Trimming of whitespace in attribute

[fred-wang]

This was mentioned in w3c/mathml#63

From https://w3c.github.io/mathml/#syntax-notation-used-in-the-mathml-specification :

"Since some applications are inconsistent about normalization of whitespace, for maximum interoperability it is advisable to use only a single whitespace character for separating parts of a value. Moreover, leading and trailing whitespace in attribute values should be avoided."

Currently MathML Core does not say anything about leading/trailing whitespace in attributes. I think we should be explicit about this. Do we allow them or not?

The non-normative RelaxNG schema is inconsistent ( https://github.com/mathml-refresh/mathml-schema/tree/master/rnc ), sometimes they are allowed, sometimes they are not.

Reminder: In w3c/mathml#21 we agree to align definition on HTML5/CSS.

[fred-wang]

Regarding the "separating part of value", I guess we should rely on definitions from https://html.spec.whatwg.org/multipage/common-microsyntaxes.html#space-separated-tokens (see w3c/mathml#104 for menclose, which I think is the only one in mathml core?). These definitions explicitly says they may have leading or trailing ASCII whitespace.

https://html.spec.whatwg.org/multipage/common-microsyntaxes.html#comma-separated-tokens also has "optionally surrounded by ASCII whitespace".

[fred-wang]

We currently have a few attributes implemented in the UA stylesheet:
https://mathml-refresh.github.io/mathml-core/#user-agent-stylesheet

(From our experience in Chromium implementation, most of the rules can probably be moved to the C++ code, except maybe the one about accentunder="true" for math-script-level ?)

I don't see anything in https://drafts.csswg.org/selectors-4/#attribute-selectors that would allow trailing/leading whitespace.

Note: the "i" makes values case insensitive https://drafts.csswg.org/selectors-4/#attribute-case

[fred-wang]

So my proposal for now is to modify

https://mathml-refresh.github.io/mathml-core/#types-for-mathml-attribute-values

to be explicit that "length-percentage" and "color" attributes accept optional ASCII whitespace (https://infra.spec.whatwg.org/#ascii-whitespace) at the start/end. This matches the RelaxNG MathML3 schema as well as Gecko/WebKit/Blink implementations.

For other attributes the situation is a bit more inconsistent, so let's wait a bit before making the syntax less strict.

[fred-wang]

Attributes that can't handle whitespace trimming:

mathml-core/user-agent-stylesheet$ grep '=' *
math.css:math[display="block" i] {
math.css:math[display="inline" i] {
mtable-frame.css:mtable[frame="none" i] {
mtable-frame.css:mtable[frame="solid" i] {
mtable-frame.css:mtable[frame="dashed" i] {
scripts.css:munder[accentunder="true" i] > :nth-child(2),
scripts.css:mover[accent="true" i] > :nth-child(2),
scripts.css:munderover[accentunder="true" i] > :nth-child(2),
scripts.css:munderover[accent="true" i] > :nth-child(3) {
scripts.css:mover[accent="true" i] > :first-child,
scripts.css:munderover[accent="true" i] > :first-child {

[fred-wang]

@davidcarlisle I believe this one should still be transferred to mathml-core.