"CSI: Gopher"
summary
Code from thousands of public Go repos is parsed into a language-agnostic syntax tree and queried with UAST extracted values, run over BigQuery exports and source{d} indexes. The pipeline hits scaling limits so analysis targets ~1,000 repos, but it returns concrete claims: empty string is the most common literal (≈1M), init functions average 0.12 but some files contain 352 generated inits, and use of gorilla/context shows memory-leak patterns. Queries and heuristics prove which API additions or migrations are justified.
AI-written summary · about
