Re: [RFC] Supporting malloc_usable_size

[Siddhesh Poyarekar <siddhesh@gotplt.org>]

On 2022-12-07 11:54, Adhemerval Zanella Netto wrote:
> I tend to agree with full deprecation, systemd MALLOC_SIZEOF_SAFE is an
> example of what user needs to hack around with an ill defined interface and
> _FORTIFY_SOURCE=3 warning should be clear that keep it is not a gain in long
> term (besides adding even more trouble for portability).
> 
> Another possibility would to make malloc_usable_size always return the malloc
> size argument (although it would add some internal metadata costs).
> 

Do we want to go down this path?  I was quite reluctant until I 
remembered that there's P0401R6 for C++ that would need this kind of 
information for support anyway.  The intent of that paper is different, 
it's to optimize for realloc and that would actually need us to endorse 
the chunk size as the actual size when we return it.  That would then 
disallow chunks from being consolidated, thus making usable size 
consistent through the lifetime of the object.

Basically, it looks like one way or another, there's value in making the 
return value of malloc_usable_size consistent and maybe even usable for 
writes if we are to support something like P0401R6 in future and 
instead, help the compiler see this usable size.

Thoughts?

Thanks,
Siddhesh

[1] 
https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2021/p0401r6.html#deallocate