Make config for different step types into separate classes

[sondreso]

Currently the config for steps are vary cluttered and difficult to read. This is because the different types requires different validation, as seen below:

ert/ert3/config/_stages_config.py

Lines 83 to 112 in d1385bb

	class Step(_StagesConfig):
	name: str
	type: Literal["unix", "function"]
	script: Optional[List[str]] = []
	input: List[InputRecord]
	output: List[OutputRecord]
	transportable_commands: Optional[List[TransportableCommand]] = []
	function: Optional[Callable]
	@root_validator
	def check_defined(cls, step):
	cmd_names = [cmd.name for cmd in step.get("transportable_commands", [])]
	script_lines = step.get("script", [])
	if step.get("type") == "function":
	if cmd_names:
	raise ValueError("Commands defined for a function stage")
	if script_lines:
	raise ValueError("Scripts defined for a function stage")
	if not step.get("function"):
	raise ValueError("No function defined")
	return step
	@validator("function", pre=True)
	def function_is_valid(cls, function: str, values) -> Optional[Callable]:
	step_type = values.get("type")
	if step_type != "function" and function:
	raise ValueError(f"Function defined for {step_type} step")
	if function is None:
	return None
	return _import_from(function)

Suggestion is to instead make different sub-classes for the different types, and let them handle the validation.

POC:

# test.py
from typing import Union, List

from pydantic import BaseModel, Field


class BasicModel(BaseModel):
    name: str


class SubModelA(BasicModel):
    type: str = Field("A", const=True)
    some_property: int


class SubModelB(BasicModel):
    type: str = Field("B", const=True)
    some_other_property: str


class Model(BaseModel):
    l: List[Union[SubModelA, SubModelB]]


m = Model.parse_obj(
    {
        "l": [
            {"name": "a", "type": "A", "some_property": 3},
            {"name": "my_b", "type": "B", "some_other_property": "3"},
        ]
    }
)
assert isinstance(m.l[0], SubModelA)
print(m.l[0], type(m.l[0]))
assert isinstance(m.l[1], SubModelB)
print(m.l[1], type(m.l[1]))

$ python test.py
name='a' type='A' some_property=3 <class '__main__.SubModelA'>
name='my_b' type='B' some_other_property='3' <class '__main__.SubModelB'>

Interesting links:
pydantic/pydantic#503
https://github.com/samuelcolvin/pydantic/pull/469/files
https://stackoverflow.com/questions/58301364/pydantic-and-subclasses-of-abstract-class
pydantic/pydantic#619
pydantic/pydantic#265