@MikkoMononen @PythagoRascal @aras also, i would imagine that, because HarfBuzz is so widely used & tested, it may handle some corner cases that kb does not...?
In which case, i might not want to depend on kb in the feature-complete version of a shipping product...
*However*, if i'm in a rapid-prototyping phase, or just doing some quick one-off experiments, then i might gladly sacrifice HarfBuzz's completeness in exchange for the relative ease of dropping kb's one small header into my src dir.
Hi!
I'm currently testing it out, and I'm doing comparisons between KB and libRaqm/Harfbuzz. Currently, my performance numbers are a bit off I think.
Do any of you happen to have some benchmark codeand/or results one could have a look at? I'd like to know I've set it up correctly 🤔
@jcash @aras I dont know an existing benchmark, curious to see what you find out.
Do you have a specific metric you want to measure?
libgraphme as bidi/break related benchmarks: https://libs.suckless.org/libgrapheme/
and harfbuzz has shaping related benchmarks: https://github.com/harfbuzz/harfbuzz/blob/main/perf/README.md
Cosmic text also has a benches: https://github.com/pop-os/cosmic-text/tree/main/benches
The cosmic benches will need font fallback setup, which neither of your tested systems support, though. Maybe it could be adapted.
Thanks!
I think I'll look into running the Harfbuzz tests!
My current test om macOS M1 seems it's doing good, but not seeing the same numbers as the author.
https://github.com/JimmyLefevre/kb/issues/21