[R-SIG-Mac] GCC Wiki GFortran

Thu Aug 30 19:16:00 CEST 2012

On Aug 30, 2012, at 12:31 PM, Federico Calboli <f.calboli at imperial.ac.uk> wrote:

> On 30 Aug 2012, at 17:10, Simon Urbanek <simon.urbanek at r-project.org> wrote:
> 
>> 
>> On Aug 30, 2012, at 11:42 AM, Federico Calboli <f.calboli at imperial.ac.uk> wrote:
>> 
>>> On 30 Aug 2012, at 16:26, Simon Urbanek <simon.urbanek at r-project.org> wrote:
>>> 
>>>> 
>>>> On Aug 30, 2012, at 11:19 AM, Federico Calboli <f.calboli at imperial.ac.uk> wrote:
>>>> 
>>>>> On 22 Aug 2012, at 15:12, Simon Urbanek <simon.urbanek at r-project.org> wrote:
>>>>> 
>>>>>> 
>>>>>> On Aug 22, 2012, at 6:34 AM, Federico Calboli <f.calboli at imperial.ac.uk> wrote:
>>>>>> 
>>>>>>> Hi,
>>>>>>> 
>>>>>>> I used to be able to use the Fortran binaries provided by Simon for both R and Julia, but alas this will not be possible anymore:
>>>>>>> 
>>>>>>> "The problem is not BLAS, but your version of gfortran. Mac provides a built-in system blas, so it is not required to build our own. However, "checking for Fortran 77 name-mangling scheme... unknown" is a problem. It appears that the R version of gfortran is too old to build arpack (a quick google search reveals that gfortran older than 4.4 won't build code properly on 64 bit -- yours is 4.2.3) ."
>>>>>>> 
>>>>>>> I could install the GFortran from here: http://gcc.gnu.org/wiki/GFortranBinaries#MacOS. but (1) they will overwrite /usr/local/bin/gfortran (with a symlink) and (2) I don't know if it will work for R.
>>>>>>> 
>>>>>>> So, will the GCC binary work for R, or is it going to be utter PITA?  I'm on 10.8, Xcode 4.4.1 + CLI add-ons and all the other bells and whistles.
>>>>>>> 
>>>>>> 
>>>>>> They won't work with CRAN R binaries because they don't use Apple driver-drivers (i.e. they don't support -arch etc.)*. If you compile R from sources, you can use any Fortran binary you want - it will be tied to your particular runtime (and you can't use -arch).
>>>>> 
>>>>> I use the CRAN R binary and CRAN binary packages
>>>>>> 
>>>>>> BTW: The FSF Fortran installs in /usr/local/gfortran so it can peacefully co-exist with out CRAN Fortran -- even in our Fortran gfortran is just a symlink to gfortran-4.2 so you can easily restore it with
>>>>>> sudo ls -sfn gfortran-4.2 /usr/local/bin/gfortran
>>>>>> if you wish.
>>>>> 
>>>>> Given the above, can I leave gfortran to point to the GCC fortran?
>>>> 
>>>> Not for CRAN binary unless you do sed 's:gfortran -arch:gfortran-4.2 -arch:g' on both Makeconf files …
>>> 
>>> are you saying the binaries (R + whatever libraries I install) using fortran will not work with GCC fortran?  I am confused.
>>> 
>> 
>> The *binaries* don't care about the Fortran as R comes with its own gfortran runtime, but you can't compile packages that have Fortran code with FSF Fortran as it doesn't understand Apple flags.
>> 
>> 
>>>>> if I do not, julia throws a hissy fit every time I compile it,
>>>> 
>>>> Can't you just set FC and F77 to gfortran-4.2 when configuring Julia?
>>> 
>>> It used to be possible (I did successfully build julia on Lion with the R gfortran add on for Lion), though now I am on ML and julia has a fit trying to build arpack with a fortran < 4.4.  As I posted above from the julia dev conversation I started:
>>> 
>>> "The problem is not BLAS, but your version of gfortran. Mac provides a built-in system blas, so it is not required to build our own. However, "checking for Fortran 77 name-mangling scheme... unknown" is a problem. It appears that the R version of gfortran is too old to build arpack (a quick google search reveals that gfortran older than 4.4 won't build code properly on 64 bit -- yours is 4.2.3) ."
>>> 
>> 
>> 
>> But then it's using the wrong compiler ... did you change FC to gfortran-4.2 in Make.inc? I didn't realize that julia doesn't even have a reasonable build system, so you have to edit it by hand …
> 
> Just to test, I removed julia altogether, cloned the git repo, edited Make.inc to have FC = gfortran-4.2 and I get:
> 
> …
> …
> i686-apple-darwin8-gfortran-4.2: unrecognized option '-shared'
> Undefined symbols for architecture i386:
>  "_MAIN__", referenced from:
>      _main in libgfortranbegin.a(fmain.o)
> ld: symbol(s) not found for architecture i386
> collect2: ld returned 1 exit status
> make[2]: *** [amos/libamos.dylib] Error 1
> make[1]: *** [julia-release] Error 2
> make: *** [release] Error 2
> 
> So it looks like there is no way of building julia with R gfortran.
> 

You got that backwards - you don't want to build Julia with R Fortran. The whole point is that you separate your Fortran for julia and for R: gfortran for R (CRAN) and gfortran-4.2 (or whatever you want to call it is up to you) for julia.

Cheers,
Simon

> Cheers
> 
> F
> 
> PS
> 
>> 
>> Cheers,
>> Simon
>> 
>> 
>> 
>>> 
>>>> 
>>>> 
>>>>> though I use and need R *much more than julia*, so I'd like to avoid R binaries of any sort having a hissy fit if gfortran points to something else than gfortran-4.2.  Between R and julia I'm with R.
>>>>> 
>>>> 
>>>> Binaries will be fine, but compiling won't work (because FSF gfortran doesn't understand Apple flags).
>>>> 
>>>> Cheers,
>>>> Simon
>>>> 
>>>> 
>>>> 
>>>>> Cheers
>>>>> 
>>>>> Federico
>>>>> 
>>>>> 
>>>>> 
>>>>>> Cheers,
>>>>>> Simon
>>>>>> 
>>>>>> --- for advanced users only - if you have no idea what I'm talking about, don't do it ---
>>>>>> 
>>>>>> * - in theory you can use the Apple driver-driver of our Fortran to drive FSF-build Fortran. For example if you take the CRAN Fortran then renaming your FSF Fortran to i686-apple-darwin8-gfortran-4.2 and putting our driver instead of gfortran then you may have some luck using CRAN R with your FSF Fortran (obviously, -arch ppc won't work but fewer people care now ;)). There are some potential issues, though, for example the FSF binaries use purely dynamic setup so anything you compile can only be used on your machine. Also in theory you can have runtime version mismatch issues (two different versions of gfortran and gcc runtimes) - so your mileage may vary.
>>>>>> 
>>>>>> 
>>>>>>> BW
>>>>>>> 
>>>>>>> F
>>>>>>> 
>>>>>>> 
>>>>>>> 
>>>>>>> --
>>>>>>> Federico C. F. Calboli
>>>>>>> Neuroepidemiology and Ageing Research
>>>>>>> Imperial College, St. Mary's Campus
>>>>>>> Norfolk Place, London W2 1PG
>>>>>>> 
>>>>>>> Tel +44 (0)20 75941602   Fax +44 (0)20 75943193
>>>>>>> 
>>>>>>> f.calboli [.a.t] imperial.ac.uk
>>>>>>> f.calboli [.a.t] gmail.com
>>>>>>> 
>>>>>>> _______________________________________________
>>>>>>> R-SIG-Mac mailing list
>>>>>>> R-SIG-Mac at r-project.org
>>>>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mac
>>>>>>> 
>>>>>>> 
>>>>>> 
>>>>> 
>>>>> --
>>>>> Federico C. F. Calboli
>>>>> Neuroepidemiology and Ageing Research
>>>>> Imperial College, St. Mary's Campus
>>>>> Norfolk Place, London W2 1PG
>>>>> 
>>>>> Tel +44 (0)20 75941602   Fax +44 (0)20 75943193
>>>>> 
>>>>> f.calboli [.a.t] imperial.ac.uk
>>>>> f.calboli [.a.t] gmail.com
>>>>> 
>>>>> 
>>>> 
>>> 
>>> --
>>> Federico C. F. Calboli
>>> Neuroepidemiology and Ageing Research
>>> Imperial College, St. Mary's Campus
>>> Norfolk Place, London W2 1PG
>>> 
>>> Tel +44 (0)20 75941602   Fax +44 (0)20 75943193
>>> 
>>> f.calboli [.a.t] imperial.ac.uk
>>> f.calboli [.a.t] gmail.com
>>> 
>>> 
>> 
> 
> --
> Federico C. F. Calboli
> Neuroepidemiology and Ageing Research
> Imperial College, St. Mary's Campus
> Norfolk Place, London W2 1PG
> 
> Tel +44 (0)20 75941602   Fax +44 (0)20 75943193
> 
> f.calboli [.a.t] imperial.ac.uk
> f.calboli [.a.t] gmail.com
> 
> 