bzr: ERROR: Bad restart - attempted to skip commit :419 but matching revision-id is unknown

[Elliot Murphy]

I'm trying to do a periodic import of a hg repo to bzr, and have set this up in a cron job.
the initial import went fine. when some new revisions showed up in hg though, I got this error:

emurphy@jitter-shadow:~/tangent-conversion/tangent-conversion/libmemcached.bzr$ ~/.bazaar/plugins/fastimport/exporters/hg-fast-export.py --repo=../libmemcached/ | bzr fast-import -
17:24:34 Found 418 commits already loaded - skipping over these ...
master: Link new branch to parent [:418]
master: Exporting simple delta revision 419/423 with 0/2/0 added/changed/removed files
bzr: ERROR: Bad restart - attempted to skip commit :419 but matching revision-id is unknown
Traceback (most recent call last):
  File "/home/emurphy/.bazaar/plugins/fastimport/exporters/hg-fast-export.py", line 396, in <module>
    options.statusfile,authors=a,sob=options.sob,force=options.force))
  File "/home/emurphy/.bazaar/plugins/fastimport/exporters/hg-fast-export.py", line 344, in hg2git
    c=export_commit(ui,repo,rev,marks_cache,heads_cache,last,max,c,authors,sob,brmap)
  File "/home/emurphy/.bazaar/plugins/fastimport/exporters/hg-fast-export.py", line 243, in export_commit
    export_file_contents(ctx,man,changed)
  File "/home/emurphy/.bazaar/plugins/fastimport/exporters/hg-fast-export.py", line 122, in export_file_contents
    wr(d)
  File "/home/emurphy/.bazaar/plugins/fastimport/exporters/hg-fast-export.py", line 27, in wr
    print msg
IOError: [Errno 32] Broken pipe

I'm attaching .bzr.log and a tarball of both the hg and the bzr repos that I was using so that you can repro it locally.

[Elliot Murphy (statik)]

i'm not able to bring any new revisions over from the libmemcached repository due to this bug.
I forgot to mention that I am running the very latest revision of the bzr-fastimport plugin, and using hg-fast-export.py that is bundled with this plugin.

[Ian Clatworthy (ian-clatworthy)]

The "restart" (a.k.a. poor man's mirroring) logic depends on the exporter producing a superset of the original data by appending new data on the end. My initial guess is that hg-fast-export isn't doing this (yet). I'll take a deeper look soon if someone else doesn't beat me to it.

[Ian Clatworthy (ian-clatworthy)]

Can you please retest this with the latest trunk (rev 108)?

[ilia (ilia)]

Hi,
I have revision 136 of fastimport and the issue still exists. Session log is attached. I've run hg-fast-export, then bzr fast-import. The former complained (but completed the export):
*** failed to import extension hgext.inotify: 'module' object has no attribute 'inotify'

The later has failed with the same message as noted above:
bzr: ERROR: Bad restart - attempted to skip commit :4134 but matching revision-id is unknown

Due to this bug I'm stuck with old branch and can't continue the development. Please-please-please, resolve it or at least suggest a workaround.

[ilia (ilia)]

The first message (about hgext.inotify) seems to be unrelated to the issue and already fixed in mercurial: http://hg.intevation.org/mercurial/crew/rev/b5a6fce0bc29
I've applied the fix to mercurial and tried again with the same result.

[Ian Clatworthy (ian-clatworthy)]

I have a few ideas on how to fix this but they depend on me better understanding how hg-fast-export behaves. Can you please upload the first few commands from the output that hg-fast-export creates? It does seem strange that bzr-fast-import is trying to skip N commits but is complaining about not knowing N+1 in those loaded so far. It makes me think that perhaps hg-fast-export isn't including those original N commits in the stream it's creating?

[ilia (ilia)]

Attaching hg-fast-export.py output. The first mark is 4134.