The Power of Value Objects

Although Value Objects are still underestimated players in software design, they may have enormous impact on the quality of your code and flexibility of you domain model. When applied correctly, they increase readability and expressiveness of your code making development and future maintenance much easier. Therefore they should be an elementary tool in every software craftsman’s toolbox.

In this article I want to show:

  • when to use Value Objects
  • how to use Value Objects
  • what are the benefits of using Value Objects

Wrapping primitives

Let’s assume that you have to write a batch tool which takes invoice data stored in DB, makes some complex processing and finally passes processed data to external system. A lot of conditions depend on the invoice status which is defined as INT in the database. According to business requirements, the following rules apply to the invoice status:

Negative status of invoice is not allowed. Invoices with status of 0 to 9 are considered OPEN, and with status greater than 9 are considered CLOSED. Open/Closed classification has impact on processing workflow. Status should be passed to external system in original format (number).

The first approach to modeling could be:

public class Invoice {

    private int status;

    public Invoice(int status, ....) {
        this.status = status;
    }

    public int getStatus() {
        return status;
    }
}

and when we need to use the flag:
Invoice invoice = new Invoice(2, ...);
        ....
        if (invoice.getStatus() < 0) {
            throw new IllegalStateException(
                    "Negative invoice state is not allowed");
        } else if (invoice.getStatus() <= 9) {
            // invoice is OPEN
        } else {
            // invoice closed
        }

This solution has a few drawbacks. It promotes repetitions of code (checking the status value every time we want to use it) and therefore it leads to unmaintainable code. What’s more, each programmer in the team will probably write he’s own status-checking code. It is possible that each version will behave in a bit different way (especially in more complicated cases). How to avoid it? Let’s implement a value object which will wrap the status:
public class InvoiceStatus {

    private int status;

    public InvoiceStatus(int status) {
        if (status < 0) {
            throw new IllegalArgumentException("Negative status not allowed");
        }
        this.status = status;
    }

    public boolean isOpen() {
        return status > 9;
    }

    public boolean isClosed() {
        return !isOpen();
    }

    @Override
    public String toString() {
        return isOpen() ? "OPEN" : "CLOSED";
    }

}

Here we have it all in one place, and anyone who will need to do the “> 9″ check will obviously use the provided method. Furthermore having a dedicated value object makes adding, extending and modifying functionality easier, because the code is not spread over the whole application.

This approach has also one positive side-effect. A method which needs an invoice status as an argument will look like this:

    public void doStuff(InvoiceStatus status) {
        ....
    }

Now you don’t have to worry if someone tries to invoke your method with -123 as an argument. It is guaranteed that you will get a valid invoice status. What’s more, readability increases because you know what to pass as an argument just by looking at the method signature.

Now let’s take a look at a similar example:

public class EmailAddress {

    private final String emailAddress;
    private final String domain;

    public EmailAddress(String email) {
        this.validate(email);

        this.emailAddress = email;
        domain = this.extractDomain(email);
    }

    public boolean isInDomain(String aDomain) {
        return domain.equals(aDomain);
    }

    public String getDomain() {
        return domain;
    }

    public String getEmailAddress() {
        return emailAddress;
    }

    private void validate(String phoneNumberToValidate) {
        // do some regexp, throw runtime exception if needed
    }

    private String extractDomain(String email) {
        // do string manipulation to extract email domain
        return null;
    }

}

Here we have a string data which contains complex information. When we need to extract this information from the String, a wrapping value object seems to be a good solution. Not only does it encapsulate the parsing and string manipulation logic but also takes care about validation.

Arithmetic of Value Objects

Value objects are particularly useful when your domain model includes some kind of arithmetic calculations on money, quantities, measurements etc. In such cases it is extremely useful to define methods which will play the role of operators in expression:

        Money salaryNet = new Money(20000, Money.EUR);
        Money tax = new Money(300, Money.EUR);

        Money salaryGross = salaryNet.plus(tax);

Now let’s imagine that we need to provide calculations for multiple currencies. We can model exchange rate as a smart value object:
    class ExchangeRate {
        private final Currency   sourceCurrency;
        private final Currency   targetCurrency;
        private final BigDecimal rate;

        public ExchangeRate(final Currency sourceCurrency, final Currency targetCurrency, final BigDecimal rate) {
            this.sourceCurrency = sourceCurrency;
            this.targetCurrency = targetCurrency;
            this.rate = rate;
        }

        public Money exchange(final Money money) {
            // do the computations
            return ........;
        }
    }

Now we can use it as follows:
    public Money computeSalaryInUSD() {
        Money salary = new Money(20000, Currency.EUR);
        ExchangeRate eurToUsd = new ExchangeRate(Currency.EUR, Currency.USD, new BigDecimal(1.5));

        return eurToUsd.exchange(salary);
    }

Quite intuitive, isn’t it?

Just like Money was implemented as a pair of amount and unit (currency), also quantities and measurements should be modeled as Whole Values since unit is their fundamental part. But Unit of Measure itself is a good candidate for a Value Object:

class LengthUnit {

        public final static LengthUnit METER     = new LengthUnit("meter", BigDecimal.ONE);
        public final static LengthUnit KILOMETER = new LengthUnit("kilometer", BigDecimal.valueOf(1000));

        private final String           name;
        private final BigDecimal       baseUnitMultiplier;

        public LengthUnit(final String name, final BigDecimal aBaseUnitMultiplier) {
            this.name = name;
            this.baseUnitMultiplier = aBaseUnitMultiplier;
        }

        public BigDecimal findRate(final LengthUnit otherLengthUnit) {
            return this.baseUnitMultiplier.divide(otherLengthUnit.baseUnitMultiplier);
        }

        @Override
        public String toString() {
            return this.name;
        }
    }

Like in example above, we can provide predefined constants to improve usability even more.
Now we can create a full-blown Length model, which is NOT only a number, but a number with it’s unit:
class Length {
        private final BigDecimal quantity;
        private final LengthUnit unit;

        public Length(final BigDecimal quantity, final LengthUnit unit) {
            this.quantity = quantity;
            this.unit = unit;
        }

        public Length to(final LengthUnit targetUnit) {
            return new Length(this.quantity.multiply(this.unit.findRate(targetUnit)), targetUnit);
        }

        @Override
        public String toString() {
            return this.quantity + " " + this.unit;
        }
    }

Note how clear and lucid the code is when we need to use our classes:
        Length hundredMeters = new Length(new BigDecimal(100), LengthUnit.METER);
        Length oneKilometer = new Length(new BigDecimal(1), LengthUnit.KILOMETER);

        System.out.println(hundredMeters.to(LengthUnit.KILOMETER));
        System.out.println(oneKilometer.to(LengthUnit.METER));

We defined our own internal language of objects and operators. Everyone who will come to maintain this code will immediately catch the idea, because this arithmetic of value objects is built based on analogy to the real life. Everyone (at least every software engineer ;) ) learns at school how to add numbers, what are units and what means to convert between different units. Therefore when we need to model it in code, let’s make it explicit.

When to use Value Objects?

I will not even try to make a complete list of cases when Value Objects can be useful, partially because it all depends on the domain you are working in. But basing on examples above I can list a few cases when we should consider the need to create an explicit Object for some implicit concept.

If you are dealing with:

  • primitives with limitations
  • primitives which carry some additional information
  • primitives as arguments and return values in service methods
  • money, time, amounts, units, quantities, measurements

consider introducing explicit Value Object.

Further reading:
Power Use of Value Objects in DDD by Dan Bergh Johnsson
Domain Driven Design by Eric Evans
Alanysis Patterns by Martin Fowler

Code readability: Uncover the conditions

In this article I want to show a few techniques that can make you code more expressive and readable especially when dealing with conditions and business rules.

Let’s start with a simple example. Which of these two code snippets makes more sense to you ?

        if (ratio > 13 && ratio < 23) {
            doSomeStuff();
        } else if (ratio <= 13) {
            doOtherStuff();
        }

or
        final boolean ratioInSafeRange = ratio > 13 && ratio < 23;
        final boolean ratioTooLow = ratio <= 13;

        if (ratioInSafeRange) {
            doSomeStuff();
        } else if (ratioTooLow) {
            doOtherStuff();
        }

What are the magic numbers in the first example? Why do we need to do one thing or the other depending on these numbers? Of course, second version does not explain everything. We still must know a bit about the domain to understand this code, but it tells us that there probably is some range of values which is safe, and when the ratio is outside of this range we should act in some different way.

In second version, we gave the conditions their  names, emphasising their meaning. Some would say that we can simply add comments that will explain what is going on. That’s true, but code changes quickly and comments become outdated and misleading in time. Comments are difficult to maintain and a false comment becomes rather an obstacle than help.

The solution presented above adds more expression to the code itself. It make the comment a part of the code, and therefore makes it more flexible. As the code evolves, developer can use refactoring techniques, bulk rename and so on to keep the comment-in-code up to date.

Now lets take a look at a bit more advanced example. Consider a banking system with the following domain interface:

interface Account {
    public Date creationDate();

    BigDecimal balance();

    BigDecimal avarageMonthlyBalance();

    void debit(BigDecimal amount);

    void credit(BigDecimal amount);
}

We also have a domain service to make fund transfers:
class TransferService {
    public void transfer(final Account source, final Account destination, final BigDecimal amount) {
        source.debit(amount);
        destination.credit(amount);
    }
}

Now the business comes with a new requirement:

For every fund transfer from source account to destination account, a source account should pay a fee, unless the source account fulfills conditions of being a premium account. A premium account is an account which exists at least one year in the system and has an average monthly balance of $1000 or more for the last month.

We decide to put this logic into the TransferService, since the fee is to be paid for transfer and not for debit transaction. Let’s take a look at the first draft:

class TransferService {
    private static final BigDecimal TRANSFER_FEE = BigDecimal.ONE;

    public void transfer(final Account source, final Account destination, final BigDecimal amount) {
        source.debit(amount);
        destination.credit(amount);

        final Calendar calendar = Calendar.getInstance();
        calendar.add(Calendar.YEAR, -1);
        if (!(source.creationDate().compareTo(calendar.getTime()) <= 0 && source.avarageMonthlyBalance().compareTo(BigDecimal.valueOf(1000L)) > 0)) {
            source.debit(TRANSFER_FEE);
        }
    }
}

After extracting the date computation into utility class, and applying the named condition pattern, we get the following:
    public void transfer(final Account source, final Account destination, final BigDecimal amount) {
        source.debit(amount);
        destination.credit(amount);

        final boolean accountExistsForAtLeatOneYear = source.creationDate().compareTo(DateTimeUtils.oneYearAgo()) <= 0;
        final boolean monthlyBalanceGreaterThanRequired = source.avarageMonthlyBalance().compareTo(BigDecimal.valueOf(1000L)) > 0;

        if (!(accountExistsForAtLeatOneYear &amp;&amp; monthlyBalanceGreaterThanRequired)) {
            source.debit(TRANSFER_FEE);
        }
    }

Just a small refactoring inside the if statement and here’s the final version of the method:
    public void transfer(final Account source, final Account destination, final BigDecimal amount) {
        source.debit(amount);
        destination.credit(amount);

        final boolean accountExistsForAtLeatOneYear = source.creationDate().compareTo(DateTimeUtils.oneYearAgo()) <= 0;
        final boolean monthlyBalanceGreaterThanRequired = source.avarageMonthlyBalance().compareTo(BigDecimal.valueOf(1000L)) > 0;

        if (!accountExistsForAtLeatOneYear || !monthlyBalanceGreaterThanRequired) {
            source.debit(TRANSFER_FEE);
        }
    }

Much better, isn’t it?

What we achieved is a code which is much more readable and expressive than before. It reveals concepts that were hidden in the first version in a form of complex comparisons and logical operators.

But we can go further. In this case being or not a premium account is a business rule, a specification. It’s not programmers job to define what are the conditions and rules for an account to become premium. If so, lets make it an explicit business rule with a name which we can use talking to business people. When discussing with domain experts we will not talk in terms of IFs, but we can talk in terms of specifications, policies etc.

class PremiumAccountSpecification {

    public boolean isSatisfiedBy(final Account candidate) {
        final boolean accountExistsForAtLeatOneYear = candidate.creationDate().compareTo(DateTimeUtils.oneYearAgo()) <= 0;
        final boolean monthlyBalanceGreaterThanRequired = candidate.avarageMonthlyBalance().compareTo(BigDecimal.valueOf(1000L)) >= 0;

        return accountExistsForAtLeatOneYear &amp;&amp; monthlyBalanceGreaterThanRequired;
    }
}

and the TransferService now looks like this:
    public void transfer(final Account source, final Account destination, final BigDecimal amount) {
        source.debit(amount);
        destination.credit(amount);

        PremiumAccountSpecification premiumAccountSpecification = new PremiumAccountSpecification();
        if (!premiumAccountSpecification.isSatisfiedBy(source)) {
            source.debit(TRANSFER_FEE);
        }
    }

When writing software, we are constantly facing problem of losing important concepts in the code. It is important for the model to hide details behind abstractions, but also to make important things explicit. Business rules are a good example where explicitness pays off. The patterns presented above are useful not only in small scale where they do their job for readability and ease of maintenance of methods and classes, but also in the scale of the domain model, where they are first-class citizens (in form of Specifications) and do their contribution to the domain ubiquitous language.

Follow

Get every new post delivered to your Inbox.