InterchangeDB

Docs: start at docs/README.md (documentation map) and CONTEXT.md (domain glossary).

Development

Inner-loop commands:

cargo check --lib      # fastest signal on src/ edits
cargo test --lib       # unit tests only, no integration binaries
cargo clippy --lib     # lint without building benches
cargo test --test <harness> <filter>   # one integration suite

Full cargo test --workspace, cargo test --workspace --release, and cargo clippy --workspace --all-targets -- -D warnings are the pre-commit gates (mirrored in CI) — run them before pushing, not in the tight loop.

Test tiers

Tests run in three named tiers:

• Default tier — in-memory backends (MemoryDiskManager, WAL SyncMode::NoSync): fast on any machine; the subject of these suites is never the disk. New tests belong here unless the disk is their subject.
• Durability tier — real files, real fsync (Durable): every suite whose subject is crash durability, recovery, persistence, or fsync semantics. Enumerated in the plan doc; a test with reopen/crash semantics must live here.
• Conformance matrix — testkit's for_each_*! registries run every backend (including FileDiskManager) against the shared contracts, which is what makes the default tier's memory backends a zero-coverage-loss choice.

If your editor runs rust-analyzer, give it its own target directory so it doesn't fight terminal builds for the target/ lock (VS Code: "rust-analyzer.cargo.targetDir": true; Neovim lspconfig: settings = { ["rust-analyzer"] = { cargo = { targetDir = true } } }). Symptom of the fight: a build that sits 10–30 s doing nothing before it starts.