Measuring Language Usage

How do you measure usage of different programming languages?

There are several widely-used rankings, but they are surprisingly different. They can't all be right.

TIOBE

TIOBE is the most famous and most referenced. It's not always the most accurate.

Delphi programmers allegedly gamed the rankings in 2009 and it's still surprisingly high according to TIOBE.

There are also suggestions that TIOBE confuses generic seach terms with language names.

GitHub

GitHub is the world's biggest public code repository, so the methodology is clear. GitHub data is used in quantitative programming language research papers.

GitHub was adopted by the Ruby community early on, so the quantity of Ruby projects was not always representative of the world at large. This is reverting to the mean as GitHub grows.

GitHub's biggest weakness is that its dataset is only based on public repositories. Undoubtedly proprietary code is (slightly?) different in language popularity rankings.

Stack Overflow

Stack Overflow has tags on questions (so you can see which questions are most discussed), as well as the annual survey.

The survey is the biggest of its kind, and the best quantitative dataset for language use to date.