Operation on masked_array changes fill_value

I first raised this issue on stackoverflow (see link on the bottom)

Seems like the new masked_array should inherit the fill_value from the two masked_arrays being summed?

Can someone explain to me this behavior of a numpy masked_array? It seems to change the fill_value after applying the sum operation, which is confusing if you intend to use the filled result.

data=ones((5,5))
m=zeros((5,5),dtype=bool)

"""Mask out row 3"""
m[3,:]=True
arr=ma.masked_array(data,mask=m,fill_value=nan)

print arr
print 'Fill value:', arr.fill_value
print arr.filled()

farr=arr.sum(axis=1)
print farr
print 'Fill value:', farr.fill_value
print farr.filled()

"""I was expecting this"""
print nansum(arr.filled(),axis=1)

Prints output:

[[1.0 1.0 1.0 1.0 1.0]
 [1.0 1.0 1.0 1.0 1.0]
 [1.0 1.0 1.0 1.0 1.0]
 [-- -- -- -- --]
 [1.0 1.0 1.0 1.0 1.0]]
Fill value: nan
[[  1.   1.   1.   1.   1.]
 [  1.   1.   1.   1.   1.]
 [  1.   1.   1.   1.   1.]
 [ nan  nan  nan  nan  nan]
 [  1.   1.   1.   1.   1.]]
[5.0 5.0 5.0 -- 5.0]
Fill value: 1e+20
[  5.00000000e+00   5.00000000e+00   5.00000000e+00   1.00000000e+20
   5.00000000e+00]
[  5.   5.   5.  nan   5.]

http://stackoverflow.com/questions/18879272/why-does-sum-operation-on-numpy-masked-array-change-fill-value-to-1e20

No source file or test is named. Start by running the reported masked_array sum and filled() example, then trace the sum result's fill_value handling; done means the behavior matches the documented expectation for the masked row and its filled result.