ladyadaandLimor Fried
02fb9326be
Run sketch regressions with consistent float and double definitions across library and test translation units. Verify that exact GNSS components and double GPS results retain position changes smaller than the legacy E7 resolution, and record the precision requirement for future receiver APIs and logging. Co-authored-by: Limor Fried <ladyada@users.noreply.github.com>
Regression tests
Run every test on Linux with Python 3 and g++:
python3 extras/tests/run_tests.py
The runner discovers every .cpp and .ino in the test subdirectories.
Each C++ test supplies main(). Each Arduino sketch supplies setup(),
loop(), and a testsPassed flag that remains false until all checks pass.
The runner calls setup() and loop() once and returns failure when that
flag is false. New tests are picked up automatically without editing CI.
All tests use AddressSanitizer and UndefinedBehaviorSanitizer. Compiler warnings, failed checks, sanitizer findings, and a 30-second execution timeout fail the job. Other tests still run after an individual failure.
The Arduino sketches use the real library sources with NMEA extensions enabled,
including the marine and RMB checks. Each runs with both NMEA_FLOAT_T=float
and NMEA_FLOAT_T=double, applied consistently to the sketch and library.
extras/test_support provides serial
output and a real monotonic clock; the wakeup timeout test takes ten seconds.
I2C, SPI, GPIO, and hardware serial input abort if used. This runs the parser
and mock-stream regressions, not physical GPS/SD hardware or MCU emulation.
The sketches remain usable on Arduino boards with the normal Arduino core.
Coordinate precision tests preserve all nine retained fractional-minute digits in the GNSS core and distinguish positions about 0.185 mm apart through the GPS double output, including southern latitude and longitude near 180 degrees west. Those positions deliberately share the same legacy E7 coordinate. The default float fields and E7 fields are not full-precision RTK outputs. Future receiver APIs and logging examples must preserve the core precision through storage, public results, and formatting without requiring a global float-type override.