Testing

Baudrun ships two layers of automated tests plus a manual playbook for the parts that need a human at the keyboard.

Layer	What it covers	Where
Unit tests	Pure-data invariants: JSON round-trips, parser edge cases, protocol checksums / CRCs, hex input validation, highlight runtime.	cargo test in the repo root.
Wire-level transfer tests	Real XMODEM / YMODEM / Send-Hex code paths against a baud-paced virtual pty pair. Byte-diff against deterministic fixtures. Unix-only.	scripts/transfer-tests/
Manual playbook	Things that need eyes on the UI: import dialogs, paste safety prompts, scrollback retention, transfer cancel UX.	scripts/virtual-serial/TESTING.md in the source tree.

---

Unit tests

Standard Cargo. Runs on macOS, Windows, and Linux in CI on every push.

cargo test

The tests live alongside the code they exercise (#[cfg(test)] mod tests { ... } at the bottom of each module). No external fixtures, no I/O, no temp files. Fast — the whole suite finishes in seconds.

---

Wire-level transfer tests

Drive Baudrun’s actual send paths against a virtual serial bridge and byte-diff the output. The harness exists because the production app is the wrong shape for protocol regression testing — gpui UI, profile picker, port enumeration — but the protocol code itself (under src/data/transfer.rs) is fully decoupled.

The harness lifts transfer.rs verbatim via #[path] so it exercises the same XMODEM / YMODEM state machines the app ships. Refactors there are picked up automatically at the harness’s next build — there’s no parallel copy to keep in sync.

Setup (one-time)

# Build the bridge and the harness.
(cd scripts/virtual-serial && cargo build --release)
(cd scripts/transfer-tests && cargo build --release)

# Generate the deterministic fixture set under test/transfers/.
./scripts/transfer-tests/regen-fixtures.sh

You’ll also need lrzsz for the XMODEM / YMODEM receiver subprocess:

brew install lrzsz                # macOS — installs lrb / lrx
sudo apt install lrzsz            # Debian / Ubuntu — installs rb / rx

The harness searches both name conventions.

Running

./scripts/transfer-tests/target/release/transfer-tests              # all 11 cases
./scripts/transfer-tests/target/release/transfer-tests --quick      # skip 1 MiB T9
./scripts/transfer-tests/target/release/transfer-tests --tests hex  # subset

Results

Last verified 2026-05-15 on macOS arm64:

Test	What	Baud	Wall	Result
T1	hex ASCII (Hello)	9 600	0.01 s	ok
T3	hex binary / non-printable	9 600	0.01 s	ok
T5	hex 1 KiB random	9 600	1.35 s	ok
T11	YMODEM 512 B over slow link	9 600	5.92 s	ok
T6	YMODEM 4 KiB	115 200	3.69 s	ok
T9	YMODEM 1 MiB	115 200	126.00 s	ok
T7c	XMODEM classic (128 B / checksum)	115 200	1.56 s	ok
T7C	XMODEM-CRC (128 B / CRC-16)	115 200	1.56 s	ok
T7k	XMODEM-1K (1024 B / CRC-16)	115 200	1.55 s	ok
T8	XMODEM single-block (SUB padding)	115 200	1.08 s	ok
T10	YMODEM cancel mid-transfer	115 200	1.18 s	ok

11 / 11 passed, 143.9 s wall. T9 dominates the run time. The other ten cases finish in well under 18 seconds combined.

What each test ID exercises

The IDs (T1 … T11) match the section numbering in scripts/virtual-serial/TESTING.md, which spells out the manual version of each test. T2 (hex input format equivalence) and T4 (hex input validation) are UI-layer concerns and aren’t automated — parse_hex_string is unit-tested separately if that coverage matters to you.

ID	Category	Notes
T1	hex round-trip	”Hello” — printable ASCII baseline.
T3	hex round-trip	00 01 02 ff fe 7f — control bytes + 0xff + DEL.
T5	hex round-trip	1 KiB random — exercises the rate-limited write path at the 9600-baud floor.
T6	YMODEM	4 KiB at 115200 — the “happy path” file transfer.
T9	YMODEM	1 MiB at 115200 — many-block stress, retry-loop sensitivity.
T11	YMODEM	512 B at 9600 — slow-link timing margins, lrzsz’s own per-block timeouts.
T7c	XMODEM classic	4 KiB with 128-byte blocks + 8-bit checksum. Receiver-initiated with NAK.
T7C	XMODEM-CRC	4 KiB with 128-byte blocks + CRC-16. Receiver-initiated with 'C'.
T7k	XMODEM-1K	4 KiB with 1024-byte blocks + CRC-16. Header byte (STX vs SOH) signals block size.
T8	XMODEM single-block	6 B ("hello\n") — exercises the SUB-padding path for the last block.
T10	cancel mid-transfer	1 MiB YMODEM with cancel tripped 150 ms in. Asserts TransferError::Cancelled, verifies the bridge survives.

Why this layer exists

Unit tests catch invariant violations (the checksum of a known block must be 0x5A); the wire layer catches integration regressions (the protocol state machine actually drives the state-machine partner on the other side of the wire to completion). Both layers find things the other misses. The manual playbook then covers UI-level concerns no harness can sanely automate — does the cancel button actually look pressed during a 1 MiB transfer, does the import-pack dialog give a useful error on malformed JSON.

For the full design — why no threads in the read path, why #[path] over a library extraction, why deterministic seed — see the harness README in the source tree.