Add an include mode for infra.apply

[sdascoli]

In some cases where we know exactly on which model field an infra should depend, it would be much easier and safer to have a mode infra.apply(include_in_cache_uid).
Easier because it avoids copying all model fields.
Safer because if we subsequently add fields during the class development, we don't have to remember to add the new field in the exclude_from_cache_uid

[sdascoli]

The lack of this feature is particularly annoying in the following situation:

class Foo(BaseModel):
  foo1: str
  lots of other keys...

class Bar(BaseModel):
  bar1: str
  lots of other keys...

class MyClass(BaseModel):
  foo: Foo
  bar: Bar
  infra: TaskInfra

Imagine if we only want infra to depend on foo.bar1.
With the include logic, we could do infra.apply(include="foo.foo1")
With the current logic, it is impossible to do this: we would have to do sth horribly hacky like this (had to do in my pipeline)

def exclude(cls):
        out = [k for k in cls.model_fields.keys()]
        out.remove("foo")
        out += ["foo."+k for k in Foo.model_fields.keys()]
        out.remove("foo.bar1")

[kingjr]

The include exclude logic would become quite hard.

Can’t you simply use _ for the excluded params ? Eg _foo, _bar , this automatically exclude

[sdascoli]

The include exclude logic would become quite hard.

Can’t you simply use _ for the excluded params ? Eg _foo, _bar , this automatically exclude

No because I have multiple infras on my object which depend on different fields…
Why do you think this would be hard? If it’s too complex we could make them mutually exclusive in which case it would be easy

[kingjr]

Yes, we could add include_only.

My gut feeling is that this need is diagnostic of an improvable class design, but we'd need to discuss this with concrete examples.