Hi,
First of all, thanks again for bringing back `prange`. This is really a great feature!
When using it, it seems that some functions require the user to return an array, even if it is not actually necessary to the user (e.g. arrays that are modified in place).
For instance, the code below fails when attempting to compile the function. However, replacing the line `return` by `return x` fixes this issue. Is this an intended behavior or bug?
```
import numpy as np
import numba
N = 1000
x = np.random.rand(N)
@numba.njit(parallel=True)
def add_one( x, N ):
for i in numba.prange(N):
x[i] = x[i] + 1
return
# return x : fixes the compilation
add_one( x, N )
```