← board

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-c-conformance#shard1/6 red at 1b9b43e5b511 (auto-filed by twatch)

Repro

tools/testmgr.py --tier full --job 'test-c-conformance#shard1/6' at 1b9b43e5b511d53e9fbe55f3366e6ce9158ee0b9

Range

bad 1b9b43e5b511, last good 57b9b7148d32, 132 commit(s) in range — the watcher narrows this by idle bisect; check tstate/TSTATE.md for the current range.

Log tail

FAIL 00206.c — exit code 254 (want 0)
FAIL 00212.c — exit code 252 (want 0)
test-c-conformance: 35 pass, 2 fail, 0 skip (of 37)
test-c-conformance: FAILURES: 00206.c(exit=254) 00212.c(exit=252)

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

Root cause found and fixed — 2026-08-21 (agent-A)

Real, reproducible, and nothing to do with the commit it was blamed on (e96a698f1 / 1b9b43e5b511 are ticket edits): 00206.c and 00212.c both end main without a return.

C99 5.1.2.2.3 says reaching the closing brace of main is return 0. pxx returned whatever the uninitialised result slot happened to hold. Both programs printed EXACTLY the expected output and then exited 255 and 253 here — while the watcher recorded 254 and 252 for the same two tests. Two builds, two different wrong numbers, right output: that is stack garbage, and it is why the shard looked like it changed behaviour at a commit that changed no code.

Fixed by zeroing main's result slot in the prologue (cparser.inc), which leaves an explicit return n winning because it stores over it, and covers every way of reaching the brace — fall-through, out of a nested block, a goto to a trailing label.

Uncovered on the way: EmitZeroLocalSlotForTarget emitted nothing for a 4-byte slot on the 64-bit targets (4 div 8 = 0), which is exactly the size of a C int result. Both 64-bit arms now finish with a 4-byte store. Rounding up to 8 would have been worse than nothing — a 4-byte local is 4-aligned with the next local immediately above it.

Verified: shard 1/6 is 37 pass, 0 fail natively AND on aarch64, arm32, riscv32 and i386 under qemu. Regression guard added as test/c_main_no_return.c with a Makefile row that checks the EXIT STATUS (the only visible symptom): straight-through, out of a nested block, and via a goto to a trailing label, byte-identical to gcc -std=c99 and exiting 255 on the pre-fix compiler.