Walker sometimes fails to hide commits

[pkill]

Background

I recently upgraded GitLab from v7.3 to v7.4. One of the big changes in GitLab supplants Grit with Rugged to compute comparisons between commit objects. I'm noticing undesired behavior with computing comparisons: In some, but not all cases, the comparison returns the entire history of the branch rather than the difference between two branches. There is an active GitLab bug report that expresses others frustrations with this issue: https://github.com/gitlabhq/gitlabhq/issues/8174

This bug is causing a variety of downstream problems the administration of a GitLab v7.4 instance.

Verification

I fired up irb to isolate various Rugged versions and reproduce the behavior. Unfortunately I cannot furnish the repository used here for reproduction. As you can see, Rugged 0.19 produces the correct number of commits between those to objects but Rugged 0.21 is not properly hiding the rest of the commits and returning everything.

Just glancing at the code, it appears that the walker is not properly hiding the rest of the commits.

Rugged 0.19

2.1.2 :005 > w = Rugged::Walker.new(Rugged::Repository.new('/home/git/repositories/repo/repo-tools.git'));w.push('db6e3023debc1d6af9a73373492db11a6849b6ec');w.hide('bd30827036271ada67cb303373dfe7889490382d');w.to_a.size
 => 15 

Rugged 0.21

2.1.2 :001 > w = Rugged::Walker.new(Rugged::Repository.new('/home/git/repositories/repo/repo-tools.git'));w.push('db6e3023debc1d6af9a73373492db11a6849b6ec');w.hide('bd30827036271ada67cb303373dfe7889490382d');w.to_a.size
 => 2223 

Here is my environment details:

$ uname -a
Linux server 2.6.32-431.11.2.el6.x86_64 #1 SMP Tue Mar 25 19:59:55 UTC 2014 x86_64 x86_64 x86_64 GNU/Linux
$ ruby --version
ruby 2.1.2p95 (2014-05-08 revision 45877) [x86_64-linux]

[arthurschreiber]

Have you tried the same thing with the latest version of Rugged from git?

[pkill]

Is there a gem already released for a newer version?

[arthurschreiber]

You could try 0.21.1b2 (a beta gem).

[pkill]

0.21.1b2 looks to have solved the regression. Do you recommend 0.21.1b instead?

[pkill]

@arthurschreiber was this a problem with Rugged or libgit2?

[arthurschreiber]

This is a problem with libgit 0.21.0. In general, 0.21.1b2 should be fine, but it might include breaking API changes, so YMMV.

[gregbuehler]

@arthurschreiber Can you speak in general terms as to the timeline for a 0.21.1 release (Days, Weeks, Months)?

[arthurschreiber]

Except for potential API changes, 0.21.1b2 should be as "stable" as it gets. I think that's also what Github has been using in production since October.

Also, the 0.21.1b2 name is a bit of a lie, as the bundled libgit2 version is not API compatible with libgit2 0.21.1, but is instead based on some commit off the libgit2 master branch.

@vmg I'm not sure what to do here. I'd like to release a 0.21.2 version of Rugged, based on the libgit2 0.21.2 release, but it won't be compatible with the hashsig changes in rugged 0.21.1b2. Is that a problem?

[arthurschreiber]

Hey folks,

I just release 0.21.2, which comes bundled with libgit 0.21.2 and should fix the issue you're seeing with the walker. Let me know if you run into any issues with that release.