Architecture

Aspectran Configuration Loading Mechanism: An In-Depth Analysis of the `nodelet` Engine

Aspectran parses configuration files in XML or APON (Aspectran Object Notation) format to create a tree of *Rule objects, which serve as the blueprint for the application’s core components (Bean, Translet, Aspect, etc.). At the heart of this process is the high-performance, event-based parsing engine nodelet, specially designed for Aspectran and implemented in the com.aspectran.utils.nodelet package.

[XML/APON Config File] ---> [NodeletParser] --uses--> [AspectranNodeletGroup] ---> [*Rule Object Tree] ---> [ActivityContext]

1. The Core Parsing Engine: com.aspectran.utils.nodelet

nodelet operates on an event basis, similar to a typical SAX parser, but it is a lightweight parsing framework that allows for much more intuitive and structured processing of complex XML documents by directly mapping callbacks to specific XPath paths.

Key Components:

  • NodeletParser: The main engine that drives the parsing process. It sequentially reads an XML document, detects the start and end of each node (element), and fires the appropriate event corresponding to the current path.
  • Nodelet / EndNodelet: Event handler (callback) interfaces.
    • Nodelet: Called when the start tag of an XML element is encountered. It processes the attribute information of that element, which is passed as a parameter.
    • EndNodelet: Called when the end tag of an XML element is encountered. It processes the text content (including CDATA) contained within that element.
  • NodeletGroup: The core design of the nodelet engine. It manages Nodelets and EndNodelets in groups along with specific XPath paths. This class provides a Fluent API that acts as a kind of DSL (Domain-Specific Language) for declaratively defining parsing rules.
    • .child("elementName"): Starts defining rules for a child element of the current path.
    • .parent(): Moves from the current path to the parent path.
    • .nodelet(...): Registers a Nodelet for the start tag of the current path.
    • .endNodelet(...): Registers an EndNodelet for the end tag of the current path.
    • .with(NodeletAdder): Adds a reusable set of rules to the current group.
    • .mount(NodeletGroup): Dynamically “mounts” another NodeletGroup at the current path.
  • NodeletAdder: An interface that adds a reusable set of rules to a NodeletGroup. This allows for the modularization and reuse of parsing logic for common XML structures (e.g., <item>, <argument>) in multiple places.

The Innovative mount Feature

The mount feature of NodeletGroup is a key element that makes the nodelet engine very powerful and efficient. This feature dynamically activates a pre-defined set of rules from another NodeletGroup when a specific element appears. This provides the following important advantages:

  1. Memory Efficiency and Performance Improvement: When using mount, the parser finds the Nodelet using a relative path from the mounted point, instead of the full absolute XPath path (a/b/c/...). This dramatically reduces memory usage and improves parsing performance by avoiding the cost of generating and comparing long XPath strings, even in very deep nested structures.
  2. Maximizing Rule Reusability: Complex structures that can appear repeatedly, such as <choose>/<when>/<otherwise>, can be defined as a separate NodeletGroup (ChooseNodeletGroup), and this set of parsing rules can be reused anywhere with a single line of .mount().
  3. Removing Nesting Level Limits: Since mount effectively resets the parsing context, there is theoretically no limit to the nesting level of XML elements.

2. Aspectran’s Implementation: com.aspectran.core.context.rule.parser.xml

Aspectran uses the nodelet engine described above to parse its complex XML configuration files (e.g., aspectran-config.xml) and convert their contents into a tree of *Rule objects.

Parsing Flow:

  1. Initialization of AspectranNodeParser: This is the starting point of the parsing. It internally creates a NodeletParser and registers AspectranNodeletGroup, the master NodeletGroup where all of Aspectran’s XML rules are defined, with the parser. It also sets up AspectranDtdResolver for DTD validation.

  2. AspectranNodeletGroup and *NodeletAdder:
    • AspectranNodeletGroup is the top-level rule group for the <aspectran> root element. This group assembles the entire set of parsing rules by adding BeanNodeletAdder, AspectNodeletAdder, TransletNodeletAdder, etc., which handle each top-level element, through the with() method.
    • Each *NodeletAdder uses the Fluent API of NodeletGroup to define in detail how to parse the XML element it is responsible for and its child elements. For example, TransletNodeletAdder defines the rule for the <translet> element, and also defines the rules for child elements that can come inside it, such as <request>, <response>, and <content>.
    • In particular, Adders with ‘Inner’ in their name, like ActionInnerNodeletAdder or ArgumentNodeletAdder, or those that handle specific elements (argument, property, etc.), are made to parse common XML structures used inside various other rules (e.g., <translet>, <advice>, <bean>). This maximizes the reusability of parsing logic.
  3. State Management: AspectranNodeParsingContext and ObjectStack:
    • XML parsing is a stateful task that needs to handle parent-child relationships. For example, when parsing an <action> element, you need to know which <translet> this action belongs to.
    • Aspectran solves this problem through AspectranNodeParsingContext. This class provides a thread-safe ArrayStack (Object Stack) for each thread using a ThreadLocal variable.
    • How it works:
      1. When the Nodelet for an element’s start tag is called (e.g., <translet>), it creates a TransletRule object to hold that rule and pushes it onto the stack using AspectranNodeParsingContext.pushObject().
      2. In the Nodelet for a child element (e.g., <request>), it calls AspectranNodeParsingContext.peekObject() to get the parent object (TransletRule) at the top of the stack and connects itself (RequestRule) to the parent.
      3. When the EndNodelet for an element’s end tag is called, the completed object is removed from the stack using AspectranNodeParsingContext.popObject().

3. Parsing Process with a Real Example

Example 1: Basic <translet> Parsing and the Object Stack

Let’s assume we have the following simple XML configuration:

<translet name="/example/hello">
    <action bean="helloAction" method="sayHello" />
</translet>

The process of parsing this XML is as follows:

  1. <translet> Start:
    • NodeletParser encounters the <translet> start tag.
    • The nodelet defined in TransletNodeletAdder is executed.
    • The nodelet reads the name attribute value (“/example/hello”) and creates a TransletRule object.
    • The created transletRule object is pushed onto the Object Stack via AspectranNodeParsingContext.pushObject().
    • Object Stack State: [ TransletRule(...) ]
  2. <action> Start:
    • The parser encounters the child node <action> start tag.
    • The nodelet of ActionInnerNodeletAdder, which was added by TransletNodeletAdder via .with(ActionInnerNodeletAdder.instance()), is executed.
    • The nodelet reads the bean and method attributes and creates an InvokeActionRule object.
    • The created invokeActionRule object is pushed onto the stack.
    • Object Stack State: [ TransletRule(...), InvokeActionRule(...) ]
  3. </action> End:
    • The parser encounters the </action> end tag.
    • The endNodelet of ActionInnerNodeletAdder is executed.
    • InvokeActionRule invokeActionRule = AspectranNodeParsingContext.popObject(): The InvokeActionRule is popped from the stack.
    • HasActionRules hasActionRules = AspectranNodeParsingContext.peekObject(): The top object of the stack, the TransletRule, is checked. (TransletRule implements the HasActionRules interface.)
    • hasActionRules.putActionRule(invokeActionRule): The InvokeActionRule is added as a child to the TransletRule.
    • Object Stack State: [ TransletRule(...) ]
  4. </translet> End:
    • The parser encounters the </translet> end tag.
    • The endNodelet of TransletNodeletAdder is executed.
    • TransletRule transletRule = AspectranNodeParsingContext.popObject(): The fully configured TransletRule is popped from the stack.
    • AspectranNodeParsingContext.getCurrentRuleParsingContext().addTransletRule(transletRule): The completed rule is finally registered with RuleParsingContext.
    • Object Stack State: [ ] (empty)

Example 2: Reusing Complex Structures with the mount Feature

Now let’s look at the power of the mount feature using the nestable <choose> element as an example.

<translet name="/example/conditional">
    <choose>
        <when test="@{param:type == 'A'}">
            <action bean="actionA" />
        </when>
        <otherwise>
            <!-- Another choose can come here -->
            <action bean="actionB" />
        </otherwise>
    </choose>
</translet>
  • TransletNodeletAdder adds the <choose> rule via .with(ChooseNodeletAdder.instance()).
  • ChooseNodeletAdder, when defining the rules for the <when> and <otherwise> elements, calls .mount(ChooseNodeletGroup.instance()).
  • ChooseNodeletGroup is an independent group that defines the parsing rules for the <choose> element itself.

What this means is:

When the parser enters a <when> or <otherwise> element, the rules of the mounted ChooseNodeletGroup are dynamically activated. This means that even if another <choose> element is encountered inside <when> or <otherwise>, the parser can handle the parsing recursively in the same way as the first time.

In this way, the mount feature allows you to define the parsing rules for complex and recursive XML structures only once and reuse them wherever needed, eliminating code duplication and optimizing memory usage.

Example 3: <aspect> Rule and Complex Relationship Settings

Since AOP rules involve complex relationships between multiple components, the parsing process is more multi-layered. Let’s look at it through the following example.

<aspect id="myAspect" order="1">
    <joinpoint>
      pointcut: {
        type: "wildcard",
        +: "/example/hello@helloAction^say*"
      }
    </joinpoint>
    <advice bean="loggingAdvice">
        <before>
            <action bean="profiler" method="start"/>
        </before>
    </advice>
</aspect>

This XML defines an AOP rule that executes the profiler.start() action of the loggingAdvice bean before any method starting with say (joinpoint) of the helloAction bean is executed in the Translet named /example/hello.

In the pointcut string, the part before the @ delimiter represents the Translet’s name, the part after it represents the Bean’s ID, and the part after the ^ delimiter represents the method name to be called. Wildcards can be used in each part. If you want to target all Translets instead of a specific one, the pattern must start with the @ delimiter (e.g., + "@helloAction^say*").

  1. <aspect> Start:
    • The nodelet of AspectNodeletAdder is executed.
    • It reads the id and order attributes, creates an AspectRule object, and pushes it onto the stack.
    • Object Stack: [ AspectRule(...) ]
  2. <joinpoint> Start and End:
    • The nodelet for joinpoint defined inside AspectNodeletAdder is executed, but it does no special work when using APON format.
    • The endNodelet of </joinpoint> is executed.
    • The endNodelet receives the entire APON-formatted text inside the <joinpoint> element as the text parameter.
    • Internal logic like JoinpointRule.updateJoinpoint(joinpointRule, text) parses the APON text to create PointcutRule, etc.
    • It gets the AspectRule from the stack with peek() and sets the completed joinpointRule.
    • Object Stack: [ AspectRule(...) ]
  3. <advice> Start:
    • The nodelet for advice defined inside AspectNodeletAdder is executed.
    • It reads the bean attribute (“loggingAdvice”).
    • It gets the AspectRule with peek() and calls setAdviceBeanId("loggingAdvice") to specify the bean that will perform the Advice. Nothing is pushed onto the stack at this stage.
    • Object Stack: [ AspectRule(...) ]
  4. <before> Start:
    • The child node rules of advice are handled by AdviceInnerNodeletAdder. The nodelet for before in this Adder is executed.
    • It gets the AspectRule with peek() and calls the newBeforeAdviceRule() method to create an AdviceRule object.
    • The created adviceRule is pushed onto the stack.
    • Object Stack: [ AspectRule(...), AdviceRule(type=BEFORE, ...) ]
  5. <action> Start and End (in <before>):
    • The before rule reuses the rules of ActionInnerNodeletAdder (with(...)).
    • The nodelet of <action> is executed, creates an InvokeActionRule(“profiler.start”), and pushes it onto the stack.
    • Object Stack: [ AspectRule(...), AdviceRule(...), InvokeActionRule(...) ]
    • The endNodelet of </action> is executed.
    • It pops the InvokeActionRule, gets the AdviceRule with peek(), and adds it as a child via putActionRule().
    • Object Stack: [ AspectRule(...), AdviceRule(...) ]
  6. </before> End:
    • The endNodelet of before is executed.
    • It pops the AdviceRule. This AdviceRule is already connected to its parent AspectRule.
    • Object Stack: [ AspectRule(...) ]
  7. </advice> End:
    • No endNodelet is specifically defined for the advice element, so no action is taken.
  8. </aspect> End:
    • The endNodelet of AspectNodeletAdder is executed.
    • It pops the AspectRule, which is now filled with all information (joinpoint, advice, etc.), from the stack.
    • The completed aspectRule is finally registered with RuleParsingContext.
    • Object Stack: [ ] (empty)

As you can see, even complex AOP rules are parsed in a very systematic and predictable way through state management using Nodelet and Object Stack, and rule reuse via with().

4. Conclusion

Aspectran’s configuration loading mechanism is the essence of a sophisticated architecture designed on top of its custom nodelet parsing engine. The key advantages of this mechanism can be summarized in three keywords: flexibility, reusability, and performance.

  • Flexibility: The Fluent API of NodeletGroup and the mount feature allow for very flexible responses to complex and variable XML structures.
  • Reusability: NodeletAdder perfectly modularizes the parsing logic for common XML structures like <action> and <item>, and minimizes code duplication by easily reusing it wherever needed with the with() method.
  • Performance: The mount feature dynamically switches the parsing context, so it does not use the full XPath in deeply hierarchical XML documents, thus ensuring efficient parsing performance.

After parsing, the created *Rule objects are a kind of “blueprint”. Based on these blueprints, ActivityContext then creates the actual Bean instances, injects dependencies, and applies AOP proxies to finally build the application’s executable runtime environment. This architecture can be said to be the core foundation of the Aspectran framework’s flexibility and extensibility.