Stop committing the site; publish it from Actions #4
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
| name: statistical-tests | |
| # The Monte Carlo tests -- parameter recovery, bootstrap standard-error | |
| # calibration, confidence-interval coverage, and the Type I error rate of the | |
| # goodness-of-fit test -- are the package's only evidence that its estimates are | |
| # any good, as opposed to merely computed without error. | |
| # | |
| # They carry skip_on_cran(), because they are slower than CRAN's budget rewards | |
| # and a Monte Carlo assertion that goes marginal on one of CRAN's many platforms | |
| # is an expensive way to find out. But skip_on_cran() runs only when NOT_CRAN is | |
| # "true", and neither r-canon's reusable check nor rcmdcheck sets that -- checked, | |
| # rather than assumed: no function in the rcmdcheck namespace mentions NOT_CRAN. | |
| # | |
| # So without this workflow the tests are skipped everywhere. They were: before | |
| # this file existed they sat behind GUESS_FULL_TESTS, a variable set in no file in | |
| # the repository, and a recorded run showed `SKIP 17 | PASS 3988` with all sixteen | |
| # statistical tests among the skips. They cost about 31 seconds. | |
| on: | |
| push: | |
| branches: [main, master] | |
| pull_request: | |
| branches: [main, master] | |
| workflow_dispatch: | |
| inputs: | |
| replicates: | |
| description: "Replicates per Monte Carlo study (GUESS_MC_REPS)" | |
| required: false | |
| default: "100" | |
| jobs: | |
| monte-carlo: | |
| runs-on: ubuntu-latest | |
| env: | |
| GITHUB_PAT: ${{ secrets.GITHUB_TOKEN }} | |
| # The point of this workflow. CI is not CRAN. | |
| NOT_CRAN: true | |
| # Replicates for the studies that read `mc_reps()`. The gates in | |
| # tests/testthat/helper-gates.R derive their tolerance from this, so | |
| # raising it makes those assertions stricter without a test being edited -- | |
| # and lowering it cannot quietly weaken them, because the band widens | |
| # visibly in the failure message. Raise it on a dispatch run to pin a | |
| # result down; the default keeps the suite at about 90 seconds. | |
| GUESS_MC_REPS: ${{ github.event.inputs.replicates || '100' }} | |
| steps: | |
| - uses: actions/checkout@v7 | |
| - uses: r-lib/actions/setup-r@v2 | |
| with: | |
| use-public-rspm: true | |
| - uses: r-lib/actions/setup-r-dependencies@v2 | |
| with: | |
| extra-packages: any::testthat, any::devtools | |
| needs: check | |
| - name: Run the full test suite, statistical tests included | |
| run: | | |
| Rscript -e ' | |
| devtools::load_all(".", quiet = TRUE) | |
| results <- testthat::test_dir("tests/testthat", stop_on_failure = FALSE) | |
| frame <- as.data.frame(results) | |
| skipped <- sum(frame$skipped) | |
| failed <- sum(frame$failed) | |
| cat(sprintf("passed %d failed %d skipped %d\n", | |
| sum(frame$passed), failed, skipped)) | |
| # A skip here means the gate is dormant again, which is the failure | |
| # this workflow exists to prevent. Treat it as a failure. | |
| if (skipped > 0L) { | |
| cat("Statistical tests were skipped; NOT_CRAN is not taking effect.\n") | |
| quit(status = 1L) | |
| } | |
| if (failed > 0L) quit(status = 1L) | |
| ' |