py26-numpy +gcc44 requires gcc45 variant of atlas

[James Kyle]

> 
> Usually all scientifc ports requring gfortran should be kept in sync and
> use the same version of gcc. Recently in r71201 [1] atlas have been
> changed to +gcc45 by default because of build problems on Macs with a
> Intel Core i7 processor, see #24869 [2].

This is correct. I had moved the default for atlas to 4.5 due to reports that 4.4 was failing for i7 users. I've seen received my replacement power adapter for my macbook pro 15" i7 (bought in the last 6mths, maybe less).

I tested the atlas builds using this machine and gcc44 variant. It worked flawlessly. Though there may still be outstanding issues that need fixing, the gcc version is likely not the cause of these i5/7 build failures.

> 
> Marcus, James, and others maintaining scientific ports, please discuss
> how to resolve this issue. Either going back to gcc44 or switching *all*
> ports in category science to gcc45.

I've rolled back the default variant for atlas to gcc44 after the above testing cycle and checking what the other big math ports were using. The gcc45 variant is still available as an option that can be selected if the 4.4 build fails. 

I'm all for waiting for gcc 4.5 to mature a bit. Or at least until some errata starts appearing on the atlas build page for that compiler.

-james