Fix _fix_anno mangling bare list annotations

[jackhogan]

Bare list annotations (without type params) causes silent data corruption — e.g. ["foo", "bar"] becomes ["b", "a", "r"] — and we get an IndexError for empty lists.

e.g. the above works with list[str] but doesn't with just list

[gitnotebooks]

Found 1 changed notebook. Review the changes at https://app.gitnotebooks.com/AnswerDotAI/fasthtml/pull/841

[jph00]

@jackhogan definitely agree the current behavior is not good - although I'm wondering if the correct behavior for non-specialised list would be the same as having no type at all? Because without the generic sub-type, we don't know what it's meant to be a list of. Or do we just assume it means "list of strs"?

If we want the latter, we should probably also support subclasses of both list and tuple? is list is perhaps a bit too specific.

[jackhogan]

Good point @jph00! What do you think of this revised approach:

• Bare list/tuple now warns and returns None — consistent with how unannotated params are handled. Uses issubclass(t, (list, tuple)) rather than is list, so subclasses are covered too.
• Validate at declaration-time — factored out the annotation check logic to be used at route declaration time (in _endp) so the user sees the warning straight away when they define the endpoint, and also at runtime (in _find_p).

I also found a different bug while investigating: Optional[list[str]], list[str] | None, and other Union-wrapped list types were also silently mangling data — ['foo', 'bar'] → ['b', 'a', 'r']. This was because after unwrapping the Union in _fix_anno, origin was stale (still Union instead of list), so the _mk_list path was skipped. Fixed by re-deriving origin after the Union unwrap.

[jph00]

Thanks @jackhogan :)