Zig 2024 roadmap

I feel like there is this inevitable curve every compiler backend goes through where, unless given some external constraint, people will always find one more optimization that's just a tiny bit more expensive but makes huge gains in output quality in some scenario. And if you could just tweak this data structure a little bit they would be so much easier to write. Then you repeat that 100 times and you end up just where GCC and LLVM are with huge, slow data structures built to get every last bit of performance out at the cost of baseline compile times.

I'm a bit more optimistic about cranelift there because it always has to remain usable for interactive use with WASM. Go achieves that constraint by refusing to make a release if the compile time is slower than before. I hope Zig does something similar.