Convert strlen calls to strnlen

[tmatth]

• For strlen(x) > c comparisons, use strnlen(x, c + 1) > c (so O(1) instead of O(n))
• For strlen(x) < c comparisons, use strnlen(x, c) < c, so (so O(1) instead of O(n))

I didn't change strlen(x) > 0 comparisons (or any other comparison to 0) since compilers already optimize away the strlen call by transforming those to e.g., x[0] != '\0', see https://godbolt.org/z/KGW359ca1

N.B. Some of these calls are still followed by strlen calls, but this is always in the error and/or logging case (not the most likely path).

Inspired by https://code.videolan.org/videolan/vlc/-/merge_requests/2477/diffs

[lminiero]

I wasn't aware of this function, thanks! Is it portable enough as strlen is, or do we risk this breaking compilation in some places where it now works? Unfortunately it doesn't look like there's a g_strnlen we can piggyback to 🙂

[tmatth]

I wasn't aware of this function, thanks! Is it portable enough as strlen is, or do we risk this breaking compilation in some places where it now works? Unfortunately it doesn't look like there's a g_strnlen we can piggyback to slightly_smiling_face

It's required for POSIX so I think officially supported platforms are ok.

[atoppi]

Always cool stuff from @tmatth 😎
Kudos for the online service compiling the code in real time with different compilers and flags.
That's brilliant.

I can see the enhancement here, but being pedantic

For strlen(x) > c comparisons, use strnlen(x, c + 1) > c (so O(1) instead of O(n))

AFAIU that will not be O(1), but O(c+1).
It would theoretically be O(1) if C had a string representation like e.g. Golang, with a "length" attribute.

[tmatth]

AFAIU that will not be O(1), but O(c+1). It would theoretically be O(1) if C had a string representation like e.g. Golang, with a "length" attribute.

Where c is a constant that is bounded by a value that does not depend on the size of the input it simplifies to O(1).

But you're right that it will take up to c+1 operations 😆

[atoppi]

mh looks like strnlen has been excluded from C23 standard, I'm wondering if that might be a problem in future

[tmatth]

mh looks like strnlen has been excluded from C23 standard, I'm wondering if that might be a problem in future

That's an interesting find. I think it's ok since it's already not part of the standard, but won't be dropped from POSIX. If it does at some point become a problem it'd be trivial to add a fallback using memchr (but I wouldn't use memchr directly since it's a bit less readable).

[lminiero]

Thanks for the discussion and the contribution, merging then!