|
|
Programming With Assertions |
Documentation Contents |
An assertion is a statement in the JavaTM programming language that enables you to test your assumptions about your program. For example, if you write a method that calculates the speed of a particle, you might assert that the calculated speed is less than the speed of light.
Each assertion contains a boolean expression that you believe will be true when the assertion executes. If it is not true, the system will throw an error. By verifying that the boolean expression is indeed true, the assertion confirms your assumptions about the behavior of your program, increasing your confidence that the program is free of errors.
Experience has shown that writing assertions while programming is one of the quickest and most effective ways to detect and correct bugs. As an added benefit, assertions serve to document the inner workings of your program, enhancing maintainability.
This document shows you how to program with assertions. It covers the topics:
The assertion statement has two forms. The first, simpler form is:
assert Expression1 ;where Expression1 is a
boolean expression. When the system runs the assertion, it
evaluates Expression1 and if it is
false throws an AssertionError
with no detail message.
The second form of the assertion statement is:
assert Expression1 : Expression2 ;
where:
boolean expression. void.)
Use this version of the assert statement to provide a detail
message for the AssertionError. The system passes the value of
Expression2 to the appropriate
AssertionError constructor, which uses the string representation
of the value as the error's detail message.
The purpose of the detail message is to capture and communicate the details of the assertion failure. The message should allow you to diagnose and ultimately fix the error that led the assertion to fail. Note that the detail message is not a user-level error message, so it is generally unnecessary to make these messages understandable in isolation, or to internationalize them. The detail message is meant to be interpreted in the context of a full stack trace, in conjunction with the source code containing the failed assertion.
Like all uncaught exceptions, assertion failures are generally labeled in
the stack trace with the file and line number from which they were thrown.
The second form of the assertion statement should be used in preference to the
first only when the program has some additional information that might help
diagnose the failure. For example, if Expression1
involves the relationship between two variables x and
y, the second form should be used. Under these circumstances,
a reasonable candidate for Expression2 would be
"x: " + x + ", y: " + y.
In some cases Expression1 may be expensive to evaluate. For example, suppose you write a method to find the minimum element in an unsorted list, and you add an assertion to verify that the selected element is indeed the minimum. The work done by the assert will be at least as expensive as the work done by the method itself. To ensure that assertions are not a performance liability in deployed applications, assertions can be enabled or disabled when the program is started, and are disabled by default. Disabling assertions eliminates their performance penalty entirely. Once disabled, they are essentially equivalent to empty statements in semantics and performance. See Enabling and Disabling Assertions for more information.
The addition of the assert keyword to the Java programming
language has implications for existing code. See Compatibility With Existing Programs for
more information.
There are many situations where it is good to use assertions. This section covers some of them:
There are also a few situations where you should not use them:
Argument checking is typically part of the published specifications (or
contract) of a method, and these specifications must be obeyed whether
assertions are enabled or disabled. Another problem with using assertions for
argument checking is that erroneous arguments should result in an appropriate
runtime exception (such as IllegalArgumentException,
IndexOutOfBoundsException, or
NullPointerException). An assertion failure will not throw an
appropriate exception.
Because assertions may be disabled, programs must not assume that the
boolean expression contained in an assertion will be evaluated. Violating
this rule has dire consequences.
For example, suppose you wanted to remove all of the null elements
from a list names, and knew that the list contained one or more
nulls. It would be wrong to do this:
// Broken! - action is contained in assertion
assert names.remove(null);
The program would work fine when asserts were enabled, but would fail when
they were disabled, as it would no longer remove the null elements from the
list.
The correct idiom is to perform the action before the assertion and then
assert that the action succeeded:
// Fixed - action precedes assertion
boolean nullsRemoved = names.remove(null);
assert nullsRemoved; // Runs whether or not asserts are enabled
As a rule, the expressions contained in assertions should be free of
side effects: evaluating the expression should not affect any state
that is visible after the evaluation is complete. One exception to this rule
is that assertions can modify state that is used only from within other
assertions. An idiom that makes use of this
exception is presented later in this document.
Before assertions were available, many programmers used comments to
indicate their assumptions concerning a program's behavior. For example,
you might have written something like this to explain your assumption
about an else clause in a multiway if-statement:
if (i % 3 == 0) {
...
} else if (i % 3 == 1) {
...
} else { // We know (i % 3 == 2)
...
}
You should now use an assertion whenever you would have written a comment that asserts an invariant. For example, you should rewrite the previous if-statement like this:
if (i % 3 == 0) {
...
} else if (i % 3 == 1) {
...
} else {
assert i % 3 == 2 : i;
...
}
Note, incidentally, that the assertion in the above example may fail if
i is negative, as the % operator is not a true
modulus operator, but computes the remainder, which may be
negative.
Another good candidate for an assertion is a switch statement
with no default case. The absence of a default case
typically indicates that a programmer believes that one of the cases will
always be executed. The assumption that a particular variable will have one
of a small number of values is an invariant that should be checked with an
assertion. For example, suppose the following switch statement
appears in a program that handles playing cards:
switch(suit) {
case Suit.CLUBS:
...
break;
case Suit.DIAMONDS:
...
break;
case Suit.HEARTS:
...
break;
case Suit.SPADES:
...
}
It probably indicates an assumption that the suit variable
will have one of only four values. To test this assumption, you
should add the following default case:
default:
assert false : suit;
If the suit variable takes on another value and assertions are
enabled, the assert will fail and an AssertionError will be
thrown.
An acceptable alternative is:
default:
throw new AssertionError(suit);
This alternative offers protection even if assertions are disabled, but
the extra protection adds no cost: the throw statement won't
execute unless the program has failed. Moreover, the alternative
is legal under some circumstances where the assert statement is
not. If the enclosing method returns a value, each case in the
switch statement contains a return statement, and no
return statement follows the switch statement, then
it would cause a syntax error to add a default case with an assertion.
(The method would return without a value if no case matched and assertions were
disabled.)
The previous example not only tests an invariant, it also checks an
assumption about the application's flow of control. The author of the
original switch statement probably assumed not only that the
suit variable would always have one of four values, but also that
one of the four cases would always be executed. It points out another general
area where you should use assertions: place an assertion at any location
you assume will not be reached. The assertions statement to use is:
assert false;
For example, suppose you have a method that looks like this:
void foo() {
for (...) {
if (...)
return;
}
// Execution should never reach this point!!!
}
Replace the final comment so that the code now reads:
void foo() {
for (...) {
if (...)
return;
}
assert false; // Execution should never reach this point!
}
AssertionError.
Preconditions, Postconditions, and Class
Invariants
While the assert construct is not a full-blown design-by-contract facility, it can help support an informal design-by-contract style of programming. This section shows you how to use asserts for:
By convention, preconditions on public methods are enforced by explicit checks that throw particular, specified exceptions. For example:
/**
* Sets the refresh rate.
*
* @param rate refresh rate, in frames per second.
* @throws IllegalArgumentException if rate <= 0 or
* rate > MAX_REFRESH_RATE.
*/
public void setRefreshRate(int rate) {
// Enforce specified precondition in public method
if (rate <= 0 || rate > MAX_REFRESH_RATE)
throw new IllegalArgumentException("Illegal rate: " + rate);
setRefreshInterval(1000/rate);
}
This convention is unaffected by the addition of the assert
construct. Do not use assertions to check the parameters of a public
method. An assert is inappropriate because the method guarantees that it
will always enforce the argument checks. It must check its arguments whether
or not assertions are enabled. Further, the assert construct
does not throw an exception of the specified type. It can throw only an
AssertionError.
You can, however, use an assertion to test a nonpublic method's precondition that you believe will be true no matter what a client does with the class. For example, an assertion is appropriate in the following "helper method" that is invoked by the previous method:
/**
* Sets the refresh interval (which must correspond to a legal frame rate).
*
* @param interval refresh interval in milliseconds.
*/
private void setRefreshInterval(int interval) {
// Confirm adherence to precondition in nonpublic method
assert interval > 0 && interval <= 1000/MAX_REFRESH_RATE : interval;
... // Set the refresh interval
}
Note, the above assertion will fail if MAX_REFRESH_RATE is
greater than 1000 and the client selects a refresh rate greater than
1000. This would, in fact, indicate a bug in the library!
Classes designed for multithreaded use often have non-public methods with preconditions relating to whether or not some lock is held. For example, it is not uncommon to see something like this:
private Object[] a;
public synchronized int find(Object key) {
return find(key, a, 0, a.length);
}
// Recursive helper method - always called with a lock on this object
private int find(Object key, Object[] arr, int start, int len) {
...
}
A static method called holdsLock has been added to the
Thread class to test whether the current thread holds the lock on
a specified object. This method can be used in combination with an
assert statement to supplement a comment describing a lock-status
precondition, as shown in the following example:
// Recursive helper method - always called with a lock on this.
private int find(Object key, Object[] arr, int start, int len) {
assert Thread.holdsLock(this); // lock-status assertion
...
}
Note that it is also possible to write a lock-status assertion asserting that a given lock isn't held.
You can test postcondition with assertions in both public and nonpublic
methods. For example, the following public method uses an assert
statement to check a post condition:
/**
* Returns a BigInteger whose value is (this-1 mod m).
*
* @param m the modulus.
* @return this-1 mod m.
* @throws ArithmeticException m <= 0, or this BigInteger
* has no multiplicative inverse mod m (that is, this BigInteger
* is not relatively prime to m).
*/
public BigInteger modInverse(BigInteger m) {
if (m.signum <= 0)
throw new ArithmeticException("Modulus not positive: " + m);
... // Do the computation
assert this.multiply(result).mod(m).equals(ONE) : this;
return result;
}
Occasionally it is necessary to save some data prior to performing a
computation in order to check a postcondition. You can do this with two
assert statements and a simple inner class that saves the state
of one or more variables so they can be checked (or rechecked) after the
computation. For example, suppose you have a piece of code that looks like
this:
void foo(int[] array) {
// Manipulate array
...
// At this point, array will contain exactly the ints that it did
// prior to manipulation, in the same order.
}
Here is how you could modify the above method to turn the textual assertion of a postcondition into a functional one:
void foo(final int[] array) {
// Inner class that saves state and performs final consistency check
class DataCopy {
private int[] arrayCopy;
DataCopy() { arrayCopy = (int[]) array.clone(); }
boolean isConsistent() { return Arrays.equals(array, arrayCopy); }
}
DataCopy copy = null;
// Always succeeds; has side effect of saving a copy of array
assert ((copy = new DataCopy()) != null);
... // Manipulate array
// Ensure array has same ints in same order as before manipulation.
assert copy.isConsistent();
}
You can easily generalize this idiom to save more than one data field, and to test arbitrarily complex assertions concerning pre-computation and post-computation values.
You might be tempted to replace the first assert statement (which is executed solely for its side-effect) by the following, more expressive statement:
copy = new DataCopy();
Don't make this replacement. The statement above would copy the array whether or not asserts were enabled, violating the principle that assertions should have no cost when disabled.
The assertion mechanism does not enforce any particular style for checking invariants. It is sometimes convenient, though, to combine the expressions that check required constraints into a single internal method that can be called by assertions. Continuing the balanced tree example, it might be appropriate to implement a private method that checked that the tree was indeed balanced as per the dictates of the data structure:
// Returns true if this tree is properly balanced
private boolean balanced() {
...
}
Because this method checks a constraint that should be true before and after any method completes, each public method and constructor should contain the following line immediately prior to its return:
assert balanced();
It is generally unnecessary to place similar checks at the head of each public method unless the data structure is implemented by native methods. In this case, it is possible that a memory corruption bug could corrupt a "native peer" data structure in between method invocations. A failure of the assertion at the head of such a method would indicate that such memory corruption had occurred. Similarly, it may be advisable to include class invariant checks at the heads of methods in classes whose state is modifiable by other classes. (Better yet, design classes so that their state is not directly visible to other classes!)
Advanced Uses
The following sections discuss topics that apply only to
resource-constrained devices and to systems where asserts must not be
disabled in the field.
If you have no interest in these topics, skip to the next
section, "Compiling Files that Use
Assertions".
Removing all Trace of Assertions from Class
Files
Programmers developing applications for resource-constrained devices may wish to strip assertions out of class files entirely. While this makes it impossible to enable assertions in the field, it also reduces class file size, possibly leading to improved class loading performance. In the absence of a high quality JIT, it could lead to decreased footprint and improved runtime performance.
The assertion facility offers no direct support for stripping assertions out of class files. The assert statement may, however, be used in conjunction with the "conditional compilation" idiom described in JLS 14.20, enabling the compiler to eliminate all traces of these asserts from the class files that it generates:
static final boolean asserts = ... ; // false to eliminate asserts
if (asserts)
assert <expr> ;
Requiring that Assertions are Enabled
Programmers of certain critical systems might wish to ensure that assertions are not disabled in the field. The following static initialization idiom prevents a class from being initialized if its assertions have been disabled:
static {
boolean assertsEnabled = false;
assert assertsEnabled = true; // Intentional side effect!!!
if (!assertsEnabled)
throw new RuntimeException("Asserts must be enabled!!!");
}
Put this static-initializer at the top of your class.
In order for the javac compiler to accept code containing
assertions, you must use the -source 1.4 command-line option
as in this example:
javac -source 1.4 MyClass.javaThis flag is necessary so as not to cause source compatibility problems.
By default, assertions are disabled at runtime. Two command-line switches allow you to selectively enable or disable assertions.
To enable assertions at various granularities, use the
-enableassertions-ea-disableassertions-da
... ...For example, the following command runs a program, BatTutor,
with assertions enabled in only package com.wombat.fruitbat and
its subpackages:
java -ea:com.wombat.fruitbat... BatTutor
If a single command line contains multiple instances of these switches,
they are processed in order before loading any classes. For example, the
following command runs the BatTutor program with assertions
enabled in package com.wombat.fruitbat but disabled in class
com.wombat.fruitbat.Brickbat:
java -ea:com.wombat.fruitbat... -da:com.wombat.fruitbat.Brickbat BatTutor
The above switches apply to all class loaders. With one exception, they also apply to system classes (which do not have an explicit class loader). The exception concerns the switches with no arguments, which (as indicated above) do not apply to system classes. This behavior makes it easy to enable asserts in all classes except for system classes, which is commonly desirable.
To enable assertions in all system classes, use a different switch:
-enablesystemassertions-esa.-disablesystemassertions-dsa.
For example, the following command runs the BatTutor program
with assertions enabled in system classes, as well as in the
com.wombat.fruitbat package and its subpackages:
java -esa -ea:com.wombat.fruitbat...
The assertion status of a class (enabled or disabled) is set at the time it is initialized, and does not change. There is, however, one corner case that demands special treatment. It is possible, though generally not desirable, to execute methods or constructors prior to initialization. This can happen when a class hierarchy contains a circularity in its static initialization.
If an assert statement executes before its class is initialized, the execution must behave as if assertions were enabled in the class. This topic is discussed in detail in the assertions specification.
The addition of the assert keyword to the Java programming
language does not cause any problems with preexisting binaries
(.class files). If you try to compile an application that uses
assert as an identifier, however, you will receive
a warning or error message. In order to ease the transition from a world where
assert is a legal identifier to one where it isn't, the compiler
supports two modes of operation in this release:
assert as an identifier,
but issues warnings. In this mode, programs are not permitted to use
the assert statement.assert as an identifier. In this mode,
programs are permitted to use the assert statement.Unless you specifically request source mode 1.4 with the -source
1.4 flag, the compiler operates in source mode 1.3. If you
forget to use this this flag, programs that use the new assert
statement will not compile. Having the compiler use the old semantics as
its default behavior (that is, allowing assert to be used as an
identifier) was done for maximal source compatibility. Source mode 1.3 is
likely to be phased out over time.
Here is a collection of frequently asked questions concerning the design of the assertion facility.
AssertionError Class
Although ad hoc implementations are possible, they are of necessity
either ugly (requiring an if statement for each assertion) or
inefficient (evaluating the condition even if assertions are disabled).
Further, each ad hoc implementation has its own means of enabling and
disabling assertions, which lessens the utility of these implementations,
especially for debugging in the field. As a result of these shortcomings,
assertions have never become a part of the culture among engineers using the
Java programming language. Adding assertion support to the platform stands a
good chance of rectifying this situation.
We recognize that a language change is a serious effort, not to be
undertaken lightly. The library approach was considered. It was, however,
deemed essential that the runtime cost of assertions be negligible if they
are disabled. In order to achieve this with a library, the programmer is
forced to hard-code each assertion as an if statement. Many
programmers would not do this. Either they would omit the if statement and
performance would suffer, or they would ignore the facility entirely. Note
also that assertions were contained in James Gosling's original
specification for the Java programming language. Assertions were removed
from the Oak specification because time constraints prevented a satisfactory
design and implementation.
We considered providing such a facility, but were unable to convince ourselves that it is possible to graft it onto the Java programming language without massive changes to the Java platform libraries, and massive inconsistencies between old and new libraries. Further, we were not convinced that such a facility would preserve the simplicity that is the hallmark of the Java programming language. On balance, we came to the conclusion that a simple boolean assertion facility was a fairly straight-forward solution and far less risky. It's worth noting that adding a boolean assertion facility to the language doesn't preclude adding a full-fledged design-by-contract facility at some time in the future.
The simple assertion facility does enable a limited form of
design-by-contract style programming. The
assert statement is appropriate for nonpublic precondition,
postcondition and class invariant checking. Public precondition checking
should still be performed by checks inside methods that result in
particular, documented exceptions, such as
IllegalArgumentException and IllegalStateException.
Providing such a construct would encourage programmers to put complex assertions inline, when they are better relegated to separate methods.
assert as an identifier?
Yes, for source files. (Binaries for classes that use assert
as an identifier will continue to work fine.) To ease the transition, we
implemented a strategy whereby developers can
continue using assert as an identifier during a transitional
period.
Yes. Class files will contain calls to the new ClassLoader and Class methods, such as desiredAssertionStatus. If a class file containing calls to these methods is run against an older JRE (whose ClassLoader class doesn't define the methods), the program will fail at run time, throwing a NoSuchMethodError. It is generally the case that programs using new facilities are not compatible with older releases.
There is no compelling reason to restrict the type of this expression.
Allowing arbitrary types provides convenience for developers who, for
example, want to associate a unique integer code with each
assertion. Further, it makes this expression feel like the argument of
System.out.println(...), which is seen as desirable.
AssertionError is generated by an assert
statement in which Expression2
is absent, why isn't the program text of the asserted condition used as the
detail message (for example, height <
maxHeight"While doing so might improve out-of-the-box usefulness of assertions in
some cases, the benefit doesn't justify the cost of adding all those string
constants to .class files and runtime images.
AssertionError allow access to the object
that generated it? Similarly, why not pass an arbitrary object from the
assertion to the AssertionError constructor in place of a
detail message?
Access to these objects would encourage programmers to attempt to recover from assertion failures, which defeats the purpose of the facility.
getFile,
getline, getMethod) on
AssertionError?
This facility is best provided on Throwable, so it may be
used for all throwables, not just just assertion errors. We
enhanced Throwable with the getStackTrace
method to provide this functionality.
AssertionError a subclass of Error
rather than RuntimeException?
This issue was controversial. The expert group discussed it at at length,
and came to the conclusion that Error was more appropriate to
discourage programmers from attempting to recover from assertion
failures. It is, in general, difficult or impossible to localize the source
of an assertion failure. Such a failure indicates that the program is
operating "outside of known space," and attempts to continue execution are
likely to be harmful. Further, convention dictates that methods specify
most runtime exceptions they may throw (with @throws doc
comments). It makes little sense to include in a method's specification the
circumstances under which it may generate an assertion failure. Such
information may be regarded as an implementation detail, which can change
from implementation to implementation and release to release.
Enabling and Disabling
Assertions
It is a firm requirement that it be possible to enable assertions in the field, for enhanced serviceability. It would have been possible to also permit developers to eliminate assertions from object files at compile time. Assertions can contain side effects, though they should not, and such a flag could therefore alter the behavior of a program in significant ways. It is viewed as good thing that there is only one semantics associated with each valid Java program. Also, we want to encourage users to leave asserts in object files so they can be enabled in the field. Finally, the spec demands that assertions behave as if enabled when a class runs before it is initialized. It would be impossible to offer these semantics if assertions were stripped from the class file. Note, however, that the standard "conditional compilation idiom" described in JLS 14.20 can be used to achieve this effect for developers who really want it.
Hierarchical control is useful, as programmers really do use package hierarchies to organize their code. For example, package-tree semantics allow assertions to be enabled or disabled in all of Swing at one time.
setClassAssertionStatus return a
boolean instead of throwing an exception if it is invoked when
it's too late to set the assertion status (that is, if the named class has
already been initialized)?
No action (other than perhaps a warning message) is necessary or desirable if it's too late to set the assertion status. An exception seems unduly heavyweight.
setDefaultAssertionStatus and
setAssertionStatus?
Clarity in method naming is for the greater good. Overloading tends to cause confusion.
It's not clear that there would be any use for the resulting method. The method isn't designed for application programmer use, and it seems inadvisable to make it slower and more complex than necessary.
RuntimePermission to prevent applets
from enabling/disabling assertions?
While applets have no reason to call any of the ClassLoader
methods for modifying assertion status, allowing them to do so seems
harmless. At worst, an applet can mount a weak denial-of-service attack by
enabling assertions in classes that have yet to be initialized. Moreover,
applets can only affect the assert status of classes that are to be loaded
by class loaders that the applets can access. There already exists a
RuntimePermission to prevent untrusted code from gaining access
to class loaders (getClassLoader).
Such a construct would encourage people to inline complex assertion code, which we view as a bad thing. Further, it is straightforward to query the assert status atop the current API, if you feel you must:
boolean assertsEnabled = false; assert assertsEnabled = true; // Intentional side-effect!!! // Now assertsEnabled is set to the correct value
Few programmers are aware of the fact that a class's constructors
and methods can run prior to its initialization. When this happens, it
is quite likely that the class's invariants have not yet been established,
which can cause serious and subtle bugs. Any assertion that executes in
this state is likely to fail, alerting the programmer to the problem. Thus,
it is generally helpful to the programmer to execute all assertions encountered
while in this state.
|
Copyright © 2002 Sun Microsystems, Inc. All Rights Reserved. |
