I think -1 and 0 are equally bad (as a caller might be unaware that the function also uses the output parameter as an error indicator and may end up calling malloc(utf8_strlen(x))).

I guess a better way would be?
uint32_t utf8_strlen(char *, uint8_t * errcode);

Re-published, thanks.

I did, thanks.

Ok thanks.
I've republished it following your suggestions.

int8_t makes sense - because of simmetry with the test definition.

But why is uint32_t preferred?
Is there any 64bit platform where unsigned int is not 4 bytes long?

Thank you.

I was sure the C standard defines the default signedness (to signed), but I was wrong.
Thanks, changed to int8_t.

I've addressed both issues and re-published. Thanks.

Thanks, I've re-published with a random buffer test (and found a bug in my implementation :)).
Also made the reference solution static.

Sure, but the C source is available anyway, in Wikipedia (one search away).

And using the compiler is besides the point for someone trying to use NASM, no?
I mean most of the NASM katas can be solved in C and translated into assembler using a compiler.

Thanks, i've republished it with 0x%08x.

I don't know, it can work both ways.
Hashing a null terminated string is a special case of the hashing function that works over arbitrary input.

I thought it's better to keep the kata closer to the initial goal - port a simple C implementation to NASM.

You are right.
I think i've fixed it, thanks for you patience :).

Are you reffering to the fact that the function in the description works for binary input (and thus it requests input length) vs the required implementation that assumes input is a null terminated string?

Updated to eax, thanks.

I've changed ub4 to uint32_t, thanks