Skip to content

CI.md

Latest commit

 

History

History
99 lines (77 loc) · 3.69 KB

CI.md

File metadata and controls

99 lines (77 loc) · 3.69 KB

CI

Running CI locally with act

The repo's .github/workflows/ci.yml runs the host crate's fmt + clippy + tests plus dedicated jobs for the CAS cache and compose-protocol unit tests. While the repo is still private and GitHub-hosted runners aren't available, the same workflows run under nektos/act:

brew install act          # or follow nektos/act's install docs
act -W .github/workflows/ci.yml

act reads .actrc for our defaults: the catthehacker/ubuntu: act-latest medium image, arm64 container architecture (matches Apple Silicon hosts; flip to amd64 if you need parity with GitHub's amd64 runners), and no docker-in-docker socket forwarding.

Running a single job

act -W .github/workflows/ci.yml -j host-checks
act -W .github/workflows/ci.yml -j cache-tests
act -W .github/workflows/ci.yml -j compose-tests

Skipping the docker pull

The first run pulls ~700 MB of catthehacker/ubuntu:act-latest. Cache it locally:

docker pull catthehacker/ubuntu:act-latest

What's in CI today

Job What Why
host-checks cargo fmt --check, cargo clippy -D warnings, cargo test --all-features in host/ Catch formatting drift; clippy keeps the lint surface clean; tests cover the dispatch + SPI + compose plumbing
cache-tests cargo test --lib cache 6 CAS unit tests (blake3 + sha-256 dual-write, lookup roundtrip, purge, etc.)
compose-tests cargo test --lib compose_provider 4 protocol unit tests (manifest, query-scalar, query rows, prepare/step/finalize)

What's NOT in CI yet (deliberate)

  • WASM-side builds. Building cli, fiji-hello, and the extensions in sqlite-wasm-loader requires wasi-sdk (~150 MB) plus cargo-component. The .github/workflows/ci.yml has a commented-out wasm-builds job placeholder where this goes.

    Until that lands, the host's run_resolves_sqlite_runtime smoke test skips gracefully when fiji_hello.wasm (out-of-tree submodule artifact) is missing (prints "skipping: fiji_hello.wasm not built" and exits 0). Note: the test is currently #[ignore]'d pending the submodule rebuild against the renamed WIT world.

  • The C-side legacy build. make cli-demo-test builds sqlite-cli-demo.wasm (the legacy C cli). It needs the vendored wasi-sdk plus wit-bindgen + wac-tools. Same argument as above — adds a chunk of toolchain setup; not yet the priority.

  • cargo audit. Worth adding once we lock the deps; some of the wasmtime transitive deps have RustSec advisories that need explicit accept-or-fix per-version. Track separately when we publish.

  • Integration tests against the loader. The loader's tests live in ~/git/sqlite-wasm-loader/ — a separate repo. Adding cross-repo CI is out of scope until both repos are public and can use actions/checkout against each other.

Why the .cargo/config.toml override matters

The repo's .cargo/config.toml pins build.target = "wasm32-wasip2" because the legacy C-side build expects it. Native cargo runs need the override. The workflow sets CARGO_BUILD_TARGET="" in the job env to clear the pin; you may also need --target=x86_64-unknown-linux-gnu (or aarch64) explicitly on some workflows. Local cargo test --target aarch64-apple-darwin is the macOS equivalent.

Going public

When the repo flips public:

  1. .actrc stays — it's harmless config that helps act users.
  2. The workflows file works against GitHub's hosted runners unchanged (we use runs-on: ubuntu-latest not the catthehacker image directly).
  3. Add the wasm-build job (see the commented placeholder).
  4. Add cargo audit and pin policy.
  5. Wire branch protection so PRs require these jobs green before merge.