wiki:GeneralizedEqualCowan

Version 2 (modified by cowan, 5 years ago) (diff)

--

Generalized equal? predicate

(generalized-equal? obj1 obj2 [ = [ string=? [ char=? [ symbol=? [ recursive=? ] ] ] ] ] )

Compares obj1 and obj2 for equality. If both are pairs, vectors, or bytevectors, or belong to the same implementation-defined container types, returns #t if they are equal in length and the same call to generalized-equal? returns #t on all the components. If both are numbers, they are compared with the procedure =; if both are characters, they are compared with the procedure char=?; if both are strings, they are compared with the procedure string=?; if both are symbols, they are compared with the procedure symbol=?. The default values of these are =, char=?, string=?, and symbol=? respectively.

In all other cases, the procedure recursive=? is invoked with three arguments: obj1, obj2, and a two-argument procedure which returns what generalized-equal? returns given the seven arguments passed to or defaulted on it in this call. This allows recursive=? to return whatever it likes on atomic types it knows about, and to invoke its third argument on the components of composite types it knows how to descend into. It is recommended that when recursive=? receives arguments it does not know how to handle, that it invokes eqv? on them. The default value of recursive=? always applies eqv? to its first two arguments.

Objections

It does not permit implementation-defined container types to be different in aspects which either are not "components" (if that is narrowly defined) or which should be compared by means other than generalized-equal?. Furthermore, it requires implementation-defined container types to have a notion of length instead of shape.

The four default arguments for builtin types strikes me as a wart, in that it is not extensible to a future version which has additional builtin types.

The language for the argument to recursive=? is a little unclear.

It is not possible to interpose on specific types of containers without building a full wrapper around generalized-equal?, because the built-in-containers case comes before recursive=?. This means that it is a breaking change for an implementation to *add support for* a given container type to generalized-equal?, because an application might have been specially handling it.