2010/3/19 Alexander Smirnov <asmirnov@exadel.com>

I've thought about these problems for a long time too. Your proposal has
a disadvantage because per-row state is quite different from the
component state. Take a look for the UIInput implementation:
localValueSet field is never saved in the state while it should be vary
from row to row. The same thing happens to the UIForm.submitted
attribute, that has been eventually added to the UIData spec instead of
common solution, that would save/restore all local variables. My
proposal was a separate RowStateHolder interface with
save/restoreRowState methods there component developer can preserve
internal variables. In the your stenario, there should be some flag that
tells component to save all values, even these that are not supposed to
be preserved between requests.

On 03/19/2010 10:27 AM, Leonardo Uribe wrote:
> In short, this topic is an attempt to add full state to components
> inside UIData. I'll do a brief resume, so people can understand this one
> easily.
>
> UIData uses the same component instances to render multiple rows.
> Suppose this example:
>
> <h:dataTable id="cities" var="city" value="#{country.cities}">
> <h:column>
> <h:outputText value="#{city}" />
> </h:column>
> </h:dataTable>
>
> In the component tree it is created this hierarchy:
>
> HtmlDatatable
> UIColumn
> HtmlOutputText
>
> If we have 10 cities, the same component is used over and over to render
> all 10 cities. The reason to do that in this way is keep state as small
> as possible.
>
> Now let's suppose something like this:
>
> <h:dataTable id="cities" var="city" value="#{country.cities}">
> <h:column>
> <h:inputText value="#{city}" />
> </h:column>
> </h:dataTable>
>
> It was changed the output component for an input one. If this table is
> in a form and the values are submitted, the same component is used to
> apply request values, validate and apply them to the model (update
> process). To make this possible, UIData class has some code to preserve
> this values between phases (using EditableValueHolder interface), so
> when each phase is processed, all rows are traversed and you get the
> expected behavior.
>
> Now suppose something more complex: We have a code that use
> invokeOnComponent to change the style of my inputText. In theory, only
> one row should change. But in fact, all rows are rendered with the same
> color. Why? because we use the same component to render all rows, and we
> don't preserve the children component state between rows.
>
> There is a lot of issues, questions, and side effects related to this
> issue, but just to put some of them here:
>
> TOMAHAWK-1062 InputTextArea doesn't work properly inside facet DetailStamp
> TOMAHAWK-96 Data table Scroller not working the dataTable which was
> actually contained in other DataTable
> https://javaserverfaces-spec-public.dev.java.net/issues/show_bug.cgi?id=153
> Problems with UIData state saving
>
> Also, it is well know that one reason why people uses c:forEach in
> facelets, is because this one create "full" components per each row. It
> is very easy to find articles on internet.
>
> Now, with jsf 2.0 we have partial state saving, so we have a chance to
> fix this one once and for all. I tried fix this one per months (maybe
> years!), but talking with Martin Marinschek on JSFDays, some ideas came
> out and finally it was found a possibility to fix this one using the
> existing api and with little changes on the spec.
>
> The proposal is this:
>
> 1. Do not call UIComponent.markInitialState() on
> ComponentTagHandlerDelegate, as ComponentHandler javadoc says, instead
> call it after PostAddToViewEvent are published on vdl.buildView(). We
> need to call it from root to nodes, so the parent component is marked
> first. I know the place where this call comes is from trinidad tag
> handler, but this call needs to be fixed in a more predictable way.
> 2. Use an attribute on facesContext to identify when the VDL is marking
> the initial state (in myfaces there is already an attribute called
> "org.apache.myfaces.MARK_INITIAL_STATE"). This is necessary to indicate
> UIData instances that it is time to save the full state of all component
> children,
> 3. Allow UIData to hold a map where the key are client ids and the value
> are the deltas of all components per row. This map should be saved and
> restored.
> 4. Change UIData.setRowIndex() to restore and save the component state.
>
> I'll attach a patch on myfaces issue tracker with the algorithm proposed
> (because it is based on myfaces codebase). It was tested and it works.
> But note it is necessary to fix the javadoc for
> UIData.markInitialState(), ComponentHandler and maybe vdl.buildView(),
> so the intention is propose this change for jsf 2.0 rev A. Note this
> works only with PSS enabled because without it we don't have a place to
> notify UIData instances that it is necessary to get the full state.
> Also, note this patch preserve backward compatibility, because the old
> way to store/save is applied after the full state is restored.
>
> Really, I have the strong temptation to apply some similar code on
> myfaces UIRepeat component (because this class is private), but I prefer
> first ask to EG to know what you guys think about it.
>
> Suggestions are welcome,
>
> regards,
>
> Leonardo Uribe
>