not a valid key error

[Jelmer Vernooij]

I'm getting this error when committing to one of my repositories. This happens during repacking.

bzr: ERROR: The key '<email address hidden>' is not a valid key.

This is a regression in r6256.

[Jelmer Vernooij (jelmer)]

I can also reproduce this on another repository, where it also happens during repacking.

[Jelmer Vernooij (jelmer)]

(Pdb) print rows[-1].writer.bytes_out_len
2

[Jelmer Vernooij (jelmer)]

(Pdb) print repr(string_key)
'changelog-20061227002451-5odpmebpo2vuu6ma-7\<email address hidden>'

[Martin Pool (mbp)]

On 14 November 2011 06:27, Shannon Weyrick <email address hidden> wrote:
> ** Changed in: bzr
>     Assignee: (unassigned) => Shannon Weyrick (weyrick)

Thanks for taking this, Shannon. I might look into reverting your
previous change so that trunk does not stay broken.
--
Martin

[Martin Pool (mbp)]

bug 720853 is therefore reopened

[John A Meinel (jameinel)]

See https://bugs.launchpad.net/bzr/+bug/720853/comments/4 for why this is happening. But the short summary is:

During repack, we compress as much as we possibly can, to the point of creating a new zlib.compressobj once the page cannot hold any more content. At this point, zlib has buffered all the raw bytes and has only written out its little 2-byte header. We won't see the real content until finish() calls flush().

So instead, we should just change how we detect writing to an otherwise-empty chunk. We can do so pretty easily, which shouldn't have these false positives. (Instead of detecting how many bytes we've written, just detect that we created a new empty page.)

[Shannon Weyrick (weyrick)]

So, real fix for this should be in bug #720853, but I'll work on a regression test for this bug.