src/site/apt/roadmap.apt

~~ Licensed to the Apache Software Foundation (ASF) under one
~~ or more contributor license agreements. See the NOTICE file
~~ distributed with this work for additional information
~~ regarding copyright ownership. The ASF licenses this file
~~ to you under the Apache License, Version 2.0 (the
~~ "License"); you may not use this file except in compliance
~~ with the License. You may obtain a copy of the License at
~~
~~ http://www.apache.org/licenses/LICENSE-2.0
~~
~~ Unless required by applicable law or agreed to in writing,
~~ software distributed under the License is distributed on an
~~ "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
~~ KIND, either express or implied. See the License for the
~~ specific language governing permissions and limitations
~~ under the License.

        -------------------
         Axiom 1.3 roadmap
        -------------------

Introduction

  This page summarizes the planned changes for the next major release, i.e. Axiom 1.3.
  Note that it is not intended as a wish list for new features, but identifies a set of
  changes that break backward compatibility and therefore need to
  be postponed to the 1.3 release.

  The overall goals for the 1.3 are:
  
   * Upgrade the API to use Java 5 features, in particular generics. Make Java 6 the minimum
     supported version and get rid of the dependencies on Activation and the StAX API (which
     are included in the JRE in Java 6).
   
   * Eliminate deprecated APIs and utility classes.
   
   * Eliminate remaining API inconsistencies.
   
   * Make the API more compact by clarifying the separation between the public API
     and implementation classes and moving implementation classes out of <<<axiom-api>>>.
   
API inconsistencies to be eliminated

* Usage of <<<Object>>> instead of <<<DataHandler>>>

  There are several APIs that use <<<Object>>> although they expect or return
  a <<<DataHandler>>>. This is probably a legacy of Axis 1.x where the Activation API
  was an optional dependency. Today this no longer makes sense:
  
   * The Activation API is included in Java 6.
   
   * It is unlikely that Axiom actually works if Activation is not in the class path
     because there is nothing in the build that enforces or tests that and there are no
     known downstream projects that use Axiom without also depending on the Activation API.
  
  The following APIs will be changed to use <<<DataHandler>>> directly:
  
   * <<<OMText#getDataHandler()>>>
   
   * <<<OMFactory#createOMText(Object, boolean)>>>

* Methods declared by the wrong interface in the node type hierarchy

  Some methods are declared at the wrong level in the node type hierarchy so that they may
  be called on nodes for which they are not meaningful:
  
   * The <<<serialize>>> and <<<serializeAndConsume>>> methods that take an <<<OutputStream>>>
     or <<<Writer>>> object are only meaningful for <<<OMContainer>>> objects. However, they
     are declared by <<<OMNode>>>. Note that already in 1.2.x, these methods are deprecated.
   
   * <<<OMContainer>>> declares several methods that return child elements by name: <<<getChildrenWithLocalName>>>,
     <<<getChildrenWithName>>>, <<<getChildrenWithNamespaceURI>>> and <<<getFirstChildWithName>>>.
     Since the document element is unique, these methods are not meaningful for <<<OMDocument>>>
     and they should be declared by <<<OMElement>>> instead.

* Exception hierarchy

  The way exceptions are used in Axiom 1.2.x is not very consistent. In addition it doesn't allow application
  code to distinguish between different types of error cases. This should be improved in
  Axiom 1.3 to meet the following requirements:
  
   * Although the two implementations of the Axiom API (LLOM and DOOM) use StAX as XML parser and serializer, this
     is not a strict requirement. The only strict requirements are that the implementation is able to
     construct a tree from a <<<XMLStreamReader>>> supplied by application code, is able to serialize
     a tree to an <<<XMLStreamWriter>>> and is able to construct an <<<XMLStreamReader>>> from a tree.
     Therefore methods defined by the Axiom API should only declare <<<XMLStreamException>>> if they interact
     directly with a StAX object supplied by application code.
     
     A good example is the <<<OMContainer#serialize(OutputStream)>>> method. In Axiom 1.2.x this method
     is declared to throw <<<XMLStreamException>>>. This assumes that the implementation necessarily uses
     StAX as XML serializer. This is a wrong assumption, because the implementation is free to use its
     own serializer. Also note that the method doesn't declare <<<IOException>>> which would be much more relevant.
   
   * Axiom should have well-defined (and distinct) exceptions for at least the following two error cases:
   
     * An I/O error occurs during a deferred parsing operation. In that case, the unchecked exception should
       wrap the original <<<IOException>>> so that it can be extracted by application code.
     
     * A parser error occurs during a deferred parsing operation.
   
   * <<<OMException>>> should only be thrown by APIs in the <<<org.apache.axiom.om>>> and <<<org.apache.axiom.soap>>>
     packages (and their subpackages). However, in Axiom 1.2.x that exception is used by the MIME/attachments API
     as well.

* Methods for boolean attributes in <<<SOAPHeaderBlock>>>

  <<<SOAPHeaderBlock>>> has methods to get and set the boolean attributes <<<mustUnderstand>>>
  and <<<relay>>>. However, the set of getters and setters is not the same for both attributes:
  there is a <<<setMustUnderstand(String)>>> method, but no corresponding <<<setRelay(String)>>> method.
  Probably the <<<setMustUnderstand(String)>>> method should just be removed.

Removal of unnecessary or irrelevant APIs

  This section identifies APIs that have become unnecessary or irrelevant. Note that APIs that
  have already been deprecated in 1.2.x (and will be removed in 1.3 anyway) are not listed here.

* <<<org.apache.axiom.om.OMAttachmentAccessor>>>

  In Axiom versions prior to 1.2.9, the sequence of events produced by the <<<XMLStreamReader>>>
  returned by <<<getXMLStreamReaderWithoutCaching>>> was inconsistent if the underlying stream is
  XOP encoded (see AXIOM-255). This made it necessary for application code to (partially) handle the
  XOP processing itself. That is the reason why the <<<XMLStreamReader>>> instance (which is
  an <<<OMStAXWrapper>>>) implements <<<OMAttachmentAccessor>>>: it allowed the application code
  to retrieve the binary content for undecoded <<<xop:Include>>> elements.
  
  Starting with Axiom 1.2.9 the sequence of events produced by the
  <<<XMLStreamReader>>> is always the same (i.e. XOP is always completely decoded) and all binary
  content is reported through the <<<DataHandlerReader>>> extension, making the <<<OMAttachmentAccessor>>>
  API unnecessary in that context.
  
  The interface is used in several other contexts as well, but it is actually unnecessary in each of them:
  
   * The <<<Attachments>>> class implements <<<OMAttachmentAccessor>>>. However, it is not possible to
     substitute an <<<Attachments>>> object by an arbitrary <<<OMAttachmentAccessor>>> instance (e.g. when
     parsing an XOP/MTOM message). This would actually be a useful feature, but in that particular context,
     there is an overlap with the <<<MimePartProvider>>> interface and that API would be better suited.
   
   * The <<<OMAttachmentAccessor>>> API is required in conjunction with the <<<setInlineMTOM>>> method
     defined by <<<OMXMLStreamReader>>>. However, that method is deprecated.
   
   * The <<<XOPBuilder>>> interface extends <<<OMAttachmentAccessor>>>. However, as explained below,
     that API will be removed in Axiom 1.3.

* <<<org.apache.axiom.om.impl.builder.BuilderAwareReader>>>

  This API can't be preserved as is in Axiom 1.3. There are two reasons for this:
  
   * The only method defined by the <<<BuilderAwareReader>>> interface has a
     parameter of type <<<StAXBuilder>>>, but that class will be removed from
     <<<axiom-api>>> (see below). The parameter type would have to be changed to
     <<<OMXMLParserWrapper>>>.
   
   * It the interface is considered a public API, then it is placed in the
     wrong package.
  
  TODO: analyze the use cases for this API to see if we still need it 

* <<<org.apache.axiom.om.impl.builder.XOPBuilder>>>

  The <<<XOPBuilder>>> interface is implemented by <<<XOPAwareStAXOMBuilder>>> and <<<MTOMStAXSOAPModelBuilder>>>.
  With the changes in r1164997 and r1207662, it is no longer used internally by Axiom.
  The interface declares methods that give access to the <<<Attachments>>> object that was
  used to create the builder. That is undesirable for two reasons:
  
   * For XOP/MTOM encoded messages, optimized binary data should always be accessed
     through the <<<OMText>>> API and not by accessing MIME parts directly through
     the <<<Attachments>>> API.
   
   * The existence of the <<<XOPBuilder>>> API (and in particular the <<<getAttachments>>>
     method defined by that interface) implies that an XOP/MTOM builder is necessarily
     created from an <<<Attachments>>> object. However, it may be desirable to support other
     ways to supply the MIME parts to the builder (e.g. using an implementation of the
     <<<MimePartProvider>>> API).
  
  Therefore the <<<XOPBuilder>>> API will be removed in Axiom 1.3.

Classes to be moved from <<<axiom-api>>> to <<<om-aspects>>>

  Up to version 1.2.12, the core Axiom code was organized in three modules,
  namely <<<axiom-api>>>, <<<axiom-impl>>> and <<<axiom-dom>>>, where <<<axiom-api>>>
  contains both the public API as well as implementation classes shared by LLOM and DOOM.
  Unfortunately the distinction between the public API and these shared implementation
  classes has become somewhat blurred over time. In Axiom 1.2.13 a new module
  called <<<axiom-common-impl>>> was introduced with the specific goal of separating the
  shared implementation classes from the public API; with the introduction of AspectJ
  in Axiom 1.2.15 this module is now called <<<om-aspects>>>.
  
  However, in Axiom 1.2.x it is generally not possible to move classes from
  <<<axiom-api>>> to <<<om-aspects>>> without the risk of breaking existing code.
  A new major release gives us the opportunity to move the existing shared classes to
  <<<om-aspects>>> as well, so that in Axiom 1.3, <<<axiom-api>>> will only
  contain Axiom's public API. This is one of the important goals for Axiom 1.3
  because it has multiple benefits:
  
   * The more compact the public API is, the easier it is for users to understand the
     API and to locate the features they are looking for.
   
   * By definition, anything that is not part of the public API can be modified
     without the risk of breaking application code. Clarifying the distinction between
     the public API and internal implementation classes therefore gives the project
     more flexibility to implement changes.
   
   * Having a well defined abstract API allows to create alternative implementations
     of that API.
  
  This section identifies the classes and internal APIs that will be removed
  from <<<axiom-api>>>.

* Builder implementations

  In Axiom 1.2.13, the <<<OMXMLBuilderFactory>>> API allows to create any type of
  object model builder (plain XML, SOAP, XOP and MTOM). The API also defines two
  interfaces representing a builder: <<<OMXMLParserWrapper>>> and <<<SOAPModelBuilder>>>.
  This means that application code should no longer reference the builder implementation
  classes directly, but only <<<OMXMLBuilderFactory>>>, <<<OMXMLParserWrapper>>> and
  <<<SOAPModelBuilder>>>. In Axiom 1.3 the implementation classes can therefore be moved
  to <<<om-aspects>>>. They are:
  
   * <<<org.apache.axiom.om.impl.builder.StAXBuilder>>>
   
   * <<<org.apache.axiom.om.impl.builder.StAXOMBuilder>>>
   
   * <<<org.apache.axiom.om.impl.builder.XOPAwareStAXOMBuilder>>>
   
   * <<<org.apache.axiom.soap.impl.builder.StAXSOAPModelBuilder>>>
   
   * <<<org.apache.axiom.soap.impl.builder.MTOMStAXSOAPModelBuilder>>>
  
  TODO: there is one API that still references <<<StAXBuilder>>>, namely <<<BuilderAwareReader>>>;
        it needs to be changed (or removed)
  
  Together with these classes, the following interfaces and helper classes should also
  be moved to <<<om-aspects>>>:
  
   * TODO

Public APIs that need to be moved to another package

  Some interfaces are part of the public API although they are placed in an
  <<<impl>>> package. These interfaces need to be moved to another package to make it
  clear that they are not internal or implementation specific interfaces:
  
   * <<<org.apache.axiom.om.impl.builder.CustomBuilder>>>

APIs that need to be overhauled

* <<<OMDataSource>>>

  The <<<OMDataSource>>> API has evolved over time and there have been several additions, not all of which
  are well designed. Here is a [partial] list of issues with the design of this API:
  
   * The <<<OMDataSource#serialize(OutputStream, OMOutputFormat)>>> and <<<OMDataSource#serialize(Writer, OMOutputFormat)>>>
     are currently only used when a corresponding <<<serialize>>> method is invoked directly on the <<<OMSourcedElement>>>
     instance. This however is a rare event because a <<<OMSourcedElement>>> is usually serialized as part of a larger tree,
     in which case <<<OMDataSource#serialize(XMLStreamWriter)>>> will be used. In addition, these methods are rarely invoked
     directly by application code, and code doing this could be easily rewritten to construct a <<<OMSourcedElement>>>
     first and to use the <<<serialize>>> methods on that element.
     
     The two methods could nevertheless be useful if they were used by <<<OMSourcedElement>>> to optimize the serialization
     process for <<<OMDataSource>>> instances backed by objects that can be serialized to a byte or character stream in a
     way that is more efficient than using an <<<XMLStreamWriter>>>. This would require <<<OMSourcedElement>>> to bypass
     the <<<XMLStreamWriter>>> and let the <<<OMDataSource>>> directly write to the underlying stream. However, the current
     API is not correctly designed for this purpose:
     
      * Since the two methods are defined by <<<OMDataSource>>> (and not by an optional interface), <<<OMSourcedElement>>>
        has no way of knowing whether the data source implements an optimized serialization mechanism. In fact, most data sources
        will have a trivial implementation based on <<<XMLStreamWriter>>> and in that case, using these methods would
        actually be less efficient than using <<<OMDataSource#serialize(XMLStreamWriter)>>>.
      
      * To ensure that the optimization always produces well formed XML, the contract of the two methods would have to clearly
        specify that the implementation must not write an XML declaration and also must not write a byte order marker
        (a byte order marker may only appear at the beginning of the document). In fact, the <<<OMOutputFormat>>> argument
        is not appropriate because only the charset encoding parameter would be relevant (for the method that writes
        to an <<<OutputStream>>>).
      
      * To correctly serialize the content of the <<<OMDataSource>>> as a fragment of a larger document, the implementation
        would need information about the current namespace context. However, that information is not passed to these methods.
        Without that information, the implementation will likely generate redundant namespace declarations. It may also
        generate incorrect XML if there is a default namespace.
      
      * The two methods are not XOP aware.

* <<<MTOMXMLStreamReader>>>

  This should become an interface:
  
   * This would enable the implementation of proxies and maybe allow us to get rid of
     <<<XMLStreamWriterFilter>>>.

   * This would allow us to move the implementation code out of <<<axiom-api>>>.

* Attachment lifecycle manager

  The <<<LifecycleManager>>> API has a couple of issues that can't be fixed without breaking
  backward compatibility:
  
   * There is some overlap with the <<<AttachmentCacheMonitor>>> class. There should be a single API to
     manage the lifecycle of buffered attachments.
   
   * <<<LifecycleManager>>> is an abstract API (interface), but refers to <<<FileAccessor>>> which
     is placed in an <<<impl>>> package.

* <<<SOAPVersion>>>

  The <<<SOAPVersion>>> API uses a somewhat peculiar design pattern: it has two public implementation classes
  (<<<SOAP11Version>>> and <<<SOAP12Version>>>) which do not add any methods and that have static <<<getSingleton>>>
  methods to get the singleton instance for each of these classes. This should be changed to use a more traditional
  pattern, namely the two singleton instances should be accessible using constants defined by <<<SOAPVersion>>>.
  The implementing classes then no longer need to be public (they can be anonymous or have package access).
  
  In addition, <<<SOAPVersion>>> should be changed from an interface to an abstract class so that one can
  define static methods to get the SOAP version by envelope namespace or media type.

* <<<StAXParserConfiguration>>>

  The <<<StAXParserConfiguration>>> API relies on the assumption that the XML parser used by Axiom is
  an implementation of StAX. However, as noted above, this is not a strict requirement; an Axiom
  implementation could equally well use another API or provide its own XML parser.
  Therefore <<<StAXParserConfiguration>>> should be replaced by something more generic.

* <<<OMMetaFactory>>>

  The argument order is not consistent across methods. See e.g. the two <<<createOMBuilder>>> methods that
  take an <<<InputSource>>> argument.

* <<<OMElement>>>

  The argument order of the <<<addAttribute(String, String, OMNamespace)>>> method is inconsistent with that
  of the <<<createOMAttribute>>> method in <<<OMFactory>>>.

* <<<OMOutputFormat>>>

  This class should be made immutable (which would require introducing a builder class) so that instances
  are safe to reuse.

Miscellaneous

* Make non coalescing mode the default

  By default, Axiom configures the underlying parser in coalescing mode. The reason is purely historical.
  Axiom originally used Woodstox 3.x and that version implemented one aspect of the StAX
  specification incorrectly, namely {{{http://jira.codehaus.org/browse/WSTX-140}it configured the parser by default
  in coalescing mode}}, while the specification says otherwise. The problem is that (poorly
  written) code that uses Axiom with a parser in coalescing mode doesn't
  necessarily work with non coalescing mode. Therefore the choice was
  made to make coalescing mode the default in order to ensure
  compatibility when using a StAX implementation other than Woodstox 3.x.
  
  A new major release would be the right moment to change this and make non coalescing mode the default.
  This enables a couple of optimizations (e.g. when reading and decoding base64 from a text node) and
  ensures that an XML document can be streamed with constant memory, even if it contains large text nodes.

* Don't allow <<<addChild>>> to reorder children

  The <<<SOAPEnvelope>>> implementations in LLOM and DOOM override the <<<addChild>>> method to
  reorder the nodes if an attempt is made to add a <<<SOAPHeader>>> after the <<<SOAPBody>>>.
  This introduces unnecessary complexity in the implementation and is questionable from an OO
  design perspective because it breaks the general contract of the <<<addChild>>> method which is
  to add the node as the last child.
  
  The <<<addChild>>> implementation for <<<SOAPEnvelope>>> should not do this. Instead it should
  just throw an exception if a <<<SOAPHeader>>> is added at the wrong position.