764108 - [Azure] Cairo canvases have precision issues which cause a test fail

Status: RESOLVED WORKSFORME

(Core :: Graphics, defect)

Description

[Nick Cameron [:nrc]]

Attachment: example, highlights the error

http://philip.html5.org/tests/canvas/suite/tests/2d.path.stroke.skew.html fails. The problem can be seen more clearly on the attached example.

I believe this is due to a precision issue, we end up doing a lot of transforms because of the way we implement paths, here there is a rotation, then a large scale, which highlights the problem.

Comment 1

[Jeff Muizelaar [:jrmuizel]]

(In reply to Nick Cameron [:nrc] from comment #0)
> Created attachment 632361 [details]
> example, highlights the error
> 
> http://philip.html5.org/tests/canvas/suite/tests/2d.path.stroke.skew.html
> fails. The problem can be seen more clearly on the attached example.
> 
> I believe this is due to a precision issue, we end up doing a lot of
> transforms because of the way we implement paths, here there is a rotation,
> then a large scale, which highlights the problem.

Do you know why this doesn't impact the other backends?

Comment 2

[Nick Cameron [:nrc]]

I think it is something to do with Cairo using limited precision fixed point internally, but that is mostly a guess.

Comment 3

[Jeff Muizelaar [:jrmuizel]]

(In reply to Nick Cameron [:nrc] from comment #2)
> I think it is something to do with Cairo using limited precision fixed point
> internally, but that is mostly a guess.

Can you figure out why didn't we hit this problem before?

Comment 4

[Nick Cameron [:nrc]]

My guess is that with the Azure layer, we are doing a bunch more transforms than before (especially with paths), and this exposes the precision issue sooner. I imagine that the bug could be recreated without Azure, but that it would take more effort.

Comment 5

[Jeff Muizelaar [:jrmuizel]]

(In reply to Nick Cameron [:nrc] from comment #4)
> My guess is that with the Azure layer, we are doing a bunch more transforms
> than before (especially with paths), and this exposes the precision issue
> sooner. I imagine that the bug could be recreated without Azure, but that it
> would take more effort.

It would be nice to have more than a guess. It should be pretty easy to compare the points going into cairo in both cases.

Comment 6

[Nick Cameron [:nrc]]

This no longer looks broken, I assume that the work on Azure/Cairo paths (e.g., bug 795135) fixed this somehow. Which presumably means I was wrong about it being a precision issue, but I don't really see the point in investigating further.