Regression: Content missing in few pages of chrome canary on Mac |
|||||||||||||||
Issue descriptionChrome Version: 72.0.3612.0 , 72.0.3614.0 OS: Mac OS 10.13.6 1. Open Gmail, Open any chat window -- transparent window is seen. 2. Open Sheets, Styles section is seen missing , Click inside any cell -- Blue outline is not seen on entire cell. 3. Some of the bookmarks are seen missing. And on hovering back on bookmark bar, contents are seen. NOTE: ===== 1. Issue is not seen in equivalent dev version. i.e; Issue is not seen in 72.0.3612.0 dev and 72.0.3614.0 dev. 2. Even on restart and re-setting flags to default issue is seen. 3. On upgrading from 72.0.3612.0 to 72.0.3614.0 also issue is seen. 4. Issue is not seen on another MacBook Pro machine with above mentioned builds. Unable to provide bisect info as issue is not seen in equivalent dev builds. Attaching screenshots for reference. Thanks!
,
Nov 19
That's weird. ccameron@, any ideas?
,
Nov 19
May be related to issue 906175 ?
,
Nov 20
Issue 906779 has been merged into this issue.
,
Nov 20
I've been seeing this locally too ... I suspect that something with rasterization is going sideways. Locally I have "Out-of-process Rasterization: Hardware accelerated" in my about:gpu. chelamcherla@, do you have that setting as well? Could you bisect with --enable-oop-rasterization?
,
Nov 20
,
Nov 20
I just turned on an OOP-R canary finch trial, so a bisect probably isn't going to find anything. This seems pretty bad. I'll turn off the finch trial and get this fixed.
,
Nov 20
^ once I verify that it's OOP-R
,
Nov 20
I'm not seeing this on ToT with or without OOP-R. ccameron, can you verify that this is OOP-R related? Are there other details required to repro this?
,
Nov 20
In issue 906175, which is maybe related, c#11 noted that reverting https://chromium-review.googlesource.com/c/chromium/src/+/1341141 appears to fix it? May be you can try out, if it is the same cause?
,
Nov 20
I couldn't repro after reverting the revert so probably a red herring
,
Nov 20
,
Nov 20
Issue 906175 has been merged into this issue.
,
Nov 20
It takes a while to repro -- I suspect that #11 is the same -- both content and UI are GPU rastered.
,
Nov 20
Managed to reproduce with OOP-R enabled. Running with OOP-R disabled to see if it happens again. I suspect that it will not, because I am seeing the following GL spew every time I get a raster glitch. BrowserWorker GL ERROR : GL_INVALID_OPERATION : glRasterCHROMIUM: RasterCHROMIUM: bad op: 53 In the attached video, you can see the spew come out as the bookmarks bar flashes.
,
Nov 20
Yup, that error is definitely OOPR. We should turn off the finch experiment while this is investigated.
,
Nov 20
Hmm, BrowserWorker implies that this is something that the browser is doing. I'll try to repro more with the bookmark bar.
,
Nov 20
The finch experiment has been rolled back. Over to enne@ for investigation and re-enabling it.
,
Nov 20
I've seen this affecting web contents as well as browser UI. WRT the browser UI, I visited a bunch of sites, opened and closed menus -- also had some WebGL going (and switching dGPU to iGPU). I also have the following theme installed (as penance for a bug I introduced): https://chrome.google.com/webstore/detail/morpheon-dark/mafbdhjdkjnoafhfelkjpchpaepjknad
,
Nov 21
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/30a0cda62883f107740e2f332aa7a2c98bffff72 commit 30a0cda62883f107740e2f332aa7a2c98bffff72 Author: Adrienne Walker <enne@chromium.org> Date: Wed Nov 21 07:07:31 2018 cc: Fix bug in ClientPaintCache::Purge ClientPaintCache::Purge was not modifying the map that it was generating PurgeData from. This meant that to generate a budget to remove, it would loop over the same element repeatedly and generate a large list of the same item to purge. (This is why this is intermintent and hard to repro.) However, it would then inform the service that this one entry was removed, but the client would still think it was there. If anything used it at that point, the client would assume it was cached, but the service would fail because it would not be cached. Bug: 906510 Change-Id: I71bf22833649875ea2486f5e8518b897f354b6c1 Reviewed-on: https://chromium-review.googlesource.com/c/1345418 Reviewed-by: Khushal <khushalsagar@chromium.org> Commit-Queue: enne <enne@chromium.org> Cr-Commit-Position: refs/heads/master@{#609932} [modify] https://crrev.com/30a0cda62883f107740e2f332aa7a2c98bffff72/cc/paint/paint_cache.cc [modify] https://crrev.com/30a0cda62883f107740e2f332aa7a2c98bffff72/cc/paint/paint_cache_unittest.cc
,
Nov 21
W.r.t comment#6, tested this issue on 72.0.3612.0 dev build by enabling --enable-oop-rasterization and issue is reproducible. @ccameron: Seems like fix is landed as per c#21, does this still require bisect? Thanks!
,
Nov 26
This should be fixed.
,
Nov 27
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/e62484a35ba0306106f3f0af0965bc00b772ecef commit e62484a35ba0306106f3f0af0965bc00b772ecef Author: Adrienne Walker <enne@chromium.org> Date: Tue Nov 27 18:41:30 2018 Add crash dump error reporting for OOP-R deserialization failures There was a recent issue where this was happening due to cache bugs. This is something that should never happen, so it would be nice to have it be more noisy when it fails. Bug: 906510 Change-Id: Ie4e8aaf415c491d4ba192d1b57c115cec7d81ca9 Reviewed-on: https://chromium-review.googlesource.com/c/1351386 Reviewed-by: Antoine Labour <piman@chromium.org> Reviewed-by: Khushal <khushalsagar@chromium.org> Commit-Queue: enne <enne@chromium.org> Cr-Commit-Position: refs/heads/master@{#611217} [modify] https://crrev.com/e62484a35ba0306106f3f0af0965bc00b772ecef/cc/paint/paint_op_buffer.h [modify] https://crrev.com/e62484a35ba0306106f3f0af0965bc00b772ecef/cc/paint/paint_op_reader.cc [modify] https://crrev.com/e62484a35ba0306106f3f0af0965bc00b772ecef/gpu/command_buffer/service/raster_decoder.cc
,
Nov 27
Issue 906118 has been merged into this issue.
,
Nov 28
Able to reproduce the issue on reported version, hence verifying the fix on latest canary 72.0.3624.0. Now pages rendered properly and no content is missing. Checked with and without enabling --enable-oop-rasterization flag. As fix is working as expected, adding verified labels. Thanks!
,
Nov 28
,
Nov 29
There have been 3 digits of crashes on last 2 Mac canary with magic signature '[Dump without crash] cc::PaintOpReader::Read'. Link to the list of the crashes: ================================ https://goto.google.com/myvco The crashes are seen probably due to the CL from C#24. Re-opening for further investigation and tagging M-72/RB-Dev for tracking. Though dump without crashes are not real crashes but huge spike in crashes may skew crash metrics. Please take a look at this as M-72 is going to be branched today.
,
Nov 29
,
Nov 29
Reply to #28:[Dump without crash] cc::PaintOpReader::Read, is tracked in Issue 906453, CLis already reverted. So please verify in tomorrows canary.
,
Nov 30
Adrienne, can we close this issues? Investigation of Issue 906453 is still going on, we can track the progress if there is no pending work here.
,
Nov 30
,
Nov 30
,
Nov 30
[Auto-generated comment by a script] We noticed that this issue is targeted for M-72; it appears the fix may have landed after branch point, meaning a merge might be required. Please confirm if a merge is required here - if so add Merge-Request-72 label, otherwise remove Merge-TBD label. Thanks.
,
Dec 7
khushalsagar@, as enne@ is OOO until Jan, do you know if this change needs to be merged to 72 branch? If so can you help remove the merge-tbd label and apply the merge-request label?
,
Dec 11
khushalsagar@ friendly reminder to help confirm if a merge is needed to M72 branch
,
Dec 12
khushalsagar@ can you please help review this and confirm if needed for M72, if you are not the right one to do that pls help add the right team so we can update this bug
,
Dec 12
(Both khushalsagar@ and enne@ are on vacation) All CLs were landed before the M-72 branch. The #c24 CL has been reverted for bug 910772 and merged into M-72.
,
Dec 12
thank you wangxianzhu@ , and thanks for removing the TBD label
,
Dec 20
The problem with content blanking out, that I reported on 906175 still persists on stable Version 71.0.3578.98 and canary Version 73.0.3645.0 on Mac. The toolbar is generally fine for me so far but the contents are not especially with dynamically generated sites. I am not quite sure this bug is still being fixed somewhere or if so, which tread(s) is(are). My issue was merged into this and this is where the trail ends so I just post the problem here instead of starting a new issue.
,
Dec 20
Not quite sure about the other sites as I haven't been visiting much sites myself for the past few weeks. |
|||||||||||||||
►
Sign in to add a comment |
|||||||||||||||
Comment 1 by chelamcherla@chromium.org
, Nov 1937.4 KB
37.4 KB View Download
108 KB
108 KB View Download
64.3 KB
64.3 KB View Download