Core Java

Functional Java by Example | Part 4 – Prefer Immutability

This is part 4 of the series called “Functional Java by Example”.

In previous part we talked a bit about side effects and I’d like to elaborate a bit more about how we can prevent having our data manipulated in unexpected ways by introducing immutability into our code.

If you came for the first time, it’s best to start reading from the beginning.

It helps to understand where we started and how we moved forward throughout the series.

These are all the parts:

I will update the links as each article is published. If you are reading this article through content syndication please check the original articles on my blog.

Each time also the code is pushed to this GitHub project.

Pure functions

A small summary on what we discussed before.

  • Functional Programming encourages side-effect free methods (or: functions), to make the code more understandable and easier to reason about. If a method just accepts certain input and returns the same output every time – which makes it a pure function – all kinds of optimizations can happen under the hood e.g. by the compiler, or caching, parallelisation etc.
  • We can replace pure functions again by their (calculated) value, which is called referential transparancy.

Here’s what we currently have after the refactoring from the previous part:

class FeedHandler {

  Webservice webservice
  DocumentDb documentDb

  void handle(List<Doc> changes) {

    changes
      .findAll { doc -> isImportant(doc) }
      .each { doc ->
        createResource(doc)
        .thenAccept { resource ->
          updateToProcessed(doc, resource)
        }
        .exceptionally { e ->
          updateToFailed(doc, e)
        }
      }
  }

  private CompletableFuture<Resource> createResource(doc) {
    webservice.create(doc)
  }

  private boolean isImportant(doc) {
    doc.type == 'important'
  }

  private void updateToProcessed(doc, resource) {
    doc.apiId = resource.id
    doc.status = 'processed'
    documentDb.update(doc)
  }

  private void updateToFailed(doc, e) {
    doc.status = 'failed'
    doc.error = e.message
    documentDb.update(doc)
  }

}

Our updateToProcessed and updateToFailed are “impure” — they both update the existing document going in. As you can see by their return type, void, in Java this means: nothing comes out. A sink-hole.

private void updateToProcessed(doc, resource) {
  doc.apiId = resource.id
  doc.status = 'processed'
  documentDb.update(doc)
}

private void updateToFailed(doc, e) {
  doc.status = 'failed'
  doc.error = e.message
  documentDb.update(doc)
}

These kinds of methods are all around your typical code base. Consequently, as one’s code base grows it tends to get harder to reason about the state of the data after you’ve passed it to one of these methods.

Consider the following scenario:

def newDocs = [
  new Doc(title: 'Groovy', status: 'new'),
  new Doc(title: 'Ruby', status: 'new')
]

feedHandler.handle(newDocs)

println "My new docs: " + newDocs
// My new docs: 
// [Doc(title: Groovy, status: processed),
//  Doc(title: Ruby, status: processed)]
// WHAT? My new documents aren't that 'new' anymore

Some culprit has been mangling the status of my documents; first they’re “new” and a second later they aren’t; that’s NOT ok! It must be that darn FeedHandler. Who authored that thing? Why is it touching my data?!

Consider another scenario, where there’s more than one player handling your business.

def favoriteDocs = [
  new Doc(title: 'Haskell'),
  new Doc(title: 'OCaml'),
  new Doc(title: 'Scala')
]

archiver.backup(favoriteDocs)

feedHandler.handle(favoriteDocs)

mangleService.update(favoriteDocs)

userDao.merge(favoriteDocs, true)

println "My favorites: " + favoriteDocs
// My favorites: []
// WHAT? Empty collection? Where are my favorites????

We start with a collection of items, and 4 methods later we find that our data is gone.

In a world where everyone can mutate anything, it’s hard to reason about any state at any given time.

It’s not even “global state” per se – a collection passed into a method can be cleared and variables can be changed by anyone who gets a hold of (a reference to) your data.

Prefer Immutability

So what is it? An object is immutable if it does not change its state after it has been instantiated.

Seems reasonable, right?

ImmutabilityImage credit: Coping With and Adapting to Constant Change

There’s a ton of resources out there, about how to go about this in your particular language. Java, for instance, does not favor immutability by default; I have to do some work.

If there’s a 3rd party which is making problems and changing data along the way (such as clearing my collection) one can quickly flush out the troublemaker by passing my collection in a unmodifiable wrapper e.g.

def data = [
  ...
]

// somewhere inside 3rd-party code
data.clear()

// back in my code:
// data is empty *snif*

Preventing trouble:

def data = Collections
                 .unmodifiableCollection([])

// somewhere inside 3rd-party code
data.clear() // HAHAA, throws UnsupportedOperationException

Inside your own code base we can prevent unintended side effect (such as my data being changed somewhere) by minimizing mutable data structures.

In most FP languages like Haskell, OCaml and Scala the language itself promotes immutability by default. While not really a FP language, writing immutable JavaScript using ES6 also tends to become good practice.

Read-only first

Using the principles we’ve learned so far, and drive to prevent unintended side effects, we want to make sure our Doc class can not be changed by anything after instantiating it – not even our updateToProcessed/updateToFailed methods.

This is our current class:

class Doc {
  String title, type, apiId, status, error
}

Instead of doing all the manual labor of making a Java class immutable, Groovy comes to the rescue with the Immutable-annotation.

When put on the class, the Groovy compiler puts some enhancements in place, so NO ONE can update its state anymore after creation.

@Immutable
class Doc {
  String title, type, apiId, status, error
}

The object becomes effectively “read-only” — and any attempt to update a property will result in the aptly-named ReadOnlyPropertyException

private void updateToProcessed(doc, resource) {
  doc.apiId = resource.id // BOOM! 
  // throws groovy.lang.ReadOnlyPropertyException: 
  //  Cannot set readonly property: apiId
  ...
}

private void updateToFailed(doc, e) {
  doc.status = 'failed' // BOOM! 
  // throws groovy.lang.ReadOnlyPropertyException: 
  //  Cannot set readonly property: status
  ...
}

But wait, doesn’t this mean that the updateToProcessed/updateToFailed methods will actually fail updating a document’s status to “processed” or “failed”?

Jip, that’s what immutability brings us. How to repair the logic?

Copy second

The Haskell guide on “Immutable data” gives us advice on how to proceed:

Purely functional programs typically operate on immutable data. Instead of altering existing values, altered copies are created and the original is preserved. Since the unchanged parts of the structure cannot be modified, they can often be shared between the old and new copies, which saves memory.

Answer: we clone it!

We do not have to update the original data, we should make a copy of it — the original is not ours and should be left untouched. Our Immutable-annotation supports this with a parameter, called copyWith.

@Immutable(copyWith = true)
class Doc {
  String title, type, apiId, status, error
}

Consequently, we’ll change our methods to make a copy of the original with the altered status (and api id and error message) — and return this copy.

(The last statement in a Groovy method is always returned, doesn’t need an explicit return keyword)

private Doc setToProcessed(doc, resource) {
  doc.copyWith(
    status: 'processed',
    apiId: resource.id
  )
}

private Doc setToFailed(doc, e) {
  doc.copyWith(
    status: 'failed',
    error: e.message
  )
}

The database logic has also been moved up one level, taking the returned copy to store it.

We’ve gained control of our state!

This is it for now

If you, as a Java programmer, worry about the performance implications of excessive object instantiation, there’s a nice reassuring post here.

For reference, here’s the full version of the refactored code.

class FeedHandler {

  Webservice webservice
  DocumentDb documentDb

  void handle(List<Doc> changes) {

    changes
      .findAll { doc -> isImportant(doc) }
      .each { doc ->
        createResource(doc)
        .thenAccept { resource ->
          documentDb.update(
            setToProcessed(doc, resource)
          )
        }
        .exceptionally { e ->
          documentDb.update(setToFailed(doc, e))
        }
      }
  }

  private CompletableFuture<Resource> createResource(doc) {
    webservice.create(doc)
  }

  private boolean isImportant(doc) {
    doc.type == 'important'
  }

  private Doc setToProcessed(doc, resource) {
    doc.copyWith(
      status: 'processed',
      apiId: resource.id
    )
  }

  private Doc setToFailed(doc, e) {
    doc.copyWith(
      status: 'failed',
      error: e.message
    )
  }

}
Published on Java Code Geeks with permission by Ted Vinke, partner at our JCG program. See the original article here: Functional Java by Example | Part 4 – Prefer Immutability

Opinions expressed by Java Code Geeks contributors are their own.

Ted Vinke

Ted is a Java software engineer with a passion for Web development and JVM languages and works for First8, a Java Web development company in the Netherlands.
Subscribe
Notify of
guest

This site uses Akismet to reduce spam. Learn how your comment data is processed.

2 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
enric
enric
6 years ago

cool series of FP articles, when is expected to have all of them published?

6 years ago

Awesome series, please continue!

Back to top button