Some further digging in the MacPorts fora suggests that gmp is the real 
culprit, but I'm not through digging yet. Compiling gcc 4.8.5 from source using 
the binaries of gmp-5.1.3 et al. from r.research.att.com dies with a "gfortran 
not working" message caused by the familiar f951 illegal instruction error. 
Replacing with 6.0.0 and retrying "make" didn't help nor did using gmp-6.0.0a 
compiled locally. However, starting all over after "make distclean" actually 
produced a functioning gfortran with gmp-6.0.0a. (Thus, it is possible that it 
would have worked with the binary of 6.0.0 too, if I had tried a clean build.)  

(Building gcc/gfortran from scratch literally takes hours on the aging iMac, so 
experimenting is, um, inconvenient... In comparison, building gmp was a piece 
of cake, so if that suffices, it would be great.)

I'll see if I can dig deeper tomorrow.

-pd



> On 03 Aug 2015, at 15:57 , peter dalgaard <[email protected]> wrote:
> 
> 
> On 03 Aug 2015, at 15:10 , Berend Hasselman <[email protected]> wrote:
> 
>> 
>>> On 03-08-2015, at 13:59, peter dalgaard <[email protected]> wrote:
>>> 
>>> I have now tried doctoring $RHOME/etc/Makeconf to use a MacPorts build of 
>>> gfortran-4.8 that I had lying around, and lo and behold: It does source 
>>> installs of nleqslv, geigen, QZ, rms with no trouble at all.
>>> 
>> 
>> Good to know that.
>> 
>>> So I think the finger is pointing at the binaries on 
>>> r.research.att.com/libs.
>> 
>> Indeed as was confirmed by Simon last year here 
>> https://stat.ethz.ch/pipermail/r-sig-mac/2014-May/010895.html
>> I can use the binary of r.research.att.com/libs since I now have a more 
>> modern Mac with an i5 cpu.
>> 
>> I did some searching and found the following:
>> 
>> http://coudert.name/software.html
>> 
>> and https://gcc.gnu.org/wiki/GFortranBinaries#MacOS
>> 
>> Might be a good idea to try these versions of gfortran too on a C2D.
> 
> Yes. However, I'll have to worry about overwriting and possibly breaking my 
> build environment with 3.2.2 coming out next Friday (Simon's binary package 
> rather carefully installs gfortran-4.8, but not gfortran, I believe. Won't 
> know about the others until I try them.). So, not on this machine.
> 
>> Maybe the CRAN R could be built with one of these  gfortran binaries 
>> avoiding the need to compile an R provided gfortran and distributed on 
>> r.research.att.com/libs.
> 
> I haven't seen evidence that there is a problem with the CRAN binaries of 
> anything and we still need to provide gfortran somehow for people who need to 
> build from source. What we could try is to compile 4.8 on a C2D or 
> cross-compile it for that architecture. I think that the latter is doable, 
> but I don't recall the details (gcc --target ?).
> 
> -pd
> 
>> But that is beyond me and I have no idea about possible issues doing it.
>> 
>> Berend
>> 
> 
> -- 
> Peter Dalgaard, Professor,
> Center for Statistics, Copenhagen Business School
> Solbjerg Plads 3, 2000 Frederiksberg, Denmark
> Phone: (+45)38153501
> Office: A 4.23
> Email: [email protected]  Priv: [email protected]

-- 
Peter Dalgaard, Professor,
Center for Statistics, Copenhagen Business School
Solbjerg Plads 3, 2000 Frederiksberg, Denmark
Phone: (+45)38153501
Email: [email protected]  Priv: [email protected]

_______________________________________________
R-SIG-Mac mailing list
[email protected]
https://stat.ethz.ch/mailman/listinfo/r-sig-mac

Reply via email to