The Eclipse AspectJ implementation

The AspectJ Programming Guide describes the AspectJ language. This guide describes the AspectJ tools produced by the AspectJ team on https://eclipse.org/aspectj. The AspectJ tools include - ajc, the compiler/weaver; ajdoc, a documentation tool; Ant support for ajc; and load-time weaving support. These tools are delivered in the library folder of the AspectJ tools installation, mainly in aspectjtools.jar (tools) and aspectjrt.jar (runtime). This guide does not describe the Eclipse AspectJ development tools (AJDT). That is produced by another team (sharing some members) on https://eclipse.org/ajdt. AJDT is delivered as an Eclipse plugin, incorporating the classes in the AspectJ tools libraries along with the Eclipse plugin interface classes.

Since AspectJ 1.1, the tools have implemented the AspectJ language using bytecode weaving, which combines aspects and classes to produce .class files that run in a Java VM. There are other ways to implement the language (e.g., compiler preprocessor, VM support); the AspectJ team has always tried to distinguish the language and the implementation so other groups could build alternative implementations of AspectJ. To that end, The AspectJ Programming Guide, Implementation Notes describes how the Java bytecode form affects language semantics. VM- or source-based implementations may be free of these limits or impose limits of their own, but most should be fairly close to what’s possible in Java bytecode.

Please be careful not to confuse any description of weaving or of this implementation of the AspectJ language with the AspectJ language semantics. If you do, you might find yourself writing code that doesn’t work as expected when you compile or run it on other systems. More importantly, if you think about aspects in terms of weaving or of inserting or merging code, then you can lose many of the design benefits of thinking about an aspect as a single crosscutting module. When the text below introduces an implementation detail, it will warn if users make mistakes by applying it in lieu of the language semantics.

Bytecode weaving, incremental compilation, and memory usage

Bytecode weaving takes classes and aspects in .class form and weaves them together to produce binary-compatible .class files that run in any Java VM and implement the AspectJ semantics. This process supports not only the compiler but also IDE’s. The compiler, given an aspect in source form, produces a binary aspect and runs the weaver. IDE’s can get information about crosscutting in the program by subscribing to information produced by weaver as a side-effect of weaving.

Incremental compilation involves recompiling only what is necessary to bring the binary form of a program up-to-date with the source form in the shortest time possible. Incremental weaving supports this by weaving on a per-class basis. (Some implementations of AOP (including AspectJ 1.0) make use of whole-program analysis that can’t be done in incremental mode.) Weaving per-class means that if the source for a pure Java class is updated, only that class needs to be produced. However, if some crosscutting specification may have been updated, then all code potentially affected by it may need to be woven. The AspectJ tools are getting better at minimizing this effect, but it is to some degree unavoidable due to the crosscutting semantics.

Memory usage can seem higher with AspectJ tools. Some aspects are written to potentially affect many classes, so each class must be checked during the process of weaving. Programmers can minimize this by writing the crosscutting specifications as narrowly as possible while maintaining correctness. (While it may seem like more memory, the proper comparison would with with a Java program that had the same crosscutting, with changes made to each code segment. That would likely require more memory and more time to recompile than the corresponding AspectJ program.)

Classpath, inpath, and aspectpath

AspectJ introduces two new paths for the binary input to the weaver which you’ll find referenced in ajc, the AspectJ compiler/weaver, AspectJ Ant Tasks, and Load-Time Weaving.

As in Java, the classpath is where the AspectJ tools resolve types specified in the program. When running an AspectJ program, the classpath should contain the classes and aspects along with the AspectJ runtime library, aspectjrt.jar.

In AspectJ tools, the aspectpath is where to find binary aspects. Like the classpath, it can include archives (.jar and .zip files) and directories containing .class files in a package layout (since binary aspects are in .class files). These aspects affect other classes in exactly the same way as source-level aspects, but are themselves not affected. When deploying programs, the original aspects must be included on the runtime classpath.

In AspectJ tools, the inpath is where to find binary input - aspects and classes that weave and may be woven. Like the classpath, it can include archives and class directories. Like the aspectpath, it can include aspects that affect other classes and aspects. However, unlike the aspectpath, an aspect on the inpath may itself be affected by aspects, as if the source were all compiled together. When deploying aspects that were put on the inpath, only the woven output should be on the runtime classpath.

Although types in the inpath and the aspectpath need to be resolved by the AspectJ tools, you usually do not need to place them on the classpath because this is done automatically by the compiler/weaver. But when using the WeavingURLClassLoader, your code must explicitly add the aspects to the classpath so they can be resolved (as you’ll see in the sample code and the aj.bat script).

The most common mistake is failing to add aspectjrt.jar to the classpath. Also, when weaving with binary aspects, users forget to deploy the aspect itself along with any classes it requires. A more subtle mistake is putting a binary aspect (BA) on the inpath instead of the aspectpath. In this case the aspect BA might be affected by an aspect, even itself; this can cause the program to fail, e.g., when an aspect uses exclusion to avoid infinite recursion but fails to exclude advice in aspect BA.

The latter is one of many ways that mistakes in the build process can affect aspects that are written poorly. Aspects should never rely on the boundaries of the build specification to narrow the scope of their crosscutting, since the build can be changed without notice to the aspect developer. Careful users might even avoid relying on the implementation scope, to ensure their AspectJ code will run on other implementations.