Because GitLab caches repository archive files on disk it is important
to avoid leaving a corrupt archive when a Unicorn worker gets
interrupted. If we do not watch out for this, the next user to download
the same revision as an archive gets the unfinished file from the
interrupted worker.

This change uses a (hopefully atomic) rename to put the archive file
in its expected location.

Fix external archive compression

See merge request !11

We were seeing two test failures:
- encoding error in TarWriter;
- 'PG::Error result was cleared'.

The first is solved with an IO.new. The second error seems to have to do
with forking; this change rewrites Repository#create_archive to not to
have to fork the Ruby (Unicorn) process.

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Refactor Repository#archive_repo for tar files

Use `IO.pipe` instead of `StringIO` to archive repositories to tar files.  This addresses issue #1.

See merge request !5

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Add empty directories for submodules in both zip and tar files when
archiving a repo.

Use `IO.pipe` instead of a `StringIO` to avoid storing the whole tar
file in memory.

Cleanup gitlab_git

Remove GitStats, LogParser and Stats classes.

They are not generic and depends deeply on graphs implementation inside
GitLab. So keeping them in library does not makes sense. In order to
implement graphs in GitLab we can just use Repository#log. No need to
have such duplication in code with strange parsers

See merge request !10

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

They are not generic and depends deeply on graphs implementation inside
GitLab. So keeping them in library does not makes sense. In order to
implement graphs in GitLab we can just use Repository#log. No need to
have such duplication in code with strange parsers

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Don't set a message for lightweight tags

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Signed-off-by: Dmitriy Zaporozhets <dmitriy.zaporozhets@gmail.com>

Hound prefer single quotes.

Use Rugged in Commit, Diff, and Repository.log()

Remove Grit from the gemspec, and remove the ability to wrap Grit
objects in the Commit and Diff classes.

The previously-used version of SimpleCov doesn't seem to play well with
the forking that Repository does when archiving a repo to a file.

Use the rugged object instead of grit in the README's examples.

Implement all classes using Rugged objects instead of Grit

Return Gitlab::Git::Ref objects instead of raw references.  Also stop
caching `@heads` to avoid being out of sync with the repo.

Return an array of Gitlab::Git::Diff objects instead of a Rugged::Patch
array so that Gitlab doesn't have to deal with raw objects.

Add methods to the Repository class to reduce Gitlab's use of raw
Grit/Rugged objects.

Allow a Gitlab::Git::Commit or a Gitlab::Git::Diff object to be created
from either a Rugged or a Grit object.