mozilla :: #gfx

11 Sep 2017
01:46GankroIs anyone planning to tackle https://bugzilla.mozilla.org/show_bug.cgi?id=1397407, or otherwise think they know what's going on? I'm probably going to take a look at it tomorrow if someone more qualified in that subsystem doesn't
01:46firebotBug 1397407 UNCONFIRMED, nobody@mozilla.org Crash in mozalloc_abort | abort | core::option::expect_failed | webrender::resource_cache::ResourceC
01:47GankroMissing cached_images
01:54Gankrogw: you wrote the image cache right?
01:55Gankro(got any tips?)
01:58* gw looks
01:59gwGankro: is it this? https://github.com/servo/webrender/issues/1611
02:02Gankrogw: nah it's crashing on ".expect("Didn't find a cached resource with that ID!")"
02:03Gankrounless the code's changed a bit
02:04gwGankro: ok. If you can repro it relatively easy, best thing is probably just to log out all the inserts / removes, plus the key when get fails. I suspect that will make it fairly clear what's going on.
02:04gwGankro: it might be a bug where the key has, e.g. tiling enabled in insert but not in the get
02:04Gankrooh interesting
02:05Gankrogw: can you recommend the functions to log to catch that sort of thing?
02:07gwGankro: yea - I would just add logging to each of the methods in ResourceClassCache - e.g. insert(), get(). Hmm, I don't see a remove() there though - callers may be directly accessing that to remove items :/
02:07Gankrogw: yeah that's why I asked
02:08gwGankro: there should be very few callers (probably only one). and thr resources hash map there is not public, so it should be easy to find :)
02:11Gankrogw: oh it looks like everything goes through clear?
02:12gwGankro: yea - just looking at that myself - looks like everything goes through either clear() or clear_keys() - so should be easy to catch those
02:57Gankrogw: just gets straight up deleted, so probably an issue up in gecko
03:00gwGankro: ok - image or glyph?
03:00Gankrogw: ImageRequest { key: ImageKey(IdNamespace(1), 341), rendering: Auto, tile: None }
03:01gwGankro: ok - maybe something not quite using resource update transactions correctly in gecko then?
03:01Gankrogw: that's my guess too
10:19mstangedo we have a bug for text-shadow using wrong glyphs in webrender?
10:20mstangedata:text/html,<div style=&quot;background:gray;color:rgba(0,0,0,0.01);text-shadow:0 1px 2px black; text-align: center; font-size: 40px;&quot;>Test
10:20mstangeit should say Test
10:20mstangeit says eTts
10:20mstangethis is one of the more impressive bugs I&#39;ve encountered recently
13:09mstangeit looks like that bug is fixed in the latest build
13:43Gankroethlin_: when you were working on WRUserData, were you able to reproduce the crasher in get_cached_image?
13:43Gankro(It&#39;s the consequence of destroying certain WrUserData instances too soon)
14:58padenotjnicol, ping
14:59jnicolpadenot: hi
15:09nicaljrmuizel: are we newsletter worthy this week ?
15:09padenotjnicol, regarding 1398697, paint flashing is broken for me on OSX
15:10padenotjnicol, and I don&#39;t know how to use the prefs, I flipped them, made a profile, and then OSX deadlocked at the system level, requiring a reboot
15:11jnicolpadenot: that&#39;s odd.. what&#39;s broken about paint flashing?
15:11padenotwell nothing flashes
15:11padenoteven on pages where clearly something is being invalidated
15:12padenotside by side with nical&#39;s laptop on linux, his firefox would show repaint regions and mine would not
15:13jrmuizelnical: we will be
15:13jrmuizelnical: but I&#39;d like some crashes fixed first
15:15padenotjnicol, can you tell me quickly what to do with those prefs ?
15:15padenotso I can get you data ?
15:17jnicolpadenot: layout.display-list.dump should just cause loads of output to the terminal/stderr. just save it to a file and probably best to share it with me on google drive because it;ll be massive
15:17padenotok terminal
15:17padenotone sec, doing that
15:17jnicolpadenot: that&#39;s really strange about paint flashing!
15:21padenotjnicol, I can also look what&#39;s wrong if you tell me what to look for
15:22padenotjnicol, http://paul.cx/public/log-whatsapp-jamie.log.gz
15:22padenot21MB compressed, 500 or so uncompressed
15:24jnicolpadenot: thanks! I&#39;ll have a look now - not really looking for anything in particular i&#39;m afraid, just seeing if anything looks suspicious.
15:24padenotjnicol, also, just a note, when I look at the page, nothing is moving
15:24padenotno apparent animation or the like
15:24jnicolI think paint flashing would tell us more, it&#39;s annoying that&#39;s not working
15:24jnicolHmm
15:26jnicolpadenot: Thanks for compressing it, I&#39;m on the train from glasgow to london at the moment, would probably reach london before 500MB downloaded!
15:26padenotheh
15:38gw280jfkthame: https://bugzilla.mozilla.org/show_bug.cgi?id=1392669
15:38firebotBug 1392669 NEW, nobody@mozilla.org Left menu text does not always appear on page load for the Canvas-based game at http://s20.sfgame.pl
15:39gw280jfkthame: I was needinfod on that, and the obvious bug in the regression range is one of yours
15:39jfkthamegw280: just taking a look...
15:49jfkthamegw280: commented in bug
16:05padenotBas, ping
16:09gw280jfkthame: thanks :)
16:10gw280jfkthame: I don&#39;t have a fx dev setup anymore, so hard for me to do local testing
16:50gw280jfkthame: I liked your comment about sfgame144.js :)
16:50jfkthamecare to dig in? ;)
17:06Gankrojrmuizel: ping on oblique text position r? request
17:06jrmuizelGankro: ah right
17:19Gankroty
17:35Gankrokats: oh, do you actually know why layers-free trashes the tab color?
17:36katsGankro: not really. that&#39;s a separate issue from the one i filed
17:37katsGankro: actually wait, what do you mean by &quot;tab color&quot;?
17:37katsthe background color of the tabs in the tab bar?
17:37Gankrooh woah weird yeah your bug seems... different
17:37katsyeah
17:38GankroI thought it was this ^
17:39Gankroerr, hmm that upload failed...?
17:39Gankro*shrug*
17:45jfkthamejrmuizel: r? ping for the patches in bug 1395061, if you can find some time
17:45firebothttps://bugzil.la/1395061 NEW, myk@mykzilla.org refactor gfxFontEntry::IsSymbolFont, ::SupportsLangGroup, and ::MatchesGenericFamily into gfxFontFam
17:45jrmuizeljfkthame: I&#39;ll take a look
17:45jfkthamejrmuizel: thx!
18:20jfkthamejrmuizel: awesome, ty
18:20jrmuizeljfkthame: seemed like a nice set of patches
18:21jfkthameyeah, it&#39;s tidier than it used to be, but mostly it unblocks the Preferences-window perf issue
18:31katsmchang: are you working on any webrender things right now? if not, should i unassign you from WR bugs?
18:49pulsebotCheck-in: https://hg.mozilla.org/projects/graphics/rev/f48324bf14d0 - Miko Mynttinen - Add a profiler marker for display list resources
18:49pulsebotCheck-in: https://hg.mozilla.org/projects/graphics/rev/dc69a39af99e - Miko Mynttinen - Avoid extra call to nsLayoutUtils::GetScrollableFrameFor()
20:53Gankrokats: do you know about any bugs regarding hovering over tabs causing flickering in the text style / chrome?
20:54katsGankro: nope
20:56Gankromstange: jrmuizel just suggested you might know?
21:12mstangeGankro: https://bugzilla.mozilla.org/show_bug.cgi?id=1389476
21:12firebotBug 1389476 ASSIGNED, mstange@themasta.com On Mac if you hover the mouse over the tabs, the text from the tabs and URL and search bars becomes
21:15Gankromstange: oh you&#39;re up \o/
21:15mstangeGankro: for a few more minutes :)
21:16Gankromstange: well when you&#39;re up tomorrow, I&#39;m at a total loss on this crasher and it seems related to tab styles
21:16Gankrohoping you have a magic insight
21:16mstangeGankro: is this the same crash as the one you get when you switch to an about:config tab?
21:17mstangeI&#39;ve probably submitted around 10 crash reports for that one today
21:18mstangeGankro: e.g. https://crash-stats.mozilla.com/report/index/abe10425-6955-459b-acda-c42dd0170911 is one of mine
21:19Gankromstange: yes
21:20mstangeoh, I see!
21:21mstangebecause about:config is a very short URL
21:21mstangeso it doesn&#39;t need the fadeout mask
21:22mstangehmm, no, that&#39;s not it, I don&#39;t get the crash if I switch to other tabs with short URLs
21:24mstangeGankro: yeah no I don&#39;t know what could be going on there
21:24mstangebut even before the crash I don&#39;t see the fadeout effect do anything, i.e. there is no fadeout for me with webrender
21:24gwjrmuizel: I&#39;ve r+ed that PR - a retry only works if the sha1 hasn&#39;t changed.
21:26Gankromstange: any idea why the whole chrome flickers? (Can you repro that?)
21:27mstangeGankro: I can, and I don&#39;t know why it happens
21:27mstangeit only seems to happen when I move my mouse quickly from the content area over the toolbar to the tabs
21:27mstangebut not if I rest the mouse on the toolbar in the middle of that process
21:32jrmuizelgw: how do I make it succeed?
21:40gwDo we maintain our own fork of angle? How closely do we track upstream and do we ever add features to our angle? Or just bug fixes?
21:40mstangeGankro: hmm, no, the flickering seems to happen every time I hover a tab
21:41mstangeGankro: which is also the time when the line animation at the top of the tab plays
21:41mstangeand if I hover too many tabs too many times, I crash
21:41mstangebut &quot;switch to an about:config tab&quot; seems to be a more reliable way to reproduce crashing
21:41mstangejrmuizel: you just missed gw&#39;s question about angle again
21:42mstangejrmuizel: 23:40 <gw> Do we maintain our own fork of angle? How closely do we track upstream and do we ever add features to our angle? Or just bug fixes?
21:42jrmuizelmstange: I bet
21:43jrmuizelgw: we used to have https://github.com/mozilla/angle/
21:45gwjrmuizel: mstange: I guess jgilbert is the best person to talk to in regards to potential angle changes?
21:46mstangethat would be my guess as well
21:46jrmuizelgw: yes
21:46jrmuizelgw: what do you have in mind?
21:48gwjrmuizel: There is a GL extension that we could use to make subpixel text rendering much more efficient. There is a patch for ANGLE to support that extension, but it was merged and backed out, with no activity since. It&#39;d be nice to have access to that, which I believe is supported on underlying d3d10/11 hardware.
21:48jrmuizelgw: what&#39;s the extension/where&#39;s the patch?
21:49gwjrmuizel: https://bugs.chromium.org/p/angleproject/issues/detail?id=1085 and https://chromium-review.googlesource.com/c/angle/angle/+/346410
21:54jrmuizelgw: it seems like we could convince them to take it
21:55gwjrmuizel: yea, they seemed happy with the patch - it was just some unresolved test failures that caused it to be backed out.
12 Sep 2017
No messages
   
Last message: 9 days and 7 hours ago