Object mapping is a necessary and often unloved evil in software development projects. To communicate between layers of your application, you have to create and test mappers between a multitude of types, which can be a very cumbersome task, depending on the mapper library that is used. This article introduces reMap, yet another Java object mapper that has a unique focus on robustness and minimal testing overhead.
Specifying a Mapper
Rather than creating a mapper via XML or annotations as in some other mapping libraries, with reMap you create
a mapper by writing a few
good old lines of code. The following mapper maps all fields from a Customer
object to a
Person
object.
Mapper<Customer, Person> mapper = Mapping
.from(Customer.class)
.to(Person.class)
.mapper();
However, the above mapper specification expects Customer
and Person
to have exactly the same fields with the
same names and the same types. Otherwise, calling mapper()
will throw an exception.
Here, we already come across a main philosophy of reMap:
In your specification of a mapper, all fields that are different in the source and destination classes have to be specified.
Identical fields in the source and destination classes are automatically mapped and thus specified implicitly. Different fields have to be specified explicitly as described in the following sections. The reasoning behind this is simply robustness as discussed in more detail below.
Once you have a mapper instance, you can map a Customer
object into a Person
object by simply calling the map()
method:
Customer customer = ...
Person person = mapper.map(customer);
Omitting fields
Say Customer
has the field address
and Person
does not. Vice versa, Person
has a field birthDate
that is missing in Customer
.
In order to create a valid mapper for this scenario, you need to tell reMap to omit those fields:
Mapper<Customer, Person> mapper = Mapping
.from(Customer.class)
.to(Person.class)
.omitInSource(Customer::getAddress)
.omitInDestination(Person::getBirthDate)
.mapper();
Note that instead of referencing fields with Strings containing the field names, you use references of the corresponding getter methods instead. This makes the mapping code very readable and refactoring-safe.
Also note that this feature comes at the “cost” that mapped classes have to follow the Java Bean conventions, i.e. they must have a default constructor and a getter and setter for all fields.
Why do I have to specify fields that should be omitted? Why doesn’t reMap just skip those fields? The simple reason for this is robustness again. I don’t want to let a library outside of my control decide which fields to map and which not. I want to explicitly specify what to map from here to there. Only then can I be sure that things are mapped according to my expectations at runtime.
Mapping fields with different names
Source and target objects often have fields that have the same meaning but a different name. By using
the reassign
specification, we can tell reMap to map one field into another field of the same type. In this
example, Customer
has a field familyName
that is mapped to the name
field in Person
. Both fields are
of the same type String
.
Mapper<Customer, Person> mapper = Mapping
.from(Customer.class)
.to(Person.class)
.reassign(Customer:getFamilyName)
.to(Person::getName)
.mapper();
Mapping fields with different types
What if I need to convert a field to another type? Say Customer
has a field registrationDate
of type Calendar
that should be mapped to the field regDate
of type Date
in Person
?
private Mapper<Customer, Person> createMapper(){
return Mapping
.from(Customer.class)
.to(Person.class)
.replace(Customer::getRegistrationDate, Person::regDate)
.with(calendarToDate())
.mapper();
}
private Transform<Date, Calendar> calendarToDate() {
return source -> {
if(source == null){
return null;
}
return source.getTime();
};
}
By implementing a Transform
function that converts one type to another, we can use the replace
specification to convert a field value.
Nested Mapping
Another often-required feature of a mapper is nested mapping. Let’s say our Customer
class has a field
of type CustomerAddress
and our Person
class has a field of type PersonAddress
. First, we create
a mapper to map CustomerAddress
to PersonAddress
. Then we tell our Customer-to-Person mapper to
use this address mapper when it comes across fields of type CustomerAddress
by calling useMapper()
:
Mapper<CustomerAddress, PersonAddress> addressMapper =
Mapping
.from(CustomerAddress.class)
.to(PersonAddress.class)
.mapper();
Mapper<Customer, Person> mapper = Mapping
.from(Customer.class)
.to(Person.class)
.useMapper(addressMapper)
.mapper();
Key Philosophies
reMap has some more features that can best be looked up in the project’s documentation. However, I would like to point out some “meta-features” that make out the philosophy behind the development of reMap.
Robustness
A main goal of reMap is to create robust mappers. That means that a mapper must be refactoring-safe. A mapper must not break if a field name changes. This is why getter method references are used to specify fields instead of simple Strings.
A nice effect of this is that the
compiler already checks most of your mapping specification. It won’t allow you to specify a reassign()
for fields of a different type, for example. Another nice effect is that the compiler will tell you if you broke
a mapper by changing the type of a field.
But a mapper can be broken even if the compiler has nothing to fret about. For example, you might have overlooked a
field when specifying the mapper. This is why each mapper is validated at the earliest possible moment during runtime,
which is when calling the mapper()
factory method.
Testing
This leads us to testing. A major goal of reMap is to reduce testing effort to a minimum. Mapping is a tedious task, so we don’t want to add another tedious task by creating unit tests that manually check if each field was mapped correctly. Due to the rather brainless nature of this work, those unit tests are very error prone (in my experience, at least).
Since all validation of a mapper is done by the compiler and the mapper()
factory method, all you have to do
to test a mapper is to create an instance of the mapper using the mapper()
method. If this produces
an exception (for example when you overlooked a field or a type conversion) the test will fail.
If you want to create a fixture for regression testing, reMap supports asserting a mapper by creating an
AssertMapping
like this:
AssertMapping.of(mapper)
.expectOmitInSource(Customer::getAddress)
.expectOmitInDestination(Person::getBirthDate)
// ... other expectations
.ensure();
Calling ensure()
will throw an AssertionError
if the AssertMapping
does not match the specification of the
mapper. Having a unit test with such an assertion in place, you will notice if the specification of the mapper
does not match your expectations. This also allows test-driven development of a mapper.
Note that if you created a custom Transform
function as described above
you should include an explicit test for this transformation in your test suite, since it cannot be validated
automatically by reMap.
Performance
Performance was actually not a goal at all when developing reMap. Robustness and minimal test effort were valued much higher. However, reMap seems to be faster than some other popular mappers like Dozer and ModelMapper. The following performance test results were created on my local machine with a testing framework created by Frank Rahn for his mapper comparison blog post (beware of German language!).
Mapper | Average Mapping Time (ms) |
---|---|
JMapper | 0,01248 |
ByHand | 0,01665 |
MapStruct | 0,21591 |
Orika | 0,37756 |
Selma | 0,44576 |
reMap | 2,56231 |
ModelMapper | 4,71332 |
Dozer | 6,12523 |
Summary
reMap is yet another object mapper for Java but has a different philosophy from most of the other mappers out there. It values robustness above all else and minimal testing overhead a strong second. reMap is not the fastest mapper but plays in league of some of the other popular mappers performance-wise.
reMap is very young yet, and probably not feature-complete, so we’d love to hear your feedback and work out any bugs you might find and discuss any features you might miss. Simply drop us an issue on Github.