Alpha-tested rasterization isn't as good as FreeType's monochrome

[robbie01]

Here is a comparison of FreeType (via freetype-rs) and swash (via cosmic-text) using the free Virtue font, rendered at a size of 12 px and scaled by a factor of 4x. FreeType (on the top) is using LoadFlag::TARGET_MONO to generate a 1-bit bitmap using monochrome-optimized hinting, whereas swash (on the bottom) is using an alpha test where alpha >= 0x85. (0x86 looks identical. Lower thresholds cause the comma to look blocker, and higher ones open a hole in the bottom of "a".)

Note carefully the renderings of C, R, A, G, u, a, y, and d.

My apologies if aliased monochrome rasterization is out of scope here.

[robbie01]

Here's a test with the original Charcoal TTF from Mac OS 9.2.2. The threshold here is 0x80, but there were no values that yielded particularly satisfactory results. This demonstrates monochrome-optimized hinting in additioned to the aforementioned rasterization. (For what it's worth, this font seems to be a bit problematic overall, as FreeType spaces capitals differently than the Mac OS 9 renderer.)

[robbie01]

Finally, here's a test with Chicago, again from Mac OS 9.2.2. This actually seems to yield identical results to FreeType over a wide range of thresholds, from 0x84 to 0x8D, inclusive.

[valadaptive]

I think this is because skrifa (which swash now uses for hinting and drawing glyphs) does not currently implement horizontal hinting. I'm not sure if that's distinct from monochrome-optimized hinting, but it's part of the way there.

If/when googlefonts/fontations#1496 is merged, the relevant flag could probably be plumbed through and exposed here.