OK, so you have finally decided to write your own Check. Welcome aboard, this is really a easy thing to do. Very basic Java knowledge is required to write a Check, it is good practice for even for student. There are actually two kinds of Checks, so before you can start, you have to find out which kind of Check you want to implement.
The functionality of Checkstyle is implemented in modules that can be plugged into Checkstyle. Modules can be containers for other modules, i.e. they form a tree structure. The top level modules that are known directly to the Checkstyle kernel (which is also a module and forms the root of the tree) implement the FileSetCheck interface. These are pretty simple to grasp: they take a set of input files and fire error messages.
Checkstyle provides a few FileSetCheck implementations by default and one of them happens to be the TreeWalker. A TreeWalker supports submodules that are derived from the AbstractCheck class. The TreeWalker operates by separately transforming each of the Java input files into an abstract syntax tree and then handing the result over to each of the Check submodules which in turn have a look at a certain aspect of the tree.
Most of the functionality of Checkstyle is implemented as Checks. If you know how to write your own Checks, you can extend Checkstyle according to your needs without having to wait for the Checkstyle development team. You are about to become a Checkstyle Expert.
Suppose you have a convention that the number of methods in a class should not exceed a certain limit, say 30. This rule makes sense, a class should only do one thing and do it well. With a zillion methods chances are that the class does more than one thing. The only problem you have is that your convention is not checked by Checkstyle, so you'll have to write your own Check and plug it into the Checkstyle framework.
This chapter is organized as a tour that takes you through the process step by step and explains both the theoretical foundations and the Checkstyle API along the way.
Every Java Program is structured into files, and each of these files has a certain structure. For example, if there is a package statement then it is the first line of the file that is not comment or whitespace. After the package statement comes a list of import statements, which is followed by a class or interface definition, and so on.
If you have ever read an introductory level Java book you probably knew all of the above. And if you have studied computer science, you probably also know that the rules that specify the Java language can be formally specified using a grammar (statement is simplified for didactic purposes).
There are tools which read a grammar definition and produce a parser for the language that is specified in the grammar. In other words, the output of the tool is a program that can transform a stream of characters (a Java file) into a tree representation that reflects the structure of the file. Checkstyle uses the parser generator ANTLR but that is an implementation detail you do not need to worry about when writing Checks, as well tested parser will parse Java file for you.
Checkstyle can print Abstract Syntax Tree for Java trees. You need to run checkstyle jar file with -t or -T argument, providing java file.
For example, here is MyClass.java file:
/**
* My <b>class</b>.
* @see AbstractClass
*/
public class MyClass {
}
|
1) Using -t as parameter.
java -jar checkstyle-X.XX-all.jar -t MyClass.java |
2) Using -T as parameter.
java -jar checkstyle-X.XX-all.jar -T MyClass.java |
CLASS_DEF -> CLASS_DEF [5:0]
|--MODIFIERS -> MODIFIERS [5:0]
| `--LITERAL_PUBLIC -> public [5:0]
|--LITERAL_CLASS -> class [5:7]
|--IDENT -> MyClass [5:13]
`--OBJBLOCK -> OBJBLOCK [5:21]
|--LCURLY -> { [5:21]
`--RCURLY -> } [7:0]
|
CLASS_DEF -> CLASS_DEF [5:0]
|--MODIFIERS -> MODIFIERS [5:0]
| |--BLOCK_COMMENT_BEGIN -> /* [1:0]
| | |--COMMENT_CONTENT -> *\n * My <b>class</b>.\n * @see AbstractClass\n [1:2]
| | `--BLOCK_COMMENT_END -> */ [4:1]
| `--LITERAL_PUBLIC -> public [5:0]
|--LITERAL_CLASS -> class [5:7]
|--IDENT -> MyClass [5:13]
`--OBJBLOCK -> OBJBLOCK [5:21]
|--LCURLY -> { [5:21]
`--RCURLY -> } [7:0]
|
As you can see very small java file transforms to a huge Abstract Syntax Tree, because that is the most detailed tree including all components of the java file: object block, comments, modifiers, etc.
Still with us? Great, you have mastered the basic theory so here is your reward - a GUI that displays the structure of a Java source file. To run it type
java -cp checkstyle-${projectVersion}-all.jar com.puppycrawl.tools.checkstyle.gui.Main
on the command line. Click the button at the bottom of the frame and select a syntactically correct Java source file. The frame will be populated with a tree that corresponds to the structure of the Java source code.
In the leftmost column you can open and close branches of the tree, the remaining columns display information about each node in the tree. The second column displays a token type for each node. As you navigate from the root of the tree to one of the leafs, you'll notice that the token type denotes smaller and smaller units of your source file, i.e. close to the root you might see the token type CLASS_DEF (a node that represents a class definition) while you will see token types like IDENT (an identifier) near the leaves of the tree.
In the bottom of frame you can find buttons "Open File", "Reload File" and dropdown list with parse modes to choose. First one opens file choose window. After choosing file tree that corresponds to java source file builds in frame. Notice that only files with ".java" extension can be opened. Second one reloads chosen file from file system and rebuilds source code tree. Dropdown list allow to choose one of three parse modes: "PLAIN JAVA", "JAVA WITH COMMENTS", "JAVA WITH JAVADOC AND COMMENTS". "PLAIN JAVA" mode uses to show java source code without comments. In "JAVA WITH COMMENTS" you can see also comments blocks on the tree. When "JAVA WITH JAVADOC AND COMMENTS" chosen javadoc tree builds and attaches for every comment block, that contains javadoc.
Notice: text of a tree node and its children is selected automatically after either pressing "Enter" or double-clicking on it, so there is no need to make selection manually.
We'll get back to the details in the other columns later, they are important for implementing Checks but not for understanding the basic concepts. For now it is sufficient to know that the gui is a tool that lets you look at the structure of a Java file, i.e. you can see the Java grammar 'in action'.
If you use Eclipse you can install Checkstyle AST Eclipse Viewer plugin to launch that application from context menu on any file in Eclipse.
Ready for a bit more theory? The last bit that is missing before you can start writing Checks is understanding the Visitor pattern.
When working with Abstract Syntax Tree (AST), a simple approach to define check operations on them would be to add a check() method to the Class that defines the AST nodes. For example, our AST type could have a method checkNumberOfMethods(). Such an approach would suffer from a few serious drawbacks. Most importantly, it does not provide an extensible design, i.e. the Checks have to be known at compile time; there is no way to write plugins.
Hence Checkstyle's AST classes do not have any methods that implement checking functionality. Instead, Checkstyle's TreeWalker takes a set of objects that conform to a AbstractCheck interface. OK, you're right - actually it's not an interface but an abstract class to provide some helper methods. A Check provides methods that take an AST as an argument and perform the checking process for that AST, most prominently visitToken().
It is important to understand that the individual Checks do no drive the AST traversal (it possible to traverse itself, but not recommended). Instead, the TreeWalker initiates a recursive descend from the root of the AST to the leaf nodes and calls the Check methods. The traversal is done using a tree traversal (depth-first) algorithm.
Before any visitor method is called, the TreeWalker will call beginTree() to give the Check a chance to do some initialization. Then, when performing the recursive descend from the root to the leaf nodes, the visitToken() method is called. Unlike the basic examples in the pattern book, there is a visitToken() counterpart called leaveToken(). The TreeWalker will call that method to signal that the subtree below the node has been processed and the TreeWalker is backtracking from the node. After the root node has been left, the TreeWalker will call finishTree().
Let's get back to our example and start writing code - that's why you came here, right? When you fire up the Checkstyle GUI and look at a few source files you'll figure out pretty quickly that you are mainly interested in the number of tree nodes of type METHOD_DEF. The number of such tokens should be counted separately for each CLASS_DEF / INTERFACE_DEF.
Hence we need to register the Check for the token types CLASS_DEF and INTERFACE_DEF. The TreeWalker will only call visitToken for these token types. Because the requirements of our tasks are so simple, there is no need to implement the other fancy methods, like finishTree(), etc., so here is our first shot at our Check implementation:
package com.mycompany.checks;
import com.puppycrawl.tools.checkstyle.api.*;
public class MethodLimitCheck extends AbstractCheck
{
private static final int DEFAULT_MAX = 30;
private int max = DEFAULT_MAX;
@Override
public int[] getDefaultTokens()
{
return new int[]{TokenTypes.CLASS_DEF, TokenTypes.INTERFACE_DEF};
}
@Override
public void visitToken(DetailAST ast)
{
// find the OBJBLOCK node below the CLASS_DEF/INTERFACE_DEF
DetailAST objBlock = ast.findFirstToken(TokenTypes.OBJBLOCK);
// count the number of direct children of the OBJBLOCK
// that are METHOD_DEFS
int methodDefs = objBlock.getChildCount(TokenTypes.METHOD_DEF);
// report error if limit is reached
if (methodDefs > this.max) {
String message = "too many methods, only " + this.max + " are allowed";
log(ast.getLineNo(), message);
}
}
}
There are four methods in Check class to control the processed TokenTypes - one setter setTokens(), which is used to define a custom set (which is different from the default one) of the processed TokenTypes via config file and three getters, which have to be overridden: getDefaultTokens(), getAcceptableTokens(), getRequiredTokens().
Before Checkstyle 6.0, there was no comments in AST tree as comments are not a code and ignored by compiler, so it was not a primary focus of Checkstyle. But more and more requests appear to do validation of code vs javadoc or comments. So there was a simple solution to receive plain text of file in Check, do own parsing and searching for required comments in file.
Since 6.0, there is a new method in AbstractCheck class that allow you to see or not comment nodes in AST Tree - isCommentNodesRequired(). It should return TRUE if Check want to see comments in AST Tree.
It is done as optional because it is change for a AST Tree structure, and non of existing Checks ready for this. Checkstyle does not do re-parse file one more time, comments were already in grammar and parsed, but skipped during AST nodes creation.
Before execution, all Checks are divided into 2 groups (base on isCommentNodesRequired method): "java only Checks", "comments Checks". Checkstyle execute "java only Checks" first, as all them finish, append to AST Tree missed comment AST nodes and call "comments Checks". All Javadoc Checks that are child of AbstractJavadocCheck are "comments Checks".
Comment tokens: SINGLE_LINE_COMMENT, BLOCK_COMMENT_BEGIN, BLOCK_COMMENT_END, COMMENT_CONTENT.
In the example above you already saw that the DetailsAST class provides utility methods to extract information from the tree, like getChildCount(). By now you have probably consulted the API documentation and found that DetailsAST additionally provides methods for navigating around in the syntax tree, like finding the next sibling of a node, the children of a node, the parent of a node, etc.
These methods provide great power for developing complex Checks. Most of the Checks that Checkstyle provides by default use these methods to analyze the environment of the ASTs that are visited by the TreeWalker. Don't abuse that feature for exploring the whole tree, though. Let the TreeWalker drive the tree traversal and limit the visitor to the neighbours of a single AST.
OK Mr. Checkstyle, that's all very nice but in my company we have several projects, and each has another number of allowed methods. I need to control my Check through properties, so where is the API to do that?
Well, the short answer is, there is no API. It's magic. Really!
If you need to make something configurable, just add a setter method to the Check:
public class MethodLimitCheck extends AbstractCheck
{
// code from above omitted for brevity
public void setMax(int limit)
{
max = limit;
}
}
With this code added, you can set the property max for the MethodLimitCheck module in the configuration file. It doesn't get any simpler than that. The secret is that Checkstyle uses JavaBean reflection to set the JavaBean properties. That works for all primitive types like boolean, int, long, etc., plus Strings, plus arrays of these types.
Detecting errors is one thing, presenting them to the user is another. To do that, the Check base class provides several log methods, the simplest of them being Check.log(String). In your Check you can simply use a verbatim error string like in log("Too many methods, only " + mMax + " are allowed"); as the argument. That will work, but it's not the best possible solution if your Check is intended for a wider audience.
If you are not living in a country where people speak English, you may have noticed that Checkstyle writes internationalized error messages, for example if you live in Germany the error messages are German. The individual Checks don't have to do anything fancy to achieve this, it's actually quite easy and the Checkstyle framework does most of the work.
To support internationalized error messages, you need to create or reuse existing a messages.properties file alongside your Check class (example) , i.e. the Java file and the properties files should be in the same directory. Add a symbolic error code and an English representation to the messages.properties. The file should contain the following line: too.many.methods=Too many methods, only {0} are allowed. Then replace the verbatim error message with the symbolic representation and use one of the log helper methods to provide the dynamic part of the message (mMax in this case): log("too.many.methods", mMax);. Please consult the documentation of Java's MessageFormat to learn about the syntax of format strings (especially about those funny numbers in the translated text).
Supporting a new language is very easy now, simply create a new messages file for the language, e.g. messages_fr.properties to provide French error messages. The correct file will be chosen automatically, based on the language settings of the user's operating system.
The great final moment has arrived, you are about to run your Check. To integrate your Check, add a new subentry under the TreeWalker module of your configuration file. Use the full classname of your Check class as the name of the module. Your configuration file config.xml should look something like this:
<?xml version="1.0"?>
<!DOCTYPE module PUBLIC
"-//Puppy Crawl//DTD Check Configuration 1.3//EN"
"http://checkstyle.sourceforge.net/dtds/configuration_1_3.dtd">
<module name="Checker">
<module name="TreeWalker">
<!-- your standard Checks that come with Checkstyle -->
<module name="UpperEll"/>
<module name="MethodLength"/>
<!-- your Check goes here -->
<module name="com.mycompany.checks.MethodLimitCheck">
<property name="max" value="45"/>
</module>
</module>
</module>
To run the new Check on the command line compile your Check, create a jar that contains the classes and property files, e.g. mycompanychecks.jar. Then run (with the path separator system property ("path.separator") adjusted to your platform: Linux/Unix - ":", Windows - ";"):
For Linux/Unix OS:
java -classpath mycompanychecks.jar:checkstyle-${projectVersion}-all.jar \
com.puppycrawl.tools.checkstyle.Main-c config.xml myproject
For Windows OS:
java -classpath mycompanychecks.jar;checkstyle-${projectVersion}-all.jar ^
com.puppycrawl.tools.checkstyle.Main-c config.xml myproject
Did you see all those errors about "too many methods" flying over your screen? Congratulations. You can now consider yourself a Checkstyle expert. Go to your fridge. Have a beer.
Please consult the Checkstyle configuration manual to learn how to integrate your Checks into the package configuration so that you can use MethodLimit instead of the full class name.
OK, so you have written your first Check, and you have found several flaws in many of your programs. You now know that your boss does not follow the coding conventions he wrote. And you know that you are the king of the world. To become a programming god, you want to write your second Check - now wait, first you should know what your limits are.
There are basically only few of them:
This means that you cannot implement some of the code inspection features that are available in advanced IDEs like Eclipse, IntelliJ IDEA, FindBugs, Sonarqube.
For example you will not be able to implement:
- a Check that finds redundant type casts or unused public methods.
- a Check that validate that user custom Exception class inherited from java.lang.Exception class.
Writing a FileSetCheck usually required when you do not need parse Java file to get inner structure, or you are going to validate non "*.java" files.
Writing a FileSetCheck is pretty straightforward: Just inherit from AbstractFileSetCheck and override the abstract processFiltered(java.io.File, java.util.List) method and you're done. A very simple example could fire an error if the number of files exceeds a certain limit. Here is a FileSetCheck that does just that:
package com.mycompany.checks;
import java.io.File;
import java.util.List;
import com.puppycrawl.tools.checkstyle.api.*;
public class LimitImplementationFiles extends AbstractFileSetCheck
{
private static final int DEFAULT_MAX = 100;
private int fileCount;
private int max = DEFAULT_MAX;
public void setMax(int aMax)
{
this.max = aMax;
}
@Override
public void beginProcessing(String aCharset)
{
super.beginProcessing(aCharset);
//reset the file count
this.fileCount = 0;
}
@Override
public void processFiltered(File file, List<String> aLines)
{
this.fileCount++;
if (this.fileCount > this.max) {
// log the message
log(0, "max.files.exceeded", Integer.valueOf(this.max));
// you can call log() multiple times to flag multiple
// errors in the same file
}
}
}
Note that the configuration via bean introspection also applies here. By implementing the setMax() method the FileSetCheck automatically makes "max" a legal configuration parameter that you can use in the Checkstyle configuration file.
There are virtually no limits what you can do in FileSetChecks, but please do not be crazy.
Checkstyle can cache external configuration resources of any kind which are used by your check. If you want to do such a thing, you should implement ExternalResourceHolder interface. Such module must declare external resource locations as a set of Strings which will be returned from getExternalResourceLocations method. This will allow Checkstyle to invalidate (clear) cache when the content of at least one external configuration resource of your check is changed.
ATTENTION!
That's probably our fault, and it means that we have to provide better documentation. Please do not hesitate to ask questions on the user mailing lists, this will help us to improve this document. Please ask your questions as precisely as possible. We will not be able to answer questions like "I want to write a Check but I don't know how, can you help me?". Tell us what you are trying to do (the purpose of the Check), what you have understood so far, and what exactly you are getting stuck on.
We need your help to keep improving Checkstyle. Whenever you write a Check or FileSetCheck that you think is generally useful, please consider contributing it to the Checkstyle community and submit it for inclusion in the next release of Checkstyle.