Mapping to aspect-ratio with zero dimension.

[emilio]

https://html.spec.whatwg.org/#map-to-the-aspect-ratio-property uses https://html.spec.whatwg.org/#rules-for-parsing-non-zero-dimension-values, which means that width=0 or height=0 shouldn't map to aspect-ratio, but that doesn't match Chrome's implementation nor the WPT tests.

Given how this should work in CSS is well defined, and that in #4961 we probably want to use the attributes for <canvas> (whose width / height attributes have different parsing rules, https://html.spec.whatwg.org/#rules-for-parsing-non-negative-integers), I suggest the aspect-ratio mapping code should not re-parse the attribute value with different rules.

[emilio]

cc @cbiesinger @tabatkins

[emilio]

This is also relevant because some tokens are valid when parsing as non-negative integers but not vice versa, so dimensions aren't always a super-set of what the attribute will parse.

[emilio]

I'm ambivalent on whether zero should be allowed or not, but we should parse the attributes with the same rules we use for the element (this is relevant for <canvas>).

[annevk]

So interesting test cases here would be:

• +1
• 1.5

Are there any other differences?

[emilio]

10% is valid (and means 10) as per the integer parsing rules, but not valid as the dimension parsing rules.

[cbiesinger]

I agree that the rules should be the same for the two. But what do you mean with "re-parse"? Whether parsing happens twice is just an implementation issue, right?

I'm certainly happy to change the attribute mappings to disallow zero, and to use ParseHTMLNonNegativeInteger for canvas. Or whatever the outcome here is.

[emilio]

I agree that the rules should be the same for the two. But what do you mean with "re-parse"? Whether parsing happens twice is just an implementation issue, right?

Yeah, true. I generally mean "parsing with different algorithms for different purposes". I don't care about parsing multiple times if the result is the same of course :-)

I'm certainly happy to change the attribute mappings to disallow zero, and to use ParseHTMLNonNegativeInteger for canvas. Or whatever the outcome here is.

I don't mind so much about which direction we go in the zero issue, I just noticed it diverged from the spec. But the integer vs. dimension issue I think is worth solving.

[annevk]

Okay, so if we want to parse consistently, we'd indeed move to the zero-allowing dimension parser for img, video, etc. and use the non-negative integer parser for canvas. Is that the proposal?

I can do a PR if that sounds good as I see we already have some tests.

[emilio]

Yes, I think so. That'd be great. Note that the tests in web-platform-tests/wpt#28229 would need to be adjusted to not have the double expectation, but that's a very simple change.

[annevk]

Created #6529.