x86 prod gate: the console-prompt count assertion aborts via set -e instead of failing through the verdict trap (PROMPT_BEFORE=0 under host contention)

[ryanbreen]

Summary

The x86 production-profile boot test script (docker/qemu/run-x86-prod-profile-boot-test.sh) has a bare test assertion that checks the console prompt count, which aborts via set -e instead of routing through the verdict trap. This is the #802 class of pre-empting exit that bypasses proper failure handling and reporting.

What the Assertion Checks

Lines ~1248-1249:

test "$PROMPT_BEFORE" -eq 1
test "$PROMPT_AFTER" -eq 2

These assertions validate the liveness signal: a steady-state prompt is printed once at start (PROMPT_BEFORE=1), and exactly one additional prompt is earned by the stimulus (PROMPT_AFTER=2). This proves the kernel is executing and responding to UART input.

Observed Failure

During PR #883 landing on beast (attempt 1): the script aborted with set -e at the PROMPT_BEFORE assertion when PROMPT_BEFORE=0 and PROMPT_AFTER=1, indicating the steady-state prompt was never printed (typical under host contention). The abort bypassed the verdict trap, leaving no proper error reporting, GATE_BOOT_FACTS record, or ended_by marker.

Why This Is #802 Class

PR #805 fixed a similar issue: AF_UNIX preflights in this same script were aborting via bare exit before the ERR trap was installed, pre-empting the verdict path. PR #818 widened that fix to six sibling gate scripts. The PROMPT assertions carry the same pre-empting shape — they run after the trap is armed but under set -e, so a failure kills the script without invoking report_gate_failure.

The Ask

1. Route through the verdict trap: Convert the bare test assertions to a check-and-false idiom that routes failures through report_gate_failure, similar to PR gates: every gate script with a verdict trap fails through it - six sibling gates repaired, 66 pre-empting exits removed, the ratchet widened by shape (the #802 class) #818's pattern (e.g., use || groups and false).
2. Record failure facts: Ensure the error captures PROMPT_BEFORE and PROMPT_AFTER values in the diagnosis output and/or a distinct FAIL text for this class of failure.
3. Extend the ratchet: Update tests/teardown_structure.rs to include run-x86-prod-profile-boot-test.sh in the verdict-trap shape census (similar to the test at line ~6815 for other x86 gates), verifying that:
   • The script installs its ERR trap before any assertion that could fail
   • All failures route through the trap (no bare exits or set -e aborts pre-empting verdict)
   • A verdict is computed and explicitly tested

Related

#802 #805 #818 #826 #865 #871

[ryanbreen]

Landed in #895 (merge commit d18c46f53cdba941fc57e6c108dd61af9091abaf on main).

Fix: converted the bare test "$PROMPT_BEFORE" -eq 1 / test "$PROMPT_AFTER" -eq 2 assertions in
docker/qemu/run-x86-prod-profile-boot-test.sh to the ||-plus-false
idiom so a liveness failure now routes through the script's ERR trap into
report_gate_failure (distinct ended_by=prompt_absent /
ended_by=prompt_stall facts) instead of aborting via set -e.
report_gate_failure's preservation block also now copies
gate_boot_facts.txt into $failure_dir. Extended
tests/teardown_structure.rs with a verdict-trap shape census for this
gate script, matching the existing pattern for the sibling x86 gates.

Evidence (paths below are under
docs/planning/green-program/gates/884-x86-prod-prompt-verdict/ unless
otherwise noted, on main):

• Forced-failure repro (BREENIX_X86_PROD_FORCE_PROMPT_ABSENT=1)
  reproducing the exact ended_by=prompt_absent failure the fix targets:
  docs/planning/green-program/gates/884-x86-prod-prompt-verdict/force-prompt-absent-proof.txt,
  and the F6 fix-round re-run:
  docs/planning/green-program/gates/884-x86-prod-prompt-verdict/f6-fix-proof-gate-boot-facts-preserved.txt.
• Minimal isolated repro of the ERR-trap-fires-not-set -e-dies shape:
  minimal-isolated-repro.sh / minimal-isolated-repro-output.txt.
• Pre-landing x86 prod gate passes at this branch's own HEAD:
  x86-prod-gate-head-pass.txt, x86-prod-gate-fixround-pass.txt, and
  boot tests x86-boot-tests-head-pass.txt.
• Landing re-smoke at the merged head (main a9d4bd3e merged in as
  20b00edaa6a846b4387ad03b21a55b5b8bb27e5c, 0 conflicts, 0 kernel/
  path collisions): 50/50 tests/*_structure.rs suites locally (0
  failures), x86 prod gate x1 on beast PASS
  (x86-prod-gate-merged-head-landing.txt, PROMPT_BEFORE=1/
  PROMPT_AFTER=2 through the verdict trap), run-x86-boot-tests.sh 1 on
  beast PASS (x86-boot-tests-merged-head-landing.txt) -- see the round
  doc's "Landing re-smoke" section.

Not claimed: a second independent run of either beast gate beyond the x1
each step asked for; an aarch64 re-smoke (this branch and its merge touch
no aarch64-only path).