Comment by burjui

One day I needed to parse and generate RISC-V instructions. I wrote a custom bitfield implementation, nothing fancy: no derive macros or any crap like that, only u32 bitfields specified by an array — [(RangeInclusive<u32> /src/, usize /length/, RangeInclusive<u32> /dst/]. One function + a bunch of tests. It worked great. Then I tried the `bitfield` crate, which claims to be as fast as code I would've written by hand. It had dependencies and was two orders of magnitude slower. Basically, it was as fast as code that I would've written by hand in Python.