(onProcessKilled) Crash while browsing: IllegalArgumentException: Display not attached

[ekager]

Just ran into this while testing something else in Fenix. Don't have exact STR but I was opening/closing/switching tabs and toggling PB and then backgrounded the app I think? (did NOT have to do with custom tabs though)
Going to file just to keep open in case someone else runs into it as well.
(cc @pocmo )

java.lang.IllegalArgumentException: Display not attached
        at org.mozilla.geckoview.GeckoSession.releaseDisplay(GeckoSession.java:2438)
        at org.mozilla.geckoview.GeckoView.releaseSession(GeckoView.java:393)
        at org.mozilla.geckoview.GeckoView.setSession(GeckoView.java:436)
        at mozilla.components.browser.engine.gecko.GeckoEngineView.rebind(GeckoEngineView.kt:150)
        at mozilla.components.browser.engine.gecko.GeckoEngineView.access$rebind(GeckoEngineView.kt:27)
        at mozilla.components.browser.engine.gecko.GeckoEngineView$observer$1.onProcessKilled(GeckoEngineView.kt:68)
        at mozilla.components.browser.engine.gecko.GeckoEngineSession$createContentDelegate$1$onKill$2.invoke(GeckoEngineSession.kt:649)
        at mozilla.components.browser.engine.gecko.GeckoEngineSession$createContentDelegate$1$onKill$2.invoke(GeckoEngineSession.kt:612)
        at mozilla.components.support.base.observer.ObserverRegistry.notifyObservers(ObserverRegistry.kt:137)
        at mozilla.components.concept.engine.EngineSession.notifyObservers(Unknown Source:7)
        at mozilla.components.browser.engine.gecko.GeckoEngineSession$createContentDelegate$1.onKill(GeckoEngineSession.kt:649)
        at org.mozilla.geckoview.GeckoSession$2.handleMessage(GeckoSession.java:370)
        at org.mozilla.geckoview.GeckoSession$2.handleMessage(GeckoSession.java:359)
        at org.mozilla.geckoview.GeckoSessionHandler.handleMessage(GeckoSessionHandler.java:90)
        at org.mozilla.gecko.EventDispatcher$3.run(EventDispatcher.java:356)
        at android.os.Handler.handleCallback(Handler.java:883)
        at android.os.Handler.dispatchMessage(Handler.java:100)
        at android.os.Looper.loop(Looper.java:214)
        at android.app.ActivityThread.main(ActivityThread.java:7356)
        at java.lang.reflect.Method.invoke(Native Method)
        at com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run(RuntimeInit.java:492)
        at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:930)

┆Issue is synchronized with this Jira Task

[pocmo]

This is an interesting stacktrace.

The call goes through onProcessKilled() which means that the content process got killed by Android (It did not crash, but Android is reclaiming system resources by killing it), which also makes it very likely that you backgrounded the app, as you said.

In that situation, under the hood, we are creating a new GeckoSession, restoring the last state for it, and then we try to render the new GeckoSession. And that's where it is crashing with Display not attached here.

We implemented this in:

• Recover when content process gets killed android-components#3310
• https://bugzilla.mozilla.org/show_bug.cgi?id=1557096
• [Bug] Fenix doesn't handle content process dying #1577

Back then this situation was easy to reproduce by using kill -9 on the content process. Maybe that can help reproducing this crash here now.

I'm going to move this into the AC sprint too.

[pocmo]

I tried to reproduce this with the STR from #1577 but wasn't successful.

[pocmo]

I found some similar stacktraces on Sentry and merged them together.

[pocmo]

Sentry issue: FENIX-53P

[pocmo]

After mozilla-mobile/android-components#7867 lands in AC we can rewrite that logic in AC. Instead of recreating a GeckoSession and restoring it, we will be able to "suspend" the whole tab, only keeping it's EngineSessionState. So that an individual tab will only be restored when it is needed the next time.

[liuche]

We're doing a crash sprint, and wondering what the next work is here, and whether we can pick up/split any of this work.

[pocmo]

Patch landed in AC.
Fenix counter part: #14823

[ekager]

Should we close this since #14823 has landed?

[pocmo]

Yes, let's close it. I have some other issues for other white screen problems.. :)