Re: [Fink-users] [Fink-devel] Switch to gcc43 (Was Re: octave.info & gcc-4.3)
On Friday, March 21, 2008, at 02:02PM, "Jean-François Mertens" <[EMAIL PROTECTED]> wrote: > >On 20 Mar 2008, at 00:35, Ben Abbott wrote: >> On Mar 14, 2008, at 2:46 AM, Martin Costabel wrote: >> >>> Jean-François Mertens wrote: >>> [] but right, if true it would encourage a fast update of anything in fink to gcc43 >>> >>> I too have come round to the conclusion that this is indeed the best >>> plan for action at this moment. I would suggest that we band >>> together and do this update rather quickly. >>> Martin >> >> I continue to have problems with building the bleeding-edge octave >> sources. Unfortunately the "bus error" I encounter leaves gdb useless >> to me :-( >> >> In a nut shell, I'm unable to execute "A\B" or "mldivide(A,B)" in >> octave for more than 96 eqns/unknowns. >Is this with octave-atlas from fink unstable, unmodified ? >Else it is your problem... > >> It is has been pointed out that using gcc 4.0.1 (Apple's vecLib) will >> conflict with gfortran 4.1+, due to type mismatches resulting from how >> gfortran treats functions. I haven't looked at the detail, but when I >> googled, "gfortran ff2c type mismatch" quite a bit showed up. > This is exactly the reason why we >upgraded fast all basic fortran packages to gcc43. >For programs that would use both Apple's vecLib and gcc43, >I agree there is a risk. >But I'm sure octave-atlas doesn't use veclib. > >> The suggested solution is to add "-ff2c" to the fortran compile >> parameters. >'Solution to what ? > > >> I can't vouch for any of this, but if someone can tell me what to >> change in the atlas.info to pass "-ff2c" to gfortran, I can test >> suggested solution and report back. >This would really be the worst thing to do.. >atlas is what allows you to be in straight gcc43 setting > >JF Mertens > JF, When I wrote the prior email, the atlas.info file on my mac didn't mention gcc42 or gcc43 (at least I didn't see it). I selfupdated since, and now see gcc43 is needed to build. Regarding -ff2c, I had been advised to build octave using -ff2c with gcc4.1/2/3 so that the result would be compatible with gcc 4.0.1. In any event, thanks for the comments. One final note, the octave developers have made some changes that allows the bleeding-edge octave to build correctly using Fink's octave.info file (with the necessary change to the version numbers). Ben - This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse012070mrt/direct/01/ ___ Fink-users mailing list Fink-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/fink-users
Re: [Fink-users] [Fink-devel] Switch to gcc43 (Was Re: octave.info & gcc-4.3)
On 20 Mar 2008, at 00:35, Ben Abbott wrote: > On Mar 14, 2008, at 2:46 AM, Martin Costabel wrote: > >> Jean-François Mertens wrote: >> [] >>> but right, if true it would encourage a fast update of anything in >>> fink to gcc43 >> >> I too have come round to the conclusion that this is indeed the best >> plan for action at this moment. I would suggest that we band >> together and do this update rather quickly. >> Martin > > I continue to have problems with building the bleeding-edge octave > sources. Unfortunately the "bus error" I encounter leaves gdb useless > to me :-( > > In a nut shell, I'm unable to execute "A\B" or "mldivide(A,B)" in > octave for more than 96 eqns/unknowns. Is this with octave-atlas from fink unstable, unmodified ? Else it is your problem... > It is has been pointed out that using gcc 4.0.1 (Apple's vecLib) will > conflict with gfortran 4.1+, due to type mismatches resulting from how > gfortran treats functions. I haven't looked at the detail, but when I > googled, "gfortran ff2c type mismatch" quite a bit showed up. This is exactly the reason why we upgraded fast all basic fortran packages to gcc43. For programs that would use both Apple's vecLib and gcc43, I agree there is a risk. But I'm sure octave-atlas doesn't use veclib. > The suggested solution is to add "-ff2c" to the fortran compile > parameters. 'Solution to what ? > I can't vouch for any of this, but if someone can tell me what to > change in the atlas.info to pass "-ff2c" to gfortran, I can test > suggested solution and report back. This would really be the worst thing to do.. atlas is what allows you to be in straight gcc43 setting JF Mertens - This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse012070mrt/direct/01/ ___ Fink-users mailing list Fink-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/fink-users
Re: [Fink-users] [Fink-devel] Switch to gcc43 (Was Re: octave.info & gcc-4.3)
On 14-Mar-08, at 02:46 , Martin Costabel wrote: > Jean-François Mertens wrote: > [] >> but right, if true it would encourage a fast update of anything in >> fink to gcc43 > > I too have come round to the conclusion that this is indeed the best > plan for action at this moment. I would suggest that we band together > and do this update rather quickly. > > I am CCing fink-devel so that the maintainers of the 28 packages > concerned get a chance to act, but I think we should just test this > upgrade for each of these packages and then simply go ahead and change > any package that builds correctly with gcc43. As far as I know, no > problem with the switch from gcc42 to gcc43 has yet been reported. > > Here are the package descriptions concerned: > > sci/avl.info > sci/qprop.info > sci/xfoil.info I have confirmed that avl, qprop and xfoil build with gcc43 on Intel. I am in the process of testing them to be sure they still run correctly. I'll commit the updates once they have passed my run tests. -- Kevin Horton Ottawa, Canada - This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse012070mrt/direct/01/ ___ Fink-users mailing list Fink-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/fink-users