Unified Arrays in Python?

Travis Oliphant Oliphant.Travis at mayo.edu
Fri May 26 16:29:08 CEST 2000


> > I thought there was a rumour some time ago that said unified arrays for
> > Python were on their way.  These would presumably fix the inconsitencies
> > and annoyances involved in moving data between NumPy and PIL.  I've just
> > read the Python 1.6 page, but have found no announcement on this topic.
> > Perhaps there was never any truth in the rumour?
> 
> it's definitely on the todo list, but not for 1.6.
> 
> the problem appears to be that there is no really great candidate; the code
> in numpy is considered unmaintainable, and I'm not sure if anyone is working
> on an alternative at this time.
> 

Well I'll chime in here.  I am currently working on an alternative that 
I want to get ready by Python 1.7.  The numpy code isn't UNmaintainable
--- but it is pretty involved.  The difficulty is trying to make a
significant body of users happy.   

The solution I'm working on is to make the NumPy Object a bonified Python
Class (my solution to the type/Class dichotomy is to make everything
classes).  The addition of the buffer-type to Python makes this a workable
solution now.  I'm projecting completion by the end of the summer.  (But,
you know how dates get pushed...)

The NDArray class will still be fast due to carefully placed C-code that
will borrow from the current code-base.   But, it should be considerably
easier to maintain and enhance (not to mention the advantage of
subclassability --- with C-extensions to NumPy not caring).

If anybody would like to help me.  I can easily set up a CVS tree for 
co-development work. 







More information about the Python-list mailing list