David Douard wrote:
Nice.
On my side, I've spent some time to do a simiar job (in the train in
which I spend 2 hours a day), but with the SVN version of numpy. I've
done the job the quick'n dirty way (no deep testing, and so), but I plan
to do a cleaner refactoring allowing to choose at install time which
numeric module is used (the way matplotlib does) by Veusz.

I was also wondering if that was worthwhile. If it is, there is the question of which array interface to use in the wrapper (i.e. NumPy-like or Numarray-like). I would favour NumPy-like but it does have the significant disadvantage that documentation is less readily available. My only worry is that it will produce a significant extra maintainability burden over time.

The other thing I did, which I have forgot to mention is changed all the setting objects so that all the convertTo methods actually do a conversion to the relevant python type and raise ValueError if the conversion fails for whatever reason. I also changed the tests to match this. I don't know if there's a reason that this change is a bad idea. Jeremy?

Do you want me to send the patches I produce ?

Yes :)

--
"You see stars that clear have been dead for years
But the idea just lives on..." -- Bright Eyes

_______________________________________________
Veusz-discuss mailing list
[email protected]
https://mail.gna.org/listinfo/veusz-discuss

Répondre à