mozilla :: #gfx

 
17 Oct 2017
13:25katskvark: is the line at https://github.com/servo/webrender/blob/master/webrender/src/device.rs#L710 necessary? the entire impl is wrapped in that anyway. it looks like maybe it was leftover from a previous version where you conditioned the function implementations rather than the whole impl
13:30katskvark: made a PR, feel free to comment there
15:09katsrhunt: https://github.com/servo/webrender/pull/1887/commits/845901543c6a98705f97569e56376836bf5f8f55 is the fix for the problem. basically having #[cfg(...)] on a statement inside a function was causing the problem
15:10rhuntkats: thanks
15:24jnicolsnorp: ping
15:54Aryx[17:36:08] Aryx: hi, there are so far 2 crashtest failures for the same test on autoland (more assertions than expected).
15:54Aryx[17:36:33] Aryx: none of the commits looks like an obvious cause for this
15:54Aryx[17:36:45] Aryx: But maybe you can identify it? https://treeherder.mozilla.org/logviewer.html#?job_id=137575193&repo=autoland
15:54Aryx[17:37:11] Aryx: ASSERTION: bad aListVisibleBounds: 'r.GetBounds().IsEqualInterior(aListVisibleBounds)', file /builds/worker/workspace/build/src/layout/painting/nsDisplayList.cpp, line 2039
15:54Aryx[17:37:40] Aryx: this test fails in both crashtest failures: https://dxr.mozilla.org/mozilla-central/source/docshell/base/crashtests/914521.html?q=path%3A914521.html&redirect_type=single
16:04katsAryx: i think that crashtest was just recently re-enabled
16:04katsAryx: it's possible that was premature
16:05Aryxi backed out 464450, seems to have started with that
16:05katsAryx: hm interesting
16:24snorpjnicol: pong
16:32jnicolsnorp: I think I need the function ANativeWindow_setBuffersTransform, which only exists in the newest ndk
16:33snorpjnicol: is that for setting the texture transform?
16:33jnicolsnorp: yes
16:33jnicolsnorp: the surfacetexture is returning identity on some devices and flip vertically on others
16:33snorpjnicol: wait it's just flags
16:34snorpoh
16:34snorpjnicol: yeah, it's very weird...
16:34jnicolsnorp: I'm assuming I can initialize it from the producer side with this function
16:34jnicolsnorp: though documentation is sparse
16:35snorpjnicol: you can dlopen it
16:36snorpjnicol: could add a wrapper in AndroidNativeWindow
16:36snorpjnicol: when was it added?
16:37snorpoh very recently apparently
16:38jnicolsnorp: I'm not sure. It's in ndk 15c, but it's only defined #if __ANDROID_API__ >= __ANDROID_API_O__
16:38jnicolso very recent
16:38snorpah you won&#39;t be able to even dlopen < O I guess
16:39jnicolBut the equivalent internal functions have been around a while I think
16:39snorppossibly
16:39snorpbut you can&#39;t dlopen private stuff anymore
16:39jnicolah
16:40snorpjnicol: I would say ignore the texture transform on ST we use for tiles
16:40jnicolwell I suppose I can add a flag to the SurfaceTextureHost telling it to ignore the SurfaceTexture.getTransformMatrix
16:40jnicolha exactly!
16:40snorp:)
16:40jnicolit&#39;s just a bit eww
16:40snorpit is indeed
16:40snorpjnicol: I saw the same weirdness when I implemented the ST blit stuf
16:41snorpwhich I am trying to land now
16:41snorpthe thing I have seems to work, but wouldn&#39;t be surprised if it&#39;s flipped for some people :/
16:43snorpjnicol: things are coming along, though!
16:43snorpjnicol: I hope?
16:46jnicolsnorp: think it&#39;s all working now. but the code needs improved
16:46snorpjnicol: very nice
17:13pulsebotCheck-in: https://hg.mozilla.org/projects/graphics/rev/902fb74ca400 - Miko Mynttinen - Convert WeakFrame to nsIFrame* earlier
18:03kvarkGankro: jrmuizel https://play.rust-lang.org/?gist=5721fcbef858d5346a84dd66155df3c8&version=stable
18:18jrmuizelkvark, Gankro: https://play.rust-lang.org/?gist=b28658218b617ac2ee6a8ba07f9eef1f&version=stable
18:39bdahlmstange: did you ever have any luck with https://bugzilla.mozilla.org/show_bug.cgi?id=1365844
18:39firebotBug 1365844 REOPENED, nobody@mozilla.org ASSERTION: Bounds computation mismatch: &#39;mContainerBounds.IsEqualInterior(mAccumulatedChildBounds)&#39;,
18:41mstangebdahl: Oh, I did not, and I completely forgot about the bug :(
18:42bdahlmstange: oh okay, i started getting this in headless mode on macos recently
18:42mstangebdahl: I see... on the same test?
18:43bdahlmstange: seems like startup of any tests and then throughout the test
18:44mstangebdahl: that&#39;s interesting! Can you file a new bug about that? It may have a different cause, and it seems higher priority.
18:46bdahlmstange: will do. i don&#39;t think it was happening before so i&#39;ll try to bisect, but unfortunately my icecc appears to no longer work after xcode 9
18:46bdahlso will be slow
18:50mstangebdahl: you can work around that by using an older macos SDK
18:50mstangeI thought I had written a note abou tthat on https://developer.mozilla.org/en-US/docs/Mozilla/Developer_guide/Using_Icecream but I can&#39;t find it
18:54bdahlmstange: is that just a mozconfig change?
18:54mstangebdahl: yes
18:55mstangebdahl: e.g. ac_add_options --with-macos-sdk=/Users/mstange/SDKs/MacOSX10.11.sdk
18:55mstangebdahl: the biggest problem is how to obtain the SDK
18:55mstangebdahl: you can usually get it from somebody who hasn&#39;t updated their XCode yet, at /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/
18:56bdahlk, thx!
18:57mstangeI remember now, https://bugzilla.mozilla.org/show_bug.cgi?id=1401667#c3 is where I put the workaround
18:57firebotBug 1401667 NEW, nobody@mozilla.org Xcode 9.0 - Foundation NSUUID.h doesn&#39;t compile
20:04mattwoodrowGankro: nsDisplayText::CreateWebRenderCommand doesnt appear to take mOpacity into account, is that a known issue?
20:05jrmuizelmstange: https://pastebin.mozilla.org/9070345
20:06jrmuizelmattwoodrow: I think it&#39;s supposed to
20:06jrmuizelmattwoodrow: Gankro knows
20:06Gankromattwoodrow: it should be passed in to the textRun drawing code
20:07Gankromattwoodrow: http://searchfox.org/mozilla-central/source/layout/generic/nsTextFrame.cpp#5206
20:07GankroThat bubbles out to us in the text color when we get called back in FillText
20:08Gankro*FillGlyphs
20:08mattwoodrowGankro: Hmm, indeed it does
20:08mattwoodrowI got backed out for causing this failure: https://treeherder.mozilla.org/logviewer.html#?job_id=137388953&repo=mozilla-inbound&lineNumber=3012
20:09mattwoodrowI think the test version is using nsDisplayOpacity, the ref is using mOpacity on nsDisplayText
20:09Gankromattwoodrow: it&#39;s likely a bug in webrender&#39;s blend mode selection
20:10Gankro(common)
20:10Gankromattwoodrow: text on linux has no AA, but there&#39;s a bunch of places where we see &quot;oh hey opacity, use greyscale AA&quot;
20:10gwmattwoodrow: agree with Gankro - likely WR bug - there&#39;s two alpha + subpixel bug fixes in WR master that haven&#39;t made it through to gecko yet
20:11mattwoodrowgw: Gankro: https://hg.mozilla.org/mozilla-central/raw-file/tip/layout/tools/reftest/reftest-analyzer.xhtml#logurl=https://queue.taskcluster.net/v1/task/X4Ku3KJ4QuW0sYdlIMUI2A/runs/0/artifacts/public/logs/live_backing.log&only_show_unexpected=1
20:11mattwoodrowThats the actual image, they are both greyscale, but quite different
20:12mattwoodrowAnyway, is it ok if I just mark it as fuzzy if-webrender?
20:12Gankromattwoodrow: mark it as fails, kats can revert it with next wr update
20:13mattwoodrowok
20:13Gankromattwoodrow: the reference in that link appears to have no AA at all (which is the correct result)
20:16mstangejrmuizel: https://pastebin.mozilla.org/9070347
20:45gwmstange: What am I supposed to do with those buzilla bugs that get assigned to me as RESOLVED FIXED ?
20:45* gw has never learnt bugzilla workflow
20:45mstangegw: nothing
20:45gwmstange: ah, cool. thanks!
20:45mstangegw: it just tracks that you were the one who fixed the bug
20:45gwright
20:46mstangegw: usually the bug gets assigned when somebody starts working on it, but once I see that you&#39;re working on it, the fix has usually already landed
 
17 Oct 2017
   
Last message: 1 minute ago