Re: proposed new module careadlinkat (renamed from breadlinkat)

[Ben Pfaff]

Paul Eggert <address@hidden> writes:
> +struct allocator
> +{
> +  /* Call MALLOC to allocate memory, like 'malloc'.  On failure MALLOC
> +     should return NULL, though not necessarily set errno.  When given
> +     a zero size it may return NULL even if successful.  */
> +  void *(*malloc) (size_t);
> +
> +  /* If nonnull, call REALLOC to reallocate memory, like 'realloc'.
> +     On failure REALLOC should return NULL, though not necessarily set
> +     errno.  When given a zero size it may return NULL even if
> +     successful.  */
> +  void *(*realloc) (void *, size_t);
> +
> +  /* Call FREE to free memory, like 'free'.  */
> +  void (*free) (void *);
> +
> +  /* If nonnull, call DIE if MALLOC or REALLOC fails.  DIE should
> +     not return.  */
> +  void (*die) (void);
> +};

It might be a good idea to mark each function with appropriate
GCC attributes, e.g. __attribute__((malloc)) for malloc and
realloc and __attribute__((noreturn)) for die.

I was a little surprised to see the first proposed use of this
actually copy out all of the pointers into local variables.  If
that's the way it's going to be used, I guess that would make
the attributes useless, unless they were applied to the local
variables too.

I'm not sure why die is a separate member.  Couldn't xmalloc and
xrealloc simply be used as the malloc and realloc functions?

I would have guessed that there would be a global instance of
this allocator, something like:

struct allocator standard_allocator = { malloc, realloc, free, NULL };

and possibly another one that uses the x* functions:

struct allocator xstandard_allocator = { xmalloc, xrealloc, free, NULL };

or 

struct allocator xstandard_allocator = { malloc, realloc, free, xalloc_die };

With global instances, functions can pass in one of those instead
of a null allocator, and then the functions that call into the
allocator don't have to have special cases.
-- 
Ben Pfaff 
http://benpfaff.org