diff --git a/.cargo/mutants.toml b/.cargo/mutants.toml index 25a8ad1f..ad40b9f2 100644 --- a/.cargo/mutants.toml +++ b/.cargo/mutants.toml @@ -40,7 +40,7 @@ exclude_globs = ["**/*_tests.rs"] # - equivalent: identical on every reachable input, panics and side # effects included; # - unreachable: the code runs for no language; -# - non-terminating under mutation; +# - non-terminating, or allocating without bound, under mutation; # - same output, different work: no metric moves, and the entry names # the work the mutant skips or adds, which a test-support counter can # observe; @@ -233,6 +233,15 @@ exclude_re = [ # stale line stops excluding this mutant, and a line one off excludes # the live one. '^big-code-analysis-ast/src/checker\.rs:1105:9: replace \|\| with && in rust_attribute_run_member$', + # Allocates without bound under mutation. With `*` a set's last word + # is its offset times its length, so `union_with` reserves that far + # and every merge up a space tree multiplies the span again. The + # suite does notice: several hundred tests fail on the wrong rows. + # But a nested fixture then asks for a span of gigabytes, and on a + # hosted runner the test binary stalls past the timeout before it can + # report (run 43437925608: 168 GB requested, judged a timeout; run + # 43278927651 judged the same mutant caught). + '^src/metrics/loc/line_set\.rs:\d+:\d+: replace \+ with \* in LineSet::last_word$', # Non-terminating under mutation. The loop then breaks when # `goto_next_sibling` succeeds and repeats when it fails, so a list # with one child is visited forever. Its one caller walks the values diff --git a/.github/workflows/mutation-test.yml b/.github/workflows/mutation-test.yml index 6191029a..453e11d4 100644 --- a/.github/workflows/mutation-test.yml +++ b/.github/workflows/mutation-test.yml @@ -46,10 +46,11 @@ jobs: # about 165 min at those slow-runner rates; the margin to 270 is for # the runner variance that made identical shards differ 1.6x. The # second complete run (43278927651, sixteen shards of 341 at - # 68b70b08) took 125 to 222 min, two shards past 200, so the count is - # eighteen: ~292 per shard of the 5249 the list holds once the test - # modules and the excluded mutants are out, about 190 min at that - # run's slowest rate of 39 s per mutant. + # 68b70b08) took 125 to 222 min, and the third (43437925608, eighteen + # shards of 291 at 3fd38e5d) 108 to 237: its two slowest shards held + # no escape, so the 2.2x spread is the runners', at 22 to 49 s per + # mutant. The count is twenty: ~261 per shard of the 5227 the list + # holds, about 213 min at that slowest rate. prepare: name: shard list runs-on: ubuntu-latest @@ -61,7 +62,7 @@ jobs: - id: define run: | set -euo pipefail - shards='[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17]' + shards='[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19]' echo "shards=${shards}" >> "$GITHUB_OUTPUT" echo "count=$(jq -n "${shards} | length")" >> "$GITHUB_OUTPUT" diff --git a/docs/development/mutation_testing.md b/docs/development/mutation_testing.md index f5d34332..490d42d0 100644 --- a/docs/development/mutation_testing.md +++ b/docs/development/mutation_testing.md @@ -128,12 +128,15 @@ artifact that records per-mutant `duration` — prices a root-crate mutant at ~28 seconds and an ast-crate one, which rebuilds both crates, at ~39, with a missed mutant about twice either since both suites then run to completion. The ast crate holds about 31 % of the -list since #1728 widened the selection. The current eighteen rests on -the second complete run (43278927651, 2026-11-12, at 68b70b08): its -sixteen shards of 341 mutants run for 125 to 222 minutes, two of them -past 200, and the slowest prices a mutant at 39 seconds. The list holds -5249 mutants with the test modules and the excluded mutants out, so -eighteen shards hold about 292 each, roughly 190 minutes at that rate. +list since #1728 widened the selection. The current twenty rests on +the third complete run (43437925608, 2026-11-13, at 3fd38e5d): its +eighteen shards of 291 mutants run for 108 to 237 minutes, two of them +past 200. Those two hold no escape, so the spread is the runners': 22 +to 49 seconds per mutant for the same work. The list holds 5227 +mutants with the test modules and the excluded mutants out, so twenty +shards hold about 261 each, roughly 213 minutes at the slowest rate +seen. Size for that rate, not the median: a shard that passes 270 is +cancelled and the run does not count. The laptop figure the count was first sized from, about ten seconds per mutant, is roughly half the hosted cost. Both halves of `utils/mutation-report.py` are covered by `make