using xsd:any for extensible schema - java

Until now, I've been handling extensions by defining a placeholder element that has "name" and "value" attributes as shown in the below example
<root>
<typed-content>
...
</typed-content>
<extension name="var1" value="val1"/>
<extension name="var2" value="val2"/>
....
</root>
I am now planning to switch to using xsd:any. I'd appreciate if you can help me choose th best approach
What is the value add of xsd:any over my previous approach if I specify processContents="strict"
Can a EAI/ESB tool/library execute XPATH expressions against the arbitrary elements I return
I see various binding tools treating this separately while generating the binding code. Is this this the same case if I include a namespace="http://mynamespace" and provide the schema for the "http://mynamespace" during code gen time?
Is this WS-I compliant?
Are there any gotchas that I am missing?
Thank you

Using <xsd:any processContents="strict"> gives people the ability to add extensions to their XML instance documents without changing the original schema. This is the critical benefit it gives you.
Yes. tools than manipulate the instances don't care what the schema looks like, it's the instance documents they look at. To them, it doesn't really matter if you use <xsd:any> or not.
Binding tools generally don't handle <xsd:any> very elegantly. This is understandable, since they have no information about what it could contain, so they'll usually give you an untyped placeholder. It's up the the application code to handle that at runtime. JAXB is particular (the RI, at least) makes a bit of a fist of it, but it's workable.
Yes. It's perfectly good XML Schema practice, and all valid XML Schema are supported by WS-I
<xsd:any> makes life a bit harder on the programmer, due to the untyped nature of the bindings, but if you need to support arbitrary extension points, this is the way to do it. However, if your extensions are well-defined, and do not change, then it may not be worth the irritation factor.

Regarding point 3
Binding tools generally don't handle
very elegantly. This is
understandable, since they have no
information about what it could
contain, so they'll usually give you
an untyped placeholder. It's up the
the application code to handle that at
runtime. JAXB is particular (the RI,
at least) makes a bit of a fist of it,
but it's workable.
This corresponds to the #XmlAnyElement annotation in JAXB. The behaviour is as follows:
#XmlAnyElement - Keep All as DOM Nodes
If you annotate a property with this annotation the corresponding portion of the XML document will be kept as DOM nodes.
#XMLAnyElement(lax=true) - Convert Known Elements to Domain Objects
By setting lax=true, if JAXB has a root type corresponding to that QName then it will convert that chunk to a domain object.
http://bdoughan.blogspot.com/2010/08/using-xmlanyelement-to-build-generic.html

Related

How to use XSD for non namespaced documents

Recently I've encountered a service that returns its results in XML, in sort of following fashion
<event>
<event-header>
...
</event-header>
<event-body>
...
</event-body>
</event>
Notice that the document does not have a namespace definition. As a result, there is no "official" schema that I can use.
I have written a schema definition that I can use to generate classes that are usable in code to interact with equivalent elements in the document. From observation I can tell that the document format does not change (field order remains the same, fields are not introduced or go away). But question stands, can I still deserialize the provided document using my schema? As far as I know, schemas must define a namespace, and in theory the documents above and below
<event xmlns="http://saltyjuice.lt/dragas/event-service/1.0/event-schema.xsd">
<event-header>
...
</event-header>
<event-body>
...
</event-body>
</event>
are not equivalent.
For reference, I'm using stax and woodstox 6 as implementation.
You can have a schema for a no-namespace document, I don't know why you thought otherwise. It's not ideal, because a namespace can guide people to the right schema. But it's allowed. Anyway, even with a namespace, it's quite possible to have several schemas for the same namespace (usually, versions and variants).

How to prevent JAXB from writing unused namespaces during marshalling

Has someone ever been able to remove unused namespaces during marshal of an object using JAXB? Here is a link of the requested feature: https://github.com/javaee/jaxb-v2/issues/103 (see description)
Is there a property for configuring JAXB for this?
Has this been fixed in MOXy?
I am currently traversing the object that needs to be marshalled and extract all classes that need to be bound into Class[] classesToBeBound.
Then I create a new JAXBContext.newInstance(classesToBeBound)
The unused namespaces now are not included in the XML.
I know that xml validation is valid even with the unused namespaces but to me this is something a framework should handle.
The following link https://blogs.oracle.com/enterprisetechtips/entry/customizing_jaxb mentions various fixed (see middle of the text somewhere) but when trying to find the solution in those links either the link is broken or no-one really solved it.
Any comments are welcome.
(EDIT)
Plain text:
GIVEN
a new instance of JAXBContext and add 2 classes with each a separate namespace.
WHEN
marshalling a class that has these 2 classes as a property but only 1 of them is not null
THEN
I expect only the namespace of the property that is not null to be visible in the XML.
BUT the ACTUAL is
that both namespaces are in the xml.
So my question was how can I remove or tell JAXB to NOT write the unused namespaces?
To put it in java-code:
GIVEN
public class Foo{
private Bar bar; //namespace something2
private User user; //namespace user
}
WHEN
JAXBContext c = JAXBContext.newInstance(Foo.class, Bar.class, User.class);
...
Foo foo = new Foo();
foo.setBar(null);
foo.setUser(new User("Bob"));
marshaller.umarshal(foo);
THEN I expect the xml to be
<foo xmlns="something1" xmlns:user="user">
<user:name>Bob</user:name>
</foo>
BUT the ACTUAL is (note the something2 namespace)
<foo xmlns="something1" xmlns:user="user" xmlns:bar="something2">
<user:name>Bob</user:name>
</foo>
Of course this is a simplified example and our specification of a type has around 30 different namespaces.
As far as I know, this is indeed not possible in JAXB - and is actually a well-known issue. As you have noticed it, the list of produced namespaces are the ones that have been registered in your JAXBContext, and not the ones that are effectively used when marshalling :-(
I the past, I used the same workaround as you (identify the various used classes and narrow the JAXBContext to this limited set of classes).
Another typical workaround is a 2-step processing: a first marshalling with JAXB, followed by a XSLT transformation to get rid of let's says "polluting" namespaces.
This may not be possible as while marshaling of this objects hierarchy happen, at the time of creating root tag, information about which objects are null v/s not null may not be available. Any attempt to get this information in advance may also have side-effects associated with it as respective accessor methods are invoked. Hence JAXB will statically use info from JAXBContext to have this information populated.
You can try using a different javax.xml.bind.Marshaller implementation.
For example org.eclipse.persistence.jaxb.JAXBMarshaller implementation deals well with this case and remove all unnecessary namespaces when marshall the object.
To do so you need to do the next steps:
Add eclipselink-2.6.5.jar to the classpath in order to use the org.eclipse.persistence.jaxb.JAXBMarshaller. If you're using gradle you can add compile 'org.eclipse.persistence:eclipselink:2.6.5' to your dependencies.
Create a jaxb.properties file in the same package where you've the objects to marshall (following the example in your question - JAXBContext c = JAXBContext.newInstance(Foo.class, Bar.class, User.class);, in the package of one of these classes Foo, Bar or User).
In the jaxb.properties file, add the follow property which specify the desired Context factory:
javax.xml.bind.context.factory=org.eclipse.persistence.jaxb.JAXBContextFactory
Doing this, the org.eclipse.persistence.jaxb.JAXBMarshaller will be used as a javax.xml.bind.Marshaller implementation on Runtime. And then no unnecessary namespaces will appear when you marshall the objects.
I tried the solution albciff suggested in this thread and it turns out that Eclipse Moxy handles this much better than the reference implementation (org.glassfish.jaxb).
Here's info on how to switch to the Moxy implemenation of JAXB:
https://wiki.eclipse.org/EclipseLink/Examples/MOXy/JAXB/SpecifyRuntime
The documentation doesn't specify it but you can also change the jaxb implemenation with just a single configuration file instead of a jaxb.properties in each package where your jaxb annotated classes exists. Just create a file META-INF/services/javax.xml.bind.JAXBContext (yep unconventional filename) with the contents:
org.eclipse.persistence.jaxb.JAXBContextFactory
This makes the jaxb ContextFinder use the Eclipse Moxy implementation for all jaxb marshalling in the jvm.
Another option is to use a system property -Djavax.xml.bind.context.factory=org.eclipse.persistence.jaxb.JAXBContextFactory
Disclaimer though: The namespaces declaration aren't minimized/optimized for the current payload, but at least it doesn't include ALL namespaces that are part of the jaxb grammar. The integration I'm working on went from a staggering 700+ declared namespaces (about 60KB worth of useless overhead per sent message) to at best 3 declarations in a message. Though for messages which have a lot of different types in them, all of the namespaces which are valid in that particular message is declared. That means in some case I still get ~30 declared namespaces when only one would suffice for the current payload.
I guess that SOAP isn't the way to go if you need to optimize on bandwidth.
Yes, they could be omitted. I'm not sure I understood the problem you face correctly. But there is no problem to marshal an object without namespaces.
Try something like this marshaller.setProperty(Marshaller.JAXB_NO_NAMESPACE_SCHEMA_LOCATION, "class for which namepsace not needed"); in your case it should be
marshaller.setProperty(Marshaller.JAXB_NO_NAMESPACE_SCHEMA_LOCATION, bar.class);

How to make JAXB recognize #XmlElement(default='something') annotation parameters?

Having no luck in generating XML with fixed element values using JAXB 2.1 RI.
We're generating XML bound code using xjc and marshall the results.
It works for attributes using <xs:attribute fixed='something'/> and the JAXB customization property fixedAttributeAsConstantProperty.
For elements we figured that there was no way to do the same. Or is there?
As a workaround, we used <xs:element default='something'/> which is turned into #XmlElement(default='something'). Now my guess is that you can tell the marshaller somehow to interpret the default parameter and generate an element with the corresponding content, like <element>something</element>.
I've looked at the standard and RI vendor specific marshaller configuration properties without finding something useful.
There seems to be an xjc plug-in that does something similar[1], but I'd be kinda surprised if there is no standard JAXB way to do so.
Any pointers are much appreciated, thanks.
[1] http://fisheye5.cenqua.com/browse/~raw,r=1.5/jaxb2-commons/www/default-value/index.html
The XML schema specification deals with unmarshalling only, i.e. determining the post validation infoset of the source XML representation.
JAXB appears to mirror this in its support for defaults, i.e. for reading mostly. For instance, the support for default attributes does not extend to writing defaults, just to reading (or accessing from Java using the getter)
As I see it, the best bet for a solution is to extend the default value plugin to handle fixed elements as well, by making the field final and removing the setter.
The code for the plugin is very straightforward, really!
There is an maven-plugin one can use to generate constants: https://github.com/fbdo/jaxb2-fixed-value

Duplicate namespace declarations in JAXB generated XML

I am using JAXB to generate XML from Java objects, it's a realtime, quite high message rate application and works fine most of the time. However occassionally and without any obvious clues as to why, I am getting duplicate namespace declarations in the generated XML. eg:
<UpdateRequest xmlns="http://xml.mycomp.com/ns/myservice"
xmlns="http://xml.mycomp.com/ns/myservice">
<field1>value</field1>
...
</UpdateRequest>
Has anyone seen this behaviour before?
Check if the xsd code of this class allow the creation of more than 1 instance of the repeated attribute. if so, you can avoid this repetitions setting the number of instances of the xmlns attribute for each UpdateRequest object.
If the problem is your code (maybe there is being created this attribute twice) and you have limited the number of instances of the attribute (as i said above), the program will show an error at runtime complaining that you are trying to insert an attribute already defined.
A solution might be available at this link.
here's the relevant section quoted verbatim from the above link that may be relevant for you:
Similar explicit inclusion of a schema
type in an instance document's element
occurs if you instantiate a JAXB
element using an object of some
(abstract) XML schema base type so
that the element would have the
element tag of the base type.
Second, avoid xs:anySimpleType since
this will also create multiple
references to the namespaces bound to
xsi and xs, and type attributes
containing the actual type. And you
lose JAXB's advantage of having typed
fields in your Java classes so that
you lose all the checks the Java
compiler might do, and for
unmarshalling you'll have to handle
all the conversions yourself.

JAVA: Build XML document using XPath expressions

I know this isn't really what XPath is for but if I have a HashMap of XPath expressions to values how would I go about building an XML document. I've found dom-4j's
DocumentHelper.makeElement(branch, xpath) except it is incapable of creating attributes or indexing. Surely a library exists that can do this?
Map xMap = new HashMap();
xMap.put("root/entity/#att", "fooattrib");
xMap.put("root/array[0]/ele/#att", "barattrib");
xMap.put("root/array[0]/ele", "barelement");
xMap.put("root/array[1]/ele", "zoobelement");
would result in:
<root>
<entity att="fooattrib"/>
<array><ele att="barattrib">barelement</ele></array>
<array><ele>zoobelement</ele></array>
</root>
I looked for something similar a few years ago - a sort of writeable XPath. In the end, having not found anything, I hacked up something which essentially built up the XML document by adding new nodes to parent expressions:
parent="/" element="root"
parent="/root" element="entity"
parent="/root/entity" attribute="att" value="fooattrib"
parent="/root" element="array"
parent="/root" element="ele" text="barelement"
(This was itself to be governed by an XML configuration file, hence the appearance of above.)
It would be tempting to try an automate some of this to just take the last path element, and make something of it, but I always felt that there were XPath expressions I could write which such a dumbheaded approach would get wrong.
Another approach I considered, though did not implement (the above was "good enough"), was to use the excellent Jaxen to generate elements that did not exist, on the fly if it didn't already exist.
From the Jaxen FAQ:
The only thing required is an implementation of the interface org.jaxen.Navigator. Not all of the interface is required, and a default implementation, in the form of org.jaxen.DefaultNavigator is also provided.
The DOMWriterNavigator would wrap and existing DOMNavigator, and then use the makeElement method if the element didn't exist. However, even with this approach,
you'd probably have to do some pre/post processing of the XPath query for things like attributes and text() functions.
The best I was able to come up with is to use a JAXB implementation, which will marshall/unmarshal objects to xml and then I used Dozer (http://dozer.sourceforge.net/documentation/mapbackedproperty.html) to map the xpaths which were keys in a map to the JAXB object method setters.
<mapping type="one-way" map-id="TC1">
<class-a>java.util.Map</class-a>
<class-b>org.example.Foo</class-b>
<field>
<a key="root/entity/#att">this</a>
<b>Foo.entity.att</b>
<a-hint>java.lang.String</a-hint>
</field>
It's more of a two step solution, but really worked for me.
I also wanted same kind of requirement where nature is so dynamic and dont want to use XSLT or any object mapping frameworks, so i've implemented this code in java and written blog on it please visit,
http://ganesh-kandisa.blogspot.com/2013/08/dynamic-xml-transformation-in-java.html
or fork code at git repository,
https://github.com/TheGanesh/DynamicXMLTransformer

Categories