feat: Adding addition of 0

[henryiii]

Adding with a false-like, such as 0 or None, is now supported. This allows syntax like sum(histogram_list) without answering the hard question of what it means to add a constant to a histogram with flow bins and unevenly sized bins. That can be done much more clearly using the view or [...] = syntax.

Closes #261.

[henryiii]

@HDembinski, thoughts? Is this fine?

[HDembinski]

It is weird that adding None and False also works. Adding 0 makes sense, but why allow the others?

[henryiii]

Partially because it is natural in construction (you can just check truthiness, and don't have to require a specific type check; any false-like does not affect the histogram when added - you can have WeightedSum(0,0) be false-like an then it would be immediately included). We could do type checks though if it's better to restrict things like None + hist == hist. Let me know if you want it restricted, and I'll do it.

False will always work, as it's a subclass of integer with value 0 (It was just an integer before Python 2.2 or 2.3, IIRC). 1 + False is valid Python syntax. As is 1 + True == 2. isinstance(False, int) is True.

[HDembinski]

Ok, I understand better now. If the argument to __radd__ is not another histogram, I would again push the responsibility to what should happen to the numpy view of the internal storage. This way, we get all the numpy broadcasting rules automatically consistent, including broadcasting of scalars.

The question then again is what should happen with the flow bins. The consistent thing is to include them, since the other operators also work on the flow bins.

[henryiii]

Okay, will place on hold until we have more support for operators on views. We need at least addition then for all views. Is it legal to add a double/int to each of the 3 accumulators?

[HDembinski]

Is it legal to add a double/int to each of the 3 accumulators?

That should work for WeightedSum out of the box, but not for Mean or WeightedMean.

[henryiii]

So Mean and WeightedMean should not support sum(histogram_list)?

[HDembinski]

They should, but they cannot support addition of constants in general, unless we find a reasonable interpretation of adding a number to a mean. So far, we should only allow the addition of 0, which we define to do nothing.

[HDembinski]

You probably already know this:

import numpy as np
a = np.zeros(3)
a + 0 # ok
a + False # ok, False interpreted as zero
a + None # Error: unsupported operand type(s) for +: 'float' and 'NoneType'

We should also raise an error in case someone adds None.

[henryiii]

Okay, we should just check for int/float 0 and only allow that, then.