[strengejacke]

@jmgirard What would be best practice to have vdiffr tests for both Windows and Linux? Snapshot SVG files usually slightly differ on both OS, so you can't have just one test for both OS.

[strengejacke]

@copilot please review and check whether this is an appropriate way to add tests specifically for linux.

[Copilot]

@copilot please review and check whether this is an appropriate way to add tests specifically for linux.

Yes, this is functionally appropriate: the OS guard restricts the tests to Linux, and variant = "linux" keeps the Linux SVG snapshots separate. The repository already uses this variant pattern for Linux-specific vdiffr snapshots. The main tradeoff is duplicated test setup/assertions, which can drift from test-plot-facet.R; consider sharing the plot-generation code if these tests will evolve. I’d also use the repository’s usual .R extension rather than .r. Neither point invalidates the current approach.

[Copilot]

🟢 Approval recommended

No active review issues remain for the Linux visual regression test additions and platform-specific snapshots.

No open issues 🎉

What changed in this PR

Extends modelbased’s visual regression tests to Linux, using separate snapshot variants where rendering differs from Windows. No production code changes are included.

Changes:

• Enables ordinal plotting tests on Linux and separates Linux-specific bracl and nnet snapshots.
• Adds Linux-only facet plotting tests, including categorical predictors and the emmeans backend.
• Adds Linux SVG baselines for ordinal and faceted plots.

File	Description
tests/​testthat/​test-plot-ordinal.R	Enables Linux execution and adds platform-specific snapshot checks.
tests/​testthat/​test-plot-facet-linux.r	Adds Linux facet plotting tests.
tests/​testthat/​_snaps/​linux/​plot-ordinal/​plot-ordinal-nnet.svg	Linux baseline for multinomial predictions by gender.
tests/​testthat/​_snaps/​linux/​plot-ordinal/​plot-ordinal-nnet-2.svg	Linux baseline for multinomial interaction predictions.
tests/​testthat/​_snaps/​linux/​plot-ordinal/​plot-ordinal-bracl.svg	Linux baseline for adjacent-category model predictions.
tests/​testthat/​_snaps/​linux/​plot-facet-linux/​plot-interaction-facets-cat-4emmeans.svg	Categorical facet baseline using emmeans.
tests/​testthat/​_snaps/​linux/​plot-facet-linux/​plot-interaction-facets-cat-1.svg	Categorical facet baseline without connecting lines.
tests/​testthat/​_snaps/​linux/​plot-facet-linux/​plot-interaction-facets-cat-1-lines.svg	Categorical facet baseline with connecting lines.
tests/​testthat/​_snaps/​linux/​plot-facet-linux/​plot-interaction-facets-2.svg	Baseline for continuous predictions with opacity grouping.
tests/​testthat/​_snaps/​linux/​plot-facet-linux/​plot-interaction-facets-10.svg	Baseline for multi-factor facets with confidence bands.
tests/​testthat/​_snaps/​linux/​plot-facet-linux/​plot-interaction-facets-1.svg	Baseline for two-panel interaction predictions.

_{Review effort: Balanced}

[jmgirard]

Sorry I missed this

[strengejacke]

No worries. This is still "work in progress", and this was just a test. So I still need to figure out how I can add variants of visual snapshot for windows and linux, so that testing on one system does not remove those snapshots for the other system?

[jmgirard]

@strengejacke Right now the Linux facet tests live in their own file, test-plot-facet-linux.R, which is skipped on Windows, while test-plot-facet.R is skipped on Linux. The catch is that when you run the full test suite locally, testthat deletes any snapshot that no test used during that run. It doesn't do this on CI. So a full local run on Windows removes the Linux facet snapshots, and a full run on Linux removes the Windows ones. I tried this on a small test package to be sure.

The fix is to use one expect_doppelganger() call per plot and let the variant argument depend on the OS. testthat then sees the same snapshot name on both systems and keeps both copies. That's also why test-plot-ordinal.R doesn't have this problem, because its Windows and Linux blocks are in the same file and use the same names.

You'd add a helper to tests/testthat/helper.R:

snapshot_variant <- function() {
  if (identical(Sys.info()[["sysname"]], "Linux")) "linux" else NULL
}

and each call in test-plot-facet.R becomes:

vdiffr::expect_doppelganger(
  "plot-interaction-facets-1",
  plot(pr, show_data = FALSE),
  variant = snapshot_variant()
)

Then we'd remove "linux" from that file's skip_on_os(), delete test-plot-facet-linux.R, and move the Linux SVGs from _snaps/linux/plot-facet-linux/ to _snaps/linux/plot-facet/. The Windows snapshots stay where they are. The same approach would also let test-plot-ordinal.R drop its duplicated Linux block.

Want me to open a PR for this?

[strengejacke]

Sounds good! But I would wait for #704 (and #705) to be resolved first.