1224974 - widgets don't paint correctly after mouseovers on GTK with non-compositing WM

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Nathan Froyd [:froydnj]]

STR:

1. Go to any site (I used bugzilla.mozilla.org for an https site and amazon.com for an http site)
2. Click on the circled "i" in the address bar.
3. Move mouse over the ">" button for displaying further information.

Expected result:

The button highlights.

Actual result:

The button completely disappears and one can see the address bar, bookmark bar, etc. beneath it.

Clicking on the button does produce the expected result (information about the security of the page is displayed, etc.), but it's extremely surprising to have it disappear.  I occasionally see the globe icon and the lock icon in the info box disappear when I mouse into the box, but I can't get that to reproduce reliably currently.

Comment 1

[Robert O'Callahan (:roc) (email my personal email if necessary)]

I don't see this on Linux in 7cd2d806bd06 (m-c today).

Comment 2

[Nathan Froyd [:froydnj]]

Somewhere in https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=cc473fe5dc512c450634506f68cbacfb40a06a23&tochange=3cc3b1968524248450c465c4ea2ee5596ffa65f2 things go badly wrong for me.  The 11 Nov nightly does not exhibit the problem and the 12 Nov nightly does.

Comment 3

[Nathan Froyd [:froydnj]]

I also see this with the hamburger menu: opening it and mousing over the menu icons cause them to no longer be displayed.

Comment 4

[Nathan Froyd [:froydnj]]

This appears to be the fault of bug 1210560.

Comment 5

[Nathan Froyd [:froydnj]]

[Tracking Requested - why for this release]: Poor Linux user experience (possibly only limited to some versions of GTK?)

Comment 6

[Nathan Froyd [:froydnj]]

Bas, any ideas why this might be?

Comment 8

[Bas Schouten (:bas.schouten)]

One would think this is a sort of bug in the linux cairo backend or something like that considering it's restricted to linux. I'll ask Lee to take a look at this later today.

Comment 10

[Lee Salzman [:lsalzman]]

I couldn't manage to reproduce this locally.

Can we get some more info on this? Graphics section of about:support, OS/version, and GTK3 version would be helpful to know so we can try to identify any causative factors.

Comment 11

[Nathan Froyd [:froydnj]]

about:support graphics:

Adapter Description	X.Org -- Gallium 0.4 on AMD PITCAIRN
Asynchronous Pan/Zoom	wheel input enabled
Device ID	Gallium 0.4 on AMD PITCAIRN
Driver Version	3.0 Mesa 10.3.2
GPU Accelerated Windows	0/4 Basic (OMTC)
Supports Hardware H264 Decoding	No;
Vendor ID	X.Org
WebGL Renderer	X.Org -- Gallium 0.4 on AMD PITCAIRN
windowLayerManagerRemote	true
AzureCanvasBackend	cairo
AzureContentBackend	cairo
AzureFallbackCanvasBackend	none
AzureSkiaAccelerated	0
CairoUseXRender	1

This is Ubuntu 15.04...I think?  Maybe 14.10.

GTK+3 version is 3.12.2-0-ubuntu15.2.

This also happens on a different machine with GTK+3 3.14.5-1+deb8u1.  about:support for that machine:

Adapter Description	VMware, Inc. -- Gallium 0.4 on llvmpipe (LLVM 3.5, 256 bits)
Asynchronous Pan/Zoom	none
Device ID	Gallium 0.4 on llvmpipe (LLVM 3.5, 256 bits)
Driver Version	3.0 Mesa 10.3.2
GPU Accelerated Windows	0/1 Basic (OMTC)
Supports Hardware H264 Decoding	No;
Vendor ID	VMware, Inc.
WebGL Renderer	VMware, Inc. -- Gallium 0.4 on llvmpipe (LLVM 3.5, 256 bits)
windowLayerManagerRemote	true
AzureCanvasBackend	cairo
AzureContentBackend	cairo
AzureFallbackCanvasBackend	none
AzureSkiaAccelerated	0
CairoUseXRender	1

(That one is running over an SSH connection; not sure if that makes any difference.)

Bug 1225719 comment 4 says you need a non-compositing window manager to reproduce.

Comment 12

[Lee Salzman [:lsalzman]]

While the initial STR didn't work for me, I did manage to reproduce it via the hamburger menu. Investigating it...

Comment 13

[Lee Salzman [:lsalzman]]

Attachment: fix GTK shaped window drawing

The problem was introduced in a regression in bug 1210560 patch 7 that modified widget/gtk/nsWindow.

nsWindow::UpdateAlpha was rendering into the temporary image buffer at an offset, when it should have just been rendering into it at the origin. This patch fixes that.

I also noticed that the gfxContext constructor doesn't initialize the DrawTarget's matrix with the GetDTTransform() which is necessary to utilize device offset passed in for shaped windows. It was just initializing with identity instead. I changed that to use the DT transform instead now.

There are also some incidental cleanups I batched in here that I did while I was trying to track down the problem.

Comment 14

[Bas Schouten (:bas.schouten)]

Comment on attachment 8689750 [details] [diff] [review]
fix GTK shaped window drawing

Review of attachment 8689750 [details] [diff] [review]:
-----------------------------------------------------------------

::: gfx/thebes/gfxContext.cpp
@@ +81,5 @@
>  
>    mStateStack.SetLength(1);
>    CurrentState().drawTarget = mDT;
>    CurrentState().deviceOffset = aDeviceOffset;
> +  mDT->SetTransform(GetDTTransform());

Funny how apparently never before anyone used an offset context before changing the transform :-). Nice catch!

::: widget/gtk/nsWindow.cpp
@@ +2232,5 @@
>          // call UpdateTranslucentWindowAlpha once. After we have dropped
>          // support for non-Thebes graphics, UpdateTranslucentWindowAlpha will be
>          // our private interface so we can rework things to avoid this.
>          boundsRect = region.GetBounds();
> +        dt->PushClipRect(Rect(boundsRect));

ooh right.. that works now, nice!

Comment 15

[Lee Salzman [:lsalzman]]

Try results: https://treeherder.mozilla.org/#/jobs?repo=try&revision=13914471eacc

Comment 16

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/1ca5537062ab

Comment 17

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

https://hg.mozilla.org/mozilla-central/rev/1ca5537062ab

Comment 18

[Chris AtLee [:catlee]]

Confirmed fixed on 2015-11-25's nightly build.