Re: [libobjc] Rewrite message forwarding to use libffi

[Mark Mitchell <mark at codesourcery dot com>]

Alexander Malmberg wrote:

> What ffi type should I map long long ints to? ffi has macros for all 
> other basic int types, but not for long long ints. Currently, I assume 
> that long long ints are always 64 bits and map to ffi_type_[su]int64.

You should at least add an assertion and comment to that affect.

> Unfortunately, the patch also causes four new failures compared to a 
> clean tree.

I think those issues need to be resolved.

> I don't know how this will affect other platforms. I'd expect the 
> ffi-based forwarding to work just as well on all platforms supported by 
> libffi, but I don't know what to expect from __builtin_apply on other 
> platforms, or if there are platforms unsupported by libffi where 
> __builtin_apply works well. However, from what I understand, 
> __builtin_apply is pretty evil. :)

I think you are asking the right questions, but you need to get the 
answers.  Which platforms do not support libffi, and are those platforms 
important?

> Mark: Andrew told me to tell you that this is a regression, that 
> rewriting libobjc to use libffi is the easiest way of fixing it, and 
> that he's approved the libobjc parts.
>
> OK for mainline?

I've delegated decisions like this to the appropriate maintainers, so I 
think it's up to the Objective-C people to make a decision.  However, I 
would encourage them to be conservative.  My personal opinion is that 
this patch is a great step in the right direction, but doesn't quite go 
far enough.  For example, on platforms where libffi is not supported, 
perhaps you should fall back to the original mechanism.  Perhaps you 
should fix the double/long double problem.  In other words, I feel like 
there are some loose ends that should be tied off before this patch is 
checked in.

--
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304