cluts - posix_memalign test

cluts - posix_memalign test

[Rich Felker]

I recently fixed the issue with musl's posix_memalign not returning
failure when the align argument is smaller than sizeof(size_t), but
cluts is still throwing an error in that case:

W:posix_memalign returned EINVAL, but gave a non-NULL pointer (alignment=1)

I believe this error is mistaken. posix_memalign is not giving a
non-NULL pointer to the caller. Rather the caller is seeing the value
it wrote to the pointer before making the call.

I suppose some versions of posix_memalign might clear the pointer to
null if allocation fails, but I could find no text in the standard
that requires or even allows posix_memalign to modify the pointer when
it fails, so at present musl does not touch it.

I'd welcome input from anyone who has a good argument that this is
incorrect, but short of that, I think cluts is checking a condition
that need not (and maybe can't) be true.

Rich

Re: cluts - posix_memalign test

[Luka Marčetić]

On 07/05/2011 02:44 AM, Rich Felker wrote:
> W:posix_memalign returned EINVAL, but gave a non-NULL pointer (alignment=1)
> I suppose some versions of posix_memalign might clear the pointer to
> null if allocation fails, but I could find no text in the standard
> that requires or even allows posix_memalign to modify the pointer when
> it fails, so at present musl does not touch it.
>
> I'd welcome input from anyone who has a good argument that this is
> incorrect, but short of that, I think cluts is checking a condition
> that need not (and maybe can't) be true.
>
> Rich

Technically, it was a warning (prints to stdout and doesn't set an 
error), but you're right. It's what malloc and calloc should do, not 
posix_memalign, so it shouldn't be there in the first place. Removed it. 
Thanks.

Speaking of malloc.c, I changed my mind about rewriting it this week, in 
favor of writing more new code (framework and tasks). The test seems to 
work even though it's awful(ly slow), so it'll have to do for now.
-Luka