# Testing without a tinySA The library exposes configuration, protocol, transport, acquisition and CSV boundaries independently of the terminal. Tests use small synthetic byte transcripts rather than opening a serial device. They are regression checks, not proof of behavior on a particular model or firmware. ```console cargo test --locked --all-targets --all-features cargo test --locked --doc ``` | Boundary | Regression cases | | --- | --- | | Configuration | Embedded defaults, legacy flat keys, partial nested overrides, unknown/malformed values, numeric and path bounds, round trips, and failed replacement preservation. | | Protocol | Basic mode boundaries, point/repeat validation, truncated/extra records, alignment, closing brace, binary delimiter bytes, little-endian levels, offset and both frequency-axis conventions. | | Transport | One-byte/fragmented reads, multiple responses in one read, retained bytes after prompts, partial writes, invalid command lines, EOF, timeout, response size limits and raw-body framing. | | Acquisition | Exact command ordering, optional legacy pre-scan, progress completion, successful decoding, resume after malformed frames, and preservation of acquisition plus recovery errors. | | UI/startup | Missing/stale ports, numeric overflow and clamping, configured baud rates, navigation, one-time result transfer, every screen on small/headless terminals, and invalid config before terminal startup. | | Export | Header/units, deterministic CSV values, propagated write failures and collision-safe output files. | `Shell` accepts `T: Read + Write`. To add a regression, provide a transcript reader and capture its writes; vary read chunk size so a token may cross any read boundary. Test both expected output and error behavior. Keep command side effects outside logging macros because disabled log levels do not evaluate format arguments. `decode_frame` receives exactly the framed body (`{`, records, `}`), without shell echo or prompt. Use raw bytes in tests: converting binary through UTF-8 can destroy the failure being reproduced. Check every truncated prefix when a parser changes, and include payload bytes that equal framing characters. `write_csv` accepts a writer; use an in-memory vector or a writer that fails partway through. Filesystem tests use separate temporary directories and clean up after themselves. They must never write shared public assets. Run the coverage alias and badge tests as described in [security and CI](security-and-ci.md). Keep the minimum tied to an actual measurement over production code. Report test counts and hardware limitations alongside percentages; a high line percentage does not test RF correctness. Hardware validation still requires a known model/version, safe RF input, and a bounded capture of the real command/response sequence. In particular, verify frequency-axis policy, the legacy preparatory scan, zero-offset acceptance and resume behavior against the connected firmware before changing their defaults. ## Library feature matrix and mock integration ```console cargo test --locked --no-default-features --lib cargo check --locked --no-default-features --features serial --lib cargo test --locked --no-default-features --features mock --lib --test mock_device ``` The mock suite includes real loopback sockets, so the test environment must permit localhost listeners. Tests use ephemeral ports, bounded socket timeouts and fresh sessions. See [mock scope and faults](mock-device.md). No mock result certifies device command compatibility or RF accuracy. Full hardware acceptance is tracked in the [roadmap](roadmap.md). Standalone example CLI and socket smoke test: ```console cargo build --locked --no-default-features --features mock --example mock-tinysa-server python3 scripts/test-mock-server.py ``` With a custom `CARGO_TARGET_DIR`, pass its `debug/examples/mock-tinysa-server` executable path to the script (append `.exe` on Windows). This checks argument validation, dynamic-port binding, identity, binary scans and `--once` shutdown. Local Linux validation on 2026-10-04 passed; Windows/macOS runtime, actual hardware and Jenkins execution remain separate unverified acceptance checks.