← board

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 12 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-c-conformance-aarch64#shard0/6 red at b695bcb4b192 (auto-filed by twatch)

Repro

tools/testmgr.py --tier full --job 'test-c-conformance-aarch64#shard0/6' at b695bcb4b19264847cedc6c01678d4e298d14cfb

Range

The named sha b695bcb4b192 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 b695bcb4b192, last good 4d3f8a4eac00, 7 commit(s) in range — the watcher narrows this by idle bisect; check tstate/TSTATE.md for the current range.

Log tail

self-host fixedpoint: verified — 1 round(s), 82583d52f672 --shard 0/6
FAIL 00175.c — output mismatch:
    --- library_candidates/c-testsuite/tests/single-exec/00175.c.expected	2026-08-29 16:16:00.000000000 +0000
    +++ /tmp/pxx_c_conformance.3650240/out.txt	2026-08-30 14:03:15.227717970 +0000
    @@ -4,9 +4,9 @@
     int: 97
     int: 98
     int: 99
    -float: 97.000000
    -float: 98.000000
test-c-conformance-aarch64: 36 pass, 1 fail, 0 skip (of 37)
test-c-conformance-aarch64: FAILURES: 00175.c(output)

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

CAUSE IDENTIFIED AND FIXED — but DO NOT CLOSE on that alone

The cause is fixed; this job is not yet observed green. Those are different claims and the distinction is the ticket's, not a formality.

frankA found one expression behind both symptoms — the >8-param refusal and the conformance output mismatch:

if ProcExternal[procIdx] or ProcCdecl[procIdx] then        { the bug }
if ProcExternal[procIdx] or (ProcCdecl[procIdx] and (not CProgramMode)) then   { the fix }

The C frontend marks every C function ProcCdecl, and a C-defined function is not ProcExternal — so the aarch64/arm32 direct-call sites dragged every function of every C program onto the C-ABI marshalling path, where cparser.inc's own prologue spill is positional on exactly those two targets. A 9-param C function hits the C-ABI path's arg limit (the refusal); mix(int,double,int,double) makes an AAPCS call into a positional prologue and prints garbage (the mismatch). x86-64 was never affected because cparser's x86-64 spill really is SysV — the two targets whose halves disagree are exactly the two that broke.

Fixed forward, not reverted: the prologue arms are correct and tested, only the call-site predicate was wrong.

Why this ticket stays open

The lua tree and the c-testsuite corpus are both ABSENT from this checkout. frankA reproduced the mechanism from synthetic C — a new ccross_cdecl_cmode.c that fails to build at 83a767151ffa with both reported error messages verbatim, verified by stashing the fix and rebuilding at that sha rather than by reasoning — and says so itself:

My evidence is that the cause is fixed — not that those five jobs are green.

So: Track T re-runs these jobs against current HEAD and that verdict closes them. A fixed cause plus an unrun job is exactly the shape that produced today's other findings; do not substitute one for the other.

Log