This article is more than 1 year old

First among equals

The contract metaphor is an effective way of approaching API design

Kevlin HenneyColumn Whether formal or informal, a contract defines an (in principle) enforceable agreement between two or more parties with respect to a specific undertaking. The same is also true in code. The contract metaphor is an effective way of approaching API design and use [1]: "A contract is effectively an agreement on the requirements fulfilled by a component between the user of the component and its supplier".

One way of capturing a functional contract is through the use of pre- and postconditions that state what must be true before an operation is to be called, for it to be called correctly, and what must be true after an operation has returned, for it to be considered correct [2]. To consider contracts only with respect to pre- and postconditions, however, offers a slightly limiting and incomplete — albeit common — view of the contract metaphor, although it clearly offers a great deal of coverage, utility and insight [3].

Comparing two objects for some kind of equality offers fertile ground for exploring different conventions and contracts [4, 5].

Beyond pre- and postconditions

The contract for the equals method in Java can be phrased most easily and clearly in terms of named constraints, each one stated as a simple truth:

reflexive:
a.equals(a)
symmetric:
a.equals(b) if and only if b.equals(a)
transitive:
a.equals(b) && b.equals(c) implies a.equals(c)
consistent:
a.equals(b) returns the same as long as a and b are unmodified
null inequality:
!a.equals(null)
hashCode equality:
a.equals(b) implies a.hashCode() == b.hashCode()

This contract is binding on any override of the Object.equals method. If you try to state this in terms of preconditions — what must be true before a successful call to equals — and postconditions — what must be true after a successful call to equals — from a strictly object-centric viewpoint you will find a loss of clarity as well as a loss of part of the contract. Assuming that the argument to equals is named other:

postcondition where other == null:
The result is false
postcondition where other == other:
The result is true
postcondition otherwise:
The result is the same as the result of other.equals(this), and where true then hashCode() == other.hashCode().

As you can see, there is no useful precondition and the postcondition is in part partitioned by the named constraints we had before, but not as clearly. More reading between the lines is needed to get the full sense of the contract. The consistency constraint is a difficult one to express at the best of times, but its temporal nature means that it cannot be properly expressed as the postcondition of an operation, which describes only what is true at the point of completion of the method call, and nothing beyond.

There is also a subtle loop here: if the result of calling equals is required to be the same as the result of making the same call but with the argument and receiving object switched round, what is the requirement on that second call? It would be the first call. Any literal implementation of this would find itself caught in infinite recursion.

Although the contract for overriding Object.equals in Java cannot be as well stated in terms of pre- and postconditions, the truths it uses are assertible from a testing perspective. For a given a, b and c that are supposed to compare equal we can assert the following, using the Java 1.4 assertion mechanism:

assert a.equals(a)                  : "reflexive";
assert !a.equals(null)              : "null inequality";
assert a.hashCode() == b.hashCode() : "hashCode equality";
assert a.equals(b) && b.equals(a)   : "symmetric";
assert b.equals(c) && a.equals(c)   : "transitive";
assert a.equals(b)                  : "consistent";

Of course, the consistency constraint check cannot be a thorough one, but it simply reinforces that in the simple test suite shown yet another equality comparison between a and b will yield true.

It is worth noting that contrary to the way that it is often presented — including in Sun's own documentation — the relationship between equals and hashCode is more properly a part of the equality contract, which deals with relations between objects, than it is of the hashing contract, which deals with individual objects.

More about

TIP US OFF

Send us news


Other stories you might like