Re: Exploring DataObject mixins

From: Andrus Adamchik (andru..bjectstyle.org)
Date: Tue Nov 02 2010 - 01:03:57 UTC

  • Next message: Andrus Adamchik: "3.1 MySQL testing parameters"

    I sort of want to keep mixins out of the Modeler until the use patterns become more clear, so I kind of got stuck on how we can cgen the extra properties outside of it. I like the idea of annotated annotation, just wish we could use it in cgen somehow...

    Another thought (a bit unrelated) ... per repeating discussions on the user list, many people (me too actually) would want to have a simple way to access an object diff when processing a lifecycle callback for this object. Cayenne has this information of course, and its a shame it is not easily accessible. Moreover sometimes the diff is needed AFTER the commit, when ObjectContext already disposed of it. So we may provide a mixin handler "context" containing such diff data. This is not even mixin-specific, I just feel reluctant to add another API layer to the core, and a mixin module may provide a sandbox for experimentation.

    Andrus

    On Nov 1, 2010, at 7:49 PM, Robert Zeigler wrote:

    > Interesting. Except for the bit with having to implement interfaces + cover methods; you've now "polluted" your code with mixin details. All-in-all, a nice idea, though. I wonder if we could incorporate mixins (custom or otherwise?) directly into the modeler + class generation so that the interface + cover methods are auto-generated into the DO superclass? Not sure about this, just "thinking out loud". Maybe if the DO mixins could, themselves, be annotated, like:
    >
    >..ayenneMixin(interfaceClass=Referenceable.class, template="SomeFile.txt")//both interfaceClass and template would be optional... template would contain some blurb that the superclass generator "mixes in" to superclass templates.
    > public..nterface ReferenceableMixin {
    > }
    >
    > Now you potentially have a way to scan for mixins/gain additional information about them... anyway, just considering ways to write less duplicate code. :)
    >
    > Robert
    >
    > On Nov 1, 2010, at 11/16:27 PM , Andrus Adamchik wrote:
    >
    >> Been thinking about work-related design issues, and came up with an idea of DataObject "mixins". So what are the problems:
    >>
    >> 1. Often you'd like to associate a certain set of common properties and/or behavior with a group of persistent objects that are unrelated and not a part of the same inheritance hierarchy. Here is an abstracted real-life example:
    >>
    >> * entities A, B, C are "referenceable" (they have a public UUID)
    >> * entities A, C, D are "auditable" (when a user changes their data in some way, a system must record the change history)
    >> * entities A, C and E are "access-controlled" (a certain permission level is required for a user to view or edit them).
    >>
    >> 2. In a layered architecture, DataObject itself may not be the right place to implement complex behavior. All non-data methods should ideally be pluggable (normally meaning must be IoC-driven), but still attached to a specific DataObject.
    >>
    >> In other words there's no multiple inheritance in Java, and anyways we want to push most of the logic to the higher app layers. So... I am experimenting with something I called "mixins", based on DataObject dynamic nature + listener capabilities. I checked in to github a simple mixin extension for Cayenne:
    >>
    >> http://github.com/andrus/cayenne-mixin/tree/master/other/cayenne-mixin/
    >>
    >> and a proto-CMS using it:
    >>
    >> http://github.com/andrus/cayenne-mixin/tree/master/oc/
    >>
    >> I may actually keep developing the CMS piece for my own needs, but here a very basic version of it is used for a mixin demo. A mixin in the simplest form is just a custom annotation placed on a DataObject:
    >>
    >>..eferenceableMixin
    >>..uditableMixin
    >> public class Article extends _Article { }
    >>
    >> It is up to the application to attach lifecycle handlers for a given type of mixin. MixinHandlerManager class from cayenne-mixin module provides API to bind such handlers:
    >>
    >> MixinHandlerManager handlerManager = new MixinHandlerManager(entityResolver);
    >> handlerManager.addMixinHandler(handler);
    >>
    >> A handler implements MixinHandler interface and does whatever is needed to the object during its lifecycle. Below is a mixin handler that calculates and injects a UUID property in a referenceable object:
    >>
    >> class ReferenceableMixinHandler implements MixinHandler<ReferenceableMixin> {
    >>
    >> ..verride
    >> public Class<ReferenceableMixin> getMixinType() {
    >> return ReferenceableMixin.class;
    >> }
    >>
    >> ..verride
    >> public void setupListeners(LifecycleCallbackRegistry registry,
    >> Class<? extends DataObject> entityType) {
    >>
    >> registry.addListener(LifecycleEvent.POST_PERSIST, entityType, this,
    >> "initUuidCallback");
    >> registry.addListener(LifecycleEvent.POST_LOAD, entityType, this,
    >> "initUuidCallback");
    >> }
    >>
    >> void initUuidCallback(DataObject object) {
    >> int id = DataObjectUtils.intPKForObject(object);
    >> String uuid = object.getObjectId().getEntityName() + ":" + id;
    >> object.writePropertyDirectly(Referenceable.UUID_PROPERTY, uuid);
    >> }
    >> }
    >>
    >> Of course any DataObject can store any transient property, so we are taking advantage of that. Similarly AuditableMixinHandler is used to create ContentVersion records when an object changes. I am still exploring various uses of mixins and ways to extract common handler patterns in cayenne-mixin. A few early observations:
    >>
    >> * Mixins provide a better way to organize and understand listeners. From experience, ad-hoc mapping of listeners quickly results in a mess - each listener maps to more than 1 entity and more than 1 type of events. Very hard to remember and manage that stuff. I'd rather not think about event types at all, and just think that e.g. a listener manages 'uuid' property for a set of referenceable objects.
    >>
    >> * Mixins may be a better way for inheritance mapping. I am still to explore this idea, but I suspect in many cases mixins may be more scaleable than e.g. vertical inheritance. Besides of course they allow for many "superclasses" at once.
    >>
    >> * Mixins may provide a facility for custom relationship faulting. E.g. an injected property can be a lazy collection backed by a custom query (maybe try overriding standard Cayenne relationships this way?)
    >>
    >> * There are some limitations, most obviously all mixin properties have to be accessed via DataObject.readProperty(..) which may not fly well with scripting (e.g. "article.uuid"). This can be solved by an optional mixin interface and manual cover methods. Not ideal, but works:
    >>
    >> public class Article extends _Article implements Referenceable {
    >>
    >> ..verride
    >> public String getUuid() {
    >> return (String) readProperty(Referenceable.UUID_PROPERTY);
    >> }
    >> }
    >>
    >>
    >> Anyways, this is something to explore. Mixins solve a whole class of design problems for me, and I am sure we can find other uses.
    >>
    >> Cheers,
    >> Andrus
    >>
    >>
    >>
    >
    >



    This archive was generated by hypermail 2.0.0 : Tue Nov 02 2010 - 01:04:25 UTC