AP_AHRS: use direct active_backend access in place of state structure entries

[peterbarker]

Summary

Stops copying the data simply available from the backend pointer into state, just take it straight from the pointer

Classification & Testing (check all that apply and add your own)

• Checked by a human programmer
• Non-functional change
• No-binary change
• Infrastructure change (e.g. unit tests, helper scripts)
• Automated test(s) verify changes (e.g. unit test, autotest)
• Tested manually, description below (e.g. SITL)
• Tested on hardware
• Logs attached
• Logs available on request

Description

We don't need this indirection - remove it.

The matrix identity wasn't surviving long anyway, so remove it.

[peterbarker]

Board                    AP_Periph  antennatracker  blimp  bootloader  copter  heli  iofirmware  plane  rover  sub
CubeOrange-periph-heavy  *                                 *                                                   
Durandal                            -240            -200   *           -552    -544              -200   -104   -280
Hitec-Airspeed           *                                 *                                                   
KakuteH7-bdshot                     -256            -224   *           8       -8                424    64     -72
MatekF405                           -360            -248   *           -224    -224              520    -200   -288
Pixhawk1-1M-bdshot                  -272            -360               -64     -72               32     -96    -200
SITL_x86_64_linux_gnu               -144            -144               16      16                16     16     16
YJUAV_A6SE                          -248            -224   *           16      24                56     -112   -216
f103-QiotekPeriph        *                                 *                                                   
f303-MatekGPS            *                                 *                                                   
f303-Universal           *                                 *                                                   
iomcu                                                                                *                         
revo-mini                           -288            -256   *           -160    -176              -104   -264   -312
skyviper-v2450                                                         -80                                     
speedybeef4                         -328            -256   *           -208    -208              960    -240   -296

[github-actions]

Global Size Summary (Total Flash delta in bytes)

Built from a3b3a84 rebased onto afeb9c4

Board	Periph	tracker	blimp	bootloader	copter	heli	iofw	plane	rover	sub
DrotekP3Pro	_	_	_	*	_	_	_	_	_	_
Durandal	_	-160	-124	_	-4	+8	_	+48	-52	-272
Hitec-Airspeed	*	_	_	_	_	_	_	_	_	_
KakuteF7	_	-252	-176	_	-32	+260	_	-196	-200	-260
MambaH743v4	_	-272	-288	_	-64	-32	_	+8	-140	-332
MatekF405-CAN	_	_	_	*	_	_	_	_	_	_
MatekF405	_	-252	-176	_	-32	+260	_	-196	-200	-260
MatekH743-bdshot	_	-160	-124	_	+60	+32	_	+84	-80	-292
Pixhawk1-1M	_	-224	-188	_	-20	-52	_	-40	-140	-292
disco	_	-284	-272	_	-180	-136	_	-952	-348	-388
f103-GPS	*	_	_	_	_	_	_	_	_	_

[AP-Review]

Deprecated — see below for the updated review.

Previous review (2026-10-18)

Automated review note — AI-generated (Claude+Codex), validated against the live diff. Please sanity-check before acting.
Verdict: REQUEST CHANGES

Reviewed at head e6f83268c4.
Full report: https://firmware.ardupilot.org/Tools/APReview/DevCallReviews/PRReviews/ardupilot/ardupilot/33608/1.html#pr33608

Thanks. Replacing the state copies with direct active_estimates reads looks right for the main-thread and semaphore-holding callers. One problem needs fixing first, though.

get_gyro_drift() and get_gyro() (AP_AHRS.h:95, AP_AHRS.h:92) now alias the backend Estimates. AP_AHRS::update() zeroes those (libraries/AP_AHRS/AP_AHRS.cpp:572/583) before get_results() refills them. Copter's fast-rate thread reads ahrs.get_gyro_drift() without the AHRS semaphore (ArduCopter/rate_thread.cpp:265), and so does the Pixart flow callback for get_gyro() (libraries/AP_OpticalFlow/AP_OpticalFlow_Pixart.cpp:291). Either can pick up zeros in that window, which the old state copy prevented. Could you keep a published copy of those two values, or stage the results so these readers never see the cleared struct?

Minor, no change needed: get_velocity_NED() now leaves the caller's vector untouched when velocity is invalid, where it used to copy the last valid value. About ten callers ignore the return value, for example ArduPlane/quadplane.cpp:3682. It may be worth a line in the commit message, since 'Non-functional change' is unticked.

I also checked the identity removal. Only the AHRS_View constructor sees the zero matrix before the first update, which agrees with your note.

[AP-Review]

Automated review note — AI-generated (Claude+Codex), validated against the live diff. Please sanity-check before acting.
Verdict: COMMENT

Reviewed at head a3b3a8495b.
Full report: https://firmware.ardupilot.org/Tools/APReview/DevCallReviews/PRReviews/ardupilot/ardupilot/33608/2.html#pr33608

Thanks for restoring the gyro copies. get_gyro() and get_gyro_drift() now return state again (AP_AHRS.h:92, AP_AHRS.h:95), so the fast-rate thread and Pixart reads are fine.

Two smaller cases of the same reset window remain. Neither is blocking.

• iBus telemetry runs as a timer process, above main priority on ChibiOS. It reads get_accel()/get_accel_bias() without the AHRS semaphore (libraries/AP_IBus_Telem/AP_IBus_Telem.cpp:599-607). If it pre-empts update() between *configured_estimates = {} and get_results() (libraries/AP_AHRS/AP_AHRS.cpp:577-578), it can send IMU 0's accel with zero bias (AP_AHRS.h:465, AP_AHRS.h:712). This is cosmetic, and only reachable when ACC_X/Y/Z fall within iBus IDs 1-15.
• get_rotation_body_to_ned() (AP_AHRS.h:669) is read unlocked by DDS body_to_earth() (libraries/AP_DDS/AP_DDS_ExternalControl.cpp:72) and by the DroneCAN moving-baseline gate (libraries/AP_GPS/GPS_Backend.cpp:406). Those threads sit below main on ChibiOS, so it is only a problem on multi-core Linux and SITL. There, a body-frame DDS command could be turned into zero velocity.

Filling a temporary in update() and then assigning it to the stored Estimates would avoid the zero window for all of these without bringing the state copies back.

The get_velocity_NED() note from last time still stands: it no longer writes the output when the estimate is invalid. It may be worth a line in the commit message.