[SciPy-user] [Numpy-discussion] [ANN] NumPy 0.9.6 released
Travis Oliphant
oliphant at ee.byu.edu
Wed Mar 15 13:32:05 EST 2006
mfmorss at aep.com wrote:
>>We need some of these people to convert earlier than that, though, so we
>>can make sure that 1.0 really is ready for prime time. I think it's
>>very close already or the version number wouldn't be so high. We are
>>just waiting for more people to start using it and report any issues
>>that they have before going to 1.0 (I'd also like scalar math to be
>>implemented pre 1.0 as well in case that requires any C-API additions).
>>
>>
>
>I can understand that, but here as a potential industrial users of Numpy,
>we can't really afford the risk. We're looking at Numpy as a key piece of
>a Python replacement of commercial software for critical daily production.
>If we build Numpy/Scipy into our system, it has to work. We don't want to
>be anyone's beta testers.
>
>
>
That's fine. In your situation, you may need to wait until 1.0 before
fully switching to NumPy. But, of course you can start using it now.
We will not be changing standard interfaces in NumPy. Most of these are
derived from Numeric which has been around for a long time. The closer
we get to 1.0, the less willing the developers will be to see things
change. On the Python-level, for example, I don't expect anything to
change except for better support of array sub-classes (i.e. making sure
they are preserved through more operations).
On the C-level, there may be an additional C-API or two, but I don't see
any significant changes to the C-API itself. Because of the way the
C-API is loaded, extension modules must be re-compiled if the C-API
changes. This is the only real headache at this point I think.
Basically, until 1.0 you need to re-build extension modules whenever a
new version of NumPy comes out (SciPy consists of a bunch of extension
modules and so must be re-built). Once we hit 1.0, you will only need
to rebuild on minor increment changes (i.e. 1.0 to 1.1 and bug-fixes
will not change the C-API).
-Travis
More information about the NumPy-Discussion
mailing list