1134027 - Flash HW Accelerated 3D Graphics (Stage3D) content no longer renders in Firefox 34 and higher on MacOS

Status: RESOLVED INCOMPLETE

(Core Graveyard :: Plug-ins, defect)

Description

[Jeromie Clark]

Attachment: Firefox_Stage3D_Content.jpg

User Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_9_5) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/40.0.2214.111 Safari/537.36

Steps to reproduce:

Go Here


Actual results:

Only a grey rectangle is displayed


Expected results:

Flash content should have rendered

Comment 1

[Jeromie Clark]

Attachment: Flash content that doesn't use the GPU renders fine

Comment 2

[Jeromie Clark]

Attachment: Flash Stage3D content in Chrome renders as expected

Comment 3

[Loic]

Does it work if you disable HWA in Flash?
http://www.macromedia.com/support/documentation/en/flashplayer/help/help01.html

If that doesn't change anything, could you install mozregression to find a regression range, please.
See http://mozilla.github.io/mozregression/ for details (no need to bisect, only changelog is enough).

Comment 4

[Jeromie Clark]

Yes, disabling hardware acceleration in Flash causes the content to render in Firefox 34 or higher.  

In the Firefox ESR (and all versions below the 34 branch), content renders as expected with hardware acceleration enabled in Flash.

Comment 5

[Benjamin Smedberg]

I'd love to see a nightly regression range.

Comment 6

[Steven Michaud [:smichaud] (Retired)]

I just tested in FF 35.0.1 with Flash's hardware acceleration on in OS X 10.7.5 and 10.8.5 (on two different machines).  I didn't see this bug in either test.

So it looks like there are some additional, unknown factors here -- possibly including the version of OS X.

My 10.8.5 doesn't have a Retina display.  My 10.7.5 machine does.

Comment 7

[Chris Peterson [:cpeterson]]

I bisected this regression to Nightly 34 build 2014-07-31:

https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=f61a27b00e05&tochange=005424a764da

We don't have archived inbound builds that old, so mozregression can't bisect any further. I am testing Flash 16.0.0.305 on a MacBook Pro running OS X 10.10.2.

Comment 8

[Benjamin Smedberg]

Suspects, although neither jumps out at me:
70935e906273	Ralph Giles — Bug 1045128 - Bump MacOS X SDK to 10.7 for release builds. r=ted
477ba7091557	Jeff Gilbert — Bug 1045957 - Improve code in GLReadTexImageHelper.cpp. - r=mattwoodrow

Comment 9

[Ralph Giles (:rillian)]

Hmm. If it's my sdk bump you could try testing a build of the revision right before that landed. SDK build skew seems unlikely though.

Comment 10

[Jeromie Clark]

You're right.  I'm not seeing this on my non-retina iMac running 10.10.1 and Firefox 35.0.1 with Flash Player 16.0.0.305.

Comment 11

[Steven Michaud [:smichaud] (Retired)]

Testing with FF 35.0.1 and recent mozilla-central nightlies, I see this bug on OS X 10.9.5 and 10.10.2, but only when HiDPI support is turned on.  One way to turn it off is right-click on a distro, choose Get Info, then check "Open in Low Resolution".

I'm unable to test on OS X 10.8.5 with a Retina display.  But as I mentioned above (in comment #6) I don't see this bug on OS X 10.7.5, even with a Retina display and HiDPI support on.

The trigger for this bug is https://hg.mozilla.org/mozilla-central/rev/70935e906273

 Bug 1045128 - Bump MacOS X SDK to 10.7 for release builds. r=ted
author	Ralph Giles <[e-mail]>
	Mon, 28 Jul 2014 10:36:52 -0700

I tested by doing "hg update -r 042fa33c3f5c" (the rev just before 70935e906273) and building twice (on OS X 10.7.5) -- once with the 10.6 SDK and once with the 10.7 SDK.  When run on OS X 10.9.5, the latter build (made with the 10.7 SDK) had this bug, but not the former (made with the 10.6 SDK).

It will be very difficult to figure out why the SDK with which you build Firefox makes a difference here, and probably impossible for someone outside Adobe.  Surely the crucial factor is something in Flash's internal workings while Flash-specific HW acceleration is on.  Though it's also very likely that there may be an Apple bug here -- something Adobe will need to learn how to work around.

Furthermore, I think it's very unlikely we'll be able to go back to using the 10.6 SDK.

Sorry for the bad news.  I doubt there's anything Mozilla can do here.

Comment 12

[Jeromie Clark]

No problem.  I didn't catch earlier in the thread that this was related to the Mac SDK.  Thanks for the clarification.  We'll chase it down.

Comment 13

[Jeromie Clark]

This is ADBE 3937398.  We currently still support OSX 10.6 so we have a dependency there, but we're looking at dealing with this in either the May or June release since that population has tailed off significantly at this point in favor of Yosemite.

Comment 14

[Steven Michaud [:smichaud] (Retired)]

> We currently still support OSX 10.6 so we have a dependency there

Do you mean that Adobe's code currently links to the 10.6 SDK?  And if so, can you "fix" this bug (and/or bug 1146036) by making Adobe's code link to the 10.7 SDK (or some other higher version)?

Comment 15

[Jeromie Clark]

That's what we're looking at.  The concern is that changing SDK versions has implications for multiple browsers on Mac.  We're currently investigating a change to the 10.7 SDK, and that work is targeted for our May release.