scipy: scipy.stats.chisquare test does not check that observed and expected frequencies add to same total

The function scipy.stats.chisquare performs the chi-square test on a vector of observed and expected frequencies. For the test to make sens, the observed and expected frequency vectors must sum to the same total, otherwise the result is nonsense since the inputs are incompatible. So there are two options: either an error should be thrown if they do not, or f_exp need to be rescaled so they sum to the total of f_obs. The first is probably better, or the second with a warning thrown.

Reproducing code example:

# In the following example, one vector is an exact multiple of the other.
# this means that the observed and expected frequencies are exactly
# proportional.  This should give a p-value of 1 (not significant at all).  
# Instead, you get the following:
from scipy.stats import chisquare
chisquare(f_obs=[10,20], f_exp=[30,60])

# Power_divergenceResult(statistic=40.0, pvalue=2.5396285894708634e-10)
# the statistic of 40 is calculated as following, which directly follows the 
# formula.  This is then plugged into a chi squared distribution to get a 
#p-value of close to 0, which is the opposite of the significance you should
# get.
# ((10-30)**2 /30) + ((20-60)** 2 / 60) = 40

# Instead, here is what should happen
import numpy as np
fobs = np.array([10,20])
fexp = np.array([30,60])
# adjust the totals
# gives array([10., 20.]), the same as observed
fexp = fexp * (np.sum(fobs)/np.sum(fexp)) 
chisquare(f_obs=fobs, f_exp=fexp)
# the correct result
# Power_divergenceResult(statistic=0.0, pvalue=1.0)

Scipy/Numpy/Python version information:

1.4.1 1.18.4 sys.version_info(major=3, minor=7, micro=4, releaselevel='final', serial=0)