899798 - [B2G] With Skia/GL, canvases get re-composited with B and R channels swapped

Status: RESOLVED FIXED

(Firefox OS Graveyard :: Hardware, defect)

Description

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

Attachment: testcase

Attached testcase just draws a red quad.

On B2G with Skia/GL (as enabled by attachment 782802 [details] [diff] [review]) this is first painted correctly, then (after half a second) re-painted in blue i.e. with R and B channels swapped.

This is in particular at least one of the problems happening in bug 899341.

Comment 1

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

Attachment: testcase updated to print the RGBA values from the canvas after it's been composited

This shows that the RGBA values read back from the canvas are correct (255, 0, 0, 255) even after the canvas has been re-rendered as blue. This means that the problem is purely a compositing problem, not an issue with the canvas' backbuffer actually getting modified.

Comment 2

[Peter Chang[:pchang]]

This issue was related to hwc composition which happened on leo and hamachi.
You can use below steps to disable HWC composition.

Right now we are waiting for bug 886415 to land.

1. adb shell mv /system/lib/hw/hwcomposer.msm7627a.so /system/lib/hw/hwcomposer.msm7627a.so_bak
2. adb shell stop b2g
3. adb shell start b2g

Comment 3

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

Peter, now that bug 886415 is fixed, what are the next steps here?

Comment 4

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

Milan, this might be the most 'open' of the remaining blockers for Skia/GL/B2G1.2. Need to figure whom this bug should be assigned to. I feel that I should finish work on the other blockers I'm already working on, before looking more into this one, so this might be better assigned to someone else than me.

Comment 5

[Milan Sreckovic [:milan] (needinfo for best results)]

Peter, any chance you can take a look at this?  It seems very related to the work you've previously done.

Comment 6

[Peter Chang[:pchang]]

I will import QCT release patch to B2G soon.

Comment 7

[Peter Chang[:pchang]]

Attachment: import RBSwap patch from CAF's patch queue

Comment 8

[Peter Chang[:pchang]]

Verified the imported patch with my leo device and didn't have problem.

Paste the pull request link.
https://github.com/mozilla-b2g/gonk-patches/pull/2/commits

Comment 9

[Peter Chang[:pchang]]

Comment on attachment 793824 [details] [diff] [review]
import RBSwap patch from CAF's patch queue

Michael, could you help me to grant my git pullrequest?

Comment 10

[Peter Chang[:pchang]]

Since I just imported the patch which got review+ on bug 881460 from QCT CAF, I think mwu's r+ is enough.

Comment 11

[Peter Chang[:pchang]]

Michael, do you know who can help to grant the git pullrequest?

Comment 12

[Michael Wu [:mwu]]

I'll do it on Monday if you can't find someone in the Taipei office with access to the repo.

Comment 13

[Peter Chang[:pchang]]

(In reply to Michael Wu [:mwu] from comment #12)
> I'll do it on Monday if you can't find someone in the Taipei office with
> access to the repo.

I see, I will find mchen in TPE office first.

Comment 14

[Peter Chang[:pchang]]

patch was merged, close case.

https://github.com/mozilla-b2g/gonk-patches

Comment 16

[lecky]

Hi Wayne :
 Please help us to push to merge this case into V1.1 HD. It blocked the Helix Device .If you can not resolve it , please get the authorization from carrier.

Comment 17

[Milan Sreckovic [:milan] (needinfo for best results)]

In case this is confusing others as it has me - even though this bug was specific to 1.2, as SkiaGL wasn't on 1.1 or earlier, the fix is the same as for the duplicated bug 915869, and that is the one that is 1.1 specific.

Comment 18

[Milan Sreckovic [:milan] (needinfo for best results)]

(In reply to Milan Sreckovic [:milan] from comment #17)
> ...
> for the duplicated bug 915869, and that is the one that is 1.1 specific.

That would be bug 906534 instead.