[ty] Introduce UnionType::try_from_elements and UnionType::try_map

[AlexWaygood]

Summary

An increasingly common pattern in our codebase is to do something like this, where we iterate over a sequence of types and apply a fallible mapping operation to each element in turn:

                let mut builder = UnionBuilder::new(db);
                for element in union.elements(db) {
                    builder = builder.add(element.to_instance(db)?);
                }
                Some(builder.build())

While the mapping function continues to return Some(Type<'db>) for each element, we build up a union; but as soon as the mapping operation returns None for any element, we short-circuit and return None.

This PR adds some helper functions to UnionType to make this pattern less verbose: UnionType::try_from_elements() and UnionType::try_map.

Unfortunately there are a couple of places where a mapping function returns a Result:

ruff/crates/ty_python_semantic/src/types/infer.rs

Lines 7465 to 7480 in 27eee5a

	(Type::Union(union), other) => {
	let mut builder = UnionBuilder::new(self.db());
	for element in union.elements(self.db()) {
	builder =
	builder.add(self.infer_binary_type_comparison(*element, op, other, range)?);
	}
	Ok(builder.build())
	}
	(other, Type::Union(union)) => {
	let mut builder = UnionBuilder::new(self.db());
	for element in union.elements(self.db()) {
	builder =
	builder.add(self.infer_binary_type_comparison(other, op, *element, range)?);
	}
	Ok(builder.build())
	}

But I don't think this approach can be generalized to support mapping operations that return Results as well as mapping operations that return Options until the Try trait is stabilised in Rust.

Test Plan

cargo test -p ty_python_semantic

[github-actions]

mypy_primer results

No ecosystem changes detected ✅

[MichaReiser]

Maybe: try_map and try_map_result or try_map and try_map_ok?

and try_from_elements and try_from_result_elements?

I also think it's important to document what "fallible" means because it wasn't clear to me if it bails if there's any None value or if it skip sover None values.

[AlexWaygood]

Maybe: try_map and try_map_result or try_map and try_map_ok?

and try_from_elements and try_from_result_elements?

We could, but I don't think it's worth it right now considering that there's only one callsite that uses Results. We can add it later if that pattern becomes more common, I think.

I also think it's important to document what "fallible" means because it wasn't clear to me if it bails if there's any None value or if it skip sover None values.

👍 I'll push an update

[MichaReiser]

It's unfortunate that the Try trait isn't stabilized :(