Libertem: Handle empty ROI gracefully

[sk1p]

To reproduce, run a UDF with roi=np.zeros(ds.shape.nav, type=bool). Currently raises a weird exception in the internals...

Edit: specifically happens with the PickUDF - maybe the error in the UDF runner can also be improved?

[uellue]

Discussion with @sk1p :

We should probably emulate NumPy's behavior with arrays that have a zero in their shape. That means we create empty result buffers of the appropriate shape if the ROI is empty. Editing title accordingly.

Example code for NumPy empty arrays:

In [1]: import numpy as np

In [2]: a = np.ones((0, 1, 0))

In [3]: a.shape
Out[3]: (0, 1, 0)

In [4]: a
Out[4]: array([], shape=(0, 1, 0), dtype=float64)

[AnandBaburajan]

To reproduce, run a UDF with roi=np.zeros(ds.shape.nav, type=bool). Currently raises a weird exception in the internals...

I've managed to reproduce this error:
ValueError: invalid extra_shape (0, 16, 16): cannot contain zeros

That's the error right? I'm familiarising myself with UDFs and I've got @uellue's comment.

@sk1p I'd like to work on this. Anything else I should know or keep in mind except uellue's comment? Thanks!

[sk1p]

That's the error right?

Sound right, yes.

Anything else I should know or keep in mind except uellue's comment?

Don't think so - let me know if you have further questions when you start your implementation. Btw. a good approach for changes like these is to practice TDD and start by writing tests that initially fail, and then work to make the tests succeed.