Provides high-performance collection classes and miscellaneous utilities; although
this package provides very few collection classes, they are substitutes for
most of java.util.* classes (for example, java.util.IdentityHashMap would be
a {@link javolution.util.FastMap FastMap} with an {@link javolution.util.FastComparator#IDENTITY
identity} key comparator).
Javolution collections are compliant with standard collections
(generic when built with the ant target 1.5 ) and they can safely be used
with RTSJ virtual machines (e.g. if the capacity of
a collection increases, the extension part is allocated from the same memory
area as the collection itself).
They support direct iterations with the following advantages:
- Faster than iterators, see benchmark.
- No object creation not even the iterator object itself. For example, visiting a tree structure using
iterators creates as many iterators as they are nodes in the tree:[code]
public static void visit(Collection node) {
for (Collection i : node) { // Creates iterator.
visit(i);
}
}[/code]
Not so with direct iterations:[code]
public static void visit(FastCollection node) {
for (FastCollection.Record r = node.head(), end = node.tail(); (r = r.getNext()) != end;) {
visit(node.valueOf(r));
}
}[/code]
- Used to implement most of {@link javolution.util.FastCollection FastCollection} base class methods
(including {@link javolution.util.FastCollection#iterator iterator()}).
- Support forward/backward iterations from the start (head) or from the end (tail)
- Thread-safe as long as objects are not inserted/removed during iterations. Objects can safely be
append/prepend by the current thread or other threads.
(Note: {@link javolution.util.FastMap#setShared Shared FastMap} are always thread-safe even when entries are removed).
- Fully integrated with the JDK1.5+ generic framework (strong typing) and still compatible
with other platforms (J2ME, 1.4, GCJ).
Here are few examples of direct iterations:[code]
FastList list;
for (FastList.Node n = list.head(), end = list.tail(); (n = n.getNext()) != end;) {
String value = n.getValue(); // No typecast necessary.
}
...
FastMap map;
for (FastMap.Entry e = map.head(), end = map.tail(); (e = e.getNext()) != end;) {
String key = e.getKey(); // No typecast necessary.
Thread value = e.getValue(); // No typecast necessary.
}[/code]
Users may provide a read-only view of any {@link javolution.util.FastCollection FastCollection}
(or {@link javolution.util.FastMap FastMap}) instance using the
{@link javolution.util.FastCollection#unmodifiable() FastCollection.unmodifiable()}
(or {@link javolution.util.FastMap#unmodifiable FastMap.unmodifiable()}) method.
For example:[code]
public class Polynomial {
private final FastSet _terms = new FastSet();
// Read-only view (also thread-safe as terms are not "deleted").
public Set getTerms() {
return _terms.unmodifiable();
}
}[/code]
Collection/maps of primitive types can be created using the
{@link javolution.util.Index Index} class. It avoids the overhead
of wrapping primitives types (for reasonably small int values).
For example:[code]
public class SparseVector {
FastMap _elements = new FastMap();
...
}[/code]
Although all collections capacity increases smoothly (no resizing/copy or rehashing ever performed),
it is nevertheless possible to specify an initial capacity; in which case, all necessary storage
is allocated at creation. For RTSJ VMs, all
collections/maps can reside in ImmortalMemory (e.g. static )
and be used by all threads (including NoHeapRealtimeThread ) without resulting into memory leaks
or illegal access errors. For example:[code]
public class XmlFormat {
// RTSJ Unsafe! Memory leaks (when entries removed) or IllegalAssignmentError (when new entries while in ScopedArea).
static HashMap ClassToFormat = HashMap();
// RTSJ Safe! Removed entries are internally recycled, new entries are in ImmortalMemory.
static FastMap ClassToFormat = FastMap();
}[/code]
For more details, please read Javolution-Collection.pdf .
Temporary collection classes can be recycled (e.g. throw-away collections) to avoid the creation
cost. For example:[code]
static void removeDuplicate(List persons) {
FastSet tmp = FastSet.newInstance(); // Possibly recycled instance.
tmp.addAll(persons);
persons.clear();
persons.addAll(tmp);
FastSet.recycle(tmp); // Recycles the temporary instance.
}[/code]
Here is a summary of the collection classes with their defining characteristics:
Javolution Collections Classes
|
Ordering |
Duplication Allowed |
Custom Comparators |
Record Type |
Miscellaneous |
{@link javolution.util.FastTable FastTable} |
Insertion Order |
Yes |
{@link javolution.util.FastTable#setValueComparator setValueComparator(FastComparator)} |
{@link javolution.util.Index Index} |
Thread-safe random access collection
No array resize/copy ever performed |
{@link javolution.util.FastList FastList} |
Insertion Order |
Yes |
{@link javolution.util.FastList#setValueComparator setValueComparator(FastComparator)} |
{@link javolution.util.FastList.Node Node} |
Recycle their own nodes (no adverse effect on GC) |
{@link javolution.util.FastSet FastSet} |
Insertion Order |
No |
{@link javolution.util.FastSet#setValueComparator setValueComparator(FastComparator)} |
{@link javolution.util.FastCollection.Record Record} |
Based on {@link javolution.util.FastSet FastMap} (same characteristics) |
FastTree |
Comparator |
No |
setValueComparator(FastComparator) |
TreeNode |
(not implemented) |
{@link javolution.util.FastMap FastMap} |
Insertion Order |
Key: No Value: Yes |
{@link javolution.util.FastMap#setKeyComparator setKeyComparator(FastComparator)}
{@link javolution.util.FastMap#setValueComparator setValueComparator(FastComparator)} |
{@link javolution.util.FastMap.Entry Entry} |
Thread-safe when marked as {@link javolution.util.FastMap#setShared shared}
No rehash/resize ever performed
Recycle their own entries (no adverse effect on GC) |
- ArrayList may throw ConcurrentModificationException,
but Javolution FastTable does not, why?
FastTable (or any Javolution collection/map) do support concurrent modifications
as long as these are not insertions at an arbitrary position or deletions (Note: Shared FastMap
does support concurrent deletions).
In other words you can safely iterate (using iterators or not) through a FastList, FastMap
(entries, keys values), FastTable, etc. while new elements/entries are being added
(by you or another thread). You can also export a {@link javolution.util.FastCollection#unmodifiable() read-only}
view over your collection and still add more elements to it.
Disallowing concurrent modifications (standard java util) has proven to be a performance
killer for many (forcing users to work with copies of their whole collections). Furthermore the additional checks required
directly impact performance (e.g. ArrayList iterations about 3x slower than FastTable iterations).
- Do you have a test case showing any scenario of concurrent modification where
ArrayList "fails" and FastTable doesn't?
Let's say that you have a collection of "Units", and you want to provide users
with a read-only view of these units. The following code will fail miserably:[code]
public class Unit {
static ArrayList INSTANCES = new ArrayList();
public static List getInstances() {
return Collections.unmodifiableList(INSTANCES);
}
}[/code]
Why? Because, it the user iterates on the read-only list of units while a new unit is added
to the collection (by another thread) a ConcurrentModificationException is
automatically raised. In other words, it is almost impossible to provide a "read-only" view
of non-fixed size collections with the current java.util classes (e.g. you will have to replace
the whole collection each time a new unit is added).
Now with FastTable the following is completely safe even when new units are added:[code]
public class Unit {
static FastTable INSTANCES = new FastTable();
public static List getInstances() {
return INSTANCES.unmodifiable();
}
}[/code]
- Do checks for concurrent modifications make your code safer?
Not really. The current checks for concurrent modifications do not "guarantee" that concurrent
modifications will not occur! You can imagine two threads one updating a collection
and the other one iterating the collection. As long as the update is not performed
while the other thread is iterating, everything is fine (no ConcurrentModificationException)!
But, if for a reason or another the timing changes (e.g. in the user environment) and
iterations are performed at the wrong time then your application crashes...
Not a good thing and very high probability for this to happen!
- Are {@link javolution.util.FastMap#setShared shared maps} valid substitutes for
ConcurrentHashMap ?
Unlike ConcurrentHashMap access to a
shared FastMap never blocks. Retrieval reflects the map state not older than the last
time the accessing threads have been synchronized* (for multi-processors
systems synchronizing ensures that the CPU internal cache is not stale).
In practice, it means that most well-written concurrent programs should
be able to use shared FastMap in place of ConcurrentHashMap as
threads are already synchronized to ensure proper behavior.
* It is important for both threads to synchronize on the same monitor
in order to set up the happens-before relationship properly. It is not the case
that everything visible to thread A when it synchronizes on object X becomes visible
to thread B after it synchronizes on object Y. The release and acquire have to
"match" (i.e., be performed on the same monitor) to have the right semantics.
Otherwise, the code has a data race.
- Are all Javolution collection thread-safe?
Collections/Maps are thread-safe with regard to access (no need to
synchronize reading even if the collection is modified concurrently).
But the modifications themselves require either the collection/map to be
marked shared or synchronization to be used.
- What is the overhead in term of performance when
FastMap.setShared
is set to true ?
Marking the map shared avoid synchronizing when possible (e.g. put when
entry already exists or remove when entry does not exist), if a new entry
is created and added, synchronization is performed internally. In all
cases there is no impact on reading (never synchronized).
|