You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
ci(integration-tests): diagnostic step round 2 — tar probe first, set +e
Round 1 (91772ad) had `set -e` inherited from `bash -e {0}` shell
invocation. The diagnostic script halted at the first failing probe
(`head -c 50` on a mode-600 file), so the actual tar probe — the
operation that docker cp uses internally and is the proximate cause
of fixture teardown failures — never ran.
QA SubAgent caught the gap before I committed a fix on incomplete
evidence (the "setfacl will work" hypothesis was extrapolated from
"file is mode 600 + runner can't read" without measuring that tar
actually fails the same way).
Changes in round 2:
- `set +e` at the top so each probe runs independently
- Tar probe MOVED TO PROBE 1 (highest diagnostic value, runs even
if downstream probes fail)
- Added `getfacl` + `mount` to confirm filesystem supports ACLs
- Added `which setfacl` to confirm the tool is installed
- `exit 0` at end so this step never gates downstream steps
Goal: this single CI run produces enough evidence to either
(a) confirm tar-fails-on-mode-600 hypothesis, in which case
setfacl with default-inheritance is the right fix, OR
(b) reveal a different mechanism entirely.
No prediction. The next commit waits on data.
0 commit comments