Re: Component resolution question

[Ryan Holmes]

Hi Patrick,

I was referring to the class resolution ambiguity you encountered in  
your Navigation example. Had you specified the 'type' parameter, you  
would be forced to indicate whether you meant 'Navigation' or  
'members/Navigation'.

Specifically, I was talking about the ambiguity in the component  
resolution logic in ComponentAnnotationWorker when you don't specify  
a 'type' parameter. If you look at the ComponentAnnotationWorker  
source, you can see that Method.getReturnType().getSimpleName() is  
used to guess the component type if the 'type' parameter is an empty  
string. It's sort of a "best effort" approach, apparently provided as  
a convenience for those cases where the 'type' parameter is  
completely redundant with your method's return type.

As Andreas mentioned, many components have BaseComponent as their  
class. Obviously, Tapestry cannot resolve a method return type of  
BaseComponent to a specific component. In those cases, the behavior  
is so ambiguous that it's useless.

Even when your component has a specific class, you could conceivably  
have situations where your Tapestry component type differs from the  
return type of your @Component annotated method (although I can't  
honestly think of a use-case that makes good sense, even with a  
component class hierarchy).

Having said all that, 'type' is indeed redundant in many cases and it  
would be great if it there was a better way. In the meantime, I would  
actually prefer to have only one way to specify the type of a  
component (not that I'm really complaining about the current  
"component type guessing" behavior -- I just probably won't use it).  
In short, getting rid of the 'type' parameter or significantly  
changing component resolution logic seems more like a Tap 5 thing and  
I'm more interested in seeing 4.1 stabilize.


Regarding Eclipse refactoring, the 'type' parameter's value does not  
need to be fully qualified to be detected. For example, the type  
parameters in the following @Component methods will be updated  
correctly if you check the "Update textual occurrences in comments  
and strings" box:

  @Component(type = "FlushLiquidSystem", bindings =  
"poolSet=prop:poolSet")
  public abstract FlushLiquidSystem getFlushLiquidSystem();

-- or --

  @Component(type = "directoryThatDoesNotMatchThePackageName/ 
FlushLiquidSystem", bindings = "poolSet=prop:poolSet")
  public abstract FlushLiquidSystem getFlushLiquidSystem();


I'm using Eclipse 3.2 and this works for @InjectPage annotations as  
well. It's obviously not as robust as regular type refactoring and  
you have to pay attention to the preview, but it does make the  
process pretty easy. Let me know if your experience is different,  
because this is an extremely handy feature.

I don't mean to dampen your enthusiasm to make improvements and I  
think everyone has noticed that 'type' feels oddly redundant. It's  
just that it makes sense when you step back and realize that a  
Tapestry component type is not the same as Java type, except when it  
is ;)

-Ryan

On Oct 16, 2006, at 10:22 AM, Patrick Moore wrote:

Hi Ryan --

Eclipse can look for fully specified names but 'members/Navigation'
isn't fully specified.
The component specification looks like:

@Component(type="member/Navigation")
public Navigation getNavigation();

(FYI, in my example I was only giving the full flassname for  
illustration only)

You talk about the ambiguity if the 'type' parameter isn't specified.
What ambiguity are you refering to? I haven't encountered a single
example of such ambiguity yet. The only case I have heard of is the
template-only components. But that isn't an issue for me as those
components I specify anonymously in my html template.

So in my situation, the type parameter is merely a source of run-time
bugs when Tapestry can find a component.

-Pat

On 10/14/06, Ryan Holmes <[EMAIL PROTECTED]> wrote:
Actually, eclipse will pick up @Component 'type' values during
refactoring if you check the "Update textual occurrences in comments
and strings" option when you rename a component class.

The "bug" you illustrate is a good example of why the type parameter
should be required, at least for the time being. If the @Component
annotation worker could resolve class names as you expect, perhaps
making 'type' optional would be nice. Of course, you also wouldn't
need to do :

@Component
public abstract
com.transparentpolitics.core.web.components.members.Navigation
getNavigation();

as it's no different than:

import com.transparentpolitics.core.web.components.members.Navigation
public abstract Navigation getNavigation();

I think there are more important things to be done in Tap 4.1 and I
prefer the predictable behavior of the 'type' parameter (which, as I
mentioned, is refactorable in Eclipse) over the more elegant but
potentially ambiguous behavior that arises when omitting 'type'.

Just my $.02

-Ryan


On Oct 13, 2006, at 10:19 AM, Patrick Moore wrote:

> I would vote just the opposite way. In this case I am returning the
> exact type that is declared on the abstract method. I am not  
returning
> a BaseComponent. I hate the 'type' parameter. It interfers with
> refactoring because eclipse doesn't know it should do anything with
> the type parameter. So far 100% of the times where I had to specify
> the type hasn't needed any clarification. The coding style I am  
using
> means that I never use the .jwc/.page files, only annotations.
>
> The current behavior points out a Tapestry bug as well.
>
> 1. Declare a method:
>
> @Component
> public abstract
> com.transparentpolitics.core.web.components.members.Navigation
> getNavigation();
>
> 2. Declare the classes:
> com.transparentpolitics.core.web.components.members.Navigation
> and
> com.transparentpolitics.core.web.components.Navigation
>
> 3. Declare the component-class-packages:
>    <meta key="org.apache.tapestry.component-class-packages"
> value="com.transparentpolitics.web.components"/>
>
> If Tapestry looks at the
> com.transparentpolitics.core.web.components.members.Navigation  
class
> file, it can see that it has been annotated correctly and it should
> chose the class,
> com.transparentpolitics.core.web.components.members.Navigation, not
> com.transparentpolitics.core.web.components.Navigation.
>
> However, Tapestry gets really wigged out and throws this exception:
>
> Property navigation has already been accounted for by the  
element at
> Annotation @org.apache.tapestry.annotations.Parameter(cache=true,
> defaultValue=, required=true, name=, aliases=) of public abstract
> com.transparentpolitics.web.components.member.Navigation
>  
com.transparentpolitics.web.components.member.NonflowBorder.getNaviga 
t
> ion().
>
> -Pat
>
> On 10/13/06, Norbert Sándor <[EMAIL PROTECTED]> wrote:
>> > If this causes confusions , i'm 100% for making type required
>> again.
>>
>> I would vote a +1 for changing "type" back to required, mainly
>> because of new users.
>>
>> Discarding "type" results in less readable code for example when
>> compared to omitting @InjectObject, which has a more implicit
>> meaning.
>>
>>  >So, in a word, Tapestry cannot use the class name (neither the
>> simple
>> nor the full)
>> This may not be evident for new users, especially when they read
>> in the
>> docs that Tapestry supports pure-java, annotation-only  
components...
>>
>> IMO
>>
>> Regards,
>> Norbi
>>
>> andyhot wrote:
>> > Patrick Moore wrote:
>> >
>> >> To my untrained eye, it looks like the problem is that the
>> >> _componentResolver on line 390 of
>> >> org.apache.tapestry.pageload.PageLoader doesn't have the full
>> class
>> >> name.
>> >>
>> >> On 10/12/06, Patrick Moore <[EMAIL PROTECTED]>  
wrote:
>> >>
>> >>> Hi there --
>> >>>
>> >>> I just shifted over to Tap 4.1.1 and I was hoping I could get
>> rid of
>> >>> the use of 'type' in my @Component annotation. But no such  
luck.
>> >>>
>> >
>> > It's not a matter of luck... You can simply have many components
>> all
>> > sharing the same class. Think for instance all those template- 
only
>> > components...
>> > their class is BaseComponent.
>> >
>> > So, in a word, Tapestry cannot use the class name (neither the
>> simple
>> > nor the full)
>> > in order to make apart a component. It always needs the type.
>> >
>> > type has been made optional to facilitate cases where it matches
>> the
>> > class name.
>> > I believe that's what stated at
>> > http://tapestry.apache.org/tapestry4.1/tapestry-annotations/
>> index.html
>> > If this causes confusions , i'm 100% for making type required
>> again.
>> >
>> >
>> >
>> >>> In my application file I indicate that the components are  
in the
>> >>> 'com.transparentpolitics.web.components' directory (or its
>> >>> subdirectories). However, Tap doesn't find components that  
are in
>> >>> child directories of the 'com ... components' directory. So
>> component
>> >>> references like this:
>> >>>
>> >>>   @Component
>> >>>   public abstract Navigation getNavigation()
>> >>>
>> >>> don't work but this does work :
>> >>>
>> >>>    @Component(type="utils/Navigation")
>> >>>    public abstract Navigation getNavigation()
>> >>>
>> >>> (Navigation is
>> >>> 'com.transparentpolitics.core.web.components.util.Navigation')
>> >>> Now I don't understand why Tap can't find the component as the
>> method
>> >>> call returns the exact component class. Is this just a known
>> temporary
>> >>> limit? Or would changing this current behavior to look at the
>> actual
>> >>> class supplied cause problems?
>> >>>
>> >>> I do know that I can list out each child component directory
>> but that
>> >>> has its own problem as there are some cases of duplicate class
>> names.
>> >>> In any case all the information is on that annotated method  
call.
>> >>>
>> >>> -Pat
>> >>>
>> >>>
>> >>
>>  
---------------------------------------------------------------------
>> >> To unsubscribe, e-mail: [EMAIL PROTECTED]
>> >> For additional commands, e-mail: [EMAIL PROTECTED]
>> >>
>> >>
>> >>
>> >
>> >
>> >
>>
>>
>>  
---------------------------------------------------------------------
>> To unsubscribe, e-mail: [EMAIL PROTECTED]
>> For additional commands, e-mail: [EMAIL PROTECTED]
>>
>>
>
>  
---------------------------------------------------------------------
> To unsubscribe, e-mail: [EMAIL PROTECTED]
> For additional commands, e-mail: [EMAIL PROTECTED]
>




---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]