Division by zero possible in Oscillator and YinPitchDetector

[user1303836]

Problem

Two division-by-zero paths exist that can produce Inf/NaN and propagate through the entire signal chain:

1. Oscillator: sr = 0

Oscillator::updateIncrement() (source/dsp/Oscillator.cpp:52) computes phaseIncrement = freq / sr. If a host passes sampleRate = 0 to prepareToPlay, sr becomes 0.0 and division produces Inf, which propagates to the audio output.

2. YinPitchDetector: betterTau = 0

YinPitchDetector::analyse() (source/dsp/YinPitchDetector.cpp:109) computes float freq = static_cast<float>(analysisSR) / betterTau. After parabolic interpolation (lines 97-107), betterTau could theoretically reach 0 or near-zero, producing Inf.

3. PitchSmoother: sr = 0

PitchSmoother::recomputeAlpha() (source/dsp/PitchSmoother.h:48) computes alpha = 1.0f - std::exp(-1.0f / (sr * tau)). If sr = 0, division by zero inside exp argument yields NaN for alpha, and every subsequent process() call produces NaN.

Corroborated by: 2/8 audit agents (stability-edge, pitch-detection)

Suggested Fixes

1. Oscillator: Guard prepare(): sr = std::max(1.0, sampleRate);
2. YinPitchDetector: Guard after parabolic interpolation: if (betterTau < 1.0f) { lastResult = {0.0f, 0.0f}; return; }
3. PitchSmoother: Guard prepare(): sr = std::max(1.0f, static_cast<float>(sampleRate));

Impact

• Stability: Can produce Inf/NaN in audio output, causing silence, loud noise, or downstream plugin crashes
• Severity: High -- defensive guards are trivial to add