setup.gate-blocks-controllers asserted on the machine which steps the image's table requires (GET /wiz's missing list) and recommends (each not run yet, never the gate's parts or the optional steps). Both are the setup table's decisions over a record, software with no hardware in them, and forgectrl's setup_test now proves both on the image's own table: the required steps in order (10b), the missing list emptying when they are done (4b), and the recommended steps in order, each not run yet (10b). The catalog keeps what needs the machine: the supervisor holding and spawning the controller, the process, and the Grbl port. - The two missing-list checks and the recommended-list check leave the test's body; its description says where they are proven. - RECOMMENDED_STEPS and NEVER_RECOMMENDED go with them (no test read them any more), and the meta-test case that only compared them. Proof: the forgetest host suite passes under Linux (test_setup_suite with the case removed among it); the coverage lint is unchanged. Acceptance: every setup.py test (the constants were the module's shared text), setup.gate-blocks-controllers among them.
forgetest - the ForgeFIRM release acceptance tool
The daemon behind http://<machine>:8090/ on the dev image: runs the
acceptance catalog against the machine, keeps the append-only result log,
decides which results still apply to the image that is running, exports
the release artifact scripts/release.sh gates on, and serves the bench
diagnostics page. The contract - catalog, campaigns, fingerprints,
inheritance, the gate, the coverage rule - is
the Acceptance page of the documentation site.
Run the host tests
cd forgetest
python3 -m unittest discover -s tests -v
Run the daemon on a workstation (against a mock or a manifest file)
FORGETEST_DATA=/tmp/ft FORGETEST_MANIFEST=../tree-manifest.json \
FORGECTRL_URL=http://<machine> python3 -m forgetest --port 8090
scripts/manifest-from-tree.py produces tree-manifest.json from the recipe
pins; the coverage lint is python3 -m forgetest.coverage --manifest ....
Environment
| Variable | Default | Purpose |
|---|---|---|
FORGETEST_DATA |
/data/forgetest |
results.jsonl, bench.jsonl, token, export/ |
FORGETEST_MANIFEST |
/etc/forgefirm-manifest.json |
the image manifest |
FORGETEST_PORT, FORGETEST_HOST |
8090, 0.0.0.0 | listener |
FORGETEST_BENCH_DIR |
/usr/share/forgetest/bench |
the installed bench scripts |
FORGETEST_BENCH_DATA |
<FORGETEST_DATA>/bench |
passed to bench tools: where they keep their data files (with GF_HOST=127.0.0.1 and the panel token in GF_TOKEN) |
FORGETEST_MARKER |
/run/forgetest.active |
takeover marker |
FORGECTRL_URL, FORGECTRL_TOKEN_FILE |
http://127.0.0.1, /data/forgefirm/panel.token |
forgectrl client (HTTP; the token authorizes writes from the board) |
FORGECTRL_TLS_URL |
https://127.0.0.1 |
forgectrl over HTTPS (self-signed, unverified), for the login test |
GF_SYSFS_ROOT |
/sys/glowforge/ |
kernel module sysfs |
GRBL_HOST, GRBL_PORT |
127.0.0.1, 23 | Grbl TCP |
Adding a test
Register it in the subsystem module under forgetest/suite/ with
@test(...): id subsystem.name, kind, hardware, mode (the controller
mode the test needs; the runner switches to it first), covers,
requires, always, steps. The body gets a Context (log, check, fail,
prompt, confirm, instruct, sleep, evidence, forgectrl, sysfs,
grbl, takeover). Return normally for PASS, raise runner.Failed for
FAIL. Then run the unit tests and the coverage lint.