Dave Bristor's Sun Weblog

Notable changes in the Java SE 6 Core Libraries

This blog summarizes notable changes to the core libraries in Java SE 6. Think of Java's "core libraries" as lang, io, util, and so on: the stuff of which all Java programs are comprised. For good measure, throw in jar, zip, regular expressions, concurrency, and reflection too.

There's lots more information about Java SE 6's features and compatibility with earlier releases. The platform overview describes the entire Java platform briefly, with links to more detailed guides. And of course, you can download the Java SE 6 release itself at java.sun.com.

Changes in I/O

Reading passwords without echoing their text

New to Java SE 6 is the ability to read text from a terminal without having it echo on the screen, through java.io.Console:

import java.io.Console;
...
    char[] password = System.console().readPassword("password: ");
    // use password
    ...
    // clear password
    java.util.Arrays.fill(password, '\0');

A few things are worth noting:

• System.console() will return null if either standard input or standard output is not a terminal.
• readPassword() returns an array, so that you can null it out as shown.
• readPassword() uses the character set of the terminal device, which might not be that of Charset.defaultCharset().

Display of the Critical Message Box

Chances are you've seen a dialog on Microsoft Windows presenting a critical message such as

        Abort, Retry, Ignore

In the java.io.File class, any method which accesses removable media such as a floppy drive could cause a dialog with that message to appear. This could be a real problem if you're accessing the system remotely! So we've changed the functionality here, so that by default for the JVM (including native code) and all processes it launches, the dialog doesn't appear. In a case like this:

    new File("A:\\tmp.txt").exists();

if there's no floppy in the A: drive, exists() returns false.

Interruptible I/O

What does this do?

    public class Interrupt {
        public static void main(String[] args) {
            Thread.currentThread().interrupt();
            System.out.format("Hello?");
        }
    }

Brownie points if you answered "it depends". On Microsoft Windows and Linux operating systems, it will print Hello?. On the Solaris Operating System, it will hang. That's because I/O is truly interruptible on Solaris only: once the interrupt flag is set, you won't get anything written to stdout nor stderr. Note that you can turn it off, by running java with -XX:-UseVMInterruptibleIO. This will be the default behavior in future releases of the JDK, for better cross-platform compatibility.

Finding out about disk space

We've added some new method to java.io.File so you can find out how much disk space is available:

• getTotalSpace()
• getFreeSpace()
• getUsableSpace()

Note that the returned values are only estimates at the time of the call. They don't take into account any background I/O activity (by other processes, etc.)

A few notes are in order

• getUsableSpace() takes quotas into account only on Microsoft Windows.
• Each returns the result as a long. The return values don't imply anything about access controls, i.e. the space might not be writable.
• A return value of zero could mean that there's no space, or it could mean that the partition doesn't exist.

Changing access modes

In Java SE 6, java.io.File provides methods to change access of files:

• setExecutable()
• setReadable()
• setWritable()

Each returns true if the operation succeeded.

Microsoft Windows devices are no longer considered files

Prior to Java SE 6, File.isFile() returned true for Microsoft Windows devices such as CON, NUL, and LPT. Now it returns false.

File.toURL is now deprecated

Creating a URL from some files can result in an illegal URL. Consider a file named "%-2": really, those characters should be escaped. The correct way to do this is to use toURI() as a go-between:

    URL u = File.toURI("%-2").toURL()

Long path names on Microsoft Windows

Java SE 6 supports long path names on Microsoft Windows. Each path name element is limited to 260 characters, which is the platform limitation.

Other noteworthy changes to java.io.File

• 6348207: File.length() reports a length of 0 for special files such as pagefile.sys.
• 6198547: File.createNewFile() on an existing directory incorrectly throws IOException. Now it returns false on all platforms.
• 6395581: File.listFiles() is unable to read nfs-mounted directory. Now it returns the correct results.
• 4809375: File.deleteOnExit() should be implemented with shutdown hooks. Prior to Java SE 6, each invocation of deleteOnExit() added the file to a list: in particular, a file could be added multiple times and during JVM shutdown, deletion was attempted for each File in the list. In Java SE 6, instead of a list, File objects are added to a LinkedHashSet, guaranteeing that each is deleted only once.

Improved selector scalability on Linux

Linux 2.6 kernel provides epoll(4), which is more scalable than poll(2). The JDK detects the platform on which it is running, and so for Linux 2.6 or later kernels, will use epoll(4) in its SelectorProvider. This should scale well when there are thousands of selectable channels registered with a selector.

Multiple locks on the same file

What does this code do?

    public class FileLock {
        public static void main(String[] args) throws Throwable {
            new FileOutputStream("foo").getChannel().lock();
            new FileOutputStream("foo").getChannel().lock();
        }
    }

Prior to Java SE 6, this code would deadlock on Microsoft Windows, and return two locks on Linux and Solaris. Now, on all of those platforms, the second lock attempt throws an OverlappingFileLockException, as should have been the case all along since the behavior of lock() is specified that way.

Changes in collections

Performance improved in java.util.HashMap

We were able to improve the performance hashing function used by java.util.HashMap by about 5%. A side effect is that the order of the results returned by iterating over values() has changed from the previous release. Of course, as per the specification of iterator(), clients should not depend on the order from an iterator!

Performance improved in copying arrays

We've added new methods to java.util.Arrays to copy arrays and sub-arrays:

• copyOf
• copyOfRange

(The links above are to the int methods, but there are methods for each primitve type as well as type-parameterized methods for objects.) These methods efficiently resize, truncate, or copy subarrays of all types. They simplify your code a bit too. Before Java SE 6, you might write:

    int[] newarray = new int[oldArray.length];
    System.arraycopy(oldArray, 0, newArray, 0, oldArray.length);

Now you can instead write this:

    int[] newarray = Arrays.copyOf(oldArray, oldArray.length);

Not only is the code slightly simpler, but there's a performance benefit: The old way created an array and initialized its values to zero, then overwrite those zeroes with values from oldArray. By providing a means to do this in one step, we avoid the zero initialization. Finally, the new methods will truncate or pad the new array as required.

Bidirectional navigable collections

We've added some new interfaces to enhance Maps and Sets:

• NavigableMap
• NavigableSet
• ConcurrentNavigableMap

These extend their related Sorted classes with navigation methods reporting closest matches (e.g. higher) and allow for traversal in either ascending or descending order. We've retrofitted the related implementations TreeMap, TreeSet, and added the new concurrent classes ConcurrentSkipListMap, and ConcurrentSkipListSet.

Double-Ended Queues

To the Queue interface in Java SE 5, we added the Deque and BlockingDeque interfaces. We updated LinkedList to implement Deque since it had everything needed to do so. We also added new classes ArrayDeque, which is an efficient implementation of Deque, and the concurrent implementation LinkedBlockingDeque.

Changes in jar and zip

Timestamps on files extracted by jar

Prior to Java SE 6, the date and time of files extracted by the jar command were the current time. Other de/compression tools use the time noted in the archive file itself. So for Java SE 6, we changed jar to conform with other tools

if the old behavior is needed, run the JVM with sun.tools.jar.useExtractionTime=true

Number of open ZIP files

We removed a limitation on the number of concurrently open ZIP files on Microsoft Windows. The maximum used to be 2036, but now it's whatever the platform will support.

Number of entries in a ZIP file

The ZIP file format has a 2-byte field to record the number of entries in the file, artificially imposing a 64k limit. Our implementation now ignores that field, and instead just counts the entries. Prior to this change, you could count the entries with ZipInputStream or ZipFile , but get differing results if there were more than 64k entries in the file.

Long Zip File Names on Microsoft Windows

We now use the same support for long file names in ZIP files as we use in the java.io package.

Other changes

Support for plugable service providers

We've added java.util.ServiceLoader as the standard way for clients to load plugable code. It maintains a (clearable) cache of loaded services.

Array syntax checking

Final quiz: what does this do?

    public class LoadArray {
        public static void main(String[] args) throws Throwable {
            String name = (new String[0]).getClass().getName();
            LoadArray.class.getClassLoader().loadClass(name);
        }
    }

The key observations are that name is the name of an array class, and that we're using the single-argument overload of loadClass(). In earlier releases, this was allowed, but in Java SE 6, throws a ClassNotFoundException. It really is a bug: loadClass() only works for binary names, and the Java Language Specification, Third Edition, Chapter 13 defines binary names for classes and interfaces, not arrays.

Notable changes in Runtime.exec()

We had a number of bugs all related to attempts to predict whether an exec would succeed. In Java SE 6, we instead exec whatever is given, and communicate over a pipe with the exec'd process. If it fails for any reason, such as in these bugs, we can detect the failure:

• 4052517: Runtime.exec won't execute programs belonging to other groups on Unix
• 4811767: Runtime.exec should throw IOException when workdir does not exist (Unix)
• 5033302: Can't execute Solaris NFS programs with uid>64k on Linux-amd64

Posted at 09:29AM Dec 11, 2006 by daveb in Sun  |  Comments[3]