Re: [tlug] Fortran --> Python (was linux engineer)

["jrjohansson@example.com" <jrjohansson@example.com>]

Hi

>> More seriously, I hear (but don't know) that there is a slow move
>> in the scientific community from Fortran to Python. Next month
>> there is a conference for SciPy (Scientific Python) in Austin,
>> Texas, USA.

I will be going to the scipy conference in Austin next month. Python
is definitely gaining in popularity for scientific computing. Largely
thanks to scipy and numpy which provides great bindings to low level
numerical libraries for linear algebra etc (usually written in C or
fortran), which makes it possible to get high performance using
python. Ruby for example, which is a nice enough language, does not
have any equivalent of numpy/scipy and is therefore out of the
question for scientific computing (unfortunately).

But as Stephen points out this is not so much of a move from Fortran
or C to python, rather that python is used as a framework to access
various low level packages in a unified way. Similar to what
MATLAB/Octave/Mathematica provides.

> There's a rather clear division of labor between the
> algorithm hackers (in low-level languages) and the "science" side of
> the computation.

I think such a division is very important. It makes sense to put a lot
of effort on creating highly efficient low level numerical libraries
(in C or fortran), but the applications that use those libraries
should be easy to create, modify and play around with. That's what the
python environment excel at. And often enough the "algorithm hackers"
and the "science side users" are the same people.

> So I'd say that rather than a *move* from FORTRAN to Python, I'd say
> that FORTRAN (and of course C/C++) are still in heavy use, but Python
> bindings are making heavy-duty computation available to scientists
> without much in the way of programming skills.

I agree, except with the comment on programming skills ;-) I think it
takes as much programming skills to write good python programs as it
takes to write good fortran programs... fortran is not difficult, it's
just arcane.

Rob