Add presence tracking to structure design

[JordonPhillips]

This will be needed due to smithy-lang/smithy#1286

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[nateprewitt]

Two small comments, otherwise things look good!

[nateprewitt]

I think this looks good for the code changes. It looks like both Pants and the Python Codegen build are failing though. Do we know what's going on there?

[mtdowling]

Are these sentinel values? What about just adding a dict that track which values were explicitly set vs which were defaulted?

[JordonPhillips]

They're just wrapper classes. It's not exactly a sentinel because they do actually have values.

Using a tracking dict is problematic for a few reasons. Probably the biggest is that there's a lot of ways to modify an object's attributes so making that airtight is difficult. They could, for instance, just modify the __dict__ of the object directly and thus bypass __setattr__. Maybe that's a far-fetched scenario to account for, but it is still a problem we don't need to have.

It also means that you have to set the default value to None in the object definition everywhere since there's no way to distinguish between a defaulted value and an explicitly set one. This means that the type for the member is now int | None instead of int for example, which sort of defeats one of the intents of having defaults. This also makes auto-docs like the kind that show up during dot-completion worse since they can't show you the real default value now. Actually now that I think of it, it would also be impossible to distinguish between an explicit and implicit None for document types.

It also just takes a lot more code. See below for a simplified example. While that's not quite as pressing an issue for python as for, say, JavaScript, it still is an issue.

class WithWrappers:
    def __init__(
        self,
        *,
        default_int: int = _DEFAULT_INT(7),
        default_str: str = _DEFAULT_STR("foo")
    ):
        self.default_int = default_int
        self.default_str = default_str


class WithoutWrappers:
    def __init__(
        self, *, default_int: int | None = None, default_str: str | None = None
    ):
        self._has = {}
        if default_int is None:
            self.default_int = 7
            self._has["default_int"] = False
        else:
            self.default_int = default_int
            self._has["default_int"] = True

        if default_str is None:
            self.default_str = "foo"
            self._has["default_str"] = False
        else:
            self.default_str = default_str
            self._has["default_str"] = True

    def __setattr__(self, name: str, value: Any) -> None:
        object.__setattr__(self, name, value)
        self._has[name] = True

    def _hasattr(self, name: str) -> bool:
        return self._has[name]


print(inspect.signature(WithWrappers.__init__))
# (self, *, default_int: int = 7, default_str: str = 'foo')


print(inspect.signature(WithoutWrappers.__init__))
# (self, *, default_int: int | None = None, default_str: str | None = None)

[mtdowling]

Ok sounds good. Why generate these for each service?

[JordonPhillips]

Why generate these for each service?

We don't want people to use them, so they need to be private.

[JordonPhillips]

I think this looks good for the code changes. It looks like both Pants and the Python Codegen build are failing though. Do we know what's going on there?

There's a separate PR to get the build passing on latest.

[JordonPhillips]

Hmm.... I forgot you can't subclass None or bool so wrapping may not work ideally either. :/