GuitarPedal

make_one_noise() walks 'disable' towards 'disable_all' one sample at a time so that switching bypass is a 100ms crossfade, and passes single_sample() disable / EFF_ENABLE_STEPS as the mix. But single_sample() takes its mix the other way up: 1.0 is the processed signal and 0.0 the untouched input, which is what bypass() passes. So the ramp runs backwards. Engaging bypass cuts straight to the dry input, fades the effects back in over 100ms, and then cuts to dry again once bypass() takes over. Releasing it is the mirror image. Two steps per footswitch tap, with a tenth of a second of the wrong signal between them. Nothing in Validation/ sees it, because the bench calls single_sample(1.0f) directly and never goes through the ramp. Driving the real make_one_noise() on the host instead, with Test Tone routed and silence on the input, the tone's level in 5ms windows after a tap is before: 1.00 1.00 0.89 0.10 0.15 0.19 ... 0.94 1.00 0.89 0.00 after: 1.00 1.00 0.99 0.95 0.90 0.84 ... 0.09 0.05 0.00

Two comments in WebMIDI/app.js no longer match the code. No behaviour change. EQ graph gain helper. The comment says the firmware's takes the linear gain and square-roots it, so the app and firmware use different units. That was true until d3dcddc ("Have the biquads take A, like the cookbook does"), which didn't touch app.js. Both take A now. I checked it with , comparing the firmware's coefficients with the app's functions. The test covered 300 filters: peaking/loshelf/hishelf, 40 Hz–12 kHz, Q 0.5–8, −20..+20 dB. They agree to 2.4e-5. If the app did what the comment describes (passing 10^(dB/20)), the gain would be off by up to 20.65 dB. I fixed the comment and renamed to to match . Positional anchors. Above there is an older comment saying the anchors are the first and last effects. That stopped being true with 666abc7 and 6cd5aa4, and the comment below it already describes the current behaviour. Removed. Tested with , which passes. test-webmidi draws every graphed effect, so the renamed function is exercised.

The build instructions leave the reader in the repository root, where 'make check' fails because the target lives in Validation/Makefile. State which directory to use for the commands in the Testing section.