Re: [patch]: Sandro's GCJ on Darwin/i386 patches pt.2

[Sandro Tolaini <tolaini at libero dot it>]

On 16/set/2006, at 21:20, Geoffrey Keating wrote:

> Hi Jack,
>
> 2006-03-28 Sandro Tolaini <tolaini@libero.it>
>
> 	* configure.ac: Enable Darwin/i386 as a supported target
> 	* src/x86/ffi.c: Port to Darwin/i386
> 	* src/x86/sysv.S: Likewise
> 	* src/prep_cif.c: Likewise
>
> The ChangeLog entry for prep_fic.c should specify which routine is
> being altered.
>
> I think this patch needs some re-work to ensure it doesn't affect
> other platforms.  The prep_cif change looks like it would affect
> powerpc-darwin; is this intentional?  The sysv.S changes look, in
> part, like they would affect all x86 platforms.  It also looks like
> the sysv.S changes need a corresponding change to the DWARF
> information, do they?

The affected function in prep_cif.c is ffi_prep_cif. You are right:  
the change would affect PPX/Darwin. Attached is the modified patch.

The sysv.S changes that are not platform specific are needed because  
the original source had a bug in passing structs smaller than 4 bytes  
in a reg instead of on the stack. The bug never show up on other  
platforms because only x86/Darwin use this modification to the  
original SYSV specs. Other platform should not be affected because  
they never get the FFI_TYPE_SINT* tyep value, only FFI_TYPE_STRUCT.

About exception unwinding DWARF infos, there should be no needed  
changes IMHO.

Cheers,
  Sandro Tolaini

Attachment: libffi.patch
Description: Binary data