Represent source code in Java for analysis

[mkshiblu]

To perform refactoring detection, it is necessary to represent the source code in a preferred format.

• One way to do it is to present the source code using a composite pattern where each node represents a code element

• The technical complexity includes parsing the source files using Java with the help of J2V8 plugin from eclipse. Though it's documentation is a bit dated, it's actively being maintained and more performant than other Js Java engine. This bridge between the js code adds additional overhead such as proper releasing of memory etc.

Alternatively, we could parse the source code in JS script and store them in a database. Then these tokens could be represented in Java. This also would allow us to cache the results since commits are permanent of nature.

[mkshiblu]

A seemingly interesting idea could be representing the code elements such as functions, identifiers, etc as nodes in graphs. The program structure will probably not be composite in such a case.

Possible Pros of graph approach:

• Storing of parsed code elements in graph DB (Such as neo4j or in-memory like TinkerPop)
• Good graph traversal support from google guava graph library
• It should be much faster than a trivial composite structure or simple comparison?
• Should allow scaling.

Cons:

• Yet to evaluate how this will affect the Diffing of previous and current program versions to detect refactoring. Theoretically, using graph DB or graph structure should improve the retrieval of entries with many to many relationships
• Need more R&D and effort on deciding which graph DB to choose from, integrating libraries.

[mkshiblu]

Due to more flexibility and theoretically increase customized performance, It's decided not to go with any databases rather creating own graph-like structure.