| Figure 1: Type hierarchy for the Nullness type system. Java's Object is expressed as @Nullable Object. Programmers can omit most type qualifiers, because the default annotation (Section 3.2.2) is usually correct. |
The Checker Framework ManualMIT Program Analysis Group |
For the impatient: Section 1.2 describes how to install and use pluggable type-checkers.
The Checker Framework enhances Java's type system to make it more powerful and useful. This lets software developers detect and prevent errors in their Java programs.
The Checker Framework comes with 4 checkers for specific types of errors:
These checkers are easy to use and are invoked as arguments to javac.
The Checker Framework also enables you to write new checkers of your own; see Sections 7 and 9.
The Checker Framework supports adding pluggable type systems to the Java language in a backward-compatible way. Java's built-in typechecker finds and prevents many errors — but it doesn't find and prevent enough errors. The Checker Framework lets you run an additional typechecker as a plug-in to the javac compiler. Your code stays completely backward-compatible: your code compiles with any Java compiler, it runs on any JVM, and your coworkers don't have to use the enhanced type system if they don't want to. You can check only part of your program, and type inference tools exist to help you annotate your code.
A type system designer uses the Checker Framework to define type qualifiers and their semantics, and a compiler plug-in (a “checker”) enforces the semantics. Programmers can write the type qualifiers in their programs and use the plug-in to detect or prevent errors. The Checker Framework is useful both to programmers who wish to write error-free code, and to type system designers who wish to evaluate and deploy their type systems.
This document uses the terms “checker”, “checker plugin”, “type-checking compiler plugin”, and “annotation processor” as synonyms.
This section describes how to install the binary release of the Checker Framework. The binary release contains everything that you need, both to run checkers and to write your own checkers. As an alternative, the source release (Section 10.3) is useful if you wish to examine or modify the implementation of checkers or of the framework itself.
Requirement: You must have JDK 6 or later installed. You can get JDK 6 from Sun or elsewhere. If you are using Apple Mac OS X, you can either use Apple's implementation or SoyLatte.
cd $JSR308 wget http://groups.csail.mit.edu/pag/jsr308/current/jsr308-checkers.zip unzip jsr308-checkers.zip
The binaries and scripts will be installed in $JSR308/checkers/binary. When using Windows, the files may be in $JSR308/jsr308-checkers/checkers/binary. Add the correct folder to your PATH environment variable.
$ $JSR308/checkers/binary/javac -version javac 1.7.0-jsr308-0.8.8
That's all there is to it! Now you are ready to start using the checkers.
The next section walks you through a simple example. More detailed instructions for using a checker appear in Section 2.
To run a checker on a source file, just run javac as usual, passing the -processor flag. For instance, use the following command (where javac is the JSR 308 compiler):
javac -processor ProcessorName source-files ...
import checkers.nullness.quals.*;
public class GetStarted {
void sample() {
@NonNull Object ref = new Object();
}
}
javac -processor checkers.nullness.NullnessChecker GetStarted.java
The compilation should complete without any errors.
@NonNull Object ref = null;
GetStarted.java:5: incompatible types.
found : @Nullable <nulltype>
required: @NonNull Object
@NonNull Object ref = null;
^
1 error
The type qualifiers (e.g. @NonNull) are permitted anywhere that would write a type, including generics and casts.
@Interned String intern() { ... } // return value
int compareTo(@NonNull String other) { ... } // parameter
@NonNull List<@Interned String> messages; // non-null list of interned Strings
Finding bugs with a checker plugin is a two-step process:
The syntax of type qualifier annotations in Java 7 is specified by JSR 308 [Ern08]. Ordinary Java permits annotations on declarations. JSR 308 permits annotations anywhere that you would write a type, including generics and casts. You can also write annotations to indicate type qualifiers for array levels and receivers. Here are a few examples:
@Interned String intern() { ... } // return value
int compareTo(@NonNull String other) { ... } // parameter
String toString() @ReadOnly { ... } // receiver ("this" parameter)
@NonNull List<@Interned String> messages; // generics: non-null list of interned Strings
@Interned String @NonNull [] messages; // arrays: non-null array of interned Strings
myDate = (@ReadOnly Date) readonlyObject; // cast
Sometimes, your code needs to be compilable by people who are not using the JSR 308 or Java 7 compiler.
A Java 4 compiler does not permit use of annotations, and a Java 5 compiler only permits annotations on declarations (but not on generic arguments, casts, method receiver, etc.).
For backward compatibility, you may write any annotation inside a /*…*/ Java comment, as in List</*@NonNull*/ String>. The JSR 308 compiler treats the code exactly as if you had not written the /* and */. In other words, the JSR 308 compiler will recognize the annotation, but your code will still compile with pre-JSR-308 compilers.
In a single program, you may write some annotations in comments, and others without comments.
By default, the compiler ignores any comment that contains spaces at the beginning or end, or between the @ and the annotation name. This feature enables backward compatibility with code that contains comments that start with @ but are not annotations. (The ESC/Java [FLL+02], JML [LBR06], and Splint [Eva96] tools all use “/*@” or “/* @” as a comment marker.) Compiler flag -Xspacesincomments causes the compiler to parse annotation comments even when they contain spaces. You may need to use -Xspacesincomments if you use Eclipse's “Source > Correct Indentation” command, since it inserts space in comments. But the annotation comments are less readable with spaces, so you may wish to disable inserting spaces: in the Formatter preferences, in the Comments tab, unselect the “enable block comment formatting” checkbox.
When writing source code with annotations, it is more convenient to write a short form such as @NonNull instead of @checkers.nullness.quals.NonNull. There are two ways to do this.
A potential disadvantage of this is that everyone who compiles the code (even using a non-JSR-308 compiler) must have the annotation definitions (e.g., the checkers.jar or checkers-quals.jar file) on their classpath. The reason is that a Java compiler issues an error if an imported package is not on the classpath. See Section 2.1.2.
In bash, you could write export jsr308_imports='checkers.nullness.quals.*', or prefix the javac command by jsr308_imports='checkers.nullness.quals.*' . Alternately, you can set the system variable via the javac command line argument -J-Djsr308_imports="checkers.nullness.quals.*".
You can specify multiple packages separated by the classpath separator (same as the file path separator: ; for Windows, and : for Unix and Mac.). For example, to implicitly import the Nullness and Interning qualifiers, set jsr308_imports to checkers.nullness.quals.*:checkers.interning.quals.*.
If your code contains any annotations outside of comments, or any import statements for the annotations, then your code has a dependency on the annotation declarations. You also will need to provide the annotation declarations as well, if you decide to distribute your project.
For your convenience, the checkers distribution comes with a jar file, checkers-quals.jar, that only contains the distributed qualifiers. You may include the jar file in your distribution.
Your clients need to have the annotations jar in the classpath when compiling your project. When running it though, they most likely don't require the annotations declarations, unless the annotation classes are loaded via reflection.
To run a checker plugin, run the JSR 308 compiler javac as usual, but pass the -processor plugin_class command-line option. Two concrete examples (using the Nullness checker) are:
javac -processor checkers.nullness.NullnessChecker MyFile.java javac -processor checkers.nullness.NullnessChecker -sourcepath checkers/jdk/nullness/src MyFile.java
For a discussion of the -sourcepath argument, see Section 8.2.
The checker is run only on the Java files specified on the command line (or created by another annotation processor). The checker does not analyze other classes (e.g., pre-compiled, or source code is available on the classpath), but it does check the uses of those classes in the source code being compiled.
You can always compile the code without the -processor command-line option, but in that case no checking of the type annotations is performed.
You can also direct the JSR 308 compiler javac to always run a checker plugin, without explicitly invoking the checker at every compilation. You can specify the checkers to run via a configuration file that you place in the CLASSPATH.
You need to create a configuration file named META-INF/services/javax.annotation.processing.Processor, which contains the names of the checker plugins to be used, listed one per line. For instance, to run the Nullness and the Interning checkers automatically, the configuration file should contain:
checkers.nullness.NullnessChecker checkers.interning.InterningChecker
You can disable this auto-discovery mechanism by passing -proc:none option to javac.
Future releases of the Checker Framework would make the distributed supported checkers auto-discoverable by default.
If you use the Ant build tool to compile your software, then you can add an Ant task that runs a checker. We assume that your Ant file already contains a compilation target that uses the javac task. Duplicate that target, then modify it slightly as indicated in this example:
<property environment="env"/>
<target name="check-nullness" description="Check for nullness errors." depends="clean">
<javac ...
fork="yes"
executable="${env.JSR308}/jdk1.7.0/bin/javac">
<compilerarg value="-version"/>
<compilerarg line="-target 5"/>
<compilerarg value="-proc:only"/>
<compilerarg line="-processor checkers.nullness.NullnessChecker"/>
<compilerarg line="-sourcepath ${env.annotations}/checkers/jdk/nullness/src"/>
<classpath>
<pathelement location="${env.annotations}/checkers/checkers.jar"/>
</classpath>
...
</javac>
</target>
The property target makes environment variables (such as your home directory) available to Ant.
In the example, the target is named check-interning, but you can name it whatever you like.
The target assumes the existence of a clean target that removes all .class files. That is necessary because Ant's javac target doesn't re-compile .java files for which a .class file already exists.
The executable and fork fields of the javac task ensure that an external javac program is called. Otherwise, Ant will run javac via a Java method call, and there is no guarantee that it will get the JSR 308 version.
The -version compiler argument is just for debugging; you may omit it.
The -target 5 compiler argument is optional, if you use Java 5 in ordinary compilation when not performing pluggable type-checking.
The -proc:only compiler argument is optional, to work around a known bug (see Section 10.1.1).
The -processor ... compiler argument indicates which checker to run. You can supply additional arguments to the checker as well.
Adam Warski has written a Maven2 plugin that runs a checker. The plugin is available at http://www.warski.org/checkersplugin.html.
There are two ways to run a checker from within the Eclipse IDE: via Ant or using an Eclipse plug-in.
Add an Ant target as described in Section 2.2.2. You can run the Ant target by executing the following steps (instructions copied from http://www.eclipse.org/documentation/?topic=/org.eclipse.platform.doc.user/gettingStarted/qs-84_run_ant.htm):
A prototype Eclipse plug-in for running a checker is available at http://groups.csail.mit.edu/pag/jsr308/eclipse/. The website contains instructions for installing and using the plug-in. The plug-in is experimental now, but some people have used it successfully (and we have fixed all bugs that have been reported so far).
Sometimes, you wish to type-check only part of your program. You might focus on the most mission-critical or error-prone part of your code. When you start to use a checker, you may not wish to annotate your entire program right away. You may not have source code (or enough knowledge to annotate) the libraries that your program uses.
If annotated code uses unannotated code, then the checker may issue warnings. For example, the Nullness checker (Section 3) will warn whenever an unannotated method result is used in a non-null context:
@NonNull myvar = unannotated_method(); // WARNING: unannotated_method may return a null value
If the call can return null, you should fix the bug in your program by removing the @NonNull annotation in your own program.
If the library call never returns null, there are several ways to eliminate the compiler warnings.
Section 8 discusses adding annotations to signatures when you do not have source code available. Section 2.4 discusses suppressing warnings.
If you annotate additional libraries, please share them with us so that we can distribute the annotations with the Checker Framework; see Section 10.2.
You may wish to suppress checker warnings because of unannotated libraries or un-annotated portions of your own code, because of application invariants that are beyond the capabilities of the type system, because of checker limitations, because you are interested in only some of the guarantees provided by a checker, or for other reasons. You can suppress warnings via
You can suppress specific errors and warnings by use of the @SuppressWarnings("annotationname") annotation, for example @SuppressWarnings("interning"). This may be placed on program elements such as a class, method, or local variable declaration. It is good practice to suppress warnings in the smallest possible scope. For example, if a particular expression causes a false positive warning, you should extract that expression into a local variable and place a @SuppressWarnings annotation on the variable declaration. As another example, if you have annotated the signatures but not the bodies of the methods in a class or package, put a @SuppressWarnings annotation on the class declaration or on the package's package-info.java file.
You can suppress all errors and warnings at all uses of a given class.
Set the checkers.skipClasses Java property to a
regular expression that matches classes for which warnings and errors
should be suppressed. For example, if you use
“-Dcheckers.skipClasses=^java\.” on the command line when invoking
javac, then the checkers will suppress all warnings within those
classes, all warnings relating to invalid arguments, and all warnings
relating to incorrect use of the return value.
(Note that if your javac is a script rather than a binary, it may
not support JVM flags such as -D; in that case, you may need to edit
javac script itself to pass the -D flag. This is a flaw in
the OpenJDK build process, which we will try to correct in a future release.)
You can suppress an entire class of warnings via javac's -Alint command-line option. The -Alint option uses the same syntax as javac's -Xlint option. Following -Alint=, write a list of option names. If the option name is preceded by a hyphen (-), that disables the option; otherwise it enables it. For example: -Alint=-dotequals causes the Interning checker (Section 4) not to output advice about when a.equals(b) could be replaced by a==b.
You can also compile parts of your code without use of the -processor switch to javac. No checking is done during such compilations.
Finally, some checkers have special rules. For example, the Nullness checker (Section 3) uses assert statements that contain null checks to suppress warnings (Section 3.3.1).
The Checker Framework fully supports qualified Java generic types (also known in the literature as “parametric polymorphism”). Before running a checker, we recommend that you eliminate raw types (e.g., List as opposed to List<...>) from your code. Using generics helps prevent type errors just as using a pluggable type-checker does.
When instantiating a generic type, clients supply the qualifier along with the type argument, as in List<@NonNull String>.
The declaration (that is, the implementation) of a generic class may use the extends clause to restrict the types and qualifiers that may be used for instantiating. For example, given the declaration class MyClass<T extends @NonNull Object> {...}, a client could use MyClass<@NonNull String> but not MyClass<@Nullable String>.
Style note: When using the Nullness checker (Section 3), programmers sometimes write extends @NonNull Object even though it's the default. The reason is that code with no extends clause, like
class C<T> { ... }
typically means that class C can be instantiated with any type argument at all. But in the Nullness type system, to permit all type arguments, to obtain that effect one must write
class C<T extends @Nullable Object> { ... }
A type annotation on a generic type variable overrides/ignores any type qualifier (in the same type hierarchy) on the corresponding actual type argument. For example, @Nullable T applies the type qualifier @Nullable to the (unqualified) Java type of the type argument T.
Here is an example of applying a type annotation to a generic type variable:
class MyClass2<T> {
...
@Nullable T = null;
...
}
The type annotation does not restrict how MyClass2 may be instantiated (only the optional extends clause on the declaration of type variable T would do so). In other words, both MyClass2<@NonNull String> and MyClass2<@Nullable String> are legal, and in both cases @Nullable T means @Nullable String. In MyClass2<@Interned String>, @Nullable T means @Nullable @Interned String.
The Checker Framework also supports type qualifier polymorphism for methods, which permits a single method to have multiple different qualified type signatures.
A polymorphic qualifier's definition is marked with @PolymorphicQualifier. For example, @PolyNull is a polymorphic type qualifier for the Nullness type system:
@PolymorphicQualifier
public @interface PolyNull { }
A method written using a polymorphic qualifier conceptually has multiple versions, somewhat like a template in C++. In each version, the polymorphic qualifier has been replaced by another qualifier from the hierarchy. See the examples in Section 2.5.3.
The method body must type-check with all signatures. A method call is type-correct if it type-checks under any signature.
Polymorphic qualifiers can be used within a method body. They may not be used on classes or fields.
As an example of the use of @PolyNull, method Class.cast returns null if and only if its argument is null:
@PolyNull T cast(@PolyNull Object obj) { ... }
This is like writing:
@NonNull T cast( @NonNull Object obj) { ... }
@Nullable T cast(@Nullable Object obj) { ... }
except that the latter is not legal Java, since it defines two methods with the same Java signature.
As another example, consider
@PolyNull T max(@PolyNull T x, @PolyNull T y);
which is like writing
@NonNull T max( @NonNull T x, @NonNull T y); @Nullable T max(@Nullable T x, @Nullable T y);
One way of thinking about which one of the two variants is selected is that the nullness annotations of (the declared types of) both arguments are unified to a type that is a subtype of both. If both arguments are @NonNull, their unification is @NonNull, and the result is @NonNull. But if even one of the arguments is @Nullable, then the result is @Nullable.
It does not make sense to write only a single instance of a polymorphic qualifier in a method definition, as in
void m(@PolyNull Object obj)
which expands to
void m(@NonNull Object obj) void m(@Nullable Object obj)
which is no different than writing just
void m(@Nullable Object obj)
The benefit of polymorphic qualifiers comes when one is used multiple times in a method, since then each instance turns into the same type qualifier. Most frequently, the polymorphic qualifier appears on both the return type and at least one formal parameter. It can also be useful to have polymorphic qualifiers on (only) multiple formal parameters, especially if the method side-effects one of its arguments.
A checker sometimes treats a type as having a slightly different qualifier than what is written on the type — especially if the programmer wrote no qualifier at all. The following steps determine the effective qualifier on a type — the qualifier that the checkers treat as being present.
At this point, every type has a qualifier.
Most readers can skip this section on first reading, because you will probably find the system simply works as you would prefer, without forcing you to write too many qualifiers in your program.
A programmer can cause unannotated references to be treated as if they had a default annotation.
The default for unannotated references is sometimes determined by the type system; in such cases, specifying a default is not sensible. For example, the Interning type system has unqualified types and @Interned types; no different meaning for unannotated types may be specified. However, the Nullness type system has @Nullable types and @NonNull types, with no built-in meaning for unannotated types; a user may specify a default qualifier.
The user specifies a default qualifier by writing the @DefaultQualifier annotation on a package (via the package-info.java file), class, method, or variable declaration. The argument to @DefaultQualifier is the fully qualified String name of an annotation, and its optional second argument indicates where the default applies. If the second argument is omitted, the specified annotation is the default in all locations. See the Javadoc of DefaultQualifier for details.
The user could specify multiple default qualifiers by writing @DefaultQualifiers annotations in all the locations that accept @DefaultQualifier. DefaultQualifiers accept an array of DefaultQualifier arguments.
This example shows @DefaultQualifier and a @DefaultQualifiers annotations for the Nullness type system (Section 3) and the IGJ type system (Section 5):
@DefaultQualifiers({
@DefaultQualifier("checkers.nullness.quals.NonNull"),
@DefaultQualifier("checkers.igj.quals.Mutable")
})
class MyClass {
public boolean compile(File myFile) {// myFile type "@NonNull @Mutable File"
if (!myFile.exists()) // no warning: myFile is non-null
return false;
@Nullable File srcPath = ...; // must annotate to specify "@Nullable File"
...
if (srcPath.exists()) // warning: srcPath might be null
...
}
@DefaultQualifier("checkers.igj.quals.ReadOnly")
public boolean isJavaFile(File myfile) { // myFile type "@NonNull @ReadOnly File"
...
}
}
In order to reduce the burden of annotating types in your program, the checkers soundly treat certain variables and expressions as having a subtype of their declared or defaulted (Section 2.6.1) type. This functionality never introduces unsoundness or causes an error to be missed: it merely suppresses false positive warnings.
By default, all checkers, including new checkers that you write, can take advantage of this functionality. Most of the time, users don't have to think about, and may not even notice, this feature of the framework. The checkers simply do the right thing even when a programmer forgets an annotation on a local variable, or when a programmers writes an unnecessarily general type in a declaration.
If you are curious or want more details about this feature, then read on.
As an example, the Nullness checker (Section 3) can automatically determine that certain variables are non-null, even if they were explicitly or by default annotated as nullable. A variable or expression can be treated as @NonNull from the time that it is either assigned a non-null value or checked against null (e.g., via an assertion, if statement, or being dereferenced), until it might be re-assigned (e.g., via an assignment that might affect this variable, or via a method call that might affect this variable).
As with explicit annotations, the implicitly non-null types permit dereferences and assignments to explicitly non-null types, without compiler warnings.
Consider this code, along with comments indicating whether the Nullness checker (Section 3) issues a warning. Note that the same expression may yield a warning or not depending on its context.
// Requires an argument of type @NonNull String
void parse(@NonNull String toParse) { ... }
// Argument does NOT have a @NonNull type
void lex(String toLex) {
parse(toLex); // warning: toLex might be null
if (toLex != null) {
parse(toLex); // no warning: toLex is known to be non-null
}
parse(toLex); // warning: toLex might be null
toLex = new String(...);
parse(toLex); // no warning: toLex is known to be non-null
}
If you find instances where you think a value should be inferred to have (or not have) a given annotation, but the checker does not do so, please submit a bug report (see Section 10.2) that includes a small piece of Java code that reproduces the problem.
Type inference is never performed for method parameters of non-private methods and for non-private fields, because unknown client code could use them in arbitrary ways. The inferred information is never written to the .class file as user-written annotations are.
The inference indicates when a variable can be treated as having a subtype of its declared type — for instance, when an otherwise nullable type can be treated as a @NonNull one. The inference never treats a variable as a supertype of its declared type (e.g., an expression of @NonNull type is never inferred to be treated as possibly-null).
A checker can guarantee that a particular property holds throughout the code. For example, the Nullness checker (Section 3) guarantees that every expression whose type is a @NonNull type never evaluates to null. The Interning checker (Section 4) guarantees that every expression whose type is an @Interned type evaluates to an interned value. The checker makes its guarantee by examining every part of your program and verifying that no part of the program violates the guarantee.
There are some limitations to the guarantee.
A checker can be useful in finding bugs or in verifying part of a program, even if the checker is unable to verify the correctness of an entire program.
If you find that a checker fails to issue a warning that it should, then please report a bug (see Section 10.2).
In the checkers distributed with the Checker Framework, an annotation on a constructor invocation is equivalent to a cast on a constructor result. That is, the following two expressions have identical semantics: one is just shorthand for the other.
new @ReadOnly Date() (@ReadOnly Date) new Date()
If the Nullness checker issues no warnings for a given program, then running that program will never throw a null pointer exception. This guarantee enables a programmer to prevent errors from occurring when his program is run. See Section 3.3 for more details about the guarantee and what is checked.
You can control the behavior of the Nullness checker via the -Alint options flow, cast, and cast:redundant.
Four qualifiers are part of the Nullness type system.
During execution of a constructor, every field of non-primitive type starts out with the value null. If the field has @NonNull type, the value null violates the type. If the constructor makes a method call (passing this as a parameter or the receiver), then the called method could observe the object in an illegal state.
The @Raw type annotation represents a partially-initialized object. If a reference has @Raw type, then all fields fields are treated as @Nullable. Within the constructor, this has @Raw type and can only be passed to methods when the corresponding parameter is annotated with @Raw. Similar restrictions apply to assigning this to a field.
The Nullness checker does not yet implement strengthening of @Raw types — permitting a class to access its superclass's @NonNull fields as @NonNull as soon the superclass is fully initialized. Most users do not need such strengthening.
The name “raw” comes from a research paper that proposed this approach [FL03]. The @Raw annotation has nothing to do with the raw types of Java Generics.
As described in Section 2.6, the Nullness checker adds implicit qualifiers, reducing the number of annotations that must appear in your code. For example, enum types are implicitly non-null, so you never need to write @NonNull MyEnumType.
For a complete description of all implicit nullness qualifiers, see the Javadoc for NullnessAnnotatedTypeFactory.
Unannotated references are treated as if they had a default annotation (see Section 2.6.1, which also gives an example of how to specify a default by use of the @DefaultQualifier annotation).
There are three possible defaults:
The NNEL default leads to the smallest number of explicit annotations in your code [PAC+08]. It is what we recommend. If you do not explicitly specify a different default, then NNEL is the default.
By default, reference types are nullable in Java, but sometimes it is
Figure 1: Type hierarchy for the Nullness type system. Java's Object is expressed as @Nullable Object. Programmers can omit most type qualifiers, because the default annotation (Section 3.2.2) is usually correct.
Unannotated references are treated as if they had a default annotation; see Section 3.2.2.
It can be tedious to write annotations in your code. Two tools exist that can automatically infer annotations and insert them in your source code. (This is different than type qualifier refinement for local variables (Section 2.6.2), which infers a more specific type for local variables and uses them during type-checking but does not insert them in your source code. Type qualifier refinement is always enabled, no matter how annotations on signatures got inserted in your source code.)
Your choice of tool depends on what default annotation (see Section 3.2.2) your code uses. You only need one of these tools.
The checker issues a warning in two cases:
This example shows both sorts of problems:
Object obj; // might be null
@NonNull Object nnobj; // never null
...
obj.toString() // checker warning: dereference might cause null pointer exception
nnobj = obj; // checker warning: nnobj may become null
Parameter passing and return values are checked analogously to assignments.
In addition to the other ways of suppressing warnings (Section 2.4), the Nullness checker assumes that assertions succeed. For example, it assumes that no null pointer exception can occur in code such as
assert x != null; ... x.f ...
(Another way of stating the Nullness checker's use of assertions is as an additional caveat to the guarantees provided by a checker (Section 2.7). The Nullness checker prevents null pointer errors in your code under the assumption that assertions are enabled, and it does not guarantee that all of your assertions succeed.)
To try the Nullness checker on a source file that uses the @NonNull qualifier, use the following command (where javac is the JSR 308 compiler):
javac -processor checkers.nullness.NullnessChecker examples/NullnessExample.java
Compilation will complete without warnings.
To see the checker warn about incorrect usage of annotations (and therefore the possibility of a null pointer exception at run time), use the following command:
javac -processor checkers.nullness.NullnessChecker examples/NullnessExampleWithWarnings.java
The compiler will issue three warnings regarding violation of the semantics of @NonNull.
The Nullness checker itself is annotated with @NonNull.
In addition, you can run the Nullness checker on the annotation scene library, another library that has been fully annotated with @NonNull. To run the Nullness checker on the annotation scene library, first download the scene library suite (which includes build dependencies for the scene library as well as its source code) and extract it into your checkers installation. The checker can then be run on the annotation scene library with Apache Ant using the following commands:
cd checkers ant -f scene-lib-test.xml
You can view the annotated source code, which contains @NonNull annotations, in the checkers/scene-lib-test/src/annotations/ directory.
The Checker Framework @NonNull annotation is similar, but not identical, to the @NotNull annotation of IntelliJ IDEA, the @NonNull annotation of FindBugs, the non_null modifier of JML, and annotations proposed by JSR 305, among others. Also see Section 10.5 for a comparison to other tools.
If the Interning checker issues no warnings for a given program, then all reference equality tests (i.e., “==”) in that program operate on interned types. Interning is a design pattern in which the same object is used whenever two different objects would be considered equal. Interning is also known as canonicalization or hash-consing, and it is related to the flyweight design pattern. Interning can save memory and can speed up testing for equality by permitting use of ==; however, use of == on non-interned values can result in subtle bugs. For example:
Integer x = new Integer(22); Integer y = new Integer(22); System.out.println(x == y); // prints false!
The Interning checker helps programmers to prevent such bugs. The Interning checker also helps to prevent performance problems that result from failure to use interning. (See Section 2.7 for caveats to the checker's guarantees.)
Two qualifiers are part of the Interning type system.
Figure 2: Type hierarchy for the Interning type system.
In order to perform checking, you must annotate your code with the @Interned type annotation, which indicates a type for the canonical representation of an object:
String s1 = ...; // type is (uninterned) "String" @Interned String s2 = ...; // Java type is "String", but checker treats it as "Interned String"
The type system enforced by the checker plugin ensures that only interned values can be assigned to s2.
To specify that all objects of a given type are interned, annotate the class declaration:
public @Interned class MyInternedClass { ... }
This is equivalent to annotating every use of MyInternedClass, in a declaration or elsewhere. For example, enum classes are implicitly so annotated.
As described in Section 2.6, the Interning checker adds implicit qualifiers, reducing the number of annotations that must appear in your code. For example, String literals and the null literal are always considered interned, and object creation expressions (using new) are never considered @Interned unless they are annotated as such, as in
@Interned Double internedDoubleZero = new @Interned Double(0); // canonical representation for Double zero
For a complete description of all implicit interning qualifiers, see the Javadoc for InterningAnnotatedTypeFactory.
Objects of an @Interned type may be safely compared using the “==” operator.
The checker issues a warning in two cases:
This example shows both sorts of problems:
Object obj;
@Interned Object iobj;
...
if (obj == iobj) { ... } // checker warning: reference equality test is unsafe
iobj = obj; // checker warning: iobj's referent may no longer be interned
The checker also issues a warning when .equals is used where == could be safely used. You can disable this behavior via the javac -Alint command-line option, like so: -Alint=-dotequals.
For a complete description of all checks performed by the checker, see the Javadoc for InterningVisitor.
To try the Interning checker on a source file that uses the @Interned qualifier, use the following command (where javac is the JSR 308 compiler):
javac -processor checkers.interning.InterningChecker examples/InterningExample.java
Compilation will complete without warnings.
To see the checker warn about incorrect usage of annotations, use the following command:
javac -processor checkers.interning.InterningChecker examples/InterningExampleWithWarnings.java
The compiler will issue a warning regarding violation of the semantics of @Interned.
The Daikon invariant detector (http://groups.csail.mit.edu/pag/daikon/) is also annotated with @Interned.
IGJ is a Java language extension that helps programmers to avoid mutation errors (unintended side effects). If the IGJ checker issues no warnings for a given program, then that program will never change objects that should not be changed. This guarantee enables a programmer to detect and prevent mutation-related errors. (See Section 2.7 for caveats to the guarantee.)
IGJ permits a programmer to express that a particular object should never be modified via any reference (object immutability), or that a reference should never be used to modify its referent (reference immutability). Once a programmer has expressed these facts, an automatic checker analyzes the code to either locate mutability bugs or to guarantee that the code contains no such bugs.
Figure 3: Type hierarchy for three of IGJ's type qualifiers.
To learn the details of the IGJ language and type system, please see the ESEC/FSE 2007 paper “Object and reference immutability using Java generics” [ZPA+07]. The IGJ checker supports Annotation IGJ (Section 5.5), which is slightly different dialect of IGJ than that described in the ESEC/FSE paper.
The following qualifiers are part of the IGJ type system.
For additional details, see [ZPA+07].
The IGJ checker issues an error whenever mutation happens through a readonly reference, when fields of a readonly reference which are not explicitly marked with @Assignable are reassigned, or when a readonly expression is assigned to a mutable variable. The checker also emits a warning when casts increase the mutability access of a reference.
As described in Section 2.6, the IGJ checker adds implicit qualifiers, reducing the number of annotations that must appear in your code.
For a complete description of all implicit nullness qualifiers, see the Javadoc for NullnessAnnotatedTypeFactory.
The default annotation (for types that are unannotated and not given an implicit qualifier) is as follows:
interface List<T extends Object> { ... }
is defaulted to
interface List<T extends @Readonly Object> { ... }
This default is not backward-compatible — that is, you may have to explicitly add @Mutable annotations to some type parameter bounds in order to make unannotated Java code type-check under IGJ. However, this reduces the number of annotations you must write overall (since most variables of generic type are in fact not modified), and permits more client code to type-check (otherwise a client could not write List<@Readonly Date>).
The IGJ checker supports the Annotation IGJ dialect of IGJ. The syntax of Annotation IGJ is based on JSR 308 annotations.
The syntax of the original IGJ dialect [ZPA+07] was based on Java 5's generics and annotation mechanisms. The original IGJ dialect was not backward-compatible with Java (either syntactically or semantically). The dialect of IGJ checked by the IGJ checker corrects these problems.
The differences between the Annotation IGJ dialect and the original IGJ dialect are as follows.
Vector<Mutable, Integer> <: Vector<ReadOnly, Integer>
<: Vector<ReadOnly, Number>
<: Vector<ReadOnly, Object>
@I is a template annotation over IGJ Immutability annotations. It acts similarly to type variables in Java's generic types, and the name @I mimics the standard <I> type variable name used in code written in the original IGJ dialect. The annotation value string is used to distinguish between multiple instances of @I — in the generics-based original dialect, these would be expressed as two type variables <I> and <J>.
A class annotated with @I could be declared with any IGJ Immutability annotation. The actual immutability that @I is resolved to dictates the immutability type for all the non-static appearances of @I with the same value as the class declaration.
Example:
@I
public class FileDescriptor {
private @Immutable Date creationData;
private @I Date lastModData;
public @I Date getLastModDate() @ReadOnly { }
}
...
void useFileDescriptor() {
@Mutable FileDescriptor file =
new @Mutable FileDescriptor(...);
...
@Mutable Data date = file.getLastModDate();
}
In the last example, @I was resolved to @Mutable for the instance file.
For example, it could be used for method parameters, return values, and the actual IGJ immutability value would be resolved based on the method invocation.
For example, the below method getMidpoint returns a Point with the same immutability type as the passed parameters if p1 and p2 match in immutability, otherwise @I is resolved to @ReadOnly:
static @I Point getMidpoint(@I Point p1, @I Point p2) { ... }
The @I annotation value distinguishes between @I declarations. So, the below method findUnion returns a collection of the same immutability type as the first collection parameter:
static <E> @I("First") Collection<E> findUnion(@I("First") Collection<E> col1,
@I("Second") Collection<E> col2) { ... }
To try the IGJ checker on a source file that uses the IGJ qualifier, use the following command, where javac is the JSR 308 compiler.
javac -processor checkers.igj.IGJChecker examples/IGJExample.java
The IGJ checker itself is also annotated with IGJ annotations.
Javari [TE05, QTE08] is a Java language extension that helps programmers to avoid mutation errors that result from unintended side effects. If the Javari checker issues no warnings for a given program, then that program will never change objects that should not be changed. This guarantee enables a programmer to detect and prevent mutation-related errors. (See Section 2.7 for caveats to the guarantee.) The Javari webpage (http://groups.csail.mit.edu/pag/javari/) contains papers that explain the Javari language and type system. By contrast to those papers, the Javari checker uses an annotation-based dialect of the Javari language.
The Javarifier tool infers Javari types for an existing program; see Section 6.2.2.
Also consider the IGJ checker (Section 5). The IGJ type system is more expressive than that of Javari, and the IGJ checker is a bit more robust. However, IGJ lacks a type inference tool such as Javarifier.
Figure 4: Type hierarchy for Javari's ReadOnly type qualifier.
Five annotations are part of the Javari type system.
A programmer can write five annotations: @ReadOnly, @Mutable, @Assignable, @PolyRead, and @QReadOnly.
As described in Section 2.6, the Javari checker adds implicit qualifiers, reducing the number of annotations that must appear in your code.
For a complete description of all implicit nullness qualifiers, see the Javadoc for JavariAnnotatedTypeFactory.
It can be tedious to write annotations in your code. The Javarifier tool (http://groups.csail.mit.edu/pag/javari/javarifier/) infers Javari types for an existing program. It automatically inserts Javari annotations in your Java program or in in .class files.
This has two benefits: it relieves the programmer of the tedium of writing annotations (though the programmer can always refine the inferred annotations), and it annotates libraries, permitting checking of programs that use those libraries.
The checker issues an error whenever mutation happens through a readonly reference, when fields of a readonly reference which are not explicitly marked with @Assignable are reassigned, or when a readonly expression is assigned to a mutable variable. The checker also emits a warning when casts increase the mutability access of a reference.
To try the Javari checker on a source file that uses the Javari qualifier, use the following command, where javac is the JSR 308 compiler, or specify just one of the test files.
javac -processor checkers.javari.JavariChecker tests/javari/*.java
The compiler should issue the errors and warnings (if any) specified in the .out files with same name.
To run the test suite for the Javari checker, use ant javari-tests.
The Javari checker itself is also annotated with Javari annotations.
The Basic checker enforces only subtyping rules. It operates over annotations specified by a user on the command line. Thus, users can create a simple type checker without writing any code beyond definitions of the type qualifier annotations.
The Basic checker can accommodate all of the type system enhancements that can be declaratively specified (see Section 9). This includes type introduction rules (implicit annotations, e.g., literals are implicitly considered @NonNull) via the @ImplicitFor meta-annotation, and other features such as flow-sensitive type qualifier inference (Section 2.6.2) and qualifier polymorphism (Section 2.5.2).
The Basic checker is also useful to type system designers who wish to experiment with a checker before writing code; the Basic checker demonstrates the functionality that a checker inherits from the Checker Framework.
For type systems that require special checks (e.g., warning about dereferences of possibly-null values), you will need to write code and extend the framework as discussed in Section 9.
The Basic checker is used in the same way as other checkers (using the -processor option; see Section 2), except that it requires an additional annotation processor argument via the standard “-A” switch:
The annotations listed in -Aquals must be accessible to the compiler during compilation, either on the classpath or sourcepath or as one of the .java files passed to the compiler.
Consider a hypothetical Encrypted type qualifier, which denotes that the representation of an object (such as a String, CharSequence, or byte[]) is encrypted. To use the Basic checker for the Encrypted type system, follow three steps.
package myquals;
/**
* Denotes that the representation of an object is encrypted.
* ...
*/
@TypeQualifier
@SubtypeOf(Unqualified.class)
public @interface Encrypted {}
public @Encrypted String encrypt(String text) {
// ...
}
// Only send encrypted data!
public void sendOverInternet(@Encrypted String msg) {
// ...
}
void sendText() {
// ...
@Encrypted String ciphertext = encrypt(plaintext);
sendOverInternet(ciphertext);
// ...
}
void sendPassword() {
String password = getUserPassword();
sendOverInternet(password);
}
You may also need to add @SuppressWarnings annotations to the encrypt and decrypt methods.
$ javac -processor checkers.basic.BasicChecker -Aquals=myquals.Encrypted YourProgram.java
YourProgram.java:42: incompatible types.
found : java.lang.String
required: @myquals.Encrypted java.lang.String
sendOverInternet(password);
^
When annotated code uses an unannotated library, a checker may issue warnings. As described in Section 2.3, the best way to correct this problem is to add annotations to the library. This section tells you how to add annotations to a library for which you have no source code, because the library is library distributed only binary (.class or .jar) form.
The Checker Framework distribution contains annotations for popular libraries, such as the JDK. If you annotate additional libraries, please share them with us so that we can distribute the annotations with the Checker Framework; see Section 10.2.
You can determine the correct annotations for a library either automatically by running an inference tool, or manually by reading the documentation. Presently, type inference tools are available for the Nullness (Section 3.2.3) and Javari (Section 6.2.2) type systems.
You can make the annotations known to the JSR 308 compiler (and thus to the checkers) in two ways.
A stub file contains “stub classes” that contain annotated signatures. A checker uses those annotated signatures at compile time, instead of or in addition to annotations that appear in the library.
Section 8.1.1 describes how to create stub classes. Section 8.1.2 describes how to use stub classes. These sections illustrate stub classes via the example of creating a @Interned-annotated version of java.lang.String. (You don't need to repeat these steps, since such a stub class is already included in the Checker Framework distribution; see file checkers/src/checkers/interning/jdk.astub, which is reproduced in Section 8.1.3.)
cd nullness-stub java checkers.util.stub.StubGenerator java.lang.String > String.astub
Supply it with the fully-qualified name of the class for which you wish to generate a stub class. The stub class generator prints the stub class to standard out, so you may wish to redirect its output to a file.
import checkers.interning.quals.Interned;
@Interned String intern();
You may also remove irrelevant parts of the stub file; see Section 8.1.3.
When you run javac with a given checker/processor, you can specify a list of the stub files or directories using -Astubs=file_or_path_name. The stub path entries are delimited by File.pathSeparator (`:' for Linux and Mac, `;' for Windows). When you supply a stub directory, the checker only considers the enclosed stub files whose names end with .astub.
The -Astubs argument causes the Checker Framework to read annotations from annotated stub classes in preference to the unannotated original library classes.
javac -processor checkers.interning.InterningChecker -Astubs=String.astub:stubs MyFile.java MyOtherFile.java ...
The stub file format is designed for simplicity, readability, and compactness. It reads like a Java file but contains only the necessary information for type checking.
As an illustration, the stub file for the Interning type system (Section 4) is as follows. This file appears as checkers/src/checkers/interning/jdk.astub in the Checker Framework distribution.
import checkers.interning.quals.Interned;
package java.lang;
// All instances of Class are interned.
@Interned class Class<T> { }
class String {
// The only interning-related method in the JDK.
@Interned String intern();
}
You can use a regular Java file as a stub file. However, you can omit information that is not relevant to pluggable type-checking; this makes the stub file smaller and easier for people to read and write. You can also put annotated signatures for multiple classes in a single stub file.
The stub file format differs from Java source code in the following ways:
The Checker Framework stub file reader has several limitations:
The Checker Framework distribution contains annotated JDKs at the path checkers/jdk/[checker-name]/src. These are in another format called “skeleton classes”. We are currently working on converting the skeleton files into stub files.
The -sourcepath argument causes the compiler to read annotations from annotated skeleton classes in preference to the unannotated original library classes. However, the compiler will use the originals on the classpath if no file is available on the sourcepath.
javac -processor checkers.nullness.NullnessChecker -sourcepath checkers/jdk/nullness/src my_source_files
This section describes how to extend the Checker Framework to create a checker — a type-checking compiler plugin that detects bugs or verifies their absence. After a programmer annotates a program, the checker plugin verifies that the code is consistent with the annotations. If you only want to use a checker, you do not need to read this section.
Writing a simple checker is easy! For example, here is a complete, useful type checker:
@TypeQualifier
@SubtypeOf(Unqualified.class)
public @interface Encrypted {}
This checker is so short because it builds on the Basic Checker (Section 7). See Section 7.2 for more details about this particular checker.
You can also customize a typestate checker. Two of these are available. One is by Adam Warski: http://www.warski.org/typestate.html. The other is by Daniel Wand: http://typestate.ewand.de/.
The rest of this section contains many details for people who want to more write powerful checkers. You won't need all of the details, at least at first. In addition to reading this section of the manual, you may find it helpful to examine the implementations of the checkers that are distributed with the Checker Framework, or to create your checker by modifying another one. The Javadoc documentation of the framework and the checkers is in the distribution and is also available online at http://pag.csail.mit.edu/jsr308/current/doc/.
If you write a new checker, let us know so we can mention it here, link to it from our webpages, or include it in the Checker Framework distribution.
The Checker Framework provides abstract base classes (default implementations), and a specific checker overrides as little or as much of the default implementations as necessary. Sections 9.2–9.5 describe the components of a type system as written using the Checker Framework:
A type system designer specifies the qualifiers in the type system and the type hierarchy that relates them.
Type qualifiers are defined as Java annotations [Dar06]. In Java, an annotation is defined using the Java @interface keyword. Write the @TypeQualifier annotation on the annotation definition to indicate that the annotation represents a type qualifier (e.g., @NonNull or @Interned) and should be processed by the checker. For example:
// Define an annotation for the @NonNull type qualifier.
@TypeQualifier
public @interface NonNull { }
(An annotation that is written on an annotation definition, such as @TypeQualifier, is called a meta-annotation.)
The type hierarchy induced by the qualifiers can be defined either declaratively via meta-annotations (Section 9.2.1), or procedurally through subclassing QualifierHierarchy or TypeHierarchy (Section 9.2.2).
To set a default annotation (which a user may override), use the QualifierDefaults.setAbsoluteDefaults method. You may do this even if you have declaratively defined the qualifier hierarchy; see the Nullness checker's implementation for an example. Recall that defaults are distinct from implicit annotations; see Sections 2.6 and 9.3.
Declaratively, the type system designer uses two meta-annotations (written on the declaration of qualifier annotations) to specify the qualifier hierarchy.
@TypeQualifier
@SubtypeOf( { Nullable.class } )
public @interface NonNull { }
(The actual definition of NonNull is slightly more complex.)
@SubtypeOf accepts multiple annotation classes as an argument, permitting the type hierarchy to be an arbitrary DAG. For example, in the IGJ type system (Section 5.2), @Mutable and @Immutable induce two mutually exclusive subtypes of the @ReadOnly qualifier.
As a special case, the root qualifier needs to be annotated with @Subtype( ). The root qualifier is the qualifier that is a supertype of all other qualifiers. Nullable is the root of the Nullness type system, hence is defined as:
@TypeQualifier
@SubtypeOf( { } )
public @interface Nullable { }
All type qualifiers, except for polymorphic qualifiers, need to be properly annotated with SubtypeOf.
@TypeQualifier
@PolymorphicQualifier
public @interface PolyNull { }
For a description of polymorphic qualifiers, see Section 2.5.2. A polymorphic qualifier needs no @SubtypeOf meta-annotation and need not be mentioned in any other @SubtypeOf meta-annotation.
The declarative and procedural mechanisms for specifying the hierarchy can be used together. In particular, when using the @SubtypeOf meta-annotation, further customizations may be performed procedurally (Section 9.2.2) by overriding the isSubtype method in the checker class (Section 9.5). However, the declarative mechanism is sufficient for most type systems.
While the declarative syntax suffices for many cases, more complex type hierarchies can be expressed by overriding, in BaseTypeChecker, either createQualifierHierarchy or createTypeHierarchy (typically only one of these needs to be overridden). For more details, see the Javadoc of those methods and of the classes QualifierHierarchy and TypeHierarchy.
The QualifierHierarchy class represents the qualifier hierarchy (not the type hierarchy), e.g., Mutable is a subtype of ReadOnly. A type-system designer may subclass QualifierHierarchy to express customized qualifier relationships (e.g., relationships based on annotation arguments).
The TypeHierarchy class represents relationships between annotated types, rather than merely type qualifiers, e.g., @Mutable Date is a subtype of @ReadOnly Date. The default TypeHierarchy uses QualifierHierarchy to determine all subtyping relationships. The default TypeHierarchy handles generic type arguments, array components, type variables, and wild-cards in a similar manner to the Java standard subtype relationship but with taking qualifiers into consideration. Some type systems may need to override that behavior. For instance, the Java Language Specification specifies that two generic types are subtypes only if their type arguments are identical: for example, List<Date> is not a subtype of List<Object>, or of any other generic List. (In the technical jargon, the generic arguments are “invariant”.) The Javari type system overrides this behavior to allow some type arguments to change covariantly in a type-safe manner (e.g., List<@Mutable Date> is a subtype of List<@QReadOnly Date>).
For some types and expressions, a qualifier should be treated as present even if a programmer did not explicitly write it. For example, every literal (other than null) has a @NonNull type.
The implicit annotations may be specified declaratively and/or procedurally.
The @ImplicitFor meta-annotation indicates implicit annotations. When written on a qualifier, ImplicitFor specifies the trees (AST nodes) and types for which the framework should automatically add that qualifier.
In short, the types and trees can be specified via any combination of five fields:
For example, consider the definitions of the @NonNull and @Nullable type qualifiers:
@TypeQualifier
@SubtypeOf( { Nullable.class } )
@ImplicitFor(
types={TypeKind.PACKAGE},
typeClasses={AnnotatedPrimitiveType.class},
trees={
Tree.Kind.NEW_CLASS,
Tree.Kind.NEW_ARRAY,
Tree.Kind.PLUS,
// All literals except NULL_LITERAL:
Tree.Kind.BOOLEAN_LITERAL, Tree.Kind.CHAR_LITERAL, Tree.Kind.DOUBLE_LITERAL, Tree.Kind.FLOAT_LITERAL,
Tree.Kind.INT_LITERAL, Tree.Kind.LONG_LITERAL, Tree.Kind.STRING_LITERAL
})
public @interface NonNull { }
@TypeQualifier
@SubtypeOf({})
@ImplicitFor(trees={Tree.Kind.NULL_LITERAL})
public @interface Nullable { }
For more details, see the Javadoc for the ImplicitFor annotation, and the Javadoc for the javac classes that are linked from it. (You only need to understand a small amount about the javac AST, such as the Tree.Kind and TypeKind enums. All the information you need is in the Javadoc, and Section 9.8 can help you get started.)
The Checker Framework provides a representation of annotated types, AnnotatedTypeMirror, that extends the standard TypeMirror interface but integrates a representation of the annotations into a type representation. A checker's type factory class, given an AST node, returns the annotated type of that expression. The Checker Framework's abstract base type factory class, AnnotatedTypeFactory, supplies a uniform, Tree-API-based interface for querying the annotations on a program element, regardless of whether that element is declared in a source file or in a class file. It also handles default annotations, and it optionally performs flow-sensitive local type inference.
AnnotatedTypeFactory inserts the qualifiers that the programmer explicitly inserted in the code. Yet, certain constructs should be treated as having a type qualifier even when the programmer has not written one. The type system designer may subclass AnnotatedTypeFactory and override annotateImplicit(Tree,AnnotatedTypeMirror) and annotateImplicit(Element,AnnotatedTypeMirror) to account for such constructs.
A type system's rules define which operations on values of a particular type are forbidden.
The framework provides a base visitor class, BaseTypeVisitor, that performs type-checking at each node of a source file's AST. It uses the visitor design pattern to traverse Java syntax trees as provided by Sun's Tree API, and issues a warning whenever the type system induced by the type qualifier is violated.
A checker's visitor overrides one method in the base visitor for each special rule in the type qualifier system. Most type-checkers override only a few methods in BaseTypeVisitor. For example, the visitor for the Nullness type system of Section 3 consists of a single 4-line method that warns if an expression of nullable type is dereferenced, as in:
myObject.hashCode(); // invalid dereference
By default, BaseTypeVisitor performs subtyping checks that are similar to Java subtype rules, but taking the type qualifiers into account. BaseTypeVisitor issues these errors:
In particular, in every assignment and pseudo-assignment, the left-hand side of the assignment is a supertype of (or the same type as) the right-hand side. For example, this assignment is not permitted:
@Nullable Object myObject;
@NonNull Object myNonNullObject;
...
myNonNullObject = myObject; // invalid assignment
A checker's entry point is a subclass of BaseTypeChecker. This entry point, which we call the checker class, serves two roles: an interface to the compiler and a factory for constructing type-system classes.
Because the Checker Framework provides reasonable defaults, oftentimes the checker class has no work to do. Here are the complete definitions of the checker classes for the Interning and Nullness checkers:
@TypeQualifiers({ Interned.class, PolyInterned.class })
@SupportedLintOptions({"dotequals"})
public final class InterningChecker extends BaseTypeChecker { }
@TypeQualifiers({ Nullable.class, Raw.class, NonNull.class, PolyNull.class })
@SupportedLintOptions({"flow", "cast", "cast:redundant"})
public class NullnessChecker extends BaseTypeChecker { }
The checker class must be annotated by @TypeQualifiers, which lists the annotations that make up the type hierarchy for this checker (including polymorphic qualifiers), provided as an array of class literals. Each one is a type qualifier whose definition bears the @TypeQualifier meta-annotation (or is returned by the BaseTypeChecker.getSupportedTypeQualifiers method).
The checker class bridges between the compiler and the checker plugin. It invokes the type-rule check visitor on every Java source file being compiler, and provides a simple API, report, to issue errors using the compiler error reporting mechanism.
Also, the checker class follows the factory method pattern to construct the concrete classes (e.g., visitor, factory) and annotation hierarchy representation. It is a convention that, for a type system named Foo, the compiler interface (checker), the visitor, and the annotated type factory are named as FooChecker, FooVisitor, and FooAnnotatedTypeFactory. BaseTypeChecker uses the convention to reflectively construct the components. Otherwise, the checker writer must specify the component classes for construction.
A checker can customize the default error messages through a Properties-loadable text file named messages.properties that appears in the same directory as the checker class. The property file keys are the strings passed to report (like type.incompatible) and the values are the strings to be printed (cannot assign ...). The messages.properties file only need to mention the new messages that the checker defines. It is also allowed to override messages defined in superclasses, but this is rarely needed.
[This section should discuss the testing framework that is used for checking the distributed checkers.]
The Checker Framework provides debugging options that can be helpful when writing checker. These are provided via the standard javac “-A” switch, which is used to pass options to an annotation processor.
The following example demonstrates how these options are used:
$ javac -processor checkers.interning.InterningChecker \
examples/InternedExampleWithWarnings.java -Ashowchecks -Anomsgtext -Afilenames
[InterningChecker] InterningExampleWithWarnings.java
success (line 18): STRING_LITERAL "foo"
actual: DECLARED @checkers.interning.quals.Interned java.lang.String
expected: DECLARED @checkers.interning.quals.Interned java.lang.String
success (line 19): NEW_CLASS new String("bar")
actual: DECLARED java.lang.String
expected: DECLARED java.lang.String
examples/InterningExampleWithWarnings.java:21: (not.interned)
if (foo == bar)
^
success (line 22): STRING_LITERAL "foo == bar"
actual: DECLARED @checkers.interning.quals.Interned java.lang.String
expected: DECLARED java.lang.String
1 error
You can use any standard debugger to observe the execution of your checker. Set the main class to com.sun.tools.javac.Main and the bootclasspath to include the JSR308 langtools.
The implementation of Sun's javac compiler can be a bit daunting to a newcomer, and its documentation does not particularly help a newcomer to get oriented. But do not lose heart! This section helps you to understand the small part of javac that you need in order to write a checker. Other useful resources include the Java Infrastructure Developer's guide at http://wiki.netbeans.org/Java_DevelopersGuide and the compiler mailing list archives at http://news.gmane.org/gmane.comp.java.openjdk.compiler.devel (subscribe at http://mail.openjdk.java.net/mailman/listinfo/compiler-dev).
The Checker Framework uses Sun's Tree API to access a program's AST. This is specific to the Sun JDK. In the future, the Checker Framework can be migrated to use the Java Model AST of JSR 198 (Extension API for Integrated Development Environments) [Cro06], which gives access to the entire source code of a method in an implementation-neutral way.
A Tree is an AST node; it represents an arbitrary code snippet such as a method definition, a block, a statement, etc.
The Tree interface has many subinterfaces, that specify what kind of node is being handled. Trees are usually processed by a class implementing the TreeVisitor interface, through the accept method on Tree. Common implementations of TreeVisitor that you may want to extend are SimpleTreeVisitor, that visits a single node based on its type, TreeScanner, that visits all subnodes recursively, and TreePathScanner, that visits all subnodes recursively and stores the TreePath corresponding the the currently visited Tree. (Also note that the iterator given by TreePath used to have an implementation bug.)
In order to determine the kind of an object that extends Tree, use the getKind method, as opposed to the instanceof operator, since a Tree implementation might opt to implement more than one interface from this API. There is a utility class to perform operations on trees, Trees, but the framework is intended to do all the low-level tree processing, so you probably should not need to use this class.
An Element represents a program element such as packages, classes or methods. Element has 5 subinterfaces: ExecutableElement represents methods, constructors or initializers (anything invocable); PackageElement represents package elements, and contain package information; TypeElement represents the element of a class or an interface (note that TypeElement is an Element, not a Type; the corresponding Type is represented by DeclaredType; TypeParameterElement represents an element of a formal type parameter of a something with generics, and VariableElement represents the element associated with a variable. There is an ElementVisitor interface for visiting objects that Element, in a similar manner to the Tree visitors, with similar provided implementations. Use the asType method from Element to obtain a TypeMirror for the element.
Again, Element is an interface, so use getKind() to obtain the kind of an Element, as opposed to the instanceof operator, since an implementation of Element might also implement other element interfaces. There is an utility class for handling elements, Elements; the appropriate instance can be obtained by using the getElementUtils method on the ProcessingEnvironment object visible on factories and checkers. The framework should do most of the element processing that requires Elements, unless you are doing something non-trivial.
A TypeMirror represents a Java type. It is yet another interface you should be familiar with, with various subinterfaces, notable ones being DeclaredType for class and interface types, and ExecutableType for method, constructor and initializer types.
Note that a MethodTree resolves into a ExecutableType, while a MethodInvocationTree resolves into a DeclaredType if the return type is a class or an interface, an ArrayType if the return type is an array, a NoType if the return type is void, or a PrimitiveType if the return type is primitive.
Not every Tree corresponds to an Element (such as a BlockTree), not every Tree corresponds to a TypeMirror (again, such as a BlockTree), and not every TypeMirror has a corresponding Element (such as primitive types or arrays).
As one could expect by this point, TypeMirror is an interface, so use the appropriate getKind() method to distinguish the types, as opposed to the instanceof operator, since those are interfaces, and more than one can be implemented by a same object.
Note that the TypeMirror API makes no guarantees that the same type will always be represented by the same object; use the method recommended on the API if you need to compare two types.
TypeVisitor and implementations of visitors for TypeMirror are provided, but those classes should not be used or extended directly on the framework, since all checker plugin classes are meant to visit AnnotatedTypeMirror instead, modifying the annotations as needed. A Types utility class is provided by the ProcessingEnvironment as well, if you need to do more complex operations with types. In general, you should use AnnotatedTypeMirror and its subclasses as opposed to using TypeMirror and its subinterfaces.
An AnnotatedTypeMirror (defined in the Checker Framework, not in javac) represents an annotated type — a type along with all its annotations. It is modeled after Sun's TypeMirror. Similarly modeled visitors are presented: a AnnotatedTypeVisitor interface, implemented by SimpleAnnotatedTypeVisitor for visiting just one node, AnnotatedTypeScanner for visiting every node recursively.
In short: a Tree represents some snippet of code, an Element represents some program element, and a TypeMirror represents a Java type, but you usually should use AnnotatedTypeMirror, provided by the Checker Framework, instead of TypeMirror, as our implementation carries along with the types the annotation information at every node level. The AnnotatedTypeFactory (or its extension on your framework plugin) is responsible for producing AnnotatedTypeMirror objects for Tree and Element parameters it receives; those AnnotatedTypeMirror objects are then processed by the visitor class and checked by the checker class on your checker plugin.
If you get the error
com.sun.tools.javac.code.Symbol$CompletionFailure: class file for com.sun.source.tree.Tree not found
then you are using the source installation and file tools.jar is not on your classpath; see the installation instructions (Section 1.2).
If you get an error such as
package checkers.nullness.quals does not exist
despite no apparent use of import checkers.nullness.quals.*; in the source code, then perhaps jsr308_imports is set as a Java system property, a shell environment variable, or a command-line option (see Section 2.1.1). You can solve this by unsetting the variable/option, or by ensuring that the checkers.jar file is on your classpath.
Consider this MyEnum class:
public enum MyEnum {
PRIVATE();
MyEnum() { }
}
Any checker will produce this error:
MyEnum.java:3: call to super not allowed in enum constructor
MyEnum() { }
^
1 error
Another example is
./Test.java:10: method asList in class java.util.Arrays cannot be applied to given types
required: T[]
found: java.lang.Object[]
List<Object> a_list = Arrays.<Object>asList (new Object[] {a});
^
The workaround is to pass the argument -proc:only to javac when using the checkers. However, this prevents the compiler from generating classfiles; to generate classfiles, you will separately need to run the Java compiler without any pluggable type checker.
@NonNull String value;
if (myMap.containsKey(key)) {
value = myMap.get(key);
}
for (String keyInMap : myMap.keySet()) {
value = myMap.get(keyInMap);
}
The Nullness checker can sometimes fail to issue a warning if the map is modified or re-assigned between the check of containsKey and the call to get.
If you have a problem with any checker, or with the Checker Framework, please let us know at jsr308-bugs@lists.csail.mit.edu. In addition to bug reports, we welcome suggestions, annotated libraries, bug fixes, new features, new checker plugins, and other improvements.
Please ensure that your bug report is clear and that it is complete. Otherwise, we may be unable to understand it or to reproduce it, either of which would prevent us from fixing the bug. Your bug report will be most helpful if you:
The binary release (Section 1.2) contains everything that most users need, both to use the distributed checkers and to write your own checkers. This section describes how to install the source release. Doing so permits you to examine and modify the implementation of the distributed checkers and of the checker framework. It may also help you to debug problems more effectively.
The following commands install the JSR 308 javac compiler and the Checker Framework, or update an existing installation. It currently works only on Linux. For more details, or if anything goes wrong, see the comments in the Makefile-jsr308-install file.
cd wget -nv -N http://groups.csail.mit.edu/pag/jsr308/current/Makefile-jsr308-install make -f Makefile-jsr308-install
The following instructions give detailed steps for installing the source release of the Checker Framework.
cd $JSR308 wget http://groups.csail.mit.edu/pag/jsr308/current/jsr308-checkers.zip unzip jsr308-checkers.zip
export JSR308=${HOME}/jsr308
export CLASSPATH=${CLASSPATH}:$JSR308/jdk1.7.0/lib/tools.jar:$JSR308/checkers/checkers.jar
cd checkers ant all-tests
JSR 308 extends the Java language to permit annotations to appear on types, as in List<@NonNull String>. This change is planned to be part of the Java 7 language.) We recommend that you write annotations in comments, as in List</*@NonNull*/ String> (see Section 2.1). The JSR 308 compiler still reads such annotations, but this syntax permits you to use a compiler other than the JSR 308 compiler. For example, you can compile your code with a Java 5 compiler, and you can use a checker as an external tool in an IDE such as Eclipse.
Building (compiling) the checkers and framework from source creates the checkers.jar file. A pre-compiled checkers.jar is included in the distribution, so building it is optional. It is mostly useful for people who are developing compiler plug-ins (type-checkers). If you only want to use the compiler and existing plug-ins, it is sufficient to use the pre-compiled version.
cd checkers ant
The technical paper “Practical pluggable types for Java” [PAC+08] (http://www.cs.washington.edu/homes/mernst/pubs/pluggable-checkers-issta2008.pdf) gives more technical detail about many aspects of the Checker Framework and its implementation. The technical paper also describes a few features that are part of the distribution but are not yet documented in this manual. Finally, the technical paper describes case studies in which each of the checkers found previously-unknown errors in real software.
Several other tools have a similar goal to the Checker Framework: helping you to prevent or detect errors in your code.
Section 6 of the paper “Practical pluggable types for Java” [PAC+08] (http://www.cs.washington.edu/homes/mernst/pubs/pluggable-checkers-issta2008.pdf) compares the Checker Framework's nullness checker with those of FindBugs, JLint, and PMD.
A pluggable type-checker such as those created by the Checker Framework differs from a bug detector in several ways:
A bug detector aims to find some errors. If it reports no null pointer errors, then there may still be some in your code.
Both types of tools may issue false positive warnings; see Section 2.4.
The Checker Framework distribution was developed in the MIT Program Analysis Group, with prime contributions from Mahmood Ali, Telmo Correa, and Matthew M. Papi. Many users have provided valuable feedback, for which we are grateful.
Differences from previous versions of the checkers and framework can be found in the changelog-checkers.txt file. This file is included in the checkers distribution and is also available on the web at http://groups.csail.mit.edu/pag/jsr308/current/changelog-checkers.txt.