Re: [commit] new observer.[hc] files

[Daniel Jacobowitz <drow@mvista.com>]

On Fri, Feb 28, 2003 at 11:17:10AM -0500, Andrew Cagney wrote:
> >static void
> >>observer_normal_stop_notification_stub (const void *data,
> >>					const void *unused_args)
> >>{
> >>  observer_normal_stop_ftype *notify = (observer_normal_stop_ftype *) 
> >>  data;
> >>  (*notify) ();
> >>}
> >
> >
> >Is this extra indirection really necessary?  Because I'm 99% sure it
> >won't work on several 64-bit platforms.  Function pointers and data
> >pointers are not required to have the same size; on IA-64 I believe
> >that a function pointer is 128 bits and a data pointer is 64 bits.
> 
> Like the PowerABI?  That has a 32 bit pointer but a 64 bit function 
> descriptor.  void* ends up containing the address of the function 
> descriptor.  Anyway, "defs.h" has the comment:
> 
> /* NOTE: cagney/2000-03-04: This typedef is strictly for the
>    make_cleanup function declarations below. Do not use this typedef
>    as a cast when passing functions into the make_cleanup() code.
>    Instead either use a bounce function or add a wrapper function.
>    Calling a f(char*) function with f(void*) is non-portable. */
> typedef void (make_cleanup_ftype) (void *);
> 
> Once the code gets converted into a SED script, it can always be changed 
> to something more strict.
> 
> >Why not require all observer functions to take the same arguments
> >instead?
> 
> Per the original thread, it is to ensure strongly typed interfaces.

OK, that makes good sense.  Anyway, my concerns about casting a
function pointer to void* seem to have be moot; we can eliminate some
of the grossness when/if this becomes a generated file.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer