Re: probing for compiler attributes?

[Benoit SIGOURE]

On Aug 9, 2007, at 3:29 AM, Joshua Hoblitt wrote:

> On Tue, Aug 07, 2007 at 08:43:54PM -0400, John W. Eaton wrote:
> > Ugh.  Don't do it like that.  Instead, write something like
> >
> >   your-config-header.h:
> >   --------------------
> >   #ifdef __GNUC__
> >   #define ATTR_MALLOC __attribute__((malloc))
> >   #else
> >   #define ATTR_MALLOC
> >   #endif
> >
> > and then in your code, just use
> >
> >   #include "your-config-header.h"
> >   ...
> >   void *myfunc (char *myvar) ATTR_MALLOC;
>
> The more I think about it the more I think this is a bad idea for
> library code.  It certainly makes the code more tidy but doesn't this
> end up screwing any application that tries to link against your  
> code?  I
> don't ever install config.h.

That's true although I've already seen some packages doing this  
(prefixing all the #defines of config.h with the name of the package,  
rename the file itself and install it).
Now I *think* (wild guess :D) that it won't screw anything up at  
least with the malloc __attribute__ because this attribute does not  
change anything WRT the ABI, it merely helps the compiler in doing  
its alias analysis job (which can lead to better optimizations at  
call site).

http://gcc.gnu.org/onlinedocs/gcc-4.2.1/gcc/Function- 
Attributes.html#index-g_t_0040code_007bmalloc_007d-attribute-1869
"The malloc attribute is used to tell the compiler that a function  
may be treated as if any non-NULL pointer it returns cannot alias any  
other pointer valid when the function returns. This will often  
improve optimization." [...]

So, really, I hardly see how this can harm.  Some other attributes  
however would be more problematic.  For instance, I guess that if you  
declare a function __attribute__ ((always_inline)), the symbol for  
that function won't exist in the library compiled, and if someone  
uses a compiler that doesn't support this attribute tries to link to  
your library, the linker may complain that the symbol could not be  
found.

Cheers,

--
Benoit Sigoure aka Tsuna
EPITA Research and Development Laboratory

PGP.sig
Description: This is a digitally signed message part