Hi @dnkl, for a project (#X11 with #XRender using #xcb, still in experimentation phase), I was looking for a sane way to render #SVG fonts with #freetype. Looking for a sane library to do the rasterizing, I finally found the solution (#nanosvg 👍) in your #fcft project (thanks!).

I tried to "steal" your metrics calculations at first, but the result was completely broken with the "Twitter Color Emoji" font, so I started over (meanwhile ditching most floating-point stuff in favor of freetype's fixed-point formats), and now the result looks sane. In case you're interested, here's my nanosvg/freetype adapter:
https://github.com/Zirias/xmoji/blob/master/src/bin/xmoji/svghooks.c

But of course I have doubts now, maybe the font I'm testing with is "unusual"? Which fonts do you test with? Thanks!

#NanoSVG is a tiny #SVG #library.

NanoSVG is a parser and rasterizer for scalable vector graphics (SVG) data. NanoSVG loads SVG into a tree of objects composed of Bézier curves, which can be iterated through. Coordinates are converted into uniform coordinates specified by the user. NanoSVG's rasterizer renders simple flat SVG into pixel data.

Website 🔗️: https://github.com/memononen/nanosvg

#free #opensource #foss #fossmendations #programming #graphics

My project #fnott (notification daemon) uses #nanosvg to render SVG icons.

I've always thought the rendered icons looked bad, compared to the ones rendered by librsvg. I chalked it up to nanosvg being "simple stupid" [sic].

Turns out /I'm/ stupid.

Nanosvg produces RGBA output that is NOT pre-multiplied.

But, pixman expects pre-multiplied color channels.

Blend a non-premultplied image as if it was pre-multiplied, and the result is, well, crap.

https://github.com/memononen/nanosvg