Repository navigation
regression: compiler/optimizer.t27 and port/tools/gft_deep_demo.t27 stopped compiling on 2026-10-10 #8552
Description
Two specs that compiled and passed their tests on 2026-10-09 stopped compiling on 2026-10-10. Their seals still record the pass. tri seals drift --fix would reseal them as blocked, which writes the regression into the record as truth, so the re-mint in #8551 left them stale on purpose.
| spec | sealed on master | now (t27c test-report) |
last change |
|---|---|---|---|
specs/compiler/optimizer.t27 |
pass, failed 0 | spec.zig:21:5: error: duplicate enum member |
#8502 |
specs/port/tools/gft_deep_demo.t27 |
pass, failed 0 | spec.zig:608:33: error: expected ')' |
#8494 |
optimizer.t27 is also one of the 5 UNEXPECTED FAILURES of master's Corpus ratchet (parse-no-discard).
Acceptance criteria
t27c test-reportcompiles both specs and reportsFAIL 0.tri seals driftlists neither spec.- The Corpus ratchet no longer lists
specs/compiler/optimizer.t27.
Boundary
specs/compiler/optimizer.t27
specs/port/tools/gft_deep_demo.t27
.trinity/seals/
User Scenarios
- Given the specs
specs/compiler/optimizer.t27andspecs/port/tools/gft_deep_demo.t27compiled and passed tests on 2026-10-09, when runningt27c test-reporton 2026-10-10, then both specs fail to compile with errors in generatedspec.zig. - Given the Corpus ratchet on master lists
specs/compiler/optimizer.t27as an UNEXPECTED FAILURE underparse-no-discard, when the regression is fixed, then the Corpus ratchet no longer lists that spec.
Requirements
- FR-001: The compiler MUST accept
specs/compiler/optimizer.t27without emitting "duplicate enum member" in generated Zig. - FR-002: The compiler MUST accept
specs/port/tools/gft_deep_demo.t27without emitting "expected ')'" in generated Zig. - FR-003:
t27c test-reportMUST compile both specs and reportFAIL 0. - FR-004:
tri seals driftMUST list neither spec after the fix. - FR-005: The Corpus ratchet MUST no longer list
specs/compiler/optimizer.t27as an UNEXPECTED FAILURE.
Success Criteria
t27c test-reportcompilesspecs/compiler/optimizer.t27andspecs/port/tools/gft_deep_demo.t27and exits 0 withFAIL 0.tri seals driftoutputs no lines referencingspecs/compiler/optimizer.t27orspecs/port/tools/gft_deep_demo.t27.- Corpus ratchet run reports 0 UNEXPECTED FAILURES for
specs/compiler/optimizer.t27underparse-no-discard.
Activity
seals: the first frontier --reseal run on master, 41 specs (Closes #8551
gHashTag commentedon Oct 10, 2026
More evidence for
compiler/optimizer.t27: spec-guards' duplicate-declaration check now fails every PR with "NEW: specs/compiler/optimizer.t27 declares 2 name(s) twice and is not in the baseline" (seen on #8572). The zig errorduplicate enum memberis the same defect. Fixing the spec clears three gates at once: zig compile, the Corpus ratchet's parse-no-discard, and spec-guards.seal --save refuses to record a regression without --force (Closes #8576
gHashTag commentedon Oct 10, 2026
Update: #8564 ("Reseal the #8415 FROZEN_HASH wave") resealed
port/tools/gft_deep_demo.t27asblockedon master, so its regression is now in the record as the current state.compiler/optimizer.t27may be the same. Since #8577,t27c seal --saverefuses pass -> blocked without--force, so this cannot recur silently. The specs themselves still need their fixes (criteria above).