BUG: Disable hex(np.floating) and oct(np.floating) on python 2 #10765
Conversation
94591ee to
9918e22
Compare
| * cfloat, cdouble, clongdouble)*2# | ||
| * #oper = oct*7, hex*7# | ||
| */ | ||
| #define @name@_@oper@ NULL |
There was a problem hiding this comment.
There's precedent for this approach with #define @name@_invert NULL
|
I wonder if this is worth doing at this late point when we are about to drop 2.7 support. It has the potential to cause some pain downstream and I don't see that we gain much benefit. |
|
For users planning to upgrade to python 3 anyway, this is pain they'll experience when they transition - arguably it's worth getting some of the pain out of the way. |
|
I'd be surprised if this causes pain downstream - |
|
We should either close this or add tests and get it in |
|
@mattip: Into 1.16.x, you mean? |
|
yes |
9918e22 to
7654b69
Compare
|
Updated with tests |
This makes them match the behavior of hex(float) and oct(float)
7654b69 to
c123a5d
Compare
|
Tests are failing with infinite recursion |
|
Bizarre... |
|
Ah, this is the fun scalar type recursion where Longdouble seem to be the only type that forms a loop here (#8547) |
|
Leave be to avoid churn. |
This makes them match the behavior of hex(float) and oct(float) on python 2
Note that
np.float16.__hex__still exists, which callsnp.generic.__hex__→ndarray.__hex__→np.float16::tp_number.nb_hex. Unfortunately #9972 means__hex__andnb_hexare out of sync.