Develop Reference

Builder pattern vs encapsulation of a data

How should I retain Encapsulation Principle of OOP, when using builder pattern? I mean fact that builder should provide abstraction layer between object and the code that uses it, so that it can be constructed part-by-part, requires either making setter for every parameter of an object that we would normally pass in the constructor. That again may be undesirable in some cases, as I don't want client to be able to modify value that I have to via builder. Example to picture what I mean is below:
public class Cat
{
private string _race;
private string _name;
public Cat()
{
_race = "common";
_name = string.Empty;
}
public void setRace(string race) { _race = race; }
public void setName(string name) { _name = name; }
}
public class CatBuilder
{
private Cat _objectUnderConstruction;
public CatBuilder() { _objectUnderConstruction = new Cat(); }
public CatBuilder WithName(string name)
{
_objectUnderConstruction.setName(name);
return this;
}
public CatBuilder OfRace(string race)
{
_objectUnderConstruction.setRace(race);
return this;
}
}
This is not production code, I wrote it now with presentation in mind, so do not get mad on how it is constructed.
In the example above there is need to set cat's race, as we need that information for the purpose of object filling, so we need to pass info into it. At the same time I don't want anyone to ever change race of my cat during its lifetime (e.g. it would change from egyptian to british in the middle of processing) Normally I would get rid of accessor method, but I need for the builder. This way, encapsulation of data is hurt (because straight get and set aren't encapsulating anything), and I want to avoid it.
This example is simple and I could pass parameter in constructor, but imagine bigger class, where there is a lot of such fields, what in this case? Should I pass some configuration object inside (which is almost like builder, but simpler, hence builder is pointless) or pass the builder itself to the constructor (which is weird, but what do I know)?
How I should do that?

If your builder is tightly-coupled with your class you can make Builder subclass of the object being constructed:
public class Cat
{
private string _race;
private string _name;
public Cat()
{
_race = "common";
_name = string.Empty;
}
private void setRace(string race) { _race = race; }
private void setName(string name) { _name = name; }
public class Builder
{
private Cat _objectUnderConstruction;
public CatBuilder() { _objectUnderConstruction = new Cat(); }
public CatBuilder WithName(string name)
{
_objectUnderConstruction.setName(name);
return this;
}
public CatBuilder OfRace(string race)
{
_objectUnderConstruction.setRace(race);
return this;
}
}
}
This way, you'll be able in Builder to access private fields and methods of Cat and use it like new Cat.Builder().OfRace("").OfName("").Build().

Builder Design Pattern with sub-classing and required parameters?

Recently I came into a situation where the builder pattern was very strong, but I had the need to subclass. I looked up some solutions and some suggested generics while others suggested normal subclassing. However, none of the examples I looked at had required fields in order to even begin building an object. I wrote a tiny example to illustrate where I'm getting stuck. At every turn I kept running into a wall of problems where things would return the wrong class, can't override static methods, returning super() returns the wrong data type, etc. I have a feeling there is no way out except excessive use of generics.
What is the correct way to go in this situation?
Tester
import person.Person;
import person.Student;
public class Tester
{
public static void main(String[] args)
{
Person p = Person.builder("Jake", 18).interest("Soccer").build();
// Student s = Student.builder(name, age) <-- It's weird that we still have access to pointless static method
// Student s = Student.builder("Johnny", 24, "Harvard", 3).address("199 Harvard Lane") <-- returns Person builder, not student
Student s = ((Student.Builder)Student.builder("Jack", 19, "NYU", 1).address("Dormitory")).build(); // really bad
}
}
Person Class
package person;
import java.util.ArrayList;
import java.util.List;
public class Person
{
// Required
protected String name;
protected int age;
// Optional
protected List<String> interests = new ArrayList<>();
protected String address = "";
protected Person(String name, int age)
{
this.name = name;
this.age = age;
}
public String getName() { return name; }
public int getAge() { return age; }
public List<String> getInterests() { return interests; }
public String getAddress() { return address; }
// person.person does not allow builder construction
// unless all required fields are provided
/* Problem: I have to repeat the constructor fields here, very annoying */
public static Builder builder(String name, int age)
{
Person p = new Person(name, age);
return new Builder(p);
}
public static class Builder
{
Person reference;
protected Builder(Person reference)
{
this.reference = reference;
}
public Builder address(String address)
{
reference.address = address;
return this;
}
public Builder interest(String interest)
{
reference.interests.add(interest);
return this;
}
public Person build()
{
return reference;
}
}
}
Student Class
package person;
import java.util.ArrayList;
import java.util.List;
public class Student extends Person
{
// Required
protected String school;
protected int year;
// Optional
protected List<String> subjects = new ArrayList<>();
// This looks good
public Student(final String name, final int age, final String school, final int year)
{
super(name, age);
this.school = school;
this.year = year;
}
public String getSchool() { return school; }
public int getYear() { return year; }
public List<String> getSubjects() { return subjects; }
/* Here's where my issues are:
* Override doesn't compile on static methods but how else can I describe that I want to
* override this functionality from the Person class?
*
* Extending 'Person' does not enforce that I need to provide 'name', 'age', etc like it would
* if this was a normal design pattern using the 'new' keyword. I have to manually drag fields
* from 'person' and place them here. This would get VERY messy with an additional class
*
* User can STILL call the Person builder on a Student object, which makes no sense. */
/*#Override*/ public static Builder builder(String name, int age, String school, int year)
{
Student s = new Student(name, age, school, year);
return new Builder(s);
}
public static class Builder extends Person.Builder
{
// Student reference; <--- this should not be needed since we already
// have a variable for this purpose from 'Person.Builder'
public Builder(final Student reference)
{
super(reference);
}
/* Things begins to get very messy here */
public Builder subject(String subject)
{
((Student)reference).subjects.add(subject);
// I guess I could replace the reference with a student one, but
// I feel like that infringes on calling super() builder since we do the work twice.
return this;
}
#Override public Student build()
{
// I can either cast here or
// rewrite 'return reference' every time.
// Seems to infringe a bit on subclassing.
return (Student)super.build();
}
}
}

What you write here :
Student s = ((Student.Builder)Student.builder("Jack", 19, "NYU", 1).address("Dormitory")).build(); // really bad
is indeed not very natural and you should not need to cast.
We expect rather something like :
Student s = Student.builder("Jack", 19, "NYU", 1).address("Dormitory")).build();
Besides all casts you did in the implementation of Student.Builder are also noise and statements that may fail at runtime :
/* Things begins to get very messy here */
public Builder subject(String subject) {
((Student)reference).subjects.add(subject);
return this;
}
#Override public Student build() {
return (Student)super.build();
}
Your main issue is the coupling between the Builder classes and the building methods.
A important thing to consider is that at compile time, the method binding (method selected by the compiler) is performed according to the declared type of the target of the invocation and the declared type of its arguments.
The instantiated type is considered only at runtime as the dynamic binding is applied: invoking the method bounded at compile time on the runtime object.
So this overriding defined in Student.Builder is not enough :
#Override public Student build() {
return (Student)super.build();
}
As you invoke :
Student.builder("Jack", 19, "NYU", 1).address("Dormitory").build();
At compile time, address("Dormitory") returns a variable typed as Person.Builder as the method is defined in Person.Builder :
public Builder address(String address){
reference.address = address;
return this;
}
and it not overriden in Student.Builder.
And at compile time, invoking build() on a variable declared as Person.Builder returns a object with as declared type a Person as the method is declared in Person.Builder as :
public Person build(){
return reference;
}
Of course at runtime, the returned object will be a Student as
Student.builder("Jack", 19, "NYU", 1) creates under the hood a Student and not a Person.
To avoid cast to Student.builder both from the implementation and the client side, favor composition over inheritancy :
public static class Builder {
Person.Builder personBuilder;
private Student reference;
public Builder(final Student reference) {
this.reference = reference;
personBuilder = new Person.Builder(reference);
}
public Builder subject(String subject) {
reference.subjects.add(subject);
return this;
}
// delegation to Person.Builder but return Student.Builder
public Builder interest(String interest) {
personBuilder.interest(interest);
return this;
}
// delegation to Person.Builder but return Student.Builder
public Builder address(String address) {
personBuilder.address(address);
return this;
}
public Student build() {
return (Student) personBuilder.build();
}
}
You can now write :
Student s = Student.builder("Jack", 19, "NYU", 1)
.address("Dormitory")
.build();
or even that :
Student s2 = Student.builder("Jack", 19, "NYU", 1)
.interest("Dance")
.address("Dormitory")
.build();
Composition introduces generally more code as inheritancy but it makes the code
both more robust and adaptable.
As a side note, your actual issue is enough close to another question I answered 1 month ago.
The question and its answers may interest you.

A few thoughts as background
Static methods are not so great,
they make unit testing more difficult.
It is fine to put the builder as a static, nested class, but if you are using a builder to construct a class you should make the constructor not-public.
I prefer to have the builder be a separate class in the same package and to make the constructor (of the class that is created by the builder) package access.
Limit the builder constructor parameters.
I'm not a fan of using a class hierarchy for builders.
The Person and Student classes each have a builder.
Some Code
public class PersonBuilder
{
private String address;
private int age;
private final List<String> interestList;
private String name;
public PersonBuilder()
{
interestList = new LinkedList<>();
}
public void addInterest(
final String newValue)
{
// StringUtils is an apache utility.
if (StringUtils.isNotBlank(newValue))
{
interestList.add(newValue);
}
return this;
}
public Person build()
{
// perform validation here.
// check for required values: age and name.
// send all parameters in the constructor. it's not public, so that is fine.
return new Person(address, age, interestList, name);
}
public PersonBuilder setAddress(
final String newValue)
{
address = newValue;
return this;
}
public PersonBuilder setAge(
final int newValue)
{
age = newValue;
return this;
}
public PersonBuilder setInterestList(
final List<String> newValue)
{
interestList.clear();
if (CollectionUtils.isNotEmpty(newValue))
{
interestList.addAll(newValue);
}
return this;
}
public PersonBuilder setName(
final String newValue)
{
name = newValue;
return this;
}
}
public class Person
{
private Person()
{
}
Person(
final String addressValue,
final int ageValue,
final List<String> interestListValue,
final String name)
{
// set stuff.
// handle null for optional parameters.
}
// create gets or the fields, but do not create sets. Only the builder can set values in the class.
}

Proper Constructors to Run all Methods in Java

I have this class and need to know which constructor is needed to create an object that may immediately use all its methods without error
public class Robot {
private boolean fuelEmpty = true;
private int roboID;
private String greeting;
private String securityProtocol;
//insert robot constructor here
public void destroyAllHumans(){
while (fuelEmpty == false) {
//robot begins to destroy all humans
}
}
public int getRoboID(){
return roboID;
}
public void greet(){
System.out.println(greeting);
}
public void setSecurityProtocol(String proto){
securityProtocol = proto;
}
}
For example should look like this:
public Robot(int id, String greet) {
roboID = id;
greeting = greet;
}
or this:
public Robot(int id, String greet) {
roboID = id;
greeting = greet;
fuelEmpty = false;
}
or:
public Robot(boolean full, int id, String greet, String proto) {
roboID = id;
greeting = greet;
fuelEmpty = full;
securityProtocol = proto;
}
Which of these (or something else different) is needed so that all the other methods can run without an error?

You can overload the constructor as much as you need, the important thing is
the object gets properly instantiated after you create a new one...
a way can be:
public Robot() {
this(false, 0, "", "");
}
public Robot(int id) {
this(false, id, "", "");
}
public Robot(boolean fuelEmpty, int roboID, String greeting, String securityProtocol) {
this.fuelEmpty = fuelEmpty;
this.roboID = roboID;
this.greeting = greeting;
this.securityProtocol = securityProtocol;
}
so look how all other constructors will at the end call internally the
public Robot(boolean fuelEmpty, int roboID, String greeting, String securityProtocol)
that will give you the waranty that no matter which constructor is invoked, the Robot is fully created and can invoke all those methods without crashing

The solution works like this:
you look at each of your methods
you check which fields each method is using
you check more closely, if the method breaks when that field has its default value (like null for Objects, or false for booleans)
When you do that for all methods, you get a list of those fields that you need to initialize somehow. Then you could go forward and define a corresponding constructor.
But of course, that is the wrong approach.
The real answer goes like this: you don't put fields into a class because you can. You add them because they are required so that this class can implement the requirements (responsibilities) that you want it to implement. Meaning: you focus on the methods that your class should provide. Then you clarify which fields you need in order to implement these methods.
In other words: you have exactly those fields in your class that your class needs. If you have fields in there that go unused - then you get rid of them.

How to enable enum inheritance

I'm writing a library, which has a predefined set of values for an enum.
Let say, my enum looks as below.
public enum EnumClass {
FIRST("first"),
SECOND("second"),
THIRD("third");
private String httpMethodType;
}
Now the client, who is using this library may need to add few more values. Let say, the client needs to add CUSTOM_FIRST and CUSTOM_SECOND. This is not overwriting any existing values, but makes the enum having 5 values.
After this, I should be able to use something like <? extends EnumClass> to have 5 constant possibilities.
What would be the best approach to achieve this?

You cannot have an enum extend another enum, and you cannot "add" values to an existing enum through inheritance.
However, enums can implement interfaces.
What I would do is have the original enum implement a marker interface (i.e. no method declarations), then your client could create their own enum implementing the same interface.
Then your enum values would be referred to by their common interface.
In order to strenghten the requirements, you could have your interface declare relevant methods, e.g. in your case, something in the lines of public String getHTTPMethodType();.
That would force implementing enums to provide an implementation for that method.
This setting coupled with adequate API documentation should help adding functionality in a relatively controlled way.
Self-contained example (don't mind the lazy names here)
package test;
import java.util.ArrayList;
import java.util.List;
public class Main {
public static void main(String[] args) {
List<HTTPMethodConvertible> blah = new ArrayList<>();
blah.add(LibraryEnum.FIRST);
blah.add(ClientEnum.BLABLABLA);
for (HTTPMethodConvertible element: blah) {
System.out.println(element.getHTTPMethodType());
}
}
static interface HTTPMethodConvertible {
public String getHTTPMethodType();
}
static enum LibraryEnum implements HTTPMethodConvertible {
FIRST("first"),
SECOND("second"),
THIRD("third");
String httpMethodType;
LibraryEnum(String s) {
httpMethodType = s;
}
public String getHTTPMethodType() {
return httpMethodType;
}
}
static enum ClientEnum implements HTTPMethodConvertible {
FOO("GET"),BAR("PUT"),BLAH("OPTIONS"),MEH("DELETE"),BLABLABLA("POST");
String httpMethodType;
ClientEnum(String s){
httpMethodType = s;
}
public String getHTTPMethodType() {
return httpMethodType;
}
}
}
Output
first
POST

Enums are not extensible. To solve your problem simply
turn the enum in a class
create constants for the predefined types
if you want a replacement for Enum.valueOf: track all instances of the class in a static map
For example:
public class MyType {
private static final HashMap<String,MyType> map = new HashMap<>();
private String name;
private String httpMethodType;
// replacement for Enum.valueOf
public static MyType valueOf(String name) {
return map.get(name);
}
public MyType(String name, String httpMethodType) {
this.name = name;
this.httpMethodType = httpMethodType;
map.put(name, this);
}
// accessors
public String name() { return name; }
public String httpMethodType() { return httpMethodType; }
// predefined constants
public static final MyType FIRST = new MyType("FIRST", "first");
public static final MyType SECOND = new MyType("SECOND", "second");
...
}

Think about Enum like a final class with static final instances of itself. Of course you cannot extend final class, but you can use non-final class with static final instances in your library. You can see example of this kind of definition in JDK. Class java.util.logging.Level can be extended with class containing additional set of logging levels.
If you accept this way of implementation, your library code example can be like:
public class EnumClass {
public static final EnumClass FIRST = new EnumClass("first");
public static final EnumClass SECOND = new EnumClass("second");
public static final EnumClass THIRD = new EnumClass("third");
private String httpMethodType;
protected EnumClass(String name){
this.httpMethodType = name;
}
}
Client application can extend list of static members with inheritance:
public final class ClientEnum extends EnumClass{
public static final ClientEnum CUSTOM_FIRST = new ClientEnum("custom_first");
public static final ClientEnum CUSTOM_SECOND = new ClientEnum("custom_second");
private ClientEnum(String name){
super(name);
}
}
I think that this solution is close to what you have asked, because all static instances are visible from client class, and all of them will satisfy your generic wildcard.

We Fixed enum inheritance issue this way, hope it helps
Our App has few classes and each has few child views(nested views), in order to be able to navigate between childViews and save the currentChildview we saved them as enum inside each Class.
but we had to copy paste, some common functionality like next, previous and etc inside each enum.
To avoid that we needed a BaseEnum, we used interface as our base enum:
public interface IBaseEnum {
IBaseEnum[] getList();
int getIndex();
class Utils{
public IBaseEnum next(IBaseEnum enumItem, boolean isCycling){
int index = enumItem.getIndex();
IBaseEnum[] list = enumItem.getList();
if (index + 1 < list.length) {
return list[index + 1];
} else if(isCycling)
return list[0];
else
return null;
}
public IBaseEnum previous(IBaseEnum enumItem, boolean isCycling) {
int index = enumItem.getIndex();
IBaseEnum[] list = enumItem.getList();
IBaseEnum previous;
if (index - 1 >= 0) {
previous = list[index - 1];
}
else {
if (isCycling)
previous = list[list.length - 1];
else
previous = null;
}
return previous;
}
}
}
and this is how we used it
enum ColorEnum implements IBaseEnum {
RED,
YELLOW,
BLUE;
#Override
public IBaseEnum[] getList() {
return values();
}
#Override
public int getIndex() {
return ordinal();
}
public ColorEnum getNext(){
return (ColorEnum) new Utils().next(this,false);
}
public ColorEnum getPrevious(){
return (ColorEnum) new Utils().previous(this,false);
}
}
you could add getNext /getPrevious to the interface too

#wero's answer is very good but has some problems:
the new MyType("FIRST", "first"); will be called before map = new HashMap<>();. in other words, the map will be null when map.add() is called. unfortunately, the occurring error will be NoClassDefFound and it doesn't help to find the problem. check this:
public class Subject {
// predefined constants
public static final Subject FIRST;
public static final Subject SECOND;
private static final HashMap<String, Subject> map;
static {
map = new HashMap<>();
FIRST = new Subject("FIRST");
SECOND = new Subject("SECOND");
}
private final String name;
public Subject(String name) {
this.name = name;
map.put(name, this);
}
// replacement for Enum.valueOf
public static Subject valueOf(String name) {
return map.get(name);
}
// accessors
public String name() {
return name;
}

Capturing method parameter in jMock to pass to a stubbed implementation

I wish to achieve the following behavior.
My class under test has a dependency on some other class, I wish to mock this dependency with jMock. Most of the methods would return some standard values, but there is one method, where I wish to make a call to a stubbed implementation, I know I can call this method from the will(...) but I want the method to be called by the exact same parameters that were passed to the mocked method.
Test
#Test
public void MyTest(){
Mockery context = new Mockery() {
{
setImposteriser(ClassImposteriser.INSTANCE);
}
};
IDependency mockObject = context.mock(IDependency.class);
Expectations exp = new Expectations() {
{
allowing(mockObject).methodToInvoke(????);
will(stubMethodToBeInvokedInstead(????));
}
};
}
Interface
public interface IDependency {
public int methodToInvoke(int arg);
}
Method to be called instead
public int stubMethodToBeInvokedInstead(int arg){
return arg;
}
So how do I capture the parameter that were passed to the method being mocked, so I could pass them to the stubbed method instead?
EDIT
Just to give another example, let's say I wish to mock the INameSource dependency in the following (C#) code, to test the class Speaker
public class Speaker
{
private readonly string firstName;
private readonly string surname;
private INameSource nameSource ;
public Speaker(string firstName, string surname, INameSource nameSource)
{
this.firstName = firstName;
this.surname = surname;
this.nameSource = nameSource;
}
public string Introduce()
{
string name = nameSource.CreateName(firstName, surname);
return string.Format("Hi, my name is {0}", name);
}
}
public interface INameSource
{
string CreateName(string firstName, string surname);
}
This is how it can be done in Rhino Mocks for C# I understand it can't be as easy as this since delegates are missing in Java

The solution from Duncan works well, but there is even a simpler solution without resort to a custom matcher. Just use the Invocation argument that is passed to the CustomActions invoke method. At this argument you can call the getParameter(long i) method that gives you the value from the call.
So instead of this
return matcher.getLastValue();
use this
return (Integer) invocation.getParameter(0);
Now you don't need the StoringMatcher anymore: Duncans example looks now like this
#RunWith(JMock.class)
public class Example {
private Mockery context = new JUnit4Mockery();
#Test
public void Test() {
final IDependency mockObject = context.mock(IDependency.class);
context.checking(new Expectations() {
{
// No custom matcher required here
allowing(mockObject).methodToInvoke(with(any(Integer.class)));
// The action will return the first argument of the method invocation.
will(new CustomAction("returns first arg") {
#Override
public Object invoke(Invocation invocation) throws Throwable {
return (Integer) invocation.getParameter(0);
}
});
}
});
Integer test1 = 1;
Integer test2 = 1;
// Confirm the object passed to the mocked method is returned
Assert.assertEquals((Object) test1, mockObject.methodToInvoke(test1));
Assert.assertEquals((Object) test2, mockObject.methodToInvoke(test2));
}
public interface IDependency {
public int methodToInvoke(int arg);
}

Like Augusto, I'm not convinced this is a good idea in general. However, I couldn't resist having a little play. I created a custom matcher and a custom action which store and return the argument supplied.
Note: this is far from production-ready code; I just had some fun. Here's a self-contained unit test which proves the solution:
#RunWith(JMock.class)
public class Example {
private Mockery context = new JUnit4Mockery();
#Test
public void Test() {
final StoringMatcher matcher = new StoringMatcher();
final IDependency mockObject = context.mock(IDependency.class);
context.checking(new Expectations() {
{
// The matcher will accept any Integer and store it
allowing(mockObject).methodToInvoke(with(matcher));
// The action will pop the last object used and return it.
will(new CustomAction("returns previous arg") {
#Override
public Object invoke(Invocation invocation) throws Throwable {
return matcher.getLastValue();
}
});
}
});
Integer test1 = 1;
Integer test2 = 1;
// Confirm the object passed to the mocked method is returned
Assert.assertEquals((Object) test1, mockObject.methodToInvoke(test1));
Assert.assertEquals((Object) test2, mockObject.methodToInvoke(test2));
}
public interface IDependency {
public int methodToInvoke(int arg);
}
private static class StoringMatcher extends BaseMatcher<Integer> {
private final List<Integer> objects = new ArrayList<Integer>();
#Override
public boolean matches(Object item) {
if (item instanceof Integer) {
objects.add((Integer) item);
return true;
}
return false;
}
#Override
public void describeTo(Description description) {
description.appendText("any integer");
}
public Integer getLastValue() {
return objects.remove(0);
}
}
}
A Better Plan
Now that you've provided a concrete example, I can show you how to test this in Java without resorting to my JMock hackery above.
Firstly, some Java versions of what you posted:
public class Speaker {
private final String firstName;
private final String surname;
private final NameSource nameSource;
public Speaker(String firstName, String surname, NameSource nameSource) {
this.firstName = firstName;
this.surname = surname;
this.nameSource = nameSource;
}
public String introduce() {
String name = nameSource.createName(firstName, surname);
return String.format("Hi, my name is %s", name);
}
}
public interface NameSource {
String createName(String firstName, String surname);
}
public class Formal implements NameSource {
#Override
public String createName(String firstName, String surname) {
return String.format("%s %s", firstName, surname);
}
}
Then, a test which exercises all the useful features of the classes, without resorting to what you were originally asking for.
#RunWith(JMock.class)
public class ExampleTest {
private Mockery context = new JUnit4Mockery();
#Test
public void testFormalName() {
// I would separately test implementations of NameSource
Assert.assertEquals("Joe Bloggs", new Formal().createName("Joe", "Bloggs"));
}
#Test
public void testSpeaker() {
// I would then test only the important features of Speaker, namely
// that it passes the right values to the NameSource and uses the
// response correctly
final NameSource nameSource = context.mock(NameSource.class);
final String firstName = "Foo";
final String lastName = "Bar";
final String response = "Blah";
context.checking(new Expectations() {
{
// We expect one invocation with the correct params
oneOf(nameSource).createName(firstName, lastName);
// We don't care what it returns, we just need to know it
will(returnValue(response));
}
});
Assert.assertEquals(String.format("Hi, my name is %s", response),
new Speaker(firstName, lastName, nameSource).introduce());
}
}

JMock doesn't support your use case (or any other mocking framework I know of in java).
There's a little voice in my head that says that what you're trying to do is not ideal and that your unit test might be to complicated (maybe it's testing too much code/logic?). One of the problems I see, is that you don't know which values those mocks need to return and you're plugging something else, which might make each run irreproducible.