bazaar-vcs.org/
>
> bzr comes with ABSOLUTELY NO WARRANTY.  bzr is free software, and
> you may use, modify and redistribute it under the terms of the GNU
> General Public License version 2 or later.
>

I get
Created a standalone tree (format: 2a)
added a
Committing to: /tmp/foo/A/
added a
Committed revision 1.
added a/b
a/b => a/c
a/c => a/d
a/d => a/a
added:
  a/a/

with bug bug 403322 branch. Is that what you're expecting fullermd ?