fix for a keyerror in finding least common ancestors

[Michael Hudson]

I have an branch of launchpad with pretty complicated merge history.
Attempting to merge a particular revision of launchpad head into it
yields an exception like:

Traceback (most recent call last):
  File "/home/mwh/bzr/bzrlib/commands.py", line 729, in run_bzr_catch_errors
    return run_bzr(argv)
  File "/home/mwh/bzr/bzrlib/commands.py", line 691, in run_bzr
    ret = run(*run_argv)
  File "/home/mwh/bzr/bzrlib/commands.py", line 389, in run_argv_aliases
    return self.run(**all_cmd_args)
  File "/home/mwh/bzr/bzrlib/builtins.py", line 2709, in run
    change_reporter=change_reporter)
  File "/home/mwh/bzr/bzrlib/builtins.py", line 3724, in _merge_helper
    merger.set_base(base_revision)
  File "/home/mwh/bzr/bzrlib/merge.py", line 268, in set_base
    self.base_rev_id = graph.find_unique_lca(*revisions)
  File "/home/mwh/bzr/bzrlib/graph.py", line 279, in find_unique_lca
    lca = self.find_lca(*revisions)
  File "/home/mwh/bzr/bzrlib/graph.py", line 140, in find_lca
    return self._filter_candidate_lca(border_common)
  File "/home/mwh/bzr/bzrlib/graph.py", line 244, in _filter_candidate_lca
    del active_searchers[candidate]
KeyError: 'pqm at pqm.ubuntu.com-20070712051010-4o5ij0wgrvhvdf0o'

I think the most obvious fix (attached) is actually the right fix (it's
not at all obvious to me that the candidate being deleted here hasn't
already been deleted by line 249).  In a half an hour of trying, I
haven't been able to make a test case, but that's not surprising as I
spent most of that time trying to understand the terminology of this file...

Cheers,
mwh
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: find-lca-problem.bundle
Url: https://lists.ubuntu.com/archives/bazaar/attachments/20070730/ce546052/attachment.diff