gmp 4.2 fails tests on G5

[Daniel Lord]

On Apr 26, 2006, at 12:49, Torbjorn Granlund wrote:

> If an assembly code patch for x86 Macs will fix an issue with powerpc
> Macs, then somebody needs to explain me how computers work.
>
> Until then, I will claim that perhaps something along the same lines
> is needed, but surely that very patch is irrelevant.
>
> --
> Torbjörn

Okay it's a deal...
I'll call you on your claim and raise you one: we have now show,  
informally, that GMP 4.2 can compile on both PPCC Intel Macs and that  
it is a general PPC problem that affect Linux PPC as well
So _I_ claim:
1) a change of compiler flags will create a clean and checked build  
of GMP 4.2 on Mac OS X PPC, and
2) a simple three-line patch will build a clean and checked build of  
GMP 4.2 on Mac OS X 10.4 i386

So how can I assist you in getting these integrated into the source  
so it correctly compiles on both Apple platforms.
As I told you earlier today. I want to help out. let me know how...

	If you rebuild gmp 4.2 on a G5 with...

	./configure --enable-cxx CC='gcc -fPIC' CXX='g++ -fPIC'

	...the resulting build passes make check fine. We ran into a lot of
	this in ppc-linux. Any shared libs built on ppc without -fPIC are
	subject to weird hard to debug crashes. I seriously doubt that
	-fno-common provides the same compiler behavior as -fPIC. Often
	on ppc compiles I have seen both used together.

Daniel Lord
daniellord at mac.com
---
"You will never regret getting up too early,
and you'll always regret getting up too late,
but sometimes you may regret giving up too late."
                           -- Mountaineer's Adage




-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://gmplib.org/list-archives/gmp-bugs/attachments/20060427/fd98914c/attachment-0001.html