• Create DB control class by extending JdbcControl
• Declare DB control variable in page controller and use @Control annotation to have it automatically instantiated and injected
• Use the DB control

The above works fine. But what happens if we would like to use our DB control class from within JSP page which is not part of our Beehive application and has no access to our controller?

The following text presents one possible approach.

Custom DB control class:

Classic approach, instantiate via @Control annotation in controller:

But what if we need to instantiate the DB control class manually? One possible approach might be creating a factory like this:

It simply reads all GIF images in specified directory, does some custom transformation of pixels, and then writes images back onto disk. It is really quick'n'dirty so be warned

Source code: ImageManipulation.java

Reading BLOB from database

This is relatively simple. Just use ResultSet.getBlob() or ResultSet.getClob() method (usually in our custom Spring's RowMapper implementation) and read data using getBinaryStream() (for BLOBs) or getCharacterStream() (for CLOBs).

Inserting BLOB (CLOB) to database

Inserting must be done in two steps. First, we insert an empty BLOB (CLOB) into table row (along with other columns). Second, we lock & select currently inserted row from the table, open the stream and write data.

It is good to do both steps in one transaction (e.g. to prevent somebody from seeing empty BLOB in the meantime - but be sure to take transaction isolation level into account - READ COMMITTED or higher level is needed).

Let's say we declare jdbcTemplate bean in our spring.xml:

where dataSource is (got from JNDI):

Running both steps of BLOB (CLOB) insertion code in transaction can be done like this:

Where MyTransaction for inserting BLOB could be (Java 1.4 this time):

Writing CLOB is very similar - just use oracle.sql.CLOB instead of oracle.sql.BLOB everywhere and write data to CLOB using getCharacterOutputStream() method.

• Classpath hell debugging
• Oracle XML parser crashes on SOAP response
• Using custom AQ message payload (Oracle ADT message)

If anyone encountered such problems, read the rest of the article for tips and solutions.

Classpath hell

Really useful technique for solving classpath issues is to turn on OC4J class load tracing. It can be done by passing special arguments to VM.

If starting OC4J manually via oc4j/bin/oc4j.cmd you can run the following one-line command before:

SET OC4J_JVM_ARGS=-Dclass.load.trace=class+loader
-Dclass.load.log.file=c:/classloadlog.txt
-Dclass.load.log.level=all


If you're running OC4J server directly from Eclipse's Server view, put these arguments into VM arguments section of Run dialog.

Oracle XML parser crashes on SOAP response

Oracle XML parser is somewhat buggy and crashes on manually created SOAP response if the message body isn't surrounded with <Envelope> element (the same one which surrounds the whole SOAP message).

One way of getting rid of this is to completely replace Oracle XML parser with e.g. Apache Xerces parser. If you have full access to application server you can try adding Xerces as OC4J's shared library. As this wasn't our case (we weren't allowed to change the configuration of production server) we needed to find another way.

The first step was to add Xerces jars to our application.

If you only deploy manually via server's web interface, second step is quite easy as it is sufficient to remove oracle.xml shared library at the very last step of the deployment process.

If you use Eclipse's automatic publishing (which you probably are because nobody wants to manually deploy the application after every modification when developing) things get more complicated.

Upon every publishing, new orion-application.xml file for our application is created. We need it to keep containing the following snippet:

This can be achieved by:

1. creating deployment plan with removed reference to oracle.xml shared library (via manual web deployment)
2. saving the plan to some place on disk
3. modifying oracle.10.1.3.xml file in [eclipse_dir]/plugins/org.eclipse.jst.server.generic.oc4j/buildfiles:
   
   PLAIN TEXT
   XML:

   1. ...
   2. <target name="deploy.j2ee.ear"
   3. depends="check.skip.ear.deploy"
   4. unless="skip.deploy">
   5. ...
   6. <oracle:deploy ... deploymentPlan="path_to_deployment_plan" ...
   7. ...

Using custom AQ message payload (Oracle ADT message)

Using standard JMS message types for AQ works just fine (see this great techblog's article for description). To turn on JTA transactions, add the following to your Spring configuration file:

A little more advanced is to use custom objects as message payload. Follow these steps:

1. Create custom type & queue in your Oracle database
2. Use jpub to generate Java class for the type
3. Sending message is pretty straightforward
4. For receiving message, an extension of Spring's DefaultMessageListenerContainer is needed

Ad 1)

For type:
CREATE OR REPLACE TYPE MyCustomType...

For queue table:
execute dbms_aqadm.create_queue_table(
queue_table => 'MyQueueTable',
queue_payload_type => 'MyCustomType'
...


Ad 2)

Download JPublisher, unpack it to e.g. c:\oracle\sqlj and run it:

SET CLASSPATH = <path_to_driver>\ojdbc14.jar; c:\oracle\sqlj\lib\translator.jar; c:\oracle\sqlj\lib\runtime12.jar;

SET PATH = c:\Program Files\Java\j2sdk1.4.2_16\bin

java -cp %CLASSPATH% oracle.jpub.Doit -user=autopub/autopub -url=<url to your db> -sql=<name of type> -case=mixed -compatible=CustomDatum -usertypes=oracle


Ad 3)

Ad 4)

We need to extend DefaultMessageListenerContainer's createConsumer method to slip payload factory which can handle our new custom object type:

• Run some code on plugin startup?
• Set text file encoding?

If the answer to any of these questions is yes, be sure to read on!

How to run code when plugin starts

Sometimes it is necessary to run custom code on plugin startup. Two steps are needed to do this. First, add the following to your plugin.xml:

Second, create the startup class:

How to set text file encoding

We needed to support special text files that have character encoding specified inside the file. In custom editors this can be achieved:

Server side

First thing we need is a small utility script called bugzilla-submit which can be found in BUGZILLA DIRECTORY/contrib/bugzilla-submit folder. It's simple command line interface for submitting new bugs. This script reads .netrc file in the home directory (/var/www in our case) which must contain bugzilla system credentials:

machine "http://bugzilla.our-company.com"
login <an email address of bugzilla user>
password <password>


Second thing we need is to create a simple php script that will live on some publicly accessible server and will process POSTed bugs and submit them using the mentioned utility into bugzilla server (which may or may not be on the same machine). Basic skeleton of such script:

As you can see, the script is very simple as it allows only bug description to be modified (by our Eclipse plugin reporting system) and other fields are hard coded. It also isn't secured in any way so anyone who knows URL of the script (and guesses the format of POST request) can submit bugs!

There are many possibilities of how to secure it (e.g. some kind of one time passwords, https, limiting the number of allowed submit from one IP address etc.).

Client side

The third (and last) thing is the code in our Eclipse plugin that handles exceptions and asks user if he wants to report them. We do this simply by listening to workspace log events and reporting (or asking user if he wants to report) if one of the following conditions holds:

1. The plugin where an error occured is ours
2. The exception stacktrace contains our class

We consider plugin or class "ours" if its name starts with "com.our-company.productname". The following code must be run at startup (see how to do it):

And that's probably all

Works nice when used in conjunction with org.apache.xerces.parsers.SAXParser (but wasn't tested with anything else so take care...).

If you're interested, download the original xom-1.1.zip file, unpack and unpack all files from my xom-1.1.position.patch.zip over original ones.

• Custom event listeners
• Soft reference cache
• Preventing memory leaks when using cached objects & listeners

Creating and using custom event listeners

Listeners are great when it comes to the communication of objects that should stay decoupled to a high degree. They are used just about everywhere (e.g. SWT) and sometimes it would be nice to add custom listeners to our objects. And that's showtime for ListenerList class from org.eclipse.core.runtime package. It is fully thread-safe list designed for storing lists of listeners (optimized for frequent reads and infrequent writes).

Typical custom class would contain something like:

With IMyListener being:

So when our event happens in our class ClassWithListeners we notify all listeners (that have been previously registered by addMyEventListener()) by calling fireMyEvent().

Its always good practice to unregister by calling removeMyEventListener() when we are no longer interested in the event or when the owning objects are going to be disposed. To see how to achieve this automatically read on.

Creating soft reference cache

Common scenario is to have some objects that are very expensive (time consuming) to create but once created they can be used many times as they don't change (or at least not too often).

Such objects are ideal candidates for some kind of caching. Naive approach is to create these objects at the start of program and keep them in memory until the end of the program. It sure works but the price of slower program startup and bigger memory footprint may be too high.

First improvement would be lazy initialization (do not create object until first needed). Second and here described improvement is soft reference caching whose main advantage is that cached objects can be garbage-collected when needed (to release memory for other objects) and recreated when they are to be used again.

The simplest cache might look like this:

Of course, there is room for improvements (generics, cache entries invalidation etc.). Without much trouble a whole generic cache can be written.

Preventing memory leaks when using cached objects & listeners

If we combine both above described steps and cache objects that register some listeners to other objects we might end up having memory leaks in our application. Sample scenario:

The problem with this scenario is that ExpensiveObject will never be garbage-collected even if memory runs low. That's because ObjectThatLivesTheWholeApplicationLife holds instance of IMyListener that holds strong (non garbage-collectable) reference to an instance of ExpensiveObject.

Possible way to solve it is to wrap IMyListener into weak reference (which is garbage-collectable if there are no other strong references to the object). Moreover the wrapper will automatically unregister event handler if its owner (ExpensiveObject) is disposed.

Such wrapper could look like this:

And the usage could be either on the side of ExpensiveObject:

or on the side of ObjectThatLivesTheWholeApplicationLife:

• How to get the index of selected column in TableViewer
• How to create and use TreeViewer with multiple equal items

Getting the index of selected column in TableViewer

Let's say we want do some action when user double clicks on the row in TableViewer. Achieving this is easy, the code might look like this:

If we need to know not only which row but also which exact column was double clicked, things get more complicated then that because there doesn't seem to be direct standard support for this:

Creating TreeViewer with multiple equal items

There are situations where we want a TreeViewer to contain multiple items that are equal (i.e. X.equals(Y) returns true for items X, Y).

An example of such situation would be tree containing classes (as root nodes) with every class having its super class as a child node. Then it is possible to imagine having class A as a root node and class B as its child node (which means class A extends class B) - but class B might be also root node (with another class C as a child).

We can see that class B is contained twice in the TreeViewer and this may lead to problems (one of which is incorrectly working treeViewer.setSelection(B) - it will randomly select one of two B class nodes depending on circumstances).

The solution consists of these steps:

1. Use ITreePathContentProvider instead of commonly used ITreeContentProvider content provider
3. Before selecting the item in tree do treeViewer. setSelection(null) first - this is needed to get around what I am not sure is feature or bug - there is some internal hash table used and if currently selected item has the same hashCode() as the new one, the new selection is ignored. Setting null before real selection helps.
4. For selecting item in tree use setSelection(treePath) (instead of setSelection(item)) - simply put, TreePath is path consisting of items from root to item we want to select (inclusive)