Failed to parse index (GIT_EOBJCORRUPTED (-28))

[kaisellgren]

I received this error message:

  _message    "An error was raised by libgit2. Error code = GIT_EOBJCORRUPTED (-28).\r\nFailed to parse index \n\t- Failed to parse index. Extension size is zero"    string

When doing this:

LibGit2Sharp.Repository repo = new LibGit2Sharp.Repository(FullPath);

Here's a test case, a little test repo you can try: https://docs.google.com/open?id=0B-3-X85VysdNSHRDVWFTNzM0enc

I merged topic-2 branch into master, and then I committed a new file to topic-2 and it keeps crashing after that.

I'm using the latest released version. Should I build and try a recent version?

Update: It works fine in Git cli and TortoiseGit.

Just for reference:

b8b8819 HEAD@{0}: commit: Added test.txt
50a2725 HEAD@{1}: merge master: Merge made by recursive.
ec42713 HEAD@{2}: checkout: moving from master to topic-2
788bd4c HEAD@{3}: merge topic-1: Merge made by recursive.
634ec14 HEAD@{4}: checkout: moving from master to master
634ec14 HEAD@{5}: checkout: moving from topic-2 to master
ec42713 HEAD@{6}: commit: Added file.
fda8d19 HEAD@{7}: checkout: moving from master to topic-2
634ec14 HEAD@{8}: commit: Added file.
fda8d19 HEAD@{9}: checkout: moving from topic-1 to master
88bca54 HEAD@{10}: commit: Added file.
fda8d19 HEAD@{11}: checkout: moving from master to topic-1
fda8d19 HEAD@{12}: commit (initial): Initial commit.

[nulltoken]

Thanks a lot for this report.

Just a quick question: Does git.git complain when dealing with this repo?

[kaisellgren]

What do you mean git.git?

[kaisellgren]

Ok. Now I am confused. The error kept repeating until I checkout out master (I was on topic-2 branch). I'll try to reproduce it and find the specific problem causing this.

[kaisellgren]

Hmm. It occurs after I've committed something and checkout another branch. But I think it may be my fault... I have a program open which monitors changes to the repository folder and as soon as the event fires it reloads the repository for latest info... maybe the event fires too quickly and the checkout is only partially complete... though I would believe file locks are being used...? but it almost feels like a mixed-state problem because as soon as I close my program and re-open it, this time it seems fine even though it just threw an error and I did not change the repository.

[nulltoken]

What do you mean git.git?

Sorry. "git.git" is a term to refer to the original git and disambiguate from tortoisegit, gitsharp, ...

My question was meaning something like "How does git handle this? What happens if you run a $ git status on this repo in a Bash console. Does everything look fine? Does git complain?"

BTW, @carlosmn thinks that this very issue might have already been fixed in libgit2

[kaisellgren]

I tried git status, git log. git fsck --full and nothing seems to implicate any problems. TortoiseGit also works fine in every place.

[carlosmn]

Yeah, this was an off-by-one that triggered in a particular layout of the cache that git keeps for directories. It was fixed in the library on the 20th and the fix was cherry-picked into development a few days later.

[nulltoken]

@kaisellgren The fix should then be part of the soon-to-be-released next version :)

Let's keep this issue open, I'll make sure everything works ok.
Please make sure the link will still be valid until this issue is closed.

[kaisellgren]

Sure, the link will stay valid.

[nulltoken]

Checked against latest vNext tip and indeed fixed. @carlosmn rulz.

Closing... and thanks for the report!