on 7/29/06 7:46 PM, Bill Northcott at email@hidden wrote:
> On 30/07/2006, at 5:04 AM, Joe Koski wrote:
>> Back when I was trying to build octave-2.1.7x with OS X 10.3.x, on
>> my G5
>> Mac, I ran into the missing libraries saveFP and restFP when
>> linking. As
>> Yogi Berra said, "It's deja vu all over again." Now I'm trying to
>> build
>> octave-2.9.7 on OS X 10.4.7, and I'm getting the same error.
>
> Just for the nth time:
> This error is caused by trying to mix objects built with gcc3 and
> objects built with gcc4.
>
> It usually comes up building mixed Fortran and C. If you use g77 for
> the Fortran, you must have gcc3.3 as the default compiler. (selected
> with gcc_select). If you use gfortran then gcc4 must be the default.
>
> You can also get into this strife if you download a binary containing
> a static library, because there is no way I know to easily tell which
> compiler was used for the library. Either use dylibs, which are not
> compiler specific or build everything from source with a compatible
> compiler combination.
>
> Don't stuff about with symlinks as advocated in lots of messages in
> the archives. That is all done automagically if you have the right
> default compiler. Fudging the symlinks might let something build,
> but it is likely to crash in a weird way while being run.
>
> g95 is probably better avoided. It is now no better than gfortran in
> what it will build and to all account the resulting code on Darwin is
> generally not as fast at runtime.
>
> Bill Northcott
>
Bill,
Thanks for taking the time to respond. My limited experience with open
source software (OS X 10.2 => OS X 10.3, and OS X 10.3 => OS X 10.4) is that
the challenge is to port Linux developed software to the Mac with a
constantly changing version of Apple's gcc and ld approach, and no
officially supported Fortran version. We do the best we can. Your points
about not mixing gcc-3 and gcc-4 are well taken. I delayed upgrading my OS X
10.3.9 for several months because of these issues, and now at 10.4.7, OS X
seems to be capable again of building open source code without too many
tricks. For a while 10.4.x was impossible, hence the reliance on the gcc-3
approach of 10.3.9.
As for g95 vs. gfortran, my experience is that, depending on the benchmark,
one or the other will win the speed race, but there is no consistent
pattern. I use g95 because the unformatted output from gfortran is not
compatible with our post-processing software. It seems to me that gfortran
needs a compiler flag to address this.
Joe
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Unix-porting mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden
