RFC: __attribute_alloc_size__ on allocation functions (BZ#23741)

[Paul Eggert]

DJ Delorie wrote:
> Perhaps a good compromise would be to retain current behavior for 32-bit
> targets (where allocations bigger than 2Gb are realistic) but limit them
> to PTRDIFF_MAX on 64 bit targets, (where a 5 exabyte allocation is not
> reasonable).

I'm not sure I see the point of doing it that way. A 2**63 allocation already 
fails now. The main practical problem here is 2**31 allocations, which can cause 
applications to fail later in mysterious ways on 32-bit hosts. We've by and 
large insulated GNU core utilities against this problem, but it was a bit of a 
pain (and the job is not completely done) and it'd be better if glibc did it for us.

Perhaps a memory-allocation option of some sort? The default should be a 
PTRDIFF_MAX limit, but applications needing larger allocations could set the 
option first.