gcc builtins mframe etc.

[Bill Northcott]

On 17/03/2004, at 4:31 PM, Andrew Pinski wrote:
> The reason why __builtin_return/__builtin_apply_args were broken so 
> long were because
> they were added just for GNU's libobjc and really no one should be 
> using them.  I hope
> to remove the usage of these builtins in a future libobjc so they can 
> be removed from
> GCC the next release after the one which uses libffi.
On reflection, perhaps there is more to be said about this.

I am aware that discussions on the gcc lists before about 2001 seem to 
take the above line that the builtins were a temporary which should be 
allowed to die.
However, in the last twelve months the tone of the discussions has been 
that as these are documented functions in all versions of gcc including 
Apple's, they should be fixed.  Indeed in the last three months there 
have been a lot of patches for them checked into the gcc cvs.

These patches are coming from people who do not seem to be active on 
either GNUStep or Swarm.  So I conclude that some other packages must 
be using them.

It may well make sense to discontinue them in favour of libffi, but if 
so, it would be good to get agreement about it and make it clear in the 
gcc documentation that this is intended.  It would also be very helpful 
to have some documentation and maintenance system for libffi.  The only 
docs I can find so far are very out of date and only talk about using 
it within lisp.

My two pennyworth
Bill Northcott