Java try/catch/finally best practices while acquiring/closing resources

Question

While working on a school project, I wrote the following code:

FileOutputStream fos;
ObjectOutputStream oos;
try {
    fos = new FileOutputStream(file);
    oos = new ObjectOutputStream(fos);

    oos.writeObject(shapes);
} catch (FileNotFoundException ex) {
    // complain to user
} catch (IOException ex) {
    // notify user
} finally {
    if (oos != null) oos.close();
    if (fos != null) fos.close();
}

The problem is that Netbeans is telling me the resource.close() lines throw an IOException and therefore must either be caught or declared. It also is complaining that oos and fos might not yet be initialized (despite the null checks).

This seems a little strange, seeing as how the whole point is to stop the IOException right there.

My knee-jerk fix is to do this:

} finally {
    try {
        if (oos != null) oos.close();
        if (fos != null) fos.close();
    } catch (IOException ex) { }
}

But deep down this bothers me and feels dirty.

I come from a C# background, where I would simply take advantage of a using block, so I am unsure of what the "right" way is to handle this.

What is the right way to handle this problem?

Answer 1

Note that the following is only applicable for Java 6 and earlier. For Java 7 and later, you should switch to using try-with-resources ... as described in other answers.

If you are trying to catch and report all exceptions at source (in Java 6 or earlier), a better solution is this:

ObjectOutputStream oos = null;
try {
   oos = new ObjectOutputStream(new FileOutputStream(file));
   oos.writeObject(shapes);
   oos.flush();
} catch (FileNotFoundException ex) {
    // complain to user
} catch (IOException ex) {
    // notify user
} finally {
    if (oos != null) {
        try {
            oos.close();
        } catch (IOException ex) {
            // ignore ... any significant errors should already have been
            // reported via an IOException from the final flush.
        }
    }
}

Notes:

• The standard Java wrapper streams, readers and writers all propagate close and flush to their wrapped streams, etc. So you only need to close or flush the outermost wrapper.
• The purpose of flushing explicitly at the end of the try block is so that the (real) handler for IOException gets to see any write failures¹.
• When you do a close or flush on an output stream, there is a "once in a blue moon" chance that an exception will be thrown due to disc errors or file system full. You should not squash this exception!.

If you often have to "close a possibly null stream ignoring IOExceptions", then you could write yourself a helper method like this:

public void closeQuietly(Closeable closeable) {
    if (closeable != null) {
        try {
            closeable.close();
        } catch (IOException ex) {
            // ignore
        }
    }
}

then you can replace the previous finally block with:

} finally {
    closeQuietly(oos);
}

Another answer points out that a closeQuietly method is already available in an Apache Commons library ... if you don't mind adding a dependency to your project for a 10 line method.

But be careful that you only use closeQuietly on streams where IO exceptions really are irrelevant.

UPDATE : closeQuietly is deprecated in version 2.6 of the Apache Commons API. Java 7+ try-with-resources makes it redundant.

---

On the issue of flush() versus close() that people were asking about in comments:

• The standard "filter" and "buffered" output streams and writers have an API contract that states that close() causes all buffered output to be flushed. You should find that all other (standard) output classes that do output buffering will behave the same way. So, for a standard class it is redundant to call flush() immediately before close().

• For custom and 3rd-party classes, you need to investigate (e.g. read the javadoc, look at the code), but any close() method that doesn't flush buffered data is arguably broken.

• Finally, there is the issue of what flush() actually does. What the javadoc says is this (for OutputStream ...)

  If the intended destination of this stream is an abstraction provided by the underlying operating system, for example a file, then flushing the stream guarantees only that bytes previously written to the stream are passed to the operating system for writing; it does not guarantee that they are actually written to a physical device such as a disk drive.

  So ... if you hope / imagine that calling flush() guarantees that your data will persist, you are wrong! (If you need to do that kind of thing, look at the FileChannel.force method ...)

Answer 2

Current best practice for try/catch/finally involving objects that are closeable (e.g. Files) is to use Java 7's try-with-resource statement, e.g.:

try (FileReader reader = new FileReader("ex.txt")) {
    System.out.println((char)reader.read());
} catch (IOException ioe) {
    ioe.printStackTrace();
}

In this case, the FileReader is automatically closed at the end of the try statement, without the need to close it in an explicit finally block. There are a few examples here:

http://ppkwok.blogspot.com/2012/11/java-cafe-2-try-with-resources.html

The official Java description is at:

http://docs.oracle.com/javase/7/docs/technotes/guides/language/try-with-resources.html

Answer 3

Java 7 will add Automatic Resource Management blocks. They are very similar to C#'s using.

Josh Bloch wrote the technical proposal, which I highly recommend reading. Not just because it will give you a leg up on an upcoming Java 7 language feature, but because the specification motivates the need for such a construct, and in doing so, illustrates how to write correct code even in the absence of ARM.

Here's an example of the Asker's code, translated into ARM form:

try (FileOutputStream fos = new FileOutputStream(file);
        ObjectOutputStream oos = new ObjectOutputStream(fos)) 
{
    oos.writeObject(shapes);
}
catch (FileNotFoundException ex) 
{
    // handle the file not being found
}
catch (IOException ex) 
{
    // handle some I/O problem
}

Answer 4

I usually have small class IOUtil with method such as:

public static void close(Closeable c) {
    if (c != null) {
        try {
            c.close();
        }
        catch (IOException e) {
            // ignore or log
        }
    }
}

Answer 5

How about this guys? No null check, no surprise. Everything is cleaned upon exit.

try {
    final FileOutputStream fos = new FileOutputStream(file);
    try {
        final ObjectOutputStream oos = new ObjectOutputStream(fos);
        try {
            oos.writeObject(shapes);
            oos.flush();
        }
        catch(IOException ioe) {
            // notify user of important exception
        }
        finally {
            oos.close();
        }
    }
    finally {
        fos.close();
    }
}
catch (FileNotFoundException ex) {
    // complain to user
}
catch (IOException ex) {
    // notify user
}

Answer 6

Unfortunately there is no language level support. But there are lot of libraries out there that makes this simple. Check commons-io library. Or modern google-guava @ http://guava-libraries.googlecode.com/svn/trunk/javadoc/index.html

Answer 7

You're doing it right. It bothers the crap out of me too. You should initialize those streams to null explicitly - it's common convention. All you can do is join the club and want for using.

Answer 8

Not a direct answer to your point but it is an unfortunate fact that because finally and catch are both associated with try people think that they belong together. The best design for try blocks is to either have a catch or a finally but not both.

In this case your comments hint that something is wrong. Why, in a method dealing with file IO, are we complaining about anything to the user. We might be running deep on a server somewhere with nary a user in sight.

So, the code you present above should have a finally to fail gracefully when things go wrong. It lacks the capacity to deal intelligently with errors however, so your catch belongs somewhere higher up on the call chain.