You are here: Home ยป Blog

Examining nested stack traces

Posted on Tuesday, April 21 2015 at 09:06 | Category: Java | 0 Comment(s)

I often read Java Stack traces bottom up when I examine them for the first time - simply because this is the code path which was executed when the exception occurred. However, it can happen that the last line of a stack trace shows something like ... 2 more - so, one might ask why can't the runtime just dump those missing lines, along with all the other stack trace elements? Real life server stack traces sometimes contain dozens of lines, it should not matter to print those additional lines, right? And often those lines contain just the information you require to see from where the problematic code which caused the exception was called ...

The thing is: those lines are actually in the stack trace. Lets consider this example:

Exception in thread "main" java.lang.RuntimeException: java.lang.RuntimeException: Exception thrown
	at com.example.TraceTest.doSomething(TraceTest.java:13)
	at com.example.TraceTest.run(TraceTest.java:6)
	at com.example.TraceTest.main(TraceTest.java:22)
Caused by: java.lang.RuntimeException: Exception thrown
	at com.example.TraceTest.throwAnException(TraceTest.java:18)
	at com.example.TraceTest.doSomething(TraceTest.java:11)
	... 2 more

As you can see, the last line reads ... 2 more, but it might be crucial for the further analysis to know from where the doSomething() method was called. In order to get this information, we need to look further up in the stack trace: There, we again find the doSomething() method at the top and see that it was called from the run() method. In other words, the initial entry point for the code flow is the last element of the first stack trace block - from there, we can follow to the next stack trace block to see where the exception was finally thrown:

 

The reason for this is that the original exception was wrapped as nested exception into another exception. The following is the code which was used for the test above:

package com.example;

public class TraceTest {
    
    public void run() {
        doSomething();
    }

    private void doSomething() {
        try {
            throwAnException();
        }catch(RuntimeException re) {
            throw new RuntimeException(re);
        }
    }

    private void throwAnException() {
        throw new RuntimeException("Exception thrown");
    }

    public static void main(String[] args) {
        new TraceTest().run();
    }
}

Real stack traces might also contain more than one nested exception, so it might be necessary to follow them more than once.

In any case, the stack trace still contains the whole code path from the entry point (usually main) to the place where the exception was thrown.

See also how to print the full stacktrace in java on StackOverflow.


Eclipse Luna crashes with KDE on Ubuntu 14.04 LTS (Trusty)

Posted on Monday, February 16 2015 at 20:47 | Category: Java | 1 Comment(s)

This is an issue I came across today, after I upgraded my system to Ubuntu 14.04 LTS (Trusty) some days ago and after I also upgraded to Eclipse Luna 4.4.1 (using JDK 8u31):

When importing an existing project into the workspace, the "Import Projects" dialog (where you can select the root directory) still opens, but then the JVM crashes with this assertion:

java: /build/buildd/gtk2-engines-oxygen-1.4.5/src/animations/oxygencomboboxdata.cpp:87: 
void Oxygen::ComboBoxData::setButton(GtkWidget*): Assertion `!_button._widget' failed.

After some searching, I found that this is a known KDE bug (there is also a corresponding Eclipse bug) and it has already been fixed in oxygen-gtk2-1.4.6. This version is already available in Ubuntu 15.04 (Vivid), but not yet in 14.04, so I crated a backport package - you can get it from my Personal Package Archive. After installing the upgraded package, the issue is gone (what remains is a message like Oxygen::WindowManager::wmButtonPress - warning: a button was already set for this combobox which is obvious when we look at the diff (have a look at the oxygencomboboxdata.cpp file).

A different workaround in case you do not want or can install the new oxygen-gtk version is to set a different GTK style in the KDE control center:

Open the KDE System settings and select "Application Appearance". In the dialog choose the GTK category and choose a different GTK style in the "Select a GTK2 Theme" dropdown, e.g. "Ambiance" as in the example below:


Some background on Generics: dumping a Map

Posted on Wednesday, December 17 2014 at 11:16 | Category: Java | 0 Comment(s)

In Java, dumping a Map to see what key/value mappings it contains is quite easy. Lets assume we have a factory method which returns a map, containing some String to Long mappings:

Map<String, Long> map = createMap();

The toString() method of the Collection classes usually provide a useful implementation, so that we can simply use

System.err.println(map);

to dump the map. If we want to have more control over the output format, e.g. one line per map entry, we can use a simple loop to iterate over the entries:

Set<Entry<String, Long>> entries = map.entrySet();
for (Entry<String, Long> entry : entries) {
    String key = entry.getKey();
    Long value = entry.getValue();
    System.err.printf("%s=%s\n", key, value);
}

When running this code with the map returned from the createMap() method, we get

Exception in thread "main" java.lang.ClassCastException: com.example.Key cannot be cast to java.lang.String
	at com.example.MapSample.run(MapSample.java:25)
	at com.example.MapSample.main(MapSample.java:54)

Ehm... wait a moment ... why do we get this exception in the line which executes String key = entry.getKey();? We did not get any compile time errors or even warnings, and entry.getKey() is declared to return a String due to the definition of entry as Entry<String, Long>. Also, the Map is declared as Map<String, Long> - so each key element in the map must be a String, right?

No.

Lets examine how the map is created in the createMap() factory method:

private Map<String, Long> createMap() {
    Map result = new HashMap();
    result.put("one",  1L);
    result.put(new Key(), 3L);
    result.put("two",  2L);
    result.put("five",  5L);
    result.put("four",  4L);
    return result;
}

Obviously, we are allowed to use different types than String as a key, an instance of the Key class in this case. This is possible since Map and HashMap are used as raw types here, instead of parameterized types. The Java compiler will issue some warnings, but will still compile the method as such (and even the warnings can be switched off with an annotation like @SuppressWarnings({ "unchecked", "rawtypes" })).

Remember that Generics (paremeterized types) are syntactic sugar only - internally, during runtime, everything is raw types, with the necessary casts automatically applied where necessary. These casts succeed when the parameterized types are used throughout the application. If this was properly done in the createMap() method, the compiler would not let us use Key as a key value. However, due to the usage of Map as raw type, we can. And then, during runtime, the casts mentioned before will fail horribly. In the example above, entry.getKey() still returns an Object reference, but due to the type parameters the compiler automatically casts the result to a String. You can verify this in the byte code which contains a corresponding checkcast instruction after entry.getKey() has been invoked.

The general rule is: do not use raw types. Especially with large legacy code bases, there are sometimes occasions where Collections are still created without the proper usage of type parameters (but later returned through parameterized types). Thus, the learning is: Even if you get a reference to a properly parameterized Collection type, do not be surprised if there are other types stored in it when you iterate over the Collection values or keys.

Back to our dumping method, we can simply solve the issue by not forcing any implicit casts:

Set<Entry<String, Long>> entries = map.entrySet();
for (Entry<String, Long> entry : entries) {
    Object key = entry.getKey();
    Object value = entry.getValue();
    System.err.printf("%s=%s\n", key, value);
}

When run, this will produce the expected output:

four=4
one=1
com.example.Key@15db9742=3
two=2
five=5

Incompatible Java 8 changes in the Collections runtime classes

Posted on Tuesday, December 09 2014 at 13:15 | Category: Java | 0 Comment(s)

Suppose that we want to implement a HashMap (let's call it OrderedHashMap) where the values can also retrieved by their index, where the index refers to the order in which the key/value pairs have been inserted. The following unit test shows how such a class shall be used:

@Test
public void testPutGet() {
    OrderedHashMap<String, String> ohm = new OrderedHashMap<>();
    ohm.put("Z", "LetterZ");    // 0
    ohm.put("A", "LetterA");    // 1
    ohm.put("D", "LetterD");    // 2
    ohm.put("K", "LetterK");    // 3

    assertEquals(ohm.get(0), "LetterZ");
    assertEquals(ohm.get(3), "LetterK");
}

The straight forward implementation of the class OrderedHashMap is to inherit from one of the existing Collection classes and simply implement the additional functionality, e.g. like

public class OrderedHashMap<K, V> extends LinkedHashMap<K, V> {
    private static final long serialVersionUID = 6354582314971513369L;

    private List<V> items = new ArrayList<>();

    public V get(int index) {
        return items.get(index);
    }

    @Override
    public V put(K key, V value) {
        items.add(value);
        return super.put(key, value);
    }
}

With this code, the above unit test succeeds. Lets add an additional unit tests for one of the other methods inherited from LinkedHashMap, putAll():

@Test
public void testPutAll() {
    OrderedHashMap<String, String> ohm = new OrderedHashMap<>();
    ohm.put("Z", "LetterZ");    // 0
    ohm.put("A", "LetterA");    // 1
    ohm.put("D", "LetterD");    // 2
    ohm.put("K", "LetterK");    // 3

    OrderedHashMap<String, String> ohm2 = new OrderedHashMap<&ht;();
    ohm2.putAll(ohm);

    assertEquals(ohm2.get(0), "LetterZ");
    assertEquals(ohm2.get(3), "LetterK");
}

This unit test also succeeds and shows that we can copy the elements from one OrderedHashMap to a second one, using putAll().

Unless we switch to Java 8.

If we run the above sample with the Java 8 runtime, the test fails with an exception:

java.lang.IndexOutOfBoundsException: Index: 0, Size: 0
	at java.util.ArrayList.rangeCheck(Unknown Source)
	at java.util.ArrayList.get(Unknown Source)
	at com.example.OrderedHashMap.get(OrderedHashMap.java:13)
	at com.example.test.OrderedHashMapTest.testPutAll(OrderedHashMapTest.java:37)
	...

This is an issue I came across when I was trying to translate a grammar with ANTLR (not using the latest version) - see also https://github.com/antlr/antlr4/issues/337.

The background is that the LinkedHashMap class (or one of its super classes) changed the implementation of putAll() so that it no longer calls put() to insert the new elements in the destination map. Hence, our own implementation is no longer able to catch this call to update the List at the same time a new key/value pair is added to the map.

If you do similar things, you should check if there are such differences between the Java 7 and the Java 8 runtime which might affect your code. 

In general, this is a good example to Favor Composition over Inheritance in Java and Object Oriented Programming. The main problem is that the sub class depends on the behaviour of the superclass, and hence becomes fragile (and breaks when the superclass behaviour changes in an incompatible way). Collections are always a bad candidate for subclassing - instead, create a new class which implements the desired interfaces and then delegates to the necessary Collection classes. This is more effort at the beginning (to implement all the necessary interface methods), but makes the class much more resistent against modifications in the runtime classes. See also Effective Java, Second Edition: Item 16: Favor composition over inheritance.


Displaying results 5 to 8 out of 23