Knit header error when attempting to bzr merge

[Misha Dorman]

Have a local repo containing a devo branch on laptop; doing star merges to a central repo on a server (accessed via Windows file share).

Attempted to bzr merge from server and got:
> bzr: ERROR: Knit header error: '' unexpected

Doing a fresh clone from the server in a new repo, I have subsequently been able to check-in, push, and pull from the server repo; the problem is with the local repo and/or branch and/or checkout.

Creating a clone of the borked devo branch within the same local repo gets the same problem, so it is not a problem with the local checkout, but rather with the branch or repo.

Creating a clone of the clean server branch within the same local repo gets the same problem, so it must be a problem with the local repo not the particular branch.

I suspect that it became corrupted when I interrupted a bzr operation with ctrl-C (I thought it was wedged, though subsequently investigation suggests that another process was thrashing the disk at the time and it was an innocent victim).

How to proceed with investigation?

(This is using bzr 0.8.2 native Win32 on XP Pro on NTFS, and has also been replicated using bzr 0.9.0dev0 r1750 cygwin on XP Pro on NTFS.)

[Michael Ellerman (michael-ellerman)]

What does 'bzr check' say?

[John A Meinel (jameinel)]

It sounds like you had an empty .kndx file. Theoretically, deleting that file and trying again would probably work.

Is this still a bug?

[Reinhard Tartler (siretart)]

I've just experienced this bug myself, and indeed, I indentified an empty knit file in .bzr/04/l10nflist.c-20060605152223-c144340b278be354.kndx.

Deleting this file helped.

bzr check says:

bzr: ERROR: No WorkingTree exists for /srv/scratch/packages/xine/.bzr/checkout/.

This is:
ii bzr 0.8.2-1ubuntu3 bazaar-ng, the next-generation distributed version control system

[Misha Dorman (mishad-work)]

>It sounds like you had an empty .kndx file. Theoretically, deleting that
>file and trying again would probably work.

Thanks -- yes I had one 0 length .kndx, and deleting it fixed the problem
(the .kndx was recreated).

>Is this still a bug?

I haven't had a recurrence yet, so can't say whether or not it there is
still a bug there. I suspect that there 0-length file was created when I
interrupted a merge or pull operation, and probably only occurs if the
interruption occurs at specific points, so it might take a while to trigger
again in any case.

[John A Meinel (jameinel)]

In bzr 0.14, empty .kndx files will not be treated as corrupt, rather they will be considered un-initialized. So this should not require manual intervention.