Since one or two days, there's a regression where the alpha gradient at the end of tab titles fades to black instead of fading to transparent. This seems to be irregular, specific to some tabs in some cases, and possibly triggered by custom themes (I used mart3ll's Quantum theme [1]). The attachment shows the effect. I ran a mozregression, which landed on a recent WebRender upgrade: https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=1a4f40cd0bea7855f76b2237819e07e89fd89d75&tochange=6d9dc65ca0ed1a374dde7592a5b4191a7a10759c [1] https://addons.mozilla.org/en-US/firefox/addon/quantum-launch/

I'm not able to reproduce this. Would you be able to bisect using the intermediate builds from bug 1430829? From the screenshot it looks like you're on OS X so here are the OS X opt builds from oldest to newest. If you can, please run these builds and identify the pair where it goes from good to bad. cset 9cad76acef7c1da00ba4beb1e01e764eee8fe4d5: https://queue.taskcluster.net/v1/task/AO3Zy_ToTZ27fYidESHHhg/runs/0/artifacts/public/build/target.dmg cset 8811f69c4940b8d37b09aacf449f1b1932bf1e5e: https://queue.taskcluster.net/v1/task/Ex-tD4PrQxSd4xQzxCI_PQ/runs/0/artifacts/public/build/target.dmg cset 2547f6fe91da018cd3374acd661effb5f8fbe022: https://queue.taskcluster.net/v1/task/D-fd_FVUSU-40_oeM3iUSQ/runs/0/artifacts/public/build/target.dmg cset 46bd3a07cd002401005d8387f98fe187cc3b8cdf: https://queue.taskcluster.net/v1/task/PBPJeSHQQT-Y5Vg58FYPKw/runs/0/artifacts/public/build/target.dmg cset 5fae1ac37d9c9adf45601cfad83b31a3d6bafd68: https://queue.taskcluster.net/v1/task/L60OoTkJT5KJwrEJ1tnBUw/runs/0/artifacts/public/build/target.dmg cset e9269c7e06e20363be0b2a2a1be98d292ff7acca: https://queue.taskcluster.net/v1/task/C_1Z-muXSGWDYRWThoyVLQ/runs/0/artifacts/public/build/target.dmg

Here are my results: cset 9cad76acef7c1da00ba4beb1e01e764eee8fe4d5: https://queue.taskcluster.net/v1/task/AO3Zy_ToTZ27fYidESHHhg/runs/0/artifacts/public/build/target.dmg -> bad cset 8811f69c4940b8d37b09aacf449f1b1932bf1e5e: https://queue.taskcluster.net/v1/task/Ex-tD4PrQxSd4xQzxCI_PQ/runs/0/artifacts/public/build/target.dmg -> bad cset 2547f6fe91da018cd3374acd661effb5f8fbe022: https://queue.taskcluster.net/v1/task/D-fd_FVUSU-40_oeM3iUSQ/runs/0/artifacts/public/build/target.dmg -> bad cset 46bd3a07cd002401005d8387f98fe187cc3b8cdf: https://queue.taskcluster.net/v1/task/PBPJeSHQQT-Y5Vg58FYPKw/runs/0/artifacts/public/build/target.dmg cset 5fae1ac37d9c9adf45601cfad83b31a3d6bafd68: https://queue.taskcluster.net/v1/task/L60OoTkJT5KJwrEJ1tnBUw/runs/0/artifacts/public/build/target.dmg cset e9269c7e06e20363be0b2a2a1be98d292ff7acca: https://queue.taskcluster.net/v1/task/C_1Z-muXSGWDYRWThoyVLQ/runs/0/artifacts/public/build/target.dmg I started with the D-fd_FVUSU-40_oeM3iUSQ; downloaded, installed on top of my regular Nightly install, then opened with my own profile -- still see the problem. Then tried Ex-tD4PrQxSd4xQzxCI_PQ: same result, as well as for AO3Zy_ToTZ27fYidESHHhg. As I understand, this makes testing the remaining three builds useless. Hope that helps!

Thanks. If even the first one was bad, but the build prior to bug 1430829 landing was good, then the range must be the one I've indicated below. I'm not as confident in this range though; it seems kind of unlikely given the changes in the range. Dirkjan, can you try one more thing, just to confirm: install the AO3Zy_ToTZ27fYidESHHhg build the same way you did last time, and when you start it up, go to about:buildconfig and check the "Built from" URL. For that build it should say https://hg.mozilla.org/try/rev/0c4cea5ad05fb531fa50a113b5fbd86c26b8ef7a I think. That will give me a bit more confidence that installing the builds over top of the Nightly install actually did what we would expect. $ git log --oneline --graph eb9e1702df4b6dc036b649b3dd32ccc4bfbe43bf..9cad76acef7c1da00ba4beb1e01e764eee8fe4d5 * 9cad76ac Auto merge of #2313 - glennw:race-cond, r=kvark |\ | * db9a055c Fix for race condition when caching render tasks. |/ * 39abbda0 Auto merge of #2305 - kvark:capture2, r=glennw |\ | * 712e62fc Capture: refined and clarified the associated TODOs, re-exposed the read_pixels_into API | * cd01eaa0 Capture: proper handling of external images on the Renderer side | * f0d98658 Capture: raw test update | * ddde21fd Capture: share the push_iter implementation | * 7c7c0611 Capture: Wrench loading support | * 207e39ba Capture: GPU cache | * 33d81223 Gleam update | * 4b6a5925 Capture: device texture init refactor, texture loading | * ec26b520 Capture: GPU cache | * f485d663 Capture: cached textures on the Renderer | * 42e8cb42 Capture: tiling frame deserialization | * 02bc8895 Capture: deserialization derives, glyphs support for load_capture | * 2f691b93 Capture: refactored resource cache serialization, separate glyph blobs | * 1392cc39 Capture: resource cache serialization | * 8e3d65fe Capture: separate module | * e0357ade Capture: PlainFrame serialization under capture2 feature |/ * ef420106 Auto merge of #2306 - glennw:seg-regions, r=kvark * 7fef3730 Add support for regions to the segment builder.

Debian Testing, KDE, Radeon RX480 I can reproduce now! :) Wait until all tabs are loaded. Then install the theme. The other tab titles should get immediately black at the end. mozregression --good 2018-01-12 --bad 2018-01-23 --pref gfx.webrender.all:true startup.homepage_welcome_url:"https://addons.mozilla.org/en-US/firefox/addon/quantum-lights-dynamic/|http://www.spiegel.de/politik/deutschland/spd-mitgliederentscheid-wie-die-jusos-die-groko-noch-verhindern-wollen-a-1189420.html|https://reviewboard.mozilla.org/r/215060/diff/1#index_header" > 13:56.44 INFO: Last good revision: 1a4f40cd0bea7855f76b2237819e07e89fd89d75 > 13:56.44 INFO: First bad revision: 6d9dc65ca0ed1a374dde7592a5b4191a7a10759c > 13:56.44 INFO: Pushlog: > https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=1a4f40cd0bea7855f76b2237819e07e89fd89d75&tochange=6d9dc65ca0ed1a374dde7592a5b4191a7a10759c > 6d9dc65ca0ed Kartikaya Gupta — Bug 1430829 - Reduce fuzz in reftest from WR PR 2308. r=jrmuizel > 6a281573abf4 Kartikaya Gupta — Bug 1430829 - Update for removal of ImageFormat::Invalid in WR PR 2317. r=jrmuizel > ce7010ea6de7 Kartikaya Gupta — Bug 1430829 - Update Cargo lockfiles and re-vendor rust dependencies. r=jrmuizel > d1a22efb9e43 Kartikaya Gupta — Bug 1430829 - Update webrender to commit e9269c7e06e20363be0b2a2a1be98d292ff7acca. r=jrmuizel > WR @ 9cad76acef7c1da00ba4beb1e01e764eee8fe4d5 > https://treeherder.mozilla.org/#/jobs?repo=try&revision=0c4cea5ad05fb531fa50a113b5fbd86c26b8ef7a mozregression --repo try --launch 0c4cea5ad05fb531fa50a113b5fbd86c26b8ef7a --pref gfx.webrender.all:true startup.homepage_welcome_url:"https://addons.mozilla.org/en-US/firefox/addon/quantum-lights-dynamic/|http://www.spiegel.de/politik/deutschland/spd-mitgliederentscheid-wie-die-jusos-die-groko-noch-verhindern-wollen-a-1189420.html|https://reviewboard.mozilla.org/r/215060/diff/1#index_header" good > WR @ 8811f69c4940b8d37b09aacf449f1b1932bf1e5e > https://treeherder.mozilla.org/#/jobs?repo=try&revision=e7839088a9fb83f04c92216088ee4ab71537b4dd mozregression --repo try --launch e7839088a9fb83f04c92216088ee4ab71537b4dd --pref gfx.webrender.all:true startup.homepage_welcome_url:"https://addons.mozilla.org/en-US/firefox/addon/quantum-lights-dynamic/|http://www.spiegel.de/politik/deutschland/spd-mitgliederentscheid-wie-die-jusos-die-groko-noch-verhindern-wollen-a-1189420.html|https://reviewboard.mozilla.org/r/215060/diff/1#index_header" bad > WR @ 2547f6fe91da018cd3374acd661effb5f8fbe022 > https://treeherder.mozilla.org/#/jobs?repo=try&revision=8915a19fda816b4a7e040c30fda26241dd0f96eb mozregression --repo try --launch 8915a19fda816b4a7e040c30fda26241dd0f96eb --pref gfx.webrender.all:true startup.homepage_welcome_url:"https://addons.mozilla.org/en-US/firefox/addon/quantum-lights-dynamic/|http://www.spiegel.de/politik/deutschland/spd-mitgliederentscheid-wie-die-jusos-die-groko-noch-verhindern-wollen-a-1189420.html|https://reviewboard.mozilla.org/r/215060/diff/1#index_header" bad > WR @ 46bd3a07cd002401005d8387f98fe187cc3b8cdf with bustage fix > https://treeherder.mozilla.org/#/jobs?repo=try&revision=375987120b09b3a5b7e4f42192237c9a0afcf13b mozregression --repo try --launch 375987120b09b3a5b7e4f42192237c9a0afcf13b --pref gfx.webrender.all:true startup.homepage_welcome_url:"https://addons.mozilla.org/en-US/firefox/addon/quantum-lights-dynamic/|http://www.spiegel.de/politik/deutschland/spd-mitgliederentscheid-wie-die-jusos-die-groko-noch-verhindern-wollen-a-1189420.html|https://reviewboard.mozilla.org/r/215060/diff/1#index_header" bad > WR @ 46bd3a07cd002401005d8387f98fe187cc3b8cdf > https://treeherder.mozilla.org/#/jobs?repo=try&revision=d13fe8c4f00dbc5071624b7f035464162595371c mozregression --repo try --launch d13fe8c4f00dbc5071624b7f035464162595371c --pref gfx.webrender.all:true startup.homepage_welcome_url:"https://addons.mozilla.org/en-US/firefox/addon/quantum-lights-dynamic/|http://www.spiegel.de/politik/deutschland/spd-mitgliederentscheid-wie-die-jusos-die-groko-noch-verhindern-wollen-a-1189420.html|https://reviewboard.mozilla.org/r/215060/diff/1#index_header" bad > WR @ 5fae1ac37d9c9adf45601cfad83b31a3d6bafd68 > https://treeherder.mozilla.org/#/jobs?repo=try&revision=dad0ba506f8aa8065af431056303d50c5b40ff61 mozregression --repo try --launch dad0ba506f8aa8065af431056303d50c5b40ff61 --pref gfx.webrender.all:true startup.homepage_welcome_url:"https://addons.mozilla.org/en-US/firefox/addon/quantum-lights-dynamic/|http://www.spiegel.de/politik/deutschland/spd-mitgliederentscheid-wie-die-jusos-die-groko-noch-verhindern-wollen-a-1189420.html|https://reviewboard.mozilla.org/r/215060/diff/1#index_header" bad > Err the R4-opt unexpected pass only showed up at WR @ e9269c7e06e20363be0b2a2a1be98d292ff7acca: > https://treeherder.mozilla.org/#/jobs?repo=try&revision=750c40072847e3ac2f072042699fe4a2a48e25f3 mozregression --repo try --launch 750c40072847e3ac2f072042699fe4a2a48e25f3 --pref gfx.webrender.all:true startup.homepage_welcome_url:"https://addons.mozilla.org/en-US/firefox/addon/quantum-lights-dynamic/|http://www.spiegel.de/politik/deutschland/spd-mitgliederentscheid-wie-die-jusos-die-groko-noch-verhindern-wollen-a-1189420.html|https://reviewboard.mozilla.org/r/215060/diff/1#index_header" bad ----- > WR @ 9cad76acef7c1da00ba4beb1e01e764eee8fe4d5 is still good > WR @ 8811f69c4940b8d37b09aacf449f1b1932bf1e5e is still bad https://github.com/servo/webrender/compare/9cad76acef7c1da00ba4beb1e01e764eee8fe4d5...8811f69c4940b8d37b09aacf449f1b1932bf1e5e

So that points to this range: * 8811f69c Auto merge of #2316 - kvark:capture-fix, r=staktrace |\ | * 3befcf77 Capture: workaround for cbindgen parsing failure * e5514a40 Auto merge of #2311 - glennw:seg-bs, r=kvark * 9b939a40 Optimize box-shadow rendering (use segments for pictures). which seems more likely to me.

(I was also able to reproduce using your STR in comment 4 and confirmed your result). Filed a GH issue for this.

This appears to have been fixed in 60.0a1 (2018-01-30).

STR from comment 4: mozregression --find-fix --bad 2018-01-24 --good 2018-01-30 --pref gfx.webrender.all:true startup.homepage_welcome_url:"https://addons.mozilla.org/en-US/firefox/addon/quantum-lights-dynamic/|http://www.spiegel.de/politik/deutschland/spd-mitgliederentscheid-wie-die-jusos-die-groko-noch-verhindern-wollen-a-1189420.html|https://reviewboard.mozilla.org/r/215060/diff/1#index_header" 20180130102929 (9746e0a0a81cc089ff65e30ae902864846cd1b94) @ Debian Testing (KDE, Radeon RX480) bad Still reproducible.

I see this, except the tab labels fade to white instead of black. I guess it's theme-dependent.

This bug seems much more likely to happen on HiDPI rather than normal-resolution screens (on the same rMBP).

(In reply to Dirkjan Ochtman (:djc) from comment #11) > This bug seems much more likely to happen on HiDPI rather than > normal-resolution screens (on the same rMBP). I see this very consistently on a 1080p display at 1.25x scale on Linux. I guess that counts as HiDPI?

Gankro, I was able to reproduce this on Linux. The problem is the way the image-mask clip is interacting with text shadows. I've assigned to you since it seems likely to be resolved by the fix for https://github.com/servo/webrender/issues/2321, which I believe you're working on?

Added a reduced test case of this bug. I'm not really sure at this stage whether the bug is in how WR is handling the image mask on the text shadow, or something to do with the way the gecko DL was built. But it is based on a recording of the website in question, and shows the shadow not fading out with the mask, as intended.

The bugs I'm handling are all about special behaviour for local clip-rects (them needing to be offset), while this seems to be the opposite case: non-local clips breaking.

There's been a bunch of work touching this area recently. I'd like to wait to see if the fix for Bug 1412375 fixes this. (alternatively, has this already been fixed? I don't know what glenn's test case used to look like)

This has been fixed for me (original reporter) -- and for pretty long, too.

Install the theme, switch to the second and third tab. mozregression --find-fix --bad 2018-01-30 --good 2018-03-01 --pref gfx.webrender.all:true startup.homepage_welcome_url:"https://addons.mozilla.org/en-US/firefox/addon/quantum-launch/|http://www.spiegel.de/politik/deutschland/spd-mitgliederentscheid-wie-die-jusos-die-groko-noch-verhindern-wollen-a-1189420.html|https://reviewboard.mozilla.org/r/215060/diff/1#index_header" > 11:15.90 INFO: First good revision: 5eb19c81c7fd7d5ffbc3eac211cf20136c02c3d7 > 11:15.90 INFO: Last bad revision: d49f13abd54ad351ee1665a639127f14b938bc25 > 11:15.90 INFO: Pushlog: > https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=d49f13abd54ad351ee1665a639127f14b938bc25&tochange=5eb19c81c7fd7d5ffbc3eac211cf20136c02c3d7 > 3a32ca63e380 Kartikaya Gupta — Bug 1439565 - Update webrender to commit 8a19316a733a484bf9bafb8257e3008b1418bfe4. r=jrmuizel Fix range: https://github.com/servo/webrender/compare/e8d2ffb404a85651fe08a6d09abbece9bd2b9182...8a19316a733a484bf9bafb8257e3008b1418bfe4