Switch to typing-inspection
#556
Conversation
Viicos
force-pushed
the
typing-inspection
branch
from
d93b372 to
f1c9d3c
Compare
| return str(obj) | ||
|
|
||
| if not isinstance(obj, (typing_base, WithArgsTypes, type)): | ||
| origin = get_origin(obj) |
There was a problem hiding this comment.
This function is still a big mess, in the future I'd like to add a string repr function in typing-inspection
| annotation = root_annotation | ||
| except TypeError: | ||
| pass | ||
| if annotation is not None and _lenient_issubclass(annotation, RootModel) and annotation is not RootModel: |
There was a problem hiding this comment.
_lenient_issubclass takes care of the TypeError.
Also, checking for annotation is not RootModel and relying on model_fields['root'].annotation is more robust
| @@ -2365,22 +2364,23 @@ def _annotation_contains_types( | |||
|
|
|||
|
|
|||
| def _strip_annotated(annotation: Any) -> Any: | |||
There was a problem hiding this comment.
Python takes care of flattening Annotated forms already.
| if sys.version_info < (3, 10): | ||
| _WithArgsTypes = tuple() | ||
| else: | ||
| _WithArgsTypes = (_GenericAlias, types.GenericAlias, types.UnionType) |
There was a problem hiding this comment.
This is taken from Pydantic 2.9. The < (3, 10) branch was actually wrong (it was missing _GenericAlias), and pydantic-settings relied on this inconsistency.
There was a problem hiding this comment.
Actually this was reported here.
There was a problem hiding this comment.
I'm keeping the inconsistency here because the metavar format function is really messy and I can't reason properly when looking into it. It needs to be refactored and as mentioned in another comment this will probably be done in typing-inspection.
Viicos
force-pushed
the
typing-inspection
branch
from
f1c9d3c to
ceb3253
Compare
Viicos
force-pushed
the
typing-inspection
branch
from
ceb3253 to
5fadd43
Compare
Also fix some logic to be a bit more robust.
I'll note that the code base rely heavily on inspection of typing objects, but this fragile for several reasons (one example is when determining if a union is complex: this doesn't take into account PEP 695 type aliases).
Ideally most of the logic should rely on the core schema instead, but this isn't trivial as well.
Closes #553.