Validate workflow job args against hook runtime fixtures

[yngve-sk]

Issue
Resolves 10334

Approach
Inspect the .run(...) arguments to identify attempts at getting fixtures that are unavailable at the given HookRuntime

[codspeed-hq]

CodSpeed Performance Report

Merging #10341 will improve performances by 11.75%

_{Comparing yngve-sk:25.03.17.workflowhooks-add-configvalidation (a1dc63d) with main (0e8c9f2)}

Summary

⚡ 1 improvements
✅ 24 untouched benchmarks

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	test_load_from_context[gen_x: 20, sum_x: 20 reals: 10]	6.5 ms	5.8 ms	+11.75%

[oyvindeide]

Can probably put almost everything except the class into a fixture?

[oyvindeide]

Can we refactor this a bit so that it is easier to keep in sync? Or make sure that the run_models use the same? Think it might be easy for this to come out of sync?

[yngve-sk]

I tried some stuff but I don't think there is a nice solution to it. We would have to have an enumeration over all hooks and their args, and somehow generically make those work across the runmodels, which would require a bigger change, and probably not be sensible with mypy. I tried this:

    def run_workflows_pre_simulation(self, fixtures: PreSimulationFixtures) -> None:
        self.run_workflows(HookRuntime.PRE_SIMULATION, fixtures)

    def run_workflows_post_simulation(self, fixtures: PostSimulationFixtures) -> None:
        self.run_workflows(HookRuntime.POST_SIMULATION, fixtures)

    def run_workflows_pre_first_update(self, fixtures: PreFirstUpdateFixtures) -> None:
        self.run_workflows(HookRuntime.PRE_FIRST_UPDATE, fixtures)

    def run_workflows_pre_update(self, fixtures: PreUpdateFixtures) -> None:
        self.run_workflows(HookRuntime.PRE_UPDATE, fixtures)

    def run_workflows_post_update(self, fixtures: PostUpdateFixtures) -> None:
        self.run_workflows(HookRuntime.POST_UPDATE, fixtures)

    def run_workflows_post_experiment(self, fixtures: PostExperimentFixtures) -> None:
        self.run_workflows(HookRuntime.POST_EXPERIMENT, fixtures)

and this:

class WorkflowFixtures(TypedDict, total=False):
    ensemble: Ensemble | None
    storage: Storage
    random_seed: int | None
    reports_dir: str
    observation_settings: UpdateSettings
    es_settings: ESSettings
    run_paths: Runpaths
    workflow_args: list[Any]
    parent: QWidget | None


class PreExperimentFixtures(WorkflowFixtures, total=False):
    random_seed: int | None


class PreSimulationFixtures(WorkflowFixtures, total=False):
    storage: Storage
    ensemble: Ensemble | None
    reports_dir: str
    random_seed: int | None
    run_paths: Runpaths


PostSimulationFixtures = PreSimulationFixtures


class PreFirstUpdateFixtures(WorkflowFixtures, total=False):
    storage: Storage
    ensemble: Ensemble|None
    reports_dir: str
    random_seed: int | None
    run_paths: Runpaths


PreUpdateFixtures = PreFirstUpdateFixtures
PostUpdateFixtures = PreFirstUpdateFixtures


class PostExperimentFixtures(WorkflowFixtures, total=False):
    random_seed: int | None

This is OK for mypy, but then to get the enumeration we would have to get the keys of the typeddict subclass, to get the list of accepted fixtures, and we'd also still have to maintain the mapping. I think this can be done with some "meta" style programming but I think it will add more complexity than it removes.