Improving text rendering

[tdewolff]

Instead of reinventing the wheel, we should be using existing solutions. Especially important to me is using HarfBuzz for text shaping (rendering glyphs correctly in all languages, such as direction, accents, kerning, etc.). See https://github.com/grisha/hbshape for an example of a port.

See https://harfbuzz.github.io/ch01s03.html for what HarfBuzz does, and https://harfbuzz.github.io/what-harfbuzz-doesnt-do.html for what it doesn't do. The latter is what we need to implement here:

• Line-break, word break, justification: Knuths algorithm. Note that "boxes" can also be a path (not just text) to support inline images, and that the text box boundaries may not be rectangular but a path (hard?)
• Breaking text in pieces of different font faces, we already do this
• Solving bidirectionality, no idea, probably use https://github.com/fribidi/fribidi
• Rendering, we already do this but the rasterizer is slow! Why not use FreeType?

For an example see https://github.com/anoek/ex-sdl-cairo-freetype-harfbuzz/. (NB: they don't use FreeType for rasterization?)

Additionally, but with less priority, I'd like to support the TCC file format (how does that pan out with fc-match?). And finally I'd like to support OpenGL rendering using Blinn and Loop algorithm or something similar (see https://lambdacube3d.wordpress.com/2014/11/12/playing-around-with-font-rendering/ as well). This needs proper tesselation of paths which is not that trivial. Low priority.

[mewmew]

Instead of reinventing the wheel, we should be using existing solutions. Especially important to me is using HarfBuzz for text shaping (rendering glyphs correctly in all languages, such as direction, accents, kerning, etc.). See https://github.com/grisha/hbshape for an example of a port.

@tdewolff you may also be interested in checking out github.com/npillmayer/gotype/engine/text/textshaping for text rendering using HarfBuzz.

In general, I think both you and @npillmayer are working on a similar set of problems (check out https://github.com/npillmayer/gotype) and you may consider getting in touch and discuss how to best collaborate your efforts, where applicable. I feel really happy and grateful to have stumbled on both your work today Taco, and the work of Norbert. You are both a huge source of inspiration as I'm just starting to get into type-setting and reading up on this domain with excitement.

Wish you a great start of the summer!

With kindness,
Robin Eklind

[npillmayer]

Hi Robin, hi Taco, I am well aware of Taco’s great work and integrating Canvas is on my todo-list. I have a quick-and-dirty canvas type in place, which was to be replaced as soon as someone would take it on him/her to implement some Cairo-style functionality in Go. Taco’s project is actually far more comprehensive than I could hope to find. The Harfbuzz-adapter I wrote is a quick shot, done and „tested“ in a couple of hours. I just wanted to prove that using HB from Go is doable, and it turned out to be fairly straightforward. If someone would be willing to create a standalone module, i.e. a generic HB interface in Go, I think that would be something to benefit the whole Golang community. I know it’s always a hard decision to include non-Go code, but HB is such a well established standard and solves such a huge amount of challenges for more exotic languages, that rolling your one is not really a feasible option. LuaTeX does it (while XeTeX relies on HB), and it’s… complex. There’s a project by grisha (https://github.com/grisha/hbshape) as a possible starting point. Currently I’m not working on the output layer, but rather concentrate on the higher abstraction layers: HTML/CSS layout, Unicode annexes / text layout, etc. I’ll re-visit the output layer later this year, I guess. That said, thanks for pointing me to Taco and Canvas, Robin! You’re right, canvas and gotype (not the final name) are complementing projects. Taco is obviously high-energy on his OpenSource projects, while I’m pursuing a more chilled out stop-and-go approach, but nevertheless I am looking forward to rebase my output layer onto Canvas. All the best, —Norbert

…

Am 24.05.2020 um 01:25 schrieb Robin Eklind ***@***.***>: Instead of reinventing the wheel, we should be using existing solutions. Especially important to me is using HarfBuzz for text shaping (rendering glyphs correctly in all languages, such as direction, accents, kerning, etc.). See https://github.com/grisha/hbshape for an example of a port. @tdewolff you may also be interested in checking out github.com/npillmayer/gotype/engine/text/textshaping for text rendering using HarfBuzz. In general, I think both you and @npillmayer are working on a similar set of problems and you may consider getting in touch and discuss how to best collaborate your efforts, where applicable. I feel really happy and grateful to have stumbled on both your work today Taco, and the work of Norbert. You are both a huge source of inspiration as I'm just starting to get into type-setting and reading up on this domain with excitement. Wish you a great start of the summer! With kindness, Robin Eklind — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[tdewolff]

@mewmew Thank you for bringing this up! The textshaping repository is definitely interesting and will help. I think people like you are needed more in the community, to bring together developers and save time!

@npillmayer Nice work on gotype! I was thinking along the same lines with Harfbuzz that it'd be a lot of work for nothing to reimplement. Given that Harfbuzz is pure C is really nice, I think we shouldn't be afraid of using a little C to use a industry-standard approach to text visualization. In the same line of thought I was considering adding fribidi, but I'll see how it goes. I might get started with your port of Harfbuzz and work it out a little further (in whichever repo it might end up). I'd be glad to work together with you on this!

I might seem high-energy on my open source projects, but I'm pretty much balancing between a job and relationship too ;-), though the quarantaine has given me extra time I'd otherwise spend socializing haha. Though I really need more hands on canvas to make it better. Anyways, I can always help you out with any problem you encounter while using canvas, as edge-cases may need to get worked on!

Keep it up!

[mewmew]

@mewmew Thank you for bringing this up! The textshaping repository is definitely interesting and will help. I think people like you are needed more in the community, to bring together developers and save time!

More than happy to :)

In the same line of thought I was considering adding fribidi, but I'll see how it goes.

@tdewolff, for bidi you may also look into golang.org/x/text/unicode/bidi. It seems bidi.ReverseString could be used in place of the visual_str return value of fribidi_log2vis. However, fribidi still seems more powerful as it also handles visual positions of characters, so x/text/unicode/bidi only covers a subset of the functionality.

[tdewolff]

I've gone ahead and implemented bindings for HarfBuzz and FriBidi, based on the implementation from @npillmayer, thanks a lot! It's currently in the develop branch and will be tracked in the roadmap in #74

[mewmew]

Very exciting to follow this development! Glad you both continue to facilitate and build upon each others work :)