That is an excellent point. I did an initial implementation of this and found that the world did not work well. I had thought that there where other users of the constants that made presumptions that failed ... IE they didn't juse use buildTopic. It could also have been the issue you are pointing out ... although presumably Eclipse would have rebuilt all required files.
I have not had a chance to spend more time on it.... Likely because I have not had to do any event stuff yet. I know, a pretty poor excuse. Having to build a parallel structure makes me cringe ... but as you point out it might be the only way forward if we need to be compatible with previously compiled .class files. And it does save me having to look through all references to these constants to find out which ones where using them directly and causing problems. Do we have to be compatible with previously compiled .class files ... I'm assuming the answer is going to be yes here. [email protected] Brian de Alwis <[email protected]> Sent by: [email protected] 11/09/2011 06:50 PM Please respond to E4 Project developer mailing list <[email protected]> To E4 Project developer mailing list <[email protected]> cc Subject Re: [e4-dev] Suggestion to restructure UIEvents to increase clarity and performance I was just writing some event code, and again cringed at the buildTopic(), and remembered Dean's email on simplifying the approach. I still think this is a worthy change. I wondered about backwards compatibility — if we changed the attribute definitions to incorporate the topic directly, such as UIEvents.UIElement. VISIBLE, are we guaranteed that all existing compiled classes that referenced UIEvents.UIElement.VISIBLE would have had the string inlined? Or should we just define a parallel set of strings (e.g., EMF-esque like UIEvents.UIElement_VISIBLE). Brian. On 22-Sep-2011, at 3:30 PM, Dean Roberts wrote: While trying to understand UIEvents and how the topics are built up for subscribe operations I found the buildTopic() methods a little cumbersome and a barrier to understanding what was really going on. I would like to make a suggestion that I think would 1) Make the code look cleaner 2) Make the concepts clearer 3) Improve run time performance (albeit marginally) For illustration purposes, here is an example of an interface definition from UIEvents for the UIElement model object generated by GenTopic public static interface UIElement { public static final String TOPIC = UIModelTopicBase + "/ui/UIElement"; //$NON-NLS-1$ public static final String ACCESSIBILITYPHRASE = "accessibilityPhrase"; //$NON-NLS-1$ public static final String CONTAINERDATA = "containerData" ; //$NON-NLS-1$ public static final String CURSHAREDREF = "curSharedRef"; //$NON-NLS-1$ public static final String ONTOP = "onTop"; //$NON-NLS-1$ public static final String PARENT = "parent"; //$NON-NLS-1$ public static final String RENDERER = "renderer"; //$NON-NLS-1$ public static final String TOBERENDERED = "toBeRendered"; //$NON-NLS-1$ public static final String VISIBLE = "visible"; //$NON-NLS-1$ public static final String VISIBLEWHEN = "visibleWhen"; //$NON-NLS-1$ public static final String WIDGET = "widget"; //$NON-NLS-1$ } To subscribe to an event that tracks a UIElements visible change you would write eventBroker.subscribe(UIEvents.buildTopic(UIEvents.UIElement.TOPIC, UIEvents.UIElement.VISIBLE), visibilityHandler); To subscribe to all changes to UIElement you would write eventBroker.subscribe(UIEvents.buildTopic(UIEvents.UIElement.TOPIC), visibilityHandler); I think it would be much nicer to write, for the first case, eventBroker.subscribe(UIEvents.UIElement.VISIBLE), visibilityHandler); or, for the second case, eventBroker.subscribe(UIEvents.UIElement.ALL), visibilityHandler); To make this possible I propose that we would generate the UIElement interface definition like this. public static interface UIElement { public static final String ALL= UIModelTopicBase + "/ui/UIElement/*"; //$NON-NLS-1$ public static final String VISIBLE = UIModelTopicBase + "/ui/UIElement/" + "visible"; //$NON-NLS-1$ ... } This would allow us to delete two of the buildTopic() methods. We could leave the 3 argument version of buildTopic() to retain construction of topics that filter based on event type ( although nobody calls this internally at this time so do we really need it?). The code would also be more efficient as we would be removing method calls that do string concatenation with simple static references. In fact, I believe the compiler will just in line the references at compile time. Although I would be a fan of simply making these changes and requiring consumers to make a few small simplifying changes, we could provide a stepping stone by regenerating the new shape, deprecating the buildTopic() methods and all the TOPIC fields. We would then change the implementation of the buildTopic methods to return the appropriate constant directly without concatenating. Admittedly there is higher priority work to be done but since this class is essentially API, if we don't change it now, we will never be able to make the change. And I think the clarity in the subscribe statements make the change worth it. This change seems to fit well with the Eclipse 4 mantra that simple things should be simple. I also believe the effort required to change GenTopic to generate the new shape is small, and I volunteer to create the patch. Admittedly changing the calling code to use the new shape is more wide spread, but a very simple task. Although an actual committer probably wants to do that work as it would be a LOT of patches to apply :-) What does the community think? Does it make sense? Am I totally off base? [email protected] _______________________________________________ e4-dev mailing list [email protected] https://dev.eclipse.org/mailman/listinfo/e4-dev _______________________________________________ e4-dev mailing list [email protected] https://dev.eclipse.org/mailman/listinfo/e4-dev
_______________________________________________ e4-dev mailing list [email protected] https://dev.eclipse.org/mailman/listinfo/e4-dev
