MSVC fixes#5207
Conversation
The functions are macros in msvc which we undefine and redeclare in npy_math. Since we now generate the ldexpl and frexpl ufuncs in numpygh-4852 this was broken by the extra logic for them in npy_config.h
cbrt does not set the invalid flag on nan input as all inputs are valid. Fix the fallback to behave the same.
There was a problem hiding this comment.
Not sure about this one, the same code is in v1.8.2. What is the motivation? I'm sure there is a reason...
There was a problem hiding this comment.
I don't think its necessary. The motivation was, AFAICT, that since frexp and ldexp were wrapped by hand the logic for using the double versions for long double also needed to be by hand. In npy_math we just wrap these like any other now, meaning that we always call the double version for long double if we don't have an explicit l version, so this logic shouldn't be necessary.
|
Thanks for cleaning this up, Julian. This breakage certainly goes on me. |
|
@ewmoore OK, we will give it a shot. The previous was also a gamble... |
|
Thanks Julian. |
fix undefined reference of ldexpl/frexpl and the cbrt fallback raising an invalid flag on nan input.