Migrate to cosmic-text

[alice-i-cecile]

What problem does this solve or what need does it fill?

Our font rendering is subpar. See #2404 for an example of the rendering problems and #1325 / #1017 for some missing features.

What solution would you like?

Migrate to the well-made, well-supported [cosmic-text] crate that is quickly becoming the ecosystem standard.

What alternative(s) have you considered?

We could try and fix all of these issues in the assorted font crates we depend on.

Additional context

Iced's migration should be informative: iced-rs/iced#1697

[tigregalis]

Some things to consider / possible limitations, when using cosmic_text's existing high level APIs:

It is not clear how to mix spans of different font sizes, see pop-os/cosmic-text#64

You can't currently layout text like this

<div>
  <span>Some text is quite </span>
  <span class="big">big</span>
  <span> but some text is quite </span>
  <span class="small">small</span>
  <span>, and sometimes text has </span>
  <span class="wide-lines">wide lines, </span>
  <span>but other times text has </span>
  <span class="thin-lines">thin lines.</span>
</div>
<style>
div {
  width: 140px;
  font-size: 16pt;
  line-height: 1.5;
}

.big {
  font-size: 32pt;
}

.small {
  font-size: 8pt;
}

.wide-lines {
  line-height: 4;
}

.thin-lines {
  line-height: 1;
}
</style>

It is not clear how you supply your own font (i.e. as bytes, needed to load a font as an asset), see comment https://github.com/iced-rs/iced/pull/1610/files#r1054666431 - Iced seems to tackle this elsewhere as part of its implementation

Perhaps we could reimplement cosmic-text's layout and font-loading functionality using some of its lower level APIs but that doesn't feel great.

[grovesNL]

It is not clear how you supply your own font (i.e. as bytes, needed to load a font as an asset)

You can use load_font_data to load a font from bytes

[tigregalis]

It is not clear how you supply your own font (i.e. as bytes, needed to load a font as an asset)

You can use load_font_data to load a font from bytes

The challenge I had with it was that...

cosmic_text starts with a FontSystem and borrows the FontSystem for all of its high level APIs:

// A FontSystem provides access to detected system fonts, create one per application
let font_system = FontSystem::new();

// A SwashCache stores rasterized glyphs, create one per application
let mut swash_cache = SwashCache::new(&font_system);

// Text metrics indicate the font size and line height of a buffer
let metrics = Metrics::new(14, 20);

// A Buffer provides shaping and layout for a UTF-8 string, create one per text widget
let mut buffer = Buffer::new(&font_system, metrics);

FontSystem holds the fontdb::Database privately.

To add fonts as the application is running (such as when the asset server loads font files) we would need to access the inner Database somehow.

FontSystem only exposes a method to access the Database immutably via db; there is no mutable method. Actually this makes sense given there are multiple borrowers, starting with SwashCache.

Note I am assuming here that FontSystem and SwashCache are long-lived and owned by the ECS.

As a really hacky workaround:

• We could perhaps clone the Database.
• We could then replace the existing FontSystem resource (presuming it is a resource) with a new FontSystem constructed using new_with_locale_and_db.
  • Using that method though we would also have to duplicate the locale checking logic which you don't get out of the box, as you would with new.
• We would also need to replace the SwashCache resource (presuming it is a resource) as well, as it holds a reference to the original FontSystem, but it doesn't seem ideal.
  • Actually can replacing the SwashCache even be done?
  • If it can, can it be done cheaply?
  • Does SwashCache holding a reference to FontSystem play nicely with Bevy's ECS that wants to own and control all of its data?
  • If that's not workable, would we have to effectively reimplement SwashCache?

The easiest solution would probably be to fork cosmic_text and expose the APIs we need exposed, and contribute back upstream if and when they are open to the changes. Otherwise I don't know how you support Bevy's existing text functionality.

[nicoburns]

The easiest solution would probably be to fork cosmic_text and expose the APIs we need exposed, and contribute back upstream if and when they are open to the changes.

FWIW, cosmic-text seems very open to changes in general. They have already accepted PRs from both the Iced and Vizia developers. Discussion of cosmic-text development happens partly in it's github issues, but also partly in this thread in Iced's discord https://discord.com/channels/628993209984614400/1027584473631555604

[hecrj]

@tigregalis The lifetime dependencies will be most likely gone soon (see pop-os/cosmic-text#39), but in the meantime you can use ouroboros to safely deal with self-referential types.

For font loading, you can leverage FontSystem::into_locale_and_db to mutate the Database and then create a new FontSystem with new_with_locale_and_db. Here's how iced does it!

[tigregalis]

@tigregalis The lifetime dependencies will be most likely gone soon (see pop-os/cosmic-text#39),

Great news. That looks exactly like the design I was envisioning to resolve this. Glad to learn someone's already working on it.

but in the meantime you can use ouroboros to safely deal with self-referential types.

In this case, I guess it means we store the SwashCache and FontSystem together under a single resource, at least until then?

For font loading, you can leverage FontSystem::into_locale_and_db to mutate the Database and then create a new FontSystem with new_with_locale_and_db. Here's how iced does it!

Ah, that's a smart workaround. And once pop-os/cosmic-text#39 is merged, with nothing long-borrowing the FontSystem, it should be possible to expose a method to mutate the database directly right?

[jackpot51]

pop-os/cosmic-text#97 has been merged, this should now be possible.

[tancop]

FontSystem::db_mut is now released.

[Zaplime]

Is this migration for 0.11 or should we expect it for another update?

[VitalyAnkh]

I'm working on this, but the progress is slow. Since 0.11 will be released one or two weeks later, it's hopefully done in 0.12.

[argxentakato]

It is very disappointing that we are unlikely to meet the 0.12 milestone. Do you have anything I can contribute?

[alice-i-cecile]

my PR is blocked on my own work on multi-size text upstream in cosmic-text (which is blocked on IRL commitments)
More than happy for you to adopt the PR. There's been a lot of changes upstream in cosmic-text which will simplify this PR too, so that could be a good start.

From the OP on Discord. pop-os/cosmic-text#150 appears to be the linked PR.

Adopting either #8808 or that PR would be incredibly useful: I'm sure the OP will be around to give feedback and review.

[viridia]

@alice-i-cecile Apparently there is a new contender for text layout: https://github.com/dfrg/parley/blob/master/doc/concept.md

[alice-i-cecile]

Neat! IMO for this sort of dependency we should prioritize stability and longevity though, and cosmic-text has a funded open source team with a long history behind it.

[viridia]

Agreed. The reason this popped up on my radar is that I've been chatting with Raph Levien, and apparently the Xilem team is going to be working on this.

[alice-i-cecile]

Ah, that's also a sensible team to rely on. Probably worth evaluating on the merits then.

[alice-i-cecile]

We've opened a new working group to handle this migration, and decide between cosmic-text and parley.

Draft design doc: https://hackmd.io/@alice-i-cecile/rJhG8KR40/edit

Open membership working group (join us!): https://discord.com/channels/691052431525675048/1248074018612051978