zgo.at/git-stats

git-stats displays aggregate statistics for git repos.

There's a public instance at https://gitstats.arp242.net, with various popular repos and some things I found interesting, updated semi-regularly. There isn't really any way to add something to that (yet), partly because this isn't really finished yet, and partly because importing a repo can be relatively slow so I need to build something to manage that a bit better. But if you ask nicely I can add something.

Note there are MANY inaccuracies with all of this:

• A "commit" can mean different things in different projects; I've seen people merge small bugfixes with 20 commits, and I've seen people do a huge feature in a single commit.

• "Number of commits" doesn't tell the full story; Linus Torvalds is ranked #57 for Linux and he's not even the top Linus – Linus Walleij is ranked #22 with 4,338 commits (vs. Torvalds' 2,586).

• Domains may be misleading; for example the top two committers for ElasticSearch work for Elastic, but neither use an @elasticsearch.com email address.

• Committing code isn't the only way to contribute to a project.

• Some commits aren't code; for example README typo fixes, i18n updates, and things like that. For many projects this is a negligible amount of commits, but for some it's a large number. It filters some of these out but there is no guarantee it filters everything.

• People can use multiple accounts to commit to a project. It tries to merge as much as reasonably safe, but this is not 100%. There is also a small chance of a false positive in cases where two committers have exactly the same name (projects where two notable committers have exactly identical names are probably very rare though).

• People's affiliations are not fixed. There are people who committed to Go before they were employed at Google, while they were employed at Google, and after they were employed at Google. Or any combination of the above.

• Some projects use git "incorrectly" and don't record authorship information in the Author header. For example PostgreSQL, Vim, NeoVim, bash, probably more.

But all of that said, I still feel it's useful. Linux is not the average project, there tends to be a large amount of overlap between the top contributors and the people doing reviews, support, and the like, and the rest can be solved by actually look at the data and use your brain before uncritically accepting any numbers here.

As an aside, GitHub's graph isn't accurate as it only shows commits that can be linked to a GitHub account and excludes everything else. It's not too uncommon people commit with an email address not associated with their account, have their accounts deleted, or things like that. This can be fixed with a .mailmap file, but many don't bother.

A good example is mpv: compare with git-stats, where a major mpv author deleted their GitHub account, and many of the MPlayer/MPlayer2 authors don't have a GitHub account. The "top" author on GitHub is actually #11 (and the margin is huge). The chart also starts in 2010 instead of 2001. All combined it gives a completely misleading picture.

Also this codebase isn't super great; I quickly wrote much of this about 4 years ago to prove a point about something, and I quickly hacked up some more stuff to look at the contributors for the recent Redis license change. There's tons of obvious things that can be improved, made less ugly, etc.

Installation

Install from source with:

% go install zgo.at/git-stats@latest

Or just use go build (or install) after a git clone.

Other than a PostgreSQL database, there are no other dependencies.

Usage

You will need a PostgreSQL database, as getting the data out of git is too slow for many repos, and this drastically reduces storage requirements.

You can also use git-cache on local repositories, or it can automatically clone remote repositories to a cache directory.

The git repo is stored in a cache directory (default: /tmp/git-stats). This is /tmp by default as it's ephemeral: you don't need to keep the cache around; the last commit will be recorded, and on an update it will fetch a shallow clone from there.

Insert data in the database:

% git-stats update https://github.com/golang/go

Or a local repo:

% git-stats update ~/code/my-git-repo

There are two interfaces: the CLI and web UI. Start the web UI with:

% git-stats serve

And most of the interface should be self-explanatory.

You can mix the web and CLI usage: both are backed by the same database.

CLI usage

You can get stats with e.g.:

% git-stats author https://github.com/golang/go

Or with the short name:

% git-stats author go

The authors command is the main meat, as this was my primary interest, but there are also some others:

% git-stats ls go
% git-stats activity go

The web interface has a few more features though, basically just out of laziness.

Alternatives

• https://git-quick-stats.sh

• gitdm – https://lwn.net/Articles/290957/