GNU Cauldron 2018 Agenda

On Friday September 7th, 2018 the GNU C Library community held a BoF at GNU Cauldron 2018.

GNU Cauldron 2018 Notes

We started with math, with dummy data.
Current goal: “relevant” benchmark workloads
glibc microbenchmarks was added to the openbenchmarking.org project
Are you doing any performance optimization work?
How do you use the benchmarks to guide your work?
Should we gate performance patches on microbenchmarks?
- General feeling: no absolute requirement, depends on patch and on expert opinion.
Benchmarks should be suggested on on libc-alpha. Little implementation overhead (for generating JSON output).
But users report that running individual benchmarks in-tree is difficult.
Florian might contribute his unwinder example code.
Carlos suggests to add benchmarks for RSS usage in functions (printf).

Proposal: Reviewed-By: only change to posted patch. No agreement or strong disagreement.
ChangeLog automation and simplified review process is desirable.
Joseph explains the situation with GNU standards (see mailing list), and the need for an auto-generation tool for ChangeLog entries for the described changes. This would also useful for the offline case.
Commit message contents would have to be reviewed. Information in the commit message would have to replace useful parts (the non-auto-generatable parts) of the ChangeLog entry.
No one wants the present state. Room is split 50/50 between ChangeLog entry in commit messages and dumping “git log”-like output into the tarballs.

Can we remove awk from the bootstrap? No, GCC relies on it.
Some in the room object to requiring Python 3 for the build. (“It makes the build environment bigger. It makes bootstrap longer and harder and maintain it so keep it working.”)
Disagreement about the difficulty of bootstrapping Python for a new architecture (and the difficulty of cross-building), particularly in comparison to Perl.