← board

Track guessed as P from the FAILING STEP — line 1 of 2, ./compiler/pascal26 test/test_nested_class_type_scoping.pas /tmp/test_nested_cls_type26, which names test/test_nested_class_type_scoping.pas. Not from the job's name or its src: those describe what the job is ABOUT, and this job's recipe spans 2 source file(s). The ranker reads frontmatter, so this line — not the body — decides who works it; correct it if the guess is wrong.

origin/master has advanced 2 commit(s) since this sha. Re-verify at current HEAD before acting — the callback is tagged to the sha that was tested, which may no longer be the state of the tree.

regression: test-core#src:test/test_nested_class_type_scoping.pas at d81b90a991e9 in step 1/2, ./compiler/pascal26 test/test_nested_class_type_scoping.pas /tmp/test_nested_cls_type26 (auto-filed by twatch)

Repro

tools/testmgr.py --tier native --job 'test-core#src:test/test_nested_class_type_scoping.pas' at d81b90a991e9eb2266c31c9d2f3889173c7ab8df

Range

bad d81b90a991e9, last good baad7e842cd8, 1 commit(s) in range — the watcher narrows this by idle bisect; check tstate/TSTATE.md for the current range.

Log tail

pascal26:104: error: "w": no such member on this record/class
pascal26:104: error: "v": no such member on this record/class
pascal26:105: error: "w": no such member on this record/class
pascal26:106: error: "v": no such member on this record/class
(tail)
pascal26:104: error: "w": no such member on this record/class
  near: tinner . Create ; b . >>> w := 22 
pascal26:104: error: "v": no such member on this record/class
  near: w := 22 ; b . >>> v := 33 
pascal26:105: error: "w": no such member on this record/class
  near: Chk ( 'other-field-w' , b . >>> w , 22 
pascal26:106: error: "v": no such member on this record/class
  near: Chk ( 'other-field-v' , b . >>> v , 33

Stub ticket: signal only. Track T agent (face 2) enriches or a dev track takes it from the repro line.

Verified dead at HEAD (2026-09-08, frankS)

Cause and fix, so the range does not have to be re-derived: d81b90a99 put an EatQualifiedTypePrefix ahead of five FindArrayType(CurTok.SVal) probes, which for a qualified spelling read the OWNER token. ParseTypeKindInner carries its own copy of that strip, and that copy does one thing the shared helper does not: when the member is a nested CLASS or RECORD it REWRITES the name to the registered row. Stripping ahead of it consumed the qualifier that arm reads, the arm never fired, and the bare name bound to whichever class registered first -- which is exactly the defect this test was written for, reintroduced through a different door.

5ea212e36 adds EatQualifiedArrayTypePrefix: it decides from the token stream WITHOUT consuming and delegates only when the member names a row in ArrType*. Not a narrowing for safety -- FindNestedType answers -1 for an array member, so ParseTypeKindInner would strip and rewrite nothing there. Same QualTypeOwnerCi, same final token. The guard is "strip exactly where the two copies agree".

Before and after both measured in this checkout. At d81b90a99 my own full tier died on this source with four rows -- "w": no such member on this record/class and the same for "v", at lines 104-106. At 5ea212e36, compiler 29e343715a4a: total ok 9 / 9, zero FAIL rows.

test/test_a_qualified_nested_array_type_in_a_declaration.pas now carries the guard's own positive control -- two owner classes each declaring a nested class named TIn, with the fields proving which row was picked -- so the next person to widen the strip fails on the row that names the reason rather than here.

Log