Thanks, looking back it's most probably too clever. Negative detection could simply be xneg = (x < 0); more readable and code generation would be same or better because it wouldn't confuse optimizers about the intent.

You're welcome :D
But it's optimization theater, really. Most compilers optimize simple ?: expressions to branchless. I wrote it this way bc I was bored.

meh, tabs vs. spaces strike again..

Don't think so. Why would it?

Just checked on Godbolt for GCC x86 and Clang Arm64, -O3 and they don't.

Even if they did, it still would be bad practice, there's no reason to add such cost in debug mode.

Besides, copy-pasting strlen(str) 3 times instead of putting it in one variable looks awful.

Call strlen for each iteration? How is this best practices??

It runs the same as n%2 because compilers turn it into a bit test, it only runs faster with optimizer turned off. I used it because once you're familiar with bit ops, it becomes a self-explanatory idiom.

I accidentally left a printf; ignore this solution and look for my other one.

Yeah, well it's there because I wanted the whole upper 64 part of the ascii table in the map for a simpler masking condition. Although I don't know whether masking improves any performance over just (*s >= *map).

Yes, if returning a static label. It points to program data segment.

Any sane reason why use 32 instead of a char literal ' '?

It'll fail on platforms where strings can be placed in low memory. Besides it's slow: chooses overload at runtime instead of compile-time.

Very elegant

Yeah, how returning pointer to stack got top score in best practices is beyond me...

Should return an int, not pointer.