fe0898cc14
We removed calls from our code to Rugged::Repository#fetch_attributes:
<340e111e04
>
However, we didn't remove calls from within Linguist. This method is only called
when calculating the languages for a repository on the Charts page:
<https://github.com/github/linguist/blob/v4.7.0/lib/linguist/lazy_blob.rb#L33-L36>
We can safely use our own Gitlab::Git::Attributes here. On staging, for the
GitLab CE repo, this makes the calculation take about a third of the time:
# Before
Benchmark.realtime do
Linguist::Repository.new(repository.rugged,
repository.rugged.head.target_id).languages
end
#=> 23.67193900188431
# After
Benchmark.realtime do
Linguist::Repository.new(repository.rugged,
repository.rugged.head.target_id).languages
end
#=> 8.945212290156633
25 lines
982 B
Ruby
25 lines
982 B
Ruby
# We don't want to ever call Rugged::Repository#fetch_attributes, because it has
|
|
# a lot of I/O overhead:
|
|
# <https://gitlab.com/gitlab-org/gitlab_git/commit/340e111e040ae847b614d35b4d3173ec48329015>
|
|
#
|
|
# While we don't do this from within the GitLab source itself, the Linguist gem
|
|
# has a dependency on Rugged and uses the gitattributes file when calculating
|
|
# repository-wide language statistics:
|
|
# <https://github.com/github/linguist/blob/v4.7.0/lib/linguist/lazy_blob.rb#L33-L36>
|
|
#
|
|
# The options passed by Linguist are those assumed by Gitlab::Git::Attributes
|
|
# anyway, and there is no great efficiency gain from just fetching the listed
|
|
# attributes with our implementation, so we ignore the additional arguments.
|
|
#
|
|
module Rugged
|
|
class Repository
|
|
module UseGitlabGitAttributes
|
|
def fetch_attributes(name, *)
|
|
@attributes ||= Gitlab::Git::Attributes.new(path)
|
|
@attributes.attributes(name)
|
|
end
|
|
end
|
|
|
|
prepend UseGitlabGitAttributes
|
|
end
|
|
end
|