Skip to content
This repository has been archived by the owner on Sep 30, 2020. It is now read-only.

Latest commit

 

History

History
269 lines (201 loc) · 11.2 KB

README.md

File metadata and controls

269 lines (201 loc) · 11.2 KB
Raw

co.unruly.control

Build Status Release Version Javadocs

⚠️ This repo had been archived. Development continues at https://github.com/writeoncereadmany/control ⚠️

Control is a collection of functional control-flow primitives and utilities, built around a Result type.

Installation

Using Maven

Add this dependency to your pom.xml.

<dependency>
    <groupId>co.unruly</groupId>
    <artifactId>control</artifactId>
    <version>0.8.12</version>
</dependency>

Documentation

Examples

Result

What is a Result? It's a representation of the outcome of an operation which may have failed. Kind of like java.util.Optional.

Like Optional, it wraps a value, and like Optional, it doesn't allow you to use that value directly. After all, if it's representing a failed operation, there may not be a value. In order to extract a value, you have to define how to handle the failure cases.

However, an Optional is either present, and contains a value, or absent, conveying no information. A Result, however, is either a success, containing a value, or a failure, containing information about the failure.

Creating a Result from a value: Introducers

The simplest way to create a Result is simply to instantiate either a success or failure value, as appropriate. For example, the following code creates a successful Result, followed by a failed Result.

Result<Integer, String> firstResult = Result.success(42);
Result<Integer, String> secondResult = Result.failure("What is six times nine?");

There are also many common repeated patterns of operations which can fail. For example, you may have some code which creates Optionals, which you would like to convert to Results in order to better track failure causes.

For this, co.unruly.control.result.Introducers contains a selection of useful functions which yield Results:

public Result<Integer, String> asResult(Optional<Integer> maybeNumber) {
    return with(maybeNumber, fromOptional(() -> "No number found. :("));    
}

Composing operations on a Result: Transformers

Result is most useful when modelling an operation as a sequence of smaller operations, each using the output of the last step, and some of which can fail. The most common operations here are onSuccess() and attempt().

Invoking onSuccess(f) on a Result will apply f to the value in the Result if it's a success, yielding a new success. If the original Result was a failure, f will not be invoked and the original failing Result will be returned.

If you have a function f which can fail - i.e. a function which returns a Result - then invoking attempt(f) will apply f to the value in the Result if it's a success, yielding a new Result which may be either a success or failure. If the original Result was a failure, f will not be invoked and the original failing Result will be returned.

This allows us to chain together a sequence of calls to onSuccess and attempt to specify what to do on the happy path, and cascade any failure cases together to be handled once.

For example, if we want to write a bestselling novel, then there are various steps towards getting it published. We need to have an idea, secure an advance, write the manuscript, get it edited, get it published, and rocket up the bestseller charts.

Getting the idea, securing an advance and finishing the manuscript are definitely steps which can fail. However, given success in the other steps, editing and publishing are mechanical steps we can have confidence will succeed, and once those are complete we can then release the novel and (eventually) count the total sales.

If any of the steps fail, no novel is published, and we can look at the error message from whenever the process failed.

We could therefore model the process as follows:

public static String describeNovelSales(Author author, Publisher publisher, Editor editor, Retailer retailer) {
    return author.getIdea()
            .then(attempt(publisher::getAdvance))
            .then(attempt(author::writeNovel))
            .then(onSuccess(editor::editNovel))
            .then(onSuccess(publisher::publishNovel))
            .then(onSuccess(retailer::sellNovel))
            .then(onSuccess(sales -> format("%s sold %d copies", sales.novel, sales.copiesSold)))
            .then(collapse());
}

Whilst onSuccess() and attempt() are the most common ways to transform a Result into another Result, other use cases exist. A collection of such functions exists in co.unruly.control.result.Transformers.

Extracting a value from a Result: Resolvers

The simplest way to extract a value from a Result is to simply describe what to do with a failure value. For example, the following method takes a Result of either an Integer or a String describing the failure, and returns the wrapped integer if it was a success, or -1 if it was a failure:

public Integer extractValue(Result<Integer, String> count) {
    return count.then(ifFailed(x -> -1));
}

Sometimes, the success and failure types are the same. In that case, you can simply collapse both cases to a single value:

public static void main(String ...args) {
    printResult(Result.success("This was a triumph!"));
    printResult(Result.failure("The cake is a lie"));
}


public static String printResult(Result<String, String> value) {
    System.out.println(value.then(collapse()));
}

Which outputs:

> This was a triumph!
> The cake is a lie

ifFailed and collapse can both be found in co.unruly.control.result.Resolvers , along with a selection of other functions to convert Results into non-Result values.

Functional API

None of the operations described above are methods on the Result object. Instead, Result presents only two methods: Result.either() and Result.then().

Result.either()

Result.either takes two functions, one for the success case and one for the failure case. If the Result is a success, it applies the first function to its success value and returns the result. If it was a failure, instead it applies the second function.

public static void main(String ...args) {
    printResult(Result.success(99));
    printResult(Result.failure("bananas"));
}


public static String printResult(Result<String, String> value) {
    System.out.println(value.either(
        success -> success + " bottles of beer on the wall",
        failure -> "Yes sir, we have no " + failure    
    ));
}

Which outputs:

> 99 bottles of beer on the wall
> Yes sir, we have no bananas

Note that the argument types of the functions must match the corresponding success and failure types, and both functions must return the same type.

Result.then()

Result.then takes a function from a Result to a value, and then applies that function to the result. The following lines of code are (other than generics inference issues) equivalent:

ifFailed(x -> "Hello World").apply(result);
result.then(ifFailed(x -> "Hello World"));

By structuring the API like this, instead of having a fixed set of methods available, we can create more specialised functions for domain-specific use cases and mix them seamlessly with standard operations, and customise how we interact with Results.

The functions included in this library are all higher-order functions, returning instances of Function. This includes functions to create Results from non-Result values - both for consistency, and compatibility with idiomatic Stream usage. As non-Result values don't implement then(), and there are both readability and generics issues, we also include Piper.

Piper

This approach is slightly more cumbersome than it could be, as Java's ability to infer generic types is not particularly great. For example, the following code doesn't compile:

public Result<Integer, String> asResult(Optional<Integer> maybeNumber) {
    return fromOptional(() -> "No number found. :(").apply(maybeNumber);    
}

This fails to compile because the type of fromOptional is inferred to be Function<Optional<Object>, Result<Object, String>, because the type inference engine can't infer the success type based on subsequent operations (namely, the type passed to apply()).

Enter Piper. A Piper is simply a box for a value:

Piper<Integer> pipe = Piper.pipe(42);

You can then chain functions on the pipe:

Piper.pipe(42)              // yields a Pipe containing 42
     .then(x -> x + 10)     // yields a Pipe containing 52
     .then(x -> x * 2)      // yields a Pipe containing 104
     .resolve()             // returns 104

This allows us to rewrite our failing-to-compile example from above as:

public Result<Integer, String> asResult(Optional<Integer> maybeNumber) {
    return pipe(maybeNumber)                                // Pipe of Optional<Integer>
        .then(fromOptional(() -> "No number found. :("))    // Pipe of Result<Integer, String>
        .resolve();                                         // Result<Integer, String>               
}

For short examples like this, there's also a resolveWith() method:

public Result<Integer, String> asResult(Optional<Integer> maybeNumber) {
    return pipe(maybeNumber)                                
        .resolveWith(fromOptional(() -> "No number found. :("));
}

Piper.then() functions in the same way as Result.then(): it applies the provided function to its contents. This means we can build a functional pipeline which includes both regular values and Results:

Result<Integer, String> result = pipe("a=1234;")                            // Piper<String>
        .then(pattern::matcher)                                             // Piper<Matcher>            
        .then(ifIs(Matcher::find, m -> m.group(1)))                         // Piper<Result<String, Matcher>>
        .then(onFailure(__ -> "Could not find group to match"))             // Piper<Result<String, String>>
        .then(attempt(tryTo(Integer::parseInt, ex -> ex.getMessage())))     // Piper<Result<Integer, String>>
        .resolve();                                                         // Result<Integer, String>

Note that the use of Piper.then() is identical to the way we'd use Result.then(), except now we can directly chain from non-failable states into failable states.