Prevent fixtures web server from blocking on unread output

[aik099]

Fixes #64
Closes #48

The test fixtures web server (php -S, started in tests/bootstrap.php) writes an access log line per request to its stderr. The Symfony Process never read that output, so once the OS pipe buffer filled up the server blocked on write() and stopped serving requests. From that point every page load in the browser hung until the 180s WebDriver timeout, failing all remaining tests in the run.

This is why the failures:

• only happen in the full suite (depends on total request count, not on a specific test);
• usually land on WindowTest (it's late in the run), but can hit other tests too;
• were never reproducible locally when running tests in isolation.

Confirmed locally on selenium/standalone-chromium:4: during the hang the fixtures server didn't respond even to curl from the host, and its main thread was stuck in fprintf to stderr. With the fix the full suite passes in ~90s instead of hanging.

Likely also related to #48 (window tests crashing only on GitHub Actions), and to the extra timeouts seen in this CI run of #65.

The remaining Chrome-only Form test failures (page content read before form submission navigation completes) are a separate issue and aren't addressed here.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.06%. Comparing base (66a0bd1) to head (2e34abb).

Additional details and impacted files

@@             Coverage Diff              @@
##               main      #67      +/-   ##
============================================
+ Coverage     88.33%   89.06%   +0.72%     
  Complexity      197      197              
============================================
  Files             1        1              
  Lines           506      494      -12     
============================================
- Hits            447      440       -7     
+ Misses           59       54       -5     

Flag	Coverage Δ
chrome	88.46% <ø> (+1.50%)	⬆️
chromium	87.85% <ø> (+4.25%)	⬆️
edge	88.46% <ø> (+4.46%)	⬆️
firefox	88.05% <ø> (+0.70%)	⬆️
selenium-2.53.1	87.04% <ø> (+1.47%)	⬆️
selenium-3	88.05% <ø> (+0.70%)	⬆️
selenium-4	88.66% <ø> (+0.91%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[aik099]

Comparison of the reference main build (run) vs this PR (run):

Matrix entries	main	This PR	Result
Selenium 2.53.1: chrome, firefox	✅ pass	✅ pass	no change
Selenium 3: chrome, firefox	❌ WindowTest hangs (Resize window, Switching to missing window), 6-12 min	✅ pass, ~1.5 min	fixed
Selenium 4: chrome (PHP 7.4-8.3), chromium, edge	❌ WindowTest hangs (3-5 tests) + random Form failures, 12-17 min	❌ random Form failures only, ~2 min	hangs fixed
Selenium 4: firefox	❌ WindowTest hangs (Resize window, Switching to missing window) + HoverTest (Double click hover, Right click hover)	❌ HoverTest only	hangs fixed
Static analysis	❌ unmatched ignoreErrors patterns	❌ same	unrelated

The remaining failures are not caused by this PR and will be addressed separately:

• random Form failures (1-3 per job, different each run: Submit empty textarea, Advanced form second submit, Basic get form, etc.) - page content is read before form submission navigation completes;
• HoverTest on Firefox + Selenium 4 - fails consistently;
• PHPStan ignoreErrors patterns no longer matching.

[aik099]

I've also removed the testDeprecatedShortPageLoadTimeoutThrowsException skip (2e34abb), that was added in #38 (tracked by #48) for Chromium-based browsers and Selenium 2 Firefox on GitHub Actions.

That freeze had the same root cause as the hanging WindowTest: setting page load timeout several times was never a problem by itself, it just added more requests to the fixtures web server, which made its output pipe fill up sooner. That's also why it was "cumulative", only happened on GitHub Actions and never in isolation.

With the skip removed the test passes in all 12 test jobs (run) without any hangs.

[stof]

the failure for the form tests in Chrome is caused by this bug in ChromeDriver 134+: https://issues.chromium.org/issues/402796660?pli=1
The last comment on this issue says they have a fix in progress, but it does not seem to be merged yet.

[aik099]

the failure for the form tests in Chrome is caused by this bug in ChromeDriver 134+: https://issues.chromium.org/issues/402796660?pli=1 The last comment on this issue says they have a fix in progress, but it does not seem to be merged yet.

Thank you. The #68 uses the proposed workaround. Ready for review as well.

Merging this one.

[uuf6429]

The Symfony Process never read that output, so once the OS pipe buffer filled up the server blocked on write() and stopped serving requests.

Nice catch.