summaryrefslogtreecommitdiff
path: root/numpy/lib/arraysetops.py
diff options
context:
space:
mode:
authorDavid Cournapeau <cournape@gmail.com>2009-07-20 04:42:13 +0000
committerDavid Cournapeau <cournape@gmail.com>2009-07-20 04:42:13 +0000
commitc378f4a0179b9c2a0d700e9239744ebc07a389cb (patch)
treee120c75def44e6fba6f0add8c073fc8bbdf7874c /numpy/lib/arraysetops.py
parent2917b5710dc9957b7aa84df074b29a1a0a86476a (diff)
downloadnumpy-c378f4a0179b9c2a0d700e9239744ebc07a389cb.tar.gz
Change how we decorate npy_math functions.
For platforms without the necessary math functions, we used to first implement them by ourselves, and then decorate the math function with npy_ prefix, whether the undecorated functions are from the implementation or from ourselves. This is not always practical, because we may want to use our own implementation even if the platform has one implementation. Instead, when we define our own implementation, we directly define the decorated function, and only decorate functions when we use the platform implementation. Example: say the function hypot is available on the platform, but buggy. If we undefine HAVE_HYPOT, in the former scheme, we would have first: static double hypot(double x, double y); And then: double npy_hypot(double x, double y); But this would clash with the platform hypot. By defining directly hpy_hypot, and put our implementation there, we avoid this clash.
Diffstat (limited to 'numpy/lib/arraysetops.py')
0 files changed, 0 insertions, 0 deletions