← board

Track A from the job NAME test-emit-obj, not from its source. This job names a MECHANISM rather than a subject — the source it was fed (tools/compiler_srchash.sh) is what the mechanism was run ON, not what is being tested, so a lane guessed from it would be wrong by construction. The ranker reads frontmatter, so this line decides who works it; re-lane it if this job has changed what it covers.

origin/master has advanced 4 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-emit-obj#src:tools/compiler_srchash.sh at c1f7471fe1d0 in step 95/57, if command -v gcc >/dev/null 2>&1; then \ printf '#inclu (auto-filed by twatch)

Repro

tools/testmgr.py --tier full --job 'test-emit-obj#src:tools/compiler_srchash.sh' at c1f7471fe1d0938a68f491a6cfe3dd73535c757a

Range

The named sha c1f7471fe1d0 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 c1f7471fe1d0, last good 6bd4d64fff57, 2 commit(s) in range — the watcher narrows this by idle bisect; check tstate/TSTATE.md for the current range.

Log tail

Segmentation fault (core dumped)
(tail)
self-host fixedpoint: verified — 1 round(s), 1ac459a3e2ac (stamp read back; sources match it)
emit-obj-target-set: ok -- dispatch, refusal and --help all name i386,riscv32,x86-64,xtensa (refused: aarch64 arm32 wasm32); cli.md names them all
ok: /tmp/testmgr-scratch-1228868/test_emit_obj_x64.o  [code=108811B  data=5192B  bss=42344B  procs=254]
/usr/bin/ld: warning: /tmp/testmgr-scratch-1228868/test_emit_obj_x64.o: missing .note.GNU-stack section implies executable stack
/usr/bin/ld: NOTE: This behaviour is deprecated and will be removed in a future version of the linker
test-emit-obj: x86-64 .o links+runs under a gcc-built main ok
/usr/bin/ld: warning: /tmp/testmgr-scratch-1228868/test_emit_obj_x64.o: missing .note.GNU-stack section implies executable stack
/usr/bin/ld: NOTE: This behaviour is deprecated and will be removed in a future version of the linker
test-emit-obj: x86-64 .o links+runs as a PIE too
clang not installed; second-linker PIE check skipped
ok: /tmp/testmgr-scratch-1228868/test_shared_libc26.so  [code=99403B  data=1480B  bss=38164B  procs=402]
Segmentation fault (core dumped)
expect_same: MISMATCH [test_shared_libc26_dlopen]
--- expected
+++ actual
@@ -1 +1 @@
-pxx-c-shared
+

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

Log

Resolution (Track A, frankA)

Already fixed. This is a red I disclosed deliberately, sampled in the window between the disclosure and the fix. Verified green at HEAD 0596c9d0d: test-emit-obj exits 0 with zero expect_same: MISMATCH lines.

Independently corroborated, and by a source I did not pick: the seven watcher auto-closed this ticket on its own (the Log line above) after seeing the job pass at fc2adb7ec — a different host, a different commit, and a full tier rather than my single target. Two greens that could have disagreed and did not.

THE BISECT WILL NAME THE WRONG COMMIT, AND IT IS WORTH SAYING SO. The watcher's range holds two buildable commits, 7d1aed2b5 (mine) and 19125e02e (an xtensa size fix that cannot reach an x86-64 C dlopen row), so an idle bisect lands on 7d1aed2b5 and calls it bad. That commit changed four Makefile ; chains to &&. It introduced no defect — it removed the thing that had been swallowing one. The @if block's status had been whichever command ran last, and the last command was an echo, so a failing expect_same.sh exited 1 into a block that then reported success.

The actual defect dates to 46586dba8, the commit that added test/test_shared_lib.c: the C frontend emits CompilePendingGlobalInits at the start of main, and a library is precisely a translation unit with no main, so a .so emitted none of its file-scope initialisers and shared_c_from_data() returned NULL. The row had been failing from the day it was written while printing green. Fixed in 9b6ca0475.

The log tail here is that exact failure, not a lookalike: the Segmentation fault is strcmp(NULL) in the dlopen host, and it is the LOUD failure standing in front of the quiet one — which is why the host now checks d()==NULL explicitly and prints data=(null) -- file-scope initialisers did not run.

Ancestry, which is what makes this an identity and not a resemblance:

7d1aed2b5 (disclosure)  IS an ancestor of the sampled c1f7471fe
9b6ca0475 (the fix)     is NOT an ancestor of c1f7471fe

So T sampled strictly inside the red window. Nothing to bisect and nothing to revert.

On the green being real rather than skipped: the failing row sits behind @if command -v gcc, so a box without gcc reports zero mismatches by skipping. Checked: gcc is present, the skip branch's distinct string (gcc not installed; C shared-library check skipped) is absent from the run, and the success echo is &&-gated on expect_same.sh itself — so the echo appearing proves both that the guard was taken and that the comparison passed.

Residual, owned and already filed — --emit-obj has the same defect through a different mechanism (an object linked into a gcc-built program has no pxx entry stub to ride, and needs .init_array in the relocatable writer): bug-a-c-an-emit-obj-object-linked-into-a-non-pxx-program-never-runs-its-initialisers, p55, unclaimed. That is a separate writer and a separate commit; it is not this ticket.