gmp-5.0.0 fails to build on x86_64-apple-darwin10

[Jack Howarth]

On Fri, Jan 15, 2010 at 05:08:11PM +0100, Torbjorn Granlund wrote:
>   > I am not aware of any external patches affecting ABI=32 except the one I
>   > wrote and mailed yesterday.  And yes, my own patches do have my blessing.
>   > :-)
> 
>   Torbjörn,
>       I believe David is referring to the patch I proposed in...
>   
>   http://gmplib.org/list-archives/gmp-bugs/2010-January/001787.html
>   
> That patch does not affect an ABI=32 build.  (I am not trying to be a
> nitpick, just trying to make sure I have not missed to evaluate a
> proposed patch.)
> 
>   Unlike the previously proposed patches for this problem, this one
>   eliminates all testsuite failures and the changes are contained to
>   darwin to work around their assembler bugs.
> 
> Yes, it is crucial for ABI=64.
> 
> I believe that all Mac-related problems are now fixed in the repository
> (http://gmplib.org:8000/gmp/) except this one:
> 
>   http://gmplib.org/list-archives/gmp-bugs/2010-January/001746.html
> 
> I have looked at that only briefly, and have failed to reproduce it on
> other systems even with the compiler in C99 mode.
> 
> It is crucial that you grab a snapshot and check that things now indeed
> work.  Snapshots are available from ftp://ftp.gmplib.org/pub/snapshot/.
> The first snapshot that will have these fixes is gmp-5.0.0-20100116.tar.lz.
> 
> -- 
> Torbjörn

Torbjörn,
    The current hg pull of gmp trunk still shows a testsuite failure
on core2-apple-darwin10 for the ABI=64 build...

/bin/sh: line 1:  1625 Segmentation fault      ${dir}$tst
FAIL: t-fat

If I use the change...

--- mpn/x86_64/darwin.m4.org	2010-01-15 20:58:05.000000000 -0500
+++ mpn/x86_64/darwin.m4	2010-01-15 20:58:20.000000000 -0500
@@ -19,7 +19,7 @@
 define(`DARWIN')
 
 define(`LEA',`
-	mov	$1(%rip), $2
+	lea     $1(%rip), $2
 ')
 
 dnl  Usage: CALL(funcname)

as recommended by the Apple programmers, this testsuite failure
disapppers and the testsuite completely passes.
                  Jack