614832 - Changing orientation causes fennec to stop drawing with opengl

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Comment 1

[Jeff Muizelaar [:jrmuizel]]

There are two possible approaches that I see for this:
 1. On resize throw away our layer manager and opengl context. It will be recreated for the new buffer.
 2. Use the new buffer with the existing gl context.

I have prototypes of both approaches, and think I prefer the second one.

Comment 2

[Jeff Muizelaar [:jrmuizel]]

Attachment: Basic easy implementation

Comment 3

[Jeff Muizelaar [:jrmuizel]]

Attachment: More involved version that just changes the target buffer

Comment 4

[Matt Woodrow (:mattwoodrow)]

Both of these patches break text entry for me.

When typing into the address bar no letters show up, it does still work correctly though.

I think this has to do with the second window triggering a resize and being reset as well. When my onscreen keyboard appears I get a resize for an area half the size of the screen (debugging using glClear to bright green :)

Comment 5

[Matt Woodrow (:mattwoodrow)]

It also makes the phone respond to input slowly. Haven't checked the CPU usage, will try do that in the morning

Comment 6

[Jeff Muizelaar [:jrmuizel]]

Attachment: cleaned up version of the second approach

This version does the resize on draw, perhaps it fixes the text input problem.

Comment 7

[Jeff Muizelaar [:jrmuizel]]

My guess is that when we were ignoring the resizes earlier we would get lucky and our buffer would still be valid.

Comment 8

[Jeff Muizelaar [:jrmuizel]]

During orientation change we get SurfaceCreated/SurfaceDestroyed events, we should probably use those to renew the EGLSurface instead of hanging off of resize. That way we won't interfere with the text input resizing that happens.

Comment 9

[Jeff Muizelaar [:jrmuizel]]

Attachment: This should work better

Comment 10

[Matt Woodrow (:mattwoodrow)]

Comment on attachment 495966 [details] [diff] [review]
This should work better

This fix this and bug 613002 for me.

Comment 11

[Jeff Muizelaar [:jrmuizel]]

Attachment: refactor EGL surface creation

This factors CreateSurfaceForWindow() out of CreateForWindow() so that we can reuse it in RenewSurface(). For this to work we also need to factor out CreateConfig(). The QT implementation is also pulled out because it didn't share any code.

Comment 12

[Jeff Muizelaar [:jrmuizel]]

Attachment: Renew the surface after it's been destroyed

Uses the new RenewSurface() to renew the surface when it is destroyed.

Comment 14

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Comment on attachment 496237 [details] [diff] [review]
Renew the surface after it's been destroyed

looks fine I think, maybe sans printfs?

Comment 15

[Jeff Muizelaar [:jrmuizel]]

Attachment: Actual cleaned up version I meant to upload

Comment 16

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Comment on attachment 496560 [details] [diff] [review]
Actual cleaned up version I meant to upload

even better!

Comment 17

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Comment on attachment 496233 [details] [diff] [review]
refactor EGL surface creation

some printfs/#if 0s etc in this one also?

Comment 18

[Jeff Muizelaar [:jrmuizel]]

Attachment: refactor EGL surface creation v2

Indeed, that was the wrong patch too. Sorry about the mess ups.

Comment 19

[Jeff Muizelaar [:jrmuizel]]

http://hg.mozilla.org/mozilla-central/rev/4b233c7bff2c
http://hg.mozilla.org/mozilla-central/rev/783e6a4850dd