[android] Restore prior GLContext if exists #4716
Conversation
| @@ -196,6 +201,14 @@ void NativeMapView::render() { | |||
| } | |||
|
|
|||
| deactivate(); | |||
|
|
|||
| if (currentContext != context && currentContext != EGL_NO_CONTEXT) { | |||
There was a problem hiding this comment.
Why don't you just comment out the call to deactivate instead? It's equivalent.
There was a problem hiding this comment.
👍
But I think it is better deactivate() if there is no prior GLContext. Gonna change that.
tmpsantos
force-pushed
the
tmpsantos-android_not_rendering
branch
from
5f247c6
to
1fdc3aa
Compare
| throw std::runtime_error("eglMakeCurrent() failed"); | ||
| } | ||
| } else { | ||
| deactivate(); |
There was a problem hiding this comment.
For the record, on my Nexus 5x Android 6.0.1 deactivate() gets called but on my Nexus 5 Android 4.4.4 the context gets replaced with the prior context.
There was a problem hiding this comment.
@tmpsantos My guess is that on "Nexus 5 Android 4.4.4", either the View Framework or some other opengl view is changing the opengl state prior to call to NativeMapView::render. Is it possible to do the following experiment in your sandbox with the following two changes? If the 'screen is white bug' goes away, its an issue with not handling the opengl state correctly.
- Instantiate mbgl::Map with GLContextMode::Shared as the fourth argument here. This will ensure that gl state is reset every time by mbgl::Map.
- Call glViewPort unconditionally by commenting out if(sizeChanged) (The current code will consistently cause crashes/wrong behavior if there are multiple Opengl views in the same application. Its also missing the glError call that MBGL_ERROR_CHECK provides.).
There was a problem hiding this comment.
@tmpsantos My guess is that on "Nexus 5 Android 4.4.4", either the View Framework or some other opengl view is changing the opengl state prior to call to NativeMapView::render. Is it possible to do the following experiment in your sandbox with the following two changes? If the 'screen is white bug' goes away, its an issue with not handling the opengl state correctly.
Instantiate mbgl::Map with GLContextMode::Shared as the fourth argument here. This will ensure that gl state is reset every time by mbgl::Map.
Call glViewPort unconditionally by commenting out if(sizeChanged) (The current code will consistently cause crashes/wrong behavior if there are multiple Opengl views in the same application. Its also missing the glError call that MBGL_ERROR_CHECK provides.).
What really happens is there is another GL context in use that we cannot use/share because is not configured with the attributes we expect. GLContextMode::Shared didn't work in this case (I tested). We really have to use our own context and after we are done rendering, restore the old context that Android expects to be active.
@mb12 thanks for the feedback!
tmpsantos
force-pushed
the
tmpsantos-android_not_rendering
branch
from
1fdc3aa
to
9efa5f9
Compare
|
If saving and restoring the specific display, context, and surfaces is necessary, it seems like that should be part of |
|
@tmpsantos and @jfirebaugh Is there any plan to fix this? All this complexity will go away when this is fixed. |
I still want to work at this issue, but I don't have a timeline for that. Now that Android renders on the same thread as the mbgl::Map object that is feasible (we need to create the mbgl::Map on the GLSurfaceView.Renderer thread). The challenge is to proxy every getter and setter from the UIThread to this renderer thread, specially because some of the getters are synchronous. |
Apparently some versions of Android, notable 4.4.4 running on my Nexus 5, creates a GLContext on the Android UIThread which is the one we are now using for rendering after #2909. If the context is not restored, nothing gets rendered (or sometimes partially, or artifacts) because Android will try to do GL stuff on the context used by Mapbox GL and mess things up.
tmpsantos
force-pushed
the
tmpsantos-android_not_rendering
branch
from
9efa5f9
to
7d80690
Compare
👍 |
Apparently some versions of Android, notable 4.4.4 running on my Nexus 5, creates a GLContext on the Android UIThread which is the one we are now using for rendering after #2909.
If the context is not restored, nothing gets rendered (or sometimes partially, or artifacts) because Android will try to do GL stuff on the context used by Mapbox GL and mess things up.