a breaking instance of bzr push

[Michael Hudson-Doyle]

This seems much simpler than the other similar bugs. I checked out pydoctor using

bzr get http://codespeak.net/svn/user/mwh/pydoctor/trunk pydoctor-bzr-2

I made a trivial change (deleted an item in doc/todo.txt), did 'bzr ci' and then 'bzr push http://codespeak.net/svn/user/mwh/pydoctor/trunk'.

This was the result:

$ bzr push http://codespeak.net/svn/user/mwh/pydoctor/trunk
This transport does not update the working tree of: http://codespeak.net/svn/user/mwh/pydoctor/trunk
bzr: ERROR: exceptions.KeyError: u'bin/makedocsys'

Traceback (most recent call last):
  File "/Library/Frameworks/Python.framework/Versions/2.4/lib/python2.4/site-packages/bzrlib/commands.py", line 650, in run_bzr_catch_errors
    return run_bzr(argv)
  File "/Library/Frameworks/Python.framework/Versions/2.4/lib/python2.4/site-packages/bzrlib/commands.py", line 612, in run_bzr
    ret = run(*run_argv)
  File "/Library/Frameworks/Python.framework/Versions/2.4/lib/python2.4/site-packages/bzrlib/commands.py", line 304, in run_argv_aliases
    return self.run(**all_cmd_args)
  File "/Library/Frameworks/Python.framework/Versions/2.4/lib/python2.4/site-packages/bzrlib/builtins.py", line 795, in run
    push_result = br_from.push(br_to, overwrite)
  File "/Library/Frameworks/Python.framework/Versions/2.4/lib/python2.4/site-packages/bzrlib/decorators.py", line 127, in read_locked
    return unbound(self, *args, **kwargs)
  File "/Library/Frameworks/Python.framework/Versions/2.4/lib/python2.4/site-packages/bzrlib/branch.py", line 1661, in push
    stop_revision, _hook_master=master_branch)
  File "/Library/Frameworks/Python.framework/Versions/2.4/lib/python2.4/site-packages/bzrlib/decorators.py", line 127, in read_locked
    return unbound(self, *args, **kwargs)
  File "/Library/Frameworks/Python.framework/Versions/2.4/lib/python2.4/site-packages/bzrlib/branch.py", line 1520, in push
    target.update_revisions(self, stop_revision)
  File "/Users/mwh/.bazaar/plugins/bzrsvn/branch.py", line 219, in update_revisions
    push_as_merged(self, other, rev_id)
  File "/Users/mwh/.bazaar/plugins/bzrsvn/commit.py", line 381, in push_as_merged
    new_tree = target.repository.revision_tree(target.last_revision())
  File "/Users/mwh/.bazaar/plugins/bzrsvn/repository.py", line 308, in revision_tree
    return SvnRevisionTree(self, revision_id, inventory)
  File "/Users/mwh/.bazaar/plugins/bzrsvn/tree.py", line 60, in __init__
    self.id_map = repository.get_fileid_map(self.revnum, self.branch_path)
  File "/Users/mwh/.bazaar/plugins/bzrsvn/repository.py", line 234, in get_fileid_map
    self.revision_fileid_renames)
  File "/Users/mwh/.bazaar/plugins/bzrsvn/fileids.py", line 197, in get_map
    revmap[p] = map[p][0]
KeyError: u'bin/makedocsys'

bzr 0.15.0candidate0 on python 2.4.1.final.0 (darwin)
arguments: ['/Library/Frameworks/Python.framework/Versions/Current/bin/bzr', 'push', 'http://codespeak.net/svn/user/mwh/pydoctor/trunk']

** please send this report to <email address hidden>

Thanks,
mwh

[Michael Hudson-Doyle (mwhudson)]

I dug a little into this and found that it seems to be something to do with the history of the repository. The call to follow_branch_history in FileIdMap.get_map() doesn't seem to be returning enough of the history, and that's probably because the history of the repo looks a little like this:

/user/mwh/docextractor was created
/user/mwh/docextractor was svn mv-ed to /user/mwh/pydoctor
the contents of /user/mwh/pydoctor were svn mv-ed to /user/mwh/pydoctor/trunk

follow_branch_history doesn't seem to be following the mv from docextractor to pydoctor.

I don't really see why the ancient history of the repo is relevant, but that's probably just my limited understanding :-)

Of course, you've probably figured out this much already...

[Jelmer Vernooij (jelmer)]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Thu, 15 Mar 2007 10:36:48 -0000
Michael Hudson <email address hidden> wrote:

> I dug a little into this and found that it seems to be something to do
> with the history of the repository. The call to follow_branch_history
> in FileIdMap.get_map() doesn't seem to be returning enough of the
> history, and that's probably because the history of the repo looks a
> little like this:
>
> /user/mwh/docextractor was created
> /user/mwh/docextractor was svn mv-ed to /user/mwh/pydoctor
> the contents of /user/mwh/pydoctor were svn mv-ed
> to /user/mwh/pydoctor/trunk
>
> follow_branch_history doesn't seem to be following the mv from
> docextractor to pydoctor.
>
> I don't really see why the ancient history of the repo is relevant,
> but that's probably just my limited understanding :-)
Thanks, that helps. The history is needed so we can always use the same
file id for each file.

Cheers,

Jelmer
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iQCVAwUBRf/P4Qy0JeEGD2blAQIdVQP/eube24GUMzMYVCXb+AbFTMG9r+r0G5hP
yE7qC5drhHFW2A3WSo42gk9aSqkIwzcG+khndIgtXZCWowOKy+cA4/3E7XsboNtn
hiy0tmDbc8VVwe+DibeQNpwBGktkwGINsiN8JzWGg+/PC1RgTC2dmFZm44W4QHlt
oIe2dqM39wk=
=L+3o
-----END PGP SIGNATURE-----

[Michael Hudson-Doyle (mwhudson)]

I added a failing test in https://code.launchpad.net/~micahel/bzr-svn/devel

Hope it helps :)

[Jelmer Vernooij (jelmer)]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Michael Hudson wrote:
> I added a failing test in https://code.launchpad.net/~micahel/bzr-
> svn/devel
>
> Hope it helps :)
I think this has been fixed in the upstream 0.3 branch. I could
reproduce your bug with Bazaar 0.15 and your branch, but once I merged
it into the current 0.3 branch it suddenly went away, presumably because
of a performance improvement to commit I made recently.

Can you confirm that this bug has been fixed in the current 0.3 branch
(http://people.samba.org/bzr/jelmer/bzr-svn/0.3) ?

Cheers,

Jelmer
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iQCVAwUBRhhFcwy0JeEGD2blAQIzCQP/VttUr0ItJOwAo0ajXWkmgd/RfWuolwyF
bZEOo5OKit7pi7402ugI2jTIqxugg6MnUj9oSq6FGkvzTyjIZ2H9sBi2GHSYdpii
FGKofIsoXsYmKIdVFntEXVDSV94C5G0Wo7n20iKXFnFLljjeZumOvWX3pzpOqdSH
xMWixBlALQU=
=uhx6
-----END PGP SIGNATURE-----

[Michael Hudson-Doyle (mwhudson)]

Grr! The test case passes with current 0.3, but my original command line still fails in the same way :(

[Michael Hudson-Doyle (mwhudson)]

However, if I trash my working copies and rebranch, it works. Thanks :)

All a bit mysterious though....

[Michael Hudson-Doyle (mwhudson)]

The status should be "Fix Committed" until 0.4 is out, right?

[Jelmer Vernooij (jelmer)]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Michael Hudson wrote:
> The status should be "Fix Committed" until 0.4 is out, right?
Yep, exactly. That might be 0.3.3 though.

Cheers,

Jelmer
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iQCVAwUBRhl3MQy0JeEGD2blAQJZgQP9G+08SbW/dr6LWcgTSXsxyCa6sJbGXcsp
MH+ydLzzE/xSbtEKrNuT9tr5r/21WLr3JEuZp5dIndKz08Q4OFMJgxlsNhwA+71T
PM3v2zzlwzvFzX19p2LOk1CIas5dJdWX/ixATHtyIV8K4csoBUDbxrfVhH39jbGz
9fCqvFjDxxg=
=Lj/T
-----END PGP SIGNATURE-----