Fix the Ahem assumption test

[gsnedders]

8624366d45 tried to change the test in line with the modern requirement that Ahem is not a system font.

However, in actuality it just ended up testing that:

<span style="font: 15px/1 Ahem">X</span>

!=

<span style="font-size: 15px; line-height: 1;">X</span>

This test fails in Safari, because unsurprisingly 15px / 15px serif is the same as 15px / 15px Ahem when Ahem is not installed. It seems like it is actually testing that Ahem is installed, or at least is not a no-op.

Instead, simply test that Ahem's X is the same as serif's X, as this should be true without the font installed.

[gsnedders]

Though under our currently documented assumptions, we probably can't actually guarantee Ahem isn't installed, which implies we can't have any test here at all?

[jonathan-j-lee]

Though under our currently documented assumptions, we probably can't actually guarantee Ahem isn't installed, which implies we can't have any test here at all?

Right, I think that's what the Firefox infrastructure/ failure indicates, so this PR's new test is too restrictive as-is. TC installs Ahem as a system font via wpt run --install-fonts.

It seems like it is actually testing that Ahem is installed, or at least is not a no-op.

I suspect the original -notref.html mismatch reftest was intended to check that serif is some human-readable font and not Ahem, which seemed reasonable (e.g., for https://wpt.live users). (Hence why https://github.com/web-platform-tests/wpt/commit/8624366d459e5dbb9fd0fcbe1bfb43c30f657243 keeps the test.) If we still want to check this property, the old ahem.html can be fixed by loading Ahem as a webfont.

However, this is technically not a documented assumption, so as you noted it might be better to not have a test at all.

[gsnedders]

I suspect the original -notref.html mismatch reftest was intended to check that serif is some human-readable font and not Ahem, which seemed reasonable (e.g., for https://wpt.live users). (Hence why 8624366 keeps the test.) If we still want to check this property, the old ahem.html can be fixed by loading Ahem as a webfont.

I guess that's true. It does have some problem about different versions of Ahem — which implies we probably would want to check a single character (like the classic "X").

However, this is technically not a documented assumption, so as you noted it might be better to not have a test at all.

This seems very much a technicality, and I don't think is even worthy of an RFC.