Re: [Numpy-discussion] ANN: numarray-1.3.3

[Todd M. <jm...@st...>]

On Sun, 2005-07-10 at 11:43 -0400, Les Schaffer wrote:
> Todd Miller wrote:
> 
> >numarray-1.3.3 is a "maintenance release" which solves a couple esoteric
> >(embedded Python + numarray) but serious (core dumping) problems.
> >
> >  
> >
> cool
> 
> >I think there's a good chance of adding something for the next release
> >but I don't think it will be called keys() because that's a dictionary
> >cue and RecArrays aren't dictionaries.
> >  
> >
> 
> tho as column spaces, they essentially act as dicts.
> 
> >I'm in favor of adding .names to make _names public.
> >
> 
> 
> that would work, at least it tells outsiders .names is public and stable 
> attribute.
> 
> and even better is help us subclass easily by making the constructor 
> functions fromarray, fromrecord, and array take an optional argument 
> which is the RecArray class to use.  as it stands RecArray cannot be 
> subclasses without re-doing all those constructor functions.
> 
> i would be happy to contribute code if i knew it would be used.

What you want to do sounds reasonable so chances are good a patch would
be used.  I would:

1. Make .names another reference to ._names

2. Either convert the functions you want to be subclassable into "class
functions" (and add backward compatible globals) or else add a "kind"
parameter for passing in the subclass (see strings.py) to the current
functions.  I'm not sure which approach is better.

Regards,
Todd