clippy-ci-parity

Clippy / CI Parity

CI treats clippy warnings as errors (-D warnings), so any lint the local workflow misses becomes a hard build failure on push. Plain cargo clippy only warns, which lets drift slip through.

Run Locally what CI Runs

cargo clippy --all-targets -- -D warnings
cargo fmt -- --check
cargo test

--all-targets covers lib, bins, tests, and examples — matching CI's surface area. Without it, test-only code (which is most of tests/) goes unchecked.

Lints that Have Bitten Us

  • collapsible_match — match X { Variant => { if cond { … } } } must be written as Variant if cond => { … } when there's a catchall arm. Hit in src/tui/form.rs callout title edit handler (4 arms: Backspace, Left, Right, Char(c) each wrapping an if around the body).
  • unused_mut across cfg boundaries — a let mut x is fine on Windows (where a #[cfg(target_os = "windows")] block mutates it) but flags as unused on Linux CI. Scope the allow to the platforms where it's genuinely unused: #[cfg_attr(not(target_os = "windows"), allow(unused_mut))]. Don't blanket-allow. Hit in src/init.rs::detect_obsidian_vaults — local Windows build was happy, Linux CI was not.

Rule of Thumb

If rustc is happy but clippy isn't, CI will reject the push. Pre-push check should be the three-command block above, not just cargo build.

Test Flakes from Shared Mutexes

Tests that mutate process-wide state (env vars, CWD, etc.) commonly serialize via a static Mutex. Two gotchas the suite has hit:

  1. Every test touching the shared state must acquire the mutex, not just the ones clustered near its definition. Missing just one creates a race that Ubuntu's scheduler may hide but Windows' will expose. Hit when load_with_pour_config_env_var_nonexistent_file and load_from_pour_config_env_var mutated POUR_CONFIG without SECRETS_ENV_LOCK, racing the guarded secrets tests.
  2. Prefer .lock().unwrap_or_else(|e| e.into_inner()) over .lock().unwrap(). A plain unwrap panics on a poisoned mutex — which happens any time another test panics mid-critical-section — so one genuine failure cascades into N false PoisonError failures, burying the real signal. Using into_inner keeps the real failure visible and lets the rest of the suite run.

Integration test isolation note: each file under tests/ compiles to a separate binary and runs as its own process, so env vars don't leak across files. Within a single file, though, tests share a process and run in parallel threads.