py26-numpy +gcc44 requires gcc45 variant of atlas

[Michelle Gill]

Greetings MacPorts users,

I am a fresh convert from Fink to MacPorts and am currently muddling my way through a setup of numpy and various other scientific packages. I'm wondering if someone could let me know why the gcc45 variant of atlas seems to be the default install for the gcc44 variant of py26-numpy:

% port rdeps py26-numpy
The following ports are dependencies of py26-numpy @1.5.0_0+gcc44:
  python26
    gettext
      libiconv
        gperf
      ncurses
        ncursesw
      expat
    zlib
    openssl
    sqlite3
      readline
    db46
    gdbm
    bzip2
  fftw-3
  py26-nose
    py26-distribute
  atlas
    gcc45
      gmp
      mpfr
      libmpc
    gzip
      texinfo
        lzmautils
  gcc44

Since it seems there is a variant of atlas available for gcc44:

% port variants atlas
atlas has the variants:
   gcc42: build using macports-gcc-4.2
     * conflicts with gcc43 gcc44 universal
   gcc43: build using macports-gcc-4.3
     * conflicts with gcc42 gcc44 universal
   gcc44: build using macports-gcc-4.4
     * conflicts with gcc42 gcc43
[+]gcc45: build using macports-gcc-4.5
     * conflicts with gcc42 gcc43 gcc44
   universal: Build for multiple architectures

My guess is that this is because gcc45 is the default variant for atlas, while gcc44 is the default for py26-numpy. Further, there is not a gcc45 variant available for py26-numpy. More importantly, will the issue of having two copies of gcc be avoided by first installing the gcc44 variant of atlas and then installing py26-numpy or do I need to force an install and/or edit config files? (The second option is no problem for me, I'm just not yet familiar with how the config/build system works in MacPorts.)

I did search for this issue on the users mailing list and I suspect it's related to this: 
http://www.mail-archive.com/macports-users@lists.macosforge.org/msg19958.html

I have a related newbie question is about the use of universal vs gcc4X builds. Is there a rule of thumb as to which is preferable for someone installing a large number of python and scientific packages? I guess my criteria for "preferable" include ease of building packages and avoiding having multiple gcc versions, if possible. I'm on an i5 MacBook Pro.

Thanks for your help.

Kindest regards,
Michelle Gill