Skip to content

henrik42/deeto

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

70 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Deeto

A Java dynamic proxy factory for interface-typed data transfer objects

Quick start

See Usage below for the details.

package deeto_user;

import deeto.Deeto;

public interface HelloWorld {

    HelloWorld name(String x);

    static HelloWorld newInstance() {
        return Deeto.factory().newInstance(HelloWorld.class);
    }

    public static void main(String[] args) {
        System.out.println("Hello " + newInstance().name("World"));
    }
}

Gives you:

Hello {:type deeto_user.HelloWorld, :value {"Name" "World"}}

Long story

Deeto is a Clojure [1] library for Java developers. With Deeto you can define your data transfer object (DTO [2, 3]) types via interfaces in Java. You do not need to implement these interfaces. Instead you can ask Deeto to analyze (via reflection [4]) the interface class I and then give you a factory [5] Deeto.factory().newInstance(I) for I. On each invovation this factory will return a new (initializd; see below) instance i with the following properties:

  • i's class P is a Java dynamic proxy class [6] which implements I, Cloneable and Serializable [11].

  • i uses a (Clojure based) invocation handler. This handler will process method invocations on i. i's handler is stateful. Calling setter void set<q>(Q) (of I) on i will set the property (see below) q of type Q. Calling build-mutator I <q>(<Q>) will also set property q but return i. Calling Q get<q>() (of I) on i will return i's current value of property q of type Q [9]. These stateful semantics are implemented in the handler.

  • P's boolean equals(Object o) implementation (of boolean Object.equals(Object)) returns true if o and i are of the same Class and if all properties are Q.equals(Object) [10]. I.e. the equals semantics for P are implemented in the handler -- the equals semantics for each of the properties are implemented in each properties' class/type Q.

  • P's int hashCode() implementation (of int Object.hashCode()) is consistent with P's boolean equals(Object) implementation. The P.hashCode() semantics are implemented in the handler and it's based on Q.hashCode() of the properties. So again, P.hashCode is consistent with P.equals(Object) only if each Q.hashCode() is consistent with Q.equals(Object).

  • P's P clone() implementation (of Object Object.clone()) will return a serialization copy of i [8]. I.e. the copy will be created by deserializing the serialized i. Note that this will usually mean that the copy is a deep clone/copy of the original i (but there is no guarantee since technically there are ways to explicitly and intentionally screw-up on this).

[1] https://clojure.org/
[2] https://en.wikipedia.org/wiki/Data_transfer_object
[3] https://martinfowler.com/bliki/LocalDTO.html
[4] https://www.oracle.com/technetwork/articles/java/javareflection-1536171.html
[5] https://en.wikipedia.org/wiki/Factory_method_pattern
[6] https://docs.oracle.com/javase/8/docs/technotes/guides/reflection/proxy.html
[7] https://www.journaldev.com/17129/java-deep-copy-object
[8] Implementation note: at the moment clone() does not use serialization but just creates a new proxy with the original's value. This may change in future releases though.
[9] If Q is boolean (not Boolean!) getters of the form boolean is<q>() are also recognized.
[10] See Native-typed properties below for more detail on when two DTOs are considered being equal.
[11] Deeto uses the Java serialization mechanism internally to create deep-copies of instances and it exposes DTO implementations which are Serializable. Note that serialization or rather de-serialization introduces a security risk! [12, 13] That's why Clojure's proxy does not support Serializable. Deeto has deeto.SerializableInvocationHandler (see core.clj) to (de-)serialize the DTOs. You must ensure, to ever only consume (i.e. de-serialize) data from a trusted source. Make sure you understand the threat before consuming serialized data.
[12] https://clojure.atlassian.net/browse/CLJ-2204
[13] https://www.infoworld.com/article/3003197/library-misuse-exposes-leading-java-platforms-to-attack.html

Properties

Data transfer objects are mutable containers -- used to transport a bunch of values (properties) that belong together somehow [1]. They usually have a setter and a getter for each property q (of matching types Q). Java Beans [2] have similiar semantics but support the observer pattern and more.

Deeto analyzes the interface I and derives the set of properties and their types. The handler then implements the stateful get-and-set-semantics.

On construction these properties are set to an initialization value (see below).

Note: Deeto supports interface-definitions with only a getter or setter for a property even if those have limited usability.

[1] I don't want to argue about what DTOs exactly are, what to use them for and why you shouldn't and how they are different from value objects. Deeto is for those who do use DTOs for one reason or the other.
[2] https://en.wikipedia.org/wiki/JavaBeans

Native-typed properties

Deeto supports native-typed properties (like boolean and double). Internally these values are stored as instances of their wrapper classes (like Boolean and Double).

floating point datatypes

Java has a mismatch when it comes to compare floating point values [1]. You have 0.0 = -0.0 is true but new Double(-0.0).equals(new Double(0.0)) is false.

So for the implementation of equals(Object) Deeto will compare native-typed (floating point) properties (float and double) via =. But for properties of the wrapper types (Float and Double) Deeto will use the type's equals(Object) method.

This is probably closest to what a Java programmer would do, if she was asked to code an implementation for equals(Object) and the class had a float/double typed field.

Special care must be taken when introducing Deeto and refactoring existing Java code which does not follow this path.

Note thet Deeto's implementation of hashCode() is consistent with these semantics.

As a side note: notice that Arrays.equals(new double[] { -0.0 }, new double[] { 0.0 }) is false which puzzles me. But the Java docs for Arrays.equals(double[], double[]) say just that.

[1] https://stackoverflow.com/questions/45544180/signed-zero-double-equals-as-in-but-double-comparedouble-double-0/45544483

initialization of native-typed properties

Java initializes reference-typed fields (incl. arrays) with null (i.e. nil). For native-typed fields (like boolean and float) Java defines an initialization value for each type (see JLS 4.12.5 Initial Values of Variables [1]).

Deeto mimics this behavior. So when constructing a DTO, properties will be bound to their corresponding type's initialization value.

Currently there is no way for clients to define this initialization value for Deeto's DTOs (like through an annotation or so). You'll have to put that logic into your factories (see below).

[1] https://docs.oracle.com/javase/specs/jls/se7/html/jls-4.html#jls-4.12.5

Fluent interface/builder pattern

A builder [1] is a stateful-container that lets you set/build/collect state (values) and then finally use it as a factory for some object. Usually the builder will provide "setter"-methods that can be chained (fluent API [2]).

Deeto supports this kind of usage by providing/supporting implementations for (build-mutator) methods of the form I <q>(<Q>).

Example: here the property foo of some SomeDto dto can be set via dto.foo(String) which returns dto (i.e. the mutated instance that the method was invoked on).

interface SomeDto extends IDeeto<SomeDto> {

    String getFoo();
    void setFoo(String x);
    SomeDto foo(String x);

With Deeto you're not using a seperate builder but you're using the stateful DTO directly. So all this feature gives you is the option to use method chaining on build-mutators instead of calling setters.

Note: the build-mutators are mutators and not factory methods. So they act like setters and they return the mutated instance (proxy) on which they are invoked.

The build-mutators (methods) of some interface I are discovered by Deeto through their signature:

  • their return type is I
  • they take one argument
  • their (one) argument is q's type Q

The following rules are not enforced at the moment (but may in future releases!):

  • their name matches the name q of one of I´s properties
  • There must at least be a getter for q when there is a build-mutator for q.

Note: Deeto supports interface-definitions with just a build-mutator for a property and no getter even if those have limited usability.

[1] https://en.wikipedia.org/wiki/Builder_pattern
[2] https://en.wikipedia.org/wiki/Fluent_interface

Read-only views to DTOs

Sometimes you may want to construct a DTO but then give clients just read-only access to the DTO. At the moment Deeto has no special feature for this use-case.

But you can implement this by using two interfaces to define the DTO: one public interface for the read-only access for the client and an additional builder interface for your factory [1].

Example: Here we define FooDto with two properties. The factory method FooDto.newInstance uses (the just locally visable) Builder with the mutating builder methods. Clients access the DTO through the read-only view FooDto.

package deeto_user;

import deeto.Deeto;
import deeto.IDeeto;

interface Builder extends FooDto {

    Builder foo(long x);
    Builder bar(boolean x);

}

public interface FooDto extends IDeeto<Builder> {

    long getFoo();
    boolean isBar();

    static FooDto newInstance(long foo, boolean bar, double doo) {
        return Deeto.factory().newInstance(Builder.class).foo(foo).bar(bar);
    }
}

Note that this DTO is not immutable per se. Deeto has no feature that let's you express that. It's just the case that there aren't any mutators accessible to clients (other than using reflection ...).

[1] Deeto could have used true factory-like builder-methods instead of making them a mutator. Then you could have put them into just one interface and still implement the read-only view. But then builders would create a new instance/clone each time they're called which introduces a performance penalty (which I haven't measured -- of course ;-) Anyway, they are mutators and the pattern from above should work fine.

Map-based access

DTOs behave almost like mutable maps with typed key/values (properties). Deeto supports converting between maps and DTOs.

The DTO's interface can/may define

java.util.Map<String, Object> toMap();

When this method is called on a Deeto proxy it returns a map which maps each/all property name (capitalized string; e.g. "FooBar") to its cloned/copied value (possibly nil). I.e. the map will contain an entry for each (all!) property.

Note that values for properties with native type (like int and float) will be returned with their wrapper class (like Integer and Float) [1].

The DTO's interface can/may also define

<T> T fromMap(java.util.Map<String, Object> source);

When this method is called on a Deeto proxy it will use the entries in source to set the property values (clones/copies of source's values!) of the instance on which the method was called. It then returns the (mutated) instance. So this is a mutator and not a factory.

The source map need not contain all of the DTO's properties. If source contains keys for which there is no property in the DTO, an exception is thrown.

Again you have to supply wrapper-typed values for properties with native type. If a native-typed property value in source has null value, an exception is thrown [2].

Note that this method throws an exception if any given value is not type-compatible (as of assignableFrom) with the corresponding property. Wrapper-types are handled as exspected. No other conversion, transformation, cast etc. is done though.

Mutation of the DTO when consuming source is done in one step (just one assignment). If fromMap throws an exception the state which the DTO was in when the method was called will be unchanged (it won't change at all).

[1] https://www.w3schools.com/java/java_wrapper_classes.asp
[2] Currently there is no way for clients to ask Deeto about whether a given property (name) has a native data-type.

Immutability

Deeto's proxys are mutable, stateful containers with getters and setters for access to their properties values. And eventhough the properties are mutable, their values are meant to be immutable.

So Deeto's setter implementation creates a serialization copy (see above) or defensive copy [1, 2] of its argument value (which therefore must be Serializable [4]). So there is no way for users of Deeto to give a setter method (or one of the mutators) a mutable object and keep a reference to that object through which the client could afterwards change that very object and thus change the DTO's value behind the scene. The only way to change the property is by using the setters/mutators. There just is no way you can change a value [3].

The same is true for Deeto's getter/toMap which also return only a defensive copy of their internal (possibly mutable) value object.

[1] http://www.javacreed.com/what-is-defensive-copying/
[2] http://www.javapractices.com/topic/TopicAction.do?Id=15
[3] https://clojure.org/about/state
[4] Deeto supports Serializable typed properties only. One could imagine to support Cloneable typed properties also. But Cloneable/clone is implemented easily in a way that introduces reference leaks and thus breaks with the kind of immutablility which Deeto aims to support. I hope that DTOs usually contain Serializable typed values so that this restriction does not hinder the applicability of Deeto.

Concurrency

Deeto proxies are not thread-safe per se! I.e. the Clojure stuff is thread-safe [1] but serialization of the mutable property values may or may not be thread-safe depending on the classes involved.

Note that the critical part is the cloneing/copying of argument values via serialization (which may or may not be affected by race-conditions) when invoking setters and mutators. Once the values have been copied any further access to the DTO (even the mutation of the DTO's internal state) is thread-safe!

So Deeto can make no guarantee on the overall thread-safeness.

[1] Internally Deeto uses persistent/immutable maps [2] to store property values and swap! [3] to mutate this state atom [4]. All that is thread-safe.
[2] https://clojure.org/reference/data_structures#Maps
[3] https://clojure.org/reference/refs
[4] https://clojure.org/reference/atoms

Mapping DTOs

When using DTOs you often need to map between DTOs and other classes/containers (like JPA entities) back and forth.

Mapping means to use a getter on one container to retrieve a value and then use that value with a setter on another container.

The source and target properties in those containers may or may not have the same name. If they have different types then one needs to convert the source-typed value to a target-typed value before using the setter.

In some cases one may even want/have to apply some sort of business logic to the mapping (like special/default values, handling of nulls and more).

Deeto does not support mappings in any special way. There are other Java tools for that [1]:

Here is an example for Spring's BeanUtil [2]:

package deeto_user;

import java.util.Arrays;
import org.springframework.beans.BeanUtils;
import deeto.Deeto;

public interface HelloWorld {

    HelloWorld name(String x);

    HelloWorld fooBar(String x);
    void setFooBar(String x);
    String getFooBar();

    static HelloWorld newInstance() {
        return Deeto.factory().newInstance(HelloWorld.class);
    }

    public static void main(String[] args) {

        System.out.println("Hello " + newInstance().name("World"));

        System.out.println(Arrays.asList(BeanUtils.getPropertyDescriptors(HelloWorld.class)));

        HelloWorld foo = newInstance().fooBar("FOO");
        System.out.println(foo);

        HelloWorld bar = newInstance().fooBar("BAR");

        BeanUtils.copyProperties(bar, foo);
        System.out.println(foo);

    }
}

Note that Spring does not recognize name(String) and fooBar(String) as (non-standard) setters. So for this example to work, we have to define setFooBar(String) and getFooBar().

[1] https://www.baeldung.com/java-performance-mapping-frameworks
[2] https://docs.spring.io/spring/docs/5.1.8.RELEASE/spring-framework-reference/core.html#beans-beans

Performance

I haven't done any performance measurements. But I expect Deeto to be orders of magnitude slower than a Java implementation of the DTO classes.

This may or may not be a problem for your use-case.

[1] https://dzone.com/articles/java-reflection-but-faster

Usage

You can download Deeto JAR from clojars: https://clojars.org/repo/deeto/deeto/

In addition you need Clojure JAR [1]: https://repo1.maven.org/maven2/org/clojure/clojure/1.8.0/clojure-1.8.0.jar

For Maven users (you need to add clojars to your SNAPSHOT repos):

<dependency>
  <groupId>deeto</groupId>
  <artifactId>deeto</artifactId>
  <version>0.1.0</version>
</dependency>

In your Java code you define an interface for your DTO (SomeDto) with getters and setters. If you like, you can put a static factory method in there too (which will be ignored by Deeto).

Note that SomeDto extends deeto.IDeeto so that the DTO type implements Cloneable and Serializable and has a public clone method so that clone can be called without problems (and no need to cast). The interface also has definitions for fromMap and toMap (see above).

Using deeto.IDeeto is totally optional for using Deeto.

package deeto_user;

import deeto.Deeto;
import deeto.IDeeto;

interface SomeDto extends IDeeto<SomeDto> {

    String getFoo();
    void setFoo(String x);
    SomeDto foo(String x);

    int getBar();
    void setBar(int x);
    SomeDto bar(int x);

    static SomeDto newInstance() {
        return Deeto.factory().newInstance(SomeDto.class);
    }
}

public class DeetoExample {

    public static void main(String[] args) throws Exception {

        SomeDto foo = SomeDto.newInstance();
        System.out.println("foo = " + foo);

        foo.setFoo("FOO!");
        System.out.println("foo = " + foo);

        System.out.println("foo.equals(null) = " + foo.equals(null));
        System.out.println("foo.equals(\"\") = " + foo.equals(""));

        SomeDto bar = foo.clone();
        System.out.println("bar.equals(foo) = " + bar.equals(foo));

        SomeDto fooCopy = Deeto.factory().copyOf(foo);
        System.out.println("fooCopy = " + fooCopy);

        SomeDto barCopy = Deeto.factory().copyOf(bar);
        System.out.println("barCopy = " + barCopy);

        System.out.println("barCopy.equals(fooCopy) = " + barCopy.equals(fooCopy));

        foo.foo("y").foo("x");
        bar.foo("x");

        System.out.println("foo.equals(bar) = " + foo.equals(bar));

        Map fooMap = new HashMap();
        fooMap.put("Bar", 42);
        System.out.println("foo.fromMap(fooMap) = " + foo.fromMap(fooMap));

        System.out.println("foo.toMap() = " + foo.toMap());

    }
}

[1] I've tested version 1.8.0 but it should work with any recent Clojure release.

Notes

  • setters and mutators do only some checks on their arguments. These checks will become stricter in future releases.

  • toString() implementation is not stable yet. So you have to expect changes in future releases.

  • the Clojure code is pretty bad at the moment. It will be refactored but sematics will not change (hopefully :-)

  • serial form will change in the future. You must expect that stored serial forms will not be compatible with future releases.

TODOS

  • add documentation about serial form