fix(gwf): add field default for package singletons

[mjreno]

fix #230

[wpbonelli]

nice, this goes part of the way to addressing Joeri and Marnix's point about it not being clear what is/isn't required when constructing (any subset of) a simulation.

[wpbonelli]

could also modify the type hints like Ic | None for each of these singles, to make clear the package isn't required in e.g. the gwf model initializer.

[mjreno]

do you mean like ic: Ic | None = field(block="packages", default=None)

[wpbonelli]

Yeah exactly.

[mjreno]

When I try that there seems to be a xattree issue, here's the unaltered quickstart test result:

cls = <class 'flopy4.mf6.gwf.ic.Ic'>

    def _find_field(cls: type) -> str:
        matches = set()
        for name, field in parent_spec.items():
            if isinstance(field, Child) and isclass(field.type) and issubclass(cls, field.type):
                matches.add(name)
        match len(matches):
            case 0:
>               raise TypeError(
                    f"Class '{parent_cls.__name__}' has no fields of type {cls.__name__}"
                )
E               TypeError: Class 'Gwf' has no fields of type Ic

../.pixi/envs/dev/lib/python3.11/site-packages/xattree/__init__.py:734: TypeError

Maybe this is related to top down initialization- gwf is created without Ic and then passed in as the parent when Ic is created.

[wpbonelli]

Ah, shoot. I think it's probably failing to parse the child type correctly when it inspects the class to build the xatspec, then getting confused when it can't find any Ic children. The introspection pass probably doesn't "unwrap" optionals (i.e. union of something and None) correctly. I'll have a look later, unless you want to. It'll be in _get_xatspec somewhere but that is a prickly pile of code.

[wpbonelli]

Sorry I'm on mobile so might be wrong, but I think if right below this line, still in the optional condition https://github.com/wpbonelli/xattree/blob/develop/xattree%2F__init__.py#L628 if we do if has(type_) to see if the unwrapped type is another xattree cls, then set is_child = True and child_kind = "only"? That might do it?

[wpbonelli]

Ok it's fixed in xattree, if you update the pixi env it should be good to go.

[mjreno]

nice!

[mjreno]

nice, this goes part of the way to addressing Joeri and Marnix's point about it not being clear what is/isn't required when constructing (any subset of) a simulation.

maybe this means npf and ic should not set the default? some packages are implicitly required by mf6

[wpbonelli]

maybe this means npf and ic should not set the default? some packages are implicitly required by mf6

yeah, then those shouldn't have the optional type hint or the none default. do we know which packages are required? I think this came up previously and you said the IO guide doesn't explicitly state this. I don't see it either. It does say how many of certain packages are allowed.

[mjreno]

maybe this means npf and ic should not set the default? some packages are implicitly required by mf6

yeah, then those shouldn't have the optional type hint or the none default. do we know which packages are required? I think this came up previously and you said the IO guide doesn't explicitly state this. I don't see it either. It does say how many of certain packages are allowed.

Yeah maybe we should leave everything optional (except dis?) for now and deal with that somehow in the spec when it becomes relevant to distinguish.