Collection of Collections Is a Code Smell

From WikiContent

Revision as of 09:34, 6 July 2009 by Kcpeppe (Talk | contribs)
Jump to: navigation, search

Every once in a while I run into a design problem to which the answer is combination of a collection of collections. While some things, such a matrices, are naturally represented this way, more often than not using a collection of collections is a code smell that indicates that you are missing an abstraction that is key to your domain.

Code that "smells" (Kent Beck), is used to describe code that is awkward or otherwise doesn't, well, smell quite right. Awkwardness in code tends to point to some deeper underlying problem with either the underlying implementation or the overall design. In the case of a collection of collections, we often see an indication that one needs an abstraction to represent the relationship that is being expressed in the collection of collections. Quite often this missing abstraction is translated to a new class in your domain that represents some vocabulary that is missing from your model. In the following code the key in the outer map is keyed on a persons last name and the inner collection is keyed on a persons first name:

public class AllPersons {
    private HashMap<String, HashMap> allPersons = new HashMap<String, HashMap>();

    public void addPerson(Person person) {
        HashMap<String, HashMap> persons = allPersons.get(person.getLastName());
        if (persons == null) {
            persons = new HashMap<String, Person>();
            allPersons.add(person.getLastName(), persons);
        persons.add(person.getFirstName(), person);

The code smell in the example is that we are keying on two different values, one after the other. The question is, what is the missing design element if there is one?

One of the roles that a collection can play is to implicitly create an index over a collection much in the same way we'd create an index in a database table. If we were to create an index on a single column, that would be a simple key. If we combine two or more simple keys to create another index, we have created a compound key. And this is exactly what we are doing in this example, creating an index based on two fields. From this we can conclude that the missing design element is a compound key. The following code demonstrates the effect of implementing our newly discovered abstraction:

public class AllPersons {
    private HashMap<CompoundKey, Person> allPersons = new HashMap<CompoundKey, Person>();

    public void addPerson(Person person) {
        allPersons.add(new CompoundKey(person.getFirstName(), person.getLastName()), person);

If you found the second listing to be more readable than the first, then you've already experienced the biggest benefit, the added abstraction's effect on readability. Another benefit is that the added abstraction often results in less code. Unfortunately this point isn't that clear when presented in a short example such as this. The benefits are realized only after repeated use of the abstraction. However the end results is that not only do we have less code to read, the code is more readable to begin with. In this case the abstraction also gives us a better memory utilization profile.

The next time you come across a collection or collections, think code smell. And then think, what is the missing abstraction. If you pay attention to these code smells, you'll quickly start to notice an improvement in your code base.

By Kirk Pepperdine

This work is licensed under a Creative Commons Attribution 3

Back to 97 Things Every Programmer Should Know home page

Personal tools