JoeJoeflyn8/6/2026, 5:33:02 PM

- Replace ts-morph with oxc-parser for TS/JS (5x faster parse)
- Skip AST walk during indexing (parse only for chunking, not graph)
- Move FTS insert after main transaction with triggers down
- Truncate FTS search_text to 400 chars
- page_size=16384 at DB creation (fewer page splits)
- journal_mode=MEMORY + locking_mode=EXCLUSIVE during bulk write
- mmap_size=512MB for larger memory-mapped I/O
- Move setOptimizedWriteMode(false) after finalize to avoid
  mid-finalize fsync flushing 70MB of dirty pages
- Use PASSIVE checkpoint to avoid blocking fsync on small repos
- Migrate TEXT PKs to INTEGER PRIMARY KEY (auto-increment)
- Use RETURNING id for graph node batch insert (fixes empty graph bug)
- Add C to graph builder language filter
- Add TS/JS graph building via OxcParser with call/import extraction
- Fix function pointer typedef name extraction (compare_fn_t not a)
- Skip struct/union/enum type references without body (no false symbols)
- Keep qualified call names (obj->method instead of method)
- Remove symbol/call caps in graph builder (was 30 symbols, 20 calls)
- Fix OxcParser symbol extraction for export wrappers and variable decls
- Add import edge resolution (file-to-file graph edges)
- Switch hashContent from SHA-256 to Bun.hash (xxHash64, 2x faster)
- Linux kernel: 88s -> 23.9s (3.7x faster)

Parentsdfed44618a688d