@bagder nice optimization. Another way to write this might've been to use `strspn`/`strcspn`. Depending on how many characters end up in the accept/reject set `clang` used to be able to do some pretty impressive optimizations on that (IIRC if the accept set < 64 and in a zone of 64 contiguous chars, then it can be implemented with a shift and bitmask test). Although that might be brittle because you can't be sure the optimization will kick in, and if not then the library function may be slower, also I haven't checked whether that optimization would be possible for the particular accept set here. You could see the optimized assembly and translate it back to C manually though :)

@bagder
Is this due to whatever ISUNRESERVED (a macro, I presume) does?

Replaced with a lookup table?

Not going to tell you how to write, but:

reject_limit = (ctrl == REJECT_CTRL) ? 0x20 :
(ctrl == REJECT_ZERO) ? 1 : 0;

…makes me itch, because years back I started putting the constant as the lvalue of equality tests for that rare/frequent mistake of dropping an '=' and creating an accidental assignment…

I know it messes with readability for some people…