Re: GSoC 2020 - Porting font-rs rendering engine to FreeType

[Raph Levien]

I agree that for FreeType it makes more sense to port the algorithms to C than to complicate the build system by bringing in Rust. (I think Rust has a lot of potential for this kind of work, but I see that as a quite separate project).

I have come up with an extremely good algorithm for cubic to linear conversion, it's at https://levien.com/tmp/flatten.html. The lyon people have ported this to Rust, see https://github.com/nical/lyon/pull/552. It's based on ideas from my blog post https://raphlinus.github.io/graphics/curves/2019/12/23/flatten-quadbez.html but goes beyond it. I believe that this algorithm is very near optimal in both speed and quality. If selected as a mentor, I will write it up as a paper (ideally for submission to jctg), and there would be an opportunity for the student to coauthor that.

Experience on resvg is a good qualification for this work :)

Raph

On Wed, Mar 25, 2020 at 7:42 AM Werner LEMBERG <address@hidden> wrote:

Hello Moazin,


> I am applying to GSoC this year as well and I am particularly
> interested in the project "Port the font-rs/fontdue rendering engine
> to FreeType".

Great!

> Firstly, I saw that the project obviously requires knowing Rust.
> While I am not a Rust programmer, I do know many of the basics due
> to my experience with `resvg' last year.  I'm able to easily read
> and understand Rust code.  I'm also going to be improving my
> knowledge of Rust as I further prepare the proposal.  I guess this
> should be sufficient?

Yes, I think.

> [...] The actual raster part is mostly in `src/raster.rs' which is
> quite small.  Also, creating a C wrapper for Rust code is an easy
> process (compared to other programming languages).  Given the small
> size of the rendering engine it seems that implementing it from
> scratch in C inside FreeType isn't too hard either.  Though I don't
> know if there is any advantage to doing that? The dependency issue
> maybe?

Please always have embedded systems in mind.  On such platforms it's
rather unlikely to find a Rust compiler, thus a conversion to C sounds
sensible.

> Another thing I saw was that this engine can't handle Cubic Bezier
> curves, only Quadratic.  FreeType does support font formats which
> use Cubic Bezier curves. So maybe I can add this functionality to it
> as part of this project.  (unless there is something in the
> algorithm that can't be applied on Cubic curves or other font
> formats in general)

I think this would be a very useful addition.  Raph?

> I think that porting can be done by:
>
> 1. Creating a C wrapper for font-rs (for the rendering part at
>    least).

Well, *only* the rendering part, since the rest is done by FreeType
already.

> 2. Creating a new rendering module in FreeType that can provide the
>    necessary "glue" code.

Yes.  As mentioned above, a conversion to C would be useful, too.

I don't have any special tips or requirements for a proposal. :-)


    Werner