write-tests

# Write Tests Test fails when behavior breaks, passes when it doesn't — nothing more. Run at end, report honestly. ## Step 1 — Match repo style - Tests live in-file: `#[cfg(test)] mod tests` at bottom, `use super::*`. - Builder helpers for fat structs (see `dev()` in `src/usb.rs` tests) — reuse existing helper, extend it, don't build parallel one. - Plain `assert_eq!`/`assert!`. No assertion crates, no mock frameworks. - Raw descriptor bytes: inline arrays with `#[rustfmt::skip]` + comment per chunk (existing pattern). ## Step 2 — Choose level Cheapest level that catches the regression: - **Unit `#[test]`** — pure logic: tree flatten/fold, parsing (usb.ids, sysfs values, descriptors), diff. Default. No hardware, no I/O. - **Smoke** — `./target/release/usbtree --demo --dump`: exercises scan→tree→render path, no hardware. For wiring bugs, not logic bugs. - **TUI in pty** — `--demo` in pty for interaction (keys, fold, filter). Manual/scripted, last resort — most TUI logic is testable as pure functions on App state; prefer extracting that. Pure parse bug tested through pty = wrong. Keybinding wiring tested as unit = also wrong. Match level to where bug lives. ## Step 3 — Write - **Behavior, not implementation.** Assert outputs and state, not internals. - **Name by behavior**: `max_power_parses_sysfs`, `filter_finds_devices_in_collapsed_hubs` — not `test1`. - **Independent tests**: own data, no ordering deps. No shared mutable state — tests run parallel by default. - Filesystem cases: `tempfile::TempDir` if already a dep, else skip fs test and test the parse fn on a `&str` instead — push I/O to edge, test the pure part. - Cover happy path, interesting failures (malformed input, missing sysfs file, truncated descriptor), boundary values. Skip permutations hitting same branch. - Don't test nusb/ratatui — test your use of them. ## Step 4 — Run and report - `cargo test name_fragment` first, then full `cargo test`. - Report pass/fail, verbatim output for red. - Test caught real bug in code under test → fix code (or flag), never the test. - Never weaken assertion or skip to get green. Never report green without running.