diff options
author | Sebastian Berg <sebastian@sipsolutions.net> | 2013-02-23 12:32:58 +0100 |
---|---|---|
committer | Sebastian Berg <sebastian@sipsolutions.net> | 2013-03-03 12:55:35 +0100 |
commit | 2e8fcc0814486accb71aacb7d60ebcb3203f3b7a (patch) | |
tree | 2471652fb409be80a4b6287641057f8eaccf89c9 /numpy/oldnumeric/functions.py | |
parent | 985b2674760176d7ec8d53ecc6a1a4f835a94b15 (diff) | |
download | numpy-2e8fcc0814486accb71aacb7d60ebcb3203f3b7a.tar.gz |
API: Make nditer support ndim == 0 for 0-d iterations.
There are relatively few changes necessare here. However there is a
conceptionally there are no axes for the 0-d case, and no axesdata
needs to be used. This still uses the first axisdata. Which means that
in a few places ndim == 0 is special cased or special cased to act like
ndim == 1.
It would probably be a little cleaner to to use the base pointers directly
in the 0-d case and no axes iteration at all. That would require similar
special cases though.
This also makes oa_ndim == -1 the "correct" way to signal that no op_axes
are given with oa_ndim == 0 being, for the time being, dual use. Either
meaning that nothing may be given, or if something something was given
enforcing a 0-d iteration.
The necessary changes to the ufunc machinery are also done.
Documented that the dtype transfer functions do not handle the scalar
case unless even shape is set.
Diffstat (limited to 'numpy/oldnumeric/functions.py')
0 files changed, 0 insertions, 0 deletions