java.lang.Objectjava.util.prefs.Preferences
java.util.prefs.AbstractPreferences
Direct Known Subclasses:
WindowsPreferences, FilePreferencesImpl, RegistryPreferencesImpl, FileSystemPreferences
This class is for Preferences implementers only. Normal users of the Preferences facility should have no need to consult this documentation. The Preferences documentation should suffice.
Implementors must override the nine abstract service-provider interface (SPI) methods: #getSpi(String) , #putSpi(String,String) , #removeSpi(String) , #childSpi(String) , #removeNodeSpi() , #keysSpi() , #childrenNamesSpi() , #syncSpi() and #flushSpi() . All of the concrete methods specify precisely how they are implemented atop these SPI methods. The implementor may, at his discretion, override one or more of the concrete methods if the default implementation is unsatisfactory for any reason, such as performance.
The SPI methods fall into three groups concerning exception behavior. The getSpi method should never throw exceptions, but it doesn't really matter, as any exception thrown by this method will be intercepted by #get(String,String) , which will return the specified default value to the caller. The removeNodeSpi, keysSpi, childrenNamesSpi, syncSpi and flushSpi methods are specified to throw BackingStoreException , and the implementation is required to throw this checked exception if it is unable to perform the operation. The exception propagates outward, causing the corresponding API method to fail.
The remaining SPI methods #putSpi(String,String) , #removeSpi(String) and #childSpi(String) have more complicated exception behavior. They are not specified to throw BackingStoreException, as they can generally obey their contracts even if the backing store is unavailable. This is true because they return no information and their effects are not required to become permanent until a subsequent call to Preferences#flush() or Preferences#sync() . Generally speaking, these SPI methods should not throw exceptions. In some implementations, there may be circumstances under which these calls cannot even enqueue the requested operation for later processing. Even under these circumstances it is generally better to simply ignore the invocation and return, rather than throwing an exception. Under these circumstances, however, all subsequent invocations of flush() and sync should return false, as returning true would imply that all previous operations had successfully been made permanent.
There is one circumstance under which putSpi, removeSpi and childSpi should throw an exception: if the caller lacks sufficient privileges on the underlying operating system to perform the requested operation. This will, for instance, occur on most systems if a non-privileged user attempts to modify system preferences. (The required privileges will vary from implementation to implementation. On some implementations, they are the right to modify the contents of some directory in the file system; on others they are the right to modify contents of some key in a registry.) Under any of these circumstances, it would generally be undesirable to let the program continue executing as if these operations would become permanent at a later time. While implementations are not required to throw an exception under these circumstances, they are encouraged to do so. A SecurityException would be appropriate.
Most of the SPI methods require the implementation to read or write information at a preferences node. The implementor should beware of the fact that another VM may have concurrently deleted this node from the backing store. It is the implementation's responsibility to recreate the node if it has been deleted.
Implementation note: In Sun's default Preferences implementations, the user's identity is inherited from the underlying operating system and does not change for the lifetime of the virtual machine. It is recognized that server-side Preferences implementations may have the user identity change from request to request, implicitly passed to Preferences methods via the use of a static ThreadLocal instance. Authors of such implementations are strongly encouraged to determine the user at the time preferences are accessed (for example by the #get(String,String) or #put(String,String) method) rather than permanently associating a user with each Preferences instance. The latter behavior conflicts with normal Preferences usage and would lead to great confusion.
Josh
- Bloch1.4
- Field Summary | ||
---|---|---|
final AbstractPreferences | parent | Our parent node. |
protected boolean | newNode | This field should be true if this node did not exist in the backing store prior to the creation of this object. The field is initialized to false, but may be set to true by a subclass constructor (and should not be modified thereafter). This field indicates whether a node change event should be fired when creation is complete. |
protected final Object | lock | An object whose monitor is used to lock this node. This object is used in preference to the node itself to reduce the likelihood of intentional or unintentional denial of service due to a locked node. To avoid deadlock, a node is never locked by a thread that holds a lock on a descendant of that node. |
Fields inherited from java.util.prefs.Preferences: |
---|
MAX_KEY_LENGTH, MAX_VALUE_LENGTH, MAX_NAME_LENGTH |
Constructor: |
---|
|
Method from java.util.prefs.AbstractPreferences Summary: |
---|
absolutePath, addNodeChangeListener, addPreferenceChangeListener, cachedChildren, childSpi, childrenNames, childrenNamesSpi, clear, exportNode, exportSubtree, flush, flushSpi, get, getBoolean, getByteArray, getChild, getDouble, getFloat, getInt, getLong, getSpi, isRemoved, isUserNode, keys, keysSpi, name, node, nodeExists, nodeListeners, parent, prefListeners, put, putBoolean, putByteArray, putDouble, putFloat, putInt, putLong, putSpi, remove, removeNode, removeNodeChangeListener, removeNodeSpi, removePreferenceChangeListener, removeSpi, sync, syncSpi, toString |
Methods from java.util.prefs.Preferences: |
---|
absolutePath, addNodeChangeListener, addPreferenceChangeListener, childrenNames, clear, exportNode, exportSubtree, flush, get, getBoolean, getByteArray, getDouble, getFloat, getInt, getLong, importPreferences, isUserNode, keys, name, node, nodeExists, parent, put, putBoolean, putByteArray, putDouble, putFloat, putInt, putLong, remove, removeNode, removeNodeChangeListener, removePreferenceChangeListener, sync, systemNodeForPackage, systemRoot, toString, userNodeForPackage, userRoot |
Methods from java.lang.Object: |
---|
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait |
Method from java.util.prefs.AbstractPreferences Detail: |
---|
This implementation merely returns the absolute path name that was computed at the time that this node was constructed (based on the name that was passed to this node's constructor, and the names that were passed to this node's ancestors' constructors). |
|
|
|
Finally, it is guaranteed that the named node has not been returned by a previous invocation of this method or #getChild(String) after the last time that it was removed. In other words, a cached value will always be used in preference to invoking this method. Subclasses need not maintain their own cache of previously returned children. The implementer must ensure that the returned node has not been removed. If a like-named child of this node was previously removed, the implementer must return a newly constructed AbstractPreferences node; once removed, an AbstractPreferences node cannot be "resuscitated." If this method causes a node to be created, this node is not guaranteed to be persistent until the flush method is invoked on this node or one of its ancestors (or descendants). This method is invoked with the lock on this node held. |
This implementation obtains this preference node's lock, checks that the node has not been removed, constructs a TreeSet initialized to the names of children already cached (the children in this node's "child-cache"), invokes #childrenNamesSpi() , and adds all of the returned child-names into the set. The elements of the tree set are dumped into a String array using the toArray method, and this array is returned. |
This method is invoked with the lock on this node held. If this node throws a BackingStoreException, the exception will propagate out beyond the enclosing #childrenNames() invocation. |
This implementation obtains this preference node's lock, invokes #keys() to obtain an array of keys, and iterates over the array invoking #remove(String) on each key. |
|
|
This implementation calls a recursive helper method that locks this node, invokes flushSpi() on it, unlocks this node, and recursively invokes this method on each "cached child." A cached child is a child of this node that has been created in this VM and not subsequently removed. In effect, this method does a depth first traversal of the "cached subtree" rooted at this node, calling flushSpi() on each node in the subTree while only that node is locked. Note that flushSpi() is invoked top-down. If this method is invoked on a node that has been removed with the #removeNode() method, flushSpi() is invoked on this node, but not on others. |
If this node throws a BackingStoreException, the exception will propagate out beyond the enclosing #flush() invocation. |
This implementation first checks to see if key is null throwing a NullPointerException if this is the case. Then it obtains this preference node's lock, checks that the node has not been removed, invokes #getSpi(String) , and returns the result, unless the getSpi invocation returns null or throws an exception, in which case this invocation returns def. |
This implementation invokes get(key, null) . If the return value is non-null, it is compared with "true" using String#equalsIgnoreCase(String) . If the comparison returns true, this invocation returns true. Otherwise, the original return value is compared with "false", again using String#equalsIgnoreCase(String) . If the comparison returns true, this invocation returns false. Otherwise, this invocation returns def. |
|
Finally, it is guaranteed that the named node has not been returned by a previous invocation of this method or #childSpi after the last time that it was removed. In other words, a cached value will always be used in preference to invoking this method. (The implementor needn't maintain his own cache of previously returned children if he chooses to override this method.) This implementation obtains this preference node's lock, invokes #childrenNames() to get an array of the names of this node's children, and iterates over the array comparing the name of each child with the specified node name. If a child node has the correct name, the #childSpi(String) method is invoked and the resulting node is returned. If the iteration completes without finding the specified name, null is returned. |
This implementation invokes get(key, null) . If the return value is non-null, the implementation attempts to translate it to an double with Double#parseDouble(String) . If the attempt succeeds, the return value is returned by this method. Otherwise, def is returned. |
This implementation invokes get(key, null) . If the return value is non-null, the implementation attempts to translate it to an float with Float#parseFloat(String) . If the attempt succeeds, the return value is returned by this method. Otherwise, def is returned. |
This implementation invokes get(key, null) . If the return value is non-null, the implementation attempts to translate it to an int with Integer#parseInt(String) . If the attempt succeeds, the return value is returned by this method. Otherwise, def is returned. |
This implementation invokes get(key, null) . If the return value is non-null, the implementation attempts to translate it to a long with Long#parseLong(String) . If the attempt succeeds, the return value is returned by this method. Otherwise, def is returned. |
Generally speaking, this method should not throw an exception under any circumstances. If, however, if it does throw an exception, the exception will be intercepted and treated as a null return value. This method is invoked with the lock on this node held. |
|
This implementation compares this node's root node (which is stored in a private field) with the value returned by Preferences#userRoot() . If the two object references are identical, this method returns true. |
This implementation obtains this preference node's lock, checks that the node has not been removed and invokes #keysSpi() . |
This method is invoked with the lock on this node held. If this node throws a BackingStoreException, the exception will propagate out beyond the enclosing #keys() invocation. |
This implementation merely returns the name that was passed to this node's constructor. |
This implementation obtains this preference node's lock and checks that the node has not been removed. If path is "", this node is returned; if path is "/", this node's root is returned. If the first character in path is not '/', the implementation breaks path into tokens and recursively traverses the path from this node to the named node, "consuming" a name and a slash from path at each step of the traversal. At each step, the current node is locked and the node's child-cache is checked for the named node. If it is not found, the name is checked to make sure its length does not exceed MAX_NAME_LENGTH. Then the #childSpi(String) method is invoked, and the result stored in this node's child-cache. If the newly created Preferences object's #newNode field is true and there are any node change listeners, a notification event is enqueued for processing by the event dispatch thread. When there are no more tokens, the last value found in the child-cache or returned by childSpi is returned by this method. If during the traversal, two "/" tokens occur consecutively, or the final token is "/" (rather than a name), an appropriate IllegalArgumentException is thrown. If the first character of path is '/' (indicating an absolute path name) this preference node's lock is dropped prior to breaking path into tokens, and this method recursively traverses the path starting from the root (rather than starting from this node). The traversal is otherwise identical to the one described for relative path names. Dropping the lock on this node prior to commencing the traversal at the root node is essential to avoid the possibility of deadlock, as per the locking invariant . |
This implementation is very similar to #node(String) , except that #getChild(String) is used instead of #childSpi(String) . |
|
This implementation obtains this preference node's lock, checks that the node has not been removed and returns the parent value that was passed to this node's constructor. |
|
This implementation checks that the key and value are legal, obtains this preference node's lock, checks that the node has not been removed, invokes #putSpi(String,String) , and if there are any preference change listeners, enqueues a notification event for processing by the event dispatch thread. |
This implementation translates value to a string with String#valueOf(boolean) and invokes #put(String,String) on the result. |
|
This implementation translates value to a string with Double#toString(double) and invokes #put(String,String) on the result. |
This implementation translates value to a string with Float#toString(float) and invokes #put(String,String) on the result. |
This implementation translates value to a string with Integer#toString(int) and invokes #put(String,String) on the result. |
This implementation translates value to a string with Long#toString(long) and invokes #put(String,String) on the result. |
This method is invoked with the lock on this node held. |
This implementation obtains this preference node's lock, checks that the node has not been removed, invokes #removeSpi(String) and if there are any preference change listeners, enqueues a notification event for processing by the event dispatch thread. |
This implementation checks to see that this node is the root; if so, it throws an appropriate exception. Then, it locks this node's parent, and calls a recursive helper method that traverses the subtree rooted at this node. The recursive method locks the node on which it was called, checks that it has not already been removed, and then ensures that all of its children are cached: The #childrenNamesSpi() method is invoked and each returned child name is checked for containment in the child-cache. If a child is not already cached, the #childSpi(String) method is invoked to create a Preferences instance for it, and this instance is put into the child-cache. Then the helper method calls itself recursively on each node contained in its child-cache. Next, it invokes #removeNodeSpi() , marks itself as removed, and removes itself from its parent's child-cache. Finally, if there are any node change listeners, it enqueues a notification event for processing by the event dispatch thread. Note that the helper method is always invoked with all ancestors up to the "closest non-removed ancestor" locked. |
|
This method is invoked with the lock held on this node and its parent (and all ancestors that are being removed as a result of a single invocation to Preferences#removeNode() ). The removal of a node needn't become persistent until the flush method is invoked on this node (or an ancestor). If this node throws a BackingStoreException, the exception will propagate out beyond the enclosing #removeNode() invocation. |
|
This method is invoked with the lock on this node held. |
This implementation calls a recursive helper method that locks this node, invokes syncSpi() on it, unlocks this node, and recursively invokes this method on each "cached child." A cached child is a child of this node that has been created in this VM and not subsequently removed. In effect, this method does a depth first traversal of the "cached subtree" rooted at this node, calling syncSpi() on each node in the subTree while only that node is locked. Note that syncSpi() is invoked top-down. |
If this node throws a BackingStoreException, the exception will propagate out beyond the enclosing #sync() invocation. |
|