Search for the best match
Now that we can define an ordering on the tuple of types by using the distance comparators and tuple of types distance, let’s see how we search for the best matching method. The algorithm is based on the definition of the multimethod in multimethod definition.
Algorithm
We invoke M on a tuple of objects that has the tuple of types t. The algorithm to select the method to apply is made up of two parts.
Calculate the compatible methods
The subset C of M is defined as the set of compatible methods {mj} for which dj = distance(tj, t) is defined.
- If C is empty then the invoke call is a failure. There is no matching method.
- Otherwise we can continue for the search.
Calculate the minimum methods
In C that is non-empty, we search for the minimum distances dj using the distance comparator.
- If there is a single method mj with minimum dj then the method is selected to be applied. If tj = t, it is the perfect match. Otherwise it is the best match.
- If there are many minimum dj that are equals, there is an ambiguity because we cannot select one method.
Notice that the calculation of the distance is fixed in EVL but the minimum methods set may vary depending on the comparator used.
Examples
As examples, we use the type hierarchy defined in tuple of types distance.
To simplify, we do not show the return type and the non-virtual parameter types.
We apply the couple (D, C) to multimethods of dimension 2 defined in each line of the table:
| Name | Match methods | Compatible methods | Distances | Comparator | Selected method |
|---|---|---|---|---|---|
| M1 | (A, A), (A, D) | - | - | asymmetric | failure |
| M2 | (A, A), (A, B), (A, C) | (A, B), (A, C) | (1, 1), (1, 0) | asymmetric | (A, C) |
| M3 | (A, B), (D, B), (A, C) | (A, B), (A, C), (D, B) | (1, 1), (1, 0), (0, 1) | asymmetric | (D, B) |
| M4 | (A, B), (D, B), (A, C) | (A, B), (A, C), (D, B) | (1, 1), (1, 0), (0, 1) | symmetric | ambiguity |
| M5 | (A, B), (D, B), (A, C), (D, C) | (A, B), (A, C), (D, B), (D, C) | (1, 1), (1, 0), (0, 1), (0, 0) | symmetric | (D, C) |
These are typical different cases:
- For M1, there is no compatible method thus no method can be applied.
- For M2, (A, C) is the single minimum. It is the best match but not the perfect match.
- For M3, (D, B) is the single minimum because it is less than (A, C) for the asymmetric comparator. It is the best match but not the perfect match.
- For M4 which differs from M3 by the comparator which is symmetric, there is an ambiguity because (A, C) and (D, B) are both minimum and equals for the comparator.
- For M5, the ambiguity is solved by adding (D, C) which is the perfect match.