Initial implementation of shape checking method

[ksunden]

Would like to add tests and some usage of this

[ksunden]

Some thoughts:

• Do we wish to return something other than true/false? e.g. return the values of the variables and error if incompatible

  • This is more similar to e.g. np.broadcast_shapes, which returns the resultant shape and errors if they are incompatible
  • May have some oddities when it comes to broadcast=True where a variable may be omitted or have to have a fill value when all inputs that use it are broadcastable

• There is some thought towards using e.g. flat variadic args that are then passed through itertools.batched rather than as tuples.

  • I think that that would be better for the case of using interactively, but probably actually worse in most production use. I think most production use will tend to do something like:

expected = {
    "x": ("M",),
    "y": ("N",),
    "z": ("M", "N",),
}
found = cont.describe()
if Desc.check_shapes(*[(expected[i], found[i]) for i in expected
]):
    ...
else:
   raise ...

And the extra unpacking to flatten that further is actually more verbose at that point.

• Instead of paired tuples, perhaps we accept the two dictionaries directly, which also validates the keys then.

• This usage pattern is also the reason why it currently has expected as a "raw" shape and found as a Desc object, which I don't love from an API POV, but think it needlessly complicates the code using check_shapes to require Desc objects for expected.

• Options here:

  • accept either tuples or Description objects (on both ends) and sort it out internal to check_shapes
  • Force consistency (in either direction), which would mean forcing expected to be Desc objects or doing found[i].shape to get the shape tuple out.
    • Some interplay with the idea of using dict instead of 2-tuple pairs here, as that would make the change to enforce consistency more complicated, since you'd just be passing the output of describe and so would have to do a whole comprehension rather than a small edit to an existing comprehension.