Re: Warnings in mingw64 builds on master

[Paul Eggert <eggert@cs.ucla.edu>]

On 8/16/20 8:39 AM, Eli Zaretskii wrote:
> the warning it emits has nothing to
> do with correctness of the code.

I'm not sure I follow. Although it's true that adding __attribute__ ((malloc)) 
does not turn formerly-incorrect code into correct code, the same thing is true 
of other function attributes like _Noreturn. But that doesn't mean we shouldn't 
use function attributes.

The main reason I like __attribute__ ((malloc)) has little to do with improving 
performance (is that what you meant by "improving the code"?); it has to do with 
understanding the code and using that understanding to avoid and/or fix bugs. 
When I see code like this:

    char *p = xmalloc (s);
    memcpy (p, q, r);

I mentally have to answer questions like "is r <= s?" because otherwise the 
behavior is undefined. But I do not have to mentally answer questions like "are 
*p and *q aliases?" - even though the behavior of memcpy would be undefined if 
they were aliases - because malloc has __attribute__ ((malloc)) and this means 
*p cannot alias with anything else.

This is not simply a matter of memcpy vs memmove. In C one must constantly worry 
about aliasing. These worries are lessened with __attribute__ ((malloc)), so the 
attribute is typically a win.