Local-Search / tests

Purpose

Public CLI regression tests. Exercise real binaries against isolated local state, fake browser endpoints, and fake registries rather than personal browser data.

Mental model & key files

• cli.rs checks help, welcome text, output modes, and structured errors.
• connection.rs checks explicit browser selection, background work tabs, disconnection, and cleanup boundaries with local fake CDP endpoints.
• updates.rs checks update JSON, aliases, package-manager detection, opt-out, timeout, and macOS interactive startup via /usr/bin/script.

Patterns to follow / invariants

• Set LOCAL_SEARCH_CONFIG_DIR to a test-owned temporary directory.
• Never connect to a real signed-in browser or fetch a registry in unit/integration tests.
• Keep stdout contracts parseable, and assert stderr and exit status separately.
• Use subprocess-specific environment overrides, not global environment mutations.
• Bound waits and shut down fake listeners and child processes at test completion.
• Serialize update-check subprocess/PTY fixtures because their real two-second deadlines must not compete with sibling cold process launches on loaded hosts.

Common tasks → first action

• Run cargo test --test updates for release-check changes.
• Run cargo test --test connection for browser authority and lifecycle changes.
• Finish with cargo fmt --check, cargo test, and cargo clippy --all-targets --all-features -- -D warnings.

Gotchas

• The npm wrapper has its own version; fixtures must not equate it with the native crate.
• Interactive update checks require both output streams to be terminals. Pipes intentionally do not exercise that path.
• Unix fake executables need execute permission. PTY assertions are macOS-only; pure update/cache tests also run on other platforms.