I still have yet to sit down and properly try Bun, but even from the sidelines everything the team puts out always looks so fun and interesting, and this is no exception! This truly uses C as a “portable assembler”, and I love seeing practical uses for TinyCC. The compilation benchmark in their demo is really impressive!
…but all that said, I’m unconvinced about this. Bun already has FFI for directly calling dynamic libraries, so they’re selling C compilation as a way to help write glue code to bigger libraries. This really only fills the niche where you want a little bit of C code to glue stuff together, but not so much C code that you’d want a more mainstream optimizing compiler like Clang– in which case you’d need to juggle your own pre-built native libraries anyway. I feel they recognized a real pain point with JS tooling (juggling project-specific native code), and shipping a small C compiler to do this kind of native glue is really neat and novel. But I dunno, it feels so limited and just feels like the wrong layer to solve that pain point to me (but full disclaimer: I’m biased because I’m working on tooling that also deals with this kind of problem)
I still have yet to sit down and properly try Bun, but even from the sidelines everything the team puts out always looks so fun and interesting, and this is no exception! This truly uses C as a “portable assembler”, and I love seeing practical uses for TinyCC. The compilation benchmark in their demo is really impressive!
…but all that said, I’m unconvinced about this. Bun already has FFI for directly calling dynamic libraries, so they’re selling C compilation as a way to help write glue code to bigger libraries. This really only fills the niche where you want a little bit of C code to glue stuff together, but not so much C code that you’d want a more mainstream optimizing compiler like Clang– in which case you’d need to juggle your own pre-built native libraries anyway. I feel they recognized a real pain point with JS tooling (juggling project-specific native code), and shipping a small C compiler to do this kind of native glue is really neat and novel. But I dunno, it feels so limited and just feels like the wrong layer to solve that pain point to me (but full disclaimer: I’m biased because I’m working on tooling that also deals with this kind of problem)