On Thu, 2007-03-22 at 12:22 +0100, Jim Meyering wrote:
David Edmondson <dme@sun.com> wrote:
On Wed, Mar 21, 2007 at 04:38:00PM +0100, Jim Meyering wrote:
I interpret "wrappers", above, to mean more than just a calloc-like wrapper.
A malloc (not calloc, of course) wrapper that always initializes can mask what would have otherwise been a used-uninitialised error, and what would still be a logical U.I. error.
That seems silly. If the wrapper is defined as zero-initalising then it cannot be an error to assume that it zero-initalises.
What seems silly? A malloc() wrapper that initializes the memory it allocates? That's the case in which errors can be masked. A function intended to be used as a malloc or realloc replacement should not initialize its memory -- at least not by default. A calloc-wrapper _must_ do that. Not the others.
Wow, this is a weird little side-track :-) I think the confusion is just around what the term "malloc() wrapper" means. I mean it purely as "a memory allocation function", you seem to interpret it as "a function which overrides the malloc() symbol and has the same semantics as malloc()". e.g. I would define g_malloc(), g_malloc0(), g_new() and g_new0() etc.[1] as malloc() wrappers ... but you wouldn't? They have different semantics from malloc() and some of them have the semantic "zero initialise the newly allocated memory". So, sure, a zero-initialising replacement for the malloc(3) symbol would be stupid. Cheers, Mark. [1] - http://developer.gnome.org/doc/API/2.0/glib/glib-Memory-Allocation.html