958025 - No window controls and decorations / nsChildView::DrawWindowOverlay doesn't draw anything / square window corners / missing window buttons and fullscreen button

Status: RESOLVED FIXED

(Core :: Widget: Cocoa, defect)

Description

[Markus Stange [:mstange]]

Starting with today's OS X Nightly (20140109), windows look very strange: They're missing their window buttons and have square corners.

This seems to be a regression from bug 956401. Changing the ScopedScissorRect line in nsChildView::DrawWindowOverlay to this fixes it:
ScopedScissorRect scopedScissorRectState(gl, aRect.x, aRect.y, aRect.width, aRect.height);

But we should find out why a wrong scissor rect is set when this function is called and fix that instead.

Comment 1

[Markus Stange [:mstange]]

Attachment: screenshot

Comment 2

[Markus Stange [:mstange]]

When nsChildView::DrawWindowOverlay is called, there's no ScopedScissorRect on the stack, so the active scissor rect is the one that was read from the context during the GLContext initialization, and that's 0,0,0,0.

Comment 3

[Markus Stange [:mstange]]

Attachment: disable scissor test in DrawWindowOverlay

This seems like the simplest solution. We don't need a scissor rect here.
The only reason I can see for wanting a scissor rect here in the future is if we want to recomposite only parts of the window for performance reasons, but then we'll have to revisit our scissor rect setup anyway.

Comment 6

[Benoit Jacob [:bjacob] (mostly away)]

Comment on attachment 8357738 [details] [diff] [review]
disable scissor test in DrawWindowOverlay

Review of attachment 8357738 [details] [diff] [review]:
-----------------------------------------------------------------

Many thanks for debugging this.

From our IRC conversation, it seems that the problems are caused by Apple's OpenGL libraries having glGetIntegerv(SCISSOR_BOX) returning (0,0,0,0) in the initial state. That is not spec-compliant: see for example http://www.khronos.org/registry/gles/specs/2.0/es_full_spec_2.0.25.pdf Section 4.1.2 "Scissor Test" page 93, which says:

> In the initial state left = bottom = 0; width and height are determined by the size of the GL window.

Let's take your patch to fix the short-term emergency, so we can take our time to implement what I believe would be the best fix, namely: to work around specifically the above-mentioned Apple GL bug, so that we can rely on glGetIntegerv(SCISSOR_BOX) on Mac rather than having to live with this big caveat.

Comment 8

[Markus Stange [:mstange]]

Attachment: scoped scissor test enabling

Comment 9

[Markus Stange [:mstange]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/a4ce571996e4
https://hg.mozilla.org/integration/mozilla-inbound/rev/c952d0c14bb1

Comment 10

[Benoit Jacob [:bjacob] (mostly away)]

The rationale for this second patch is as follows:

BeginFrame was enabling the SCISSOR_TEST but EndFrame was not disabling it. That was bad. Since the only place in CompositorOGL that is setting and unsetting a scissor rect was DrawQuadInternal, that is also the place that should enable and disable the scissor test.

Comment 11

[Ryan VanderMeulen [:RyanVM]]

Markus landed the first patch on m-c as well so we can run new OSX nightlies.

https://hg.mozilla.org/mozilla-central/rev/effa95701c83

Comment 13

[Ryan VanderMeulen [:RyanVM]]

(In reply to Ryan VanderMeulen [:RyanVM UTC-5] from comment #11)
> Markus landed the first patch on m-c as well so we can run new OSX nightlies.
> 
> https://hg.mozilla.org/mozilla-central/rev/effa95701c83

The respin is finished.

Comment 14

[RobertJ]

Kudos to the team. Just updated the nightly and the window controls, decorations .... are back.

Comment 15

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/a4ce571996e4
https://hg.mozilla.org/mozilla-central/rev/c952d0c14bb1