myxce/.wolf/bugs/bug-1902.md

id	status	status_updated	status_note	component	severity	timestamp	last_seen	occurrences	files	tags	related	fix_commits	repro_attempts
bug-1902	fixed	2026-10-02		harness	noise	2026-10-01T11:22:43Z	2026-10-02T04:14:22Z	2	src/harness/testsuite/test_test_cli.c	harness test flaky pty		daeb9041	0/5 reruns, 2026-10-01 (4 targeted meson reruns + reviewer full run)

bug-1902: harness pty test test_source_missing_file_rejected_with_error timed out once after quit

Description

Observed: during task 4.3 of spec job-array-one-based-index, one run of the harness suite under ASAN failed in the pty CLI test:

test_test_cli.c: test_source_missing_file_rejected_with_error: FAIL: xce_test did not exit within the harness's safety timeout after 'quit'

Expected: after quit, xce_test exits within the test's safety timeout and the test passes, as it does in every other run. Impact: noise: an intermittent false failure of the harness suite; no product impact observed. Found by: test (harness suite run during task 4.3, 2026-10-01)

Environment

• OS: no /etc/os-release; Docker container; NixOS kernel; kernel Linux 6.18.33 x86_64
• libc: glibc 2.42-47 (interpreter /nix/store/wb6rhpznjfczwlwx23zmdrrw74bayxw4-glibc-2.42-47/lib/ld-linux-x86-64.so.2)
• Toolchain: gcc (GCC) 15.2.0, meson 1.9.1
• Build: asan (src/build.jai)
• Code: scheduler-framework @ 1812a80d-dirty (task 4.3 in progress)
• Deployment: none (harness_unit_test, pty-driven xce_test, no controller)

Reproduction

1. Build the harness (bash src/tools/xce-build.sh harness asan build.jai --jobs 8 --incremental).
2. Run the pty test binary 16 at a time to load the machine (32 cores here): for i in $(seq 16); do ASAN_OPTIONS=detect_leaks=0 src/build.jai/harness/harness_unit_test > run$i.log 2>&1 & done; wait, three rounds.
3. Some runs fail a different test_test_cli.c test each time with xce_test did not exit within the harness's safety timeout after 'quit'.

• Frequency: 12/48 runs under 16-way parallel load, 2026-10-01 (1/6 and 1/127-test full suite runs earlier, unloaded reruns pass)
• Evidence: not kept (reproduces on demand with the steps above)

Root cause

Not an xce_test hang. clitest_wait_for_exit() and clitest_wait_for_text() (src/harness/testsuite/test_test_cli.c) did not measure time: they added CLITEST_POLL_SLICE_MS (100 ms) per loop iteration. After quit, xce_test closes its pty before it becomes reapable, so poll() returns at once on every call (hang-up, nothing to read) and the loop spun: the 5000 ms budget ran out within milliseconds while the ASAN child was still tearing down, which takes longer under load. Measured with a temporary wall-clock probe over 2287 exits under load: real time from quit to reap at most 6 ms, while the counter reached up to 58800 ms (9 exits would have tripped the 5 s limit).

Fix

Both waits use a CLOCK_MONOTONIC deadline (clitest_now_ms()), and clitest_pump_once() naps 10 ms when poll() wakes with nothing to read, so a hung-up pty no longer spins. Test-harness only; xce_test is unchanged.

Verification

• Test: the load reproduction above is the RED (12/48 runs failing); GREEN after the fix: 0/48 and 0/48 in two further rounds of the same 16-way parallel load
• Suites: harness meson suite 7/7 OK (asan, src/build.jai)
• Live: n/a (test-harness timing; no product code changed)

References

• Related: none