test: drive servers in memory with test_server() / testServer()

[schloerke]

py-shiny#2470 added shiny.testserver.test_server(), the Python counterpart to R's testServer(): it runs an app's server against a mock connection, no browser and no subprocess.

That suits shinyreact unusually well. A ui.tsx server contains only reactive computation, so "input X produces output Y" is the server, and the JSON a test asserts is exactly what useShinyOutputValue() receives:

with test_server(APP) as ts:
    ts.set_inputs(bins=9)
    assert ts.get_output("dist_caption") == "272 eruptions in 9 bins"

Package tests

• pkg-py/tests/test_in_memory_server.py — three tests, reusing the Playwright fixture apps as-is:

  • reactive_output publishing through a real session. test_reactive_output.py only reaches Renderer.transform(), so nothing pinned the wire path.
  • module namespace isolation via ts.make_scope("a") — the in-memory twin of playwright/test_module_namespaces.py, which spends a browser on a purely server-side claim.
  • output error statuses and messages — the server half of playwright/test_output_error.py.

  It needs no tests-e2e extras, so it belongs in the unit suite rather than the Playwright subtree.

• pkg-r/tests/testthat/test-render.R — the R mirrors of the last two, cross-referenced by name, with the divergence stated where it exists: R's testServer() raises (expect_error(output$answer, class = "shiny.silent.error")) where Python's reports .status.

Example tests

One Python test file per example with a server (01, 02, 05, 06, 07, 08, 10), plus R counterparts for the two that ship an app.R — 07-plotly had no R test directory at all, so it gains tests/testthat.R too.

Two FEATURES.md leaves turned out to be wrong, and are now corrected and pinned: 05-temperature and 08-input-handler both claimed input.x() is None before the client's first message, so the server returns None / "—". It raises a silent exception — those branches are unreachable from a real client, and the outputs never render. That is the kind of claim only a test that drives the app can falsify, which is the argument for this whole PR in one leaf.

(test) markers added across seven example trees for what these pin: wire shapes, zone thresholds, pluralization, the two-step event handling, the AsIs wrapping that makes a one-bin R result a JSON array, and the plotly payload's plotly-main-but-not-plotly-binding dependency split.

examples/01-hello/tests/test_faithful.py opened with "app.py calls set_react_page() at module scope, so the two outputs' logic cannot be unit tested". No longer true anywhere in the tree.

Docs and skills

• pkg-py/docs/articles/testing.qmd and its R mirror pkg-r/vignettes/articles/testing.Rmd are retitled Testing and gain a server-testing half ahead of the wire-tap material, tabbed py/R. Both render; headings and tabsets verified in the built HTML.
• The shipped skills were actively wrong. shinyreact-build-app/references/testing.md said "[py] has no testServer() equivalent … This is a real gap, not an oversight in your app." It now documents test_server() alongside R's, and shinyreact-convert-app's verify phase goes from three layers to four. make update-skills run.
• Four things are documented because each cost time to find: pass an absolute Path (a relative one resolves against the test file's directory); an untyped id needs no :shinyreact.default suffix (both Python handlers are no-ops) but a typed id does, since the suffix is what runs the handler; an event input needs two set_inputs calls, mirroring the client's register-at-mount then send; and an unset input means status == "silent", not a None value.

test_server() is newer than shiny 1.7.0 — which this repo already depends on via the py-shiny main git dep — and the docs and skills say so explicitly.

Two upstream issues filed from this work

• posit-dev/py-shiny#2492 — a silent req() failure is invisible in memory: get_output() keeps the previous value with status="ok" while the browser blanks the output, so a test can pass asserting a value the user cannot see. Includes R evidence that testServer() does not retain the stale value, which is the parity rationale the current behavior cites.
• posit-dev/py-shiny#2493 — status == "silent" means "never rendered", the exact inversion of what Shiny already taught req()/silent to mean. Suggests "never-rendered", which also frees the name for #2492.

Neither blocks this PR; the one leaf that would have asserted the missing behavior is written as a comment naming the issue instead.

Drive-by

pkg-r/tests/testthat/test-examples.R read every file in an example directory and then filtered to app.R + www/, so the new __pycache__ produced 208 readLines() warnings. It filters first now.

Verification

• uv run pytest — 195 passed; tox py310 and py314 — 195 each
• Rscript -e 'testthat::test_local()' — 320 pass, 0 warn (was 208 warn)
• make r-check — 0 errors, 0 warnings, 0 notes
• each example's tests pass from the app directory: pytest, shiny::runTests()
• both doc sites render