Skip to content

testing.md

Latest commit

 

History

History
54 lines (40 loc) · 4.02 KB

testing.md

File metadata and controls

54 lines (40 loc) · 4.02 KB
Raw

Testing

The suite is ~700 tests across the workspace, run by CI on Linux, macOS, and Windows. This page documents how it's organized, the conventions to follow when adding tests, and the known coverage gaps.

Running

./scripts/check.sh        # fmt + clippy (-D warnings) + tests, mirrors CI
cargo test                # all tests
cargo test -p void-core   # one crate
cargo test --locked       # exactly what CI runs

CI (.github/workflows/ci.yml) also runs, on every push and PR:

  • fmt — cargo fmt --all --check
  • check matrix — clippy -D warnings + tests on ubuntu / windows / macOS, all --locked
  • msrv — cargo check on the declared MSRV (pinned in Cargo.toml rust-version); fails when a dependency raises the real floor, signalling a bump is needed
  • deny — cargo deny check (license allow-list + RUSTSEC advisories)
  • coverage — cargo llvm-cov → Codecov, non-blocking

Layout

Tests are inline #[cfg(test)] mod modules next to the code (so they can reach private items), except void-cli binary tests which live in crates/void-cli/tests/ as integration tests. Shared DB seed fixtures (make_conversation, make_message, …) live in void_core::test_fixtures (feature test-fixtures).

Area Where What
Binary CLI contract void-cli/tests/cli_contract.rs every command's --help exits 0; required-arg violations exit non-zero
Read paths void-cli/tests/read_paths.rs seeds an on-disk void.db in a tempdir, runs inbox/search/messages/… asserting seeded content
Read-path JSON snapshots void-cli/tests/read_paths_snapshots.rs insta snapshots of inbox / conversations JSON envelopes (layout regressions)
First run void-cli/tests/first_run.rs empty store / missing config never panics; doctor --non-interactive exits cleanly
Sync engine void-core/src/sync/ mock Connector drives orchestration, failure isolation, cancellation, LOCK release
Database void-core/src/db/ FTS5 search (incl. proptest fuzzing), bulk_archive_before, schema snapshot + migration data-preservation, dedup, mute
Hooks void-core/src/hooks/ trigger matching, cron scheduling, active windows, placeholders, and execute_hook against a stub agent binary
Remote store void-core/src/store/ fake ssh/scp on PATH verify argv, staging order, error surfacing; cache TTL
Config void-core/src/config/ legacy [[accounts]] migration, unknown-type errors
Connectors each void-* crate API-response parsing happy + error paths (401/429/5xx/malformed) over wiremock; message/media extraction

Conventions

  • Determinism: no real network, no wall-clock (Utc::now()) in assertions — inject fixed chrono instants. No real user filesystem — use tempfile::tempdir().
  • Mock Connector: an in-crate test double implementing the async Connector trait, recording calls via Arc<Mutex<…>>/atomics with configurable behavior (succeed / fail / block-until-cancelled). See void-core/src/sync/tests.rs.
  • Stub agent (hooks): a shell script written to a tempdir emitting canned Claude-style stream-json, gated #[cfg(unix)].
  • Fake ssh/scp (remote store): scripts on a prepended PATH, gated #[cfg(unix)], serialized on a mutex since PATH is process-global.
  • HTTP connectors: wiremock::MockServer via each client's with_base_url(...) test constructor (including Hacker News HnClient::with_base_url). For a 429 retry test, set Retry-After: 0 so retries exhaust without sleeping.
  • No #[ignore]: a test that can't run is removed or #[cfg]-gated, not left ignored.

Known coverage gaps

Honest list of what is not covered and why — good first contributions:

  • Telegram/WhatsApp live sync (start_sync, authenticate): still require a live MTProto/WhatsApp session; health_check and pure extract/send helpers are now unit-tested, but full orchestration is not driven end-to-end without a real client.

When you close one of these, update this section.