ci: let Dependabot scan the tend-setup and build-js composite actions - #6177
ci: let Dependabot scan the tend-setup and build-js composite actions#6177prql-bot wants to merge 1 commit into
Conversation
prql-bot
left a comment
There was a problem hiding this comment.
Diff is correct. Verified independently:
- The new list is exactly
ls .github/actions/(six entries, sorted), and both newly-listed actions are live —build-jsis consumed byrelease.yaml,tend-setupby all eighttend-*workflows. go-task/setup-task@v2withrepo-token+version: 3.52.0is already running inbuild-web.yaml,test-php.yamlandtest-prqlc-c.yamlsince #5891, so the input block genuinely needs no change.build-js's three refs are at current majors (setup-node@v7matches the #6080 bump), so listing it is a no-op today, as the description says.- No conflict with #6144 — that PR only appends to the final
run:block; this one touches line 15.
One trade-off the description doesn't state, worth a maintainer's eye before merge: directories: fans a shared-action bump out into one PR per listed directory, not one PR overall. actions/upload-artifact 5→7 landed as five separate PRs across the five directories listed at the time — #5889, #5892, #5893, #5894, #5895. Going from five listed directories to seven raises that ceiling accordingly, and baptiste0928/cargo-install and Swatinem/rust-cache are now each referenced from several listed directories, so a major bump on either would fan out the same way.
That cost is clearly worth paying — the alternative is what this PR is fixing, tend-setup sitting on a stale ref for three months with no watcher — and the fan-out is a pre-existing property of the config rather than something this change introduces. Flagging it only because "two more Dependabot PRs per shared-action major bump" is the kind of thing that's better decided now than discovered later. If it becomes noisy, a groups: block on this ecosystem entry is the lever to reach for.
Self-authored, so no approval from me — this needs a human.
For the
github-actionsecosystem, Dependabot'sdirectory: "/"only covers.github/workflows/. Composite actions under.github/actions/are scanned only if they are listed explicitly — which is why four of them already are. Two are not:.github/actions/tend-setupand.github/actions/build-js. Theiruses:refs have never been scanned.That gap has already cost one update. Dependabot's #5891 (
chore: bump go-task/setup-task from 1 to 2, merged 2026-05-14) touched exactly three files —build-web.yaml,test-php.yaml,test-prqlc-c.yaml— all of them under/..github/actions/tend-setup/action.yamluses the same action and was left on@v1, where it has sat for nearly three months whilev2.0.0,v2.1.0andv2.2.0shipped. Every one of the eighttend-*workflows runs through that action, so it is the most-executeduses:ref in the repo and the one that had no watcher.This PR adds both directories to the list (sorted, so
ls .github/actions/is a checkable invariant, with a comment recording why/isn't enough) and applies the one bump the gap swallowed.On the
v1→v2bump specifically: the only breaking change inv2.0.0is the runtime moving from Node 20 to Node 24. The input surface atv2is a superset ofv1—version,repo-tokenandmax-retriesare all still there, withchecksumadded inv2.2.0— so the existingrepo-token/version: 3.52.0block needs no change. The three workflow files bumped by #5891 have been running@v2with the identicalversion: 3.52.0input since May, which is the strongest available evidence that the bump is a no-op here.build-js's three refs (baptiste0928/cargo-install@v3,actions/setup-node@v7,Swatinem/rust-cache@v2) are all at their current majors, so listing it changes nothing today — it just stops the same silent drift from starting there.The cost, stated plainly:
directories:fans a shared-action bump out into one PR per listed directory, not one PR overall —actions/upload-artifact5→7 landed as five separate PRs (#5889, #5892, #5893, #5894, #5895). Going from five listed directories to seven raises that ceiling, and bothbaptiste0928/cargo-installandSwatinem/rust-cacheare now referenced from several listed directories, so a major bump on either fans out the same way. That seems clearly worth paying against three months of an unwatched stale ref, and the fan-out is a pre-existing property of this config rather than something this change introduces — but it's a maintainer's call, and if it turns noisy agroups:block on this ecosystem entry is the lever.How this surfaced, and scope notes
Found during the daily
review-runssweep (31369430836). The symptom was half-surfaced a run earlier: the 08-09tend-weeklysession (31306003041) noticed the@v1/@v2split while doing itssetup-taskversion-pin task and correctly left it alone as out of scope — "the major-version drift may be unintentional". It reported the drift but not the cause; the cause is this config gap.The commented-out macOS job in
tests.yamlalso referencesgo-task/setup-task@v1(around line 297), but it is inside a fully commented-out block, so it's dead text — left untouched rather than adding diff noise.No overlap with the other open PR touching this file: #6144 modifies the final
run:block oftend-setup/action.yaml, this touches the firstuses:. Different hunks.