Host writes to te::AutomatableParameter

Host code that pushes a value into a TE plugin parameter — UI sliders, controllers, project state restore, modifier-driven sync — must call setParameterFromHost(value, nt), not setParameter(value, nt).

If you call setParameter and any modifier is producing nonzero output on that param, your write is silently dropped.

Why

Upstream TE commit a2ac9e8acd1 ("Fix macro-linked parameter drift caused by plugin feedback loop") added a guard at the top of AutomatableParameter::setParameter:

void AutomatableParameter::setParameter (float value, juce::NotificationType nt)
{
    if (currentModifierValue.load() != 0.0f)
        return;
    ...
}

The guard exists because some plugins (synths with their own internal LFOs, etc.) report their modulated value back to the host through the parameter callback. With a host-side modifier already adding on top, that creates a feedback loop and the base drifts to maximum within a few audio blocks.

But TE has no flag distinguishing "plugin echoed its modulated value" from "user moved the host slider" — so the guard drops both. Without the patch, every parameter with any modifier attached (LFO, macro, envelope) silently ignores host writes. Affects 4OSC, every TE built-in, and every external VST/AU equally.

The patch

MAGDA's TE fork (Conceptual-Machines/tracktion_engine, branch magda_minimal) carries commit cc7ecae8b44 which adds:

// Header: tracktion_AutomatableParameter.h
void setParameterFromHost (float value, juce::NotificationType);

Implementation is the same body as setParameter minus the modifier guard. The original setParameter still exists and still drops plugin echoes — leave it for that purpose.

How to call from MAGDA

// Wrong — gets silently dropped when a modifier is active
params[paramIndex]->setParameter(value, juce::sendNotificationSync);

// Right
params[paramIndex]->setParameterFromHost(value, juce::sendNotificationSync);

When to use which

Caller	Method
DeviceProcessor::setParameterByIndex (UI slider, controller, automation playback)	setParameterFromHost
ExternalPluginProcessor::syncFromDeviceInfo (project state restore)	setParameterFromHost
ControllerParamWriter (MIDI controller)	setParameterFromHost
RackSyncManager / PluginManager macro setters, sync paths	setParameterFromHost
applyLFOProperties and other modifier-internal param setup	setParameterFromHost
Internal MAGDA plugin ValueTree → TE param sync (Arpeggiator, StepSequencer, MagdaSampler, DrumGrid)	setParameterFromHost
Plugin-internal callback echoing its own modulated value back to host	setParameter (so the guard catches it)

The default for any new MAGDA-side write is setParameterFromHost. Only use setParameter if the call originates from inside the plugin's own callback (which MAGDA rarely writes — mostly upstream TE territory).

Adding a new processor

If you add a new DeviceProcessor subclass that wraps a TE plugin:

void MyNewProcessor::setParameterByIndex(int paramIndex, float value) {
    if (!plugin_) return;
    auto params = plugin_->getAutomatableParameters();
    if (paramIndex >= 0 && paramIndex < params.size()) {
        params[paramIndex]->setParameterFromHost(value, juce::sendNotificationSync);
    }
}

Match the pattern in FourOscProcessor::setParameterByIndex etc. in magda/daw/audio/DeviceProcessor.cpp.

After a TE submodule bump

If make debug fails with 'setParameterFromHost' is not a member of 'tracktion::AutomatableParameter', the patch was lost in the bump. Re-apply:

1. cd third_party/tracktion_engine
2. Check the patch commit is present: git log --oneline | grep setParameterFromHost
3. If missing, cherry-pick cc7ecae8b44 from the magda_minimal branch of the fork, or re-apply the diff:
   • Header: add void setParameterFromHost (float value, juce::NotificationType); next to setParameter in tracktion_AutomatableParameter.h
   • Cpp: extract everything after the guard in setParameter into a new setParameterFromHost, and replace the post-guard body of setParameter with setParameterFromHost(value, nt);
4. Commit on the fork's magda_minimal branch and push, then bump the parent submodule pointer.

How to verify the regression hasn't reintroduced

Set up an LFO on a parameter (any plugin's filter cutoff is a clean test), then move the underlying parameter slider while the LFO is producing audible modulation:

• Working: the modulation envelope shifts up/down with the slider — you can hear the sweep range move.
• Broken (regression): the slider has no audible effect; the sweep stays pinned where it was.