Re: Stable behavior ids

[Sven Meier]

Hi Edmond,

+1 from me.

Thanks
Sven


On 15.05.20 09:01, Emond Papegaaij wrote:

Hi All,

With Sven's comment in mind and the fact that Andrea did not directly
understand the implications of the new method, I decided that the new
method was indeed not a good idea. It complicates the API and brings
too little benefit. I've updated the PR once more to clarify the
impact and usefulness of getBehaviorId without introducting a new
method. I did make one minor change to the API: you can no longer
request stable ids for temporary behaviors. IMHO that does not make
sense. The behavior will be gone the next request.

Can this now be merged?

Best regards,
Emond

On Mon, May 11, 2020 at 11:12 PM Emond Papegaaij
<emond.papega...@gmail.com> wrote:

Hi Sven,

I agree with you that this method is not very nice. It indeed complicates the 
API even more and it is very hard to explain why it is needed. It however is 
the most efficient implementation I could think of.

There is a compromise we can use: we can freeze all behavior ids when one is 
requested. This can be stored via a flag in the component. The downside is that 
after this flag has been set, it is no longer possible to compact the array of 
behaviors. This can lead to an increased serialization size when behaviors are 
removed from the component. It is a compromise I can live with. The API is 
cleaner and the increase in size is small and only in corner cases. What do you 
think?

Even with this change, we should clarify the docs.

Best regards,
Emond

Op ma 11 mei 2020 17:56 schreef Sven Meier <s...@meiers.net>:

Hi Edmond,

I agree that we should get rid of BehaviorIdList, it's probably the most
inefficient implementation.

But I'm against complicating matters even more with an additional method
- at least until we know how the implementation will be improved post-9.0.

Why would anyone need a non-stable id? How often does this happen?
What if the method returns true, but getId() is never actually called?

IMHO this is the wrong timing to adjust the API.

Have fun
Sven


On 08.05.20 07:59, Emond Papegaaij wrote:

Hi Sven,

Yes, this is all about optimization. It already was. It is not difficult to
keep stable ids for behaviors, you can simply store an id per behavior.
However, this is very inefficient. A long time ago, the decision was made
to only store these ids when needed. Unfortunately, there is no way of
knowing when this is actually needed. Currently, wicket will start
recording these ids when a single id is requested. The getBehaviorId method
has this as a side effect. The method is also called when a behavior that
is not stateless is added to a component.

There are 3 things I don't like about this: first the side effect, second
the way the ids are stored is very inefficient and last the fact that
whether a behavior is stateless or not has nothing to do with its id.
That's why I propose to introduce this new method. It allows me to solve
all 3 cases.

To complicate things even more, there is also stateless ajax, which I would
rather call ajax by luck at the moment. Stateless ajax requests a stable id
and then discards the component storing this id. There is no guarantee at
all that the behavior will get the same id in the next request. This
currently is not explained in the javadoc.

Does this make it more clear? I really feel we should have this ironed out
before the 9 release, even though it has been the way it currently is since
somewhere in 7.x. To me it feels like a loose end in the API and this is
preventing me from improving the code that implements this API.

Best regards,
Emond

Op do 7 mei 2020 22:19 schreef Sven Meier <s...@meiers.net>:

Hi Emond,

I'm not happy with that method.

It's one additional method like many others in Wicket which are not
clear what they are good for. I didn't understand the description.

What's the reason for this stable id hassle? Why can't the id be stable
when a behavior wants it?
Is this really needed for optimization?

Thanks
Sven


On 07.05.20 21:04, Emond Papegaaij wrote:

Hi all,

I've updated my pull request at
https://github.com/apache/wicket/pull/432 . I've added the method I
described and changed the current code to use this new method. Also,
the documentation is updated and clarified to better reflect the rules
a developer has to follow to keep stable ids. This PR should not
change any behavior, but it does allow me to keep on working on my
rewrite without having to break the contract of getBehaviorId. The PR
splits the stable id part from the behavior being stateful or
stateless: a stateless behavior can still request a stable id and a
stateful behavior may not need it. IMHO this change makes it more
clear to a developer what to expect. I would like to merge this into 9
before the release, even though strictly speaking it is not an API
break. What do you think?

Best regards,
Emond

On Thu, May 7, 2020 at 1:58 PM Emond Papegaaij
<emond.papega...@gmail.com> wrote:

Hi Sven and Andrea,

Sven:
All tests still pass, I've also updated the ImmutableBehaviorIdsTest
with some additional checks. However, the tests I've seen so far are
not very exhaustive on this part. Many of them also pass with a bit of
luck :)

Andrea:
What I mean with 'moved to the front' is that a behavior with a stable
id is assigned a low number. For example, when you add these 4
behaviors in this order: class-x, class-y, an ajax link and a special
stateless behavior which uses its id, the ajax link will currently get
id 0. If you then request the id of class-y, it will get id 2 (just
after the special one). In my current implementation, the ajax link
will get id 2 and class-y will get 1 (using the index). I'm planning
to change this to move behaviors to the front when a stable id is
required. The special behavior can override this method to return
true, in which case the id's will be 0 for link, 1 for special, 2 for
class-x, 3 for class-y. The difference is this is that for the last 2,
this id is not guaranteed to be stable over requests. For example,
when class-x is removed somewhere during a request, class-y will get
id 2 on the next request, because the array storing the behaviors is
compacted, removing null-slots.

The default implementation of the new method will simply look at
getStatelessHint on the behavior, stateless means not a stable id. On
ajax behaviors, this can be overridden to return true, even though the
behavior can be stateless. The point is, that for a stateless
behavior, this information is still stored in the component, which may
also be stateless and not being able to store state at all. This is
already the case in the current implementation, and will remain the
same in my implementation. I'm trying to keep things as close to what
it already was, but without having to store the additional
BehaviorIdList.

Emond

On Thu, May 7, 2020 at 1:13 PM Sven Meier <s...@meiers.net> wrote:

Hi,

we have a test for stateless behaviors:

org.apache.wicket.ajax.markup.html.form.StatelessAjaxSubmitLinkTest

We should just make sure it still works.

Have fun
Sven


On 07.05.20 12:13, Andrea Del Bene wrote:

On 07/05/20 10:37, Emond Papegaaij wrote:

Hi Martin,

I know what these id's are for, and this still works as expected.
However, your claim about stable id's on stateless pages currently
does not work as you describe. The id's of behaviors is stored in the
component they are added to, also when the component is stateless. As
you said, a stateless page is discarded at the end of a request,
including the components and the recording of the stable id's. On the
next request, the page is reconstructed, but Wicket cannot guarantee
that the behavior id's are the same. For example, when you
conditionally add a behavior, it influences the id's of the behaviors
added after it. I want the documentation to reflect this, and use

this

to optimize the code. IMHO there's no point in storing id's for
stateless behaviors for the next request. If you absolutely need this
guarantee, you should make your behavior stateful.

One difference between my implementation and the current one is that
behaviors with stable id's are currently moved to the front.

Hi Emond,

I don't perfectly understand what you mean with 'moved to the front'.
Could please explain it?
To be clear about stateless behavior and ids, actually we can have a
"predictable" id for them if we avoid things like conditional adding,
etc... In short we can have stable ids for stateless behaviors but the
responsibility for them is up to the developer and not to the
framework. This is critical to make complex stateless components (for
example AJAX links) properly work. I guess your PR preserve this
condition. And yes, I agree there's no need for storing stateless
behaviors ids.

Andrea.

This
makes the id's a bit more stable. I'm planning to do something
similar, via a new method on Behavior. This should all be API
compatible. I hope this clarifies my intent.

Emond

On Thu, May 7, 2020 at 9:23 AM Martin Grigorov <mgrigo...@apache.org
wrote:

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