Repository navigation
Description
Summary
gate-61 listener-work-placement is structurally incapable of reporting a result on any --full / workflow_dispatch run. It reports NOT APPLICABLE, which does not count against coverage, so it is invisible in the COVERAGE: line as well. On openregister this hides 3 real findings.
This is the falsely-NOT-APPLICABLE dead-gate shape: the exemption reason names a scope condition that does not hold on the run that printed it.
Observed
openregister full-scope run 31488412842, Hydra Gates job 93770857036, gate package b8c7eade (@main). The same log contains both of these lines:
[hydra-gates] Base ref: n/a — --full requested, scanning the entire tree.
...
[gate-61] listener-work-placement: NOT APPLICABLE — the diff against 'origin/development'
put every post-event registration out of scope, so NONE were inspected.
There was no diff. The gate excluded itself by naming one.
Root cause (read in the source, not inferred)
bin/hydra-gates— on--full(SCOPE_DIFF=0) it builds the runner argv asset -- "${APP_DIR}"and only prepends--scope-to-diff --base "${BASE_REF}"whenSCOPE_DIFF=1. The runner never receives--base.scripts/run-hydra-gates.sh:204—BASE_REFtherefore keeps its hardcoded default"origin/development".scripts/run-hydra-gates.sh:7854— gate-61 callscheck_listener_placement.py . --base "${BASE_REF}"unconditionally. The in-file comment says this is deliberate ("this gate is always diff-scoped in both modes … Switching an unscoped run to--allwas tried and reverted here: the builder runs unscoped, so it would have surfaced the whole backlog as blocking findings on every build").- On a full-repo run of
development,origin/development...HEADis empty → every registration falls out of scope → the checker exits 3 (EXIT_EMPTY_SCOPE) → the runner calls_skip 61 … na.
The revert referenced in (3) was correct for the case it addressed. The defect is that it treats two different unscoped modes as one: the builder's implicit unscoped run, and an explicit whole-tree audit that a human asked for by pressing "Run workflow".
Positive control — the gate is blind, not clean
Same tree, same checker, same commit (openregister@fdb0d6ca8); only the scope flag differs:
$ python3 scripts/lib/check_listener_placement.py . --all
FAIL lib/Listener/SourceRecordChangeListener.php (ObjectCreatedEvent): ...
FAIL lib/Listener/SourceRecordChangeListener.php (ObjectUpdatedEvent): ...
FAIL lib/Listener/SourceRecordChangeListener.php (ObjectDeletedEvent): ...
checked 45 post-event registration(s) [all], 0 out of scope: 3 failure(s)
45 registrations exist. 0 were inspected on the run that reported NOT APPLICABLE.
Proposed fix (narrow — does not re-break the reverted case)
Distinguish unscoped by default from explicitly asked for a whole-tree audit:
bin/hydra-gates: when--fullis passed, forward a new flag (e.g.--full-scan) to the runner.run-hydra-gates.sh: parse it intoFULL_SCAN=1, and in the gate-61 block call the checker with--allinstead of--base "${BASE_REF}"whenFULL_SCAN=1.
The builder invokes run-hydra-gates.sh directly and would not pass the flag, so its unscoped run stays diff-scoped exactly as today. Only an explicit --full audit changes behaviour — which is the run type whose entire purpose is inherited debt.
Expect this to make full-scope runs redder across the fleet. That is the honest answer and it is what someone pressing "Run workflow" is asking for. Please land the false-positive fix in the companion issue first, or land both together, so the first fleet-wide gate-61 report is not ~1/3 noise.
Why this is filed rather than merged
I have merge authority on .github this session, but this change turns a permanently-green gate red in every repo at once, while other agents are mid-flight driving their repos to green, and it carries a known false-positive class. Coordinating that is worth more than landing it unilaterally.
Activity
test(gates): repo-shaped acceptance suite — scope matrix, exit-code s…
fix(gate-16): an EMPTY SCOPE is not a PASS (.github#361)
fix(gate-61): the scope-matrix invariant read the FILE scope, not whe…
feat(gates)!: scan the whole tree by default — ADR-020 superseded (#378)