[jiri-god]

With 2x host oversampling, a processor prepared at 96 kHz receives a 48 kHz sample position, and a reported 257-sample intrinsic latency is added to the parent graph as 257 samples rather than 129. This affects sample-based timing and graph delay compensation.

The conversion now lives in element::Processor, covering direct implementations (including the CLAP/native-node path) as well as AudioProcessorNode wrappers. Each processor exposes a stable playhead adapter that scales only timeInSamples by its prepared processing factor; seconds, PPQ, BPM and absent position fields are preserved. Nested graphs compose their local factors once per boundary. Multiplication is bounded to avoid signed overflow.

Intrinsic latency is stored atomically in rendering-rate samples and converted to parent-rate samples, rounding up, in Processor::getLatencySamples(). Manual delay compensation and oversampling-filter latency are already in parent-rate units and remain separate. AudioProcessorNode and ScriptNode forward the common playhead and the audio-processor wrapper reports raw plugin latency, including latency-change notifications, without a second conversion.

Validation on Apple Silicon macOS, RelWithDebInfo, JUCE 8.0.13:

The final independent branch passes all 59 CTest suites (3.42 s).

• PluginSampleUnitsTests covers wrapped plugins, direct Processor subclasses, 1x/2x/4x/8x and returning to 1x, latency notifications, manual-delay units, cached playhead rebinding/detachment, optional position fields and integer boundaries.
• Nested 2x graphs with 1x/2x/4x children verify both direct and wrapped processors receive the composite sample rate/position, while seconds and the source clock remain unchanged.
• Applying the new tests to the pre-review PR revision f1e3cf3 produces 22 failures in the direct/nested cases; the original wrapped-plugin case still passes. The new direct/nested cases pass with the revision.
• A Lua DSP regression observes the actual script position before/after reload and playhead rebinding. It reproduces four failures with the old ScriptNode forwarding override and passes with the shared clock.
• Original regression evidence on upstream d0226686: six assertions failed, including -48000 != -96000 and 257 != 129 at 2x.

The direct-subclass test exercises the shared path used by CLAP; it is not an audition of a third-party CLAP binary. Graph compensation still uses integer samples; no fractional delay lines are introduced. Linux/Windows validation is left to CI. This remains an independent host-only change based on upstream main; no private extensions are included.

The final change also merges without conflicts with #1212–#1214; all 62 CTest suites pass together (3.34 s).

[mfisher31]

Hey thanks for all the new PRs. Gonna try and get those reviewed tonight!

[mfisher31]

I like where this is going @jiri-god, but doing the playhead work in audioprocessornode will only affect VST, VST3, AU, and LV2 plugins. Clap implements element::Processor directly, and all internal nodes (mostly) are element::Processor's as well. The good thing is, all of the latency information you're using already exists in the base class. Should me a relatively simple move of this code up a level to the the Processor class.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}