Skip to content

Commit

Permalink
Initial proposal for method invokers: invoker builder features
Browse files Browse the repository at this point in the history
This commit introduces actual interesting features to the `InvokerBuilder`:

- automatic lookup of the target instance and arguments
- transformation of the target instance and arguments before invocation
- transformation of the return value and thrown exception after invocation
- wrapping the invoker into a custom piece of code for maximum flexibility
  • Loading branch information
Ladicek committed Oct 11, 2023
1 parent f86d856 commit 7ab970a
Show file tree
Hide file tree
Showing 2 changed files with 347 additions and 2 deletions.
4 changes: 4 additions & 0 deletions api/src/main/java/jakarta/enterprise/invoke/Invoker.java
Original file line number Diff line number Diff line change
Expand Up @@ -118,6 +118,10 @@ public interface Invoker<T, R> {
*
* TODO the previous 2 paragraphs refer to "assignability", which needs to be defined somewhere!
*
* TODO when the `InvokerBuilder` applies transformations, some of the requirements above
* are no longer strictly necessary, should reflect that in this text somehow (it is already
* mentioned in `InvokerBuilder`, but that likely isn't enough)
*
* @param instance the instance on which the target method is to be invoked, may only be {@code null}
* if the method is {@code static}
* @param arguments arguments to be supplied to the target method, may only be {@code null}
Expand Down
345 changes: 343 additions & 2 deletions api/src/main/java/jakarta/enterprise/invoke/InvokerBuilder.java
Original file line number Diff line number Diff line change
Expand Up @@ -11,16 +11,357 @@
package jakarta.enterprise.invoke;

/**
* Builder of {@link Invoker}s.
* Builder of {@link Invoker}s that allows configuring input lookups, input and output
* transformations, and invoker wrapping. The method for which the invoker is built is
* called the <em>target method</em>. If a lookup is configured, the corresponding input
* of the invoker is ignored and an instance is looked up from the CDI container before
* the target method is invoked. If a transformation is configured, the corresponding input
* or output of the invoker is modified in certain way before or after the target method
* is invoked. If a wrapper is configured, the invoker is passed to custom code for getting
* invoked. As a result, the built {@code Invoker} instance may have more complex behavior
* than just directly calling the target method.
* <p>
* Transformations and wrapping are expressed by ordinary methods that must have
* a pre-defined signature, as described below. Such methods are called
* <em>transformers</em> and <em>wrappers</em>.
* <p>
* Invokers may only be built during deployment. It is not possible to build new invokers
* at application runtime.
*
* <h2>Example</h2>
*
* Before describing in detail how lookups, transformers and wrappers work, let's take
* a look at an example. Say we have the following bean with a method:
*
* <pre>
* class MyService {
* String hello(String name) {
* return "Hello " + name + "!";
* }
* }
* </pre>
*
* And we want to build an invoker that looks up {@code MyService} from the CDI container,
* always passes the argument to {@code hello()} as all upper-case, and repeats the return
* value twice. To transform the argument, we can use the zero-parameter method
* {@code String.toUpperCase()}, and to transform the return value, we write a transformer
* as a simple {@code static} method:
*
* <pre>
* class Transformations {
* static String repeatTwice(String str) {
* return str + " " + str;
* }
* }
* </pre>
*
* Then, assuming we have obtained the {@code InvokerBuilder} for {@code MyService.hello()},
* we can set up the lookup and transformations and build an invoker like so:
*
* <pre>
* builder.setInstanceLookup()
* .setArgumentTransformer(0, String.class, "toUpperCase")
* .setReturnValueTransformer(Transformations.class, "repeatTwice")
* .build();
* </pre>
*
* The resulting invoker will be equivalent to the following class:
* <pre>
* class TheInvoker implements Invoker&lt;MyService, String&gt; {
* String invoke(MyService ignored, Object[] arguments) {
* MyService instance = CDI.current().select(MyService.class).get();
* String argument = (String) arguments[0];
* String transformedArgument = argument.toUpperCase();
* String result = instance.hello(transformedArgument);
* String transformedResult = Transformations.repeatTwice(result);
* return transformedResult;
* }
* }
* </pre>
*
* The caller of this invoker may pass {@code null} as the target instance, because
* the invoker will lookup the target instance on its own. Therefore, calling
* {@code invoker.invoke(null, new Object[] {"world"})} will return
* {@code "Hello WORLD! Hello WORLD!"}.
*
* <h2>General requirements</h2>
*
* To refer to a transformer or a wrapper, all methods in this builder accept:
* 1. the {@code Class} that that declares the method, and 2. the {@code String} name
* of the method.
* <p>
* Transformers may be {@code static}, in which case they must be declared directly
* on the given class, or they may be instance methods, in which case they may be declared
* on the given class or inherited from any of its supertypes.
* <p>
* It is possible to register only one transformer of each kind, or for each argument
* position in case of argument transformers. Attempting to register a second transformer
* of the same kind, or for the same argument position, leads to an exception.
* <p>
* Wrappers must be {@code static} and must be declared directly on the given class.
* It is possible to register only one wrapper. Attempting to register a second wrapper
* leads to an exception.
* <p>
* It is a deployment problem if no method with given name and valid signature is found,
* or if multiple methods with given name and different valid signatures are found. It is
* a deployment problem if a registered transformer or wrapper is not {@code public}.
* <p>
* Transformers and wrappers may declare the {@code throws} clause. The declared exception
* types are ignored when searching for the method.
* <p>
* For the purpose of the specification of transformers and wrappers below, the term
* <em>any-type</em> is recursively defined as: the {@code java.lang.Object} class type,
* or a type variable that has no bound, or a type variable whose first bound is
* <em>any-type</em>.
*
* <h2>Input lookups</h2>
*
* For the target instance and for each argument, it is possible to specify that the value
* passed to {@code Invoker.invoke()} should be ignored and a value should be looked up
* from the CDI container instead.
* <p>
* For the target instance, a CDI lookup is performed with the required type equal to the bean
* class of the bean to which the target method belongs, and required qualifiers equal to the set
* of all qualifier annotations present on the bean class of the bean to which the target method
* belongs. When the target method is {@code static}, the target instance lookup is skipped.
* <p>
* For an argument, a CDI lookup is performed with the required type equal to the type of
* the corresponding parameter of the target method, and required qualifiers equal to the set
* of all qualifier annotations present on the corresponding parameter of the target method.
* <p>
* Implementations are required to resolve all lookups during deployment. It is a deployment
* problem if the lookup ends up unresolved or ambiguous.
* <p>
* If the looked up bean is {@code @Dependent}, it is guaranteed that the instance will be
* destroyed after the target method is invoked but before the the invoker returns. The order
* in which the looked up {@code @Dependent} beans are destroyed is not specified.
* <p>
* The order in which input lookups are performed in not specified and must not be relied upon.
*
* <h2>Input transformations</h2>
*
* The target method has 2 kinds of inputs: the target instance (unless the target method is
* {@code static}, in which case the target instance is ignored and should be {@code null}
* by convention) and arguments. These inputs correspond to the parameters of
* {@link Invoker#invoke(Object, Object[]) Invoker.invoke()}.
* <p>
* Each input can be transformed by a transformer that has one of the following signatures,
* where {@code X} and {@code Y} are types:
*
* <ul>
* <li>{@code static X transform(Y value)}</li>
* <li>{@code static X transform(Y value, Consumer<Runnable> cleanup)}</li>
* <li>{@code X transform()} &ndash; in this case, {@code Y} is the type of the class that
* declares the transformer</li>
* </ul>
*
* An input transformer must produce a type that can be consumed by the target method.
* Specifically: when {@code X} is <em>any-type</em>, it is not type checked during deployment.
* Otherwise, it is a deployment problem if {@code X} is not assignable to the corresponding type
* in the declaration of the target method (that is the bean class in case of target instance
* transformers, or the corresponding parameter type in case of argument transformers). {@code Y}
* is not type checked during deployment, so that input transformers may consume arbitrary types.
* TODO this paragraph refers to "assignability", which needs to be defined somewhere!
* <p>
* When a transformer is registered for given input, it is called before the target method is
* invoked, and the outcome of the transformer is used in the invocation instead of the original
* value passed to the invoker by its caller.
* <p>
* If the transformer declares the {@code Consumer<Runnable>} parameter, and the execution
* of the transformer calls {@code Consumer.accept()} with some {@code Runnable}, it is
* guaranteed that the {@code Runnable} will be called after the target method is invoked but
* before the invoker returns. These {@code Runnable}s are called <em>cleanup tasks</em>.
* The order of cleanup task execution is not specified. Passing a {@code null} cleanup task
* to the {@code Consumer} is permitted, but has no effect.
* <p>
* If an input transformation is configured for an input for which a lookup is also configured,
* the lookup is performed first and the transformation is applied to the looked up value.
* If the looked up bean for some input is {@code @Dependent}, it is guaranteed that all
* cleanup tasks registered by a transformer for that input are called before that looked up
* {@code @Dependent} bean is destroyed.
* <p>
* The order in which input transformations are performed in not specified and must not
* be relied upon.
*
* <h2>Output transformations</h2>
*
* The target method has 2 kinds of outputs: the return value and the thrown exception. These
* outputs correspond to the return value of {@link Invoker#invoke(Object, Object[]) Invoker.invoke()}
* or its thrown exception, respectively.
* <p>
* Each output can be transformed by a transformer that has one of the following signatures,
* where {@code X} and {@code Y} are types:
*
* <ul>
* <li>{@code static X transform(Y value)}</li>
* <li>{@code X transform()} &ndash; in this case, {@code Y} is the type of the class that
* declares the transformer</li>
* </ul>
*
* An output transformer must consume a type that can be produced by the target method.
* Specifically: when {@code Y} is <em>any-type</em>, it is not type checked during deployment.
* Otherwise, it is a deployment problem if {@code Y} is not assignable from the return type of
* the target method in case of return value transformers, or from {@code java.lang.Throwable}
* in case of exception transformers. {@code X} is not type checked during deployment, so that
* output transformers may produce arbitrary types.
* TODO this paragraph refers to "assignability", which needs to be defined somewhere!
* <p>
* When a transformer is registered for given output, it is called after the target method
* is invoked, and the outcome of the transformer is passed back to the caller of the invoker
* instead of the original output produced by the target method.
* <p>
* If the target method returns normally, any registered exception transformer is ignored; only
* the return value transformer is called. The return value transformer may throw, in which case
* the invoker will rethrow the exception. If the invoker is supposed to return normally,
* the return value transformer must return normally.
* <p>
* Similarly, if the target method throws, any registered return value transformer is ignored;
* only the exception transformer is called. The exception transformer may return normally,
* in which case the invoker will return the return value of the exception transformer. If
* the invoker is supposed to throw an exception, the exception transformer must throw.
* TODO this requires that implementations catch java.lang.Throwable, which is perhaps a bit too much?
* maybe stick with java.lang.Exception?
*
* <h2>Invoker wrapping</h2>
*
* An invoker, possibly utilizing input lookups and input/output transformations, may be wrapped
* by a custom piece of code for maximum flexibility. A wrapper must have the following signature,
* where {@code X}, {@code Y} and {@code Z} are types:
*
* <ul>
* <li>{@code static Z wrap(X instance, Object[] arguments, Invoker<X, Y> invoker)}</li>
* </ul>
*
* A wrapper must operate on a matching instance type. Specifically: when {@code X} is
* <em>any-type</em>, it is not type checked during deployment. Otherwise, it is a deployment
* problem if {@code X} is not assignable from the class type of the bean class to which
* the target method belongs. {@code Y} and {@code Z} are not type checked during deployment.
* <p>
* When a wrapper is registered, 2 invokers for the same method are created. The <em>inner</em>
* invoker applies all lookups and transformations, as described in previous sections, and
* invokes the target method. The <em>outer</em> invoker calls the wrapper with the passed
* instance and arguments and an instance of the inner invoker. The outer invoker is returned
* by this invoker builder.
* <p>
* In other words, the outer invoker is equivalent to the following class:
*
* <pre>
* class InvokerWrapper implements Invoker&lt;X, Z&gt; {
* Z invoke(X instance, Object[] arguments) {
* // obtain the invoker as if no wrapper existed
* Invoker&lt;X, Y&gt; invoker = obtainInvoker();
* return SomeClass.wrap(instance, arguments, invoker);
* }
* }
* </pre>
*
* If the wrapper returns normally, the outer invoker returns its return value, unless the wrapper
* is declared {@code void}, in which case the outer invoker returns {@code null}. If the wrapper
* throws an exception, the outer invoker rethrows it directly.
* <p>
* The wrapper is supposed to call the invoker it is passed, but does not necessarily have to.
* The wrapper may call the invoker multiple times. The wrapper must not use the invoker
* in any other way; specifically, it is forbidden to store the invoker instance anywhere
* or pass it to other methods that do not follow these rules. Doing so leads to non-portable
* behavior.
*
* <h2>Type checking</h2>
*
* An invoker created by this builder has relaxed type checking rules, when compared to
* the description in {@link Invoker#invoke(Object, Object[]) Invoker.invoke()}, depending
* on configured lookups, transformers and wrapper. Some types are checked during
* deployment, as described in previous sections. Other types are checked during invocation,
* at the very least due to the type checks performed implicitly by the JVM. The lookups,
* transformers and the wrapper must arrange the inputs and outputs so that when the method
* is eventually invoked, the rules described in
* {@link Invoker#invoke(Object, Object[]) Invoker.invoke()} all hold.
* <p>
* TODO specify what happens when a transformer/wrapper declares a parameter of a primitive type
* but the actual value passed to the invoker is `null` (the transformer should get a zero value?)
* TODO specify what happens when a transformer/wrapper declares a parameter of some type
* but the actual value passed to the invoker is not assignable to it (CCE?)
*
* @param <T> type of outcome of this builder; always represents an {@code Invoker},
* but does not necessarily have to be an {@code Invoker} instance directly
* @since 4.1
*/
// TODO more kinds of transformations could be defined, expecially for argument handling
// TODO it would be possible to specify a sequence of transformations for each input/output, instead of just one
public interface InvokerBuilder<T> {
/**
* Enables lookup of the target instance.
*
* @return this builder
*/
InvokerBuilder<T> setInstanceLookup();

/**
* Enables lookup of the argument on given {@code position}.
*
* @param position zero-based argument position for which lookup is enabled
* @return this builder
* @throws IllegalArgumentException if {@code position} is greather than or equal to
* the number of parameters declared by the target method
*/
InvokerBuilder<T> setArgumentLookup(int position);

/**
* Configures an input transformer for the target instance.
*
* @param clazz class that declares the transformer
* @param methodName transformer method name
* @return this builder
* @throws IllegalStateException if this method is called more than once
*/
InvokerBuilder<T> setInstanceTransformer(Class<?> clazz, String methodName);

/**
* Configures an input transformer for the argument on given {@code position}.
*
* @param position zero-based argument position for which the input transformer is configured
* @param clazz class that declares the transformer
* @param methodName transformer method name
* @return this builder
* @throws IllegalArgumentException if {@code position} is greather than or equal to
* the number of parameters declared by the target method
* @throws IllegalStateException if this method is called more than once with the same {@code position}
*/
InvokerBuilder<T> setArgumentTransformer(int position, Class<?> clazz, String methodName);

/**
* Configures an output transformer for the return value.
*
* @param clazz class that declares the transformer
* @param methodName transformer method name
* @return this builder
* @throws IllegalStateException if this method is called more than once
*/
InvokerBuilder<T> setReturnValueTransformer(Class<?> clazz, String methodName);

/**
* Configures an output transformer for the thrown exception.
*
* @param clazz class that declares the transformer
* @param methodName transformer method name
* @return this builder
* @throws IllegalStateException if this method is called more than once
*/
InvokerBuilder<T> setExceptionTransformer(Class<?> clazz, String methodName);

/**
* Configures an invoker wrapper.
*
* @param clazz class that declares the invoker wrapper
* @param methodName invoker wrapper method name
* @return this builder
* @throws IllegalStateException if this method is called more than once
*/
InvokerBuilder<T> setInvocationWrapper(Class<?> clazz, String methodName);

/**
* Returns the built {@link Invoker} or some represention of it. Implementations are allowed
* but not required to reuse already built invokers for the same target method.
* but not required to reuse already built invokers for the same target method with the same
* configuration.
*
* @return the built invoker
*/
Expand Down

0 comments on commit 7ab970a

Please sign in to comment.