To search or not to search: it depends on the question
summary
Developers run code search about 12 sessions per day, each session averaging three to four queries, and a study categorizes searches as how (≈33%), what (≈25%), where (≈16%), why (≈16%). A prototype indexes code by tokens/context, behavior via fuzzing, and language‑agnostic AST structure with non‑dominated sorting, yielding very high precision but impractically slow; replacing behavior checks with machine‑learning embeddings restores speed without losing precision and produces a practical code‑to‑code search that answers most how and where questions.
AI-written summary · about
chapters (8)
- 0:00Start
- 0:31Many interfaces.
- 1:49Code search is frequent
- 2:50Four Distinct Needs
- 3:48"How" Example Code
- 6:06Code-to-code Search - Language
- 6:25Code-to-code Search - Structure
- 6:41Code-to-code Search - In Practice
related talks
20mDevX: History of Code Search
08mSupercharging Developer Productivity With Advanced Code Search
18mZero To Hero: Building a Developer-First Search Experience With OpenSearch
29mSearch-First Writing for Developers
30mEnhancing CTI Processes with Code Search Technology
17mSyntactic Code Search with Sequence-to-Tree Matching: Supporting Syntactic Search with(…)
50mReading code
37mBeyond Grep: Semantic Code Search for Safer Change at Scale