1222990 - Transparent canvas makes transparent element underneath disappear

Status: RESOLVED FIXED

(Core :: Graphics: Layers, defect)

Description

[Paul Adenot (:padenot)]

STR:
- Download and unzip the test case
- Open index.html from the filesystem
- Click the "Do it" button

Expected:
- The semi-transparent canvas is shown, the text is visible

Actual:
- The canvas is shown, the text is hidden

Activating layers border with the pref:
- In Firefox 41 there is two layers, one for the text, one for the canvas. This seems expected.
- In Firefox 45, there is only a layer for the canvas, and the text disappears.

Talking to nical, he told me to activate border layers, and told me there should be a layer for the text and the canvas.

Additionally, looking at the bottom of `js/talk.js`, there is a line that can be commented. It was the thing that triggered the issue for me. This is originally a slidedeck written with reveal.js, so there are (potentially 3d) transforms everywhere.

Comment 1

[Paul Adenot (:padenot)]

Attachment: testcase.zip

Simply unzip and open index.html from the filesystem. The interesting bits are in `index.html` and `js/talk.js`. I can repro on OSX and Linux.

Comment 2

[u279076]

[Tracking Requested - why for this release]: new regression introduced in Firefox 43 by bug 1097464.

mozilla-central
===============
Last good revision: 11dc79e232110ba6de5179e46dfbda77b52a88c3 (2015-09-18)
First bad revision: 4313752f69956ae248bd4e7ff3913c8dd4252698 (2015-09-19)
Pushlog:
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=11dc79e232110ba6de5179e46dfbda77b52a88c3&tochange=4313752f69956ae248bd4e7ff3913c8dd4252698

mozilla-inbound
===============
Last good revision: 7641104770a80015e63597b58cb312fefcbd9ab4
First bad revision: 621ab19e86db28c38bbbf9119fbf6831ea344c54
Pushlog:
https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=7641104770a80015e63597b58cb312fefcbd9ab4&tochange=621ab19e86db28c38bbbf9119fbf6831ea344c54

Comment 3

[Liz Henry (:lizzard) (relman/hg->git project)]

Tracking since this is a recent regression.

Comment 4

[Thinker Li [:sinker]]

Attachment: Skip visible checking for Extend3D frames during building layers

Comment 5

[Thinker Li [:sinker]]

Hi Paul, could you check this patch if it fixes the problem?

Comment 6

[Paul Adenot (:padenot)]

(In reply to Thinker Li [:sinker] from comment #5)
> Hi Paul, could you check this patch if it fixes the problem?

It does not, and it kind of makes thing worse, it invalidates even less.

Comment 7

[Thinker Li [:sinker]]

Attachment: layerscope.zip

Paul, this is layerscope file that I have got on my linux machine.  It works for me.  Could you capture a layerscope file on you machine too?

Comment 8

[Paul Adenot (:padenot)]

Attachment: layerscope of the issue

Here is a layerscope, I think the issue is repro-ed on the last couple frames.

I had more luck reproducing this issue when`layers.force-enabled` is not set to true, although I've been able to rarely repro with it set to true.

Comment 9

[Liz Henry (:lizzard) (relman/hg->git project)]

Too late to make it into 43 as we are now heading into beta 9.

Comment 10

[Thinker Li [:sinker]]

Paul, I think the patch at bug 1230072.  Please try it!

Comment 11

[Paul Adenot (:padenot)]

I think it's good now, thanks !

Comment 12

[Thinker Li [:sinker]]

Paul had confirmed bug 1230072 fixes the problem.

Comment 13

[Ritu Kothari (:ritu) (Inactive, please n-i to RyanVM, jcristau, or pascal)]

From this point forward in Beta44 cycle (a week away from RC), we are only accepting fixes for critical recent regressions, sec and stability fixes. This issue while valid does not meet that bar. wontfix for Fx44.

Comment 14

[Ryan VanderMeulen [:RyanVM]]

The change that caused this regression was backed out from Firefox 45. A current Developer Edition build should now work correctly, or you can wait until 45 goes to Beta some time next week.

https://hg.mozilla.org/releases/mozilla-aurora/rev/64ec448f156d

That said, the testcase still appears to reproduce the original problem for me on a current Nightly build, so I'm reopening it.

Comment 15

[Thinker Li [:sinker]]

See comment 10~12.

Ryan, do you mean that bug 1230072 is not working for the problem here?

Comment 16

[Ryan VanderMeulen [:RyanVM]]

Yes, I'm saying that the testcase is opaque on a current m-c build, but properly transparent on Aurora. I tested the nightly build immediately after bug 1230780 landed (since that's what bug 1230072 got duped to) and it still reproduces.

Maybe the patch from bug 1230072 is still needed?

Comment 17

[Thinker Li [:sinker]]

Attachment: extend3d-range.diff

Comment 18

[Thinker Li [:sinker]]

(In reply to Thinker Li [:sinker] from comment #17)
> Created attachment 8714227 [details] [diff] [review]
> extend3d-range.diff

Ryan, this patch should fix the problem.  But, bug 1229317 would cover the change of this patch too.  So, if you can confirm this patch fix the problem, then this bug can depend on bug 1229317.

Comment 19

[Ryan VanderMeulen [:RyanVM]]

Yes, this patch fixes the attached testcase.

Comment 20

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

Is this fixed now, with bug 1229317, or do we need additional code?  Bug 1229317 landed on central on 2/7 and got uplifted to aurora 46 on 2/26.

Comment 21

[Ryan VanderMeulen [:RyanVM]]

Yes, this testcase works properly on 46 and 47 now.