[trissim]

Numbers first. A tiny app test in tests/textual/test_initial_session_title.py cost 1.43–1.98 s, and about 0.05 s of that was the test body. The rest was Toad booting. After this PR plus OpenHCSDev/textual#256 they cost 0.93–1.04 s each. The template's first test (its warm-up) went from 10.6 s to 6.2 s.

Measured with timestamps along one test's path (runner → fork → child pytest → app construct → run_test → start worker → session ready → body → teardown → reap):

step	before
fork + child pytest collect	0.14 s
construct ToadApp	0.06 s
run_test until running	0.34–0.38 s
application start worker	0.13–0.18 s
first session ready	0.33–0.40 s
test body	~0.05 s
teardown	0.06 s

Commits

1. Test apps publish frames every 1 ms (TEXTUAL_FPS=1000). Wheel walks no longer wait a 7 ms tick per event. Full tests/textual on 8 workers: 221 s against main's 231 s. Small (4%), kept.
2. The test template imports every toad module before it forks. Each forked test imported the screens and widgets it mounted cold: 51 modules, 0.21 s cold. Importing them first also lets the template's schema warm-up cover Toad's own tables, so a child's ToadDatabase.prepare costs 4 ms instead of its cold build. Small tests: 1.43–1.98 s → 1.18–1.23 s.
3. Saved settings resolve each level's owner once for its file keys. saved_key resolved the containing record's value per field (DeclaredField.owner_type → ObjectState.get_resolved_value) only to learn the owner class. One resolve per level now. Opening a populated toad.json: 48.0 ms → 13.7 ms at every launch. Keys held unchanged; settings, forms and keymap tests: 36 passed.

Paired change, outside this repository: OpenHCSDev/textual#256. HeadlessDriver waited out the full 0.25 s palette query (textual_colours.py:354) for replies to writes that go nowhere. With it, small tests are 0.93–1.04 s.

What more workers does not fix

On 12 workers the suite took 376 s, against 221 s on 8. Each test forks its own route observer and render workers, so 12 workers oversubscribe 12 cores. The lever is per-test cost, not worker count.

Still per test, next

• Textual CSS parse: 0.077 s inside _prewarm_conversation_css. Stylesheets parse per app, and no module-level cache holds rule sets.
• Compose and styling: about 0.5 s across ~300 widgets.
• Main-thread idle: about 0.15 s waiting on the route observer's first pass over an empty bus root, the same cold pass as the launch first-paint stall.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XbFMSqJMarXn2ZE8iAL7Sd

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.