Surviving mutants — value-moving modules

• Surviving mutants
• Surviving mutants
• Surviving mutants
• Surviving mutants
• Surviving mutants
• Surviving mutants
• Surviving mutants
• Surviving mutants
• Surviving mutants

Each file here lists the mutants that survived a cosmic-ray run: the suite executed that mutated line and no assertion failed. They are the raw output of scripts/mutation_survivors.py, one row per mutant — file, line, enclosing definition, mutation operator, and the exact source change.

This is a worklist, not a defect list. A survivor means one of two things and the difference is a judgement call a human has to make:

• Nothing tests this. The branch runs, the behaviour changed, and the suite did not notice. These are the ones worth a test. When you write one, put it in tests/test_mutation_hardening.py and name the mutant class it kills, the way the existing entries there do.

• Nothing can test this. The mutant is equivalent — it does not change behaviour. The classes that dominate in this codebase are catalogued in ../../how-to/mutation-testing.md: type annotations (bytes | None → bytes - None, never evaluated under from __future__ import annotations), error-message f-strings, and interpreter-detail rewrites such as == → is on interned values.

Reading a row

Column	Meaning
Line	1-based line in the module, at the commit named below
Definition	the enclosing function or method — (module level) when there is none
Operator	the cosmic-ray operator, with its core/ prefix stripped
Original / Mutant	the single source line before and after the mutation
(last)	annot marks a BitOr-family rewrite of a type annotation — X | None becoming X + None. These cannot be killed: from __future__ import annotations makes annotations strings that are never evaluated. Start with the unmarked rows. They are flagged rather than dropped so the file stays a complete record if the heuristic is ever wrong.

A row is triaged by answering one question: if this mutant shipped, what would go wrong, and why did no test see it? If the answer is “nothing would go wrong”, it is equivalent — say so in the commit that removes it from the worklist. If the answer is anything else, it needs a test.

Provenance

Produced 2026-08-12 from a full sweep of all seven value groups — 8 171 mutants, 5 049 killed, 3 122 survived (61%) — against the module sources at commit 3115028, which are byte-identical to those at 724fe92. Per-module rates and wall times are in ../../how-to/mutation-testing.md.

These lists are generated, and they go stale the moment the source or the tests move. Line numbers are only meaningful at the commit above.

Regenerate them with:

MUTATION_SESSION_DIR=.mutation-sessions task mutate <group>
python scripts/mutation_survivors.py out.md .mutation-sessions/*.sqlite

The weekly Mutation (scheduled) workflow produces the same files as build artifacts, so the current picture is always one workflow run away even when the copies here have aged.