Re: [svg-tests] Convert SVG 1.1 marker tests to reftests. (#4015)

@nikosandronikos Depending on another test passing is practically inevitable (after all, at some level, all testes rely on some other test passing, even if it is just basic functionality like the `color` property working), so I don't think we've ever really worried about it beyond trying to minimise the complexity of references such that they're unlikely to bogusly fail.

As for anti-aliasing differences, we badly need to come up with a better solution for fuzzy matching, because just assuming near identical shapes render identically doesn't quite work.

That said, at least in the marker-002 case I'm not sure this is anti-aliasing differences: I think it's subpixel rendering differences. In the bottom left, we start off with a `scale(4)`, which is fine, but then we have a `scale(0.2)`—and 0.2 cannot be expressed exactly—so I think we end up with 20px being 17px or something like this. I'd try getting rid of all the scales and see if it still renders differently.

View on GitHub: https://github.com/w3c/web-platform-tests/pull/4015#issuecomment-261252491

Received on Thursday, 17 November 2016 13:57:36 UTC