Bug 118904

Summary: [GTK] Layout Test mathml/presentation/bug95404.html is failing
Product: WebKit Reporter: Simon Pena <spenap>
Component: Tools / TestsAssignee: Nobody <webkit-unassigned>
Status: RESOLVED FIXED    
Severity: Normal CC: cfleizach, fred.wang, gur.trio, zan
Priority: P2 Keywords: Gtk, LayoutTestFailure
Version: 528+ (Nightly build)   
Hardware: Unspecified   
OS: Unspecified   
Bug Depends on: 116600    
Bug Blocks:    

Description Simon Pena 2013-07-19 07:08:52 PDT
The following layout test is failing on GTK

mathml/presentation/bug95404.html

This test was introduced when the MathML pixel tests were changed to ref tests in r152777 <http://trac.webkit.org/changeset/152777>
Comment 1 Frédéric Wang (:fredw) 2013-07-19 12:07:38 PDT
Do you have screenshots of the failing test?

It worked for me on GTK and when I submitted it to the review queue.
Comment 2 Simon Pena 2013-07-22 01:22:20 PDT
(In reply to comment #1)
> Do you have screenshots of the failing test?
> 
> It worked for me on GTK and when I submitted it to the review queue.

Hi Frédéric, you can take a look here: 

http://build.webkit.org/results/GTK%20Linux%2064-bit%20Release/r152777%20%2839211%29/results.html.

I am not familiar with MathML, and I don't know how to interpret the results, but I'll be happy to help you in anything I can to get this test unskipped again.
Comment 3 Frédéric Wang (:fredw) 2013-07-22 01:40:56 PDT
(In reply to comment #2)
> I am not familiar with MathML, and I don't know how to interpret the results, but I'll be happy to help you in anything I can to get this test unskipped again.

This is a reftest, the pages

mathml/presentation/bug95404.html

and

mathml/presentation/bug95404-expected.html

should render exactly the same. By the naked eye I don't see any differences between the two and diff.png is just black. But it's possible that there is a very small pixel difference that one can not see by the naked eye...
Comment 4 Simon Pena 2013-07-22 01:58:45 PDT
(In reply to comment #3)
> (In reply to comment #2)
> > I am not familiar with MathML, and I don't know how to interpret the results, but I'll be happy to help you in anything I can to get this test unskipped again.
> 
> This is a reftest, the pages
> 
> mathml/presentation/bug95404.html
> 
> and
> 
> mathml/presentation/bug95404-expected.html
> 
> should render exactly the same. By the naked eye I don't see any differences between the two and diff.png is just black. But it's possible that there is a very small pixel difference that one can not see by the naked eye...

Yes, you're right. 

I will ask someone more knowledgeable than me with the bots to see if there's a way to treat this as a pass or if there's a way to make that small difference go away by maybe modifying the reftest.
Comment 5 Frédéric Wang (:fredw) 2013-07-22 02:04:00 PDT
(In reply to comment #4)
> Yes, you're right. 
> 
> I will ask someone more knowledgeable than me with the bots to see if there's a way to treat this as a pass or if there's a way to make that small difference go away by maybe modifying the reftest.

OK thanks. If the latter way is necessary, I can help to modify the reftest later.
Comment 6 Zan Dobersek 2013-07-22 02:35:24 PDT
There's a minimal difference appearing in the top left corner due to the square root sign overflowing the red box at the very bottom.

There are a couple of tests already failing in similar fashion, though I can't at the moment remember the specific ones. Still, the cause is probably the same, so a master bug for these test failures should be created.

The reference tests only pass if the test output matches the reference output 100%, so there's no real way to make this test seemingly pass unless the outputs match completely.

As for the problem itself, I think it could be font-related. Comment #1 makes me think that there are still some specifics that are leaking through the Jhbuild environment configuration.
Comment 7 Frédéric Wang (:fredw) 2013-07-22 02:43:21 PDT
(In reply to comment #6)
> There's a minimal difference appearing in the top left corner due to the square root sign overflowing the red box at the very bottom.

OK, that's what I suspected. This will probably pass after bug 116600 is fixed.
Comment 8 Frédéric Wang (:fredw) 2013-07-24 10:33:27 PDT
@Simon: can you try again to run the test now that bug 116600 is fixed?
Comment 9 Frédéric Wang (:fredw) 2013-08-11 03:43:19 PDT
(In reply to comment #8)
> @Simon: can you try again to run the test now that bug 116600 is fixed?

Actually there is another square root that is visible so bug 116600 does not solve the problem. One workaround is modify the test to paint the radical in white (but its content still in black).
Comment 10 Frédéric Wang (:fredw) 2014-03-19 05:24:14 PDT
Manually committed r165889: http://trac.webkit.org/changeset/165889

I added a 2px border around the rectangles to prevent the remaining failure due to antialiasing.