ProcessFunction: Automatically serialize Python base type inputs

[sphuber]

Fixes #5686

The to_aiida_type dispatcher is set as the serializer of the input ports of the ProcessFunction process class. This process class is generated on the fly for a function decorated by the calcfunction and workfunction decorators.

Through change, process functions can now be called with Python base types for their arguments, which will be automatically wrapped in the corresponding AiiDA data type.

[unkcpz]

@sphuber thanks! Really a quick and clear implementation. I am curious, what if the input argument is a complex datatype, which error it will raise?

[sphuber]

@sphuber thanks! Really a quick and clear implementation. I am curious, what if the input argument is a complex datatype, which error it will raise?

If a type is passed that is not known by to_aiida_type nothing is serialized and the value itself is passed. It will then go to the normal type validation on the corresponding port and if the type is incorrect, an error is raised like normal. It would be like passing a StructureData to a port that would expect an Int.

[unkcpz]

LGTM!🚀

[ltalirz]

This looks very useful to me; it will also make people wonder why they can't do the same with other processes ;-)

One comment: perhaps it makes sense to adapt some existing tests of work/calcfunctions, replacing AiiDA type inputs with python types, just to have an "integration test" as well.

[ltalirz]

No further review needed from my side.

[chrisjsewell]

Yep all good cheers

[sphuber]

This looks very useful to me; it will also make people wonder why they can't do the same with other processes ;-)

We could consider to set to_aiida_type as the default serializer for all newly created InputPorts. This would achieve the desired effect. Users will just have to realize that if they replace the serializer, the functionality is lost. But usually that is anyway when you are specifying an input port with a valid type that is not a Python base type, so I doubt this would matter much.