Hi Rony (sent privately to not bother the list with this stuff). Here's
a modified ReturnTypesList that also keeps track of all the distinct
methods that can be called:
public class ReturnTypesList extends CopyOnWriteArrayList<Class<?>> {
private final Object lock = new Object();
private volatile List<Method> methods = Collections.emptyList();
public List<Method> getMethods() {
return methods;
}
public void merge(Class<?> newType) {
for (Class<?> type : this) {
if (newType.isAssignableFrom(type)) {
// already have the same or more specific type
return;
}
}
// we need to serialize access when modifying
synchronized (lock) {
// re-check under lock as the list might have been modified
for (Class<?> type : this) {
if (newType.isAssignableFrom(type)) {
// already have the same or more specific type
return;
}
}
// add newType 1st as it is the most specific type among
related ones
// this will make list appear to contain related types for
a brief moment
// bu that's not a problem.
add(newType);
// construct new list of methods and remove the less
specific related types
List<Method> newMethods = new ArrayList<>();
Iterator<Class<?>> iter = iterator();
while (iter.hasNext()) {
Class<?> type = iter.next();
if (type != newType && type.isAssignableFrom(newType)) {
// current type is less specific than newType ->
remove it
iter.remove();
} else {
// current type is unrelated to newType or equal to
newType
// -> collect its methods
newMethods.addAll(Arrays.asList(type.getMethods()));
}
}
// publish newMethods
this.methods = newMethods;
}
}
}
So all you have to do is call
ReturnTypesList.merge(method.getReturnType()) every time you (re)-visit
the registered object returned from the method. When you search for
appropriate method to call on the object, use
ReturnTypesList.getMethods() to search for most appropriate one - this
method returns a list of unique methods (no duplicates).
Regards, Peter
On 01/07/2017 08:54 PM, Peter Levart wrote:
Hi Rony,
Implementing explicit cast is easy in this scheme (the Java side):
public static void cast(Object object, ReturnTypesList rtList,
Class<?> typeToCastTo) throws ClassCastException {
typeToCastTo.cast(object);
rtList.merge(typeToCastTo);
}
Peter
On 01/07/2017 08:46 PM, Peter Levart wrote:
Hi Rony,
As with all concurrent data structures that try to optimize
something, you can get it wrong in the first try. Here's the
corrected code:
public class ReturnTypesList extends CopyOnWriteArrayList<Class<?>> {
private final Object lock = new Object();
public void merge(Class<?> newType) {
for (Class<?> type : this) {
if (newType.isAssignableFrom(type)) {
// already have the same or more specific type
return;
}
}
// we need to serialize access when modifying
synchronized (lock) {
// re-check under lock as the list might have been modified
for (Class<?> type : this) {
if (newType.isAssignableFrom(type)) {
// already have the same or more specific type
return;
}
}
// add newType 1st as it is the most specific type among
related ones
// this may make list appear to contain related types for
a brief moment
// but that's not a problem if the logic that looks up
methods can cope
// with it (it might find duplicate methods)
add(newType);
// 2nd remove the less specific related types
Iterator<Class<?>> iter = iterator();
while (iter.hasNext()) {
Class<?> type = iter.next();
if (type != newType && type.isAssignableFrom(newType)) {
// newType is more specific -> remove less
specific one
iter.remove();
}
}
}
}
}
Regards, Peter
On 01/07/2017 08:33 PM, Peter Levart wrote:
On 01/07/2017 07:25 PM, Rony G. Flatscher wrote:
Hi Peter,
thank you again for your effort, I really appreciate it!
However, as you note at the end yourself, the problem is that any
Java object could be used concurrently in different usages of the
Java bridge, so saving the last return type with the returned
object is not feasible.
I'm not suggesting that. Saving all most specific non-related return
types is what would be needed and then using them all in sequence to
search for method(s).
E.g. each new instance of a javax.script.RexxEngine creates a new
Rexx interpreter instance. Each Rexx interpreter instance allows
any number of Rexx threads to run concurrently and it is possible
to use the Java bridge from any of these Rexx threads concurrently
and use the (identical) Java object in different use cases (e.g.
having different Event handlers implemented in Rexx serving at the
same time different Java interfaces). Add to this the possibility
that the same is possible at the Java side, where (the same or
different) RexxEngines could get exercised in different Java threads.
No problem. What you need is a special concurrent collection
implementation that keeps all the most specific method return types
you add to it which are unrelated. When types are related, you just
keep the most specific one. Like this:
public class ReturnTypesList extends CopyOnWriteArrayList<Class<?>> {
private final Object lock = new Object();
public void merge(Class<?> newType) {
for (Class<?> type : this) {
if (newType.isAssignableFrom(type)) {
// already have the same or more specific type
return;
}
}
// we need to serialize access when modifying
synchronized (lock) {
Iterator<Class<?>> iter = iterator();
while (iter.hasNext()) {
Class<?> type = iter.next();
// re-check under lock as the list might have been
modified
if (newType.isAssignableFrom(type)) {
// already have the same or more specific type
return;
}
if (type != newType && type.isAssignableFrom(newType)) {
// newType is more specific -> remove less
specific one
iter.remove();
}
}
// newType is most specific
add(newType);
}
}
}
...use merge(method.getReturnType()) to keep the list of most
specific return types updated - mostly the type will already be
found in the list and the first for loop will bail out without any
modification or synchronization, so this is quite scalable. When you
search for method, iterate the ReturnTypesList registered with the
object and collect all the methods you find on all types in the list
to select the most appropriate. I think you will find that most
objects will register a single method's return type. There will be
rare occasions where multiple types will be registered.
Regards, Peter
---rony
On 07.01.2017 19:16, Peter Levart wrote:
Hi Rony,
On 01/07/2017 03:53 PM, Rony G. Flatscher wrote:
Hi Peter,
thank you very much for your efforts!
However, in this context there is a problem at hand, that there
is no information available what Java method returned what object
and what cast was carried out, if any. To understand this, maybe
I should give a little bit more information about the Rexx-Java
bridge: Rexx/ooRexx (originally developed by IBM, now in
opensource) is an interpreter for a dynamically typed, caseless
programming language with a rather easy to learn syntax, yet
powerful implemented concepts. ooRexx is implemented in C++.
The Rexx-Java-bridge uses JNI and a Java package (for ooRexx
programmers it is an external function package called BSF4ooRexx,
which allows to camouflage all of Java as the dynamically typed,
caseless ooRexx). It is possible with this package to create Rexx
proxy objects for Java objects (and the other way around as
well). This is realized by storing proxied Java objects on the
Java side in a Map ("registry") and using a common (unique)
string value as the key.
So when the Rexx side invokes a Java method, briefly the
following steps take place (there is much more to this, but not
important in this context):
* the Rexx side uses JNI and supplies the string identifying
the Java object in the Map, the method name in uppercase
(caselessness is realized in Rexx by uppercasing all Rexx
tokens outside of quotes) and the arguments, if any,
* the Java side fetches the Java object from the Java registry
and inspects it for its available methods, picks those that
have caselessly the same name as the supplied method name,
then checks whether the arguments are type-compatible and
invokes the method; any returned Java object will be placed
in the Java "registry" and its key (a unique string) is
returned to Rexx.
Couldn't you save also the method's return type besides the result
under the same key into the registry, so next time you have to
invoke a method on such object, you retrive the object and the
type you use to find methods on?
So after returning control to Rexx, there is no information
available about the Java object in the Java registry other than
the string serving as the key to fetch that Java object on the
Java side.
Not on the Rexx side, but on the Java side in the registry. Right
where you need it, right?
Take this Rexx code as an example (the tilde is the message
operator in ooRexx and can have white space around it):
clzToolkit = bsf.import("java.awt.Toolkit")
dim = clzToolkit ~getDefaultToolkit ~getScreenSize
will be transformed internally by Rexx into:
CLZTOOLKIT=BSF.IMPORT("java.awt.Toolkit")
DIM=CLZTOOLKIT~GETDEFAULTTOOLKIT~GETSCREENSIZE
and the Java bridge gets used (via JNI) as follows:
* step 1: BSF.IMPORT() is an external Rexx function that will
use JNI and cause a Java class object to be loaded (and
stored in the Java registry) and boxed as an ooRexx proxy
class object upon return and assigned to the Rexx variable
CLZTOOLKIT,
This time the 'type' to search methods on is the same as the class
object you just "imported".
*
* step 2: the CLZTOOLKIT~GETDEFAULTTOOLKIT statement contains a
Rexx message that will cause JNI to be used and the Java
method GETDEFAULTTOOLKIT to be executed for the Java object
referenced by CLZTOOLKIT (which incorporates the unique
string for that proxied Java class object); the returned Java
object will be stored in the Java registry, its unique key (a
string) returned, boxed as an ooRexx proxy object value which
will be the receiver of the next Rexx message,
Right and if you also save the return type of the method you just
called into the registry besides the returned object on the Java
side, you can use it later...
*
* step 3: the returned value gets the GETSCREENSIZE Rexx
message sent to it causing JNI to be used and the Java method
GETSCREENSIZE to be located and executed for the Java object
returned from the previous step;
The method should then use the saved method return type from
previous step for looking up the GETSCREENSIZE method...
* the returned Java object will be stored in the Java registry,
its unique key (a string) returned, boxed as an ooRexx proxy
object value that gets assigned to the Rexx variable DIM.
You should then store the getScreenSize() method's return type
besides the returned object under the key... You see the pattern?
*
Each step gets carried out contextless, i.e. there is no Java
context available, that we (or the Java compiler) can see/infer
when looking at a Java program.
When you invoke them method you not only store the returned object
but also the method's return type.
If you need casting, then this would need to be explicit (like in
Java).
There's one problem with this scheme. What is the key you use to
register returned object? Is it based on object identity? When
methods return the same instance, is it saved under the same key?
If yes, which is understandable, then there might be a problem
when two methods with different return types return the same
instance. Which return type should you use to find methods for
following invocations then? Maybe the most specific type (if they
are related) or both (all) of them if they are not and then use
them all to search for methods.
Regards, Peter
---rony
P.S.: Also it might be interesting to know, that with that same
Rexx-Java bridge it is possible to implement Java methods from
interface or abstract classes in Rexx! In that case there is a
Java proxy class available for proxying Rexx objects and on the
Rexx side there is a Rexx Directory to maintain the proxied Rexx
objects for their Java proxies. Fun stuff! :)
On 06.01.2017 23:22, Peter Levart wrote:
Hi Rony,
On 01/06/2017 02:28 PM, Rony G. Flatscher wrote:
>The j.l.r.Method object on which you call invoke() should not be obtained by
inspecting the
>methods of the implementation class given by getDefaultToolkit().getClass().
Implementation
>classes (i.e. classes in non-exported packages) cannot be instantiated, nor
their members
>manipulated, by code outside their module.
>
>The j.l.r.Method object on which you call invoke() should be obtained by
inspecting the methods of
>the "public Java class" java.awt.Toolkit. The first argument that you pass to
invoke(), indicating
>the receiver, can still be instanceof the implementation class.
As was noted earlier, the information that some Java object xyz was created by
some public method
"getDefaultToolkit()" and hence knowing that its return type would be that of
the java.desktop
public class java.awt.Toolkit is not available at runtime.
But it is. The method Toolkit.getDefaultToolkit() has a return
type. You can use reflection to find out that return type of
that method:
Method getDefKitMeth = Toolkit.class.getMethod("getDefaultToolkit");
Class<?> tkClass = getDefKitMeth.getReturnType();
// now you can obtain the toolkit instance:
Object tkInst = getDefKitMeth.invoke(null);
// and obtain a method to be called upon it
Method getScrSizMeth = tkClass.getMethod("getScreenSize");
// and invoke it:
Object screenSize = getScrSizMeth.invoke(tkInst);
... and so on...
You see, I never had to mention java.awt.Toolkit type explicitly
to invoke getScreenSize on an object of that type (or subtype).
If you think what a programmer does when he codes this in
straight Java without using reflection, it is the following:
1. He finds out a factory method on Toolkit class:
Toolkit.getDefaultToolkit()
2. He looks up the return type of that method (in javadocs).
3. He uses that type to declare a local variable to which it
assigns the result of the method invocation:
java.awt.Toolkit tkInst = java.awt.Toolkit.getDefaultToolkit();
4. He looks up and finds an instance method to call in type
java.awt.Toolkit: java.awt.Toolkit.getScreenSize() and writes it
down:
tkInst.getScreenSize();
Above invocation is using static type java.awt.Toolkit - the
return type of Toolkit.getDefaultToolkit().
You can do similar things with reflection. Instead of using
anInstance.getClass() to get the runtime class of the instance,
you can use Method.getReturnType() of the method that was used
to obtain the instance. If API is designed so that no casts are
needed when you chain calls, then this should work.
Regards, Peter