Testing

The workspaces’ own packages hold about 1,600 test functions, plus about 140 more for the tools in tools/. Parametrized tests run once per case, so pytest collects more than that. The two cars’ suites are the same apart from two files in racerbot_launch that guard each car’s own hardware settings.

Every test is a Python function whose name starts with test_. pytest finds those functions, runs each one, and reports whether its assert statements held. A test passes when every assertion holds and fails when one does not.

Kinds of test

The tests fall into a few kinds. A package usually has more than one.

Kind	What it does	Needs ROS	Example
Unit	Calls one function with chosen inputs and checks the result against something worked out independently: a closed-form answer, an invariant, or a measured value	No	test_racing_math.py checks that a circle of radius $R$ comes out with curvature $1/R$
Node	Starts the real ROS node in the test process, publishes fake inputs on its topics, and checks what it publishes	Yes	test_gap_follow_node.py checks that releasing LB publishes speed 0
Configuration	Reads a shipped YAML, launch file or udev rule and checks it against the code that uses it, or against another file it must agree with	Mostly no	test_steering_limits_fit_servo.py checks every steering limit against the servo range in vesc.yaml
Hardware guard	Checks that each car’s own hardware settings still match its profile and the hardware reference after any change	Mostly no	test_car_profiles.py: car 1’s USB Hokuyo and car 2’s Ethernet one, each loaded only on its own car
Parity	Checks that two copies of the same logic agree: the simulator’s and the node’s, or the training code’s and the car’s	Varies	test_gap_mirror_parity.py: the simulator’s copy of gap_follow commands what the node commands
Front end	Runs the calibration wizard’s JavaScript in Node.js with the page and the network faked	No (needs Node.js)	test_wizard_ui.py
Performance	Checks that a computation fits its time budget	No	test_latency.py: the neural racer’s tick takes under 5 ms at the 99th percentile on the Jetson
Simulator	Drives whole laps in a simulator and judges the result independently of the simulator	Varies	tools/f1tenth_sim/run_validation.py
Style	flake8, pep257, copyright headers, in the vendored packages only	No	test_flake8.py

Unit tests are the fastest and most precise, so every package with non-trivial mathematics keeps that mathematics in a module that imports no ROS, where unit tests can call it directly. Node tests catch what unit tests cannot: a topic name that is wrong, a parameter that is never read, a deadman check that is never called.

Pages

Page	Read it for
Run the tests	The commands, why tests need their own ROS domain, and how to read the output when something fails
Tests by package	What each package’s tests check, and how to run them
Test quality standards	The rules for writing a test: prove it can fail, never weaken it to pass