Hibernate: A Developer's Notebook

Writing a Mapping Document

Why do I care?

You might not have any need for a new system to keep track of your music, but the concepts and process involved in setting up this mapping will translate to the projects you actually want to tackle.

How do I do that?

Fire up your favorite text editor, and create the file Track.hbm.xml in the src/com/oreilly/hh directory you set up in the previous chapter. (If you skipped that chapter, you'll need to go back and follow it, because this example relies on the project structure and tools we set up there.) Type in the mapping document as shown in Example 2-1. Or, if you'd rather avoid all that typing, download the code examples from this book's web site, and find the mapping file in the directory for Chapter 2.

The first four lines are a required preamble to make this a valid XML document and announce that it conforms to the document type definition used by Hibernate for mappings. The actual mappings are inside the hibernate-mapping tag. Starting at line 7 we're defining a mapping for a single class, com.oreilly.hh.Track, and the name and package of this class are related to the name and location of the file we've created. This relationship isn't necessary; you can define mappings for any number of classes in a single mapping document, and name it and locate it anywhere you want, as long as you tell Hibernate how to find it. The advantage of following the convention of naming the mapping file after the class it maps, and placing it in the same place on the class path as that class, is that this allows Hibernate to automatically locate the mapping when you want to work with the class. This simplifies the configuration and use of Hibernate.

In the opening of the class tag on line 7, we have also specified that this class is stored in a database table named TRACK. The next tag, a meta tag (lines 8-11), doesn't directly affect the mapping. Instead, it provides additional information that can be used by different tools. In this case, by specifying an attribute value of "class-description," we are telling the Java code generation tool the JavaDoc text we want associated with the Track class. This is entirely optional, and you'll see the result of including it in Section 2.2.

TIP: Although databases vary in terms of whether they keep track of the capitalization of table and column names, this book will use the convention of referring to these database entities in all-caps, to help clarify when something being discussed is a database column or table, as opposed to a persistent Java class or property.

The remainder of the mapping sets up the pieces of information we want to keep track of, as properties in the class and their associated columns in the database table. Even though we didn't mention it in the introduction to this example, each track is going to need an id. Following database best practices, we'll use a meaningless surrogate key (a value with no semantic meaning, serving only to identify a specific database row). In Hibernate, the key/id mapping is set up using an id tag (starting at line 13). We're choosing to use an int to store our id in the database column TRACK_ID, which will correspond to the property id in our Track object. This mapping contains another meta tag to communicate with the Java code generator, telling it that the set method for the id property should be protected—there's no need for application code to go changing track IDs.

The generator tag on line 15 configures how Hibernate creates id values for new instances. (Note that it relates to normal O/R mapping operation, not to the Java code generator, which is often not even used; generator is more fundamental than the optional meta tags.) There are a number of different ID generation strategies to choose from, and you can even write your own. In this case, we're telling Hibernate to use whatever is most natural for the underlying database (we'll see later on how it learns what database we're using). In the case of HSQLDB, an identity column is used.

NOTE You may be thinking there's a lot of dense information in this file. That's true, and as you'll see, it can be used to create a bunch of useful project resources.

After the id, we just enumerate the various track properties we care about. The title (line 18) is a string, and it cannot be null. The filePath (line 20) has the same characteristics, while the remainder are allowed to be null: playTime (line 22) is a time, added (line 26) is a date, and volume (line 30) is a short. These last three properties use a new kind of meta attribute, "field-description," which specifies JavaDoc text for the individual properties, with some limitations in the current code generator.

What just happened?

We took the abstract description of the information about music tracks that we wanted to represent in our Java code and database, and turned it into a rigorous specification in the format that Hibernate can read. Hopefully you'll agree that it's a pretty compact and readable representation of the information. Next we'll look at what Hibernate can actually do with it.

What about...

. . . Other data types, including nested classes and enumerations? Relationships between tables? Indices? Class hierarchies and polymorphism? Tables that contain rows we need to ignore? Hibernate can handle all these things, and we'll cover most of them in later examples. Appendix A lists the basic types that Hibernate supports "out of the box."

Generating Some Class

Our mapping contains information about both the database and the Java class between which it maps. We can use it to help us create both. Let's look at the class first.

How do I do that?

The Hibernate Extensions you installed in Chapter 1 included a tool that can write Java source matching the specifications in a mapping document, and an Ant task that makes it easy to invoke from within an Ant build file. Edit build.xml to add the portions shown in bold in Example 2-2.

We added a taskdef (task definition) and a new target to the build file. The task definition at line 33 teaches Ant a new trick: it tells Ant how to use the hbm2java tool that is part of the Hibernate Extensions, with the help of a class provided for this purpose. Note that it also specifies the class path to be used when invoking this tool, using the project.class.path definition found earlier in the file.

The codegen target at line 38 uses the new hbm2java task to run Hibernate's code generator on any mapping documents found in the src tree, writing the corresponding Java source. The pattern "**/*.hbm.xml" means "any file ending in .hbm.xml, within the specified directory, or any subdirectory, however deeply nested."

Let's try it! From within your top-level project directory (the folder containing build.xml), type the following command:

1    ant codegen

You should see output like this:

1    Buildfile: build.xml
2    
3    codegen:
4     [hbm2java] Processing 1 files.
5     [hbm2java] Building hibernate objects
6     [hbm2java] log4j:WARN No appenders could be found for logger (net.sf.
7    hibernate.util.DTDEntityResolver).
8     [hbm2java] log4j:WARN Please initialize the log4j system properly.

The warnings are griping about the fact that we haven't taken the trouble to set up the logging environment that Hibernate expects. We'll see how to do that in the next example. For now, if you look in the directory src/com/oreilly/hh, you'll see that a new file named Track.java has appeared, with the content shown in Example 2-3.

What just happened?

Ant found all files in our source tree ending in .hbm.xml (just one, so far) and fed it to the Hibernate code generator, which analyzed it, and wrote a Java class meeting the specifications we provided for the Track mapping.

NOTE That can save a lot of time and fairly repetitive activity. I could get used to it.

You may find it worthwhile to compare the generated Java source with the mapping specification from which it arose (Example 2-1). The source starts out with the proper package declaration, which is easy for hbm2java to figure out from the fully qualified class name required in the mapping file. There are a couple of imports to make the source more readable. The three potentially unfamiliar entries (lines 5-7) are utilities from the Jakarta Commons project that help in the creation of correctly implemented and useful toString( ), equals( ), and hashCode( ) methods.

The class-level JavaDoc at line 10 should look familiar, since it comes right from the "class-description" meta tag in our mapping document. The field declarations are derived from the id (line 17) and property (lines 20-32) tags defined in the mapping. The Java types used are derived from the property types in the mapping document. We'll delve into the full set of value types supported by Hibernate later on. For now, the relationship between the types in the mapping document and the Java types used in the generated code should be fairly clear.

One curious detail is that an Integer wrapper has been used for id, while volume is declared as a simple, unwrapped short. Why the difference? It relates to the fact that the ID/key property has many important roles to play in the O/R mapping process (which is why it gets a special XML tag in the mapping document, rather than being just another property). Although we left it out in our specification, one of the choices you need to make when setting up an ID is to pick a special value to indicate that a particular instance has not yet been saved into the database. Leaving out this unsaved-value attribute, as we did, tells Hibernate to use its default interpretation, which is that unsaved values are indicated by an ID of null. Since native int values can't be null, they must be wrapped in a java.lang.Integer, and Hibernate took care of this for us.

When it comes to the volume property, Hibernate has no special need or use for it, so it trusts us to know what we're doing. If we want to be able to store null values for volume, perhaps to indicate "no change," we need to explicitly use java.lang.Short rather than short in our mapping document. (Had we not been sneakily pointing out this difference, our example would be better off explicitly using java.lang.Integer in our ID mapping too, just for clarity.)

Another thing you might notice about these field declarations is that their JavaDoc is quite generic—you may be wondering what happened to the "field-description" meta tags we put in the mapping document for playTime, added and volume. It turns out they appear only later, in the JavaDoc for the getter methods. They are not used in the setters, the actual field declarations, nor as @param entries for the constructor. As an avid user of a code-completing Java editor, I count on pop-up JavaDoc as I fill in arguments to method calls, so I'm a little disappointed by this limitation. Of course, since this is an open source project, any of us can get involved and propose or undertake this simple fix. Indeed, you may find this already remedied by the time you read this book. Once robust field and parameter documentation is in place, I'd definitely advocate always providing a brief but accurate field-description entry for your properties.

NOTE I know, I'm a perfectionist. I only bother to pick nits because I think Hibernate is so useful!

After the field declarations come a trio of constructors. The first (line 35) establishes values for all properties, the second (line 44) allows instantiation without any arguments (this is required if you want the class to be usable as a bean, such as on a Java Server Page, a very common use for data classes like this), and the last (line 48) fills in just the values we've indicated must not be null. Notice that none of the constructors set the value of id; this is the responsibility of Hibernate when we get the object out of the database, or insert it for the first time.

Consistent with that, the setId( ) method on line 57 is protected, as requested in our id mapping. The rest of the getters and setters are not surprising; this is all pretty much boilerplate code (which we've all written too many times), which is why it's so nice to be able to have the Hibernate extensions generate it for us.

WARNING If you want to use Hibernate's generated code as a starting point and then add some business logic or other features to the generated class, be aware that all your changes will be silently discarded the next time you run the code generator. In such a project you will want to be sure the hand-tweaked classes are not regenerated by any Ant build target.

Even though we're having Hibernate generate our data classes in this example, it's important to point out that the getters and setters it creates are more than a nice touch. You need to put these in your persistent classes for any properties you want to persist, since Hibernate's fundamental persistence architecture is based on reflective access to JavaBeans©-style properties. They don't need to be public if you don't want them to; Hibernate has ways of getting at even properties declared protected or private, but they do need accessor methods. Think of it as enforcing good object design; the Hibernate team wants to keep the implementation details of actual instance variables cleanly separated from the persistence mechanism.

Cooking Up a Schema

That was pretty easy, wasn't it? You'll be happy to learn that creating database tables is a very similar process. As with code generation, you've already done most of the work in coming up with the mapping document. All that's left is to set up and run the schema generation tool.

How do I do that?

The first step is something we alluded to in Chapter 1. We need to tell Hibernate the database we're going to be using, so it knows the specific "dialect" of SQL to use. SQL is a standard, yes, but every database goes beyond it in certain directions and has a specific set of features and limitations that affect real-life applications. To cope with this reality, Hibernate provides a set of classes that encapsulate the unique features of common database environments, in the package net.sf.hibernate.dialect. You just need to tell it which one you want to use. (And if you want to work with a database that isn't yet supported "out of the box," you can implement your own dialect.)

In our case, we're working with HSQLDB, so we want to use HSQLDialect. The easiest way to configure Hibernate is to create a properties file named hibernate.properties and put it at the root level somewhere in the class path. Create this file at the top level of your src directory, and put the lines shown in Example 2-4 into it.

NOTE You can use an XML format for the configuration information as well, but for the simple needs we have here, it doesn't buy you anything.

In addition to establishing the SQL dialect we are using, this tells Hibernate how to establish a connection to the database using the JDBC driver that ships as part of the HSQLDB database JAR archive, and that the data should live in the data directory we've created—in the database named music. The username and empty password (indeed, all these values) should be familiar from the experiment we ran at the end of Chapter 1.

TIP: Notice that we're using a relative path to specify the database filename. This works fine in our examples — we're using ant to control the working directory. If you copy this for use in a web application or other environment, though, you'll likely need to be more explicit about the location of the file.

You can put the properties file in other places, and give it other names, or use entirely different ways of getting the properties into Hibernate, but this is the default place it will look, so it's the path of least resistance (or, I guess, least runtime configuration).

We also need to add some new pieces to our build file, shown in Example 2-5. This is a somewhat substantial addition, because we need to compile our Java source in order to use the schema generation tool, which relies on reflection to get its details right. Add these targets right before the closing </project> tag at the end of build.xml.

First we add a prepare target that is intended to be used by other targets more than from the command line. Its purpose is to create, if necessary, the classes directory into which we're going to compile, and then copy any properties and mapping files found in the src directory hierarchy to corresponding directories in the classes hierarchy. This hierarchical copy operation (using the special "**/*" pattern) is a nice feature of Ant, enabling us to define and edit resources alongside to the source files that use them, while making those resources available at runtime via the class loader.

The aptly named compile target at line 14 uses the built-in java task to compile all the Java source files found in the src tree to the classes tree. Happily, this task also supports the project class path we've set up, so the compiler can find all the libraries we're using. The depends="prepare" attribute in the target definition tells Ant that before running the compile target, prepare must be run. Ant manages dependencies so that when you're building multiple targets with related dependencies, they are executed in the right order, and each dependency gets executed only once, even if it is mentioned by multiple targets.

If you're accustomed to using shell scripts to compile a lot of Java source, you'll be surprised by how quickly the compilation happens. Ant invokes the Java compiler within the same virtual machine that it is using, so there is no process startup delay for each compilation.

Finally, after all this groundwork, we can write the target we really wanted to! The schema target (line 26) depends on compile, so all our Java classes will be compiled and available for inspection when the schema generator runs. It uses taskdef internally at line 31 to define the schemaexport task that runs the Hibernate schema export tool, in the same way we provided access to the code generation tool at the top of the file. It then invokes this tool and tells it to generate the database schema associated with any mapping documents found in the classes tree.

There are a number of parameters you can give the schema export tool to configure the way it works. In this example (at line 35) we're telling it to display the SQL it runs so we can watch what it's doing (quiet="no"), to actually interact with the database and create the schema rather than simply writing out a DDL file we could import later or simply deleting the schema (text="no", drop="no"). For more details about these and other configuration options, consult the Hibernate reference manual.

TIP: You may be wondering why the taskdef for the schema update tool is inside our schema target, rather than at the top of the build file, next to the one for hbm2java. Well, I wanted it up there too, but I ran into a snag that's worth explaining. I got strange error messages the first time I tried to build the schema target, complaining there was no hibernate.properties on the class path and our compiled Track class couldn't be found. When I ran it again, it worked. Some detective work using ant-verbose revealed that if the classes directory didn't exist when the taskdef was encountered, Ant helpfully removed it from the class path. Since a taskdef can't have its own dependencies, the solution is to move it into the schema target, giving it the benefit of that target's dependencies, ensuring the classes directory exists by the time the taskdef is processed.

With these additions, we're ready to generate the schema for our TRACK table.

WARNING You might think the drop="no" setting in our schematask means you can use it to update the schema—it won't drop the tables, right? Alas, this is a misleading parameter name: it means it won't just drop the tables, rather it will go ahead and generate the schema after dropping them. Much as you want to avoid the codegen task after making any changes to the generated Java source, you mustn't export the schema if you've put any data into the database. Luckily, there is another tool you can use for incremental schema updates that works much the same way, as long as your JDBC driver is powerful enough. This SchemaUpdate tool can be used with an Ant taskdef too.

Because we've asked the schema export task not to be "quiet," we want it to generate some log entries for us. In order for that to work, we need to configure log4j, the logging environment used by Hibernate. The easiest way to do this is to make a log4j.properties file available at the root of the class path. We can take advantage of our existing prepare target to copy this from the src to the classes directory at the same time it copies Hibernate's properties. Create a file named log4j.properties in the src directory with the content shown in Example 2-6. An easy way to do this is to copy the file out of the src directory in the Hibernate distribution you downloaded, since it's provided for use by their own examples. If you're typing it in yourself, you can skip the blocks that are commented out; they are provided to suggest useful logging alternatives.

TIP: With the log configuration in place, you might want to edit the codegen target in build.xml so that it, too, depends on our new prepare target. This will ensure logging is configured whenever we use it, preventing the warnings we saw when first running it. As noted in the tip about class paths and task definitions in the previous section, though, to make it work the very first time you'll have to move the taskdef for hbm2java inside the codegen target, in the same way we put schemaexport inside the schema target.

Toward the end of the schemaexport section you can see the actual SQL used by Hibernate to create the TRACK table. If you look at the start of the music.script file in the data directory, you'll see it's been incorporated into the database. For a slightly more friendly (and perhaps convincing) way to see it, execute ant db to fire up the HSQLDB graphical interface, as shown in Figure 2-1.

What just happened?

We were able to use Hibernate to create a data table in which we can persist instances of the Java class it created for us. We didn't have to type a single line of SQL or Java! Of course, our table is still empty at this point. Let's change that! The next chapter will look at the stuff you probably most want to see: using Hibernate from within a Java program to turn objects into database entries and vice versa.

NOTE It's about time? Yeah, I suppose. But at least you didn't have to figure out all these steps from scratch!

Before diving into that cool task, it's worth taking a moment to reflect on how much we've been able to accomplish with a couple of XML and properties files. Hopefully you're starting to see the power and convenience that make Hibernate so exciting.

What about...

. . . Other approaches to ID generation? Keys that are globally unique across a database or the world? Hibernate can support a variety of methods for picking the keys for objects it stores in the database. This is controlled using the generator tag, line 15 in Example 2-1. In this example we told Hibernate to use the most natural kind of keys for the type of database that it happens to be using. Other alternatives include the popular "hi/lo" algorithm, global UUIDs, leaving it entirely up to your Java code, and more. See the "generator" section in the Basic O/R Mapping chapter of the Hibernate reference documentation for details. And, as usual, if none of the built-in choices are perfect for your needs, you can supply your own class to do it exactly how you'd like, implementing the interface net.sf.hibernate.id.IdentifierGenerator and supplying your class name in the generator tag.

Connecting Hibernate to MySQL

If you were skimming through this chapter (or, more likely, the table of contents) you may not have even noticed that Hibernate connected to and manipulated a database in the previous section, Section 2.3. Since working with databases is the whole point of Hibernate, it makes this as easy as possible. Once you've set up a configuration file like the one in Example 2-4, the schema generation tool can get in and work with your database, and your Java code can use it for persistence sessions as demonstrated in Chapter 3.

NOTE This example assumes you've already got a working MySQL instance installed and running, since explaining how to do that would be quite a detour.

How do I do that?

Connect to your MySQL server and set up a new database to play with, along the lines of Example 2-8.

Make a note of the database name you create, as well as the username and password that can access to it. These will need to be entered into hibernate.properties, as shown in Example 2-9.

NOTE Hopefully you'll use a less guessable password than this in your real databases!

The URL on the third line will need to reflect your server; you won't be able to resolve my private internal domain name, let alone route to it.

Once this is all set, you can rerun the schema creation example that was set up in the previous section. This time it will build the schema on your MySQL server rather than in the embedded HSQLDB world. You'll see output like that in Example 2-10.

What just happened?

Hibernate configured itself to work with MySQL's specific features, examined the mapping document for our Track class, connected to the MySQL server, and executed the commands necessary to build a database schema for persisting Track instances.

It's interesting to compare Example 2-11 with Example 2-7. Most of the output is the same, but there are subtle differences in the SQL used to actually create the table. This is what Hibernate means by SQL "dialects."

Back on the server, you can fire up the MySQL client again, and confirm that the Track mapping schema has been created.

It's not surprising to find the table empty. We'll investigate how to populate it with data in the first part of Chapter 3.

If you've followed this example and set up a MySQL database, and you'd prefer to continue working with it throughout the rest of the book, feel free to do so, but bear in mind you'll need to know how to look at the results of the examples yourself. The text will assume you're still working with HSQLDB, and it will show you how to check your progress in that context. You will also see slight differences in the schema, as databases all have slightly different column types and features. Apart from these minor details, it really makes no difference what database you're using — that's part of the appeal of an O/R mapping layer like Hibernate.

If you do want to switch back to HSQLDB for the ease of following the discussion, change your hibernate.properties back to the values shown in Example 2-4.

What about...

. . . Connecting to Oracle, or another favorite, shared, or legacy database that doesn't happen to be MySQL or HSQLDB? You've probably figured out that it's just as easy. All you need to do is change the hibernate.dialect setting in your hibernate.properties to reflect the kind of database you want to use. There are many dialects available, covering every free and commercial database I can think of. These built-in dialects are listed in Appendix C. If you need to work with a more obscure database, you may have to write your own dialect to support it, but that seems unlikely (and check to see if anyone's already started that effort).

Once you've chosen the dialect, you'll also need to set the hibernate.connection properties (driver, URL, username, and password — the other entries in Example 2-4 and Example 2-9) to the proper values for establishing a JDBC connection to your chosen database environment. If you're porting an existing project to use Hibernate, you'll be able to obtain these from the code or configuration of that project. And, naturally, you'll need to put the database's JDBC driver into your project's library directory.