I was caught offguard by the following behavior using numpy 1.12:

Current behavior

In  [1]:  a=array([1,0])
          a[0] = 1 + 1.1  
          print(a.dtype)       
          print(a)
          print(a.dtype)               
Out [1]:  int64
          [ 2  0 ]
          int64 

Expected behavior

In  [2]:  a=array([1,0])
          a[0] = 1 + 1.1  
          print(a.dtype)       
          print(a)
          print(a.dtype)               
Out [2]:  int64
          [ 2.1  0. ]
          float64

It may be that this is intentional behavior, but I would like to suggest automatic conversion of integer arrays to floating point arrays when a float is saved to an element or slice of the array. I'm not sure how straighforward implementation of this feature would be.

At the very least, there ought to be a warning when truncation occurs, especially since I at no point in the above code explicity declare the dtype of A to be int64, so I (and I think most users) would expect it to default to float64.

I'm interested in opinions on the matter.

Thanks,
Oliver