microlinkhq / browserless / 34700594853

79%

push

github

fix(goto): run overlay dismissal in an isolated world (#925)

* fix(goto): run overlay dismissal in an isolated world

dismissOverlays ran in the page main world, so the page could read
window.__browserlessDismiss and observe every querySelectorAll the scan
performed through its own DOM hooks.

setup now registers the script with Page.addScriptToEvaluateOnNewDocument
in a named isolated world, and run reaches that same world through
Page.createIsolatedWorld, which returns the existing context, so the
idempotency guard stays shared and dialogs are never clicked twice.

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

* fix(goto): resolve the dismiss world again after a navigation

A navigation between Page.createIsolatedWorld and Runtime.evaluate
destroyed the world run() had just resolved, so it rejected with
"Cannot find context with specified id" and the post-navigation
fallback was skipped for that document. run() now retries up to three
times on stale-context errors, resolving the world on the new document.

Tests now also cover goto with autoconsent enabled, for a consent
dialog present at load and for a CMP inserted after navigation. The
re-scan test asserts the click count returned by run() itself, so it
fails if rescan stops working.

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

* perf(goto): create the dismiss world only in the top frame

Page.addScriptToEvaluateOnNewDocument with a worldName has no frame
filter, so Chrome created a browserless_dismiss world in every frame of
every document even though child frames return right away. With site
isolation disabled every iframe shares the page session: a page with 50
same-origin iframes held 51 dismiss worlds and 32.3 MB of JS heap
against 24.0 MB on master.

setup now listens to Page.domContentEventFired, which ... (continued)

1147 of 1361 branches covered (84.28%)

Branch coverage included in aggregate %.

67 of 109 new or added lines in 1 file covered. (61.47%)

5201 of 6403 relevant lines covered (81.23%)

78647.5 hits per line