mozilla :: #gfx

12 Jul 2017
00:06gwGankro: I'm not sure what you're referring to - although there's definitely several font related fixes coming soon.
00:07gwGankro: Is the DL produced by shadow-complex actually expected to ever occur? It's not quite clear to me what the semantics of that should be
00:08Gankrogw: oh sorry, I guess I never mentionned it online -- with my change (so webrender gets all glyphs from gecko), emoji show up slightly small. I believe it's an artifact of subpixel-aa snaping them smaller. hopefully the new text fixes will fix that.
00:09Gankrogw: lemme check the test
00:11Gankrogw: ah, yeah. So I don't expect you'll ever get text like this from gecko/servo -- but theoretically as an optimization we could identify two text-spans have the same shadow and merge them by annihilating a pop-push pair.
00:13Gankrogw: the expected rendering/parsing is this [(AAAA){BBBB}CCCC]
00:14GankroSo all the text has the lowest shadow, but then AAAA and BBBB have seperate shadows that only apply to them
00:15gwGankro: ok, I think that makes sense. got everything else working except that test case - will work on that today
00:16Gankrogw: it might be reasonable to punt on it -- but as you say that would have "expressible but non-implemented" semantics
00:16gwGankro: it should be able to handle this test case, it probably signals a bug in the implementation, so I'll dig into it
00:20Gankrogw: btw, how exactly does taking the shadow of a BlobImage work?
00:20GankroI had naively convinced myself that made sense but now that I've worked with that API a bit I'm not sure it does?
00:22gwGankro: I haven't started looking at that yet - initially I just want to get text shadows working with the new interface. But it's not clear to me why that would be a problem?
00:24Gankrogw: how does a red underline get turned into (say) a green shadow?
00:25gwGankro: fair point ... :/
00:26Gankrogw: when I was first doing it, I was assuming we could pass the color as an arg; do we do blobs with args at all?
00:26gwGankro: not currently - blobs are completely opaque to WR
00:27gwGankro: but I guess there's no reason the client code can't pass a color though
00:28GankroAlternatively I don't think it would be a big deal if we don't render opaque shadows until we get native support -- only "typo squiggles" seem hard given what webrender has now
00:29gwGankro: typo squiggles seem fairly trivial to add native support to WR ... maybe we should just do the "proper" solution
00:29Gankrogw: I think we can land this stuff first, and do that in follow ups
00:30GankroFor now just ignore blobs
00:30gwGankro: yep, agreed
00:31GankroI'll mock up the frontend/gecko work -- you already have a notion of "dotted" and "dashed" for borders, eh?
00:32gwGankro: yup
00:33* Gankro wonders if he can hack in decorations as borders
00:33Gankroprooobably a bad plan
01:00firebotBug 1359055 NEW, WebGL2: Upload of compressed textures from PBO fails
01:01Lenzakjgilbert: Bug 1379868
01:01firebot NEW, Disable Mochitest-gl under Windows 32-bit non-e10s mode.
03:29stshineMay I ask if gecko can build webrender display lists incrementally for now?
09:16nicalstshine not yet, incremental display list is in the work but it's not there yet (not even for gecko display lists)
10:18jyaBas: ping
12:42stshinenical: ok thanks
13:42Gankrokats-away: sorry by "seen by the test runner" I meant like, if I do a try build, will the mode my changes apply in (webrender layerless text) be checked?
13:52katsGankro: no, we are not testing layerless mode in WR automation yet. you'd have to flip the pref as part of your push
13:53katsbut also if it only affects layerless i don't think we have any existing baseline to compare it to
13:53katsand also it won't break anything when it lands, since that configuration is untested
13:54Gankrokats: so ship it as long as everything else still works?
13:54katsGankro: yeah basically
13:55mrobinsonkvark: Thanks for the review. We are going to make this WR a reality!
13:55kvarkhah, glad all the blockers are gone now!
13:55kvarkWirtual Reality :)
14:02bkellydoes anyone know why this gfx layers code is doing a blocking wait on the main thread?
14:07jrmuizelGankro: ping
14:15Gankrojrmuizel: pong
14:15jrmuizelGankro: are you coming in today?
14:15Gankrojrmuizel: yeah just having a slow morning
14:18jrmuizelGankro: great
14:47vliu_cloudjrmuizel: ping
14:47jrmuizelvliu_cloud: pong
14:48vliu_cloudjrmuizel: I proposed a patch in bug 1367689 and still waiting for your review. Could you please have a review?
14:48firebot NEW, Make Wrench work with gecko
14:48jrmuizelah right, sorry
14:48vliu_cloudjrmuizel: really thanks :)
14:48jrmuizelvliu_cloud: there you go
14:49vliu_cloudjrmuizel: thanks for your review.
15:01jyanical: maybe you know this.. Do you know which blob paint a YUV image with the D3D11 compositor?
17:23katsaosmond: ping
17:25aosmondkats: pong
17:26katsaosmond: did you see my needinfo on bug 1376657?
17:26firebot NEW, Intermittent dom/html/reftests/toblob-todataurl/toblob-quality-100.html == dom/html/reftests/toblob-
17:28aosmondkats: yes, but it was already pushed out of my mind. the way it works without WR is the combination of onload, sync decoding and drawing should ensure what is displayed is correct/final for the reftest compare.
17:29katsaosmond: is it different with WR?
17:29aosmondkats: in practice we get intermittent failures even on non-WR, which may have the same root cause (or may not). I'm not sure if these particular tests have intermittents on non-WR though.
17:29katsis there a metabug tracking the non-WR failures?
17:30aosmondkats: not afaik.
17:30katsaosmond: well if you need more reliably reproducing cases, these bugs seem to trigger fairly often in automation
17:31aosmondkats: that may be handy, as the non-WR failures have never been observed outside of automation (to my knowledge)....
18:04mchangmstange: review ping on please ?
18:04firebotBug 1372602 NEW, The hidden window spends time creating a compositor / texture factory identifier that it will never
18:04mstangemchang: sorry again, the tab is still open and I'm reminded of it every day, but I want to test it locally
18:04mchangmstange: sure np :)
18:05mstangemchang: I'm worried that it's going to change something about how we paint windows that are being opened
18:05mstangelike, it might open the window first and display black or white, and only then paint the contents, or something like that
18:05mchangmstange: yeah sure, i dont know the code very well, but i think it makes sense. I tried a bunch of windows and things it worked
18:05mchangbut yeah please test too
18:05mstangeI'd like to test it anyway
21:23mstangejrmuizel: have you moved your people somewhere else?
21:23mstangejrmuizel: or do you know if dl-test is in a bugzilla attachment somewhere?
21:23jrmuizelmstange: I have a local copy
21:23mstangejrmuizel: could you attach it to ?
21:23firebotBug 1379694 NEW, Try to understand memory access patterns during display list building
21:25jrmuizelmstange: done
21:25mstangejrmuizel: thank you
21:28jrmuizelmstange: I've moved some of the stuff to
13 Jul 2017
No messages
Last message: 69 days and 14 hours ago