core: Include a regular subsetted Noto Sans as the default font

[evilpie]

The idea is to include a regular TTF font file for font rendering, instead of first creating a SWF with the Noto Sans as an embedded font. I am using pyftsubset together with a list of unicode ranges matching specific Unicode blocks.
This should make it quite easy to extend the included font in the future. Sadly ttf_parser doesn't support any of the newer compressed formats like WOFF. That is why I decided to manually apply deflate compression, which reduces the size by almost 50%.

Fixes #14761

[evilpie]

Not quite sure what happened here. I can't reproduce this when running the test in Firefox using the extension. (The colors seem wrong anyway)

[kjarosh]

What do you mean? Note that this test clicks the "Green Only" button (see its input.json).

[evilpie]

Oh sorry. Somehow I thought it's obvious. But the = sign (e.g. in rows=5) now looks like just a dash (-).

[kjarosh]

I believe you would have to render it in 1:1 scale and quality 1 to observe that. If we up the quality to 4 the second dash would probably be shown, but it's interesting it's missing on quality 1 🤔

[evilpie]

Oh this looks bad.

[evilpie]

Not sure what is going on here. The test actually includes a DefineFont3 for Noto Sans, but it seems like we might still be falling back to the device font. I actually get the same rendering on desktop (I don't have Noto Sans installed).

[evilpie]

Apparently a known issues: #14713. Thanks @kjarosh

[kjarosh]

This shouldn't be needed after #18437.

[torokati44]

Dropped this change.

[evilpie]

This should be ready to review. I don't see any text rendering change that seems problematic.

[torokati44]

Sadly ttf_parser doesn't support any of the newer compressed formats like WOFF.

Just referencing harfbuzz/ttf-parser#122, in case it's added in the future.

[torokati44]

There are also crates like allsorts and webtype, have you considered relying on them (or anything else)? Not that there's anything wrong with simply deflating (it may be better even, as no new dependency brought in), I'm just curious.

[evilpie]

We need ttf_parser anyway for TLF's DefineFont4, so adding another crate with a different API would just be additional work.

[torokati44]

(Just a side-note: I keep getting confused whether "regular" means "not light and not bold" or "just a .ttf, not a blob extracted from a .swf"; and whether "subsetted" means "subscript" like ₜₕᵢₛ, or "not all unicode characters included" here. I assume it's the latter for both words though.)

[Dinnerbone]

Looks good to me but I'd really like to consider using metric compatible fonts at some point.