BUG: Assignment of complex rank-zero arrays produces misleading error

This all makes sense to me:

In [2]: h = zeros(5, dtype=complex)

In [3]: h[1] = 1

In [4]: h[2] = array(2)

In [5]: h[3] = 3j

In [6]: h[4] = array(4j)

In [7]: h
Out[7]: array([ 0.+0.j,  1.+0.j,  2.+0.j,  0.+3.j,  0.+4.j])

This also makes sense:

In [8]: h = zeros(5)

In [9]: h[1] = 1

In [10]: h[2] = array(2)

In [11]: h[3] = 3j
------------------------------------------------------------
Traceback (most recent call last):
  File "<ipython console>", line 1, in <module>
TypeError: can't convert complex to float

This does not make sense:

In [12]: h[4] = array(4j)
------------------------------------------------------------
Traceback (most recent call last):
  File "<ipython console>", line 1, in <module>
ValueError: setting an array element with a sequence.

The problem is not that array(4j) is a “sequence”; array(3) works fine. The problem is that it’s complex and you’re trying to put it in a real array.

Took me a long time to figure out why scipy.signal.impulse((array([1, 1]), array([1, 1]), 15)) works but scipy.signal.impulse((array([1, 1]), array([1, 1]), 15+0j)) does not.

[eric-wieser]

This now gives TypeError: can't convert complex to float in master (and probably earlier)

[seberg]

To be honest, I don’t think I would mind changing the error type. The thing is that setting an element has a different logic, so currently the fact that it is a 0-d array is just ignored, and 0-d arrays in a way just happen to work, because they mostly behave like scalars (in fact there are some quirks for 128bit floats because of that, but there are worse quirks there and 0-d arrays are rare).