Scientific Libraries in Python

Horatio Davis horatio at qpsf.edu.au
Thu Nov 15 00:47:20 CET 2001


On Wed, 14 Nov 2001, Travis Oliphant wrote:

> Please expand what you mean by all of this.   I'm not sure I understand
> your concern with the package design of scipy.

The package design is fine - see below for mild concerns on names, but I
have no issues at all with the actual structure.

> We are trying to get there.

Quite so. My preferred option would be to merge other code bases (and
their maintainers) _into_ SciPy in order to get a single unified library.
The SciPy project, with the infrastructure it's built up, seems to be the
best shot at doing this. If we are going to put all the eggs in one
basket, it might as well be the best basket.

> Please make your voice heard on the scipy list. One of the design
> decisions that has been made is to use all lowercase letters for names
> in the package to improve consistency.

Hmm. I would respectfully submit that it would be better to be consistent
with Numeric (which uses Names.With.Capitals) and Scientific Python (which
Also.Does.That). The first reason is that Numeric is already part of the
package structure of SciPy, and it's not going to change, and partial
consistency is not consistency at all.

The second reason is to lower the energy barrier for a merge of Scientific
Python into SciPy. If the change is made, the second-level packages
(Scientific.MPI, Scientific.Statistics, and their siblings) may be moved
straight across without messing about.

It is understood by me that such changes in the namespace give rise to bad
karma among a library's users. But the benefits outweigh the costs if it
is done _now_ rather than waiting until SciPy hits 1.0 and everything is
set in igneous rock.

> Am I hearing that you don't like the top-level name "scipy"

Say rather that I like the name "Scientific" more. A merge between SciPy
and Scientific Python involves choosing a name to keep and a name to
discard. Scientific fits my brain better, which makes it the more Pythonic
choice, and so the one to keep.

That said, I would live with scary_devil_monastery as the top-level
package name if doing so would buy me a complete scientific computing
library in one place. The unification is the priority, the names are not.

> Glad to see the dicussion,

Me also. Now all I need to do is figure out how to attract Konrad Hinsen's
viewpoint on all this... and possibly even some users of these libraries?

Cheers,

AHD






More information about the Python-list mailing list