[pfeerick]

Fixes #2322

Summary

An input line can reverse its stick, either with a negative weight or with a custom curve. However, the trim was still applied in the original direction at the mix stage, so the stick and trim worked against each other: e.g. moving the stick right moved the channel negative, while trimming right moved it positive.

The trim value is added in the mix stage, after the input weight and curve have already been applied, and nothing recorded that the input was reversed. With this change, the trim's direction matches the physical stick: trimming right has the same effect as nudging the stick right. The trim's size is unchanged, so it still doesn't scale with input weight or expo.

An input is treated as reversed when exactly one of these is true:

• its (evaluated) weight is negative, so a GV weight works too
• it uses a custom curve whose output at full positive stick is lower than at full negative stick. This covers both an inverted curve reference (!CVn) and a curve with decreasing points.

A negative weight together with a reversing curve cancel out. Symmetric (V or U shaped) curves have no single direction, so the trim is left as before for them. Expo, diff and function curves never reverse the stick.

Changes:

• applyExpos() records per input whether the active input line reverses it.
• getSourceTrimValue() negates the trim for such inputs. It also negates the stick value passed for the throttle idle-trim scaling, so idle-only trim still applies at the correct end of the stick travel. The existing inverted mix source handling still applies on top.
• getValueForLogicalSwitch() applies the same reversal, so logical switches on an input match the mix output.
• instantTrim() negates the measured offset for reversed inputs. Otherwise it would double the offset instead of cancelling it.

moveTrimsToOffsets() works on the final mixer output, so it needed no change. Reversals after the Inputs stage (negative mix weight, !Input mix source, mix curves, output reverse) already applied to stick and trim alike.

⚠️ Behaviour change

Models that reverse a stick in Inputs (with a negative weight, !CVn, or a decreasing custom curve) and had their trims set with the old behaviour will see those trims act in the opposite direction after updating. Those trims would need to be re-centred or re-set.

Testing

• New gtests:
  • TrimsTest.ReversedInputTrim
  • TrimsTest.ReversedInputCurveTrim
  • TrimsTest.InstantTrimReversedInput
  • an extra case in getSwitch.inputWithTrim
• tools/commit-tests.sh passes for tx16s, x9dp2019, mt12 (surface), and pl18 and h17 (stick dead zone).
• Not yet tested on hardware.

🤖 Generated with Claude Code