Skip to content

ci(quality): switch the Hydra gates ON for the first time, pinned to v1.3.0 - #72

Merged
rubenvdlinde merged 1 commit into
developmentfrom
ci/hydra-gates-enable
Aug 5, 2026
Merged

rubenvdlinde merged 1 commit into
developmentfrom
ci/hydra-gates-enable

Conversation

@rubenvdlinde

Copy link
Copy Markdown
Contributor

One line: hydra-gates-ref: v1.0.1 -> v1.3.0.

This repo's CI is red right now, and it is not this repo's fault

At 2026-08-05T13:52Z, ConductionNL/.github commit f7eaf2a (#164) flipped hydra-gates-require-full-coverage to default true in the shared quality.yml. Every repo references that workflow at @main, so the flip went live everywhere at once. But the flag requires a gate to declare itself not-applicable — and that vocabulary ships in the pinned package:

_skip calls
v1.0.1 (this repo's pin) 0
v1.3.0 36

A pin with no _skip calls cannot say "not applicable". So every gate without subject matter here — 4 (composer-audit, correctly diff-scoped out per ADR-020), 24 (integration-parity), 33 (axe-core, whose producer enable-axe is deliberately off) — became DID NOT RUN and failed the job.

The red is fleet-wide, dated, and caused by a shared-workflow flag meeting a pinned script that predates the vocabulary it needs. It is not a code defect and not anything a PR author did. The control that settles it: decidesk#407 is PHPMD-only with zero npm changes and shows the identical failure, and it passed on pre-13:52Z base runs — the base is a different vintage of the shared workflow, not of the code.

The pin was also four fixes stale

v1.0.1 is f4d9756 (2026-08-03). Every Hydra Gates run this repo has ever made executed a script in which:

PR what it fixed in v1.0.1?
#147 16 gates reported PASS when their helper never ran ❌
#148 gate-33 had no axe report to read, and never said so ❌
#149 gates 6 and 7 reported PASS on an empty scope ❌
#162 gate-5 reported a resolution failure as a security finding ❌

A gate that reports PASS without running emits a tick identical to a real one, which is why nothing in this repo's history shows the difference. Tracked in .github#159.

Measured before opening this PR

Diff-scoped against origin/development, exactly as CI scopes it. Each run in its own mount namespace with a private tmpfs — the runner writes ~50 detail logs to hardcoded /tmp/hydra-gate-<name>.log paths and reads its verdicts back out of them, so two concurrent runs on one host silently corrupt each other's counts (.github#158 item 6). Another agent was running the gates unisolated on the same host during this work, so that is not hypothetical.

v1.0.1   exit 98   FAIL — "GATES THAT DID NOT RUN: 24 33"
v1.3.0   exit 0    PASS — ALL 58 APPLICABLE GATES PASSED, and all 58 ran
                          NOT APPLICABLE: 4 6 7 24 33 — each with a reason

Verified not merely that the job is green but that gates 4, 24 and 33 each name themselves not-applicable and say why. A green that came from somewhere else would not be the fix landing.

Independently confirmed end-to-end by doriath#160, which changed this same single line and nothing else: Hydra Gates went failure -> success.

v1.3.0 is f7eaf2a = .github@main at the time it was cut.

Refs ConductionNL/.github#159

…v1.3.0

This repo left `enable-hydra-gates` at its default `false` on purpose,
with a stated precondition written into the workflow:

  gate-7 (no-admin-idor) currently false-positives on ordinary responder
  delegation — that is being fixed separately in ConductionNL/.github#149.
  Revisit once #149 lands.

#149 has landed (`fc70ea6`) and is in the pin below, so the deferral is
spent. Enabling it now.

The pin moves in the SAME commit, and that is load-bearing. The shared
quality.yml now defaults `hydra-gates-require-full-coverage` to true
(#164), and that flag requires a gate to DECLARE itself not-applicable.
v1.0.1 contains ZERO `_skip` calls; v1.3.0 has 36. Enabling the gates on
the fleet's old pin would land this repo red on arrival for gates it has
no subject matter for — precisely the "red for a reason that is not
hrmq's" the original comment was protecting against.

Measured before enabling, in a private mount namespace with a private
tmpfs, diff-scoped exactly as CI scopes it:

  exit 0 — ALL 58 APPLICABLE GATES PASSED, and all 58 of them ran.
  NOT APPLICABLE: 4 6 7 24 33, each naming itself and its reason.

Whole-tree debt is 10 gates and is written into the workflow comment
rather than discovered later. Two findings triaged by hand:

  gate-8, 13 findings, ALL FALSE POSITIVE. The gate flags
  `catch (\Throwable) { return null; }` in authorize* methods without
  reading what the CALLER does with the null. Every caller here is
  fail-CLOSED (`if ($x === null) return 404`). The null is deliberate:
  "does not exist" and "RBAC denied" collapse to one 404 so the endpoint
  is not an existence oracle. That is the OPPOSITE of the decidesk#45
  defect gate-8 targets. Reported upstream, not suppressed.

  gate-4 is REAL: this repo ships composer.json but no composer.lock, so
  `composer audit --locked` cannot run at all and its dependencies have
  never been CVE-audited. Filed separately.

No waiver, no continue-on-error, no require-full-coverage override.

Refs ConductionNL/.github#159
@github-actions

github-actions Bot commented Aug 5, 2026

Copy link
Copy Markdown
Contributor

Quality Report — ConductionNL/hrmq @ 31a3138

Check PHP Vue Security License Tests
lint ✅
phpcs ✅
phpmd ✅
psalm ✅
phpstan ✅
phpmetrics ✅
eslint ✅
stylelint ✅
build ✅
check-manifest ✅
check-widget-keys ✅
composer ✅ ✅ 99/99
npm ✅ ✅ 813/813
PHPUnit ✅
Newman ⏭️
Playwright ✅
Hydra gates ✅

Quality workflow — 2026-08-05 18:18 UTC

Download the full PDF report from the workflow artifacts.

@rubenvdlinde
rubenvdlinde merged commit 56ff5cb into development Aug 5, 2026
35 checks passed
@rubenvdlinde
rubenvdlinde deleted the ci/hydra-gates-enable branch August 14, 2026 09:26
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant