Re: [svgwg] SVG2 path data coordinates are currently spec'd as integers.

> When, how do we make up for the oversight back then?

Starting now would be perfectly fine - tests can be submitted to the WPT repo (it's not actually a mirror you link above - it's the source of truth - PRs welcome.)

FWIW, SVG has had a bit of a rough upbringing (not to say that other similar era, and complexity, specs had it easier.) I wasn't around in 2001 when 1ed SVG 1.1 was released, but I'm assuming the testsuite wasn't really at the top of the list of priorities. Even then it was probably one of the larger ones for a spec in those days, clocking in at 300-400 tests or so I think... For spec development beyond 1ed it was probably as easy to get WG members to write tests as it was to get them to edit the actual spec. (2ed and Tiny 1.2 had a few more tests, but not orders of magnitude more.)  

>From an implementer's perspective it usually comes down to resources (people and their time) - you may not have the luxury of spending all the time you want analyzing every nook and cranny of a particular section of the specification, because you need to implement another even more obscure and less well-defined part than what you're currently working on. So occasionally an edge error-case is missed in test coverage (more often than not I'd say even.) So if this particular case stuck out like a sore thumb to you, that's great! That doesn't mean that everyone else would think of it right away - especially those tasked with implementing a 300+-page specification on the clock. I take it you'd be equally astounded to hear about the number of browser tests that come to be via bugs (like in the Gecko bug referenced above.) Also, for a long time this whole "sharing tests" thingamajib (WPT) was really unheard of, which probably ended up hurting interop immensely (for SVG and everything), but things were different back then in some regards.
For path parsing tests in Presto (which I used to work on) I think we used our unit-test framework, and Blink does that too although there's likely some stray path data test in LayoutTests/.

So to loop back to the question at hand: WPT is currently best way of fixing this. Submit tests (the more the merrier.)

-- 
GitHub Notification of comment by fsoder
Please view or discuss this issue at https://github.com/w3c/svgwg/issues/335#issuecomment-322924499 using your GitHub account

Received on Wednesday, 16 August 2017 23:11:36 UTC