[SciPy-dev] Should ndimage.measurements.* should return lists if index is a list?

David Cournapeau david@ar.media.kyoto-u.ac...
Mon Apr 20 01:26:56 CDT 2009


Hi Josef,

josef.pktd@gmail.com wrote:
> There are not many complaints/tickets about the code itself, no wrong
> "images", in contrast to other subpackages of scipy.

If there are crashes that nobody can fix, that's a problem in the long
term. If we have code which is slower than the current version, but in
cython and well written, then people can improve it for speed later. If
the code is bad enough so that nobody fixes it, then it will never be
improved. If ndimage has a good testsuite, then it is possible to
replace C code with cython (it can be done gradually). In my experience,
maintainable and reliable C code for python extensions is hard, and not
that many people are capable of doing it, myself included. There is also
the issue of portability to Py3k.

I am afraid that a lot of code in scipy is in that stage that it is
useful but not that maintainable - I like the "paying the technical
debt" metaphor (http://en.wikipedia.org/wiki/Technical_debt), and I
think we are exactly in that period where we have accumulated a lot of
debt, ndimage being such an example.

cheers,

David


More information about the Scipy-dev mailing list