microlinkhq / browserless / 34700068245

79%

push

github

fix(goto): run autoconsent in an isolated world (#926)

* fix(goto): run autoconsent in an isolated world

autoconsent lived in the page's main world: exposeFunction left
puppeteer_autoconsentSendMessage, autoconsentSendMessage and
autoconsentReceiveMessage on window, and its DOM scans showed up in
stack traces captured by page hooks.

Inject it with Page.addScriptToEvaluateOnNewDocument into a named
isolated world, expose the binding only to that world, and answer
through Runtime.callFunctionOn in the calling context. Messages are
accepted only from the top frame's autoconsent context, which replaces
the nonce guard: the page cannot reach an isolated world. Rule eval
snippets still run in the main world because they read page globals.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0116zka7w2WKPSi3kqHyVAET

* fix(goto): harden isolated autoconsent setup

- cache the setup promise before the first await, so concurrent goto
  calls on one page register one listener set and answer each message
  once
- give the raw CDP binding its own name: Chrome re-installs bindings on
  a back/forward cache restore and used to replace the wrapper
- detach the session listeners when the page closes
- drop the redundant receive-message guard; autoconsent's bundle already
  skips a second instance in the same world
- drop the executionContextDestroyed handler: navigations only emit
  executionContextsCleared for the top-frame world, and context ids are
  reused across documents, so clearing is what keeps the filter correct

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0116zka7w2WKPSi3kqHyVAET

* fix(goto): consume autoconsent handler rejections and reset init per document

- a rejected message handler (e.g. the rules file failing to load during
  init) was a floating promise; under Node's default policy it crashes
  the process
- `_auto... (continued)

1134 of 1346 branches covered (84.25%)

Branch coverage included in aggregate %.

277 of 282 new or added lines in 1 file covered. (98.23%)

5145 of 6327 relevant lines covered (81.32%)

79584.93 hits per line