Thinking out loud, would isEditable still be needed if there were a lookup that returned only editable java resources, and another query that could return both binary and source resources?  Plugins that operate as editors would use the former to find resources it knows it will edit, and will use the latter in order to inspect other classes (to resolve super class or throws declarations).  Plugins that operate as readers (like scaffold plugins) would use only the latter.  Hmm...even with isEditable, does it make sense to have these two kinds of methods?




On Mon, Feb 18, 2013 at 9:43 PM, Lincoln Baxter, III <lincolnbaxter@gmail.com> wrote:
I don't like it because it's not really optimal, but... I don't really see a better way forward that doesn't break every API we have.


On Mon, Feb 18, 2013 at 9:43 PM, Lincoln Baxter, III <lincolnbaxter@gmail.com> wrote:
Well, I actually meant refactor the entire JavaParser so that the types are immutable by default, but modifiable when desired by doing an intanceof and cast. This would, as John said, be a fairly invasive, with a fair amount of downstream API impact, since all users of these APIs would now need to reference these new types instead. It would mean refactoring all of Forge.

So from the perspective of Migration, it probably makes sense to do something like add a method "isEditable()"


On Thu, Feb 14, 2013 at 2:19 PM, Thomas Frühbeck <fruehbeck@aon.at> wrote:
If I understood Lincoln correctly, he meant adding an interface EditableJavaSource, so no change in the existing implementations necessary.
So the boolean is substitutable by instanceof, your graceful error handling would be supported.

Am 14.02.2013 20:10, schrieb John Franey:
I would follow my object-oriented instinct to agree that a editable/non-editable parallel of the model would make sense.  After all, an editable java source is kind of a non-editable java source, and adds editing behavior.....

However, the parallel model seems to be a significant undertaking, and has a negative characteristic, in my opinion.   Lets say there exists a method to search for and return a JavaClass, editable or not.  In a use case where the JavaClass is to be modified, this method is inadequate. 1)  The caller would have to check if the result is instance of EditableJavaClass, then cast, or 2) an alternate query method would be provided to return only EditableJavaClass and since methods cannot be overridden by return type the alternate query method would have to be defined in a different interface.  Not taht either of these are difficult to overcome, but I think the parallel model would increase overall complexity a little and add to the level of effort, and the benefit gained is merely type protection against the runtime error of attempting to modify a non-editable java component.

To go against that impulse just a second, consider the option of a property on the base class, a boolean: editable.  This would avoid the parallel model, avoid casting and support polymorphic calls.   The runtime error of attempting to modify a non-editable component can be handled in a few different ways: 1) a runtime exception, 2) a quiet no-op.  Without the editable property, I would not use a runtime exception.  There would be no chance for the forge plugin programmer to avoid the error gracefully.  With the property, the error is the programmer's and a runtime exception could be appropriate.

Regards,
John



On Thu, Feb 14, 2013 at 1:11 PM, Lincoln Baxter, III <lincolnbaxter@gmail.com> wrote:
As I understood it, we need a way of explicitly noting when a JavaSource instance is mutable or not. Perhaps splitting the API into a JavaSource an EditableJavaSource parallel.

It was also my understanding that proxies were going to be used to do lazy-classloading for any JARs brought in via this system. I'm not sure we need to worry about this just yet. We can get it working then think about performance implications.


On Thu, Feb 14, 2013 at 12:23 PM, John Franey <jjfraney@gmail.com> wrote:
I have no intention of providing a way to modify the class definition of a java class defined within a dependency of the user's project.

Putting the question another way:

To date, the forge java model today supports only modifiable java components.  This effort introduces non-modifiable java components to the forge java model.   This raises the question:  Would the non-modifiable java components be inspected with the same api that supports the modifiable java components (JavaClass, and others).  If no, what api is used to inspect the non-modifiable java components?

My naive answer is: the non-modifiable java components would be inspected using the same api as the modifiable java components.  Methods of that api that expressly modify the java component would be inert for non-modifiable java components.

John



On Thu, Feb 14, 2013 at 12:05 PM, Thomas Frühbeck <fruehbeck@aon.at> wrote:
Hmm, I expect loading of JARs not to the problem, is it? So the loading and reflecting on the "external" class should be possible.
I was thinking of the next step, implementing kind of writable JavaClass not just ignoring the changes, but making the modified class available to the project.
Sorry if I misunderstood your quest? =)

Thomas

Am 14.02.2013 17:37, schrieb John Franey:
Thomas,

I have minimal exposure to proxy due to experience with hibernate, but my understanding is not adequate to understand how they would apply.  Do I understand correctly that the benefit of a dynamic proxy is high when a temporary class implementation is needed, and when a method of the proxy is invoked, some action is taken, perhaps instantiating another implementation of the interface.  In this use case, we don't need to invoke the methods of a project's class, we need to inspect the methods (and other members) of the class, right?

John




On Thu, Feb 14, 2013 at 11:22 AM, Thomas Frühbeck <fruehbeck@aon.at> wrote:
exactly what I was looking for :-))
Thanks George!

Am 14.02.2013 16:55, schrieb George Gastaldi:
Hi Thomas,

Have a look in Forge 2.0 source code. We're using javassist at it's best in the proxy module



Em 14/02/2013, às 13:53, Thomas Frühbeck <fruehbeck@aon.at> escreveu:

Hi John,

my two cents:
    - this feature is a must-have, if Forge should be more than a tool to iniitialize projects, really great idea
    - being pragmatic I would say this calls for proxy classes, similar to CDI decorators or the copy-on-write strategy

(AFAIK the downside to CDI decorators is that they need interfaces on the base classes, thus again requiring changes of the classes if they hadnt been designed for it firstplace.)

I have a very similar problem I am currently trying to solve with silly wrapper classes and was starting to think about dynamic proxy generation - unfortunately I have _no_ experience with such technology other than being simple user :-/

Have you thought about javassist? Is it an option at all?

Thomas


Am 14.02.2013 16:21, schrieb John Franey:
My motivation for this email is to satisfy FORGE-773.  However, this is also related to FORGE-563 and FORGE-424, and resolution could enable other features.

I have written a prototype:
1) an implementation of the forge java api interfaces which delegates to java's reflection, offering a read only perspective of java components.
2) a forge module, currently a facet, to search for a given binary class in the project's dependencies and returns the result wrapped in the above delegate.

These are demonstrable in a unit test.

My dilemma now is how to integrate these into the forge project.  There are a few different areas, but I'll start with this:


For some callers, a java class is a java class, whether it originates as source code (from the current forge project) or is a class from the dependency set.  For example, scaffolding primarily is a read only operation.  In this use case, it would be simpler for these clients to have a single interface to resolve classes because whether a class is source or binary is not relevant to the use case.

On the other hand, there is a set of classes in a user's project that are modifiable.  In these cases, a java class is not a java class.  Forge components might want the distinction somehow.  There ought the be some distinction of which class is modifiable and which is not.

Naively, I took the first thinking that the existing forge java model would be adequate.  To have separate java api for read-only and read-write java model objects seems a fundamental addition to the java model which requires much more effort.  In absence of such a model, I though to implement 'no-op' for those code changing methods  (e.g., Named.setName() would be inert).  I assumed that forge component that change source code would have necessary context to know when it is operating on a source code module, avoiding attempts to modify a binary class.

So, I'm looking for discussion and consensus on the above.  Any thoughts?

Regards,
John










_______________________________________________
forge-dev mailing list
forge-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/forge-dev

_______________________________________________
forge-dev mailing list
forge-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/forge-dev


_______________________________________________
forge-dev mailing list
forge-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/forge-dev


_______________________________________________
forge-dev mailing list
forge-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/forge-dev



_______________________________________________
forge-dev mailing list
forge-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/forge-dev


_______________________________________________
forge-dev mailing list
forge-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/forge-dev


_______________________________________________
forge-dev mailing list
forge-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/forge-dev



--
Lincoln Baxter, III
http://ocpsoft.org
"Simpler is better."

_______________________________________________
forge-dev mailing list
forge-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/forge-dev



_______________________________________________
forge-dev mailing list
forge-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/forge-dev


_______________________________________________
forge-dev mailing list
forge-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/forge-dev



--
Lincoln Baxter, III
http://ocpsoft.org
"Simpler is better."



--
Lincoln Baxter, III
http://ocpsoft.org
"Simpler is better."

_______________________________________________
forge-dev mailing list
forge-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/forge-dev