Spec-Zone .ru спецификации, руководства, описания, API

John R. Rose
Oct 4, 1996

Problem:

Java 1.1 introduces many new APIs, some of which supersede older ones. The new and old APIs co-exist in the same packages. The superseded ones add various difficulties to the programmer's job, the chief being the need to ignore and avoid them in most cases.

Partial Solution: If we mark superseded classes, methods, and fields clearly and unambiguously as "deprecated", the programmers will know where to concentrate their attention. Also, the compiler will be able to help with this process, by bringing unintentional uses of deprecated APIs to the programmer's attention.

Details:

Within documentation comments, we define the new paragraph tag "@deprecated". (See JLS, $$184.) It can appear in both class and member documentation comments, and will apply specifically the construct its comment introduces.

The tagged paragraph can be empty. If it is not, it should tell the programmer how to avoid using the deprecated feature.

There should be "@see" tagged paragraphs which refer to the new versions of the same functionality.

As far as the compiler is concerned, the presence of the string "@deprecated" at the beginning of a documentation comment line (except for whitespace) causes it to place a "Deprecated" attribute in the output classfile for the corresponding class, field, or method.

Moreover, in parallel with the access checking logic for classes and members, the compiler looks for "Deprecated" attributes of classes and members being accessed, and issues warnings when the deprecated classes or members are used.

Note that deprecation applies to classes and to individual members, not to their names. It is possible for a single method to have deprecated and non-deprecated overloadings. It is possible for a non-deprecated member to hide or override a deprecated one, which removes deprecation. It is the responsibility of the programmer to deprecate overrides of a deprecated method, if in fact they should be deprecated.

A deprecation warning is suppressed if the compilation unit containing the deprecation is being compiled at the same time as the compilation unit using the deprecated class or member. This allows legacy APIs to be built without warnings. There currently is no other way to suppress the deprecation warnings.

A hypothetical "-strict" option to the compiler could promote all warnings, including deprecation warnings, to hard errors. This is not in 1.1.

Javadoc should somehow highlight deprecation paragraphs, and perhaps should place a characteristic glyph next to references to deprecated names.

Practice:

Here is an example of the most common form of a deprecated method:

    /**
     * @deprecated
     * @see #getPreferredSize
     */
    public Dimension preferredSize() {
	return getPreferredSize();
    }

If the deprecation is a simple renaming, there is no need to say something like "this method is replaced by getPreferredSize"; the "@see"-tagged paragraph points the user to the replacement.

The deprecation may be more complex if the API was reorganized by other than a renaming of features. Here is an example of a method that is being retracted:

    /**
     * Delete multiple items from the list.
     *
     * @deprecated  Not for public use in the future.
     * This method is expected to be retained only as a package
     * private method.
     * @see #remove(int)
     * @see #removeAll()
     */
    public synchronized void delItems(int start, int end) {
    ...
    }

Valid reasons for wishing one's users to migrate to the new API include:

• the old API is insecure, buggy, or highly inefficient
• the old API is going away in a future release
• the old API encourages very bad coding practices

Not all of these reasons are of equal weight, yet deprecation is a reasonable (though not mandatory) choice in all these cases. Therefore, the use of deprecated APIs can never be made a hard error by default. Also, the deprecation comments need to help the user decide when to move to the new API, and so should briefly mention the technical reasons for deprecation.

(It is probably not good to specifically mention a timetable for phase-out of the deprecated API; this is a business decision that needs to be communicated other ways.)

It is not necessary to deprecate individual members of a deprecated class, unless of course the programmer wants to explain some specific point about one member.

When a feature is deprecated, it is a good idea to notify the engineering organization of this fact, so that other engineers can respond to the change (pro or con) in a timely manner.

Other Design Options:

This is the first time the compiler is in any way cognizant of document comment contents. (Once, it made sure that they were at the right place in the parse tree, but this is no longer true.) We did not lightly choose to make the compiler's behavior dependent on the contents of a comment.

One precedent for looking into comments is "lint", which uses comments like /*PRINTFLIKE2*/ to annotate C APIs.

Note that Java documentation comments are structured, and the structure is specified in the Java Language Specification. This means that we "own" the format of this particular sort of comment, so our design is somewhat more reasonable than the less-differentiated conventions used by lint.

We considered introducing pragmas, and using them for deprecation (e.g., <*deprecated*>) but this would open the door very wide to many sorts of incompatible extensions and, possibly, to serious abuses.

However, @deprecated does work like a pragma: It belongs in the source, expresses information about it, but does not modify its semantics. Unlike many pragmas, @deprecated is intended for a human audience.

The compiler emits warnings as a friendly and helpful reminder, but does not enforce any rules. It's up to the programmer to use the contents of the deprecation paragraph. Since the human is the primary user of the data, it belongs in comments.

We considered introducing a new keyword, something like "transient". This would solve the mechanical problem, but at the cost of modifying the language itself. In particular, we would have to take away an identifier from coders that we had not previously reserved (like "var"). Moreover, since the primary audience of deprecations is humans, a keyword-based approach is insufficient, and must be augmented by comment conventions, preferably ones which integrate with javadoc. Thus, the keyword approach does not pay.

Still, the compiler has heretofore ignored and should generally ignore documentation comments. This is an admittedly irregular exception to that practice. But the other design options for deprecation involve worse sorts of irregularity.