A semantic tree represents source-code constructs and their relationships so data-preparation tooling can preserve useful context without reducing a codebase to isolated lines. section 12.2.1
A file-processing pipeline can use JavaParser to ingest each file, build the tree, and query class declarations, variables, methods, and containing code for a fine-tuning dataset. section 12.2.1
Extracting contextual code data
Rather than storing only a method line, a dataset can retain the surrounding class declaration, declared variables, method, and containing code. This preserves context while avoiding excessive granularity. section 12.2.1
