Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: AW: [jwt-dev] dynamic properties

Hi Christian and all

I really think Christian's proposition is a very good one, however I did not my nice page ( ) for nothing, so I've tried to see where are the differences, and here is what I've come up with !

Who needs to store typed information in JWT ?
* for small, user-originated needs, Christian's is perfect since it is simple, efficient (since not dynamic), answers all requirements and moreover doesn't need additional (ecore) metamodel file : if made using EMF (which can also provide an easy way to a graphical editor), Christian's metametamodel can even be bundled within a .workflow. * for complex, vendor-originated needs, static ecore extension allow to manage arbitrarily complex additional models. These extensions are defined in plugin.xml and provided in vendor jars. Decoupling an extension's metamodel (ecore file) from JWT's metamodel is done by restraining extension on a given containment reference (like advised before).

Now, how is this enough ?
* both can share the same type system, based on Christian's proposed metametamodel, and possibly the same EMF containment reference typed by a identical base class. * for needs that are in between these two extremities... well, we'll see later if there is still such need, if dynamic EMF can do something about it, and if dynamic ecore resolution can be given a usable answer. Actually that's the beauty of it, we'd still be able to do it ^^

On a less urgent note, some additional "nice to have" extension features that have appeared since : * customizing the property editor : done using a CustomPropertyEditor (like Florian said to Mickaël). Should it be configured in the plugin.xml or in the metametamodel ? * customizing the "newChild" command list on a model element : for example with two propertye extension on Action, this would allow to the user which ones he want to see between "new Action," "new CustomActionA", "new CustomActionB" and "new CustomActionAAndB".


Marc Dutoo a écrit :
Hi Christian,

wow these are some very fruitful interactions indeed ^^

This has prodded me to improve , and to provide answers within your email below

Christian Saad a écrit :
Hi Marc,

Thank you very much for your feedback and great explanation of your ideas.

As you wrote, it would be better to implement a generic extension mechanism to enrich model elements with custom aspects instead of just adding dynamic properties. If I got your idea about the open aspect element right, it would
be possible to use a generic, abstract EObject Aspect instead of a typed
yes ! Obviously the default newChild commands should be removed in the ItemAdapter (because it would allow to create any kind of element), but if instantiation is controlled by your metametamodel it is already the case.

I feel using only properties is too much of a constraint for vendors whose features need complex configuration like Imixs or Nova Bonita (see )...

Using this technique, the value field of a dynamic property could easily be
dynamically created as an EAttribute of the corresponding type.
(Concerning the typing issue, I've got a small comment: Maybe we should
consider to limit aspects to be subclassed from ModelElement, as is
everything else in the metamodel, since some routines in JWT may depend on
I agree, that's what I did in my tests, and a meeting point with yours ;)
You wrote that a dynamically typed Aspect could not be extended with static
elements. Does this mean that a static class cannot be subclassed by
statically AND dynamically generated classes (e.g. the subclass
AbstractAspect>DynamicPropertyValue has to be created dynamically) ?
Yeah, I know, those things are tricky ^^ Actually I'll need to redo more specific tests, but I'm pretty sure a static class can only contain static OR dynamic subelements but not both. What I mean is that a dynamic extension of a static class can't contain static subelements, even inherited ones. Ex. (tried) : a dynamic LogAction extending the static Action type can't contain Points, because Points will be parsed as DynamicEObjectImpl since they are underneath a dynamic object, whereas Action.setValue(point) would like a PointImpl .
Also, I clearly forgot about the requirement to attach properties/aspects to
just some model elements. However, keeping in line with the metamodeling
approach, one could perceive an aspect as an addition to a metamodel element which is then inherited by all instances. Basically I'm also thinking that this could make the management of the aspects easier, since the general rule
would be: add it to all objects of this type ;)
Maybe a "switch" attribute could be considered to activate/deactivate it for single objects with an additional possibility to set a default behavior on
the meta level.
I like keeping it simple ;) Your ideas have pushed me to detail the requirements . I think this shows that two kind of aspects are required : those that are like subtypes of existing model elements (ex. dynamic extension of abstract Application, or on Action SCA runtime configuration, properties configuring XPDL specific features...), and those that are like transversal, additional features (ex. logging, Mail tab...). The first we could call "singleton" and require to be preinstanciated, the second we could call "feature" and let it be activated/deactivated (looks like your switch).

Besides this, I like the idea of proposing additional "newChild" commands configured to instanciate a model element with a configured list of "singleton" (preinstanciated) Aspects, but this is on the UI level and not on the model level anymore. This will allow to instanciate different kind of dynamic Applications. What I mean is, ok for preinstanciation since it simplifies things, but in this case having only one "newChild" command for each static model element does not give enough flexibility : rather, more should be configurable with different singletons... What do you think ?

Finally, I fear that adding dynamic extensions everywhere by default would be a risk for performances... Though I agree your static solution would not be a problem in this case.

I've summed it up here . Still thinking it up...
Like you mentioned, the problem when loading a model without the required metamodel extensions is certainly a problem that presents itself when using this feature. When a completely dynamic extension is used, this renders the model unloadable. However, loading dynamic properties without their external
description might just work if implemented as a static extension ;)
Yep, though their metametamodel has to be found ;) by the way, the metametamodel could even be inlined in the .workflow file if it is built in EMF...

That's why your proposition should also be done.

I've tried to list all ways here
Applying the view management to aspects would certainly be a challenging
task. Of course, because I'm not quite familiar with the domains where
dynamic aspects will be used, I cannot decide how important this issue is, but my subjective impression at the moment is that, if a user specifically adds metamodel extensions, these don't necessarily have to be hidden (please correct me if I'm wrong here, this just looks like a tempting opportunity to
skip a difficult issue).
Yep, displayObject() returns true if the object is not found in the view (tried it) ^^ In my mind view management and your metametamodel (which also controls extension behaviour and display) would nicely merge together...
Anyway, probably the best way to see how this can be implemented and to
detect additional difficulties, would be to just try out several different
approaches ;)


-----Ursprüngliche Nachricht-----
Von: jwt-dev-bounces@xxxxxxxxxxx [mailto:jwt-dev-bounces@xxxxxxxxxxx] Im
Auftrag von Marc Dutoo
Gesendet: Mittwoch, 25. Juni 2008 12:13
An: Java Workflow Toolbox
Betreff: Re: [jwt-dev] dynamic properties

Hi Christian

Thanks for the research work you've done, this feedback, and the interesting solution you propose !

Nice info on the tabbed properties feature. OK, an adapter should do the thing, once the "rest' is ok.

About dynamic EMF : you're right, it has its limitations.

As you say, it is hard to create a dynamic extension of a static class. It should be doable nonetheless, but the question is : is it worth the effort (see below).

An important limitation is that dynamic and static subelements can't be mixed in a class, meaning that you can't extend an Action with a dynamic containment reference because Action may contain static Point subelements.

[actually, our ideas have a lot in common - see , and from now on I'll try to match them, ex. "DynamicProperty" / "Aspect"]

["properties" / "aspects" containment reference]
Therefore, and here it matches one of your conclusions, the right way of extending the jwt mm is to add a static containment reference where we want it to be extensible, in your case a "properties" containment reference that is typed DynamicProperty, and in my case a more open alternative consisting in an "aspects" one that is typed EObject (easy, open) or statically AspectImpl (if we want to bother with allowing to dynamically extend a static Aspect class ; note that being typed dynamically Aspect would forbid to statically extend it, which is a nice possibility).

We agree that extending through such a containment reference also allows to manage "typed extension" i.e. a typed group of properties.

[external property description and "assigned model elements" / "profile"] What you also crucially propose is a way to manage the consistency of which extended elements are allowed / available on which element. This is akin to what I termed "profile" in the jwt wiki. I didn't propose a functional or tech spec of it, and I like yours because it is simple. You do it through the cleverly external property description - I thought about putting this info in specific properties of the EMF model like EMF/XML does with ex. "xmlns:prefix", but your way has better separation of concern and more in line with the existing .

However it lacks some features ; for example it adds extended properties on all of its "assigned model elements", though the user would only want to be able to instanciate it on some of them (ex. only configure logging for some of them).

You also propose to manage versions, which I didn't first think of and which has been pointed at since. I agree that it has to be managed at the Model level i.e. at the whole extension level (and not extension element / property by extension element / property).

Comparison of both solutions : mine is more open, yours is better managed. So how about trying to merge it ? Actually, I think it can be done ^^ like this :

Merging both :

* Opening up the extension type i.e. using extensible Aspect rather than DynamicProperty is more flexible and would provide a extension mechanisme that is consistent from simple, dynamic extensions to complex, tree-like, static extensions.

* Your solution answers simple use cases very efficiently, but it can be implemented on top of mine (by providing in the core model a DynamicProperty that is a dynamic extension of Aspect) without losing its ease of use.

* Managing manage which extension is allowed where is paramount, and using an external metaconf file like yours is a nice idea that is in line with how views are handled today (as your "visible" feature on DynamicProperty confirms), and where we're trying to make them go (an eclipse plugin that not only manages views, but also other packaging and conf related issues like - surprise - extensions).

* I'd like to add in the external metaconf besides "assigned model elements" on DynamicProperty / Aspect the following metaproperties : * "singleton" (meaning it's instanciated everywhere, like you spec'd ; if false the user will be able to add it using the newChild action), * "removable" (false by default, so singleton by default wouldn't be removable, so a feature that may be activated and deactivated would be removable and not singleton). * something like "emfType" / "instanceClassName" / Factory / emf extension file that tells what EMF type to instanciate rather than DynamicProperty, and that would by default by DynamicProperty

* your version management is great

* UI (ItemProviders) has to be adapted (maybe using Adapters) like you say and also to be able to instanciate types specified in the metaconf file. Views are already displayed in this kind of way, and I've tried some hacks on the getChildDescriptors method fueling the newChild command that tells me it is doable.

* the nowadays jwt-views app or future jwt-packager plugin will have to be able to manage metaconf files.

* the only thing that bugs me is that users WILL have to generate a view / metaconf file to be able to use their extension. This would be nice because extended models could still be opened with only the ecore and not the metaconf. * However, maybe that by "displaying and allowing to create" any other Aspect subtype "provided by in EMFs that are found in an autodiscovery directory" that "are not mentioned in the metaconf file" (just like what I advised for views), we could also achieve this goal ? * or maybe we should just add an "export extended model" feature that would zip together the model, the extended ecores and the metaconf file, to be autodiscovered ;)

* a detail : if an extended workflow model is opened but its ecore extension is not found, nowadays JWT believes this to be an error due to the version of the workflow model file. So the version detection / update algorithm would have to be improved so it would be able to display a "missing extension ecore metamodel" in this case.

So what do you think ? Good enough for a try ? I for one have some time this week ^^


PS. about dynamically extending a static class :

First the extended static class can't be interface or abstract (which muddles the model), and secondly the model factory of its container (ex. for extending an Action, ProcessFactory, since it is the factory of Activity that is its container) must be able to handle creation of the extended static class, this way :

org.eclipse.jwt.we.model.processes.ProcessesFactory extensionFactory = new
org.eclipse.jwt.we.model.processes.impl.ProcessesFactoryImpl() {
            public EObject create(EClass eClass)
               if ("LogAction".equals(eClass.getName())) {
                  org.eclipse.jwt.we.model.processes.Action action =
// the following is required to be able to access custom
LogAction features
                  // but it conflicts with Action static subelements ex.
Points -_- so doesn't work
                  return action;
               if ("Point".equals(eClass.getName())) {
               //return super.create(eClass);

Christian Saad a écrit :

I've been looking a bit into the "dynamic properties" issue and I'd like to share some early thoughts and experiences concerning this matter.

First of all, the change of the properties view to a version that supports multiple tabs (as described in view.html)
is quite simple. The "old" properties can be loaded into the view using an adapter, so there is almost no additional effort to this solution (see attachment). Concerning the new dynamic properties, probably a similar approach can be used (write a propertypage that reads the dynamic properties from the itemproviders, creates the corresponding propertyeditors and can be fitted into the propertyview using the same kind of adapter).

To make the actual additions to the JWT metamodel, I've experimented a bit with dynamic EMF. At first, it seemed that possibly the simplest solution to extend existing classes would be to directly inject dynamically created EAttributes or EReferences into the metaobjects because in this case EMF would take care of most issues concerning the management of the new items.

However, I've learned that, sadly, it's not possible to extend existing static metaelements with dynamic structuralfeatures. It is only allowed to create completely new dynamic classes or to subclass existing static classes. Unfortunately this makes the whole process of an almost automatic management of extensible metaelements quite difficult (or maybe I'm just missing an obvious solution).

The alternative would be to create a static metaelement for dynamic properties as described in task 225704 and write code that simulates the behavior of EMF EAttributes or EReferences using instances of this class.

I hope I got the basic idea that is behind this extension mechanism right. To see if this is the case, I drew a small diagram outlining a very concrete implementation of this extension of how I perceived the problem (see attachment). Please correct me if I missed the point or if this would be not viable solution.

The basic idea is that dynamic properties are specified using a small, external meta-model, in which a set of properties is part of a special container element (just like Model is in the JWT metamodel). Sets of dynamic properties can be serialized to and loaded from XMI files by the user or created by Plugins using Eclipse extension points (being converted into the corresponding properties model). This would represent a generic interface to JWT.

Once a set (or multiple sets) of dynamic properties is loaded into JWT, it (they) can be added by the user to a model file. This means that when selecting a ModelElement, the itemprovider searches all sets of activated dynamic properties for properties that were declared for this EClass-type and creates corresponding instances of DynamicPropertyValue. The displaying of the dynamic properties can done by the modified propertypage.






jwt-dev mailing list

jwt-dev mailing list

jwt-dev mailing list


Back to the top