WARN Caps are averagely centred — v metrics check

[RosaWagner]

capsHeight - hhea/typoAscender ~= - hhea/typoDescender (this is only recommended for fonts with Latin, Greek or Cyrillic as "main" script, I would go for allowing an error margin of 5% of UPM)

We also had the experience with Braah One: google/fonts#6227
So also in case of non-caps fonts, we could have a similar rule.

[eliheuer]

This would be a great check, there is a twitter thread going over the rationale in detail here: https://twitter.com/romanshamin_en/status/1562801657691672576

[eliheuer]

I made a pull request for a new check covering this issue here: #4226

[tphinney]

tl;dr: you might want to do more testing and investigation before implementing that check?

I just want to point out that there is a big assumption behind this, which I am very skeptical of: that all relevant environments people care about, do vertical centering of text using the same algorithm.

I am especially a bit suspicious of the implied claim that em-square as a distance from the origin is one of the metrics in question. I would also be curious as to WHICH descender metric is relevant, or whether that might vary as well.

In the case of Google’s own Flutter web framework, it appears to rely on sTypo metrics including sTypoAscender, and also possibly on… the baseline (?!). But I did not get enough feedback to be sure what was going on. See
google/material-design-icons#1500

[eliheuer]

@tphinney Thanks for the feedback! I made a codepen here to try showing why this is a real issue: https://codepen.io/eliheuer/pen/KKrOamg

This kind of vertical centering works on the Web (Chrome) which I would argue is the most relevant environment. And to me it also just intuitively makes sense that you would want to center things this way.

This is only a WARN, so no one is forced to do this. But maybe this is a good best practice if you don't have any reason to do otherwise?

Here is Martian Mono, which is perfectly centered (Top: 42px, Bottom: 42px):

Here is Shantell Sans, which is NOT perfectly centered, but within the threshold to pass this test (Top: 46px, Bottom: 46px):

Here is Amiri, which is off by enough that it would get a WARN, note that this is visibly not centered (Top: 54px, Bottom: 62px):

[tphinney]

Clarification: I was writing about Roman Shemin’s Twitter thread and his version of the vertical metrics check, which is significantly different than what Rosa proposed. Which one are you doing?

[eliheuer]

I tried to implement the check as described by Rosalie, including the error margin of 5%.

    upm = ttFont["head"].unitsPerEm
    error_margin = upm * 0.05
    cap_height = ttFont["OS/2"].sCapHeight
    ascender = ttFont["hhea"].ascent
    descender = ttFont["hhea"].descent
    top_margin = ascender - cap_height
    difference = abs(top_margin - abs(descender))

    vertically_centered = difference <= error_margin

[felipesanches]

I understand that the approach described on the Roman Shemin’s twitter thread can be useful. But isn't it too opinionated to expect that all fonts in the world should follow the same approach?

In other words, is this proposal meant to result in a WARN-level universal profile check offering a recommendation for this specific (and optional) approach for vertical metrics? Is everybody here OK with including this recommendation on the universal profile?

[m4rc1e]

I don't like the use of sCapHeight since it's often set incorrectly. I don't mind this check being included but only as a warn. It's definitely something that should only apply to fonts intended for UI and only include scripts which have similar proportions as Rosa mentioned (Amiri shows exactly why this doesn't work for most Arabic + Latin fonts).

[khaledhosny]

I feel like whatever proposed here is overly Anglo-centeric. You don’t even need Arabic, try some accented text and the perfectly centered fonts are no longer centered. This feel like trying to solve a non-font issue in fonts, since centering depends on the text and the container and it is something should be handled in CSS not in fonts.

[felipesanches]

I agree with @khaledhosny. To me this proposed check looks like trouble.

[felipesanches]

The in-person conversations at TypeCon 2023 convinced me it is a reasonable check.

[tphinney]

Yeah, as long as it is Rosalie’s version and not Roman Shemin’s it is… not crazy. 😛

[eliheuer]

This issue is a great example of this problem: mui/material-ui#29965