Breakwater

Backtest / #1253

perf(specifiers): keep range caches across canonicalization

pypa/packaging#1253 · merged 2026-06-16 · base 26b89a05f6 → head e64c20eb7a · ran on Docker

PASS

No behaviour divergence found by any probe.

† Merged after Nemotron 3 Ultra was released (2026-06-04): the model cannot have seen this PR or its fixes.

Performance · SpecifierSet#contains

No realistic input family measured slower.

Measured in backtest packaging-v2 (same commits, run bt-1253-7b2670, image bw-pair:92555005a9f709e0). The semantic results above are from a later run; perf doesn't use the semantic planner, so it was not re-measured.

head slowerhead faster1×1001,00010,000input size (characters)R1R1: 1.00× at ~23 charsR1: 1.00× at ~37 charsR1: 1.00× at ~128 charsR1: 1.03× at ~512 charsR1: 1.00× at ~2,205 charsR1: 0.99× at ~9,142 charsR1: 1.00× at ~39,862 chars
  • R1
Head/base cachegrind instructions per call, by input size (log-log). Each line is one input family Nemotron proposed; points above 1× are where the "faster" code does more work.
FamilyInputsRatio range Slope base → headVerdict
R1many_simple_specs_variant_1claimed speedup · archive0.99× – 1.03×0.93 → 0.93neutral

Receipt

TargetBackendImageRun AuditAdjudication
packaging.specifiers:SpecifierSet#containslocal Dockerbw-pair:6fade3cf7634d561bt-1253-4ad193$0.056$0.000

Nemotron cost on Nebius Token Factory. Backend: where the untrusted code ran. local Docker (an image cited by its content-addressed tag). Every image is content-addressed (repo at each commit + harness), so every card re-runs from the image and seed 0. Raw data: audit.