[R-SIG-Mac] vecLib and building R from source

Simon Urbanek simon.urbanek at r-project.org
Thu Sep 1 03:21:14 CEST 2005

On Aug 31, 2005, at 8:11 PM, Kjell Konis wrote:

> I have both the gcc3.4.4 compilers and the gcc4.0.1 compilers.  When
> I use 3.4.4 the configure script does not link against the blas in
> vecLib but it does when I use 4.0.1.

That is strange - did you use --with-blas='-framework vecLib' --with- 
lapack ? If so, please send me your config.log.
As of R-devel that is the default now, but for R 2.1.1 you need to  
specify it. You may also want to add F77=g77 if you have both  
gfortran and g77 around just to make sure.

> However, I remember seeing an email here that said use the 3.4.x  
> compilers unless you have a good reason not to (and the good reason  
> was wanting a 64 bit build which I  don't).  Is this still the  
> prevailing wisdom?

Yes, there are still problems with 4.x - mainly with gfortran (it  
still segfaults on simple Fortran code and is slower than g77). We  
have worked around some of the show-stoppers in R-devel, but it's  
still not recommended.

>   I want a safe, stable,  useable build.  Is it better to use 3.4.4  
> w/o vecLib or 4.0.1 w/ vecLib?

3.3 with vecLib is the recommended way and that is how the CRAN  
binary is built (and the proper g77 compiler is included in the binary).


More information about the R-SIG-Mac mailing list