Java SE 8 library design

Java SE 8 Library Design
Using the new features well
Stephen Colebourne
Engineering Lead, OpenGamma
October 2016

Stephen Colebourne
● Java Champion, regular conference speaker
● Best known for date & time - Joda-Time and JSR-310
● More Joda projects - http://www.joda.org
● Major contributions in Apache Commons
● Blog - http://blog.joda.org
● Worked at OpenGamma for 6 years

Strata, from OpenGamma
● Open Source market risk library
● Valuation and risk calcs for finance
○ interest rate swap, FRA, CDS
● Great example of Java SE 8 coding style
http://strata.opengamma.io/

Introduction
● Java SE 8 is a major update to Java
● Major new features
○ Lambdas
○ Streams
○ Methods on interfaces
○ Date and Time

Introduction
● Essential to rethink how you code
● Reconsider coding conventions
● Appreciate new design options

Agenda
● Lambdas
● Streams
● Design with Lambdas
● Abstraction
● Immutability
● Interfaces
● Optional
● Odds and Ends

Lambdas
● Block of code
○ like an anonymous inner class
● Always assigned to a Functional Interface
○ an interface with one abstract method
○ Runnable, Callable, Comparator
● Uses target typing
○ context determines type of the lambda

// Java 7
List<Person> people = loadPeople();
Collections.sort(people, new Comparator<Person>() {
@Override
public int compare(Person p1, Person p2) {
return p1.name.compareTo(p2.name);
}
});
Lambdas
public interface Comparator<T> {
int compare(T obj1, T obj2);
}
// Java 7
@Override
}
});

// Java 7
List<Person> people = loadPeople();
@Override
}
});
Lambdas
public interface Comparator<T> {
int compare(T obj1, T obj2);
}
// Java 8
people.sort((p1, p2) -> p1.name.compareTo(p2.name));

Top tips for Lambdas
● Use lambdas wherever appropriate
● Mostly, they just work
● Sometimes, the compiler needs a hint
○ use local variable
○ add the type to the lambda parameter

Lambdas
Lambdas affect code
but do they affect design?

Streams
● Many loops have a similar "shape"
● Repetitive design patterns
● Stream library provides a way to abstract this
● Lambdas used to pass the interesting bits

Streams
List<Trade> trades = loadTrades();
List<Money> valued = new ArrayList<>();
for (Trade t : trades) {
if (t.isActive()) {
Money pv = t.presentValue();
valued.add(pv);
}
}

Streams
List<Money> valued =
trades.stream()
.filter(t -> t.isActive())
.map(t -> t.presentValue())
.collect(Collectors.toList());

Streams
● New stream() method on Collection
● Sequence of operations on underlying data
● Logic passed in using a lambda
○ filter() to retain/remove
○ map() to change
○ reduce() to summarise
○ sorted() to sort using a comparator

Streams
trades.stream()
.map(t -> t.presentValue())

Streams
trades.stream()
.filter(new Predicate<Trade>() {
public boolean test(Trade t) {
return t.isActive();
}
})
.map(new Function<Trade, Money>() {
public Money apply(Trade t) {
return t.presentValue();
}
})

Streams
Stream API not practical
without lambdas

Exceptions in Streams
● For-each loop is a language feature
● Streams are implemented using regular methods
● Big difference in stack traces

java.lang.IllegalArgumentException: Oops
at com.opengamma.strata.calc.DefaultCalculationRunner.lambda$2(DefaultCalculationRunner.java:98)
at java.util.stream.ReferencePipeline$11$1.accept(ReferencePipeline.java:372)
at java.util.Iterator.forEachRemaining(Iterator.java:116)
at java.util.Spliterators$IteratorSpliterator.forEachRemaining(Spliterators.java:1801)
at java.util.stream.AbstractPipeline.copyInto(AbstractPipeline.java:481)
at java.util.stream.AbstractPipeline.wrapAndCopyInto(AbstractPipeline.java:471)
at java.util.stream.ReduceOps$ReduceOp.evaluateSequential(ReduceOps.java:708)
at java.util.stream.AbstractPipeline.evaluate(AbstractPipeline.java:234)
at java.util.stream.ReferencePipeline.collect(ReferencePipeline.java:499)
at com.opengamma.strata.calc.DefaultCalculationRunner.calculate(DefaultCalculationRunner.java:100)
at com.opengamma.strata.calc.DefaultCalculationRunner.lambda$0(DefaultCalculationRunner.java:86)
at java.util.Iterator.forEachRemaining(Iterator.java:116)
at java.util.Spliterators$IteratorSpliterator.forEachRemaining(Spliterators.java:1801)
at java.util.stream.AbstractPipeline.copyInto(AbstractPipeline.java:481)
at java.util.stream.AbstractPipeline.wrapAndCopyInto(AbstractPipeline.java:471)
at java.util.stream.ReduceOps$ReduceOp.evaluateSequential(ReduceOps.java:708)
at java.util.stream.AbstractPipeline.evaluate(AbstractPipeline.java:234)
at java.util.stream.ReferencePipeline.collect(ReferencePipeline.java:499)
at com.opengamma.strata.calc.DefaultCalculationRunnerTest.calculate(DefaultCalculationRunnerTest.java:49)
Stack trace of
inner stream
Stack trace of
outer stream

java.lang.IllegalArgumentException: Oops
at com.opengamma.strata.calc.DefaultCalculationRunnerTest.calculate(DefaultCalculationRunnerTest.java:49)
Stack trace of
for-each loop

Top tips for streams
● Stream not always more readable than loop
● Stream exceptions can be much worse
● My advice:
○ use streams for small, localized, pieces of logic
○ be cautious using streams for large scale logic
● Strata uses for-each loops at top level
○ solely for shorter stack traces

Design with Lambdas
● Lambda is converted to a functional interface
● Normal interface with one abstract method
● Java SE 8 adds many new functional interfaces
○ Function<T, R>
○ Predicate<T>
○ Supplier<T>
○ Consumer<T>
○ see java.util.function package
● Primitive versions only for long, int, double

Functional interfaces
● Learn the standard functional interfaces
● Only create new ones if adding additional value
○ lots of parameters
○ mix of primitive and object parameters
○ feature really needs a good name or Javadoc

Functional interface example
// API functional interface
@FunctionalInterface
public interface Perturbation {
public abstract double perturb(int index, double value);
}
// API method that can be used by a lambda
public Curve perturbed(Perturbation perturbation) { … }
// caller code
curve = curve.perturbed((i, v) -> v + 1e-4);

Functional interface example
// API functional interface
@FunctionalInterface
public interface Perturbation {
public abstract double perturb(int index, double value);
}
public Curve perturbed(Perturbation perturbation) { … }
// caller code
curve = curve.perturbed((i, v) -> v + 1e-4);
Name has meaning
Method signature is complex

Time-series example
● A time-series stores changes to a value over time
● Date-based one like Map<LocalDate, Double>
● What if you want to change the values?

Time-series example
public LocalDateDoubleTimeSeries
mapValues(DoubleUnaryOperator mapper){ … }
// caller code
ts = ts.mapValues(v -> v * 2);

Time-series example
// caller code
Multiplication - no need for
multipliedBy(double)
on the API

Time-series example
// caller code
ts = ts.mapValues(v -> v / 4);
Multiplication - no need for
multipliedBy(double)
on the API
Division - no need for
dividedBy(double)
on the API

Design with Lambdas
Method taking a lambda
can be more flexible design

Abstraction
● Two or more classes with the same methods
● Abstract using an interface?
● Lambdas provide an alternative
● Consider an example with static methods
○ no way to abstract that with interfaces...

Abstraction
// standard API producing results
public static Money pv(FraTrade trade, Market md) { … }
public static Sens pv01(FraTrade trade, Market md) { … }
public static Double par(FraTrade trade, Market md) { … }

Abstraction
// standard API producing results
public static Money pv(FraTrade trade, Market md) { … }
public static Sens pv01(FraTrade trade, Market md) { … }
public static Double par(FraTrade trade, Market md) { … }
// functional interface matching all three methods
interface Calc<T> {
public abstract T invoke(FraTrade trade, Market market);
}

Abstraction
// create abstraction to access method by "measure" key
Map<Measure, Calc> CALCS = ImmutableMap.builder()
.put(Measures.PRESENT_VALUE, FraCalcs::pv)
.put(Measures.PV01, FraCalcs::pv01)
.put(Measures.PAR_RATE, FraCalcs:: par)
.build();
// can now invoke using measure
return CALCS.get(measure).invoke(trade, market);

Abstraction
● Class being abstracted was not changed
● Provides way to abstract over code you do not own
● Less need for reflection

Design with Lambdas
Lambdas can abstract
in dynamic/flexible ways

Multi-threaded
● JDK provides many tools for concurrency
● Parallel streams makes it even easier
● But parallel code is not simple to get right

Thread problems
● What if trade modified by some other piece of code?
● Check-then-act bug
List<Money> valued =
trades.stream()
.map(t -> presentValue(t))
Check
then Act

Immutable
● Threading bugs due to shared mutable state
● One solution is to use immutable beans
● No possibility of check-then-act type bug

Immutable beans
● Class should be final
○ no subclasses
● Fields must be final
○ needed for Java Memory Model
● Field types should be immutable
○ eg. don't use java.util.Date
● Factory methods and Builders instead of constructors

Immutable beans
● IDEs help you write mutable beans
● Need better tooling for immutable beans
○ AutoValue
○ Immutables.org
○ Joda-Beans
● Strata uses Joda-Beans
http://www.joda.org/joda-beans

Joda-Beans
@BeanDefinition
public final TradeInfo implements ImmutableBean {
/** The trade identifier. */
@PropertyDefinition(validate = "notNull")
private final StandardId tradeId;
/** The trade date. */
@PropertyDefinition(validate = "notNull")
private final LocalDate tradeDate;
}

Joda-Beans
● Source code generated for
○ getters
○ builder
○ equals/hashCode/toString
○ properties - like C#
● Can add your own code to the class and still regenerate
● Built in XML, JSON and Binary serialization

Immutable beans
● Most systems better using immutability everywhere
● Java SE 8 parallelStream() pushes at this
● Threading issues mostly eliminated
● No class hierarchies, use interfaces

Use interfaces
Trade
ProductTrade ResolvableTrade
SwapTrade FraTrade
Resolved
SwapTrade
Resolved
FraTrade

Use interfaces
● Concrete classes with no hierarchies
● Interfaces provide the hierarchy
● Methods on interfaces make this practical
● All implementations of interface should be immutable
○ "Implementations must be immutable and thread-safe beans."
● Strata uses immutable beans everywhere

Immutability
It is time to move on
from mutable data objects

Interfaces
● Two changes to interfaces
● Default methods
○ normal method, but on an interface
○ cannot default equals/hashCode/toString
● Static methods
○ normal static method, but on an interface

Coding Style
● Use modifiers in interfaces
● Much clearer now there are different types of method
● Prepares for private methods in Java SE 9
public interface Foo {
public static of(String id) { … }
public abstract isEmpty();
public default isNotEmpty() { … }
}

Interfaces
● Methods on interfaces changes design
● Interfaces are part of macro-design
○ lambdas and streams affect micro-design
● Strata uses default and static methods liberally

Holiday Calendar
● Strata interface to specify which days are holidays
public interface HolidayCalendar {
// a normal abstract interface method
public abstract isHoliday(LocalDate date);
…
}

Holiday Calendar
● Default methods make the interface more useful
// check if date is a business day
public default isBusinessDay(LocalDate date) {
return !isHoliday(date);
}
}

Holiday Calendar
● Default methods make the interface more useful
// find the next business day
public default next(LocalDate date) {
LocalDate nextDay = date.plusDays(1);
return isHoliday(nextDay) ? next(nextDay) : nextDay;
}
}

Holiday Calendar
● Static methods avoid HolidayCalendarFactory
// find holiday calendar by identifier such as DKCO
public static of(String id) {
// lookup calendar
}
}

Holiday Calendar
● Interface used just as would be expected
// find holiday calendar by identifier such as DKCO
HolidayCalendar cal = HolidayCalendar.of("DKCO");
// use calendar to select the trade start date
LocalDate startDate = cal.next(tradeDate);

Interfaces
● Interface now acts as abstract class
○ Only need abstract class if need abstracted state
○ but abstracted state is generally a bad idea
● Interface can now acts as a factory
○ Not suitable for all factory use cases*
* In Strata, holiday calendars are not really fixed at startup, but it made a good example for this talk!

Package-scoped implementation
● Can the interface be the only public API?
● Can the implementation class be package-scoped?
● Strata uses this pattern a lot

Interfaces
Consider package-scoped
factory and implementation

Optional and null
● New class Optional added to Java 8
● Opinions are polarized
○ some think it is the saviour of the universe
○ others think it is useless
● Used pragmatically, can be very useful

Optional and null
● Simple concept - two states
○ present, with a value - Optional.of(foo)
○ empty - Optional.empty()

Optional and null
● Standard code using null
// library, returns null if not found
public Foo getValue(String key) { … }
// application code must remember to check for null
Foo foo = getValue(key);
if (foo == null) {
foo = Foo.DEFAULT; // or throw an exception
}

Optional and null
● Standard code using Optional
// library, returns Optional if not found
public Optional<Foo> findValue(String key) { … }
// application code
Foo foo = findValue(key).orElse(Foo.DEFAULT);
// or
Foo foo = findValue(key).orElseThrow( … );

Optional and null
● Important that a variable of type Optional is never null
● Prefer methods like map() and orElse()
● Minimise use of isPresent()

Optional
● Strata often uses set of 3-methods
public abstract Optional<T> findValue(DataId<T> id);
public default boolean containsValue(DataId<T> id) {
return findValue(id).isPresent();
}
public default T getValue(DataId<T> id) {
return findValue(id).orElseThrow(
() -> new MarketDataException());
}

Optional
● Optional is a class
● Some memory/performance cost to using it
● Not serializable
● Not ideal to be an instance variable
● JDK authors added it for return types
● Use in parameters often annoying for callers
● Use as return type gets best value from concept
http://blog.joda.org/2015/08/java-se-8-optional-pragmatic-approach.html

Optional in Strata
● Strata has no exposed nulls
● No part of the API will return null
● Optional used when something is optional
● Pragmatically, null is used within classes

Optional
Use Optional in a
pragmatic way

Java SE 8 version
● Use Java SE 8 update 40 or later
○ preferably use the latest available
● Earlier versions have annoying lambda/javac issues

Internal JDK packages
● Java SE 9 will remove access to some JDK packages
○ sun.*
○ com.sun.*
○ com.oracle.*
● Now is the time to prepare for this
○ Avoid sun.misc.Unsafe
○ Stick to the standard JDK API

Parameters
● Java SE 8 can reflect on parameter names
● Avoids need for additional libraries like paranamer
● Not enabled by default, must choose to include data

Checked exceptions
● Checked exceptions can be made to disappear
● Helper methods can convert to runtime exceptions
Unchecked.wrap(() -> {
// any code that might throw a checked exception
// converted to a runtime exception
});

Summary
● Lots of good stuff in Java SE 8
● Design and coding standards change
● Lots more potential to abstract, but don't over-use
● Methods on interfaces add a lot of power

Key Strata design features
● Immutable data objects, using Joda-Beans
● Static methods on interfaces, package-scope impls
● Make use of new abstractions
● Beware stack traces with streams
● Pragmatic use of Optional, null never returned from API

Work in Finance?
● Take a look at OpenGamma Strata
○ developed from the ground up in Java 8
○ lots of good Java 8 techniques and utilities
● High quality library for market risk
○ day counts, schedules, holidays, indices
○ models and pricing for swaps, FRAs, swaptions, FX, futures…
○ open source and stable release v1.1
http://strata.opengamma.io/

Java SE 8 library design

Recommended

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Java SE 8 library design (20)

Recently uploaded (20)

Java SE 8 library design