2021-06-01 Triage Log

Busy week, with several reverted PRs due to performance regressions, but overall a positive week.

Triage done by @simulacrum. Revision range: cdbe2888979bb8797b05f0d58a6f6e60753983d2..1160cf864f2a0014e3442367e1b96496bfbeadf4

3 Regressions, 3 Improvements, 5 Mixed

Regressions

Merge CrateDisambiguator into StableCrateId #85804

  • Moderate regression in instruction counts (up to 1.6% on full builds of html5ever-opt)
  • Seems to differ from the benchmark on PR, but held up over the rest of the week on the graph. Suggested a revert on PR.

Improvements

Don't hash thir_body #85729

  • Moderate improvement in instruction counts (up to -4.0% on incr-full builds of deep-vector-check)

Optimize proc macro bridge #85390

  • Large improvement in instruction counts (up to -5.7% on full builds of token-stream-stress-check)

Mixed

Add TrustedRandomAccess specialization for Vec::extend() #83770

  • Moderate regression in instruction counts (up to 2.1% on incr-patched: b9b3e592dd cherry picked builds of style-servo-debug)
  • Moderate improvement in instruction counts (up to -1.6% on incr-patched: println builds of cargo-debug)
  • Reverted in the next result, seems to surprisingly in practice not be an improvement on some of our larger benchmarks.

Revert “Auto merge of #83770 - the8472:tra-extend, r=Mark-Simulacrum” #85754

  • Moderate improvement in instruction counts (up to -2.1% on incr-patched: b9b3e592dd cherry picked builds of style-servo-debug)
  • Moderate regression in instruction counts (up to 1.5% on incr-patched: println builds of cargo-debug)

Simplification of query forcing #85319

  • Moderate improvement in instruction counts (up to -3.2% on full builds of unused-warnings-check)
  • Moderate regression in instruction counts (up to 1.4% on incr-unchanged builds of stm32f4-check)
  • Regressions seem to be limited to unused-warnings incremental.

Reduce the amount of untracked state in TyCtxt #85153

  • Very large regression in instruction counts (up to 1038.2% on incr-unchanged builds of regex-opt)
  • Moderate improvement in instruction counts (up to -1.3% on full builds of cranelift-codegen-check)
  • This was a pretty big regression, and wasn't detected on the PR run, presumably due to work since the merge.

Revert “Reduce the amount of untracked state in TyCtxt” #85884

  • Very large improvement in instruction counts (up to -91.2% on incr-unchanged builds of regex-opt)
  • Moderate regression in instruction counts (up to 1.4% on full builds of cranelift-codegen-check)

Nags requiring follow up