1. The equality symbol is now bidirectionally compatible with coercions. That is, in the module Binary below (see bench/typing/good/coercions.mlw), goal G is well-typed, with coercion [to_int] being applied to variable [x] both in [x=y] and in [y=x] : module Binary use import int.Int use import mach.int.Int63 goal G: forall x: int63, y: int. x = y /\ y = x end 2. Each coercion is still uniquely defined by type symbols [Ts, Ts'] [ ls: Ts1 ty_1,1 ... ty_1,n -> Ts2 ty_2,1 ... ty_2,m ], but searching for a possible coercion is now taking into account types [ ty_1,1 ... ty_1,n ] and [ ty_2,1 ... ty_2,m ] in order to compare them with the actual type of a term [t] on which coercion may apply as well as the actual expected type. (That is, now the signature of find is [t -> ty -> ty -> lsymbol list]) This is done to anticipate in a heurstic way a situation when a coercion can be technically applied but will eventually fail, because type mismatch between types in coercion signature and the types of a term on which coercion applies. For instance, given a coercion function to_list bool : list bool meta coercion function to_list the following goal is rejected by typing: goal G: forall x: list int, y: bool. x = y Indeed, [y] is expected to be of type [list int]. Since [y] is of type [bool], one may try to apply a coercion [to_list] in order to get a term [to_list y] of type [list bool]. However, this would not help, since a [list bool] mismatch with [list int]. The error message is now better: "This term (i.e. [y]) has type bool, but is expected to have type list int" compared to old version: "This term (i.e. [y]) has type list bool, but is expected to have type list int"
Showing with 138 additions and 39 deletions