Add testthat-module reusable workflow

[eliotmcintire]

Runs a SpaDES module's testthat suite. This is the CI half of the module release wave — until now only the 6 CBM modules ran tests in CI, via a workflow that depended on a personal fork; the other ~200 module repos either render an Rmd or have no workflow at all.

It is render-module-rmd.yaml with two jobs swapped out

Everything above the render step — spatial system deps, the apt retry wrapper, install-Require, install-SpaDES, and dependency resolution straight from the module's own reqdPkgs via SpaDES.core::packages() — is deliberately identical, so a fix to one is a fix to both. That includes the hard-won details: the -reusable concurrency suffix (#33), GITHUB_PAT on the job, single-indexing packages(...)[[1]], and [skip-ci] read from head_commit (#34).

There is no commit job, and the omission is load-bearing. The workflow runs convertToPackage(), which rewrites a module and is not reversible. Nothing it produces may be pushed back. Since it is a near-copy of a workflow that does commit, this is called out in the file itself so nobody restores it by symmetry.

Three decisions

The module is tested as a package. A module keeps its functions inline in <module>.R, so without conversion they are only parsed into a simList and a test cannot call one directly — you would write sim@.xData$.mods$myModule$myHelper(3L). convertToPackage(destinationPath = ) builds the rendition in a throwaway directory, leaving the module untouched, and the tests run against a real namespace, so expect_equal(myHelper(3L), 6L) works as in any package. Verified end to end locally.

A module with no tests passes. Almost every module currently carries only the dead newModule() scaffolding (test-template.R with C:/Users/yonluo/... paths). Failing hard would paint the whole family red for the absence of tests rather than for a defect. The workflow reports "no tests" in the job summary and exits clean.

Coverage goes to the job summary, not codecov. Module repos have no CODECOV_TOKEN and no codecov.yml, and ~28 modules with near-zero tests would create 28 projects reading 0–5% with no history for the house target: auto ratchet to compare against. SpaDES.core::moduleCoverage() reports against <module>.R rather than the generated R/ copy — see PredictiveEcology/SpaDES.core#441.

One temporary step

install-SpaDES installs the released SpaDES.core, which does not have convertToPackage(destinationPath = ) or moduleCoverage() (SpaDES.core #440 and #441, both merged to development). Until a release carries them, the workflow installs SpaDES.core@development explicitly. The step is commented as removable and names the PRs.

Also added

• examples/testthat-module.caller.yaml — the call site
• examples/module-tests-setup.R — a starting tests/testthat/setup.R for a module, which sets options and a scratch tree and deliberately installs nothing

Not yet done

This has not run against a real module. I intend to pilot it on one small module (historicFires) before rolling callers out to the wave, rather than opening 28 PRs against an untested workflow.

🤖 Generated with Claude Code

https://claude.ai/code/session_0125hLaqRWaPcutdboenVK2g