[SciPy-Dev] Scipy Windows wheels

[Nathaniel Smith]

On Thu, Aug 24, 2017 at 4:49 PM,  <josef.pktd at gmail.com> wrote:
>
>
> On Thu, Aug 24, 2017 at 6:56 PM, Pauli Virtanen <pav at iki.fi> wrote:
>>
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA256
>>
>> Dear all,
>>
>> Prerelease binary wheels for Scipy on Windows 32-bit & 64-bit are now
>> available in case you would like to test them.
>>
>> Currently, the plans are that binary wheels will also be provided for
>> future releases on PyPi, so that you will be able to do simply "pip
>> install scipy" also on Windows. At least, assuming we manage to test
>> these wheels well enough for which help would be useful.
>>
>> You can install the scipy prerelease packages as shown below. Note that
>> they are meant for testing only, and correspond to the current Scipy
>> development version. Please report issues found on the Scipy issue
>> tracker on github (be sure to mention how you installed scipy and
>> python).
>>
>> The wheels are meant to be used with the Python obtained from https://p
>> ython.org --- these are not meant to be used with e.g. Conda, although
>> it may be they work.
>
>
> What's the numpy requirement?
> I assume the scipy version should not be used with a currently installed
> numpy unless it is Fortran compatible.
> For example, Winpython distributes Gohlke's binaries build with MKL.
> Is there an automatic check when not installing into an empty virtual
> environment?

IIUC the only reason Gohlke's scipy builds require a specific numpy is
that they use a tricky hack where they assume the numpy package has
installed MKL at a specific location. I don't think these builds use
any hack like this and should work with any numpy. (This makes the
download size larger b/c it means scipy has to carry its own copy of
BLAS, but the trade-off in "just works" is worth it IMO.)

Also, numpy itself doesn't provide any Fortran APIs, so Fortran ABI
shouldn't matter at all.

-n

-- 
Nathaniel J. Smith -- https://vorpus.org