[Patch ObjC/Darwin 1/3] Fix wrong code gen for m64 Objective-C

[IainS]

Hello all,

This patch corrects one of the two most significant outstanding Darwin  
problems;

That is, that we generate incorrect code and meta-data for m64  
Objective-C/C++, rendering these front ends of limited use on current  
darwin/OSX systems.   We also incorrectly assign GNU runtime data to  
NeXT sections when LTO is active at m32.

---

The patch-set has been reviewed and approved off-list for the  
Objective-C, Objective-C++ and Darwin aspects.

It touches only one file outside darwin & objc -
- which is to add two required flags to c-family/c.opt (for which I  
also need approval)

===

As far as when to apply it:

Whilst we are in stage-4, I believe that Richard (G) indicated that  
wrong-code bugs were eligible for inclusion, even at this stage.

If I have understood that correctly, I would ask please might this be  
included in 4.6,
since it would be of considerable benefit to arguably the main targets  
for Objective-C.

I should add that the patch is emphatically _not_ a "last minute fix" ..
.. it has been in progress for some months, it's regrettable that it  
should be ready so late.
(but that is a problem with volunteer contributions, I guess)

===

The patch has been tested thus:
   bootstrap/regtest
     i686-linux (F13)
     x86_64-linux (U10.10) @ m64/m32
     i686-darwin{8,9}, x86_64-darwin10, powerpc-darwin9 *
     cris-elf (simulator).

    * on Darwin 9 and Darwin 10 Objective C/C++ are tested with
    RUNTESTFLAGS="--target_board=unix\{-m32,-m32/-fobjc-abi-version=1,- 
m64} "
   to confirm the NeXT  V1 ABI at m32.

   without regression and with the expected new passes.

   I have also built a version that can be used from XCode and  
verified that the compiler builds

   http://developer.snowmintcs.com/frameworks/sm2dgraphview/
   http://aldebaran.armory.com/~zenomt/macosx/MTCoreAudio/

   for m32 and m64.
   for the first time since Darwin 9 was released :-)

cheers
Iain

====

** All three parts are required, I have split it up for ease of  
potentially different reviewer focus.

====

gcc/c-family

	* c.opts (fobjc-abi-version=) New.
	(fobjc-nilcheck): New.

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: 170111-objc-m64-c-family.txt
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20110214/3b7f7164/attachment.txt>
-------------- next part --------------



gcc

	* config/darwin-c.c (darwin_cpp_builtins): Set __OBJC2__.
	* config/darwin.c (output_objc_section_asm_op): handle sections for
	objective-c ABI 1 and 2.
	(is_objc_metadata): Recognize objective-c meta-data attributes.
	(darwin_objc2_section): New.
	(darwin_objc1_section): New.
	(machopic_select_section): Improve robustness of objective-c section
	selection, deprecate the existing scheme (warn on use).
	(darwin_emit_objc_zeroed): Correct sections for Objective-c items  
placed
	into BSS.
	(darwin_output_aligned_bss): Likewise.
	(darwin_asm_output_aligned_decl_common): Likewise.
	(darwin_asm_output_aligned_decl_local): Likewise.
	* config/darwin-sections.def: Add section definitions for objective-c  
ABI 1
	and 2.
	* config/darwin.h (SUBTARGET_C_COMMON_OVERRIDE_OPTIONS):
	Set the default objective-c ABI correctly for m64 compilation.
	Adjust an over-length line.

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: 170111-objc-m64-darwin.txt
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20110214/3b7f7164/attachment-0001.txt>
-------------- next part --------------