To make value types usable, it should be possible to override the operators that are used with them. The system I sketch out here is not specific to value types, however. It is based on Brendan's ideas for multidispatch.
WARNING: This is very preliminary and not even close to thought through.
The idea is to employ multidispatch. The correct function to apply is based on the prototype of both the left and right hand sides (note that the prototype of value operands is obtained via the registry described in the value type proposal). There is some form of operator registry that lets you add pairs of prototypes and, when an operator is applied, the most specific one is chosen. Ideally, the existing coercion rules would be subsumed under this registry.
We might wish to guarantee that adding prototype pairs to the registry never disturbs existing pairs (that is, the registry grows monotonically). Alternatively, maybe allow users to freeze prototype pairs.
We do not attempt to guarantee algebraic equivalence guarantees and the like (IEEE floating point already removes most of them anyhow).
- What operators can be overloaded?
- Not
===, but yes== +,-, etc- Not
., but what about[]?
- Not