Ways to do Typing

[Moritz-Schmidt]

There are multiple ways to type this library.

1. Mashumaro (or other serialization/deserialization library)

@erwindouna started implementing mashumaro models in his rewrite. These could be used to add typing to the HTTP API responses and the return types of the PyTado functions.
This has the drawback that it would probably break backwards compatibility since the functions will return a dataclass instead of a dict. This may be avoided by either implementing functions like __getitem__ so the dataclass also acts like a dict (however I'm not sure if this will work properly) or by using mashumaro just for validation and still returning a dict.

2. TypedDict

The Models mentioned above could also be converted to TypedDict's which would be great for backwards compatibility but not as nice for readability and validation.

Personally I'd favour Mashumaro since the Models already mostly exist. I think for backwards compatibility returning dicts in the Tado Class in interface.py would be good. The new methods should return mashumaru objects.

What are other opinions on this? I can start working on typing once we decided how it should be done.

[wmalgadey]

What about typing inside out? So we use strict typing everywhere we can inside the library but do not rely on heavy typing at the outside interfaces, like data classes? Much like an API, where you have to separate the typing into a definition file (swagger/openapi)

[wmalgadey]

I like generating my popo classes 😄

[Moritz-Schmidt]

I don't see any reason to use plain old python objects (I think that's what "popo" means?). Python standard library already has dataclasses, so this style of defining data-only objects is quite well known and much nicer to use. Since we also have a lot of the work already done, I would much prefer using Mashumaro (or Pydantic, which would be quite a simple change for the existing mashumaro code, and is the more popular project).

I'm against spending a lot of effort on a more manual solution where we don't use any library for deserialisation and validation of JSON responses.

[erwindouna]

That's probably what popo means, whereas Wolfgang is most likely just joking a bit. :)
Pydantic would be the better choice now. It has a better ecosystem behind it and is more popular nowadays.

[wmalgadey]

Yes just a bit joking, I'm German and POPO means butt for little kids 😄, but I still use "plain old ... objects" a lot, because I like the idea of lightweight objects, which can be de-/serialized easily. But I think this is not necessary for python environments actually.

[Moritz-Schmidt]

alright, then I'll start working on doing it with pydantic :)

[wmalgadey]

@Moritz-Schmidt can we help you somehow?

[wmalgadey]

we merged the changes from @Moritz-Schmidt to main. I will close this discussion.