BUG: np.reshape sometimes returns scalars

In [2]: np.reshape(int(0), ())
Out[2]: array(0)

In [3]: np.reshape(np.int64(0), ())
Out[3]: 0

I would prefer the first behavior for both. In fact, I don’t know how to get an array(0) from an np.int64(0) right now.

np.__version__ is "1.11.0"

[eric-wieser]

The problem I see here, is that I do think reshape should return an array always, however, the scalar is write-only, so that is a bit strange

reshape doesn’t make any promise of returning a view, so I think we’re off the hook for returning a writeable array copy.

[seberg]

Sounds like the scalars .reshape() method is just a bit wrong? The problem I see here, is that I do think reshape should return an array always, however, the scalar is write-only, so that is a bit strange.

Not sure if gh-14489 is too directly involved, but I don’t have an immediate idea of how to best fix/think about this.