BUG: scipy.stats.norm wrong expected value of function when loc is large

The expected value of a normal distribution should be the center of the distribution, but scipy gives wrong results when loc/scale is large.

Reproducing code example:

>>> import scipy.stats as stats
>>> stats.norm.expect(loc=10, scale=1)
10.000000000000004
>>> stats.norm.expect(loc=20, scale=1)
20.0
>>> stats.norm.expect(loc=30, scale=1)
30.0
>>> stats.norm.expect(loc=36, scale=1)
3.3637320653305685e-10
>>> stats.norm.expect(loc=40, scale=1)
1.5746385460369797e-26
>>> stats.norm.expect(loc=10, scale=0.1)
2.3378433179863228e-37

Scipy/Numpy/Python version information:

Scipy 1.1.0
Numpy 1.14.3

[lovis-heindrich]

The same issue also occurs when using a small loc and a large upper or lower bound:

>>> norm.expect(loc=1, scale=1, ub=10, conditional=True)
0.9999999999982133

>>> norm.expect(loc=1, scale=1, ub=20, conditional=True)
0.9999999999999999

>>> norm.expect(loc=1, scale=1, ub=30, conditional=True)
1.0

>>> norm.expect(loc=1, scale=1, ub=40, conditional=True)
3.4007656189526604e-22

>>> norm.expect(loc=1, scale=1, ub=50, conditional=True)
8.381344573482554e-89

[adeharo9]

Could there be a way of throwing a warning or similar when this kind of inconsistencies occur? At least users would be aware that there might be problems with the obtained results.