626694 - [OGL] Inverse website colours

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Aaron Train [:aaronmt]]

Mozilla/5.0 (Android; Linux armv7l; rv:2.0b10pre) Gecko/20110118 Firefox/4.0b10pre Fennec/4.0b4pre ID:20110118042354

Seen this happen on:

http://m.digg.com
http://pastebin.com
http://news.ycombinator.com
http://m.cnn.com

Device: HTC Google Nexus One

Prefs:
layers.acceleration.disabled, false
layers.accelerate-all, true

Comment 1

[Jeff Muizelaar [:jrmuizel]]

I can also see this.

Comment 2

[Jeff Muizelaar [:jrmuizel]]

It looks like R and B are swapped for some reason.

Comment 3

[Jeff Muizelaar [:jrmuizel]]

This happens on my Galaxy S too. It must have regressed in the last little bit.

Comment 4

[Jeff Muizelaar [:jrmuizel]]

Romaxa, do you think the EGL changes broke this?

Comment 5

[Oleg Romashin (:romaxa)]

hmmm.. I see it on my Galaxy Tab too.. I'll check it tomorrow.

Comment 6

[Oleg Romashin (:romaxa)]

with backing surface path it render fine, but without backing surface it swap colors also on maemo6.

Here we are expecting BGRALayerProgramType always
http://mxr.mozilla.org/mozilla-central/source/gfx/thebes/GLContextProviderEGL.cpp#1029

and color swap happening because we are changing mShaderType here: http://mxr.mozilla.org/mozilla-central/source/gfx/thebes/GLContext.cpp#1358
and the rest of code seems does not care about that and some messup happening around

Comment 7

[Jeff Muizelaar [:jrmuizel]]

Looks like this was caused by the component alpha changes:

      // Note BGR: Cairo's image surfaces are always in what
      // OpenGL and our shaders consider BGR format.
      ColorTextureLayerProgram *basicProgram =
        aManager->GetBasicLayerProgram(mLayer->CanUseOpaqueSurface(),
                                       !mTexImage->IsRGB());

The comment is wrong, and my guess is that mTexImage doesn't have the right IsRGB().

Comment 8

[Oleg Romashin (:romaxa)]

sure, you right, IsRGB I guess already dead after recent changes related to mShaderType...

Comment 9

[Oleg Romashin (:romaxa)]

And reason why it works with backing surface is CreateBackingSurface mIsRGBFormat = PR_TRUE.

Comment 10

[Matt Woodrow (:mattwoodrow)]

Attachment: Possible Fix

I haven't actually tested this other than confirm that it compiles and doesn't horribly break anything on mac.

Romaxa: Can you please check if this fixes the issue on maemo and that I haven't accidentally broken anything else in the EGL provider.

Comment 11

[Ali Juma [:ajuma]]

This does not seem to be happening anymore. (Testing on a Nexus S, it was fixed sometime between the nightly builds of April 8 and April 10.) Any objections to marking this Resolved?

Comment 12

[Benoit Girard (:BenWa)]

Matt the issue doesn't seem to reproduce. Are the changes still needed?

Comment 13

[Benoit Girard (:BenWa)]

Taking the bug so it has an owner.

Comment 14

[Benoit Girard (:BenWa)]

This may be related to bug 661002.

Comment 15

[Matt Woodrow (:mattwoodrow)]

Removing the IsRGB() code paths still seem worthwhile as it's been replaced by GetShaderProgramType().

I'm not sure about the other hunks, probably not if the bug no longer happens.

Comment 16

[Benoit Girard (:BenWa)]

Attachment: Remove IsRGB

Comment 17

[Matt Woodrow (:mattwoodrow)]

I think you may forgotten to qref this patch BenWa. You can definitely have an r+ in it's current form though.

Comment 18

[Benoit Girard (:BenWa)]

Attachment: Remove IsRGB

Opps!

Comment 19

[Benoit Girard (:BenWa)]

Attachment: Remove IsRGB

Sorry, trying to work on too many patches at once :(

Comment 20

[Matt Woodrow (:mattwoodrow)]

Comment on attachment 566320 [details] [diff] [review]
Remove IsRGB

Review of attachment 566320 [details] [diff] [review]:
-----------------------------------------------------------------

Is that assertion really the only code reading IsRGB()? r+++++

Comment 21

[Marco Bonardo [:mak]]

https://hg.mozilla.org/mozilla-central/rev/38789564d25b

Comment 22

[Marco Bonardo [:mak]]

https://hg.mozilla.org/mozilla-central/rev/6b0b58d5da86