Track guessed as P from the test source. 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-asm#src:test/test_asm_emit_x64.pas red at 94492d162332 (auto-filed by twatch)

• Type: regression (auto-filed by Track T watcher, host seven). Untriaged.
• Found: 2026-08-30T12:18:06Z
• Test source: test/test_asm_emit_x64.pas tools/expect_same.sh

Repro

tools/testmgr.py --tier native --job 'test-asm#src:test/test_asm_emit_x64.pas' at 94492d162332c9dc40bc84b11d1ae8bc467d7c6c

Range

The named sha 94492d162332 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 94492d162332, last good 3fd296c6b38d, 2 commit(s) in range — the watcher narrows this by idle bisect; check tstate/TSTATE.md for the current range.

Log tail

pascal26:1040: error: undefined variable (LibcSyscallCallCount)
pascal26:1052: error: undefined variable (PxxDbgWants)
pascal26:1075: error: undefined variable (PxxDbgWants)
(tail)
pascal26:1040: error: undefined variable (LibcSyscallCallCount)
  in: compiler/asmtext.inc
  near:  FixCount  s0  LibcSyscallCallCount >>>  AsmTextLine  
pascal26:1052: error: undefined variable (PxxDbgWants)
  in: compiler/asmtext.inc
  near: AsmMemoPoison    if PxxDbgWants >>>  a.asmmemo  
pascal26:1075: error: undefined variable (PxxDbgWants)
  in: compiler/asmtext.inc
  near: AsmMemoReport  begin if not PxxDbgWants >>>  a.asmmemo  

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

Re-laned P -> A: the defect is an include-visibility problem in a core file (2026-08-30)

track: P was guessed from the test path (test/test_asm_emit_x64.pas), which the auto-filer marks as a guess. It is wrong. The failure is:

pascal26:1052: error: undefined variable (PxxDbgWants)
  in: compiler/asmtext.inc
pascal26:1075: error: undefined variable (PxxDbgWants)

PxxDbgWants is declared at compiler/defs.inc:5339 and called at compiler/asmtext.inc:1052 and :1075 ('a.asmmemo'). asmtext.inc is pulled at compiler.pas:113. Nothing about this is the Pascal frontend — it is core include structure, Track A.

Note the shape before assuming it is a missing forward: the same symbol is called from lexer.inc (:143, :162, :299, :2977) and dbg_filetable.inc (:130, :228) without complaint, so whatever makes it invisible is specific to how asmtext.inc is reached in this test's configuration rather than to the declaration being absent. Find out which before adding a forward — a forward added to silence a visibility difference hides the reason the configuration differs, and this file family has already produced three wrong mechanisms today from reading rather than measuring.

Surfaced by frankwasm while confirming an unrelated tstate RED was not its own. Still red; unclaimed.

FIXED — 2026-08-30, frankA. It is harness rot, not include visibility.

Compiler: self-host fixedpoint 0825ad27809f, converged after 1 round(s).

The re-laning to A was right; the mechanism was not

asmtext.inc is not reached differently in "this test's configuration" through any subtlety of include order. test/test_asm_emit_x64.pas is a standalone harness that supplies its own environment instead of including defs.inc at all — that is the whole design, and the whole difference.

The question the ticket told me to answer before adding a forward — why here and not from lexer.inc — has an exact, measured answer:

So lexer.inc "never complains" because its only standalone harness does include defs.inc. asmtext.inc's does not. Nothing is invisible; the declaration was never in scope in that configuration and never had been.

A forward declaration would have been the wrong fix, exactly as warned — it would have papered over a harness missing two pieces of the environment it deliberately owns.

What actually changed under it

The a.asmmemo memoisation (asmtext.inc:1040-1076) grew two references into defs.inc: LibcSyscallCallCount (the memo poisons an entry when emitting a line moves a counter) and PxxDbgWants (the PXXDBG channel selector, two sites). Both are legitimate; the harness simply never gained them.

Fix

Two mocks in the harness prelude, beside the ~15 it already has:

• LibcSyscallCallCount: Integer = 0 — nothing here lowers a libc syscall, so 0 is the honest value and the poison check correctly never fires.
• PxxDbgWants(...) = False — this harness asserts exact byte sequences, so a debug line on stdout would be noise inside the thing being measured.

Measured both directions, and across all five harnesses

before:  3 undefined-variable errors  (1040 LibcSyscallCallCount, 1052/1075 PxxDbgWants)
after:   0 errors, ALL X64 ASM EMIT TESTS PASSED

Then every sibling, run exactly as the test-asm recipe does (compile, then compare the last line against the expected banner):

x64    PASS      386    PASS      a64    PASS      arm32  PASS      rv32   PASS

386 and rv32 were never exposed: test_asm_emit_386.pas includes asmtext_386.inc, a different file. Checked rather than assumed, because "the sibling is green" is only evidence if the sibling could have been red.

This is a recurrence, and the Makefile already says so

The test-asm recipe carries this comment:

"They ran nowhere, and four of them had stopped COMPILING: the emitters grew Asm<arch>ProcessInlineLine after the harnesses were written, and the mock environment never gained the line pool it reads. That is the cost of an unwired test exactly — it does not fail, it rots, and it rots silently."

compiler/rel8.inc's header records the same failure a third time (test_asm_emit_rv32.pas rotting on an AIntToStr call), and is why that file was carved out to need five mocks instead of thirty.

The difference this time is that it did not rot silently — it went red. The harnesses are wired now, so the loop is closed and the cost is one prelude edit per new reference. Not folded into a shared prelude here; that is [[idea-a-fold-the-asm-emit-harness-mock-preludes-into-one]], which this is a fourth data point for.

Gate: no compiler source touched — a test-only change. make compiler/pascal26 converged (1 round, 0825ad27809f); all five asm-emit harnesses compile and print their banners.

• 2026-08-30 — resolved, commit d0a2f5a72.