MemoryReader.timeseries doesn't raise NoDataError on empty AtomGroup, while DCDReader.timeseries does

[hmacdope]

Expected behavior

The DCDReader timeseries and MemoryReader behave the same.

Actual behavior

The DCDReader raises a NoDataError if an empty atomgroup is provided as the asel argument. This is the same approach taken in #3890

However the MemoryReader just returns an empty array.

Code to reproduce the behavior

import MDAnalysis as mda
from MDAnalysis.tests.datafiles import PSF, DCD,
import numpy as np

u = mda.Universe(PSF, DCD)
u.trajectory.timeseries(u.select_atoms(None)) # raises NoDataError

coords = np.zeros((10,20,3))
u = mda.Universe(coords, format="MEMORY")
u.trajectory.timeseries(u.select_atoms(None)) # returns an empty array. 

[IAlibay]

I believe the recent consensus was that NoDataError was for missing topology attributes rather than zero sized things (see #3837 (comment))

Probably would want some other error type here?

[hmacdope]

We could go for ValueError or something, whatever people think is best. Is this best addressed in a seperate PR after #3890 where we homogenise all the call sites?

[IAlibay]

We could go for ValueError or something, whatever people think is best. Is this best addressed in a seperate PR after #3890 where we homogenise all the call sites?

I probably would just advocate for changing DCD's to ValueError and make the other two ValueError? Seems like a simple enough one to have within the scope of #3890?

[orbeckst]

I agree with moving to ValueError.

Thanks for raising the discussion @IAlibay , making the semantics of a historical code base consistent is hard work.

[IAlibay]

We really should document this somewhere, i.e. "Raised when empty input is not allowed or required data are missing." is very non-specific if we're going with a TopologyAttrs centric viewpoint. I'll raise an issue.

[hmacdope]

Fixed in #3890