[Matrix-SIG] File corrected

Konrad Hinsen hinsen@cnrs-orleans.fr
Tue, 11 Jan 2000 15:25:39 +0100


> I fixed the LLNLDistribution13.tgz at 9:35 a.m. PST

So which is now the most current NumPy version,
LLNLDistribution13.tgz or Numerical-14.tgz? Both have a date stamp
of Jan 10.

As for the ongoing discussion about distribution strategies and
inclusion of arrays in the Python distribution, I wonder if anyone
knows Guido's opinion about this. After all, we can discuss forever
what he should do, but in the end it doesn't matter...

Perhaps the main problem we need to discuss before anything can become
part of the Python disttribution is whether we are happy with NumPy
as it is, or at least with some subset. It seems that almost everyone
is unhappy about something, but we can't agree about what.

Upcasting: A while ago there was a lengthy discussion about this, and
there were some proposals that looked as if they could be acceptable
to everyone. Has anyone ever tried implementing them? It doesn't seem
worth continuing the discussion without some hands-on experience.

Rank 0 arrays: There are various philosophies as to whether rank-0
arrays should be the same as scalars or not. NumPy has followed the
APL tradition in which there is no difference, and I see no reason to
change this. So if NumPy arrays become part of Python, it would be
preferable to minimize any existing differences. This doesn't mean
that the implementation should be the same; the array overhead is
after all quite significant. But providing the basic array attributes
(shape, real, etc.) for Python scalars would be enough to remove the
distinction as seen by the Python programmer, and at the C extension
level the need for unification is much less obvious. Another annoying
difference, the different conventions for modulo calculations, should
be fixed in NumPy in my opinion. Perhaps one could add a
single-precision float scalar type to Python to increase compatibility
even further.

Konrad.
-- 
-------------------------------------------------------------------------------
Konrad Hinsen                            | E-Mail: hinsen@cnrs-orleans.fr
Centre de Biophysique Moleculaire (CNRS) | Tel.: +33-2.38.25.55.69
Rue Charles Sadron                       | Fax:  +33-2.38.63.15.17
45071 Orleans Cedex 2                    | Deutsch/Esperanto/English/
France                                   | Nederlands/Francais
-------------------------------------------------------------------------------