1094206 - crash in java.lang.NullPointerException: at org.mozilla.gecko.gfx.GeckoLayerClient.activateProgram(GeckoLayerClient.java)

Status: RESOLVED WORKSFORME

(Firefox for Android Graveyard :: Toolbar, defect)

Description

[Aaron Train [:aaronmt]]

This bug was filed from the Socorro interface and is 
report bp-357b89c0-fc0a-49ed-836c-9a7f42141104.
=============================================================

java.lang.NullPointerException
	at org.mozilla.gecko.gfx.GeckoLayerClient.activateProgram(GeckoLayerClient.java:745)
	at dalvik.system.NativeStart.run(Native Method)
	at dalvik.system.NativeStart.run(Native Method)
	at dalvik.system.NativeStart.run(Native Method)

Comment 1

[Milan Sreckovic [:milan] (needinfo for best results)]

Sotaro, anything you can do to help?

Comment 2

[Sotaro Ikeda [:sotaro]]

(In reply to Milan Sreckovic [:milan] from comment #1)
> Sotaro, anything you can do to help?

It seems that there is a case that mLayerRenderer is null.
  https://dxr.mozilla.org/mozilla-central/source/mobile/android/base/java/org/mozilla/gecko/gfx/GeckoLayerClient.java#735

Comment 3

[Sotaro Ikeda [:sotaro]]

I take a look a bit more.

Comment 4

[Sotaro Ikeda [:sotaro]]

From the code, when nsWindow::DrawWindowOverlay() should have mLayerRendererFrame and nsWindow should not be destroyed.
  https://dxr.mozilla.org/mozilla-central/source/widget/android/nsWindow.cpp#3210
And before calling the DrawWindowOverlay(), nsWindow::DrawWindowUnderlay() should be called by Compositor.

When the DrawWindowUnderlay() was called, GeckoLayerClient.mLayerRenderer seemed valid, because mLayerRendererFrame should be valid in the DrawWindowOverlay(). It seemed to mean that GeckoLayerClient.mLayerRenderer becomes invalid between DrawWindowOverlay() call and DrawWindowOverlay().

Comment 5

[Sotaro Ikeda [:sotaro]]

GeckoLayerClient dies not release GeckoLayerClient.mLayerRenderer during its life time since it get in GeckoLayerClient.onGeckoReady()
  https://dxr.mozilla.org/mozilla-central/source/mobile/android/base/java/org/mozilla/gecko/gfx/GeckoLayerClient.java#158

It seemed that GeckoLayerClient was also released when the the crash happened.

Comment 6

[Sotaro Ikeda [:sotaro]]

GeckoLayerClient and LayerRenderer are destroyed in LayerView.destroy().
  https://dxr.mozilla.org/mozilla-central/source/mobile/android/base/java/org/mozilla/gecko/gfx/LayerView.java#194

Comment 7

[Sotaro Ikeda [:sotaro]]

There is a code that to trigger to pause compositor. It is triggered from surfaceDestroyed().

SurfaceListener.surfaceDestroyed()
->LayerView.onDestroyed()
->GLController.serverSurfaceDestroyed()
->GLController.pauseCompositor()
->nsWindow::GLControllerSupport::PauseCompositor()
->CompositorChild::SendPause() // sync ipc
->CompositorParent::RecvPause()

Comment 8

[Sotaro Ikeda [:sotaro]]

FYI: diagram around GeckoView
 https://github.com/sotaroikeda/firefox-diagrams/blob/master/mobile/mobile_GeckoView_47.pdf

Comment 9

[Sotaro Ikeda [:sotaro]]

LayerView.destroy() could be called from GeckoView.onDetachedFromWindow()

https://dxr.mozilla.org/mozilla-central/source/mobile/android/base/java/org/mozilla/gecko/GeckoView.java#325

Comment 10

[Kartikaya Gupta (email:[e-mail])]

All the crash reports I looked at with this signature seem to have some memory pressure stuff on the stack, so it might be that low-memory issues are causing the LayerView to get destroyed in the middle of a composite. The native code holds a reference to the GeckoLayerClient and tries to call drawWindowUnderlay but the layer renderer has already been destroyed.

Also I'm not sure if the transition to GeckoView and Jim Chen's patches will affect this - a lot of the code was modified recently.

Comment 11

[Sotaro Ikeda [:sotaro]]

Yea, I agree. It seems to related to low-memory situation.

Comment 12

[Sotaro Ikeda [:sotaro]]

SurfaceListener.surfaceDestroyed() is normally called when SurfaceView is hidden. onDetachedFromWindow() is called when Activity is going to be destroyed. But in normal use case, onDetachedFromWindow() seems not called.

Comment 13

[Sotaro Ikeda [:sotaro]]

(In reply to Sotaro Ikeda [:sotaro] from comment #12)
> But in normal use case, onDetachedFromWindow() seems not called.

I run fennec on my phone and used it several ways, but I did not see the onDetachedFromWindow().

Comment 14

[Sotaro Ikeda [:sotaro]]

(In reply to Sotaro Ikeda [:sotaro] from comment #12)
> SurfaceListener.surfaceDestroyed() is normally called when SurfaceView is
> hidden. onDetachedFromWindow() is called when Activity is going to be
> destroyed. But in normal use case, onDetachedFromWindow() seems not called.

Other possibility is that BrowserApp(Activity) is destroyed without surfaceDestroyed().

Comment 15

[Sotaro Ikeda [:sotaro]]

Attachment: patch - Add manual LayerView.onDestroyed() callings

Comment 16

[Sotaro Ikeda [:sotaro]]

I could not reproduce the crash. But attachment 8717360 [details] [diff] [review] seems to address the problem.

Comment 17

[(inactive) Jim Chen [:jchen] [:darchons]]

Are there any recent crash reports? The latest reports I've seen are from 44, which is before the major LayerView/GLController refactoring that landed in 46. Without STR, we don't know if this still happens or not, and I'd rather wait before we make speculative changes.

Comment 18

[Sotaro Ikeda [:sotaro]]

Comment on attachment 8717360 [details] [diff] [review]
patch - Add manual LayerView.onDestroyed() callings

Ok, I cancel the review request and wait if there could be a new crash.

Comment 19

[Kartikaya Gupta (email:[e-mail])]

No recent crashes.