Re: patch: Darwin/x86 fails to build

[Stuart Hastings <stuart at apple dot com>]

> On Fri, Jan 10, 2003 at 11:20:06AM -0800, Stuart Hastings wrote:
> > How much testing would you require to resubmit the same patch without
> > those two lines?
...

Alas, it took much more than that.  This "new and improved" patch has 
survived
	an FSF-style make bootstrap check ("--enable-languages=c++,objc", with 
no new regressions)
	an Apple-style "fat" bootstrap (PPC and x86 compiler executables bound 
together)

Recall, the current FSF sources won't build a C++ compiler on 
Darwin/x86 because the common "@GOT" syntax chokes Apples' assembler.

2002-01-24  Stuart Hastings  <stuart@apple.com>

	* config/i386/i386.c (x86_output_mi_thunk) Add Darwin/x86 support.

Attachment: patch.x86.mi_thunk.2
Description: Binary data

The call to 'machopic_stub_name()' yields the "$stub" version of the 
identifier, a minor distinction necessary to keep Apples' linker happy 
when such a thunk is included in a ".dylib".  I had some discussions 
with Stan Shebs about this, and he suggested adding a suitable "%M" 
feature to output_asm_insn(), but that seems un-worthwhile unless %M 
was needed in several places.

It's worth noting that if/when Darwin supports 'dlopen()', the 
'machopic_stub_name()' behavior will become the default, and this chunk 
of code could get much shorter.

stuart hastings
Apple Computer