Tests that specialization is working correctly:

• Dispatch

  • On methods, includes:
    • Specialization via adding a trait bound
      • Including both remote and local traits
    • Specialization via pure structure (e.g. (T, U) vs (T, T))
    • Specialization via concrete types vs unknown types
      • In top level of the trait reference
      • Embedded within another type (Vec<T> vs Vec<i32>)
  • Specialization based on super trait relationships
  • On assoc fns
  • Ensure that impl order doesn't matter

• Item inheritance

  • Correct default cascading for methods
  • Inheritance works across impls with varying generics
    • With projections
    • With projections that involve input types

• Normalization issues

  • Non-default assoc types can be projected
    • Including non-specialized cases
    • Including specialized cases
  • Specialized Impls can happen on projections
  • Projections and aliases play well together
  • Projections involving specialization allowed in the trait ref for impls, and overlap can still be determined
    • Only works for the simple case where the most specialized impl directly provides a non-default associated type

• Across crates

  • For traits defined in upstream crate
  • Full method dispatch tests, drawing from upstream crate
    • Including additional local specializations
  • Full method dispatch tests, without turning on specialization in local crate
  • Test that defaults cascade correctly from upstream crates
    • Including additional local use of defaults