On 8 Oct 2010, at 13:36, Bernhard Hoisl wrote: > > Very interesting comment and I would be the first to use a "channel" metaphor > for routing messages to specific widgets. As some of you know we here at WU > Vienna have worked some time ago on IWC for exactly this kind of purpose - > namely to notify all widgets in a certain user space (we are using widgets in > Elgg, therefore all widgets displayed in a dashboard of one user). > > As I gave the code a review just recently I did come up with a new solution > wiring together widget instances sharing the same sharedDataKey and apiKey > (excerpt of new code attached). We are using the code right now because it > fulfills exactly our purpose, but if someone comes up with a generic solution > I would be even happier. > > It is by now not entirely clear where to best put the code to: Notifier, > WidgetAPIImpl or another class because all have advantages and disadvantages. > The Notifier class (as proposed by Scott Wilson) would be a good fit, but > code there should only notify widgets and not set data. I not only want to > notify other widgets of events happening in one user space but also share the > data with all of them, therefore I have to call > PropertiesController.updateSharedDataEntry() as well to set the data for all > instances. > > Therefore, I adjusted WidgetAPIImpl.setSharedDataForKey() and > appendSharedDataForKey() accordingly to notify and set data for all widgets > having the same sharedDatakey and apiKey as the widget instance data was > initially set. >
OK I got it - so you propagate the shared data update to all widgets (of any type) that have the same shareddatakey and API key. That then cascades the notifications to their siblings. > It is somehow a hack, because if you want to make it generic you should give > config variables for the sibling attributes, for example in the > local.widgetserver.properties. But anyway I liked the idea of configuring > message buses using these sibling attributes because multiple combinations > could be useful. It certainly has the benefit of simplicity, and looks like it works nicely for the UC. I do wonder though if this kind of UC can also be handled purely client-side using things like PostMessage and open-app[1]? I think Mattias Palmer was doing something with this for a widget portal project he was working on. > I just wanted to share my code with you guys and maybe hear some comments. Thanks! Always appreciated! > > BTW: Quite hard to find out was how you actually one can delete shared data. > By setting a variable to null, but there is no actual method for doing this > (only a hack in PropertiesController.updateSharedDataEntry() exists, which is > a not very consistent logic). Maybe a method deleteSharedDataEntry() would be > useful? There is also a reset() method that hasn't been implemented yet that clears all the shared state [2]. > @Scott: I completely re-designed the former approach of notifying widgets, > which now does not rely on attributes set in the ScriptSession anymore. > Therefore, your code comment at the end of Notifier.java is somehow obsolete > and if needed could be dramatically simplified (or completely deleted). > Moreover, I tried to change as less as possible at the existing code of > Wookie, therefore no changes in wookie-wrapper.js or dwr.xml or a new > DWRSessionRemote.js is needed for that (as was in the former version). Great bit of refactoring there! I'll leave the comment in Notifier just in case we need it later, but your new solution looks much simpler - as you say the trick is to see if we can generalize it a bit and maybe add some other levels of control over the "channels" that are enabled by the user. Maybe for now we might use your patch but have it deactivated by default and have a switch in widgetserver.properties to enable it? > > > Best, > Bernhard [1] http://code.google.com/p/open-app/ [2] https://issues.apache.org/jira/browse/WOOKIE-110 > > > > On 07/10/10 11:44, Scott Wilson (JIRA) wrote: >> >> [ >> https://issues.apache.org/jira/browse/WOOKIE-133?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12918864#action_12918864 >> ] >> >> Scott Wilson commented on WOOKIE-133: >> ------------------------------------- >> >> Thanks, Ivan. Your table is my number one source of info on IWC! >> >> >> >>> Implement inter-widget messaging >>> -------------------------------- >>> >>> Key: WOOKIE-133 >>> URL: https://issues.apache.org/jira/browse/WOOKIE-133 >>> Project: Wookie >>> Issue Type: New Feature >>> Components: Server, Wookie REST API >>> Reporter: Scott Wilson >>> Priority: Minor >>> Original Estimate: 168h >>> Remaining Estimate: 168h >>> >>> One of the more persistent new feature requests we've had for Wookie has >>> been to extend the mechanisms for inter-widget communication (IWC). As it >>> currently stands, Wookie supports two mechanisms for IWC: >>> 1. Wave Shared States >>> Wookie enables "sibling" widgets to share their state. This is exposed to >>> widgets using the Google Wave Gadget API, which enables widgets to submit >>> state update deltas, and to register a callback to notify them when their >>> state has been updated by another widget instance. We define siblings using >>> the algorithm in org.apache.wookie.util.SiblingPageNormalizer; essentially >>> this is that the widget instance must be for the same widget, with the same >>> shared data key, and the same API key. >>> 2. HTML5 Drag and Drop >>> While not actually part of Wookie itself, Widgets can be developed using >>> HTML5 drag and drop capability, enabling user-directed IWC. >>> Some use-cases for IWC have been collected on the Talk About Widgets >>> mailing list. >>> The most common use-cases for extending IWC proposed for Wookie I've seen >>> involve "Dashboard"-style messaging. In this model, widgets appearing in >>> the same space for the user get to share events. For example, all the >>> widgets in a single user's dashboard can send data to each other. This >>> might follow a single shared state model, but is more likely to follow a >>> "channel" metaphor, with named "channels" or "queues" between widgets. >>> There are several ways this could be implemented, for example: >>> A. User-specified channels >>> In this model, the user specifies exactly which messages are sent between >>> widgets. For example, the EzWeb project defines a "wiring" interface with >>> "slots" and "events" connected with user-created "channels". For example, >>> the user create a channel from a widget with a "weather" event to a widget >>> with a "weather" slot. >>> B. Widget-specified channels >>> In this model, widgets are automatically able to receive any events on any >>> channel that they choose to listen to. Users do not need to create channels >>> to enable IWC. >>> These two mechanisms are not necessarily exclusive, and could be supported >>> within the same Feature extension. For example, the JavaScript API may look >>> like this: >>> iwc.sendEvent(String event_name, Object event_value) >>> iwc.registerCallback(String slot_name, Function callback) >>> I would prefer having this IWC extension use its own functions in its own >>> object rather than overload the Wave and Widget objects, to avoid any >>> possible confusion. >>> In case "A", callbacks are only triggered where there are explicit channels >>> linking the sending event and the receiving slot; in case "B", events are >>> propagated to all registered callbacks that the event_name matches. >>> For example, in pseudocode, the implementation could look something like: >>> sendEvent(event_name, value, idkey){ >>> instance = find widget instance (idkey) >>> if use_channels: >>> event = find event(instance.getWidget, event_name) >>> find channels (eventinstance = instance& event = >>> event_name) >>> for channel in channels: >>> slot = channel.slot >>> send notification to slot (target, slot, value) >>> >>> Notifier.notifySingleInstance(target,"iwc.__callback({slot},{value})") >>> else: >>> >>> Notifier.notifySiblingsByUser(instance,"iwc.__callback({event_name},{value})") >>> } >>> Note that in case "A", the widget author just registers callbacks for its >>> slots without having to be concerned with what the sending widget calls >>> them, so if a user can wire a "weather" event to a "temperature" slot, the >>> sending widget calls: >>> sendEvent("weather","30C"); >>> and the receiver can call: >>> registerCallback("temperature",my_function); >>> ... and the channel wires things up despite the names not matching. In case >>> B, the receiving widget would have to know the name of the event, and >>> register a callback. >>> In case A, widgets need to declare their "events" and "slots" as extensions >>> in config.xml so they can be wired up in channels by a user, and there >>> needs to be some sort of UI where users get to do some wiring. Wookie could >>> expose an API for creating/removing/editing channels that could be >>> implemented by containers rather than provide this UI itself. >>> (Interestingly, case A would in theory support situations where widgets >>> send notifications to widgets in different containers; though I think this >>> would only really work if we went did the OpenID implementation so could >>> have some assurance that the source and target widget instances belonged to >>> the same user). >>> [1] http://groups.google.com/group/talk-about-widgets/web/use-cases-for-iwc >>> [2] http://forge.morfeo-project.org/wiki/index.php/Gadget_development_guide >> > <WidgetAPIImpl.java>
