LFX Insights available for LLVM, Rust, Swift, and other compilers

LLVM Project
1

Hey folks, one of the cooler demos I saw at OSSNA this year was the LFX Insights tool, which now has data for LLVM! Open source project analytics are not new. There are many such tools (8knot, Grimoirelab, github, openhub/ohloh, etc), but this tool does two things that are new at least to me:

  • PR metrics: Average response time and time to close are good examples. We are having a lot of conversations right now about PRs, branch protection, AI, and maintainer burden from reviews, and this information seems pretty relevant. As long as I’ve been involved in LLVM, I’ve heard complaints from newcomers that it is difficult to figure out how to find a maintainer and get a review. Now, in the year 2025, we can actually ask and answer more questions with data.
  • Organizational affiliation: The tool attempts to map individuals to organizations so we can assess which organizations are putting resources into the project, and what exactly the nature of their contributions are (writing code, reviewing code, or managing the issue tracker). I believe individual organization membership is not visible in the tool for privacy reasons. The data is imperfect, but I think it’s much better than you’d get if you simply used committer email domain names to model affiliation.

I think the organizational contribution leaderboard is off by quite a bit for two major reasons:

  • llvmbot and llvm-ci appear to be considered human users in the LLVM organization
  • Users with no other GitHub affiliation are attributed to the LLVM organization

For these reasons, the LLVM org is the leading organizational contributor, but we all know the LLVM Foundation doesn’t fund directly fund the development of LLVM the project. I filed a data quality issue, so I hope it gets better over time, but it seems useful even in its current form.

There is a collection including relevant compiler toolchain projects, so you can look at the metrics for projects in the same domain, such as Swift, Rust, and GCC.

4 Likes
2

This is some nice data, I did not find anything that really surprised me yet. I appreciate they also track issue contributions.

The only think that stands out as missing, is that they don’t have knowledge to be able to split things like front-end Vs back-end etc Looking at the data aggregated will miss some important trends that would be useful to see. Maybe there is a way to see that and I missed it.

3

I understand the potential privacy implications, but it’d be helpful if you could e.g. debug or fix your own affiliation in case it’s accounted for incorrectly. For my org, the commit count in the past year is only a third of what filtering git logs by emails suggests, for example.

4

I wish there was statistics based on labels attached to the PR. If we are to gather more data in the future, can we politely ask for this feature? IMHO, review experience can vary a lot depending on subproject.