Input / output validation for calcfunctions and workfunctions

[greschd]

Since we're dropping py2 support in #3566, we can make use of glorious new features.

One possibility would be to add (optional) input / output validation for calcfunctions and workfunctions through type hints.

Example:

from aiida import orm
from aiida.engine import calcfunction

@calcfunction
def add(x: orm.Float, y: orm.Float) -> orm.Float:
    return x + y

Should be quite straightforward to implement. Opinions on whether this would be desirable?

[chrisjsewell]

+1 from me

to elaborate on the implementation (copied from #3724):

From python 3.6 onward I believe, type annotations are stored in __annotations__, e.g.

@calcfunction
def my_calcfunction(structure: StructureData):
    pass

my_calcfunction.__annotations__ == {"structure": StructureData}

This could be used here, to set a more specific valid_type than just orm.Data:

aiida-core/aiida/engine/processes/functions.py

Line 261 in 5790d0d

spec.input(arg, valid_type=valid_type, default=default)

[greschd]

Just checked on 3.5, works there as well. The output annotation is stored in the 'return' key (since that's a keyword, it can't be an input name).

[chrisjsewell]

If you wanted to get really fancy, you could also use TypedDicts, to set output validation on multiple outputs, but that's only available natively from python 3.8

[greschd]

I guess the user can always specify that as return value, and we just need to make sure it works out with the isinstance check, right? If some user code only supports 3.8+, that shouldn't bother us.

[ltalirz]

Just mentioning that we will drop python 3.5 support with the next release , following the compatibility timeline