Description
gate-49 accepts a catch only for one of nine named domain exceptions:
TRACKED = [DoesNotExistException, MultipleObjectsReturnedException, NotFoundException,
PermissionException, ValidationException, ForbiddenException,
CustomValidationException, AppendOnlyException, ArchivalImmutableException]
catch (\Throwable $e) matches none of them — yet it catches all nine, plus everything else. So the broadest possible translation is reported as no translation at all.
Hit on ConductionNL/hermiq#162: SettingsController::index/create/load were flagged with no handling, I added catch (\Throwable) to each with a translated JSON response and a log, and the gate still reports "3 controller method(s) missing try/catch or @throws".
Why it matters beyond a false positive
The finding pushes authors toward the narrower fix. Catching only DoesNotExistException where the real risk is "the service blew up for a reason I have not enumerated" leaves the uncaught path — a framework 500 with a stack trace — exactly where it was. For index() that trace reaches a non-admin, since the method is #[NoAdminRequired].
Suggested
Treat catch (\Throwable) and catch (\Exception) as satisfying the gate. They are supersets of the tracked list; a gate that rejects the stronger guarantee teaches people to write the weaker one.
Worked around for now with the documented opt-out ([hydra-gate-controller-exception-translation exclude] …), which is the right escape hatch but should not be needed for the broadest correct handler.
Related: #203 — gate-55 and gate-60 encode two different icon registries for the same widget field, with no opt-out at all.
Activity
chore(gates): record the gate-49 opt-out where the gate actually read…
feat(hermiq): rail-based agent detail, fuller dashboard, and the CSS …
fix(gates): make the opt-outs reachable, stop rejecting \Throwable, e…
rubenvdlinde commentedon Aug 7, 2026
Fixed in #211 (merged 2026-08-07).
Each fix ships with a control proving the gate can still fail:
\Throwable/\Exception; no-catch and an unrelated class still do not_optout_textreads the PR body and the author's head SHA; verified that plaingit log -1does NOT see a tag written behind a merge commit while the helper doesNote for #204 specifically: the fix is inside the PINNED hydra-gates package, so it reaches a repo only when its
hydra-gates-refmoves to a tag containing it. Tag + consumer sweep to follow.