mega-duckdb

Collection repo for the DuckDB/DuckLake knowledge base. Every researched upstream repo lands here as a submodule; nothing is vendored by copy.

Layout

• upstream/<repo> — official corpora and candidate integrations, as submodules.

Cloning — do not recurse by default

git clone ssh://git@git.0xdao.app:2222/caoer115/mega-duckdb.git
git -C mega-duckdb submodule update --init upstream/<the-one-you-need>

--recurse-submodules pulls duckdb core at least three times — once at upstream/duckdb, and again inside upstream/duckdb-delta/duckdb and upstream/duckdb-fts/duckdb, because the extension repos vendor it as their own submodule. A full recursive clone is multi-GiB and takes many minutes. Init only the submodule you need.

Rules

• Never borrow objects. Add submodules without --reference, or with --dissociate. A --reference into a scratch dir that is later deleted leaves dangling alternates: git status then fails for every consumer of this repo, and — worse — the failure text masks push results, so a lane reads a successful push as a failure.
• Git operations are serialized by the session mutex. Full path, not the slug: $SESSION_DIR/locks/mega-duckdb-git.mutex.sh — claim <your-ticket-name>, then release <ticket>. A lock nobody can locate is worse than no lock, because seats believe they are protected.
• Never build inside this checkout — build in a worktree or scratch dir.
• No large data artifacts (*.duckdb, corpora dumps) enter this tree. Pointers only. Published corpora live outside any git tree and are referenced by fingerprint.