Type Annotations Specification (JSR 308)Michael D. Ernst |
The JSR 308 webpage is https://checkerframework.org/jsr308/. It contains the latest version of this document, along with other information such as a FAQ, the reference implementation, and sample annotation processors.
This document is available in PDF format at https://checkerframework.org/jsr308/specification/java-annotation-design.pdf.
JSR 308 extends Java’s annotation system [Blo04a] so that annotations may appear on nearly any use of a type. (By contrast, Java SE 6 permits annotations only on declarations; JSR 308 is backward-compatible and continues to permit those annotations.) Such a generalization removes limitations of Java’s annotation system, and it enables new uses of annotations. This proposal also notes a few other possible extensions to annotations (see Section D).
This document specifies the syntax of extended Java annotations, but it makes no commitment as to their semantics. As with Java’s existing annotations [Blo04a], the semantics is dependent on annotation processors (compiler plug-ins), and not every annotation is necessarily sensible in every location where it is syntactically permitted to appear. This proposal is compatible with existing annotations, such as those specified in JSR 250, “Common Annotations for the Java Platform” [Mor06], and proposed annotations, such as those to be specified in JSR 305, “Annotations for Software Defect Detection” [Pug06]. (For a comparison of JSR 305 and JSR 308, see Section D.4.)
This proposal does not change the compile-time, load-time, or run-time semantics of Java. It does not change the abilities of Java annotation processors as defined in JSR 269 [Dar06]. The proposal merely makes annotations more general — and thus more useful for their current purposes, and also usable for new purposes that are compatible with the original vision for annotations [Blo04a].
This document has two parts: a short normative part and a longer non-normative part. The normative part specifies the changes to the Java language syntax (Sections 2 and 5), the Java toolset (Section 3), and the class file format (Section 4).
The non-normative part consists of appendices that discuss and explain the specification or deal with logistical issues. It motivates annotations on types by presenting one possible use, type qualifiers (Appendix A). It gives examples of and further motivation for the Java syntax changes (Appendix B) and lists tools that must be updated to accommodate the Java and class file modifications (Appendix C). Appendix D lists other possible extensions to Java annotations, some of which are within the scope of JSR 308 (and might be included in a future revision) and some of which are not. The document concludes with logistical matters relating to incorporation in the Sun JDK (Section E) and related work (Section F).
A JSR, or Java Specification Request, is a proposed specification for some aspect of the Java platform — the Java language, virtual machine, libraries, etc. For more details, see the Java Community Process FAQ at https://jcp.org/en/introduction/faq.
A FAQ (Frequently Asked Questions) document complements this specification; see https://checkerframework.org/jsr308/jsr308-faq.html.
In Java SE 6, annotations can be written only on method parameters and the declarations of packages, classes, methods, fields, and local variables. JSR 308 extends Java to allow annotations on nearly any use of a type, and on type parameter declarations. JSR 308 uses a simple prefix syntax for type annotations, with a special case for receiver types and for constructor results.
Section B.1 contains examples of the annotation syntax.
This section summarizes the Java language grammar changes, which correspond to the rules of Section 2.1. Section 5 shows the grammar changes in detail. Additions are underlined.
| Type: |
| [Annotations] Identifier [TypeArguments] {. Identifier [TypeArguments]} {[]} |
| [Annotations] BasicType |
| Type: |
| [Annotations] Identifier [TypeArguments]{ . Identifier [TypeArguments]} {[Annotations] []} |
| [Annotations] BasicType |
Annotations may also appear on varargs (...):
| FormalParameterDeclsRest: |
| VariableDeclaratorId [, FormalParameterDecls] |
| [Annotations] ... VariableDeclaratorId |
| VoidMethodDeclaratorRest: |
| FormalParameters [Annotations] [throws QualifiedIdentifierList] ( MethodBody | ; ) |
| TypeParameter: |
| [Annotations] Identifier [extends Bound] |
Annotations may also appear on the wildcard in any type argument. A wildcard is the declaration of an anonymous (unnamed) type parameter.
| TypeArgument: |
| [Annotations] ? [( extends | super ) Type] |
Java uses the @Target meta-annotation as a machine-checked way of expressing where an annotation is intended to appear. The ElementType enum classifies the places an annotation may appear in a Java program. JSR 308 adds two new constants to the ElementType enum.
For example, in this declaration:
@Target(ElementType.TYPE_USE)
public @interface NonNull { ... }
the @Target(ElementType.TYPE_USE) meta-annotation indicates that @NonNull may appear on any use of a type.
ElementType.TYPE_PARAMETER and ElementType.TYPE_USE are distinct from the existing ElementType.TYPE enum element of Java SE 6, which indicates that an annotation may appear on a type declaration (a class, interface, or enum declaration). The locations denoted by ElementType.TYPE_PARAMETER, ElementType.TYPE_USE, and ElementType.TYPE are disjoint.
If an annotation is not meta-annotated with @Target (which would be poor style), then the compiler treats the annotation as if it is meta-annotated with all of the ElementType enum constants.
The compiler applies an annotation to every target that is consistent with its meta-annotation. The order of annotations is not used to disambiguate. As in Java SE 6, the compiler issues an error if a programmer places an annotation in a location not permitted by its Target meta-annotation. (The compiler issues the error even if no annotation processor is being run.)
When generating .class files, the compiler must emit the attributes described in Section 4.
The compiler is required to preserve annotations in the class file. More precisely, if an expression has an annotated type (and the annotation has class file or runtime retention), and that expression is represented in the compiled class file, then the annotation must be present, in the compiled class file, on the type of the compiled representation of the expression. For expressions within a field initializer or a static initializer block, the compiled representation might be in an <init> or <clinit> method, or in a helper method called by them.
If the compiler optimizes away an expression, then it may also optimize away the annotation. (Exception: when a type cast is optimized away without optimizing away its argument, the annotation remains on the argument; see Section 4.2.)
The compiler sometimes creates new methods that did not appear in the source code. The compiler should annotate these compiler-generated methods in an appropriate way. For example, a bridge method is an implementation strategy used when the erased signature of the actual method being invoked differs from that of the compile-time method declaration [GJSB05, §15.12.4.5]. The bridge method does nothing but call an existing method, and in this case, annotations should be copied from the method being invoked.
This section defines how to store type annotations in a Java class file. It also defines how to store local variable annotations, which are permitted in Java SE 6 source code but are discarded by the compiler.
The class file format represents the type of every variable and expression in a Java class, including all temporaries and values stored on the stack. (Sometimes the representation is explicit, such as via the StackMapTable attribute, and sometimes it is implicit.) Since JSR 308 permits annotations to be added to a type, the class file format should be updated to continue to represent the full, annotated type of each expression.
More pragmatically, Java annotations must be stored in the class file for two reasons.
First, annotated signatures (public members) must be available to tools that read class files. For example, a type-checking compiler plug-in [Dar06, PAC+08] needs to read annotations when compiling a client of the class file. The Checker Framework (https://checkerframework.org/) is one way to create such plug-ins.
Second, annotated method bodies must be present to permit checking the class file against the annotations. This is necessary to give confidence in an entire program, since its parts (class files) may originate from any source. Otherwise, it would be necessary to simply trust annotated classes of unknown provenance [BHP07].
In Java SE 6, an annotation is stored in the class file in an attribute [Blo04a, LY07]. An attribute associates data with a program element (a method’s bytecodes, for instance, are stored in a Code attribute of the method). The RuntimeVisibleParameterAnnotations stores formal parameter annotations that are accessible at runtime using reflection, and the RuntimeInvisibleParameterAnnotations attribute stores formal parameter annotations that are not accessible at runtime. RuntimeVisibleAnnotations and RuntimeInvisibleAnnotations are analogous, but for annotations on fields, methods, and classes.
These attributes contain arrays of annotation structure elements, which in turn contain arrays of element_value pairs. The element_value pairs store the names and values of an annotation’s arguments.
Annotations on a field are stored as attributes of the field’s field_info structure [LY07, §4.6]. Annotations on a method are stored as attributes of the method’s method_info structure [LY07, §4.7]. Annotations on a class are stored as attributes of the class’s attributes structure [LY07, §4.2].
Generic type information is stored in a different way in the class file, in a signature attribute. Its details are not germane to the current discussion.
JSR 308 introduces two new attributes: RuntimeVisibleTypeAnnotations and RuntimeInvisibleTypeAnnotations. These attributes are structurally identical to the RuntimeVisibleAnnotations and RuntimeInvisibleAnnotations attributes described above with one exception: rather than an array of annotation elements, RuntimeVisibleTypeAnnotations and RuntimeInvisibleTypeAnnotations contain an array of extended_annotation elements, which are described in Section 4.1.
A type annotation is stored in a Runtime[In]visibleTypeAnnotations attribute on the smallest enclosing class, field, or method.
Type annotations targeting declaration types (e.g. field, method return type) will appear in the Runtime[In]visibleTypeAnnotations attribute. Section B.4 discusses disambiguating type and declaration annotations.
For backward compatibility, JSR 308 uses new attributes for storing the type annotations. In other words, JSR 308 merely reserves the names of a few new attributes and specifies their layout. JVMs ignore unknown attributes. JSR 308 does not alter the way that existing annotations on classes, methods, method parameters, and fields are stored in the class file. JSR 308 mandates no changes to the processing of existing annotation locations; in the absence of other changes to the class file format, class files generated from programs that use no new annotations will be identical to those generated by a standard Java SE 6 compiler. Furthermore, the bytecode array will be identical between two programs that differ only in their annotations. Attributes have no effect on the bytecode array, because they exist outside it; however, they can represent properties of it by referring to the bytecode (including referring to specific instructions, or bytecode offsets).
In Java SE 6, the class file stores the types of elements on the stack; this eliminates the need for byte code verification to to perform type inference for local variables and stack elements. The class file format should similarly store the annotations on those types, to aid in annotation verification in class files.
The extended_annotation structure has the following format:
extended_annotation {
// Original fields from "annotation" structure:
u2 type_index;
u2 num_element_value_pairs;
{
u2 element_name_index;
element_value value;
} element_value_pairs[num_element_value_pairs];
// New fields in JSR 308:
u1 target_type; // the type of the targeted program element, see Section 4.2
union {
offset_target;
localvar_target;
methodparam_target;
field_target;
typeparam_target;
typeparam_bound_target;
typearg_target;
supertype_target;
typeindex_target;
wildcard_bound_target;
empty_target;
} target_info; // uniquely identifies the targeted program element, see Section 4.3
}
We first briefly recap the three fields of annotation [LY07, §4.8.15].
Compared to annotation, the extended_annotation structure contains two additional fields. These fields implement a discriminated (tagged) union type: field target_type is the tag (see Section 4.2), and its value determines the size and contents of target_info (see Section 4.3).
The target_type field denotes the type of program element that the annotation targets, such as whether the annotation is on a field, a method receiver, a cast, or some other location. Figure 1 gives the value of target_type for every possible annotation location.
Annotation target target_type value target_info definition method receiver 0x06 §4.3.4 …type argument or array element type 0x07 Future extension to JSR 308 method return type 0x0A §4.3.5 …type argument or array element type 0x0B +§4.3.12 method parameter 0x0C §4.3.3 …type argument or array element type 0x0D +§4.3.12 field 0x0E §4.3.5 …type argument or array element type 0x0F +§4.3.12 class type parameter bound 0x10 §4.3.7 …type argument or array element type 0x11 +§4.3.12 method type parameter bound 0x12 §4.3.7 …type argument or array element type 0x13 +§4.3.12 class extends/implements 0x14 §4.3.9 …type argument or array element type 0x15 +§4.3.12 exception type in throws 0x16 §4.3.10 …type argument or array element type 0x17 forbidden by Java wildcard bound 0x1C §4.3.11 …type argument or array element type 0x1D +§4.3.12 method type parameter 0x20 §4.3.6 …type argument or array element type 0x21 Java does not support class type parameter 0x22 §4.3.6 …type argument or array element type 0x23 Java does not support typecast 0x00 §4.3.1 …type argument or array element type 0x01 +§4.3.12 type test (instanceof) 0x02 §4.3.1 …type argument or array element type 0x03 +§4.3.12 object creation (new) 0x04 §4.3.1 …type argument or array element type 0x05 +§4.3.12 local variable 0x08 §4.3.2 …type argument or array element type 0x09 +§4.3.12 type argument in constructor call 0x18 §4.3.8 …type argument or array element type 0x19 +§4.3.12 type argument in method call 0x1A §4.3.8 …type argument or array element type 0x1B +§4.3.12 class literal 0x1E §4.3.1 …type argument or array element type 0x1F +§4.3.12
Figure 1: Values of target_type for each possible target of a type annotation. Ordinary Java annotations on declarations are not included, because they appear in annotation, not extended_annotation, attributes in the class file. Table elements such as local variable, method parameter, and field refer to the declaration, not the use, of such elements. Enumeration elements marked * will never appear in a target_type field; see Section 4.2. The top half of the table contains targets that may appear on classes and members — places that annotations already appear in the class file in Java SE 6. The bottom half of the table contains targets that only appear inside method bodies, where annotations do not appear in Java SE 6 (not even local variable annotations written by the programmer).
Enumeration elements marked * never appear in a target_type field. Sometimes this is because annotations cannot be written in that location, such as because Java prohibits writing the location itself. In other cases this is because the annotations were permitted in Java 5, so there is already a place to store them in the class file rather than placing them in the new Runtime[In]visibleTypeAnnotations attribute. The unused enumeration elements are included for completeness.
target_info is a structure that contains enough information to uniquely identify the target of a given annotation. A different target_type may require a different set of fields, so the structure of the target_info is determined by the value of target_type.
See Section 4.3.12 for handling of generic type arguments and arrays.
When the annotation’s target is a typecast, an instanceof expression, a new expression, or a class literal expression, target_info contains one offset_target which has the following structure:
offset_target {
u2 offset;
};
The offset field denotes the offset (i.e., within the bytecodes of the containing method) of the checkcast bytecode emitted for the typecast, the instanceof bytecode emitted for the type tests, the new bytecode emitted for the object creation expression, the ldc(_w) bytecode emitted for class literals, or the getstatic bytecode emitted for primitive class literals. Typecast annotations are attached to a single bytecode, not a bytecode range (or ranges): the annotation provides information about the type of a single value, not about the behavior of a code block. A similar explanation applies to type tests, object creation, and class literals.
For annotated typecasts, the attribute may be attached to a checkcast bytecode, or to any other bytecode. The rationale for this is that the Java compiler is permitted to omit checkcast bytecodes for typecasts that are guaranteed to be no-ops. For example, a cast from String to @NonNull String may be a no-op for the underlying Java type system (which sees a cast from String to String). If the compiler omits the checkcast bytecode, the @NonNull attribute would be attached to the (last) bytecode that creates the target expression instead. This approach permits code generation for existing compilers to be unaffected.
An Runtime[In]visibleTypeAnnotations attribute containing a offset_target lives in the attributes table of a method_info structure.
When the annotation’s target is a local variable, target_info contains one localvar_target, which has the following structure:
localvar_target {
u2 table_length;
{
u2 start_pc;
u2 length;
u2 index;
} table[table_length];
};
The table_length field specifies the number of entries in the table array; multiple entries are necessary because a compiler is permitted to break a single variable into multiple live ranges with different local variable indices. The start_pc and length fields specify the variable’s live range in the bytecodes of the local variable’s containing method (from offset start_pc to offset start_pc + length). The index field stores the local variable’s index in that method. These fields are similar to those of the optional LocalVariableTable attribute [LY07, §4.8.12].
Storing local variable annotations in the class file raises certain challenges. For example, live ranges are not isomorphic to local variables. Further, a local variable with no live range may not appear in the class file (but it is also irrelevant to the program).
An Runtime[In]visibleTypeAnnotations attribute containing a localvar_target lives in the attributes table of a method_info structure.
When the type annotation’s target is a method parameter, target_info contains one methodparam_target which indicates which of the method’s formal parameters is being annotated:
methodparam_target {
u1 parameter_index;
};
An Runtime[In]visibleTypeAnnotations attribute containing a methodparam_target lives in the attributes table of a method_info structure.
When the annotation’s target is a method receiver (or an inner class constructor receiver), target_info is empty.
An Runtime[In]visibleTypeAnnotations attribute targeting a method receiver lives in the attributes table of a method_info structure.
When the type annotation’s target is a method return type, a constructor result, or a field, target_info is empty.
An Runtime[In]visibleTypeAnnotations attribute targeting a field type lives in the attributes table of a field_info structure. An Runtime[In]visibleTypeAnnotations attribute targeting a method return type lives in the attributes table of a field_info structure.
When the annotation’s target is a type parameter of a class or method, target_info contains one typeparam_info which has the following structure:
typeparam_info {
u1 param_index;
};
param_index specifies the 0-based index of the type parameter.
An Runtime[In]visibleTypeAnnotations attribute containing a offset_target lives in the attributes table of a method_info structure if it
When the annotation’s target is a bound of a type parameter of a class or method, target_info contains one typeparam_bound_target which has the following structure:
typeparam_bound_target {
u1 param_index;
u1 bound_index;
};
param_index specifies the index of the type parameter, while bound_index specifies the index of the bound. Consider the following example:
<T extends @A Object & @B Comparable, U extends @C Cloneable>
Here @A has param_index 0 and bound_index 0, @B has param_index 0 and bound_index 1, and @C has param_index 1 and bound_index 0.
An Runtime[In]visibleTypeAnnotations attribute containing a typeparam_bound_target lives in the attributes table of a method_info structure if it targets a method type parameter bound, otherwise, in the attributes of ClassFile structure if it targets a class declaration type parameter bound.
When the annotation’s type is a type argument in a constructor call or a method call, target_info has one typearg_target which has the following structure:
typearg_target {
u2 offset;
u1 type_index;
};
The offset field denotes the offset (i.e., within the bytecodes of the containing method) of the new bytecode emitted for constructor call, or the invoke{interface|special|static|virtual} bytecode emitted for method invocation. Like type cast type annotations, type argument annotations are attached to a single bytecode, not a bytecode range.
type_index specifies the index of the type argument in the expression.
An Runtime[In]visibleTypeAnnotations attribute containing a typeparam_bound_target lives in the attributes table of a method_info structure.
When the annotation’s target is a type in an extends or implements clause, target_info has the following structure:
supertype_target {
u2 type_index;
};
type_index specifies the index of the targeted type in the interfaces array field of the ClassFile structure; simply the value i is used if the annotation is on the ith superinterface type (counting from zero).
-1 (65535) is used if the annotation is on the superclass type.
An Runtime[In]visibleTypeAnnotations attribute containing a supertype_target lives in the attributes table of a ClassFile structure.
When the annotation’s target is a type in a throws clause, target_info contains exception_target which has the following structure:
exception_target {
u2 type_index;
};
type_index specifies the index of the exception type in the clause in exception_index_table of Exceptions_attribute; simply the value i denotes an annotation on the ith exception type (counting from zero).
An Runtime[In]visibleTypeAnnotations attribute containing a typeparam_bound_target lives in the attributes table of a method_info structure.
When the annotation’s target is the bound of a wildcard, target_info contains one wildcard_bound_target which has the following structure:
wildcard_bound_target {
u1 wildcard_location_type; // The location of the wildcard, values as for target_type (Section 4.2)
{
...
} target_info; // uniquely identifies the targeted program element, see Section 4.3
}
Here are examples:
@A in void m(List<? extends @A String> lst) { ... } is described as:
target_type = WILDCARD_BOUND
target_info =
{
wildcard_location_type = METHOD_PARAMETER (0x0C)
wildcard_location =
{
u1 parameter_index = 0;
}
}
@B in Map<Object, ? extends List<@B String>> newMap() { ... } is described as:
target_type = WILDCARD_BOUND_GENERIC_OR_ARRAY
target_info =
{
// location of wildcard: type argument of the return type
wildcard_location_type = RETURN_TYPE_GENERIC_OR_ARRAY;
wildcard_location =
{
// the second type variable
location_length = 1;
location = { 1 };
}
// the generic information of the annotation itself
location_length = 1;
location = { 0 };
}
An Runtime[In]visibleTypeAnnotations attribute containing a wildcard_bound_target lives in the attributes table of
method_info structure.
In a parameterized type or array, there are multiple places that an annotation may appear. For example, consider the difference among: @X Map<String, Object>, Map<@X String, Object>, and Map<String, @X Object>, or the difference among @X String [] [], String @X [] [], and String [] @X []. The classfile must distinguish among these locations.
When the annotation’s target is within a parameterized type or array type, target_info contains what it normally would for a non-parameterized, non-array type plus the following fields at the end:
u2 location_length;
u1 location[location_length];
The location field is a list of numbers that represents the annotation location among the generic type arguments and array levels. If the type is viewed as a tree, then the location field can be viewed as a path in that tree. The location_length field is the length of the location field, which is also the depth of the annotation in the tree.
A parameterized or array type can be viewed as a tree in the following way. For a parameterized type, the children of the raw type are the type arguments. For an array type, the children are the types of elements at each level; each of these is itself an array type (of lesser dimensions than the main type), except for the last one which is a non-array type. For a 1-dimensional array, there is only one child, which is a non-array.
For example, the parameterized types @A Map<@B String, @C List<@D Object>> and @H String @E [] @F [] @G [] can be viewed as the following trees:
Thus the annotations are represented in the class file as follows:
| Annotation | location_length | location |
| @A | not applicable | |
| @B | 1 | 0 |
| @C | 1 | 1 |
| @D | 2 | 1, 0 |
| @E | not applicable | |
| @F | 1 | 0 |
| @G | 1 | 1 |
| @H | 1 | 2 |
Figure 2 shows the values of the location_length and location fields in an example that combines both parameterized and array types.
Indexes start at zero. For a parameterized type, they count from the left. For an array type, they count from the first dimension.
The specification does not indicate how to differentiate @A and @B uniquely in OuterClass<@A String>.InnerClass<@B String>.
Declaration: @A Map<@B Comparable<@C Object @D [] @E [] @F []>, @G List<@H Document>>
Annotation location_length location @A not applicable @B 1 0 @C 3 0, 0, 2 @D 2 0, 0 @E 3 0, 0, 0 @F 3 0, 0, 1 @G 1 1 @H 2 1, 0
Figure 2: Values of the location_length and location fields for a declaration that combines parameterized and array types.
This section gives detailed changes to the grammar of the Java language [GJSB05, ch. 18], based on the conceptually simple summary from Section 2.2. Additions are underlined.
This section is of interest primarily to language tool implementers, such as compiler writers. Most users can read just Sections 2.1 and B.1.
Infelicities in the Java grammar make this section longer than the simple summary of Section 2.2. Some improvements are possible (for instance, by slightly refactoring the Java grammar), but this version attempts to minimize changes to existing grammar productions.
| Type: |
| [Annotations] UnannType |
| UnannType: |
| Identifier [TypeArguments]{ . Identifier [TypeArguments]} {[Annotations] []} |
| BasicType |
| FormalParameterDecls: |
| [final] [Annotations] UnannType FormalParameterDeclsRest |
| ForVarControl: |
| [final] [Annotations] UnannType Identifier ForVarControlRest |
| MethodOrFieldDecl: |
| UnannType Identifier MethodOrFieldRest |
| InterfaceMethodOrFieldDecl: |
| UnannType Identifier InterfaceMethodOrFieldRest |
| MethodDeclaratorRest: |
| FormalParameters {[Annotations] []} [Annotations] [throws QualifiedIdentifierList] ( MethodBody | ; ) |
| VoidMethodDeclaratorRest: |
| FormalParameters [Annotations] [throws QualifiedIdentifierList] ( MethodBody | ; ) |
| InterfaceMethodDeclaratorRest: |
| FormalParameters {[Annotations] []} [Annotations] [throws QualifiedIdentifierList] ; |
| VoidInterfaceMethodDeclaratorRest: |
| FormalParameters [Annotations] [throws QualifiedIdentifierList] ; |
| ConstructorDeclaratorRest: |
| FormalParameters [Annotations] [throws QualifiedIdentifierList] MethodBody |
| QualifiedIdentifierList: |
| [Annotations] QualifiedIdentifier { , [Annotations] QualifiedIdentifier } |
| Primary: |
| ... |
| BasicType {[Annotations] []} .class |
| IdentifierSuffix: |
| [Annotations] [ ( ] {[Annotations] []} .class | Expression ]) |
| ... |
| VariableDeclaratorRest: |
| {[Annotations] []} [= VariableInitializer] |
| ConstantDeclaratorRest: |
| {[Annotations] []} [= VariableInitializer] |
| VariableDeclaratorId: |
| Identifier {[Annotations] []} |
| FormalParameterDeclsRest: |
| VariableDeclaratorId [, FormalParameterDecls] |
| [Annotations] ... VariableDeclaratorId |
| TypeParameter: |
| [Annotations] Identifier [extends Bound] |
One example use of annotation on types is to create custom type qualifiers for Java, such as @NonNull, @ReadOnly, @Interned, or @Tainted. Type qualifiers are modifiers on a type; a declaration that uses a qualified type provides extra information about the declared variable. A designer can define new type qualifiers using Java annotations, and can provide compiler plug-ins to check their semantics (for instance, by issuing lint-like warnings during compilation). A programmer can then use these type qualifiers throughout a program to obtain additional guarantees at compile time about the program.
The type system defined by the type qualifiers does not change Java semantics, nor is it used by the Java compiler or run-time system. Rather, it is used by the checking tool, which can be viewed as performing type-checking on this richer type system. (The qualified type is usually treated as a subtype or a supertype of the unqualified type.) As an example, a variable of type Boolean has one of the values null, TRUE, or FALSE (more precisely, it is null or it refers to a value that is equal to TRUE or to FALSE). A programmer can depend on this, because the Java compiler guarantees it. Likewise, a compiler plug-in can guarantee that a variable of type @NonNull Boolean has one of the values TRUE or FALSE (but not null), and a programmer can depend on this. Note that a type qualifier such as @NonNull refers to a type, not a variable, though JSR 308 could be used to write annotations on variables as well.
Type qualifiers can help prevent errors and make possible a variety of program analyses. Since they are user-defined, developers can create and use the type qualifiers that are most appropriate for their software.
A system for custom type qualifiers requires extensions to Java’s annotation system, described in this document; the existing Java SE 6 annotations are inadequate. Similarly to type qualifiers, other pluggable type systems [Bra04] and similar lint-like checkers also require these extensions to Java’s annotation system.
Our key goal is to create a type qualifier system that is compatible with the Java language, VM, and toolchain. Previous proposals for Java type qualifiers are incompatible with the existing Java language and tools, are too inexpressive, or both. The use of annotations for custom type qualifiers has a number of benefits over new Java keywords or special comments. First, Java already implements annotations, and Java SE 6 features a framework for compile-time annotation processing. This allows JSR 308 to build upon existing stable mechanisms and integrate with the Java toolchain, and it promotes the maintainability and simplicity of the modifications. Second, since annotations do not affect the runtime semantics of a program, applications written with custom type qualifiers are backward-compatible with the vanilla JDK. No modifications to the virtual machine are necessary.
Four compiler plug-ins that perform type qualifier type-checking, all built using JSR 308, are distributed at the JSR 308 webpage, https://checkerframework.org/jsr308/. The four checkers, respectively, help to prevent and detect null pointer errors (via a @NonNull annotation), equality-checking errors (via a @Interned annotation), mutation errors (via the Javari [BE04, TE05] type system), and mutation errors (via the IGJ [ZPA+07] type system). A paper [PAC+08] discusses experience in which these plug-ins exposed bugs in real programs.
The ability to place annotations on arbitrary occurrences of a type improves the expressiveness of annotations, which has many benefits for Java programmers. Here we mention just one use that is enabled by extended annotations, namely the creation of type qualifiers. (Figure 3 gives an example of the use of type qualifiers.)
1 @DefaultQualifier("NonNull") 2 class DAG { 3 4 Set<Edge> edges; 5 6 // ... 7 8 List<Vertex> getNeighbors(@Interned @Readonly Vertex v) @Readonly { 9 List<Vertex> neighbors = new LinkedList<Vertex>(); 10 for (Edge e : edges) 11 if (e.from() == v) 12 neighbors.add(e.to()); 13 return neighbors; 14 } 15 }
Figure 3: The DAG class, which represents a directed acyclic graph, illustrates how type qualifiers might be written by a programmer and checked by a type-checking plug-in in order to detect or prevent errors. (1) The @DefaultQualifier("NonNull") annotation (line ??) indicates that no reference in the DAG class may be null (unless otherwise annotated). It is equivalent to writing line ?? as “@NonNull Set<@NonNull Edge> edges;”, for example. This guarantees that the uses of edges on line ??, and e on lines ?? and ??, cannot cause a null pointer exception. Similarly, the (implicit) @NonNull return type of getNeighbors() (line ??) enables its clients to depend on the fact that it will always return a List, even if v has no neighbors. (2) The two @Readonly annotations on method getNeighbors (line ??) guarantee to clients that the method does not modify, respectively, its Vertex argument or its DAG receiver (including its edges set or any edge in that set). The lack of a @Readonly annotation on the return value indicates that clients are free to modify the returned List. (3) The @Interned annotation on line ?? (along with an @Interned annotation on the return type in the declaration of Edge.from(), not shown) indicates that the use of object equality (==) on line ?? is a valid optimization. In the absence of such annotations, use of the equals method is preferred to ==.
As an example of how JSR 308 might be used, consider a @NonNull type qualifier that signifies that a variable should never be assigned null [Det96, Eva96, DLNS98, FL03, CMM05]. A programmer can annotate any use of a type with the @NonNull annotation. A compiler plug-in would check that a @NonNull variable is never assigned a possibly-null value, thus enforcing the @NonNull type system.
@Readonly and @Immutable are other examples of useful type qualifiers [ZPA+07, BE04, TE05, GF05, KT01, SW01, PBKM00]. Similar to C’s const, an object’s internal state may not be modified through references that are declared @Readonly. A type qualifier designer would create a compiler plug-in (an annotation processor) to check the semantics of @Readonly. For instance, a method may only be called on a @Readonly object if the method was declared with a @Readonly receiver. (Each non-static method has an implicit parameter, this, which is called the receiver.) @Readonly’s immutability guarantee can help developers avoid accidental modifications, which are often manifested as run-time errors. An immutability annotation can also improve performance. The Access Intents mechanism of WebSphere Application Server already incorporates such functionality: a programmer can indicate that a particular method (or all methods) on an Enterprise JavaBean is readonly.
Additional examples of useful type qualifiers abound. We mention just a few others. C uses the const, volatile, and restrict type qualifiers. Type qualifiers YY for two-digit year strings and YYYY for four-digit year strings helped to detect, then verify the absence of, Y2K errors [EFA99]. Expressing units of measurement (e.g., SI units such as meter, kilogram, second) can prevent errors in which a program mixes incompatible quantities; units such as dollars can prevent other errors. Range constraints, also known as ranged types, can indicate that a particular int has a value between 0 and 10; these are often desirable in realtime code and in other applications, and are supported in languages such as Ada and Pascal. Type qualifiers can indicate data that originated from an untrustworthy source [PØ95, VS97]; examples for C include user vs. kernel indicating user-space and kernel-space pointers in order to prevent attacks on operating systems [JW04], and tainted for strings that originated in user input and that should not be used as a format string [STFW01]. A localizable qualifier can indicate where translation of user-visible messages should be performed. Annotations can indicate other properties of its contents, such as the format or encoding of a string (e.g., XML, SQL, human language, etc.). Local and remote qualifiers can indicate whether particular resources are available on the same machine or must be retrieved over the network. An interned qualifier can indicate which objects have been converted to canonical form and thus may be compared via reference equality. Type qualifiers such as unique and unaliased can express properties about pointers and aliases [Eva96, CMM05]; other qualifiers can detect and prevent deadlock in concurrent programs [FTA02, AFKT03]. A ThreadSafe qualifier [GPB+06] could indicate that a given field should contain a thread-safe implementation of a given interface; this is more flexible than annotating the interface itself to require that all implementations must be thread-safe. Annotations can identify performance characteristics or goals; for example, some collections should not be iterated over, and others should not be used for random access. Annotations (both type qualifiers and others) can specify cut points in aspect-oriented programming (AOP) [EM04]. Flow-sensitive type qualifiers [FTA02] can express typestate properties such as whether a file is in the open, read, write, readwrite, or closed state, and can guarantee that a file is opened for reading before it is read, etc. The Vault language’s type guards and capability states are similar [DF01].
The Checker Framework (https://checkerframework.org/) gives a way to create pluggable type-checkers. A pluggable type-checker verifies absence of certain bugs (and also verifies correct usage of type qualifiers). The Checker Framework is distributed with a set of example type-checkers. The Checker Framework is built on the Type Annotations (JSR 308) syntax, though it also permits annotations to be written in comments for compatibility with previous versions of Java.
In Java SE 6, annotations can be written only on method parameters and the declarations of packages, classes, methods, fields, and local variables. Additional annotations are necessary in order to fully specify Java classes and methods.
This section gives examples of the annotation syntax specified in Sections 2.1 and 5. This list is not necessarily exhaustive (but if you notice something missing, let us know so that we can add it). Section B.2 motivates annotating these locations by giving the meaning of annotations that need to be applied to these locations.
Map<@NonNull String, @NonEmpty List<@Readonly Document>> files;
o.<@NonNull String>m("...");
class Folder<F extends @Existing File> { ... }
Collection<? super @Existing File>
interface WonderfulList<@Reified E> { ... }
class UnmodifiableList<T> implements @Readonly List<@Readonly T> { ... }
void monitorTemperature() throws @Critical TemperatureException { ... }
public String toString() @Readonly { ... }
public void write() @Writable throws IOException { ... }
It is not currently possible to express constraints on the generic parameters of the receiver. See Section D.2.1.
class Invocation {
@Immutable Invocation() { ... }
...
}
Note that the result of a constructor is different from the receiver. The receiver only exists for inner class constructors. It is the containing object, and in the body of the constructor it is referred to as Supertype.this. In the constructor body, the result is referred to as this. In any non-constructor, the receiver (if any) is referred to as this.
@Readonly Document [][] docs1 = new @Readonly Document [2][12]; // array of arrays of read-only documents Document @Readonly [][] docs2 = new Document @Readonly [2][12]; // read-only array of arrays of documents Document[] @Readonly [] docs3 = new Document[2] @Readonly [12]; // array of read-only arrays of documents
This syntax permits independent annotations for each distinct level of array, and for the elements.
myString = (@NonNull String) myObject;It is not permitted to omit the Java type, as in myString = (@NonNull) myObject;; see Sections B.2 and D.2.4.
new @Interned MyObject()
For generic constructors (JLS §8.8.4), the annotation follows the explicit type arguments (JLS §15.9):
new <String> @Interned MyObject()
boolean isNonNull = myString instanceof @NonNull String;It is not permitted to omit the Java type, as in myString instanceof @NonNull; see Sections B.2 and D.2.4.
new @NonEmpty @Readonly List<String>(myNonEmptyStringSet)
Class<@NonNull String> c = @NonNull String.class;
@NonNull Type.field
This section gives examples of annotations that a programmer may wish to place on a type. Each of these uses is either impossible or extremely inconvenient in the absence of the new locations for annotations proposed in this document. For brevity, we do not give examples of uses for every type annotation. The specific annotation names used in this section, such as @NonNull, are examples only; this document does not define any annotations, merely specifying where they can appear in Java code.
It is worthwhile to permit annotations on all uses of types (even those for which no immediate use is apparent) for consistency, expressiveness, and support of unforeseen future uses. An annotation need not utilize every possible annotation location. For example, a system that fully specifies type qualifiers in signatures but infers them for implementations [GF05] may not need annotations on typecasts, object creation, local variables, or certain other locations. Other systems may forbid top-level (non-type-argument, non-array) annotations on object creation (new) expressions, such as new @Interned Object().
Generic collection classes are declared one level at a time, so it is easy to annotate each level individually.
It is desirable that the syntax for arrays be equally expressive. Here are examples of uses for annotations on array levels:
A simple example is a method that accepts a list of files to search. null may be used to indicate that there were no files to search, but when a list is provided, then the Files themselves must be non-null. Using JSR 308, a programmer would declare the parameter as @NonNull File @Nullable [] filesToSearch — more concisely, depending on the default nullness, as either File @Nullable [] filesToSearch or @NonNull File [] filesToSearch
The opposite example, of a non-null array with nullable elements, is typical of fields in which, when an array element is no longer relevant, it is set to null to permit garbage collection.
A type qualifier on a formal parameter is a contract regarding what the method may (or may not) do with that parameter. Since the method receiver (this) is an implicit formal parameter, programmers should be able to express type qualifiers on it, for consistency and expressiveness. An annotation on the receiver is a contract regarding what the method may (or may not) do with its receiver.
For example, consider the following method:
package javax.xml.bind;
class Marshaller {
void marshal(@Readonly Object jaxbElement,
@Mutable Writer writer) @Readonly {
...
}
}
The annotations indicate that marshal modifies its second parameter but does not modify its first parameter nor its receiver.
A receiver annotation is different than a class annotation, a method annotation, or a return value annotation:
Since a receiver annotation is distinct from other annotations, new syntax is required for the receiver annotation. The syntax is adopted from C++, just as the overall syntax of Java was. This syntax is cleaner than creating a parallel annotation, such as @ReadonlyReceiver, for each type annotation, and is also cleaner than changing the definition of annotations to permit writing the similar @Receiver(@Readonly).
As with Java’s annotations on formal parameters, annotations on the receiver do not affect the Java signature, compile-time resolution of overloading, or run-time resolution of overriding. The Java type of every receiver in a class is the same — but their annotations, and thus their qualified type in a type qualifier framework, may differ.
Some people question the need for receiver annotations. In case studies [PAC+08], every type system required some receiver annotations. Even the Nullness type system required them to express whether the receiver was fully initialized (only in a fully-initialized object can fields be guaranteed to be non-null). So, the real question is how to express receiver annotations, not whether they should exist.
There are two distinct reasons to annotate the type in a type cast: to fully specify the casted type (including annotations that are retained without change), or to indicate an application-specific invariant that is beyond the reasoning capability of the Java type system. Because a user can apply a type cast to any expression, a user can annotate the type of any expression. (This is different than annotating the expression itself; see Section D.2.3.)
@Readonly Object x; ... (@Readonly Date) x ...
the cast preserves the annotation part of the type and changes only the Java type. If a cast could not be annotated, then a cast would remove the annotation:
@Readonly Object x; ... (Date) x ... // annotation processor issues warning due to casting away @Readonly
This cast changes the annotation; it uses x as a non-@Readonly object, which changes its type and would require a run-time mechanism to enforce type safety.
An annotation processor could permit the unannotated cast syntax but implicitly add the annotation, treating the cast type as @Readonly Date. This has the advantage of brevity, but the disadvantage of being less explicit and of interfering somewhat with the second use of cast annotations. Experience will indicate which design is better in practice.
As a trivial example, the following cast changes the annotation but is guaranteed to be safe at run time:
final Object x = new Object(); ... (@NonNull Object) x ...
An annotation processing tool could trust such type casts, perhaps issuing a warning to remind users to verify their safety by hand or in some other manner. An alternative approach would be to check the type cast dynamically, as Java casts are, but we do not endorse such an approach, because annotations are not intended to change the run-time behavior of a Java program and because there is not generally a run-time representation of the annotations.
Annotations on type tests (instanceof) allow the programmer to specify the full type, as in the first justification for annotations on type casts, above. However, the annotation is not tested at run time — the JVM only checks the base Java type. In the implementation, there is no run-time representation of the annotations on an object’s type, so dynamic type test cannot determine whether an annotation is present. This abides by the intention of the Java annotation designers, that annotations should not change the run-time behavior of a Java program.
Annotation of the type test permits the idiom
if (x instanceof MyType) {
... (MyType) x ...
}
to be used with the same annotated type T in both occurrences. By contrast, using different types in the type test and the type cast might be confusing.
To prevent confusion caused by incompatible annotations, an annotation processor could require the annotation parts of the operand and the type to be the same:
@Readonly Object x;
if (x instanceof Date) { ... } // error: incompatible annotations
if (x instanceof @Readonly Date) { ... } // OK
Object y;
if (y instanceof Date) { ... } // OK
if (y instanceof @NonNull Date) { ... } // error: incompatible annotations
(As with type casts, an annotation processor could implicitly add a missing annotation; this would be more concise but less explicit, and experience will dictate which is better for users.)
As a consequence of the fact that the annotation is not checked at run time, in the following
if (x instanceof @A1 T) { ... }
else if (x instanceof @A2 T) { ... }
the second conditional is always dead code. An annotation processor may warn that one or both of the instanceof tests is a compile-time type error.
A non-null qualifier is a special case because it is possible to check at run time whether a given value can have a non-null type. A type-checker for a non-null type system could take advantage of this fact, for instance to perform flow-sensitive type analysis in the presence of a x != null test, but JSR 308 makes no special allowance for it.
Java’s new operator indicates the type of the object being created. As with other Java syntax, programmers should be able to indicate the full type, even if in some cases (part of) the type can be inferred. In some cases, the annotation cannot be inferred; for instance, it is impossible to tell whether a particular object is intended to be mutated later in the program or not, and thus whether it should have a @Mutable or @Immutable annotation. Annotations on object creation expressions could also be statically verified (at compile time) to be compatible with the annotations on the constructor.
Annotations on type parameter bounds (extends) and wildcard bounds (extends and super) allow the programmer to fully constrain generic types. Creation of objects with constrained generic types could be statically verified to comply with the annotated bounds.
Annotations on class inheritance (extends and implements) are necessary to allow a programmer to fully specify a supertype. It would otherwise be impossible to extend the annotated version of a particular type t (which is often a valid subtype or supertype of t) without using an anonymous class.
These annotations also provide a convenient way to alias otherwise cumbersome types. For instance, a programmer might declare
final class MyStringMap extends
@Readonly Map<@NonNull String, @NonEmpty List<@NonNull @Readonly String>> {}
so that MyStringMap may be used in place of the full, unpalatable supertype. (However, also see Section D.5 for problems with this approach.)
Annotations in the throws clauses of method declarations allow programmers to enhance exception types. For instance, programs that use the @Critical annotation from the above examples could be statically checked to ensure that catch blocks for @Critical exceptions are not empty.
As discussed in Section B.2, it is desirable to be able to independently annotate both the base type and each distinct level of a nested array. Forbidding annotations on arbitrary levels of an array would simplify the syntax, but it would reduce expressiveness to an unacceptable degree. The syntax of array annotations follows the same general prefix rule as other annotations, though it looks slightly different because the syntax of array types is different than the syntax of other Java types. (Arrays are less commonly used than generics, so even if you don’t like the array syntax, it need not bother you in most cases.)
Most programmers read the Java type String[][] as “array of arrays of Strings”. Analogously, the construct new String[2][5] is “new length-2 array of length-5 array of Strings”. After a = new String[2][5], a is an array with 2 elements, and a[1] is a 5-element array.
In other words, the order of reading an array type is left-to-right for the brackets, then left-to-right for the base type.
type: String [] [] order of reading: 2-------------> 1 ------------------------>
To more fully describe the 2x5 array, a programmer could use the type “length-2 array of length-5 array of Strings”:
type: String @Length(2) [] @Length(5) [] order of reading: 2-------------> 1 ------------------------>
The prefix notation is natural, because the type is read in exactly the same order as any Java array type. As another example, to express “non-null array of length-10 arrays of English Strings” a programmer would write
type: @English String @NonNull [] @Length(10) [] order of reading: 2-------------> 1 ----------------------->
An important property of this syntax is that adding array levels does not change the meaning of existing annotations. For example, var1 has the same annotations as the elements of arr2:
@NonNull String var1; @NonNull String[] arr2;
because in each case @NonNull refers to the String, not the array. This consistency is especially important since the two variables may appear in a single declaration:
@NonNull String var1, arr2[];
A potential criticism is that a type annotation at the very beginning of a declaration does not refer to the full type, even though declaration annotations (which also occur at the beginning of the declaration) do refer to the entire variable. As an example, in @NonNull String[] arr2; the variable arr2 is not non-null. This is actually a criticism of Java itself, not of the JSR 308 annotation extension, which is merely consistent with Java. In a declaration String[] arr2;, the top-level type constructor does not appear on the far left. An annotation on the whole type (the array) should appear on the syntax that indicates the array — that is, on the brackets.
Other array syntaxes can be imagined, but they are less consistent with Java syntax and therefore harder to read and write. Examples include making annotations at the beginning of the type refer to the whole type, using a postfix syntax rather than a prefix syntax, and postfix syntax within angle brackets as for generics.
An annotation before a method declaration annotates either the return type, or the method declaration. There is never any ambiguity regarding the programmer intention: in that location, a type annotation annotates the return type, and a declaration annotation annotates the method itself. The @Target meta-annotation indicates whether an annotation is a type annotation. Field declarations are treated similarly.
Suppose that we have these annotation declarations:
@Target(ElementType.TYPE_USE)
@interface NonNegative { }
@Target(ElementType.METHOD)
@interface Override { }
@Target(ElementType.FIELD)
@interface GuardedBy { ... }
Then, in
@Override
@NonNegative int getHeight() { ... }
@Override applies to the method and @NonNegative applies to the return type. Furthermore, in these two field declarations
@NonNegative int balance;
@GuardedBy("myLock") long lastAccessedTime;
the annotation @NonNegative applies to the field type int, not to the whole variable declaration nor to the variable balance itself. The annotation @GuardedBy("accessLock") applies to the field lastAccessedTime.
Here are a few facts that follow from the specification. For brevity, we use “type annotation” as shorthand for “an annotation that is meta-annotated with @Target(ElementType.TYPE_USE)”.
Strictly speaking, a @Target meta-annotation could indicate that an annotation is both a type annotation and a declaration annotation. In such a case, the annotation would apply to both the return type and the method declaration, and it would exist twice in the class file. We have not found an example where that is desirable; although it is legal, it is considered bad style.
When an annotation has no @Target meta-annotation (which is bad style!), it is treated as if it applies to all locations. For example, the @Deprecated annotation lacks a @Target annotation, so in this code:
@Deprecated int foo() { return 0; }
the @Deprecated annotation applies to both the foo method, and the int data type. (It will be ignored by any type annotation processor, so it does no harm on the int data type.) The @Deprecated annotation appears twice in the AST during annotation processing, and it appears twice in the classfile. Classfile processing tools, such as decompilers, must take care not to write an annotation twice in the decompiled code. This requirement does not add extra work when constructing the tool, since the tool has to handle the case where the @Target annotation explicitly included both ElementType.METHOD and ElementType.TYPE_USE.
In summary: for certain syntactic locations, which target (Java construct) is being annotated depends on the annotation. There is no ambiguity for the compiler: the compiler applies the annotation to every target that is consistent with its meta-annotation (see Section 2.3). In practice, programmers have no difficulty in understanding where a given annotation applies.
This section primarily discusses tool modifications that are consequences of JSR 308’s changes to the Java syntax and class file format, as presented in Sections 2 and 4.
The syntax extensions described in Section 2 require the Java compiler to accept annotations in the proposed locations and to add them to the program’s AST. The relevant AST node classes must also be modified to store these annotations.
Javac’s -Xprint functionality reads a .class file and prints the interface (class declarations with signatures of all fields and methods). (The -Xprint functionality is similar to javap, but cannot provide any information about bytecodes or method bodies, because it is implemented internally as an annotation processor.) This must be updated to print the extended annotations as well. Also see Section C.4.
Section 3 requires compilers to place certain annotations in the class file. This is consistent with the principle that annotations should not affect behavior: in the absence of an annotation processor, the compiler produces the same bytecodes for annotated code as it would have for the same code without annotations. (The class file may differ, since the annotations are stored in it, but the bytecode part does not differ.)
This may change the compiler implementation of certain optimizations, such as common subexpression elimination, but this restriction on the compiler implementation is unobjectionable for three reasons.
Java compilers can often produce bytecode for an earlier version of the virtual machine, via the -target command-line option. For example, a programmer could execute a compilation command such as javac -source 7 -target 5 MyFile.java. A Java 7 compiler produces a class file with the same attributes for type annotations as when the target is a version 7 JVM. However, the compiler is permitted to also place type annotations in declaration attributes. For instance, the annotation on the top level of a return type would also be placed on the method (in the method attribute in the class file). This enables class file analysis tools that are written for Java SE 5 to view a subset of the type qualifiers (lacking generics, array levels, method receivers, etc.), albeit attached to declarations.
A user can use a Java SE 5/6 compiler to compile a Java class that contains type annotations, so long as the type annotations only appear in places that are legal in Java SE 5. Furthermore, the compiler must be provided with a definition of the annotation that is meta-annotated not with @Target(ElementType.TYPE_USE) (since ElementType.TYPE_USE does not exist in Java SE 5/6), but with no meta-annotation or with a meta-annotation that permits annotations on any declaration.
The Java Model AST of JSR 198 (Extension API for Integrated Development Environments) [Cro06] gives access to the entire source code of a method. This AST (abstract syntax tree) must be updated to represent all new locations for annotations.
Sun’s Tree API, which exposes the AST (including annotations) to authors of javac annotation processors (compile-time plug-ins), must be updated to reflect the modifications made to the internal AST node classes described in Section 2. The same goes for other Java compilers, such as that of Eclipse).
Like reflection, the JSR 269 (annotation processing) model does not represent constructs below the method level, such as individual statements and expressions. Therefore, it needs to be updated only with respect to declaration-related annotations (the top of Figure 1; also see Section D.7). The JSR 269 model, javax.lang.model.*, already has some classes representing annotations; see https://docs.oracle.com/javase/6/docs/api/javax/lang/model/element/package-summary.html. The annotation processing API in javax.annotation.processing must also be revised.
To do: Complete this design.
The java.lang.reflect.* and java.lang.Class APIs give access to annotations on public API elements such as classes, method signatures, etc. They must be updated to give the same access to the new extended annotations in the top of Figure 1.
Here are a few examples (the design is not yet complete).
Reflection gives no access to method implementations, so no changes are needed to provide access to annotations on casts (or other annotations inside a method body), type parameter names, or similar implementation details.
The Mirror API com.sun.mirror.* need not be updated, as it has been superseded by JSR 269 [Dar06].
Method Method.getParameterAnnotations() returns the annotations on the parameter declarations, just as in Java SE 6. It does not return type annotations. There is no point in new methods that parallel it, such as Method.getReceiverAnnotation (for the receiver this) and Method.getReturnAnnotation (for the return value). Rather, the interface will provide a uniform mechanism for querying annotations on types.
The semantics of reflective invocation is not changed. (The changes described in this section are to APIs that query classes, method signatures, etc.) For instance, suppose that a program reflectively calls a method with a parameter whose type is annotated as @Readonly, but the corresponding argument has a declared type that is non-@Readonly. The call succeeds. This is a requirement for backward compatibility: the existence of annotations in the class file should not cause a standard JVM to behave differently than if the annotations are not present (unless the program uses reflection to explicitly examine the annotations). Likewise, other reflective functionality such as AtomicReferenceFieldUpdater can bypass annotation constraints on a field.
No modifications to the virtual machine are necessary. (The changes to reflection (Section C.3) do change virtual machine APIs in a minor way, but the representation of execution of bytecodes is unaffected.)
The javap disassembler must recognize the new class file format and must output annotations.
The pack200/unpack200 tool must preserve the new attributes through a compress-decompress cycle.
The compiler and other tools that read class files are trivially compatible with class files produced by a Java SE 5/6 compiler. However, the tools would not be able to read the impoverished version of type qualifiers that is expressible in Java SE 5 (see Section C.1). It is desirable for class file tools to be able to read at least that subset of type qualifiers. Therefore, APIs for reading annotations from a class file should be dependent on the class file version (as a number of APIs already are). If the class file version indicates Java SE 5 or 6, and none of the extended annotations defined by JSR 308 appear in the class file, then the API may return (all) annotations from declarations when queried for the annotations on the top-level type associated with the declaration (for example, the top-level return type, for a method declaration).
Javadoc must output annotations at the new locations when those are part of the public API, such as in a method signature.
Similar modifications need to be made to tools outside the Sun JDK, such as IDEs (Eclipse, IDEA, JBuilder, jEdit, NetBeans), other tools that manipulate Java code (grammars for CUP, javacc), and tools that manipulate class files (ASM, BCEL). These changes need to be made by the authors of the respective tools.
A separate document, “Custom type qualifiers via annotations on Java types” (https://checkerframework.org/jsr308/java-type-qualifiers.pdf), explores implementation strategies for annotation processors that act as type-checking compiler plug-ins. It is not germane to this proposal, both because this proposal does not concern itself with annotation semantics and because writing such plug-ins does not require any changes beyond those described in this document.
A separate document, “Annotation File Specification” (https://checkerframework.org/annotation-file-utilities/annotation-file-format.pdf), describes a textual format for annotations that is independent of .java or .class files. This textual format can represent annotations for libraries that cannot or should not be modified. We have built tools for manipulating annotations, including extracting annotations from and inserting annotations in .java and .class files. That file format is not part of this proposal for extending Java’s annotations; it is better viewed as an implementation detail of our tools.
The Type Annotations (JSR 308) specification may be extended in the future, both to improve the specification and also to add new material. Any improvement to Java’s annotation system is within bounds. The JSR is titled “Type Annotations” to avoid a vague and inelegant title such as “Extensions to Java’s annotation system”, and because the initial extensions enable annotations on types. However, non-type-related improvements to annotations are within scope and will be considered. This is especially true if the additional changes are small, there is no better venue to add such an annotation, and the new syntax would permit unanticipated future uses.
This section gives examples of further extensions to Java’s annotation system that may be considered, either for JSR 308 or in the future. Inclusion here is not an endorsement; this list is intended to be a resource for current and future language designers, explaining pros and cons of possible changes. If you care enough about one of these extensions, please volunteer to be the point person for the work. That could make them a reality.
Java has a massive worldwide user base, so the bar to modifying the Java language is quite high. Any proposed modification requires both compelling use cases (for many users, not just a niche community) that are currently impossible, and it requires a convincing demonstration (not just a claim) that there are no negative consequences, such as undesirable interactions with current language features or tools. Where possible, this section gives a brief discussion of benefits and potential problems as a starting point. (Problems that potentially apply to all extensions, such as existing code that reflects over annotations encountering unexpected structures, are not repeated for each extension.)
It is not a goal that JSR 308 is the last annotation-related JSR. JSR 308 may leave some issues to future language designers, just as JSR 175 (the previous annotations JSR [Blo04a]) did. Leaving issues unresolved is preferable to making hasty decisions, or decisions with unknown technical implications. By contrast, it is a goal of JSR 308 not to unnecessarily close off realistic future avenues of extension.
It may be desirable for some annotations to be specified more than once at a single location. Currently, this is impossible: “It is a compile-time error if a declaration is annotated with more than one annotation for a given annotation type” [GJSB05, §9.7]. (By contrast, C# supports duplicate annotations on a given program element.)
As a related motivation, array-valued annotations can be clumsy to write:
@Resources({
@Resource(name = "db1", type = DataSource.class)
@Resource(name = "db2", type = DataSource.class)
})
public class MyClass { ... }
A cleaner syntax is desirable for both purposes:
@Resource(name = "db1", type = DataSource.class)
@Resource(name = "db2", type = DataSource.class)
public class MyClass { ... }
We note two possible approaches to this problem: desugaring into arrays, or adding a getAnnotations method to two interfaces. Among the two proposals, the second is cleaner, more elegant, and easier for clients to use. However, it requires either changing existing interfaces or creating new ones.
For backward compatibility, in both approaches an annotation may be repeated only if its definition is meta-annotated with @Repeatable:
public @interface Repeatable {
public boolean value() = true;
}
In the below, assume that the definitions of A and B are meta-annotated with @Repeatable.
It is already possible in Java SE 6 for a location to have both an inherited and an explicit annotation of the same type. The @Inherited meta-annotation specification states that a getAnnotation[s] query returns only the explicit annotation. This requirement should perhaps be relaxed, even if inheritance among annotations (Section D.3.1) is not permitted. For some annotation processors, it is best to check that the explicit annotation is consistent with the overridden, inherited one and possibly merge information in the two annotations. The current design makes that inconvenient to do.
Desugar duplicate annotations into the current array syntax. For instance, desugar
@A(1) @B @A(2) Object x;
into
@AContainer({@A(1), @A(2)}) @B Object x;
This approach treats duplicate annotations as purely a syntactic convenience; it does not change annotations in any deep way. This approach is compatible with certain existing J2EE annotation processors that are already written to process both @A and @AContainer.
This proposal would need to specify what happens if only one @A annotation is present. Most likely, the desugaring would still place @A into a singleton @AContainer annotation. Otherwise, client code would become more complicated, since it would have to deal with the possibility of either an @A annotation or an @AContainer annotation.
This proposal would need to specify what happens if both @A and @AContainer annotations are present. Most likely, the desugaring would insert @A into the existing @AContainer annotation, either at the front or the back depending on whether the @A or @AContainer annotation came first. Otherwise, client code would become more complicated, since it would have to deal with the possibility of both an @A annotation or an @AContainer annotation, and with handling the relative order.
This proposal would need to specify what happens if the user queries for annotation @A: would that always throw an exception (probably the best choice), or succeed if one @A annotation is present, or succeed if an @AContainer annotation with a length-1 array is present, or return the first @A annotation?
One problem with this proposal is that it loses the ordering of differently-named annotations (even if the ordering of same-named annotations is preserved within the container). For example, it cannot distinguish these declarations:
@A(1) @B @A(2) Object x; @A(1) @A(2) @B Object x;
Another problem is that it requires defining an @AContainer annotation for each annotation @A, or else annotation @A cannot be duplicated. It would be better not to burden programmers with writing and maintaining this boilerplate code.
Another problem with this proposal is that it makes other possible extensions to annotations, such as inheritance among annotations (Section D.3.1), more difficult and clumsy to implement and use. JSR 308 should aim to preserve, not to preclude, future avenues of extension (see Section D).
Add a new getAnnotations method, which returns multiple annotations, to two interfaces: java.lang.reflect.AnnotatedElement and javax.lang.model.element.Element.
<T extends Annotation> T[] getAnnotations(Class<T> annotationClass)
that returns all annotations for the specified annotation type that are present on the receiver element. As with getDeclaredAnnotations, the return value is never null but may have zero length.
<T extends Annotation> T getAnnotation(Class<T> annotationClass)
When duplicate annotations exist, method getAnnotation could either give the first one (convenient for backward compatibility) or throw an exception (convenient to prevent erroneous processing). The latter seems like a better choice.
The arrays returned by the methods getAnnotations(Class), getAnnotations(), and getDeclaredAnnotations() are required to preserve the ordering of the annotations as they were declared in the original .java source file.
No other changes would be necessary. Existing code that uses some other workaround (like special @AContainer annotations) would continue to work. Or, it could be converted to take advantage of this new mechanism.
The problem with this proposal is that it requires adding a single method, getAnnotations, to two interfaces. While the changes would not break upward binary-compatibility, they would break source-compatibility: existing implementations of Element and AnnotatedElement could not be re-compiled under Java 7 without adding a getAnnotations method. Sun’s policy generally frowns upon such changes that break source-compatibility, though they are sometimes made nonetheless. For example, the SQL interfaces have far more implementations than the compiler or reflection APIs, but Java SE 6 added three methods to the java.sql.Statement interface and made it extend Wrapper.
An alternative to changing the Element (and AnnotatedElement) interface is to create a new Element2 interface that augments Element with the getAnnotations method, and to mark Element as deprecated. Then, programmers should rewrite their existing code to use the Element2 interface instead of the Element interface. However, their code continues to compile until they make the change. A problem with this change is that it requires making a copy (with names suffixed “2”) of the entire javax.lang.model package, which is ugly.
Another alternative is to change the Elements utility interface. The new getAnnotations methods logically belong in Element, but they can be added to the Elements interface because it is explicitly documented with “Compatibility Note: Methods may be added to this interface in future releases of the platform.” Thus, adding static methods in Elements is a path toward extending functionality without changing core interfaces/classes such as Element. A problem with this proposal is that clients must remember to use the methods of Elements when functionality is not present in Element. A programmer browsing the Element interface might not think to check for static methods in Elements.
The JSR 175 Design FAQ [Blo04b] briefly addresses why this feature was not originally included:
Why is it illegal [to] annotate a single element repeatedly with the same annotation type?A similar effect is available by annotating once with an annotation type whose sole element is of an array type. The resulting system is simpler and cleaner: you don’t have to worry about whether you’re going to get back a single instance of the annotation type or multiple instances when you query an annotation. Also it’s consistent with the usual restriction on modifiers.
The argument that “the resulting system is simpler and cleaner” is perplexing. A client must check for both the singular and plural versions of the annotation and must handle the case when both the singular and the plural versions are present. By contrast to the stated claim, a system for duplicate annotations with an interface that always returns a list (or always returns an array) is what is needed to relieve the client of “worry about whether you’re going to get back a single instance”.
It is sometimes desirable for a method to express constraints on the generic parameters of the receiver, just as is already possible for other formal parameters. This is not currently possible (see the grammar of Section 5).
For example, consider the Collection.toArray() method, which returns an array of objects, where the objects have the same annotation as the elements of the receiver. A natural way to write this would be
public @Polynull Object [] toArray() ArrayList<@PolyNull E> { ... }
except that this is illegal syntax: “ArrayList<@PolyNull E>” in the receiver position is not legal. It is permissible to write an annotation on the raw type of the receiver, as in
public Object [] toArray() @Readonly { ... }
but not on the receiver’s type parameters.
Here are other examples:
interface List<T> {
// The size() method changes neither the receiver nor any of the elements.
public int size() @Readonly List<@Readonly T> { ... }
}
class MyMap<T,U> {
// The method requires that the map's values be non-null, but the keys may be arbitrary.
public void requiresNonNullKeys() MyMap<T, @NonNull U> { ... }
}
Annotations on statements (or on some subset of statements, such as blocks or loops) would be useful for a variety of purposes, including atomicity/concurrency. Supporting annotations on statements would require defining both Java syntax and a convention for storing the information in the class file. See https://bitbucket.org/typetools/jsr308-langtools/wiki/AnnotationsOnStatements for a proposal that summarizes why statement annotations are desirable, and that proposes a Java syntax, a classfile storage format, and how other tools will accommodate them. If you would like to help this feature become a reality, then please pitch in! You can join the jsr308-discuss@googlegroups.com mailing list (via http://groups.google.com/group/jsr308-discuss), expand the partial design on the wiki, or work on the implementation.
The JSR 175 Design FAQ [Blo04b] briefly addresses why this feature was not originally included:
Why can’t you annotate arbitrary program elements such as blocks and individual statements?This would greatly complicate the annotation syntax: We would have to sacrifice the simplicity of saying that annotations are simply modifiers, which can be used on declarations.
Annotating an expression indicates some property of the computation, such as that it should be performed atomically, that it acquires no locks, or that it should be formatted specially by an IDE. JSR 308 does not support expression annotations, because we have not yet discovered compelling use cases for them that cannot be equally well supported by statement annotations. (A minor inconvenience is that the use of statement annotations may require the programmer to create a separate statement for the expression to be annotated.) Expression annotations are not type annotations and are different than annotating a type cast, which indicates a property of a value (the result of an expression).
Arbitrary values can be annotated using an annotation on a cast:
(@Language("SQL") String) "select * from foo"
A possible shorthand would be to permit the Java type to be implicit:
(@Language("SQL")) "select * from foo"
This is not permitted, nor may a cast be omitted in a type test, as in “x instanceof @NonNull”. There are several reasons for this decision:
An even shorter shorthand would drop the parentheses:
@Language("SQL") "select * from foo"
In addition to the benefits and problems noted above, such an annotation is syntactically ambiguous with an expression annotation. Whether an annotation applies to expressions or to types is clear from the annotation’s documentation and its @Target meta-annotation, similarly to how it is determined whether an annotation applies to a type or to a declaration (Section B.4).
It would be possible to permit annotations only on blocks and/or loops, as a restricted special case of statements. This is less general than permitting annotations on statements, and uses are more syntactically cluttered (for instance, this requires a statement to be converted into a block before it can be annotated). Most declarations could not be annotated as statements because enclosing the declaration in a block to annotate it would change (and limit) the variable’s scope. This limitation in flexibility does yield the advantage that there would be no syntactic ambiguity between (say) statement annotations and declaration or type annotations.
Similarly, permitting annotations on partial constructs (such as only the body of a loop) appears both more complex, and no more useful, than annotating complete constructs (such as a full statement).
In Java SE 6, annotations cannot subclass one another, and an annotation is not allowed to extend/implement any interfaces. These restrictions make it difficult to share behavior or to express similarities or relationships among annotation types. For example, it is inconvenient to define annotations with choices in their structure, such as a discriminated union that uses field names that act as explicit tags. It is impossible to create annotation processors or APIs that work with a specific set of annotations (say, all those with a given set of element names).
Permitting inheritance among annotations would solve these problems. (To work around the problem, one could meta-annotate an annotation as a “subannotation” of another, and then the annotation processor could do all the work to interpret the meta-annotation. This is clumsy and indirect.)
A potential objection is that multiple annotation processors might try to claim the same annotation, one directly and one by claiming its supertype. This can be accommodated. In general, annotation processor designers should, and can, avoid this situation by running one processor at a time for a family of annotations that are related by inheritance.
Without other changes, inheritance among annotations would not enable recursive or self-referential annotations (see Section D.3.2).
There are two general designs that permit inheritance among annotations: permitting annotations to subclass other annotations, or permitting annotations to extend/implement interfaces, or both. We briefly elaborate on the two designs.
Permitting inheritance among annotations requires that the final modifier on an annotation type declaration works just like for ordinary classes: it prevents creation of subtypes of the given annotation. A framework that defines annotation @A may not want to load an untrusted subclass @B of @A into the framework.
The specification of the getAnnotation(Class) method must be updated. Suppose that annotation types Sub1 and Sub2 are direct subtypes of Super, and Super has a value field of type int. The specification must indicate the behavior of getAnnotation(Super.class) when called on locations with all possible combinations of the three annotations, with all possible combinations of arguments. Here are a few examples:
Reflection APIs may also need to be updated.
As is clear from the examples above, subclassing annotations raises the possibility of multiple effective annotations applying at a location. Therefore, subclassing annotations makes most sense if the specification already permits duplicate annotations (see Section D.1). In that case, the new getAnnotations(Class) method returns all of the annotations. (Then, the client can decide what to do with them, such as raise an error, merge the annotations, or take some other action.)
The JSR 175 Design FAQ [Blo04b] briefly addresses why this feature was not originally included:
Why don’t you support annotation subtyping (where one annotation type extends another)?It complicates the annotation type system, and makes it much more difficult to write “Specific Tools”.
…
“Specific Tools” — Programs that query known annotation types of arbitrary external programs. Stub generators, for example, fall into this category. These programs will read annotated classes without loading them into the virtual machine, but will load annotation interfaces.
Permitting annotation types to extend/implement interfaces is less powerful than subclassing among annotations. For example, interfaces can’t define default values, nor is it possible to subclass an annotation to enhance functionality. An advantage of this approach is that there is no possibility of multiple annotations with a given supertype appearing at a location.
One use case is marker interfaces. In another use case, the interface could define several elements/methods These could define a standard set of element names that are used by a set of specific annotations. APIs that use the interface could work with any annotation from that set.
The currently permitted annotation member types (“primitive types, String, Class and any invocation of Class, an enum type, an annotation type, or an array of one of the preceding types” [GJSB05, §9.6]) should be extended to include interfaces (or perhaps only interfaces that extend java.lang.annotation.Annotation).
In Java, an annotation can have a parameter (equivalently, a member) that is an annotation (JLS §9.6 and §9.7). JLS §9.7 gives this example:
@Author(@Name(first = "Joe", last = "Hacker"))
However, an annotation type cannot have a member of its own type. “It is a compile-time error if an annotation type T contains an element of type T, either directly or indirectly” [GJSB05, §9.6]. (If inheritance among annotations is permitted (see Section D.3.1), then the natural interpretation of this is that no element’s declared type may be either the annotation itself or any supertype thereof.)
As a particular example, it is not possible to define an annotation that takes an arbitrary annotation as a parameter, as in
@DefaultAnnotation(@AnyAnnotation)
These limitations reduce the expressiveness of annotations. Here are some examples:
Two examples of such annotations are the @DefaultAnnotation example above, and an annotation that expresses that a method is polymorphic over annotations (as opposed to polymorphic over types, as generics do).
Another example is annotation parameters that do not correspond to a type parameter. Consider a container class that is designed to hold things of a specific type (say, it interacts with the elements in domain-specific ways), not of an arbitrary type. The container has no type parameter, but it may still make sense to have annotations on the contained thing. Most of the examples of generics and arrays from Section B.2 also serve as examples here. The argument applies not just to container classes but also to algorithms that process data of a specific type.
These examples require the annotation declaration to be potentially recursive. However, they do not require that any recursive annotation ever be created in practice. They could be achieved even if creation of recursive annotation instances was prohibited.
More generally, the kind of data that can be encoded in annotations is limited. It is generally possible to store data which has a fixed structure, but it is much more difficult to encode more complex data, including hierarchical data. As the success of XML, YAML, and other such technologies has shown, the ability to encode structured data is very powerful for a wide variety of uses, many of which we cannot anticipate today. Today’s annotations only fulfill a small piece of that potential.
To express the base case, recursive annotations also require either inheritance among annotations (Section D.3.1), or nullable members (Section D.3.3), or a requirement that recursion takes place through an array (an empty array would terminate the recursion).
A more modest approach that makes annotations somewhat more expressive would be to permit inheritance among annotations (see Section D.3.1) without permitting possibly self-referential annotations.
The JSR 175 Design FAQ [Blo04b] briefly addresses why this feature was not originally included:
Why is it illegal for an annotation type to contain an element of the same type?At first glance it would appear that you could never annotate with such a type, so the prohibition might seem unnecessary, but you can get yourself into trouble if you also provide a default:
@interface Foo { Foo foo() default @Foo; } @Foo int i;If this were legal, the resulting annotation would be self-referential (or infinite), as well as useless.
Currently, it is possible to choose/create a special value and to store that value in an annotation field. The proposal would permit use of null instead of an explicit user-specified value.
The proposal makes some programming idioms slightly shorter. For example, specifying the base case of a recursive data structure will require simply writing null instead of calling a constructor. The proposal does not eliminate any substantial complexity when processing a data structure, but only converts a check against a given value into a check against null.
We note some possible objections to the proposal.
The proposal doesn’t make anything possible that was not possible before.
The programmer-defined special value provides better documentation than null, which might mean “none”, “uninitialized”, null itself, etc.
The proposal is more error-prone. It’s much easier to forget checking against null than to forget checking for an explicit value.
The proposal may make the standard idiom more verbose. Currently only the users of an annotation need to check for its special values. With the proposal, many tools that process annotations will have to check whether a field’s value is null lest they throw a null pointer exception.
Annotation types cannot have positional arguments (except for the value argument, when it is the only argument). This limitation makes writing annotations with multiple arguments more verbose than necessary.
On a somewhat related topic, the “SingleElementAnnotation” form relies on the field name value. This name is not always appropriate; designers who wish to permit the “SingleElementAnnotation” form are forced to give a member a confusing name. It would be nice to permit the “SingleElementAnnotation” form to initialize a different member than value (say, via a @ValueMember annotation).
JSR 308 does not define any annotations. JSR 308 extends the Java and class file syntax to permit annotations to be written in more places, and thus makes existing and future annotations more useful to programmers.
By contrast, JSR 305 “Annotations for Software Defect Detection” aims to define a small set of annotations. Examples include type annotations such as non-nullness (@Nonnull), signedness (@Nonnegative), tainting, and string format; and also declaration annotations such as whether a method’s return value should always be checked by the caller. A programmer who cares about code quality will use both annotations defined in the JSR 305 “standard library”, and also others that are defined by third parties or by the programmer himself. For more details about JSR 305, see https://jcp.org/en/jsr/detail?id=305 and http://groups.google.com/group/jsr-305/. Since JSR 305 appears to be abandoned, this section is now primarily of historical interest.
Any type annotation, including those defined by JSR 305, is of limited use without the JSR 308 syntax. Without the JSR 308 annotation syntax, a static checker may lose track of type qualifiers whenever a program uses generic types (e.g., collection classes), whenever a method is invoked on an object, whenever a cast is performed, whenever a class is subclassed, etc. From the point of code checking, using the old Java annotation syntax is even worse than the type unsoundness of pre-generics Java, when there was no compiler-enforced type correctness guarantee for collections classes. Therefore, use of JSR 305 without JSR 308 is much less effective.
The JSR 305 team does not plan any reference implementation. This hinders both programmers who want to use the annotations, and also people trying to interpret the meaning of the specification. By contrast, the JSR 308 reference implementation and the Checker Framework for compile-time type-checking have been available since January 2007. The Checker Framework contains checkers for many of the most useful annotations proposed in JSR 305.
(This section is not about whether the definition of an annotation may inherit from other classes/interfaces. Rather, it is about whether a class/interface inherits annotations from its superclasses/interfaces.)
The annotation type java.lang.annotation.Inherited (JLS §9.6.1.3) indicates that annotations on a class C corresponding to a given annotation type are inherited by subclasses of C. This implies that annotations on interfaces are not inherited, nor are annotations on members (methods, constructors, fields, etc.).
It might be useful to permit methods, fields, etc. to inherit annotations.
It might be useful to permit an annotation to be inherited from an interface as well as from a superclass.
It might be useful to permit annotation inheritance to merge information from the current and inherited annotations, instead of always choosing the inherited one.
Specifying a default for annotations can reduce code size and (when used carefully and sparingly) increase code readability. For instance, Figure 3 uses @DefaultQualifier("NonNull") to avoid the clutter of 5 @NonNull annotations. It would be nicer to have a general mechanism, such as
@DefaultAnnotation(NonNull.class, locations={ElementType.LOCAL_VARIABLE})
Defaults for annotations are a semantic issue that is out of the scope of JSR 308. It will be taken up by JSR 305 (“Annotations for Software Defect Detection” [Pug06]).
The defaulting syntax must also be able to specify the arguments to the default annotation (in the above example, the arguments to @NonNull).
A better syntax would use an annotation, not a string or class literal, as the argument to @DefaultAnnotation, as in
@DefaultAnnotation(@MyAnnotation(arg="foo"))
In Java, it is not possible to define an annotation that takes an arbitrary annotation as a parameter; see Section D.3.2.
An issue for JSR 260 (Javadoc) and JSR 305 (Annotation semantics) is how inherited and defaulted annotations are handled in Javadoc: whether they are written out in full, or in some abbreviated form. Just as too many annotations may clutter source code, similar clutter-reduction ideas may need to be applied to Javadoc.
An annotated type may be long and hard to read; compare Map<String, Object> to @NonNull Map<@NonNull String, @NonNull Object>. Class inheritance annotations and subclassing provides a partial solution, as noted in Section B.2 with the following example:
final class MyStringMap extends
@Readonly Map<@NonNull String, @NonEmpty List<@NonNull @Readonly String>> {}
This approach limits reusability: if a method is declared to take a MyStringMap parameter, then a Map (even of the right type, including annotations) cannot be passed to it. (By contrast, a MyStringMap can always be used where a Map of the appropriate type is expected.) Goetz [GPB+06] recommends exploiting Java’s type inference to avoid some (but not all) instances of the long type name.
In summary, a built-in typedef mechanism might achieve both code readability and reusability.
Changes to the class file syntax are out of the scope of JSR 308, which, for backward compatibility, currently does not change the way that existing annotations are stored in the class file. Class file syntax changes require modification of compilers, JVMs, javap, and other class file tools (see Sections C.4 and C.5).
Annotations could be stored in the constant pool, and use constant pool references from the annotation points. That would reduce class file size, especially if an annotation is used in many places in the same class, as is more likely once JSR 308 support is in place.
In order to reduce the size of the class file, some fields may be omitted from the .class file, in which case any access of them returns their default value.
A type-checking compiler plug-in (annotation processor) must process annotations (including those in method bodies), and it also must check each use of a variable/method whose declared type is annotated. For example, if a variable x is declared as @NonNull Object x;, then every assignment to x must be checked, because any assignment x = null; would be illegal.
The JSR 269 annotation processing API does not process annotations on local variables, as it is not designed to access method bodies. This limitation makes JSR 269 insufficient for creating a type-checking compiler plug-in.
An annotation and source code processing API for JSR 308 annotations could take advantage of JSR 198’s Java Model. The Java Model defines a parsed view into the contents of a source file that is intended for construction of IDE extensions that are portable across multiple IDEs — precisely the situation with compiler plug-ins. JSR 308 may be shipped without defining this API, but defining this API may be desirable in the future (say, in a later version of Java), particularly after more experience is gained with JSR 308 annotation processors.
JSR 308 (“Type annotations”) should be included under the Java SE 7 umbrella JSR (which lists the JSRs that are part of the Java SE 7 release). However, it should be a separate JSR because it needs a separate expert group. The expert group will have overlap with any others dealing with other added language features that might be annotatable (such as method-reference types or closures), to check impact.
The specification, reference implementation, and tests developed by the JSR 308 community are freely available, with no restrictions on use. As of build M4 (see https://openjdk.org/projects/jdk7/milestones/), the reference implementation has been incorporated into Sun’s OpenJDK Java implementation. Sun will integrate the specification into the JLS (without any conceptual changes or new information), will write additional tests, etc. We do not have control over the license terms that Sun places on its changes.
To ease the transition from standard Java SE 6 code to code with the extended annotations, the reference implementation recognizes the extended annotations when surrounded by comment markers:
List</*@Readonly*/ Object> myList;
This permits use of both standard Java SE 6 tools and the new annotations even
before Java SE 7 is released. However, it is not part of the proposal;
that is, it is not required that every Java compiler parses comments.
Sun’s OpenJDK implementation does not support a switch that
makes it recognize the new annotations when embedded in comments.
The Spec# [BLS04] extension to C# can be made compilable by
a standard C# compiler in a similar way, by enclosing its annotations in
special /*^…^*/ comment markers.
The /*@ comment syntax is a standard part of the
Splint [Eva96], ESC/Java [FLL+02], and
JML [LBR06] tools (that is, not
with the goal of backward compatibility).
Edits to the Java Language Specification (JLS): We need a document, complementary to the design document, that lists every edit that is required in the JLS. A preliminary step would be a list of all the locations that must be edited (for instance, by searching the entire JLS for uses of “annotation”, but the list will be a superset of the list of locations that were edited for JSR 175). The most important locations are the following.
Edits to the Java Virtual Machine Specification (JVMS) [LY99, LY07]: We need a document, complementary to the design document, that lists every edit that is required in the JVMS. The most important of these is the following:
JSR 308 will ship with a test suite (known as a TCK, or Technology Compatibility Kit).
Each tool that needs to be tested appears in Section 3; the TCK will include tests for each of them.
For each modified tool, we will test backward compatibility by passing all of its existing tests. (We may need to modify a few of them, for instance those that depend on specific bytecodes that are created by the compiler.)
We will test most other functionality by creating a set of Java programs that include annotations in every possible location. For instance, this can be used to test all aspects of the compiler (parsing, code generation, -Xprint).
We will provide multiple annotation processors (including at least one for checking @NonNull and one for checking @Interned) that utilize the new annotations, along with a test suite for each one. Each annotation processor’s test suite consists of annotated code, along with expected output from the given annotation processor. Since the annotation processors utilize all aspects of JSR 308, this serves as an additional end-to-end test of the JSR 308 implementation. As a side benefit, the annotation processors will be useful in their own right, will thereby illustrate the utility of JSR 308, and will serve as examples for people who wish to create their own type-checking plug-ins.
JSR 308 follows an unusually open and transparent process. Any interested party may participate, decisions are made by consensus to the greatest extent possible, and discussions are publicly archived. The Java Community Process requires that an expert group formally approve the JSR at each stage. The expert group members will only decide issues on which the group cannot obtain consensus.
When posting, please act professionally and courteously. For example, your arguments should be technical, specific, and based on logic; do not rely on your status or past accomplishments to convince others, and give specific examples rather than vague descriptions. As another example, it is perfectly acceptable to criticize a technical proposal, but do not make personal attacks. Obviously, you should read the specification and, preferably, try the implementation before posting to the mailing list.
Section A.1 gave many examples of how type qualifiers have been used in the past. Also see the related work section of [PAC+08].
C#’s attributes [ECM06, chap. 24] play the same role as Java’s annotations: they attach metadata to specific parts of a program, and are carried through to the compiled bytecode representation, where they can be accessed via reflection. The syntax is different: C# uses [AnnotationName] or [AnnotationName: data] where Java uses @AnnotationName or @AnnotationName(data); C# uses AttributeUsageAttribute where Java uses Target; and so forth. However, C# permits metadata on generic arguments, and C# permits multiple metadata instances of the same type to appear at a given location.
Like Java, C# does not permit metadata on elements within a method body. (The “[a]C#” language [CCC05], whose name is pronounced “annotated C sharp”, is an extension to C# that permits annotation of statements and code blocks.)
Harmon and Klefstad [HK07] propose a standard for worst-case execution time annotations.
Pechtchanski’s dissertation [Pec03] uses annotations in the aid of dynamic program optimization. Pechtchanski implemented an extension to the Jikes compiler that supports stylized comments, and uses these annotations on classes, fields, methods, formals, local variable declarations, object creation (new) expressions, method invocations (calls), and program points (empty statements). The annotations are propagated by the compiler to the class file.
Mathias Ricken’s LAPT-javac (https://ricken.us/research/laptjavac/) is a version of javac (version 1.5.0_06) that encodes annotations on local variables in the class file, in new Runtime{Inv,V}isibleLocalVariableAnnotations attributes. The class file format of LAPT-javac differs from that proposed in this document. Ricken’s xajavac (Extended Annotation Enabled javac) permits subtyping of annotations (https://ricken.us/research/xajavac/).
The Java Modeling Language, JML [LBR06], is a behavioral modeling language for writing specifications for Java code. It uses stylized comments as annotations, some of which apply to types.
Ownership types [CPN98, Boy04, Cla01, CD02, PNCB06, NVP98, DM05, LM04, LP06] permit programmers to control aliasing and access among objects. Ownership types can be expressed with type annotations and have been applied to program verification [LM04, Mül02, MPHL06], thread synchronization [BLR02, JPLS05], memory management [ACG+06, BSBR03], and representation independence [BN02].
JavaCOP [ANMM06] is a framework for implementing pluggable type systems in Java. Whereas JSR 308 uses standard interfaces such as the Tree API and the JSR 269 annotation processing framework, JavaCOP defines its own incompatible variants. A JavaCOP type checker must be programmed in a combination of Java and JavaCOP’s own declarative pattern-matching and rule-based language. JavaCOP’s authors have defined parts of over a dozen type-checkers in their language. Their paper does not report that they have run any of these type-checkers on a real program; this is due to limitations that make JavaCOP impractical (so far) for real use.
JACK makes annotations on array brackets refer to the array, not the elements of the array [MPPD08].
Matt Papi and Mahmood Ali designed and implemented the JSR 308 compiler as modifications to Sun’s OpenJDK javac compiler, and contributed to the JSR 308 design.
The members of the JSR 308 mailing list (http://groups.google.com/group/jsr308-discuss) provided valuable comments and suggestions. Additional feedback is welcome.
JSR 308 received the Most Innovative Java SE/EE JSR of the Year award in 2007, at the 5th annual JCP Program Awards. JSR 308’s spec leads (Michael Ernst and Alex Buckley) were nominated as Most Outstanding Spec Lead for Java SE/EE in 2008, at the 6th annual JCP Program Awards. Michael Ernst won a Java Rock Star award for a presentation on the Checker Framework, which builds on the Type Annotations syntax, at JavaOne 2009.