Testing and evidence

All commands run at repository root. A passing host test is evidence for that layer; it does not satisfy a real-hardware criterion in the acceptance matrix. Record exact commit, source cleanliness, command, environment, results, skips and artifact hashes.

Layer	Command	What it establishes
Lint/compatibility	make lint	Python, native formatting, SPDX and MicroPython restrictions
Host	make test-host	protocol logic, vectors, bounds, storage model
Coverage	make coverage	branch coverage for configured host sources; UI/glue exclusions remain
Portable C	make test-c	CTest with ASan/UBSan
MCU cross compile	tools/ci/arm-cross-check.sh	portable libraries compile for Cortex-M4/M33 with warnings as errors
MicroPython	make test-upy	actual emulator runtime, native module bindings and norcow tests
Functional	make test-functional	UDP transports and debuglink flows through emulator
Fuzz	FUZZ_SECONDS=60 make fuzz	bounded smoke campaign, not proof of parser correctness
PC/SC interop	commands below	real host tools over virtual reader
Hardware	R01–R16	physical USB, OLED/camera, RAM, timing, flash and OS interoperability

make test runs host, C and functional suites, not every row above. CI runs additional ARM/fuzz checks and scheduled reproducibility. make lint does not itself run shellcheck; CI separately runs shellcheck tools/*.sh tools/ci/*.sh.

Both emulator models

make emu MODEL=T3B1
uv run pytest -q tests/functional --junitxml=build/functional-T3B1.xml
make emu MODEL=T2B1
FEURISTIC_EMU="$PWD/upstream/trezor-firmware/core/build-xtask/artifacts/T2B1/firmware-emu" \
  uv run pytest -q tests/functional --junitxml=build/functional-T2B1.xml

The harness starts separate processes/profiles and allocates ports in steps of 16 from FEURISTIC_EMU_BASE_PORT (default 22000). Use a distinct base range for concurrent pytest invocations. A missing emulator causes skips: inspect the summary and require actual executions. The harness uses a 20 MiB heap; measure memory on each MCU separately.

Targeted regressions:

uv run pytest -q tests/host/fido2/test_capacity_budget.py tests/host/core/test_storage_budget.py
uv run pytest -q tests/host/backup tests/host/importer
uv run pytest -q tests/functional/test_import.py tests/functional/test_timesync.py
uv run pytest -q tests/functional/test_upy.py

Storage details: storage-budget.md. Native corpus, fuzzer and benchmark details: fqr, fimg, fmsc, RSA and fuzz targets. Power-cut injection in a model and orderly emulator shutdown cannot replace interrupting real flash writes.

Linux interoperability

Use a dedicated Linux test host. Install packages and start PC/SC using the same sequence as CI:

sudo apt-get update
sudo apt-get install -y pcscd vsmartcard-vpcd opensc gnupg scdaemon
sudo systemctl stop pcscd.socket pcscd.service
sudo pcscd --disable-polkit
FEURISTIC_INTEROP=1 uv run pytest -v tests/functional/interop --junitxml=build/interop.xml

This changes the host PC/SC service state; record/restore its normal service configuration after the run. The fixtures create an isolated GnuPG home and start tools/vpcd_bridge.py; do not also connect a competing manual bridge. The present suite checks GnuPG detection/generation/signing, PIV detection and GET CHALLENGE. It does not constitute the complete R09/R10 OS matrix.

Failure handling

Keep emulator logs from the failing profile, JUnit XML, command and traceback. Minimize hostile inputs and preserve a hash and generation seed. Re-run the failed layer after a fix, then its affected integration tests. Never publish real test secrets, PINs, private camera backups or unredacted APDU transcripts containing credentials. Security reporting: SECURITY.md.