feat: record traces of playwright.request API contexts

[Pavel Feldman (pavelfeldman)]

--tracing only covered browser contexts, so API tests using playwright.request.new_context() produced no trace. APIRequestContext.tracing is public since playwright 1.60, so contexts created through playwright.request.new_context() are now traced per test:

• Tracing starts when the context is created. Every chunk, including the first one, is titled after the test: contexts are usually created by fixtures before the test's artifacts recorder exists, so pytest_runtest_setup stores the current node id in config.stash for that.
• Each test records its own chunk, so a session-scoped context shared by many tests (the pattern in the API testing docs) yields one trace per test. A chunk without any action is dropped only for contexts that existed before the test started (e.g. session-scoped ones the test did not use); contexts the test creates itself keep their trace, like browser contexts. The check reads Playwright's trace format and keeps the trace when in doubt.
• Errors from stop_chunk surface like those from browser contexts instead of silently losing the trace; the context is still disposed (finally) when that happens.
• The declared pytest floor becomes 7.0 for pytest-playwright and 8.2 for pytest-playwright-asyncio (what pytest-asyncio>=0.24 already requires). pytest.Parser, used since before this PR, and pytest.StashKey both need pytest 7; the plugin did not import on pytest 6 before either.
• The traces go through the same retain-on-failure handling, trace-N.zip naming and report attachments as browser traces.

Tests that use Playwright but no page or context now get the artifacts recorder through a small autouse fixture. It only does so when tracing is enabled and the test transitively uses the playwright fixture. Reviewers should know that this also sets the recorder up for browser tests before their other function-scoped fixtures, rather than as a dependency of new_context; no behavior change was found. On playwright < 1.60 nothing changes.

Tests: session- and function-scoped API contexts with --tracing on (each trace contains only its test's requests and is titled after it; an unused self-created context keeps its trace; a test with a page and an API context gets trace-1.zip and trace-2.zip), --tracing retain-on-failure, and --tracing off leaving playwright.request.new_context untouched (sync + async).

Fixes #137

_{Stack created with GitHub Stacks CLI • Give Feedback 💬}

[Copilot]

🟡 Changes recommended

It breaks declared pytest 6 compatibility and can skip API-context disposal when trace finalization fails.

4 open findings

• Use pytest 6-compatible storage or raise the minimum version · New
• Use pytest 6-compatible config storage or raise the minimum version · New
• Await original disposal even when stop_chunk fails · New
• Ensure API context disposal runs when tracing cleanup fails · New

What changed in this PR

Adds per-test tracing for Playwright API request contexts in both plugin variants.

Changes:

• Instruments API context creation and disposal for trace collection.
• Integrates API traces with existing artifact retention.
• Adds synchronous and asynchronous integration coverage.

File	Description
tests/​test_sync.py	Tests synchronous API tracing scenarios.
tests/​test_asyncio.py	Tests asynchronous API tracing scenarios.
pytest-playwright/​pytest_playwright/​pytest_playwright.py	Implements synchronous API trace recording.
pytest-playwright-asyncio/​pytest_playwright_asyncio/​pytest_playwright.py	Implements asynchronous API trace recording.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.