Lossless semantic trees
A Lossless Semantic Tree (LST) is a tree representation of code. Unlike the traditional Abstract Syntax Tree (AST), Moderne's LST offers a unique set of characteristics that make it suitable for both single-repository and mass transformation operations:
- Type-attributed. Each LST element is imbued with type information. For example, when referencing a field, the source code may just refer to it as
myField. The OpenRewrite LST element formyField, on the other hand, would contain additional information about what the type ofmyFieldis, even if it isn't defined in the same source file or even the same project. - Format-preserving. Whitespace before and after LST elements is preserved in the tree so the tree can be printed out to reconstitute the original source code without clobbering formatting. Additionally, refactoring operations that insert code are sensitive to the local style of the code around them and match the local style.
- Acyclic and serializable. Most LSTs containing type information are potentially cyclic. Cycles usually come from generic type signatures like
class A<T extends A<T>>. This kind of pattern is generally found in things like abstract builder types in Java. OpenRewrite cuts these cycles off and adds serialization annotations to its types so the LST can be serialized/deserialized with libraries like Jackson.
All of these properties are necessary for scalable organization-wide search and transformation.
Format preservation is necessary in any organization that doesn't have 100% consistent formatting across its whole codebase.
Type attribution is necessary for the accurate matching of patterns. For example, if we are looking for SLF4J log statements and we see a statement like the following, without type attribution how do we know if
logger is an SLF4J or a Logback logger?logger.info("Hi");
The production of type-attributed LSTs for a whole organization is arbitrarily computationally complex since it requires dependency resolution, parsing of the source code, and type attribution (basically Java compilation up to the point of bytecode generation). Since OpenRewrite LSTs are serializable, we can store them as a byproduct of compilation and then operate on them en masse later.
Moderne's build plugins allow for LSTs to be serialized to disk. This makes the process of consuming and editing large quantities of them much more efficient. OpenRewrite's build plugins, on the other hand, store everything in memory and need to be reparsed every time there is a change.
Once we have a serialized LST for a particular source file, and since it also contains type information, it can be refactored/searched completely independently of other source files in the same source package or repository. This makes mass search and refactoring an approximately linearly scalable operation.
Last modified 6mo ago