fix: give the playwright npm server time to shut down browser children

[jordanmiguel]

The bug

When running a suite with --headed, the pest process hangs indefinitely after the "Tests: N passed" summary prints. On CI / in shells with a pipe it keeps the shell prompt busy; locally the user has to hit Ctrl-C to reclaim the terminal.

Repro (macOS + Linux, any project with browser tests):

./vendor/bin/pest tests/Browser --headed
# Summary prints, then hangs.

Headless mode is unaffected.

Root cause

In PlaywrightNpmServer::stop():

$this->systemProcess->stop(
    timeout: 0.1,
    signal: PHP_OS_FAMILY === 'Windows' ? null : SIGTERM,
);

Symfony's Process::stop($timeout) sends SIGTERM, waits $timeout seconds, then escalates to SIGKILL. 100ms is too short for headed Chromium — its graceful close typically takes 700–1200ms because it has to flush IPC to its GPU/renderer children, persist session state, and tear down its windows.

With the old 100ms grace period the server receives SIGKILL while Chromium is still in mid-shutdown; its stdin/stdout pipes stay open on orphan descriptors, so the parent PHP process blocks in proc_close/fclose waiting on them. That is the observable hang.

Fix

Bump the grace period to 2.0 seconds. Plenty of headroom for headed Chromium on every platform tested (macOS 14 Apple Silicon, Ubuntu 22.04 x86_64) and still bounded so a truly stuck Playwright server still gets SIGKILL.

Headless runs complete faster than the grace period, so there is no perceptible overhead — stop() returns as soon as Node exits.

Before / after

# Before
./vendor/bin/pest tests/Browser --headed
# ...Tests: 19 passed (80 assertions) in 48.23s
# hangs indefinitely, Ctrl-C required

# After
./vendor/bin/pest tests/Browser --headed
# ...Tests: 19 passed (80 assertions) in 48.23s
# returns to prompt within ~1s of summary

Notes

• No behavioural change for passing headless runs.
• If Chromium genuinely hangs beyond 2s (e.g. deadlock), SIGKILL still fires at T+2s — no regression vs. the current escalate-to-SIGKILL behaviour, just pushed out 1.9s.
• Did not add a new test because the hang only reproduces with a spawned headed Chromium subprocess, which this repo's own suite does not exercise. Happy to add one (e.g. tests/CoreFeatures/HeadedShutdownTest.php that launches + closes + asserts the parent exits within N seconds) if that fits the project's testing philosophy.

[ETES-Stuttgart]

I've got the same issue but only when i use Firefox. This PR fixed it 🎊. Thanks.

[Kimtran-art]

I hit this same teardown hang on Firefox (headless) and can confirm this PR fixes it there too, not only Chromium in headed mode.

Without the change, ./vendor/bin/pest --browser firefox runs the specs, they pass, but the PHP process never exits: it wedges in symfony/process during teardown and leaves an orphaned playwright run-server behind. We had been working around it with a wrapper script that watches the JUnit log and SIGKILLs the wedged process group.

With timeout: 2.0 (this PR), the exact same run exits cleanly (exit code 0, ~18s) and leaves no orphaned run-server. So on Firefox the old 0.1s timeout was racing the browser-child teardown as well.

Environment: pest-plugin-browser v4.3.1, Playwright 1.59.x, PHP 8.3, macOS. Would be great to get this merged. Thanks @jordanmiguel!