fix: handle synthetic oneofs for proto3 optional fields

[dsociative]

Fix path generation for proto3 optional fields

Problem

Proto3 optional fields are implemented using synthetic oneofs (e.g., oneof _field { string field = 1; }). prost-validate was treating these as regular oneofs, generating incorrect paths like message._field.field instead of message.field.

Solution

Added containing_non_synthetic_oneof() helper that filters out synthetic oneofs using is_synthetic() check, ensuring correct path generation for optional fields.

Updated prost-reflect to the latest version and all its dependencies, as the is_synthetic() method on OneofDescriptor was only added in recent versions.

Changes

• Synthetic oneofs (proto3 optional): message.field
• Real oneofs: message.oneof_name.field
• Regular fields: message.field
• Updated prost-reflect and dependencies to support is_synthetic() API

[Adphi]

Thanks a lot for this contribution @dsociative!

When you get a chance, could you please rebase the PR on the latest main branch?
That should help us move it forward. Thanks again!

[dsociative]

Thanks for the review! I've rebased on the latest main.

I've encountered an edge case with this test that's now failing due to the fix. The proto defines an optional field with message.required = true:

message MessageRequiredButOptional { optional TestMsg val = 1 [(validate.rules).message.required = true]; }

With Rust's type system, we can't distinguish between an unset optional field and one explicitly set to None. What should the expected validation behavior be in this case?

[Adphi]

Thanks @dsociative !

Can you please run git rebase <main repo remote name>/main and resolve the Cargo.toml and Cargo.lock conflicts ?
There shouldn’t be any changes in them.

I’ll check the failing test.

[Adphi]

Thanks @dsociative !
The fix in #29 resolves the tests, so I’m closing this.