That's all carefully reasoned. Thank you. I only have one
remaining comment.
Paul Eggert <egg...@cs.ucla.edu> writes:
> On 04/01/2011 09:35 AM, Ben Pfaff wrote:
>> I would have guessed that there would be a global instance of
>> this allocator, something like:
>>
>> struct allocator standard_allocator = { malloc, realloc, free, NULL };
>
> I tried that at first, but it was more awkward for the common case
> where the standard functions are being used, because users of
> careadlinkat had to say "&standard_allocator" and they had to
> #include <allocator.h>. Under the proposal they just say "NULL" and
> don't need to include <allocator.h>. It's a minor point and I could
> easily be persuaded to switch back.
It's not important to me, at least.
Another choice would be to continue to accept NULL and just add
something like:
if (!alloc)
alloc = &standard_allocator;
or (probably with better choice of variable names):
struct allocator *real_alloc = alloc ? alloc : &standard_allocator;
to the functions that use it. Whether this is an improvement, in
my opinion at least, would depend on whether there are going to
be many functions that use this allocator interface. If there
aren't, I can't see that it matters much either way. If there
are, then it's not going to be nice to have
void *(*pmalloc) (size_t) = malloc;
void *(*prealloc) (void *, size_t) = realloc;
void (*pfree) (void *) = free;
void (*pdie) (void) = NULL;
if (alloc)
{
pmalloc = alloc->malloc;
prealloc = alloc->realloc;
pfree = alloc->free;
pdie = alloc->die;
}
cut-and-pasted into each one of them. Too much code duplication,
in my opinion.
--
Ben Pfaff
http://benpfaff.org
