1. Automate everything.
You’re in charge, make the computer do the work. Ant can insert configuration into properties files, run SQL scripts, change file permissions, upload to FTP sites, and lots more. RTFM and make sure you are aware of everything that Ant can do for you.
2. Make builds Atomic.
Once you’ve automated everything, make sure you can run it all with a single command.Most software engineering endeavours are part of the eternal struggle to limit the consequences of human incompetence. Unlike a computer, the human mind is not well suited to performing a long sequenece of instructions without error. People get distracted, they forget things. My own experience tells me that the longest sequence of actions that a human can reliably reproduce consists of at most one step. Anything more than that and there is a tendency to miss steps out or do things in the wrong order. A slow build is much more bearable if it is atomic because you don’t need to get involved. You can leave it to complete while you do other important stuff.
3. Make it right first, then make it fast.
A build that is fast but wrong is not fast at all; it’s just wrong. A target that does not correctly ensure that its dependencies are rebuilt (when appropriate) will save you 3 seconds on each of the 30 rebuilds that it takes to debug the problem.A useful tip for improving the runtime of slow builds is to use the ANT_OPTS environment variable to tweak the JVM settings used by Ant. For long-running builds you may find that the -server switch provides a noticeable improvement. Alternatively, increasing the heap size (via the -Xmx switch) may help.
4. If you use Ant, use only Ant.
The build.xml is the one true build script. Don’t disrespect the build.xml by using your IDE’s build process in its place. Even if you are really diligent about configuring your IDE, you are very likely to end up with a slightly different build process. Better still, when your colleague checks in a change to build.xml, he or she won’t tell you and your build will be broken. If you’re lucky this will become obvious sometime after you start complaining loudly about others checking in broken code. In the worst case your build will be subtly broken and you won’t notice. Avoid the contempt of your co-workers and make your IDE use Ant. Don’t let the old UNIX hacker in the corner use make just because he doesn’t like XML (he doesn’t like Java either, he’s just doing it because the company hasn’t had any C work in the last 6 months 10 years). If he uses make, you all have to use make (good luck), since there can be only one true build script.
5. Follow conventions.
Ant expects your build file to be called build.xml (unless you tell it otherwise). Developers expect the build file to be in the root of the project. Other than the understandable desire to confuse and irritate the rest of your team, including the lucky souls that get to maintain your code long after you get head-hunted by Google, there’s no reason not to follow these conventions. Another less well-known convention is to prefix all “internal” targets (those that should not be called directly) with a hyphen. This is suggested in the Ant manual. It has the advantage that it is not possible to invoke targets that follow this naming scheme from the command line.
6. Provide a clean-up target.
It is usual for makefiles and Ant scripts to provide a target called “clean” that removes all generated files and returns the project to its initial state. Without this functionality failed builds can be a real problem. Providing a clean target is made easier if source files (those that are under version control) and generated files (those that are derived from versioned files) are kept separate.
7. Compile test code at the same time as production code.
Test code, by definition, depends on production code (if it didn’t, your test coverage scores would not be very impressive). Therefore, changes to production code can potentially break test code. For this reason it is a good idea to enforce the rule that test classes are always compiled at the same time as production classes (you don’t have to run the tests at this point). Since you are firing up the compiler anyway, it won’t add too much overhead and it avoids breaking windows.
8. Make builds self-contained.
A build that has external dependencies is a build that is difficult to to configure. When your hard drive fails or a new developer joins the team, you want to be able to configure a new machine for building ASAP. If your build depends on tools like Checkstyle or TestNG, add these to the project repository. This makes your build simpler because you know exactly where to find the files and you don’t have to worry about version mis-matches. Other than Ant itself and the JDK, external dependencies should be eliminated wherever possible (Maven has a different approach to these kind of dependencies).
9. Parameterise essential configuration.
It’s not always practicable to eliminate all external dependencies. For example, you may need to configure a database connection as part of your build. Don’t hard-code this kind of configuration in the build.xml, use Ant’s support for properties files to allow individual developers to set configuration parameters. Don’t complicate matters by making everything configurable. Only parameterise those settings that really need to be configurable.
10. Provide sensible defaults.
Where parameters are required for building, you can make things much simpler by providing sensible default values rather than leaving everything to be configured by the individual developer. For example, if one of the parameters is a JDBC URL, a sensible default might be to point to localhost (most developers will probably be developing against their local database). If one of the required properties is the file system path to a particular tool, provide the default installation directory. By providing sensible defaults you can eliminate most of the configuration hassles.
11. Visualise dependencies.
The build script for a large project can have complex, often ad-hoc dependencies. Tools such as yWorks Ant Explorer can provide insight into dependencies by providing a graphical representation of the structure of the build. If the graph looks like one of these, consider refactoring.
12. Constrain classpaths.
Avoid the temptation to use a single global classpath for compilation. Always ensure that code is built with a classpath that contains only those classes that will be available to it at runtime. For example, if your build creates an EAR file and a Swing client, make sure that the Swing client is compiled with access only to those libraries that will be deployed with it. Do the same for the EAR file and any other modules. This approach minimises the potential for embarrassing NoClassDefFoundErrors at runtime and also helps to detect inappropriate dependencies.
13. Modularise your build.
Arrange your project into coherent, self-contained modules. For example, you may have a Swing GUI module and web module among others. Layout your project files to support this arrangement. The goal is to make modules as self-contained as possible and to minimise dependencies between them. The following example shows one possible way of arranging your project files:
myproject |__modules | |__gui | | |__lib | | |__src | | |__java | | |__main | | |__test | |__web | |__lib | |__src | |__html | |__java | |__main | |__test |__build.xml |
14. Favour convention over configuration.
If the source code for module1 is in module1/src/java, where would you expect to find the source code for module2? If files are in consistent predictable locations within each module, it’s not necessary to explicitly configure each location. By always following the same conventions about where things go in each module, you minimise complexity and make it easier to implement the next recommendation…
15. Eliminate duplication with macros.
If all of your modules are laid out identically, the common tasks (such as compiling source trees and creating JAR files) are essentially identical for each module. Ant 1.6 introduced macros to enable these operations to be defined just once and applied repeatedly with different parameters. Familiarise yourself with the Macrodef task and consider how it can be applied in your build files. The example macros below are for compiling generic modules. The first macro defines how a single source tree is compiled. Because we have followed the advice from the rest of this article, all we need to provide to run the macro is the name of the module (by convention modules are in directories of the same name, all Java source is in the src/java directory of the module and below this are the main and test source trees). The second macro simplifies things further by combining the two separate calls to the first macro. The main classes are built, then the test classes are built with the main classes added to the compiler’s classpath. Using this macro we can compile an entire module with a single line:
<compilemodule name="mymodule" /> |
<!-- This macro compiles one source tree (i.e. the main source tree or the unit test source tree) of a given module. --> <macrodef name="compiletree"> <attribute name="module"/> <attribute name="tree"/> <element name="treeclasspath" optional="true"/> <sequential> <mkdir dir="./@{module}/build/classes/@{tree}" /> <javac destdir="./@{module}/build/classes/@{tree}" debug="on" deprecation="on" optimize="on" source="1.5" target="1.5" srcdir="./@{module}/src/java/@{tree}"> <classpath> <treeclasspath/> <path refid="base.path"/> </classpath> <compilerarg value="-Xlint:unchecked" /> </javac> </sequential> </macrodef> <!-- This macro compiles all source (including unit tests) for a single module. --> <macrodef name="compilemodule"> <attribute name="name"/> <element name="moduleclasspath" optional="true"/> <sequential> <compiletree module="@{name}" tree="main"> <treeclasspath> <moduleclasspath /> </treeclasspath> </compiletree> <compiletree module="@{name}" tree="test"> <treeclasspath> <!-- Add the main classes to the classpath for unit test compilation. --> <path location="./@{name}/build/classes/main" /> <moduleclasspath /> </treeclasspath> </compiletree> </sequential> </macrodef> |
These macro definitions are somewhat verbose (due to Ant’s XML syntax), but the benefit is that they ensure that all modules are built in the same way. Rules for compilation only have to be defined once. Also, keep in mind that once you have written macros for one project, you can reuse them elsewhere. You may choose to put all of your macros in a separate file and import this wherever it is required.