How to add a new language to PMD using JavaCC grammar.
Table of Contents
Before you start…

This is really a big contribution and can’t be done with a drive by contribution. It requires dedicated passion and long commitment to implement support for a new language.

This step-by-step guide is just a small intro to get the basics started, and it’s also not necessarily up-to-date or complete. You have to be able to fill in the blanks.

After the basic support for a language is there, there are lots of missing features left. Typical features that can greatly improve rule writing are: symbol table, type resolution, call/data flow analysis.

Symbol table keeps track of variables and their usages. Type resolution tries to find the actual class type of each used type, following along method calls (including overloaded and overwritten methods), allowing to query subtypes and type hierarchy. This requires additional configuration of an auxiliary classpath. Call and data flow analysis keep track of the data as it is moving through different execution paths a program has.

These features are out of scope of this guide. Type resolution and data flow are features that definitely don’t come for free. It is much effort and requires perseverance to implement.

Steps

1. Start with a new sub-module

  • See pmd-java or pmd-vm for examples.
  • Make sure to add your new module to PMD’s parent pom as <module> entry, so that it is built alongside the other languages.
  • Also add your new module to the dependencies list in “pmd-languages-deps/pom.xml”, so that the new language is automatically available in the binary distribution (pmd-dist).

2. Implement an AST parser for your language

  • Ideally an AST parser should be implemented as a JJT file (see VmParser.jjt or Java.jjt for example)
  • There is nothing preventing any other parser implementation, as long as you have some way to convert an input stream into an AST tree. Doing it as a JJT simplifies maintenance down the road.
  • See this link for reference: https://javacc.java.net/doc/JJTree.html

3. Create AST node classes

  • For each AST node that your parser can generate, there should be a class
  • The name of the AST class should be “AST” + “whatever is the name of the node in JJT file”.
    • For example, if JJT contains a node called “IfStatement”, there should be a class called “ASTIfStatement”
  • Each AST class should have one package-private constructor, that takes an int id.
  • It’s a good idea to create a parent AST class for all AST classes of the language. This simplifies rule creation later. (see SimpleNode for Velocity and AbstractJavaNode for Java for example)
  • Note: These AST node classes are generated usually once by javacc/jjtree and can then be modified as needed.
  • You can add additional methods in your AST node classes, that can be used in rules. Most getters are also available for XPath rules, see section XPath integration below.

4. Generate your parser (using JJT)

  • An ant script is being used to compile jjt files into classes. This is in javacc-wrapper.xml file in the top-level pmd sources.
  • The ant script is executed via the maven-antrun-plugin. Add this plugin to your pom.xml file and configure it the language name. You can use pmd-java/pom.xml as an example.
  • The ant script is called in the phase generate-sources whenever the whole project is built. But you can call ./mvnw generate-sources directly for your module if you want your parser to be generated.

5. Create a PMD parser “adapter”

  • Create a new class that extends JjtreeParserAdapter.
  • This is a generic class, and you need to declare the root AST node.
  • There are two important methods to implement
    • tokenBehavior method should return a new instance of TokenDocumentBehavior constructed with the list of tokes in your language. The compile step #4 will generate a class $langTokenKinds which has all the available tokens in the field TOKEN_NAMES.
    • parseImpl method should return the root node of the AST tree obtained by parsing the CharStream source
    • See VmParser class as an example

6. Create a language version handler

  • Extend AbstractPmdLanguageVersionHandler (see VmHandler for example)
  • This class is sort of a gateway between PMD and all parsing logic specific to your language.
  • For a minimal implementation, it just needs to return a parser (see step #5).
  • It can be used to provide other features for your language like
    • violation suppression logic
    • ViolationDecorators, to add additional language specific information to the created violations. The Java language module uses this to provide the method name or class name, where the violation occurred.
    • metrics (see below “Optional features”)
    • custom XPath functions
  • See VmHandler class as an example

7. Create a base visitor

  • A parser visitor adapter is not needed anymore with PMD 7. The visitor interface now provides a default implementation.
  • The visitor for JavaCC based AST is generated along the parser from the grammar file. The base interface for a visitor is AstVisitor.
  • The generated visitor class for VM is called VmVisitor.
  • In order to help use this visitor later on, a base visitor class should be created. See VmVisitorBase as an example.

8. Make PMD recognize your language

  • Create your own subclass of net.sourceforge.pmd.lang.impl.SimpleLanguageModuleBase. (see VmLanguageModule or JavaLanguageModule as an example)
  • Add for each version of your language a call to addVersion in your language module’s constructor. Use addDefaultVersion for defining the default version.
  • You’ll need to refer the language version handler created in step #6.
  • Create the service registration via the text file src/main/resources/META-INF/services/net.sourceforge.pmd.lang.Language. Add your fully qualified class name as a single line into it.

9. Add AST regression tests

For languages, that use an external library for parsing, the AST can easily change when upgrading the library. Also for languages, where we have the grammar under our control, it is useful to have such tests.

The tests parse one or more source files and generate a textual representation of the AST. This text is compared against a previously recorded version. If there are differences, the test fails.

This helps to detect anything in the AST structure that changed, maybe unexpectedly.

  • Create a test class in the package net.sourceforge.pmd.lang.$lang.ast with the name $langTreeDumpTest.
  • This test class must extend net.sourceforge.pmd.lang.test.ast.BaseTreeDumpTest. Note: This class is written in kotlin and is available in the module “lang-test”.
  • Add a default constructor, that calls the super constructor like so:

        public $langTreeDumpTest() {
            super(NodePrintersKt.getSimpleNodePrinter(), ".$extension");
        }
    

    Replace “$lang” and “$extension” accordingly.

  • Implement the method getParser(). It must return a subclass of net.sourceforge.pmd.lang.test.ast.BaseParsingHelper. See net.sourceforge.pmd.lang.ecmascript.ast.JsParsingHelper for an example. With this parser helper you can also specify, where the test files are searched, by using the method withResourceContext(Class<?>, String).
  • Add one or more test methods. Each test method parses one file and compares the result. The base class has a helper method doTest(String) that does all the work. This method just needs to be called:

        @Test
        public void myFirstAstTest() {
            doTest("filename-without-extension");
        }
    
  • On the first test run the test fails. A text file (with the extension .txt) is created, that records the current AST. On the next run, the text file is used as comparison and the test should pass. Don’t forget to commit the generated text file.

A complete example can be seen in the JavaScript module: net.sourceforge.pmd.lang.ecmascript.ast.JsTreeDumpTest. The test resources are in the subpackage “testdata”: pmd-javascript/src/test/resources/net/sourceforge/pmd/lang/ecmascript/ast/testdata/.

The Scala module also has a test, written in Kotlin instead of Java: net.sourceforge.pmd.lang.scala.ast.ScalaParserTests.

10. Create an abstract rule class for the language

  • Extend AbstractRule and implement the parser visitor interface for your language (see AbstractVmRule for example)
  • All other rules for your language should extend this class. The purpose of this class is to implement visit methods for all AST types to simply delegate to default behavior. This is useful because most rules care only about specific AST nodes, but PMD needs to know what to do with each node - so this just lets you use default behavior for nodes you don’t care about.

11. Create rules

  • Rules are created by extending the abstract rule class created in step 9 (see EmptyForeachStmtRule for example)
  • Creating rules is already pretty well documented in PMD - and it’s no different for a new language, except you may have different AST nodes.

12. Test the rules

  • Testing rules is described in depth in Testing your rules.
    • Each rule has its own test class: Create a test class for your rule extending PmdRuleTst (see AvoidReassigningParametersTest in pmd-vm for example)
    • Create a category rule set for your language (see category/vm/bestpractices.xml for example)
    • Place the test XML file with the test cases in the correct location
    • When executing the test class
      • this triggers the unit test to read the corresponding XML file with the rule test data (see AvoidReassigningParameters.xml for example)
      • This test XML file contains sample pieces of code which should trigger a specified number of violations of this rule. The unit test will execute the rule on this piece of code, and verify that the number of violations matches.
  • To verify the validity of the created ruleset, create a subclass of AbstractRuleSetFactoryTest (see RuleSetFactoryTest in pmd-vm for example). This will load all rulesets and verify, that all required attributes are provided.

    Note: You’ll need to add your category ruleset to categories.properties, so that it can be found.

13. Create documentation page

Finishing up your new language module by adding a page in the documentation. Create a new markdown file <langId>.md in docs/pages/pmd/languages/. This file should have the following frontmatter:

---
title: <Language Name>
permalink: pmd_languages_<langId>.html
last_updated: <Month> <Year> (<PMD Version>)
tags: [languages, PmdCapableLanguage, CpdCapableLanguage]
---

On this page, language specifics can be documented, e.g. when the language was first supported by PMD. There is also the following Jekyll Include, that creates summary box for the language:


{% include language_info.html name='<Language Name>' id='<langId>' implementation='<langId>::lang.<langId>.<langId>LanguageModule' supports_cpd=true supports_pmd=true %}

XPath integration

PMD exposes the AST nodes for use by XPath based rules (see DOM representation of ASTs). Most Java getters in the AST classes are made available by default. These getters constitute the API of the language. If a getter method is renamed, then every XPath rule that uses this getter also needs to be adjusted. In order to have more control over this, there are two annotations that can be used for AST classes and their methods:

  • DeprecatedAttribute: Getters might be annotated with that indicating, that this getter method should not be used in XPath rules. When a XPath rule uses such a method, a warning is issued. If the method additionally has the standard Java @Deprecated annotation, then the getter is also deprecated for java usage. Otherwise, the getter is only deprecated for usage in XPath rules.

    When a getter is deprecated and there is a different getter to be used instead, then the attribute replaceWith should be used.

  • NoAttribute: This annotation can be used on an AST node type or on individual methods in order to filter out which methods are available for XPath rules. When used on a type, either all methods can be filtered or only inherited methods (see attribute scope). When used directly on an individual method, then only this method will be filtered out. That way methods can be added in AST nodes, that should only be used in Java rules, e.g. as auxiliary methods.

Only the following Java result types are supported:

  • String
  • any Enum-type
  • int
  • boolean
  • double
  • long
  • char
  • float

Debugging with Rule Designer

When implementing your grammar it may be very useful to see how PMD parses your example files. This can be achieved with Rule Designer:

  • Override the getXPathNodeName in your AST nodes for Designer to show node names.
  • Make sure to override both jjtOpen and jjtClose in your AST node base class so that they set both start and end line and column for proper node bound highlighting.
  • Not strictly required but trivial and useful: implement syntax highlighting for Rule Designer:
    • Fork and clone the pmd/pmd-designer repository.
    • Add a syntax highlighter implementation to net.sourceforge.pmd.util.fxdesigner.util.codearea.syntaxhighlighting (you could use Java as an example).
    • Register it in the AvailableSyntaxHighlighters enumeration.
    • Now build your implementation and place the target/pmd-designer-<version>-SNAPSHOT.jar to the lib directory inside your pmd-bin-... distribution (you have to delete old pmd-designer-*.jar from there).

Optional features

Metrics

If you want to add support for computing metrics:

  • Create a package lang.<langname>.metrics
  • Create a utility class <langname>Metrics
  • Implement new metrics and add them as static constants. Be sure to document them.
  • Implement getLanguageMetricsProvider, to make the metrics available in the designer.

See JavaMetrics for an example.

Symbol table

A symbol table keeps track of variables and their usages. It is part of semantic analysis and would be executed in your parser adapter as an additional pass after you got the initial AST.

There is no general language independent API in PMD core. For now, each language will need to implement its own solution. The symbol information that has been resolved in the additional parser pass can be made available on the AST nodes via extra methods, e.g. getSymbolTable(), getSymbol(), or getUsages().

Currently only Java provides an implementation for symbol table, see Java-specific features and guidance.

Type resolution

For typed languages like Java type information can be useful for writing rules, that trigger only on specific types. Resolving types of expressions and variables would be done after in your parser adapter as yet another additional pass, potentially after resolving the symbol table.

Type resolution tries to find the actual class type of each used type, following along method calls (including overloaded and overwritten methods), allowing to query subtypes and type hierarchy. This might require additional configuration for the language, e.g. in Java you need to configure an auxiliary classpath.

There is no general language independent API in PMD core. For now, each language will need to implement its own solution. The type information can be made available on the AST nodes via extra methods, e.g. getType().

Currently only Java provides an implementation for type resolution, see Java-specific features and guidance.

Call and data flow analysis

Call and data flow analysis keep track of the data as it is moving through different execution paths a program has. This would be yet another analysis pass.

There is no general language independent API in PMD core. For now, each language will need to implement its own solution.

Currently Java has some limited support for data flow analysis, see Java-specific features and guidance.