[css-fonts-3] Dropping "synthesizing super/sub-scripts" requirement

[svgeesus]

Fonts 3 says " In the case of OpenType fonts that lack subscript or superscript glyphs for a given character, user agents must synthesize appropriate subscript and superscript glyphs. ".

In the features at risk for the March 2018 CR:

Synthesis of the ‘font-variant-position’ property

So, how to handle dropping this?

1. Remove the sentence
2. Change must to should or even may

[fantasai]

I think this is an important requirement: losing the synthesis backup means a page's content can become garbled, either incomprehensible and/or incorrect. We need to get implementations to fix it, and I would be uncomfortable going to REC with the font-variant-position feature if it's not implemented safely.

[svgeesus]

Notice that the requirement at-risk here is not "synthesizing super/sub", which implementations have done since forever. Instead, if an OpenType font has special sup/sub glyphs for some, but not all, of the characters that a given page needs, the requirement is to synthesize all of them rather than using a mix of some synthesized and some designed sup/sub glyphs (which is readable but ugly).

In the figure below, the font has a superscript 2, [ and ] but does not have a superscript a. In red is the naive but ugly implementation where these are both superscript, but do not fit well together. The central part shows the designed superscript 2 not being used, so that the superscript 2a is consistent.

[svgeesus]

Fonts 3 introduced the font-synthesis property, covering fake-bold and fake-italic/oblique. Fonts 4 extends this to fake-smallcaps. Perhaps a better way to deal with this would be to add fake sub/sup to font-synthesis in Fonts 4?

[fantasai]

OK, that's weird and bad, but acceptable. Making it a SHOULD to keep an entire text run either synthesized or not for L3 is fine, as long as synthesizing missing glyphs at least individually is required.

[svgeesus]

OK so, change must to should.

[frivoal]

I am not sure I agree. Change the order of 2 and a in the example above, and the read case then reads like 2^(a^2), which seems it would be confusing and a source of misunderstandings.