Repository navigation
control.t27: whole-file mutation pass (sections 1-7) #7074
Description
Problem
specs/queen/control.t27 has never had a whole-file mutation pass.
specs/queen/actors.t27did: 304 mutants, 298 killed, and the 6 survivors were named as equivalent.- control.t27 section 7, the effects journal of control.t27: journal hand-off so a restarted agent does not repeat a paid call or a PR open #7006, had
tri mutate spec --fnon its 7 new functions. - The run limit of control.t27: bound how often one task re-runs a paid model call across crashes #7059 (local commit
f8119a488) had only 22 hand mutants, because the lab had no tri build of this loop. - Sections 1-6 (leases, fence, placement, dispatch, cancel) were never run through the tool with the arithmetic and return-default mutants that tri mutate spec: arithmetic-operator and return-default mutants #7022 added.
A function whose asserts pass but whose guard can be flipped unseen is a false claim of coverage, as the gaps closed in #7001, #7003 and #7006 showed.
Expected
tri mutate spec --file specs/queen/control.t27 --jobs 8 --timeout 60is run on the lab against the head ofclaude/queen-actors-next2.- Every survivor is either:
- killed by a new assert (a test gap), or removed (a dead line), or
- named as one of the three equivalence shapes in the queen-actors-loop skill, one line per survivor.
- Constants are nudged by hand, since the tool does not mutate them.
- The counts are printed by the run, not typed.
Boundary
specs/queen/control.t27 only. No change to its contract except where a survivor shows a dead line.
Done when
The survivor list is in this issue and in the commit body, and every non-equivalent survivor has a killing assert.
Activity
gHashTag commentedon Oct 6, 2026
Whole-file pass done (lab, a master build of
triafter #7043,--jobs 8 --timeout 60, one locked foreground run per spec sha).Before (spec sha 2c92fe84): 173 mutants, 161 killed, 12 survived.
8 test gaps, closed in local commit 44501d567 on
claude/queen-actors-next2(asserts added to existing tests; no new test, no new function):kind == EV_TASK_CREATEDreaction_of(EV_TASK_CREATED) == R_DISPATCHkind == EV_EVIDENCEreaction_of(EV_EVIDENCE) == R_REVIEWkind == EV_REVIEWreaction_of(EV_REVIEW) == R_DISPATCHholder != meinrenew_landsrenew_lands(9, 7, false, fence, 2) == falsefence_after_cancelreturns 0fence == 5afterfence_after_cancel(4)fence + 1->fence - 1write_lands(fence, 5) == truefits>=->>fits(1, 1, 1024, 1024) == trueAE_RETIREfromAG_IDLEagent_next(AG_IDLE, AE_RETIRE) == AG_RETIREDEach one was applied by hand: killed on the new commit, and all 8 SURVIVED on the parent (negative control). An earlier version of this comment said three; only two had been run at that point, and the other six were run after.
After (spec sha 19c9ae4b): 173 mutants, 169 killed (all by
zig test), 4 survived. All four are equivalent, because both sides give the same value at equality:capacitymax_workers < cap-><=: assigns cap its own valuecapacityby_memory < cap-><=: samefree_slotsrunning >= cap->>: at equalitycap - running= 0offers_per_wakeeligible < slots-><=: returns the equal valueThe commit lands with #7059's commit (f8119a488) as one PR after #7060 merges.
Process note: the first launch was wrapped in a retry loop. An empty ssh answer was read as "not started", so up to three passes ran at once and reported 163/173 with 2 false kills. The numbers above come from a single run that held
flock -nafter apsbusy-check, with no retry.spec(queen): bound an effect's re-runs; close the mutation gaps in fi…