MAINT: migrated code for CheckedArray/Session to use pydantic v2

[alixdamman]

fix #1075

[gdementen]

Thanks A LOT for tackling this. It would have taken me ages to do this myself. This looks very good.

There are a few error messages which now have redundant parts because of the new "Error while assigning value to variable X" part, but this can be improved later.

[gdementen]

There is probably a typo in there. CheckedArray? CheckedSession?

[alixdamman]

Yes, it was a typo. I will change it to CheckedSession.

[gdementen]

Could you use a permalink or a link to a particular release? main/master tend to change over time and the link will probably break at some point.

[alixdamman]

OK

[gdementen]

Ouch. That's a lot of internal api usage. I suppose there was no other way (honestly, I don't remember why we couldn't use stock pydantic previously), but we can only hope it stays stable 🤞.

[alixdamman]

Indeed. What we could do for now is to fix the pydantic version to 2.12 in requirements.txt and in make_release.py ?

I do not remember why I implemented a simplified version of the ModelMetaclass. I think it was to allow unloaded members after the initialization of the checked session (when reading from a file). I was also to allow the declaration of members without a default value in CheckedParameters (defined using the frozen flag of pydantic.

In the __init__ method of the CheckedSession class, we have:

            for name, field in self.__pydantic_fields__.items():
                value = input_data.pop(name, NotLoaded())

                if isinstance(value, NotLoaded):
                    if field.default is PydanticUndefined:
                        warnings.warn(f"No value passed for the declared variable '{name}'", stacklevel=2)
                        self.__setattr__(name, value, skip_frozen=True, skip_validation=True)
                    else:
                        self.__setattr__(name, field.default, skip_frozen=True)
                else:
                    self.__setattr__(name, value, skip_frozen=True)

            # --- undeclared variables
            for name, value in input_data.items():
                self.__setattr__(name, value, skip_frozen=True, stacklevel=2)

Maybe the v2 version of pydantic has a special mechanism to write a sort of dedicated __init__ method or to skip validation and set frozen to False during the initialization process only. So that we could make CheckedSession to inherit from pydantic BaseModel. But that requires more investigation in the pydantic documentation plus the time to make some proof of concept implementation. But I don't have time for the moment. But I can open an issue for that and come back to it later if you want.

[gdementen]

Yes, that would be very much appreciated. This can come later though.

[gdementen]

do the instances actually change now or did you only change the tests? If the instance do not change, I would rather keep the "is" tests because that is a stronger test.

[alixdamman]

Yes it seems that the instances change for some types (but not all). At least, they change for dict objects (but not the others). I can go back to is for all members except d (which is of type dict)

[gdementen]

Do you actually use 2.12 specific features or is it because you fear it does not work with earlier pydantic versions?

[alixdamman]

Second option. It because you fear it does not work with earlier pydantic versions. I made changes in the code and I runned all the tests with pydantic 2.12 installed on my PC.

[alixdamman]

I checked on conda (with conda search pydantic) that pydantic 2.12 is available for Python 3.9+ :

> conda search pydantic

(...)
pydantic                      2.12.2  py39haa95532_0  pkgs/main
pydantic                      2.12.3 py310haa95532_0  pkgs/main
pydantic                      2.12.3 py310haa95532_1  pkgs/main
pydantic                      2.12.3 py311haa95532_0  pkgs/main
pydantic                      2.12.3 py311haa95532_1  pkgs/main
pydantic                      2.12.3 py312haa95532_0  pkgs/main
pydantic                      2.12.3 py312haa95532_1  pkgs/main
pydantic                      2.12.3 py313haa95532_0  pkgs/main
pydantic                      2.12.3 py313haa95532_1  pkgs/main
pydantic                      2.12.3 py314haa95532_1  pkgs/main
pydantic                      2.12.4 py310haa95532_0  pkgs/main
pydantic                      2.12.4 py311haa95532_0  pkgs/main
pydantic                      2.12.4 py312haa95532_0  pkgs/main
pydantic                      2.12.4 py313haa95532_0  pkgs/main
pydantic                      2.12.4 py314haa95532_0  pkgs/main

[gdementen]

I did that check too 😉