← board

Track guessed as P from the FAILING STEP — line 1 of 2, ./compiler/pascal26 test/test_inline_generic_specialization.pas /tmp/test_inline_genspec26, which names test/test_inline_generic_specialization.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_inline_generic_specialization.pas at b2a41d5f4fb9 in step 1/2, ./compiler/pascal26 test/test_inline_generic_specialization.pas /tmp/test_inline_genspec26 (auto-filed by twatch)

Repro

tools/testmgr.py --tier native --job 'test-core#src:test/test_inline_generic_specialization.pas' at b2a41d5f4fb93afe09b23d1b7ee3999eaa906523

Range

The named sha b2a41d5f4fb9 CANNOT be the cause — it touches no buildable file (docs / tickets / tstate only). It is the sha that was TESTED, i.e. the upper bound of an untested range; the cause is somewhere below it.

bad b2a41d5f4fb9, last good 29c40526f145, 9 commit(s) in range — the watcher narrows this by idle bisect; check tstate/TSTATE.md for the current range.

Log tail

pascal26:36: error: expected 'begin' before 'MaxOf_Char'
(tail)
pascal26:36: error: expected 'begin' before 'MaxOf_Char'
  near: b := tmp ; end ; >>> MaxOf_Char as MaxCh

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

Resolution — 2026-09-06, frankD. Cause: eb2447470, and the ( it removed was ACCIDENTAL COVER.

eb2447470 correctly stopped the call-site predicate in SpecializeInlineGenericFuncUses from requiring a trailing ( on the keyword surface: specialize DoTest<TTest> is a parameterless call and the ( was a proxy for "this is not a comparison chain", which the keyword already settles. That reasoning is sound and the fix stays.

What it could not see is that the ( was doing a second, undocumented job. specialize Max<Integer> as MaxIntF; is the specialization-ALIAS DECLARATION, and it differs from a parameterless call by exactly one token — as where a call has (. With the ( no longer required, the call-site sweep matched the declaration, rewrote the run to Max_Integer and consumed the keyword, leaving Max_Integer as MaxIntF; where a declaration had been. The reader then said expected 'begin' before 'Max_Integer' — a complaint about the declaration section ending, three phases after the rewrite that caused it.

Nothing recorded that second job. The guard's own comment described the comparison chain alone, and the isDecl check beside it covers only the OTHER declaration shape (a function/procedure header) — it had exactly the same character and was documented, which is the difference.

Fix: the no-parens allowance now excludes a following as. This restores the pre-eb2447470 behaviour for the alias form exactly — it required ( then too, so a genuine specialize F<T> as TFoo cast of a parameterless call did not work before and does not work now. Unchanged, not newly refused.

Why the author's gate did not catch it: gate.sh quick does not run test-core, and the two rows that cover the alias form — this one and test_generic_func — live there. The fixture eb2447470 added is an fpc differential and cannot carry the alias form, because fpc rejects it (Syntax error, "BEGIN" expected but "identifier SPECIALIZE" found) — it is a pxx extension. So the two spellings the predicate must tell apart are necessarily in two files. Each now names the other, and the parameterless fixture says explicitly to run both.

Verified: both rows GREEN under testmgr --tier native; eb2447470's own fixture still GREEN; tgenfunc12.pp, the corpus row it un-skipped, still compiles and runs exit 0; tgenfunc19.pp still stops at :32 with undefined variable (TTest2), the wall its corrected skip reason names — unchanged.