Bug 237230
| Summary: | REGRESSION (iOS 15.4 beta) - WebGL race condition between clearing depth buffer and switching framebuffers | ||
|---|---|---|---|
| Product: | WebKit | Reporter: | Christoph Bartschat <christoph> |
| Component: | WebGL | Assignee: | Nobody <webkit-unassigned> |
| Status: | RESOLVED DUPLICATE | ||
| Severity: | Normal | CC: | bfulgham, dino, johncunningham, kbr, kkinnunen, kpiddington, webkit-bug-importer |
| Priority: | P2 | Keywords: | InRadar |
| Version: | Safari 15 | ||
| Hardware: | iPhone / iPad | ||
| OS: | iOS 15 | ||
| Bug Depends on: | 237113 | ||
| Bug Blocks: | |||
Christoph Bartschat
This is a regression between iOS 15.3.1 and 15.4.
As can be see in this demo: https://codepen.io/cmb8/pen/GROXmKp
We should see two gears spinning normally, but instead we get visual artifacts that obscure the objects from getting properly rendered.
This behavior is caused by calling:
context.clear(context.DEPTH_BUFFER_BIT)
then immediately calling:
context.bindFramebuffer(...)
If we add a finish() call between the two calls, the issue does not occur.
This problem affects many 8th Wall projects that use three.js, like
our sample project here: https://8thwall.8thwall.app/unitcube-threejs/
or more recently, this Super Bowl promotional tie-in from Kia: https://www.kiadogmentedreality.com
Disabling "WebGL via Metal" in "Experimental WebKit Features" and relaunching Safari fixes the behavior.
Video of expected behavior: https://www.youtube.com/watch?v=SAp_RXDD7_0
Video of unexpected behavior: https://www.youtube.com/watch?v=d8cr-LP40hs
| Attachments | ||
|---|---|---|
| Add attachment proposed patch, testcase, etc. |
Kenneth Russell
+Kyle and John in case they can help triage this. Wonder whether it was fixed in the most recent ANGLE roll.
Kimmo Kinnunen
Thanks for the report.
I think this has the same root cause as bug 237113.
In order to workaround, you can try if you can rework the rendering so that it does not need the `autoClear = false`. I.e. use `autoClear = true`.
Leaving this in until fix for bug 237113 is testable.
Radar WebKit Bug Importer
<rdar://problem/89839163>
Christoph Bartschat
Now that bug 237113 has been closed, do we have confirmation that the problem described here has also been fixed?
Brent Fulgham
*** This bug has been marked as a duplicate of bug 237113 ***