New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 690470 link

Starred by 8 users

Issue metadata

Status: Fixed
Owner:
OOO until 2019-02-10
Closed: May 2017
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: Android
Pri: 2
Type: Bug



Sign in to add a comment

Inconsistent value in typed array

Reported by steph...@sketchfab.com, Feb 9 2017

Issue description

Steps to reproduce the problem:
1. Visit https://debugchrome.fatvoxel.com/models/fa575790a5b440a58606600a6a9e58c1
2. Is there an issue in the shading? (cf screenshot)
3. And/or Check the console log (the line after "blabal")

What is the expected behavior?
the console log should be "0 0" (and shading should be smooth)

What went wrong?
var nx = myarray[index];
I expect nx to be equal to myarray[index] :D. (and no, it's not a NaN)

Did this work before? N/A 

Chrome version: 55.0.2883.91  Channel: stable
OS Version: 7.1.1
Flash Version: 

When the bug is there, it's 100% reproducible, however the code where the issuehappens can't be simply copy/pasted in another context.

We tested several devices:

win 10, chrome stable, canary, FF, Edge or iOS+Safari
shading OK, 0 0

nexus 5x, canary 58.0.3006.0
shading KO, log "0 0"

nexus 5x, 55.0.2883.91
shading KO, log "4.3607766521637476e+27 0"

Mac, 55.0.2883.95 (64-bit)
shading KO, log: 114 6.89057183135153e+22 0

Android chrome 55.0.2883.91
shading KO, log: 6.129218121241909e+31 0

Android canary 58.0.3004.0
shading KO, log: 0 0

Mac Chrome Version 55.0.2883.95 (64-bit)
shading KO, log: 0 0

In the console the bug occured in the "sphericalUnmapping" file (it's the only sourcemapped file so you can access it through ctrl+p).
The first part of the file is our original code (used for shading).
The code between the "TEST START" and "TEST END" should be standalone (which *should* be completely independent of the inputs), I tried to simplify to its minimum.

Adding a "debugger" statement OR adding a breakpoint will automatically solve the issue.
Doing other stuffs such as multiplying a float by 1.0 or not initializing a float typed array with 0.0 can solve the bug too.... (see sphericalUnmapping.js comments).

 
Attached files.
3dOK.jpg
61.2 KB View Download
bug3dCanaryAndroid.png
323 KB View Download
bug3dStableAndroid.jpg
163 KB View Download
exampleLog.png
67.4 KB View Download

Comment 2 by junov@chromium.org, Feb 10 2017

Components: -Blink Blink>JavaScript
Labels: Needs-Bisect Needs-TestConfirmation
I couldn't repro it on Linux Chrome 56. The shading looks ok and I always get "0 0" in the console.

Test please try to confirm and bisect.
Labels: OS-Mac OS-Windows
Cc: ram...@chromium.org
Labels: -OS-Windows

Comment 7 by ram...@chromium.org, Feb 13 2017

It looks like the issue did not repro on M53. Will bisect it.
Cc: jmukthavaram@chromium.org
Able to reproduce the issue on Mac 10.12.2 using chrome reported version-55.0.2883.95 as 55.0.2883.91 build is unavailable.Issue working as expected in stable-56.0.2924.87& latest Canary-58.0.3012.0 on Windows 7,windows 10,Mac 10.12.2 & ubuntu 14.04.

Unable to perform reverse bisect as we observed privacy error when user navigates to below URL between builds from 56.0.2884.0 to 56.0.2916.0.
https://debugchrome.fatvoxel.com/models/fa575790a5b440a58606600a6a9e58c1

Observed issue working as intended from 56.0.2917.0 to latest Canary-58.0.3012.0.
Please find the attached screencast for reference.
Thanks!!


690470.mp4
2.4 MB View Download
Status: WontFix (was: Unconfirmed)
So it seems to be fixed already. Closing according to #8
Labels: -OS-Mac
Status: Available (was: WontFix)
We are still able to repro this issue on Android Chrome version '57.0.2987.56'

Bisect info:
54.0.2823.2(GOOD Build) 
54.0.2824.0(BAD Build)

CL range;
410288(GOOD), but before 410408(BAD)

Hmm any news?

We'll probably have to kill soon our test server with the reproducible issue: https://debugchrome.fatvoxel.com/models/fa575790a5b440a58606600a6a9e58c1

Unfortunately, I couldn't reproduce this issue on our production server (maybe because the js is minified, etc..).

Labels: -Needs-Bisect
Link to bisect range is;

https://chromium.googlesource.com/chromium/src/+log/54.0.2823.2..54.0.2824.0?pretty=fuller&n=10000

Unable to further bisect because of missing builds.
Labels: -Needs-TestConfirmation
I had to shut down the test server (cost us money for nothing :)).

We can restart it if you want to debug back the issue.

I must say it's one of the weirdest issue I ever encountered in chrome, basically leading to (in some very rare unexplained circumstances though) :

var value = tablec[ 0 ]; // tablec being a Float32Array with only 0
console.log( value, tablec[ 0 ] ); // value being random
Cc: bmeu...@chromium.org
Owner: petermarshall@chromium.org
Status: Assigned (was: Available)
Peter, could you have a look please?
Quite possibly related to https://bugs.chromium.org/p/chromium/issues/detail?id=707270 which has been fixed very recently. There is no way to repro this bug without the test server, AFAIK. Are you able to reproduce still on canary? Or could you start the test server again so I can look into it further?
Thanks.
Unfortunately, I can't deploy the *exact* same version as before so we can't reproduce the issue.

You can close this ticket then, thanks!
Status: Fixed (was: Assigned)
Ok no worries - I think this could be the same issue as 707270, so hopefully it is fixed, although feel free to open a ticket if you can repro it again.

Sign in to add a comment