Repository navigation
master Corpus Ratchet: 5 unexpected failures from 2026-10-10 merges (ingest.t27 apostrophe, a/b_c, optimizer, orchestrator_bench) #8562
Copy link
Copy link
Open
Labels
Description
Master's Corpus Ratchet lists 5 UNEXPECTED FAILURES, all introduced on 2026-10-10 by other merges. The types-classification failure in the same job is mine and is fixed by #8561.
| spec | phase | last change | diagnosis |
|---|---|---|---|
specs/a/b_c.t27, specs/a_b/c.t27 |
parse-no-discard | #8496 (t27b test blocks) | the new test blocks lose characters to the parser |
specs/compiler/optimizer.t27 |
parse-no-discard | #8502 | also stopped compiling (#8552) |
specs/memory/graph/ingest.t27 |
parse | #8493 (graph slice 4) | line 414 is a ; comment containing let's; the apostrophe opens a char literal, so "unterminated string literal opened at 414:34". The comment also says "let's assert something wrong to test the negative control", so a deliberately failing assert may have landed on master |
specs/port/trinity/src/tri/orchestrator_bench.t27 |
gen-verilog | port of orchestrator_bench | not diagnosed yet |
Acceptance criteria
t27c suite --corpus-only --ratcheton master reportsUNEXPECTED FAILURES: 0.specs/memory/graph/ingest.t27parses with 0 recovery events, and no assert in it is wrong on purpose.- Each fix lands with its own owner's spec change, not a ledger bless that hides it.
Boundary
specs/a/b_c.t27
specs/a_b/c.t27
specs/compiler/optimizer.t27
specs/memory/graph/ingest.t27
specs/port/trinity/src/tri/orchestrator_bench.t27
User Scenarios
- Given the Corpus Ratchet on master reports 5 unexpected failures from 2026-10-10 merges, when
t27c suite --corpus-only --ratchetruns, then it reportsUNEXPECTED FAILURES: 0. - Given
specs/memory/graph/ingest.t27contains a;comment with an apostrophe on line 414 that the parser treats as an unterminated char literal, when the spec is parsed, then it completes with 0 recovery events and no deliberately failing assert remains. - Given each of the five failing specs has a distinct root cause introduced by a different merge, when the fixes land, then each fix is applied by its owning spec change (not a ledger bless).
Requirements
- FR-001:
specs/a/b_c.t27andspecs/a_b/c.t27MUST parse without discarding characters in the new test blocks introduced by specs: one distinct real test for a/b_c and a_b_c, replacing the #8488 twins (Closes #8508) #8496. - FR-002:
specs/compiler/optimizer.t27MUST parse without discarding characters and MUST compile successfully (addressing Census the optimizer: 4 of 7 passes have no precondition, and dead_store_elim deletes calls #8502 and regression: compiler/optimizer.t27 and port/tools/gft_deep_demo.t27 stopped compiling on 2026-10-10 #8552). - FR-003:
specs/memory/graph/ingest.t27MUST parse with 0 recovery events; the apostrophe in the line-414 comment MUST not open a char literal, and any deliberately wrong assert MUST be corrected. - FR-004:
specs/port/trinity/src/tri/orchestrator_bench.t27MUST pass the gen-verilog phase. - FR-005: Each fix MUST be committed as its own spec change by the responsible owner; no fix MAY be hidden by a ledger bless.
Success Criteria
t27c suite --corpus-only --ratchetexits withexit 0and printsUNEXPECTED FAILURES: 0.t27c parse specs/memory/graph/ingest.t27exits withexit 0and reportsrecovery events: 0.specs/memory/graph/ingest.t27contains 0 asserts that are intentionally incorrect (verified by inspection).- Each of the five boundary files is modified in a separate commit by its owning author (5 commits total, verified by git log --oneline).
Reactions are currently unavailable
Activity
gen-zig: the dead-const pass no longer ends a block at an unindented …
gen-zig: the dead-const pass no longer ends a block at an unindented …
gHashTag commentedon Oct 10, 2026
Status, 21:35 +07: the ingest.t27 defect came from a bee's salvage commit on
queen-8425(merged as #8493, commit 0355769). It still turns emit-bitexact red on every PR that touches specs (seen on #8638 by another session). A bee branchqueen-8562holds a salvage commit, f9e974e (12:52Z), that rewrites ingest.t27 (+54 -84), but no PR has been published. Publishing that branch, after review, is the fastest way to green. I am not editing the file, to avoid a third copy.