You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
analog: pin the sar_dff_q mutants its tests do not notice (2 test gaps, 8 accepted) #12083
specs/analog/sar_dff_q.t27 (tests only, appended after the last test; no function body changes), its seal .trinity/seals/analog_AnalogSarDffQ.json, a new accepted-mutants file docs/analog/mutation-accepted-sar_dff_q.txt and a record docs/analog/mutation-sar_dff_q-2026-10-25.md. Part of epic #10486 (weak spots of the analog v0.1 release).
User Scenarios
Given master at b9d83f9 and a master-built t27c, when tri mutate spec --file specs/analog/sar_dff_q.t27 --jobs 4 runs (a shell cell, load < 40), then it prints 56 mutants, 45 killed, 6 survived, 4 hung, 1 unviable (measured 2026-10-25 01:53-01:59 +07, exit 0, checkout clean).
Given the 6 survivors, when each is classified, then 2 are test gaps (while (i < LG_R_OFF) in dq_scale: nothing reads the first rv word of a scaled table; while (it < iters) in dq_solve_k: nothing runs zero sweeps) and 4 are equivalent (w + dw >= VDD, w <= VDD as f64, and the two <= bar comparisons that differ only when a gap equals the 10 uV bar to the last bit).
Given the 4 hung mutants (a dropped loop step makes the loop endless, so no test can fail), then they are named in the accepted file with that reason.
Requirements
FR-001: the spec must gain tests whose expected values come from a second source (the rail definitions and the table words read from specs/analog/mos_psp103_data.t27, not copied from the lines under test) that go red on the dq_scale<= mutant and the dq_solve_k<= mutant.
FR-002: docs/analog/mutation-accepted-sar_dff_q.txt must name exactly the equivalent survivors (113, 120, 217, 218) and the hung dropped steps (138, 142, 181, 219), each with a note, and must not name a line whose mutant a test now kills.
FR-003: the spec must be re-sealed only with the master-built t27c, and tri coverage --strict must keep gaps at 0.
Claim (analog loop tick 11, session sad-kilby-2b6441, started 2026-10-25 about 02:30 +07): adding the two missing tests and the accepted file on branch lane/analog-dffq-mutants-12083.
Tick 11 (analog-night-2026-10-25-11): done and merged. PR #12085 is MERGED (read 2026-10-25 ~03:30 +07).
sar_dff_q: 56 mutants; before 45 killed / 6 survived / 4 hung / 1 unviable, after 47 / 4 / 4 / 1; the 8 mutants no test can or must kill are in docs/analog/mutation-accepted-sar_dff_q.txt, the gate passes. The record is docs/analog/mutation-sar_dff_q-2026-10-25.md.
The same recipe is applied to sar_switch_q in PR #12102 (issue #12086). Open and unclaimed: #12096 (sar_adc), #12097 (sar_kick), #12098 (sar_cdac), #12099 (sar_comparator_psp).
Follow-up to the review of PR #12085 (tick 11, analog-night-2026-10-25-11): a review bot said the "equivalent in practice" notes at sar_dff_q lines 217 and 218 claim more than was measured. It is right: those two flip-cmp mutants differ only at an exact f64 equality that no test builds and that is not proven unreachable.
Fix, wording only, keys of the accepted file unchanged: PR #12103 for sar_dff_q (the notes now say "not constructed"), and commit c1d746b on PR #12102 for sar_switch_q (line 273 "not constructed", the two line-361 mutants "unobservable in the returned QsHold"). The other review points were checked and need no change; the reasons are in the body of #12103. The gate was not re-run for either wording change.
Tick 16 (analog-night-2026-10-25-16, 2026-10-25 05:3x +07): closing by hand, because the merger bot closes only issues a PR names with Closes.
Done and merged: PR #12085 (the two missing tests and the accepted file for sar_dff_q, merged 2026-10-24T19:44:11Z) and PR #12103 (the wording of the two flip-cmp notes, merged 2026-10-24T22:13:06Z). Nothing in the Requirements of this issue is open. The follow-up review items of the same family continue in #12116 (census of the 17 specs without an accepted file) and #12117 (sar_kick and sar_adc review follow-ups).
Activity
gHashTag commentedon Oct 24, 2026
Claim (analog loop tick 11, session sad-kilby-2b6441, started 2026-10-25 about 02:30 +07): adding the two missing tests and the accepted file on branch lane/analog-dffq-mutants-12083.
test(analog): pin the two test-gap mutants of sar_dff_q and gate the …
gHashTag commentedon Oct 24, 2026
Tick 11 (analog-night-2026-10-25-11): done and merged. PR #12085 is MERGED (read 2026-10-25 ~03:30 +07).
sar_dff_q: 56 mutants; before 45 killed / 6 survived / 4 hung / 1 unviable, after 47 / 4 / 4 / 1; the 8 mutants no test can or must kill are in
docs/analog/mutation-accepted-sar_dff_q.txt, the gate passes. The record isdocs/analog/mutation-sar_dff_q-2026-10-25.md.The same recipe is applied to sar_switch_q in PR #12102 (issue #12086). Open and unclaimed: #12096 (sar_adc), #12097 (sar_kick), #12098 (sar_cdac), #12099 (sar_comparator_psp).
docs(skills): accepted-mutant notes say "not constructed" or "unobser…
gHashTag commentedon Oct 24, 2026
Follow-up to the review of PR #12085 (tick 11, analog-night-2026-10-25-11): a review bot said the "equivalent in practice" notes at sar_dff_q lines 217 and 218 claim more than was measured. It is right: those two flip-cmp mutants differ only at an exact f64 equality that no test builds and that is not proven unreachable.
Fix, wording only, keys of the accepted file unchanged: PR #12103 for sar_dff_q (the notes now say "not constructed"), and commit c1d746b on PR #12102 for sar_switch_q (line 273 "not constructed", the two line-361 mutants "unobservable in the returned QsHold"). The other review points were checked and need no change; the reasons are in the body of #12103. The gate was not re-run for either wording change.
docs(analog): sar_dff_q accepted notes at lines 217 and 218 say "not …
gHashTag commentedon Oct 24, 2026
Tick 16 (analog-night-2026-10-25-16, 2026-10-25 05:3x +07): closing by hand, because the merger bot closes only issues a PR names with
Closes.Done and merged: PR #12085 (the two missing tests and the accepted file for sar_dff_q, merged 2026-10-24T19:44:11Z) and PR #12103 (the wording of the two flip-cmp notes, merged 2026-10-24T22:13:06Z). Nothing in the Requirements of this issue is open. The follow-up review items of the same family continue in #12116 (census of the 17 specs without an accepted file) and #12117 (sar_kick and sar_adc review follow-ups).
docs(analog): Build B board receipts, PASS on every line, for ss ff f…
spec(crons): the lab.yml verdict step as t27 law (t27os#9) (#11894)