2021-04-06 Triage Log

A pretty major week for memory usage improvements, while wall time performance largely stayed neutral, with an average of ~20% gains on memory usage for release builds, and 5% on check builds, due to an update in the default allocator used (to a more recent jemalloc).

Triage done by @simulacrum. Revision range: 4896450e7e0a522486b4d3a8d360ac4e1d2072a0..d32238532138485c80db4f2cd596372bce214e00

1 Regressions, 4 Improvements, 0 Mixed

Regressions

add OR_PATTERNS_BACK_COMPAT lint #83468

  • Moderate regression in instruction counts (up to 1.2% on incr-patched: println builds of coercions-debug)
  • Relatively small hit on a number of benchmarks. Likely largely unavoidable.

Improvements

Stream the dep-graph to a file instead of storing it in-memory #82780

panic early when TrustedLen indicates a length > usize::MAX #83726

  • Large improvement in instruction counts (up to -7.9% on incr-patched: compile one builds of regex-debug)

Use tikv-jemallocator in rustc/rustdoc in addition to jemalloc-sys when enabled. #83152

  • Large improvement in instruction counts (up to -5.4% on incr-unchanged builds of deep-vector-check)
  • Pretty nice improvements to instruction counts, though not a major contributor to wall times it seems.

Stream the dep-graph to a file instead of storing it in-memory. #82780

  • Moderate improvement in instruction counts (up to -1.8% on incr-full builds of unused-warnings-check)

Mixed

  • None

Nags requiring follow up

  • The issue from three weeks ago has stalled. Likely this needs someone to mentor the author, or to take over the investigation.