Specific container type isn't respected, e.g. list can populate set

[vreuter]

• cattrs version: 24.1.2
• Python version:

>>> import sys
>>> sys.version_info
sys.version_info(major=3, minor=11, micro=9, releaselevel='final', serial=0)

• Operating System: macOS Ventura

Description

Structure raw data into an instance of a class, in such a way that container type is respected.

What I Did

>>> @define
... class C:
...     a: int
...     b: set[str]
...
>>> instance = structure({'a': 1, 'b': ['x', 'y']}, C)
>>> instance
C(a=1, b={'x', 'y'})

What I Expect

I expect a failure, since the list[str] value for the b key doesn't conform to the set[str] type of the corresponding member of the class. This violates separation of concerns: the structure call is also deduplicating elements of the list, then, for the conversion to a set. As such, a reasonable invariant like "each collections in the raw data should have the same size as the corresponding collection once structured into the class" can be violated.

[vreuter]

Possibly related to #44 ?

[Tinche]

Yeah, the hooks for various collections only require the input type is iterable. Validating the exact input types would be less useful - you wouldn't be able to use set[T] in anything loaded from json since no json library produces sets.

I think it helps to think of cattrs as a tool for transformations, not direct validations - dicts get turned into classes, strings into enums, iterables into collections.

The use case you describe is possible. You can opt into stricter validation for lists as documented here.

[vreuter]

Thanks @Tinche