[Aschenbath]

Summary

Fixes #348.

configure_tagged_union only registered hooks for the exact union it was given. A field typed with a subset of that union (for example A | B when A | B | C was configured) fell back to the default disambiguation, which fails for members it cannot tell apart:

configure_tagged_union(A | B | C, converter=conv)

@define
class Bar:
    y: A | B

conv.structure({"y": {"x": 1, "_type": "A"}}, Bar)
# TypeError: Could not disambiguate between A, B: no unique non-default attributes

The strategy now also registers structure and unstructure hook factories that claim any other union whose members are a subset of the configured union's members. The tag alone decides the member, and every member of a subset is already in the tag map, so the same hooks are reused. An additional None member is allowed, so optional subsets (A | B | None) work too; at least two real members are required, so a plain Optional[member] field is not claimed. A subset that is configured or registered separately keeps its own direct hooks, which take precedence over these factories.

Verification

• The example from the issue now structures and unstructures through the tag, including a round trip and an A | B | None field.
• New test test_subset_union in tests/strategies/test_tagged_unions.py.
• tests/strategies/, tests/test_unions.py and tests/test_disambiguators.py: 327 passed, 15 xfailed.
• ruff check and ruff format clean on the touched files.

[Aschenbath]

Closing this one myself: I opened too many PRs here in a short time, which isn't fair to your review queue — sorry about that. I'm leaving #827 and #822 open. If this fix is something you'd want, I'm happy to reopen or rebase it later.