Grumble about too strict an attitude about backward compatibility...

Pearu Peterson pearu at cens.ioc.ee
Mon Sep 9 11:24:36 CEST 2002


On 4 Sep 2002, Skip Montanaro wrote:

> 
> I don't want to wake any sleeping dogs, however, as I'm sitting here
> watching a bunch of Fortran function names fly by as SciPy builds, I'm
> reminded of the fairly recent threads about backward compatibility,
> Python-in-a-tie, etc.   Here I am compiling with a Fortran 90/95 compiler
> (Sun's Forte thing-a-ma-bob) and see function names like
> 
>         lpni:
>         klvna:
>         chgubi:
>         cyzo:
>         klvnb:
>         rmn2so:
>         csphik:
> 
> spew forth.  So, while it's great that this same large library compiles and
> runs on compilers back to at least Fortran 77 (and probably earlier),
> programmers are still stuck with the same cryptic function and data names
> they had to deal with 30+ years ago, all in the name of backward
> compatibility.

I don't think that its is due to the backward compatibility.
There is huge amount of Fortran 77 code available and if one wants to fix
the cryptic names of functions to something more meaningful, then one has
to switch to newer Fortran standard (that allows longer names) and edit
huge amount of F77 code. I think nobody is willing to take this task due
to enormous amount of work (and it cannot be fully automated, some
brain has to deside what are meaningful names), the backward compatibility
is a secondary issue (if an issue at all).

> What's the Python connection?  Other than a reminder not to get to slavish
> about backward compatibility, I note that these same function names will
> then go on to pollute the Python namespace because all this Fortran code is
> automatically wrapped using f2py.

The Fortran function names, that you refer to, are visible only in the
extension modules that are wrapping these functions (this is due to how
Python imports shared modules). All functions that f2py generates,
have the f2py_ prefix to avoid name collision with names from Python
or any other library.
So, I don't understand your argument about name pollution.

And btw, f2py supports mapping function names so that wrapping a Fortran
function with a name `tiarcf' can be accessed in Python side as
`this_is_a_really_cool_function', for example.

Pearu





More information about the Python-list mailing list