feat: honour [skip-ci] org-wide; add a reusable revdeps workflow - #34
Merged
Merged
Conversation
Two things callers were each solving for themselves. ## [skip-ci] GitHub stops a run itself for its own keywords ([skip ci], [ci skip], [no ci], [skip actions], [actions skip]), but the hyphenated [skip-ci] is a PE convention it does not know about. Every caller that wanted it was carrying its own `if:` on a hand-rolled job -- and moving to a thin caller silently dropped it, since these workflows had no equivalent. The guard now lives here. It tests `github.event.head_commit.message`, the tip of the push, which matches GitHub's own semantics. The hand-rolled versions tested `github.event.commits[0].message` -- the OLDEST commit in the push -- so a [skip-ci] on the tip of a multi-commit push used to be ignored. null on pull_request and schedule, so those always run. ## revdeps.yaml `revdeps-check` is a composite action, so every caller had to write the same job harness around it: checkout, geospatial system libraries, R, and the dependency install. quickPlot and Require had each written one. The default is a single ubuntu-latest/release leg. quickPlot's version ran a three-OS matrix on every pull_request, which spends hours of runner time on a signal almost no PR changes; `config` takes a JSON matrix for packages that genuinely need more. Callers own their triggers -- workflow_dispatch plus a weekly schedule is the sane default, given the composite action's own warning that these checks are too heavy for standard runners. Not added to self-test.yaml: that exercises the composite actions through local `./` refs, which reusable workflows cannot use, and a revdep run is far too heavy to gate merges on. Purely additive -- no existing caller changes behaviour unless it puts [skip-ci] in a commit message. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NNLtKFXrNPN2XvuAjsy4Sy
eliotmcintire
added a commit
to PredictiveEcology/Require
that referenced
this pull request
Sep 3, 2026
Follows PredictiveEcology/actions#34, which adds both. revdeps.yaml drops from 49 lines to 24: the job harness it was carrying (checkout, geospatial system libraries, R, dependency install) is the same one quickPlot had written, so it belongs upstream rather than here. Adds the weekly schedule alongside workflow_dispatch. The [skip-ci] caveat in R-CMD-check.yaml is gone -- the shared workflows honour it now, so a thin caller no longer loses it. split-library-check, the one job we still own, moves from commits[0] to head_commit to match the shared guard: the tip of the push rather than the oldest commit in it. Depends on PredictiveEcology/actions#34 being merged; revdeps is workflow_dispatch/schedule only, so nothing here runs it before then. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NNLtKFXrNPN2XvuAjsy4Sy
eliotmcintire
added a commit
that referenced
this pull request
Sep 4, 2026
Every SpaDES module repository carries a hand-generated copy of this job, written once by SpaDES.core::use_gha() and never regenerated. The survey in #36 found 74 repositories carrying it across 96 repository/branch combinations, in two generations that had drifted apart: 46 combinations still on ubuntu-20.04 with checkout@v2 and setup-r@v1, and 50 on the newer shape pinning five actions from this repository at @v0.2, @v0 or @v0.0.1. Those pins are the main thing blocking retirement of v0.1-v0.5. A caller now supplies its triggers and the module name; the pins live here and move once. Fixes carried in by the move: - [skip-ci] tested commits[0], the OLDEST commit in a push, and there is no commits field at all on pull_request -- so it worked on neither. Now uses head_commit, as in #34. - No concurrency group, so superseded runs were never cancelled (#30). - The hand-written apt-get install had no apt-get update, no retries and no timeout. That is the shape that left two LandWebUtils jobs sitting 6h on 2026-08-19. Now uses the same retry wrapper as install-spatial-deps. - The commit step ran on pull_request too, where github.ref is refs/pull/N/merge and the push cannot succeed; it failed silently every time, swallowed by `|| echo "No changes to commit"`. Now push-only. Rendering and committing are separate jobs, as in citation.yaml (#35): rendering installs the module's dependency tree and then executes the module's own .Rmd, which must not share a job with a write-scoped token. setup-r-deps is deliberately not used here. It delegates to r-lib/actions/setup-r-dependencies, which resolves from a DESCRIPTION, and modules have none -- their dependencies come from SpaDES.core::packages(). The apt hardening is reproduced instead. The second install-Rmd-pkgs call is preserved rather than dropped: the generated file had it, installing SpaDES can move versions the first call settled, and #36 flags it for someone with the history to confirm or remove. Refs #36. Claude-Session: https://claude.ai/code/session_015ZoicL7829pkb5ZbS5Ciww Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two things callers were each solving for themselves. Both surfaced while
converting Require to thin callers (PredictiveEcology/Require#212).
Base is
main— this repo has nodevelopmentbranch.1.
[skip-ci], org-wideGitHub stops a run itself for its own keywords (
[skip ci],[ci skip],[no ci],[skip actions],[actions skip]). The hyphenated[skip-ci]is a PE convention it does not know about, so every caller that wanted it was
carrying its own
if:on a hand-rolled job — and moving to a thin callersilently dropped it, because these workflows had no equivalent. Require hit
exactly that. The guard now lives here, on all five reusable workflows.
One deliberate difference from the hand-rolled versions: this tests
github.event.head_commit.message, the tip of the push, matching GitHub'sown semantics. The old copies tested
github.event.commits[0].message— theoldest commit in the push — so a
[skip-ci]on the tip of a multi-commitpush used to be ignored.
head_commitis null onpull_requestandschedule, andcontains(null, …)is false, so those always run.2. New reusable workflow
revdeps.yamlrevdeps-checkis a composite action, so every caller has to write the samejob harness around it: checkout, geospatial system libraries, R, dependency
install. quickPlot and Require had each written one — that is the
duplication this removes.
Defaults are deliberately more conservative than the incumbent:
ubuntu-latest/releaseleg. quickPlot's version ran athree-OS matrix on every
pull_request. The composite action's own READMEwarns these checks are too heavy for standard runners, and each revdep gets a
full
R CMD checkwith its dependency closure installed first — that ishours of runner time on a signal almost no PR changes.
configtakes a JSONmatrix for packages that genuinely need more.
workflow_dispatchplus a weeklyscheduleis the sane default.
Inputs:
config,cranonly,quiet,timeout,extra-packages,extra-repositories.The
cranonlydocs note something worth writing down: revdep sets go stale inboth directions. Require's stored
revdep/cran.mdsaid "no reversedependencies" because SpaDES.core had been archived on 2026-07-13 — and
SpaDES.core is back at 3.2.1, so that cached conclusion is now wrong.
Not added to self-test
self-test.yamlexercises the composite actions through local./refs, whichreusable workflows cannot use, and a revdep run is far too heavy to gate merges
on.
Risk
Purely additive. No existing caller changes behaviour unless it puts
[skip-ci]in a commit message. Follow-up worth doing separately: convertquickPlot's hand-rolled
revdeps.yamlto a thin caller — it is also pinned toinstall-spatial-deps@v0.3andactions/checkout@v4.🤖 Generated with Claude Code
https://claude.ai/code/session_01NNLtKFXrNPN2XvuAjsy4Sy