If you are running 32-bit 3.6 on Windows, please test this

[Pavol Lisy]

On 8/31/17, Terry Reedy <tjreedy at udel.edu> wrote:
> On 8/30/2017 1:35 PM, Terry Reedy wrote:
>> https://stackoverflow.com/questions/45965545/math-sqrt-domain-error-when-square-rooting-a-positive-number
>>
>>
>>
>> reports the following:
>> -----
>> Microsoft Windows [Version 10.0.16251.1002]
>> (c) 2017 Microsoft Corporation. All rights reserved.
>>
>> C:\Users\Adam>python
>> Python 3.6.2 (v3.6.2:5fd33b5, Jul  8 2017, 04:14:34) [MSC v.1900 32 bit
>> (Intel)] on win32
>> Type "help", "copyright", "credits" or "license" for more information.
>>  >>> import math
>>  >>> math.sqrt(1.3)
>> Traceback (most recent call last):
>>   File "<stdin>", line 1, in <module>
>> ValueError: math domain error
>>  >>>
>>
>> I upgraded from version 3.6.1 to 3.6.2 to try to resolve the issue and
>> restarted my computer but it is still occurring. Some numbers are
>> working (1.2, 1.4) and some others are also not working (1.128).
>> ----
>>
>> Neither installed 64 bit 3.6.2 nor my repository 3.6 32-bit debug build
>> reproduce this.  If anyone has the python.org 32bit 3.6.1/2 releases
>> installed on Windows, please test and report.
>
> Three people have reported that math.sqrt(1.3) works in 32 bit Python on
> 64-bit Windows and no one otherwise.  I reported back on SO that the
> problem is likely local.  Thanks for the responses.

Problem is reported on win10 and I see just 2 tests on win7 (third is
maybe Terry's but on SO I don't see win version).

If I am not wrong (with analyze source code) sqrt is calling function
from "libm" which is some kind of msvcrt.dll on windows... (see
https://github.com/python/cpython/blob/a0ce375e10b50f7606cb86b072fed7d8cd574fe7/Modules/mathmodule.c#L1183
and  https://github.com/python/cpython/blob/6f0eb93183519024cb360162bdd81b9faec97ba6/Lib/ctypes/util.py#L34
)

And with "MSC v. 1900 ..."  it seems that "alternative approaches"
(see here https://bugs.python.org/issue23606 ) are used.

So I would be cautious.

PS.
BTW on my ubuntu I got this:

from ctypes import cdll
print(cdll.LoadLibrary("libcrypt.so"))
<CDLL 'libcrypt.so', handle 21d0860 at 0x7fcce2f22f28>
print(cdll.LoadLibrary("libm.so"))
...
OSError: /usr/lib/x86_64-linux-gnu/libm.so: invalid ELF header

(same with distro's python3, python2 and anaconda 3.6.2)

So this test  ->
https://github.com/python/cpython/blob/6f0eb93183519024cb360162bdd81b9faec97ba6/Lib/ctypes/util.py#L328

has to crash on some environments (see for example:
https://github.com/scipy/scipy/pull/5416/files ). And it seems like
some test failures are just ignored...