mirror of
https://github.com/meshtastic/firmware.git
synced 2026-08-01 10:58:30 -04:00
dd509a8ecf
GeoCoord's constructor unconditionally computed all five coordinate representations (DMS, UTM, MGRS, OSGR, OLC) via setCoords(), even though most callers only ever read one. The UTM conversion alone pulls in a chain of libm trig functions (atan, __kernel_tan, __ieee754_acos, __ieee754_pow, __kernel_rem_pio2/__ieee754_rem_pio2) that then have to be linked in regardless of whether anything is ever displayed. The only screen consumer (UIRenderer.cpp's GPS coordinate display) already dispatches on a single configured format and reads exactly one representation per call - never more than one. Compute DMS eagerly (cheap, most commonly needed) and defer UTM/MGRS/OSGR/OLC to first access via their own getters, tracked with per-representation mutable dirty flags, so a caller that never touches a given representation never pulls in its conversion code or the trig functions it needs. On stm32wl, GeoCoord's heavy constructor is reached only through NMEAWPL.cpp's NMEA/CalTopo serial export (GPS-gated, so wio-e5 only), which only ever reads the DMS getters - so this recovers 8,288 bytes flash there (of the 10,172-byte ceiling if the whole feature were cut) with the feature fully intact and zero behavior change on any platform. Updated test_geocoord_extreme_coords_no_oob (the existing regression test for a historical out-of-bounds crash in the UTM/MGRS conversion on extreme lat/lon) to explicitly call each representation's getter, since it previously relied on the constructor eagerly triggering all four conversions - which this change intentionally defers. Verified: full native test suite passes (586/586 test cases, including this one under ASan), and rak4631 (nRF52, screen-equipped, where the UTM/MGRS/OSGR/OLC getters are actually reachable) builds successfully. Assisted-by: Claude Sonnet 5 <noreply@anthropic.com> Signed-off-by: Andrew Yong <me@ndoo.sg> Co-authored-by: Ben Meadors <benmmeadors@gmail.com>