BadObject on repo after git.gc()

[nova77]

Example code:

>>> r = Repo.init('/tmp/foo/', bare=False)
>>> print >> open('/tmp/foo/bar.txt', 'w'), 'some content'
>>> r.index.add(['bar.txt'])
>>> r.index.commit('commit bar.txt')
>>> print r.git.status() # all clear?
# On branch master
nothing to commit (working directory clean)
>>> r.git.gc()
''
>>> r.iter_commits() # kabooom!
...
BadObject: 39154e2f45f96bdbbb86aa4f1ffc72f224b88f31

Tested on GitPython 0.3.2.RC1

[Byron]

It seems the repo didn't update its internal state, probably its a performance implementation of the (default) python backend.
Alternatively, you might try the GitCmdDB backend, it should cope with the issue, as a workaround.

In any way, this is a bug that one might want to address indeed.

[Byron]

With your current backend, you can also call

repo.odb.update_cache()

to read in the newly created packs.

[nova77]

Ah, that workaround is good enough for me, thanks :)

[cboylan]

The repo.odb.update_cache() fix appears to only work if there were pack files present when the repo object was created. If only loose files were present when the repo object was created that repo object doesn't seem to pick up the new pack files when update_cache() is called.

[Byron]

I acknowledge that. Apparently on initialization, it will add handlers for whichever storage type it finds, like loose objects, packs, or alternate info files.
If a storage type is not available, it will never look for it again. The only workaround is to recreate the repository instance.

[Byron]

I have to consider this an issue that can't be fixed automatically, as the only solution would be to watch all commands issued through the git command generator instance. However, as it doesn't have access to it's parent repository, it can't watch for commands like gc and automatically update caches.

Alternatives are as follows:

• call repo.odb.update_cache() yourself after such a call
• initialise your repository with an ODB implementation that doesn't cache, like the GitCmdObjectDB, such as in git.Repo('.', odbt=git.GitCmdObjectDB). The latter DB uses the git command to stream objects, which could also be faster depending on the task. Probably it would also cause less file handles to be used (see Merged fd leaks fix from the master to the 0.3 version #150)