I second this request. Been wondering more about this process ever since I noticed, sometime in January, that Rust (a nascent language with very little use yet) was ranked in the low-20s for language popularity. The problem likely had to do with how the Rust devs registered their language with Github's Linguist[1] project; Rust's units of compilation ("crates") have the .rc extension, and Github must have taken this to mean that every project with a .rc file (notably many Android-specific projects) contained Rust code. The Rust devs have since revoked that association from Linguist[2], but since Github takes a while to reevaluate a project's language breakdown, the language's popularity ranking on Github is still artificially inflated (sitting at #31 at the moment).

[1] https://github.com/github/linguist

[2] https://github.com/mozilla/rust/issues/1512

.rc files don't dominate a project in file count or SLOC, so I think the answer to jashkenas' question is something like this:

    pop[lang] = avg(loc(proj, lang) / loc(proj) for proj in projects)

That depends on the project:

https://github.com/CM7-GB/android_device_motorola_droid2we

Note the language breakdown, under the Graphs tab.

But yes, it all hinges on how Github measures popularity. For one, I'm curious how forked repositories are counted; if I have a 100kloc Blub project and it gets forked 100 times, does Github count that as 100,000 or 10,000,000 lines of Blub?

defunkt has confirmed that at least in the past if you had a repo with 100 Perl files and one JavaScript file it would count equally towards both languages.

My guess is the latter.

It's the number of occurrences of a language in all public repositories. Most web applications on GitHub have some javascript, for instance.