Re: [AD] malloc

[Peter Wang <tjaden@xxxxxxxxxx>]

On 2005-11-18, aj <aj@xxxxxxxxxx> wrote:
> allegro src code seems to have many instances of malloc() thruout the code.
> there is also a wrapper for malloc in allegro.c (line646)
> 
> void *_al_malloc(int size)
> {
>   return malloc(size);
> }
> 
> 
> is there any reasons why all those malloc() calls could not use the 
> _al_malloc()  ?

No.

Anyway, we need a design for user-specified malloc functions.  Grzegorz
already had one here:

    http://alleg.sourceforge.net/future/custom_memory_management.txt

I would only add:

    void *al_malloc_atomic(size_t);

It is to be used for allocating blocks of memory which are guaranteed
not to contain any pointers.  This is useful for using conservative
garbage collectors.  The object being worked on by al_realloc() will be
taken to be atomic iff the old object was atomic.  These are the
semantics used by the Boehm GC library.

I would consider deleting the `opaque' parameter from his proposal.
It doesn't seem that useful and it breaks the function signatures of
malloc/free/realloc.

This is for 4.3, but part of the patches can probably be backported to 4.2.

Peter