:gerard-majax [PTO 01/08/2023-10/09/2023] Reporter
	Description • 9 years ago

I'm pretty confident this is a regression after bug 1175463 landed: I had a build done just before it landed and this behavior was not there. Then I did a new build with updated hardware/qcom/display and the issue happens.

Basically, upon loading of apps (any, but I saw it quite often with at least Email and Twitter), you get corrupted display where the background of the app is replaced with the homscreen layer that contains apps icons and their names. The app itself is properly usable except it's totally unreadable.

Often, focusing a text field or switching apps will trigger a proper redraw.

STR:
 0. Launch app

Expected:
 Proper content

Actual:
 Sometimes, quite often indeed but I don't have figures, we get the layer with homescreen icons and app names instead of the content of the app

	:gerard-majax [PTO 01/08/2023-10/09/2023] Reporter
	Comment 1 • 9 years ago

I'll try to get a screenshot next time I reproduce.

	Sotaro Ikeda [:sotaro] Assignee
	Comment 2 • 9 years ago

:gerard-majax, can you answer the questions?
- Is Hardware composer composition enabled on your device?
- Do you flash also kernel?

	:gerard-majax [PTO 01/08/2023-10/09/2023] Reporter
	Comment 3 • 9 years ago

(In reply to Sotaro Ikeda [:sotaro] from comment #2)
> :gerard-majax, can you answer the questions?
> - Is Hardware composer composition enabled on your device?

Yes

> - Do you flash also kernel?

Yes. I do flash absolutely everything.

	Sotaro Ikeda [:sotaro] Assignee
	Comment 4 • 9 years ago

:gerard-majax, when the problem happens, does the app have scrollable frame? When a scrollable frame exist, HWC composition does not happen, instead only OpenGL composition happens.

	:gerard-majax [PTO 01/08/2023-10/09/2023] Reporter
	Comment 5 • 9 years ago

I don't know. How can I know ?

	Sotaro Ikeda [:sotaro] Assignee
	Comment 6 • 9 years ago

(In reply to Alexandre LISSY :gerard-majax from comment #5)
> I don't know. How can I know ?

If you can scroll or pan/zoom in application, an application normally have a scrollable layer.

Another way is, enabling "Draw Layer Borders" and "Draw Tile Borders" in "Developer" in Setting app. It draws layer borders and tile borders. Scrollable frame is rendered as tiled layer. Then you can see the tiles' borders as green line if the application have a tiled layer. When you see the tiles' border, hwc is not used for composition. When you see layer border, hwc is not used for composition except video(Yuv) rendering.

	Sotaro Ikeda [:sotaro] Assignee
	Comment 7 • 9 years ago

On latest master, all application's layer seems not rendered as full hwc composition on Z3C on my side. Therefore, the problem seems not related hwc's composition.

	:gerard-majax [PTO 01/08/2023-10/09/2023] Reporter
	Comment 8 • 9 years ago

Reproduced consistently:
 0. Open Email client, go to a folder
 1. Tap the "Compose" button

Sadly, a screenshot do not show it.

	Sotaro Ikeda [:sotaro] Assignee
	Comment 10 • 9 years ago

I could also reproduce the problem on latest master aries. The problem happens only when hwc composition is enabled. Therefore hwc seems to be related to the problem. But bug 1175463 is not related to the problem. Even when rollbacked the change of bug 1175463, the problem did not fix. And the changed code was not used when the problem happens.

	Sotaro Ikeda [:sotaro] Assignee
	Comment 11 • 9 years ago

Remove dependency to 1175463 from comment 10.

	Sotaro Ikeda [:sotaro] Assignee
	Comment 12 • 9 years ago

When the problem happens hwc always did partial OVERLAY composition(GL and hwc mixed composition).

	Sotaro Ikeda [:sotaro] Assignee
	Comment 13 • 9 years ago

The following code in HwcComposer2D::TryHwComposition() does not work as expected.

                    // HWC will compose HWC_OVERLAY layers in partial
>                    // Overlay Composition, set layer composition flag
>                    // on mapped LayerComposite to skip GPU composition
>                    mHwcLayerMap[k]->SetLayerComposited(true);
>                    if ((mList->hwLayers[k].hints & HWC_HINT_CLEAR_FB) &&
>                        (mList->hwLayers[k].blending == HWC_BLENDING_NONE)) {
>                        // Clear visible rect on FB with transparent pixels.
>                        hwc_rect_t r = mList->hwLayers[k].displayFrame;
>                        mHwcLayerMap[k]->SetClearRect(nsIntRect(r.left, r.top,
>                                                                r.right - r.left,
>                                                                r.bottom - r.top));
>                    }

https://dxr.mozilla.org/mozilla-central/source/widget/gonk/HwcComposer2D.cpp#688

	Sotaro Ikeda [:sotaro] Assignee
	Comment 14 • 9 years ago

This seems like a regression of Bug 1073252.

	Sotaro Ikeda [:sotaro] Assignee
	Comment 15 • 9 years ago

The patch seems to fix the problem.

	Sotaro Ikeda [:sotaro] Assignee
	Comment 16 • 9 years ago

Comment on attachment 8640761 [details] [diff] [review]
patch - Fix LayerComposite::SetClearRect() calling

:gerard-majax, can you try if the patch fix the problem?

	Sotaro Ikeda [:sotaro] Assignee
	Comment 17 • 9 years ago

Remove unnecessary check.

	Sotaro Ikeda [:sotaro] Assignee
	Comment 18 • 9 years ago

Comment on attachment 8640802 [details] [diff] [review]
patch - Fix LayerComposite::SetClearRect() calling

:gerard-majax, can you try if the patch fix the problem?

	Sotaro Ikeda [:sotaro] Assignee
	Comment 19 • 9 years ago

Comment on attachment 8640802 [details] [diff] [review]
patch - Fix LayerComposite::SetClearRect() calling

Oh, I updated incorrect patch.

	Sotaro Ikeda [:sotaro] Assignee
	Comment 20 • 9 years ago

Correct patch.

	:gerard-majax [PTO 01/08/2023-10/09/2023] Reporter
	Comment 21 • 9 years ago

Comment on attachment 8640805 [details] [diff] [review]
patch - Fix LayerComposite::SetClearRect() calling

I'm not able to ever reproduce once this is applied, thanks!

	Sotaro Ikeda [:sotaro] Assignee
	Comment 22 • 9 years ago

(In reply to Alexandre LISSY :gerard-majax from comment #21)
> Comment on attachment 8640805 [details] [diff] [review]
> patch - Fix LayerComposite::SetClearRect() calling
> 
> I'm not able to ever reproduce once this is applied, thanks!

Good. Thanks for the confirmation!

	Pulsebot
	Comment 23 • 9 years ago

https://hg.mozilla.org/integration/mozilla-inbound/rev/0bd29e2ee5fa

	Carsten Book [:Tomcat]
	Comment 24 • 9 years ago

https://hg.mozilla.org/mozilla-central/rev/0bd29e2ee5fa