The new reflection - Basics
The new reflection
Class
object to acquire Method
, Constructor
, and/or Field
objects. These objects are used primarily for invocation of a method or constructor (in the former two cases) or read/write access to a field (in the latter case). The mechanism is simple: each reflective object accepts an Object
or array of Object
and yields void
or an Object
with the result (depending on the operation). Because primitives in Java are not Object
instances, in order to pass or return values of primitive type, we rely on so-called "box objects", which are object classes whose purpose is to represent a primitive value as an Object
.
These APIs are fairly useful for a number of use cases (including non-access cases like examining annotations), but come with several limitations:
-
They're heavyweight, requiring boxing/unboxing of inputs and outputs
-
Not type safe (generally operating on
Object
and/orObject[]
) -
Access checking is generally done at every invocation/access point
-
Hard to optimize
-
Sometimes hard to use in conjunction with JPMS (modules)
Method handles
In Java 7, a new class was introduced which was intended to overcome these limitations, called `MethodHandle`
. This class, and its related classes in `java.lang.invoke`
, are all designed to support a new and more efficient means of indirect access to class members. The initial motivating use case included support for the new invokedynamic
JVM instruction as part of JSR 292
.
This efficiency is primarily achieved through a few specific mechanisms:
-
Reducing the number of implementations needed (only one per call site signature versus one per target class)
-
Hoisting access checks to creation time instead of usage time
-
Usage of signature polymorphism (explained below)
In addition to acting as a replacement for a large number of reflection use cases, method handles are also the basis of lambdas and the Foreign Function & Memory (FFM) API , among other things.
What is a method handle?
The MethodHandle
class gives you some methods which can be used to perform an invocation. Of these, these two are the most commonly used:
These methods, and several others in this package, don't have a specific signature (that is, they have no particular number or type of argument). Rather, they are signature polymorphic.
What is signature polymorphism?
In the Java language, a method is identified by the combination of its name and its argument types. Because of this, we can provide many overloads of a method with different argument types and quantity.
Signature polymorphism takes this concept one step farther. When a method qualifies as being signature-polymorphic (meeting certain specific requirements
), it automatically has every possible overload. These methods are declared as though they return Object
and accept Object...
in order to satisfy Java language requirements, but they are treated specially such that no actual varargs array creation or primitive boxing/unboxing is done.
This means that all of these methods (and more) "exist" and can be directly invoked:
-
MethodHandle.invoke(int)
-
MethodHandle.invoke(String,List,int[])
-
MethodHandle.invoke(MethodHandle,double,Console)
-
MethodHandle.invoke(Object[],Object[])
-
MethodHandle.invoke(int,int,int,int,int,int,int,int,int,int,int,int)
It doesn't stop there though. The Java language may identify methods only by their name and argument types, but the Java virtual machine additionally uses the return type to identify a method. So, even though Java won't let you create an overload which differs only by return type, these kinds overloads will nevertheless exist as well - one for every possible return type (which includes void
).
The main benefit of this is that it is possible to call an invoke
method with the exact argument types of the target without having to wrap them in an array or box the arguments, and with the exact return type of the target without having to unbox the result. This in turn makes it easier for the compiler to optimize as well.
However, because of the aforementioned restrictions, the only place you will find this kind of method is in the java.lang.invoke
package - specifically, MethodHandle
and VarHandle
(which will be covered later).
How to call a signature polymorphic method
For the most part, you can call a signature polymorphic method like any other method. However, when such a call is made, the exact method variant which is called is determined by the actual argument types and the inferred return type.
So for example, given mh
of type MethodHandle
, a call site may look like this:
int result = (int) mh.invokeExact(123, "Hello world!", Object.class);
In this case, the method being invoked is inferred to take three arguments, whose types are int
, String
, and Class
, and whose return type is int
, and so out of the infinite possible methods to call, this one (and only this one) is selected.
Ignoring the return value
This works fine as long as the arguments that you give correspond exactly to what the method expects, and that the return type is also exactly what is expected. However, consider the case where you don't care about the return value but you still want to call that same method. With a normal Java invocation, you might try this:
mh.invokeExact(123, "Hello world!", Object.class);
However, with signature polymorphic methods, this could give you a surprise: the selected method no longer has a return type of int
, but of void
, which can cause an error in some cases (explained up ahead).
To force the call to return the type you want, you have to cast the return type, and either assign it to an ignored variable, or otherwise do something with the result that is unambiguous in terms of type:
var _ = (int) mh.invokeExact(123, "Hello world!", Object.class);
Now we've forced the Java compiler to establish that the called method returns int
. Note in this example we are using the relatively recent _
feature, but on older Java versions, you can use a placeholder variable name like ignored
.
Mismatching return or argument types
Another tricky case comes up when you want to pass in or return a value whose type is different from the type on the method you want to call.
For example, imagine we want to call the same method as above, but we want the result to be stored in a long
. If we just change our variable type to long
, we know that the wrong method will be called; we specifically want the one which returns an int
. We can solve this by ensuring that the cast matches the desired return type:
long result = (int) mh.invokeExact(123, "Hello world!", Object.class);
The cast tells Java which method to call, and existing Java rules for widening values will ensure that the int
result can be stored in the long
variable.
The same thing is true with argument types: a cast can be used to specify the exact method we are interested in calling, when the actual argument value's type differs from the type of the corresponding parameter.
The double-cast
One last problem can occur when we specifically want to narrow a result value (that is, cast it to a more specific type). In such cases, we may end up having to cast the return value twice (once to tell Java which method to invoke, and once to actually narrow the return value):
String result = (String) (CharSequence) mh.invokeExact();
In this example we're calling something which is declared to return CharSequence
, but we know (through some means) that it actually will always be an actual String
. Most IDEs will understand this construct and will not warn you about the seemingly redundant cast.
Representing method types with MethodType
Every MethodHandle
instance has a method type, represented by instances of `MethodType`
and accessible using the `MethodHandle#type()`
accessor method. This class encodes the argument types and return type (as Class
instances) of a method, which suffices to identify the type of any given call site. Every call site has an implied MethodType
which goes along with it.
A MethodHandle
may be transformed to have a different type by using the `MethodHandle#asType()`
method. This method returns a new version of the given MethodHandle
, but where each argument and the return type are based on those of the given MethodType
. The number of arguments in the new type must equal the number of arguments in the old type, and the conversion must be valid. For example, if you have a method handle whose return type is String
, and you try to convert it to one which returns an int
, an exception is thrown. However, converting a method handle which returns String
to one which returns CharSequence
is perfectly valid.
Exact versus inexact invocation
A method handle may be invoked exactly or inexactly. Invoking a method handle exactly is done via the MethodHandle.invokeExact(...)
method. When performing an exact
invocation, the call site's implied type and the type of the MethodHandle
must be exactly identical (even the return type), or an exception will be thrown at run time.
Inexact invocation via the MethodHandle.invoke(...)
method is much more flexible. Rather than requiring the method handle's type to exactly match, an inexact invocation works on any method handle which can be reasonably converted to the call site type (as if by calling MethodHandle.asType(callSiteType)
with the exact type of the call site). However, this flexibility comes at a cost, since the argument types have to be checked at run time; this may be unnecessary in many cases for exact invocations.
If a method handle is only going to be used once, then an inexact call via invoke(...)
is usually going to be fine. Otherwise, it is usually preferable to use invokeExact(...)
. One common strategy is to adapt an original MethodHandle
to the expected call site type using asType()
, and store the result in some place where it can be reused thereafter.
Handling exceptions
Unfortunately, the invocation methods on MethodHandle
are declared to throw Throwable
. This is because a MethodHandle
can refer to anything in the JVM that can be called, and those things in turn can be declared to throw any kind or number of exceptions.
To cope with this, you may be tempted to wrap the Throwable
with some kind of RuntimeException
subclass and rethrow it unconditionally. Do not do this !
Instead, use this pattern always:
private void invokeIt(MethodHandle handle, int foo, String bar) {
try {
int ignored = (int) handle.invokeExact(foo, bar);
} catch (RuntimeException | Error e) {
throw e;
} catch (Throwable t) {
throw new UndeclaredThrowableException(t);
}
}
If the method handle in question has some specific checked exceptions that can be thrown, add those to the list:
private void invokeIt(MethodHandle handle, int foo, String bar) throws SpecificException {
try {
int ignored = (int) handle.invokeExact(foo, bar);
} catch (RuntimeException | Error | SpecificException e) {
throw e;
} catch (Throwable t) {
throw new UndeclaredThrowableException(t);
}
}
Acquiring a MethodHandle
: Lookup
A MethodHandle
is not useful unless you can somehow acquire one. In order to acquire a MethodHandle
, you need an instance of `MethodHandles.Lookup`
.
A Lookup
instance provides factory methods which can create method handles representing:
-
Static and instance (virtual) methods
-
"Special" methods e.g.
super.foo()
-
Static and instance field getters and setters
-
Constructors
A Lookup
functions as an access key to a given lookup class, which is the class that is associated with the Lookup
instance. The lookup class can be returned by calling the lookupClass()
getter method on a Lookup
instance.
The access power of the Lookup
depends not only on the lookup class, but also on its _lookup mode_
. The lookup mode determines which access levels are accessible by the Lookup
. For example, a Lookup
with PRIVATE
access may be used to access any private
member that is accessible from the lookup class. However a Lookup
which lacks PRIVATE
access may not access any private
members, even those which would otherwise be accessible from the lookup class. A lookup with reduced lookup modes may be created by calling the `dropLookupMode(int)`
method with the specific mode to drop.
There are several possible ways to acquire a Lookup
. The best strategy to use depends on use case.
The public Lookup
The public Lookup
is a Lookup
which can access any public
member on any public
class. It is a singleton which can be acquired by calling `MethodHandles.publicLookup()`
. No special module flags or privileges are required to acquire and use this Lookup
. The lookup class of the public Lookup
is Object.class
. This type of Lookup
is most suitable for use implementations of APIs which only require access to public
members.
Full privilege Lookup
A full-privilege lookup is available to every class by way of the `MethodHandles.lookup()`
method, and in fact this is the most common way to acquire a Lookup
other than the public Lookup
instance.
This method is static
and does not accept any arguments. The lookup class of the returned Lookup
is always that of the calling class, and the returned Lookup
always has full privileges to that class (including access to all private
fields, methods, and constructors). Therefore it is very important that this instance be secured by the caller. Specifically, store the instance only on private fields and do not share the instance with untrusted APIs without reducing its access mode first.
This type of Lookup
is suitable for a number of use cases:
-
When explicitly granting permissions to other frameworks/APIs
-
When accessing members within the same module and package
-
As a seed for gaining access to other modules
Private access Lookup
Despite what it may seem based on what has been said so far, it is in fact possible (since Java 9) to acquire a Lookup
that allows you to gain access to another class without the class explicitly providing you with its full-privilege Lookup
. This access however is mediated by Java access controls, specifically those relating to modules. It also requires an existing full-privilege lookup to act as a seed.
To acquire the private Lookup
, the instance method `lookup.privateLookupIn(Class<?>)`
is called on the seed Lookup
. This method performs an access control check which is based on the lookup class of the original Lookup
to determine whether that class is allowed to access the target class. This check will pass if the target class is in a module that is open
, or the package of the target class is open
, or the package is open
specifically to the module of the original lookup class. Note that the unnamed module (where classpath classes live) is always considered to be open
.
If this access check passes, the resultant Lookup
will have full power access to everything that is accessible to the target class.
The disadvantage of this approach is that it does require the cooperation of module authors who would have to explicitly open
the required packages.
Next...
In the next post in this series, I will cover some more intermediate-level cases and talk a bit out proper API design around method handles.