So after several days of debugging, we were eventually able to reproduce some strange interactions between composite components, ui:repeat, p:remoteCommand and partial state saving in JSF that we do not understand.
Scenario
A composite component iterates over a list of objects using ui:repeat. During each iteration, another composite component is included and arguments are passed.
<ui:composition (...)>
<ui:repeat var="myVar" value="#{cc.attrs.controller.someList}">
<namespace:myRemoteCommand someParam="SomeParam"/>
In the included composite component, there is an auto-run p:remoteCommand calling a method using parameters defined in the component's interface.
<ui:component (...)>
<p:remoteCommand actionListener="#{someBean.someMethod(cc.attrs.someParam)}"
autoRun="true"
async="true"
global="false">
However, when setting a breakpoint in someMethod(...), an empty string is passed. This only happens if partial state saving is set to false.
Solutions
We tried several solutions and the following ones appear to work (however we do not understand why and cannot foresee any further problems that could occur):
We can set partial state saving to true.
We can change the composite component pattern to ui:include.
We can remove one or both of the composite components and directly include the content instead.
Question
Why does JSF behave this way? What is this interaction between composite component, ui:repeat and argument passing that changes depending on whether we use ui:include / partial state saving or not?
We're using Primefaces 5.3, Glassfish 4.1, Mojarra 2.2.12, Java 8.
Your code is all fine. It's just that Mojarra's <ui:repeat> is broken. You're not the first one facing a state management related problem with <ui:repeat>.
Checkbox inside ui:repeat not refreshed by Ajax
Dynamically added input field in ui:repeat is not processed during form submit
Components are with the same id inside ui:repeat
<h:form> within <ui:repeat> not entirely working, only the last <h:form> is processed
Composite component with custom backing component breaks strangely when nested inside ui:repeat
ui:repeat in o:tree not working as expected
Root cause of your problem is that #{cc} is nowhere available at the moment the <ui:repeat> needs to visit the tree. Effectively, the <ui:repeat value> is null. A quick work around is to explicitly push the #{cc} in UIRepeat#visitTree() method. Given Mojarra 2.2.12, add below lines right before line 734 with pushComponentToEL(facesContext, null).
UIComponent compositeParent = getCompositeComponentParent(this);
if (compositeParent != null) {
compositeParent.pushComponentToEL(facesContext, null);
}
And add below lines right after line 767 with popComponentFromEL(facesContext).
if (compositeParent != null) {
compositeParent.popComponentFromEL(facesContext);
}
If you don't build Mojarra from source, copy the entire source code of UIRepeat into your project, maintaining its package structure and apply above changes on it. Classes in /WEB-INF/classes have higher classloading precedence than those in /WEB-INF/lib and server's /lib. I have at least created issue 4162 to address this.
An alternative is to replace Mojarra by MyFaces, or to replace the <ui:repeat> by an UIData based component which got state management right such as <h:dataTable> or <p:dataList>.
<p:dataList type="none" var="myVar" value="#{cc.attrs.controller.someList}">
<namespace:myRemoteCommand someParam="SomeParam" />
</p:dataList>
You might only want to apply some CSS to get rid of widget style (border and such), but that's trivial.
See also:
Should PARTIAL_STATE_SAVING be set to false?
Related
I have a question about the idea behind the fact, that only UIForm got the attribute prependId. Why is the attribute not specified in the NamingContainer interface? You will now probably say that's because of backward compability but I would preferre breaking the compability and let users which implement that interface, also implement methods for the prependId thing.
The main problem from my perspective about the prependId in the UIForm component is, that it will break findComponent()
I would expect that if I use prependId, then the NamingContainer behaviour would change, not only related to rendering but also when wanting to search for components in the component tree.
Here a simple example:
<h:form id="test" prependId="false">
<h:panelGroup id="group"/>
</h:form>
Now when i want to get the panelGroup component I would expect to pass the string "group" to the method findComponent(), but it won't find anything, I have to use "test:group" instead.
The concrete problem with that is, when using ajax with prependId="false". The ajax tag expects in the attributes update and process, that the values care of naming containers. It's a bit strange that when I use prependId="false" that I have to specify the full id or path, but okay.
<h:form id="test" prependId="false">
<h:panelGroup id="group"/>
</h:form>
<h:form id="test1" prependId="false">
<h:commandButton value="go">
<f:ajax render="test:group"/>
</h:commandButton>
</h:form>
Well this code will render without problems but it won't update the panelGroup because it cannot find it. The PartialViewContext will contain only the id "group" as element of the renderIds. I don't know if this is expected, probably it is but I don't know the code. Now we come to the point where the method findComponent() can not find the component because the expression passed as parameter is "group" where the method would expect "test:group" to find the component.
One solution is to write your own findComponent() which is the way I chose to deal with this problem. In this method i handle a component which is a NamingContainer and has the property prependId set to false like a normal UIComponent. I will have to do that for every UIComponent which offers a prependId attribute and that is bad. Reflection will help to get around the static definition of types but it's still not a really clean solution.
The other way would be introducing the prependId attribute in the NamingContainer interface and change the behaviour of findComponent() to work like described above.
The last proposed solution would be changing the behaviour of the ajax tag to pass the whole id, but this would only solve the ajax issue and not the programmatic issues behind the findComponent() implementation.
What do you think about that and why the hell is it implemented like that? I can't be the first having this problem, but I wasn't able to find related topics?!
Indeed, UIComponent#findComponent() as done by <f:ajax render> fails when using <h:form prependId="false">. This problem is known and is a "Won't fix": JSF spec issue 573.
In my humble opinion, they should never have added the prependId attribute to the UIForm during the JSF 1.2 ages. It was merely done to keep j_security_check users happy who would like to use a JSF form with JSF input components for that (j_security_check requires exact input field names j_username and j_password which couldn't be modified by configuration). But they didn't exactly realize that during JSF 1.2 another improvement was introduced which enables you to just keep using <form> for that instead of sticking to <h:form>. And then CSS/jQuery purists start abusing prependId="false" to avoid escaping the separator character : in their poorly chosen CSS selectors.
Just don't use prependId="false", ever.
For j_security_check, just use <form> or the new Servlet 3.0 HttpServletRequest#login(). See also Performing user authentication in Java EE / JSF using j_security_check.
For CSS selectors, in case you absolutely need an ID selector (and thus not a more reusable class selector), simply wrap the component of interest in a plain HTML <div> or <span>.
See also:
How to select JSF components using jQuery?
How to use JSF generated HTML element ID with colon ":" in CSS selectors?
By default, JSF generates unusable ids, which are incompatible with css part of web standards
I'm currently migrating a web application from JSF 1.2/Richfaces 3.3.3 to JSF 2.2 For data iteration, we used a4j:repeat from Richfaces. I now want to change the iterators to ui:repeat, because we want to throw out Richfaces.
However, I came across a very strange behaviour. Imagine a code snippet like this (simplified from the original):
<ui:repeat id="criterions" var="criterion" value="#{AdvancedSearchBean.criterionList}">
<h:panelGroup rendered="#{criterion.searchCriterion.displayType == 'PERSON'}">
<h:inputText value="#{criterion.searchString}"/>
</h:panelGroup>
</ui:repeat>
The part inside the panelGroup is evaluated, although the rendered condition definitely evaluates to false. If I change ui:repeat to a4j:repeat, it works fine, the part inside the panelGroup is NOT evaluated.
This is a real problem for our code, as the "criterion" variable can contain different objects (extending the same superclass). In this case, the criterion object does not contain a property with the name "searchString" (because it is not of type "PERSON") -> an error is thrown.
Can anyone explain this behaviour or has a solution?
I'm using the JSF version integrated in WildFly 8.0.0.final (Mojarra 2.2.5-jbossorg-3)
Thanks
Markus
Is this legal?
<h:form id="status${a.myID}" >
// ...
</h:form>
where 'a' is an object in a backing bean. It seems to sort of work, but when I look at the rendered HTML, I see the id as: :0:status for example, instead of :status0 as I would expect. My main problem is trying to reference the id from <f:ajax render=.... I'm getting "contains an unknown id..." with pretty much every combination I can think of. Is it possible to set ids using values from a backing bean reliably?
The single-letter variable name ${a} and the symptom of an iteration index like :0 being auto-appended in client ID, suggests that this <h:form> is inside a JSF iterating component such as <h:dataTable> or <ui:repeat> with a var="a" which actually is not a backing bean. It would confirm and explain all symptoms described so far. If ${a} were a real backing bean (and not an iteration variable), then it would have "worked" and you would have seen :0:status0, :1:status0, :2:status0, etc — whose usefulness is questionable though.
First of all, the id attribute of a JSF component is evaluated and set during view build time, the moment when the JSF component tree is to be built based on XHTML source code file. The var attribue of a JSF iterating component is set during view render time, the moment when the HTML output is to be generated based on JSF component tree. Thus, logical consequence is, the object set by var attribute is not available at the moment the id attribute needs to be set and thus evaluates to null/empty-string. The effect is exactly the same as when you would do
<h:form id="status">
JSF iterating components namely already auto-appends the iteration index to the generated client ID. It would not make any sense to manually set the ID of the iterated item in the component ID. There's namely physically only one <h:form> component in the JSF component tree which is in turn reused multiple times during producing the HTML output based on the current iteration round.
This Q&A should also give more food for thought: JSTL in JSF2 Facelets... makes sense?
Coming back to your concrete functional requirement of referencing a component in <f:ajax render>, you definitely need to solve this differently. Unfortunately you didn't clearly describe the context of the source component and the target component, so it's impossible to give/explain the right client ID and so I'm just giving you a link so that you can figure it out on your own: How to find out client ID of component for ajax update/render? Cannot find component with expression "foo" referenced from "bar"
Unrelated to the concrete problem, the old JSP EL style ${...} has in Facelets exactly the same effect as #{...}. In order to avoid confusion by yourself and your future maintainers it's recommend to completely ban usage of ${...} and stick to #{...} all the time. See also Difference between JSP EL, JSF EL and Unified EL
Actually ${a.myID} this is not rendering any output. As you are getting :0:status as form ID which implies, :0 is parent of :status in HTML tree structure.
Using JSF 2.0, I need to display a table wherein each row contains a link which opens a popup. I have two models: A which has id and List<B> properties and B which has id and name properties. In my backing bean, I have a List<A> property. In my view, I am using <ui:repeat> to iterate over List<A>.
The requirement is, depending on the row that the user clicks, the corresponding List<B> of A needs to be displayed. However, the <ui:repeat> does not accept a nested list to be assigned in the var attribute. Hence, I need to do a lot of workarounds which is not efficient.
How do I efficiently solve this problem?
What you need is to nest another <ui:repeat> tag in your outer iteration:
<ui:repeat value="#{bean.listOfA}" var="a">
...
<ui:repeat value="#{a.listOfB}" var="b">
...
</ui:repeat>
</ui:repeat>
The only thing left that is worth noting is that nested <ui:repeat> tags used to have problems with state management until Mojarra 2.1.15 version (details in jsf listener not called inside nested ui:repeat and in many not so recent questions and their answers), which could result in action listeners not called, etc. but if you're currently on the latest Mojarra JSF implementation - just skip this part altogether.
I searched everywhere but could not find a solution to this. I am trying to used
required=yes to validate whether a value is present or not. I am using it inside inputtext.
The problem is it does not work inside a datatable. If I put the text box outside the datatable it works. I am using JSF 1.7 so I don't have the validateRequired tag from JSF 2.0.
I even used a validator class but it is still not working. Does anyone know why does required=yes or validator='validationClass' inside a inputtext inside a datatable is not working.
I appreciate the help.
Thanks.
First of all, the proper attribute values of the required attribute are the boolean values true or false, not a string value of Yes. It's an attribute which accepts a boolean expression.
The following are proper usage examples:
<h:inputText required="true" />
<h:inputText required="#{bean.booleanValue}" />
<h:inputText required="#{bean.stringValue == 'Yes'}" />
As to the problem that it doesn't work inside a <h:dataTable>, that can happen when the datamodel is not been preserved properly (the datamodel is whatever the table retrieves in its value attribute). That can in turn happen when the managed bean is request scoped and doesn't prepare the datamodel during its (post)construction which causes that the datamodel is null or empty while JSF is about to gather, convert and validate the submitted values.
You need to ensure that the datamodel is exactly the same during the apply request values phase of the form submit request as it was during the render response phase of the initial request to display the form with the table. An easy quick test is to put the bean in the session scope. If that fixes the problem, then you definitely need to rewrite the datamodel preserving logic. You could also use Tomahawk's <t:saveState> or <t:dataTable preserveDataModel="true"> to store the datamodel in the view scope (like as JSF2's new view scope is doing).
Finally, JSF 1.7 doesn't exist. Perhaps you mean JSF 1.2?