Writing your own PMD rules
Table of Contents

PMD is a framework to perform code analysis. You can create your own rules to check for patterns specific to your codebase, or the coding practices of your team.

How rules work: the AST

Before running rules, PMD parses the source file into a data structure called an abstract syntax tree (AST). This tree represents the syntactic structure of the code, and encodes syntactic relations between source code elements. For instance, in Java, method declarations belong to a class: in the AST, the nodes representing method declarations will be descendants of a node representing the declaration of their enclosing class. This representation is thus much richer than the original source code (which, for a program, is just a chain of characters), or the token chain produced by a lexer. For example:

Sample code (Java) AST
class Foo extends Object {

└─ CompilationUnit
   └─ TypeDeclaration
      └─ ClassDeclaration "Foo"
         ├─ ModifierList
         ├─ ExtendsList
           └─ ClassType "Object"
         └─ ClassBody

Conceptually, PMD rules work by matching a “pattern” against the AST of a file. Rules explore the AST and find nodes that satisfy some conditions that are characteristic of the specific thing the rule is trying to flag. Rules then report a violation on these nodes.

Discovering the AST

ASTs are represented by Java classes deriving from Node. Each PMD language has its own set of such classes, and its own rules about how these classes relate to one another, based on the grammar of the language. For example, all Java AST nodes extend JavaNode.

The structure of the AST can be discovered through

Writing new rules

PMD supports two ways to define rules: using an XPath query, or using a Java visitor. XPath rules are much easier to set up, since they’re defined directly in your ruleset XML, and are expressive enough for nearly any task.

On the other hand, some parts of PMD’s API are only accessible from Java, e.g. accessing the usages of a declaration. And Java rules allow you to do some complicated processing, to which an XPath rule couldn’t scale.

In the end, choosing one strategy or the other depends on the difficulty of what your rule does. I’d advise to keep to XPath unless you have no other choice.

Note: Despite that fact, the Java rules are written in Java, any language that PMD supports can be analyzed. E.g. you can write a Java rule to analyze Apex source code.

XML rule definition

New rules must be declared in a ruleset before they’re referenced. This is the case for both XPath and Java rules. To do this, the rule element is used, but instead of mentioning the ref attribute, it mentions the class attribute, with the implementation class of your rule.

  • For Java rules: this is the concrete class extending AbstractRule (transitively)
  • For XPath rules: this is net.sourceforge.pmd.lang.rule.xpath.XPathRule.
  • For XPath rules analyzing XML-based languages: this is net.sourceforge.pmd.lang.xml.rule.DomXPathRule. See XPath rules in XML for more info.

Example for Java rule:

<rule name="MyJavaRule"

Example for XPath rule:

<rule name="MyXPathRule"
        <property name="xpath">

Resource index

To learn how to write a rule:

To go further:

  • Defining Properties describes how to make your rules more configurable with rule properties
  • Testing your Rules introduces our testing framework and how you can use it to safeguard the quality of your rule