diff options
author | Sebastian Berg <sebastian@sipsolutions.net> | 2016-03-20 14:32:21 +0100 |
---|---|---|
committer | Sebastian Berg <sebastian@sipsolutions.net> | 2016-03-20 14:41:43 +0100 |
commit | 582600415d966dbfb7abafd3151415f59c6c422d (patch) | |
tree | 25dc92f351133e14be790555b168c09126b5da33 /numpy/core/fromnumeric.py | |
parent | 41ec9fc8b7ee9b710c3054b9e9409a7a1f6082f7 (diff) | |
download | numpy-582600415d966dbfb7abafd3151415f59c6c422d.tar.gz |
BUG: Do not try sequence repeat unless necessary
Only if the other class actually implements sequence repeat, should
it be tried. Otherwise classes which implement neither will cause
the sequence repeat branch. This branch may fail for two reasons:
1. The scalar was not integer
2. The other object raises an error during repeat.
Previously, the first did not happen for floats, etc. and the
second was using a try/except logic. Now both will always error.
An object which claims to support sequence repeat should never
have to fall back to normal multiplication.
Note that all of this is controversial in the final behaviour.
We may actually want `[1, 2, 3] * np.float64(3.)` to return
an array with `[3., 6., 9.]` at some point.
-
It was suggested to create a better error message, this
is not covered here. The major point is fixed though, so:
closing gh-7428
Diffstat (limited to 'numpy/core/fromnumeric.py')
0 files changed, 0 insertions, 0 deletions