Automatically generate the clang-tidy exclusion list from .gitmodules
so new submodules are excluded without manual intervention.
Changes:
- Makefile: Generate CLANG_TIDY_SUBMODULES from .gitmodules paths
- Makefile: Add CLANG_TIDY_EXCLUDE_EXTRA for manual exclusions:
- src/systemcmds/tests (test code, looser style allowed)
- src/examples (educational code, not production)
- src/modules/gyro_fft/CMSIS_5 (vendored ARM DSP library)
- Delete src/systemcmds/tests/.clang-tidy (stale since 2019)
- Delete src/modules/gyro_fft/CMSIS_5/.clang-tidy (redundant)
Rationale: Submodules and vendored code should be linted in their
upstream repositories, not here. This reduces noise and focuses
clang-tidy on code that PX4 maintainers actually edit.
Contributors adding vendored (non-submodule) third-party code should
add their path to CLANG_TIDY_EXCLUDE_EXTRA in the Makefile.
Signed-off-by: Ramon Roche <mrpollo@gmail.com>
Extend the -Wno-asm-operand-widths workaround to include Linux aarch64
in addition to Apple Silicon. CMSIS DSP contains ARM Cortex-M specific
assembly that clang (but not gcc) warns about on 64-bit ARM platforms.
The assembly code is unused on POSIX builds - only the C fallback
implementations are executed in SITL.
This fixes clang-tidy CI failing on arm64 runners.
Signed-off-by: Ramon Roche <mrpollo@gmail.com>
- when estimating the peak frequency the magnitude of side buckets will
be factored in, so it doesn't make sense to potentially treat them as
separatey detected peaks
The current practice of adding topics to the default set isn't scalable,
as it affects all setups.
By making sure topics are advertised on init, logger can just discard
topics that don't exist. This does not work for all topics, so topics are
specifically marked as optional. It can be extended to more topics later
on though.
This reduces the list of topics by ~35 on a pixracer configured as quad,
and reduces RAM usage by ~1KB.
- initial frequency peak tracking SNR increased from 10->15 db
- after initial detection the threshold decreases to SNR 5db
- gyro_fft large method refactored into smaller pieces
- sensors/vehicle_angular_velocity: dynamic notch FFT make sample rate
check a percentage and relax lower bound safety threshold
- move to high priority work queue (from low priority)
- schedule slightly more often to avoid missing messages
- perf counter include all FFT processing work
- lazily allocate gyro gap perf counters on initial sensor selection
- change default FFT length (1024 -> 512)
- this doubles the update rate because half the number of samples are required for each
- decrease number of peaks (4 -> 3)
- so far 3 seems to be sufficient on most vehicles
- increase median filter window (3 -> 5)
- decrease SNR requirement (likely needs to be configurable)
Jacob Dahl