| Re: [e4-dev] Declarative UI |
Hi,See my comments below ...On Wed, Nov 12, 2008 at 11:49 AM, Tom Schindl <tom.schindl@xxxxxxxxxxxxxxx> wrote:
Hi,
Well I envision a design for client-server with a declarative model like
this.
---------------------------------------------
Client
[1] Client-Device (Browser, ...)
[2] Client-Renderer (Widget-DOM) -----action--
[3] Client-DOM (EMF-DOM) |
--------------------------------------------- |
Network /\ |
|| Model-To-Model (e.g. CDO, JSON, ...) |
--------------------------------------------- |
Server |
[4] EMF-DOM |
[5] BusinessLogic <--------------------------|
Do you have a RAP like design in mind when you talk about server side
rendering?Yes. As far as I know other frameworks do the same. JFace for example. They maintain an in-memory representation of the tree of UI components.Because RAP supports the SWT API I think it needs to do it as well, except if RAP would translate Java to _javascript_ in the same way as GWT does.GWT is what I would call a "client side rendering framework", because it does not have a representation of the UI component tree on the server. In fact GWT applications are _javascript_ applications that happen to use RCP calls to get and change data on the server.I can see how a declarative ui description could be more easily translated into code that runs on the client, but you still will need code to handle events, construct and add new UI elements dynamically at runtime, validate user input and soon ... If you don't translate Java to _javascript_ (or whatever your client side language is), you will need to run that code on the server. Because of that you probably also need a representation of the tree of UI components on the server as well.This is really not what you want for a highly interactive Web 2.0 style application. You will end up with more roundtrips and more state on the server, which will limit your performance.Regards,MarkusI don't know enough about RAP but as far as I have understood
it they have 2 representations of a widget (ServerSide,ClientSide) maybe
they could change their (under the assumption noone is directly
accessing widgets but instead is going through the DOM) design so that
they don't need the server side representation.
EMF's object representation is by the way not as bloated as an XML-DOM
and can be optimized further (Ed recently blogged about it). Naturally
if you have to restore [2] also on the server you are doubling memory
but I don't think this is needed (always under the assumption noone is
directly accessing widgets but instead is going through the DOM).
Tom
Markus Kohler schrieb:
> Hi all,
> Maybe I don't fully understand everything here, but what does it mean
> for memory usage if only the DOM API is available?
> Something like an XML DOM Tree typically has a big overhead compared to
> a dedicated statically typed UI Component Tree.
>
> As long as the the DOM is stored client side (like it's done with
> browser based applications), I don't see a big problem with this
> approach. But if e4 is going to support server side rendering, where the
> DOM would be in memory for each user, this could create scalability
> problems, because of high memory usage per user.
>
> I wonder whether saving some memory for the code does help, if using a
> DOM representation requires more memory.
>
>
> Just my 2 cents,
> Markus
>
> On Wed, Nov 12, 2008 at 9:03 AM, Tom Schindl
> <mailto:tom.schindl@xxxxxxxxxxxxxxx>>*> <tom.schindl@xxxxxxxxxxxxxxx <mailto:tom.schindl@xxxxxxxxxxxxxxx>> wrote:
>
> Hi Kevin,
>
> Can you clarify this? I don't see a problem with going through the DOM.
> I'd argue that someone is not even getting access anymore to the REALLY
> widget when he/she is using declarative UI.
>
> You may ask why does one not have access to the widget:
> a) Did you ever see a WebDeveloper trying to access the real
> Native-Gecko-Button? SWT is just a renderer of the widget
> like Gecko, ... is in a browser.
>
> b) The DOM you are seeing and the Widget-Hierarchy don't necessary have
> a one-to-one mapping. E.g. let's say I want to have fancy border
> drawing in SWT then my DOM looks like this:
>
> <composite border="black 2px" layout="grid">
> <label text="BlaBla 1" />
> <label text="BlaBla 2" />
> </composite>
>
> whereas the widget structure is like this:
>
> Composite (FillLayout=margin=2, backgroundColor=black)
> Composite(GridLayout)
> Label
> Label
>
> c) What do you script against? The real widget is not right IMHO because
> of 2 and 3
>
> The above example of a border is the reason I'm not a friend anymore of
> a low-level dialect like XSWT but favor some higher level of Declarative
> UI. As long as I programm against the high-level DOM it doesn't bother
> me how a feature gets implemented at the widget-level
> (widget-composition, gc-drawing, ....). Naturally the above border could
> get implemented also by subclassing Composite + PaintListener but I
> guess you'll get the point, right? By the way how is such a problem
> solved when the border definition is coming from an CSS-Stylesheet?
>
> When we talk about modeling the workbench it even gets worse because
> e.g. a StackedPart can get rendered by a completely different
> widget-type for example:
> - CTabFolder
> - TabFolder
> - Nebula-PShelf
> - ...
>
>
> Tom
>
> Kevin McGuire schrieb:
> >
> > Actually this approach concerns me: if I go through the model, or go
> > directly against SWT, I get different notification. It could easily
> > lead to bugs in application code. I would also argue that the
> model is
> > then no longer a model of SWT, but rather, a model of an enhanced SWT
> > which doesn't exist.
> >
> > Regards,
> > Kevin
> >
> >
> >
> >
> > *Tom Schindl <tom.schindl@xxxxxxxxxxxxxxx
> > Sent by: e4-dev-bounces@xxxxxxxxxxx> <mailto:e4-dev-bounces@xxxxxxxxxxx>
> >> <mailto:e4-dev@xxxxxxxxxxx>>
> > 11/11/2008 04:29 AM
> > Please respond to
> > E4 Project developer mailing list <e4-dev@xxxxxxxxxxx
> >
> >
> >
> > To
> > e4-dev@xxxxxxxxxxx <mailto:e4-dev@xxxxxxxxxxx>
> > e4-dev@xxxxxxxxxxx <mailto:e4-dev@xxxxxxxxxxx>> > cc
> >
> > Subject
> > Re: [e4-dev] Declarative UI
> >
> >
> >
> >
> >
> >
> >
> >
> > Angelo zerr schrieb:
> >> Hi,
> >>
> >> I would like speak about UFace <http://code.google.com/p/uface/> that
> >> I'm using into TK-UI <http://tk-ui.sourceforge.net/index.html> and I
> >> find very helfull to manage Declarative UI. With TK-UI you can
> describe
> >> UI with any XML Markup (XUL, XHTML,...) and render it into any
> renderer
> >> (SWT, SWT Forms, Swing...) Today I'm using UFace which is UI facade
> >> which provide a lot of think like Databinding (based on JFace
> >> Databinding). I like UFace because anything can be bounded :
> >>
> >> * UI widgets : any properties of (visible, text... properties if UI
> >> widgets). For instance if you call setVisible, it fire event so it's
> >> possible after to observe this property (with SWT we cannot catch
> this
> >> visble event for instance).
> >
> > This is a very interesting point Angelo is raising here. I think
> when we
> > describe our UI and directly interface with SWT people HAVE TO program
> > against the DOM (whether it is EMF or anything else) and not directly
> > against the widget because changes in the UI like setVisible(),
> > setEnabled(), ... can't be synced back live to the declarative model
> > simply because SWT is not informing us about these changes with
> events.
> >
> > I don't see this as a problem though. By the way in "Modeling the
> > Workbench" we identified only a hand full of things where we need to
> > sync back from the Widget (e.g. Resize, Iconify, ...) to the model. We
> > assume that *all* other operations are happening on model and are then
> > reflected in the UI.
> >
> > Tom
> >
> > --
> > B e s t S o l u t i o n . a t EDV
> Systemhaus GmbH
> >
> ------------------------------------------------------------------------
> > tom schindl leiter
> softwareentwicklung/CSO
> >
> ------------------------------------------------------------------------
> > eduard-bodem-gasse 8/3 A-6020 innsbruck phone ++43 512
> 935834
> > _______________________________________________
> > e4-dev mailing list
> > https://dev.eclipse.org/mailman/listinfo/e4-dev> > e4-dev@xxxxxxxxxxx <mailto:e4-dev@xxxxxxxxxxx>
> >
> >
> >
> ------------------------------------------------------------------------
> >
> > _______________________________________________
> > e4-dev mailing list
> > https://dev.eclipse.org/mailman/listinfo/e4-dev> e4-dev@xxxxxxxxxxx <mailto:e4-dev@xxxxxxxxxxx>
>
>
> --
> B e s t S o l u t i o n . a t EDV Systemhaus GmbH
> ------------------------------------------------------------------------
> tom schindl leiter softwareentwicklung/CSO
> ------------------------------------------------------------------------
> eduard-bodem-gasse 8/3 A-6020 innsbruck phone ++43 512 935834
> _______________________________________________
> e4-dev mailing list
> https://dev.eclipse.org/mailman/listinfo/e4-dev
>
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> e4-dev mailing list
> e4-dev@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/e4-dev
--
B e s t S o l u t i o n . a t EDV Systemhaus GmbH
------------------------------------------------------------------------
tom schindl leiter softwareentwicklung/CSO
------------------------------------------------------------------------
eduard-bodem-gasse 8/3 A-6020 innsbruck phone ++43 512 935834
_______________________________________________
e4-dev mailing list
e4-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/e4-dev
_______________________________________________
e4-dev mailing list
e4-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/e4-dev