Context
From the first complete full mutation-testing run (ADR-007 shared workflow, run 29074771985, 2026-07-10, main @ 94a5d44). These survivors are from the shard that was lost in the July partial runs, so they are newly visible rather than newly introduced.
Surviving mutants
src/connection/output.rs:65 — replace ActiveOutput::clear_response with ()src/connection/state.rs:78 — replace ActorState::closed_sources -> usize with 0 / 1src/connection/state.rs:82 — replace ActorState::total_sources -> usize with 0 / 1src/connection/counter.rs:32 — replace current_count -> u64 with 0 / 1
Analysis
Every existing run-loop test ends the response stream as the last live source, so whether the exhausted stream is cleared or re-polled is indistinguishable. The cfg(test) snapshot accessors closed_sources/total_sources feed ActorStateSnapshot, but actor_state_tracks_sources (tests/connection.rs) asserts only the boolean flags, never the counters. current_count is a pub(super) duplicate of the tested active_connection_count, consumed only by an info! log line.
Suggested tests
- Run an actor whose response stream ends (or fails with
WireframeError::Protocol) while a push queue remains open; assert on_command_end fires exactly once and no duplicate stream-end frame is emitted (kills clear_response).
- Extend
actor_state_tracks_sources to assert snapshot.total_sources equals the constructed source count and that closed_sources increments by one per mark_closed() (kills all four accessor mutants).
Accepted
connection/counter.rs:32 feeds a log line from a process-global atomic; asserting it under parallel tests would be nondeterministic. Accepted — though deleting the wrapper in favour of active_connection_count() would remove the mutants entirely.
Context
From the first complete full mutation-testing run (ADR-007 shared workflow, run 29074771985, 2026-07-10,
main@ 94a5d44). These survivors are from the shard that was lost in the July partial runs, so they are newly visible rather than newly introduced.Surviving mutants
src/connection/output.rs:65—replace ActiveOutput::clear_response with ()src/connection/state.rs:78—replace ActorState::closed_sources -> usize with 0 / 1src/connection/state.rs:82—replace ActorState::total_sources -> usize with 0 / 1src/connection/counter.rs:32—replace current_count -> u64 with 0 / 1Analysis
Every existing run-loop test ends the response stream as the last live source, so whether the exhausted stream is cleared or re-polled is indistinguishable. The
cfg(test)snapshot accessorsclosed_sources/total_sourcesfeedActorStateSnapshot, butactor_state_tracks_sources(tests/connection.rs) asserts only the boolean flags, never the counters.current_countis apub(super)duplicate of the testedactive_connection_count, consumed only by aninfo!log line.Suggested tests
WireframeError::Protocol) while a push queue remains open; asserton_command_endfires exactly once and no duplicate stream-end frame is emitted (killsclear_response).actor_state_tracks_sourcesto assertsnapshot.total_sourcesequals the constructed source count and thatclosed_sourcesincrements by one permark_closed()(kills all four accessor mutants).Accepted
connection/counter.rs:32feeds a log line from a process-global atomic; asserting it under parallel tests would be nondeterministic. Accepted — though deleting the wrapper in favour ofactive_connection_count()would remove the mutants entirely.