[theproducer]

Root cause

MockCordovaWebViewImpl.triggerDocumentEvent(String) (android/capacitor/src/main/java/com/getcapacitor/cordova/MockCordovaWebViewImpl.java) evaluated

eval("window.Capacitor.triggerEvent('" + eventName + "', 'document');", (s) -> {});

with no check that the page holds the bridge. Bridge.onPause() / Bridge.onResume() always forward to this (the only guard is isInitialized(), which is true for the whole life of the activity), so a pause or resume reaches whatever document the WebView currently holds.

That document is not guaranteed to have window.Capacitor:

• A non-app origin. Bridge.loadWebView() registers the bridge script with WebViewCompat.addDocumentStartJavaScript(webView, …, Collections.singleton(allowedOrigin)), where allowedOrigin is derived from appUrl only. Any other origin the WebView navigates to — e.g. an appAllowNavigation host used for an OAuth hand-off — runs without the bridge. This is the shape reported in Breaking [Bug]: Capacitor is undefined, Capacitor.triggerEvent is not a function in allowNavigation routes #7957.
• The server.errorPath page. WebViewLocalServer.handleLocalRequest returns the error page from the isLocalFile(...) || isErrorUrl(...) branch, which never calls jsInjector.getInjectedStream(...). Note this only strips the bridge on a WebView without DOCUMENT_START_SCRIPT, or when server.url puts the app on a different origin from the error page — on a current WebView with no server.url the error page is same-origin with appUrl and does get the script.
• A load still in flight, including a reload after the renderer was killed. This is the Android counterpart of the iOS case fixed in fix(ios): do not forward scene lifecycle events to the page before it has loaded #8595.

The resulting TypeError is raised inside the page, so it only surfaces through BridgeWebChromeClient.onConsoleMessage as the Capacitor/Console line in the report. No functional breakage beyond the lost Cordova pause/resume on a page that could not have consumed them anyway.

The change

Guard the call inside the evaluated snippet and log a debug line when the event is dropped. Dropping rather than deferring matches #8595: a page without the bridge has nothing to pause or resume, and a deferred resume after a fresh load would make the page react to a transition it never went through. pause/resume on a loaded app page are unchanged.

The snippet stays ES5 (window.Capacitor && window.Capacitor.triggerEvent) rather than optional chaining, because android.minWebViewVersion allows WebViews older than the Chrome 80 that introduced ?..

How it was tested

New instrumented test android/capacitor/src/androidTest/java/com/getcapacitor/android/CordovaLifecycleEventTest.java, on a Pixel 9a API 36 emulator:

• skipsLifecycleEventsWhenBridgeIsAbsent — loads about:blank into the bridge's WebView (asserting typeof window.Capacitor === 'undefined' first, so the test cannot pass vacuously), pumps onPause() / onResume(), and fails if the console records a triggerEvent error. Before the change it fails with the exact reported string:

  java.lang.AssertionError: lifecycle event reached a page with no bridge:
    ERROR: Uncaught TypeError: Cannot read properties of undefined (reading 'triggerEvent')
  

  After the change it passes, and Logcat shows D Capacitor: Skipped the 'pause' document event: no Capacitor bridge on the current page.

• stillDeliversLifecycleEventsWhenBridgeIsPresent — installs a window.Capacitor.triggerEvent stub and asserts pause:document and resume:document both still arrive.

Suites run:

• ./gradlew lint build test -b capacitor/build.gradle — BUILD SUCCESSFUL, "Lint found no new issues".
• Full instrumented suite via adb am instrument — OK (5 tests) (CordovaLifecycleEventTest ×2, HttpInterceptorNavigationTest ×2, ExampleInstrumentedTest).
• prettier with @ionic/prettier-config — both touched files clean.

connectedAndroidTest could not be used directly because the UTP host dependencies were not in the local Gradle cache; the APK was built with Gradle and driven with adb am instrument instead, which runs the same AndroidJUnitRunner.

What a reviewer should check by hand

1. A real end-to-end background/foreground on a device, on an app page, to confirm Cordova plugins still see pause/resume — the test asserts this against a JS stub, not against core/native-bridge.ts's real triggerEvent.
2. The server.errorPath path with server.url set, which is the configuration where the error page is genuinely cross-origin from the bridge script. I traced it but did not build that configuration.
3. Whether the same guard is wanted on Bridge.triggerJSEvent(...) (Bridge.java:887-893), which has the identical unguarded pattern. It is public API called explicitly by plugins rather than by the lifecycle, so I left it alone — say the word and I will follow up.
4. Whether iOS should also guard on bridge presence in addition to the load-state guard from fix(ios): do not forward scene lifecycle events to the page before it has loaded #8595, since the non-app-origin case applies there too.

Related

• fix(ios): do not forward scene lifecycle events to the page before it has loaded #8595 — the iOS fix this mirrors.
• Breaking [Bug]: Capacitor is undefined, Capacitor.triggerEvent is not a function in allowNavigation routes #7957 (closed as duplicate) — window.Capacitor undefined on allowNavigation routes; the same missing-bridge condition, with a reproduction.
• E/Capacitor/Console error #2183, bug: cap run android blank screen, “Cannot read property ‘triggerEvent’ of undefined” #7048, TypeError: window.Capacitor.triggerEvent is not a function. (In 'window.Capacitor.triggerEvent('pause', 'document')', 'window.Capacitor.triggerEvent' is undefined) #7458 — earlier reports of the same console string, closed without the cause being established.