Hi Emond,

The behavior id is something internal to Wicket.
It is used *by* Wicket to find the requested behavior in a Component.
For example if a Component has more than one Ajax behaviors (yes, Ajax
behaviors could be stateless too since 7.4.0. Before that there was a
project in WicketStuff) it uses the id (extracted from the special
parameter in the url) to find which behavior exactly should be invoked.
In case of stateful page the ids are stored with page and later
deserialized.
In case of stateless page (with stateless Ajax behaviors) the whole page is
discarded at the end of request 1 but at request 2 a new page is created
from scratch with all its components and their behaviors. Here the id
*must* be the same as in request 1, otherwise Wicket may execute another
behavior's onRequest(). So, the ids must be stable in case of stateless as
well.



On Wed, May 6, 2020 at 9:28 PM Emond Papegaaij <emond.papega...@gmail.com>
wrote:

> Hi,
>
> While running, I came up with a solution for which only a minor change
> to the contract of the method is needed: ids for stateless behaviors
> will be stable within a single requests, but can change over requests.
> I think this is reasonable, given the way stateless components work.
>
> I would like to change the documentation to read:
> Gets a stable id for the specified behavior. The id remains stable
> from the point this method is first called for the behavior until the
> behavior has been removed from the component. For {@linkplain
> Behavior#getStatelessHint(Component) stateful} behaviors, this stable
> id is retained over requests. For stateless behaviors, the id can
> change between requests.
>
> Emond
>
> On Wed, May 6, 2020 at 5:54 PM Emond Papegaaij
> <emond.papega...@gmail.com> wrote:
> >
> > Hi Andrea,
> >
> > Stateful is fine, stateless is not. A component cannot keep a stable
> > id when it's stateless. I'm trying out a change that forbids
> > getBehaviorId() for stateless behaviors, and I'm hitting a few tests
> > with stateless ajax. I don't see how this is supposed to work anyway.
> > We request the component to store the id of a certain behavior, but it
> > must do so in a stateless way. IMHO that's impossible. The component
> > will be discarded, along with its state and the stored id at the end
> > of the request. No way of guaranteeing that the same component with
> > the same behavior at the same index will exist at the next request.
> >
> > This brings me back to my suggested change in the documentation: only
> > stateful behaviors have guaranteed stable ids. You can request the id
> > of a stateless behavior, but (in my current implementation) it may
> > change when you remove behaviors from the component.
> >
> > Emond
> >
> > On Wed, May 6, 2020 at 5:41 PM Andrea Del Bene <an.delb...@gmail.com>
> wrote:
> > >
> > >
> > > On 06/05/20 16:52, Emond Papegaaij wrote:
> > > > Hi all,
> > > >
> > > > During my refactoring of the component state (WICKET-6774) I noticed
> > > > that behavior ids are currently stored in a very inefficient way: an
> > > > ArrayList is added to Component data to store references to behaviors
> > > > with a stable id. On my branch I have eliminated this ArrayList,
> > > > greatly reducing the size of components with stateful behaviors (such
> > > > as AjaxLink).
> > > >
> > > > A behavior gets a stable id when it is stateful to be able to render
> > > > this id in an URL. However, at the moment, it also gets a stable id
> > > > when Component.getBehaviorId is called for that particular behavior.
> > > > This is also documented in the method's javadoc. Do we really need
> > > > this last part? It complicates the code a lot. In our code base nor
> in
> > > > Wicket can I find a single place where this is actually used.
> > >
> > > Actually I see that Component.getBehaviorId is used in
> > > AbstractAjaxTimerBehavior.getTimerId() and
> > > AbstractDefaultAjaxBehavior.onBind() which are however stateful
> behaviors.
> > >
> > > > I would like to suggest a change in the javadoc to state that stable
> > > > ids are only guaranteed for stateful behaviors and change this in
> > > > Wicket 9. The actual change in the implementation is not yet finished
> > > > and does not need to ship in 9.0.0, but feel I cannot change the
> > > > contract of a method in a minor release.
> > > >
> > > > Best regards,
> > > > Emond
>

Reply via email to