#pragma once

#include <juce_audio_basics/juce_audio_basics.h>
#include <juce_dsp/juce_dsp.h>
#include <atomic>
#include <array>
#include <cmath>
#include <string>

// Real-time audio analyser — runs in processBlock, results read from MCP thread.
// All inter-thread communication via std::atomic — no locks in audio path.

class AudioAnalyser
{
public:
    AudioAnalyser() = default;

    void prepareToPlay(double sampleRate, int blockSize)
    {
        currentSampleRate = sampleRate;
        currentBlockSize = blockSize;

        // FFT setup (1024-point)
        fftBuffer.fill(0.0f);
        fftWritePos = 0;

        // LUFS — simplified momentary loudness (400ms window)
        lufsWindowSamples = (int)(sampleRate * 0.4);
        lufsAccumulator = 0.0;
        lufsSampleCount = 0;

        // Reset all outputs
        momentaryLUFS.store(-100.0f);
        truePeak.store(-100.0f);
        stereoWidth.store(0.0f);
        spectralCentroid.store(0.0f);
        isSilent.store(true);

        for (auto& b : bands)
            b.store(0.0f);
    }

    // Called from audio thread — MUST be lock-free
    void processBlock(const juce::AudioBuffer<float>& buffer)
    {
        const int numSamples = buffer.getNumSamples();
        const int numChannels = buffer.getNumChannels();
        if (numSamples == 0 || numChannels == 0) return;

        const float* left = buffer.getReadPointer(0);
        const float* right = numChannels > 1 ? buffer.getReadPointer(1) : left;

        // --- True Peak ---
        float peak = 0.0f;
        for (int i = 0; i < numSamples; ++i)
        {
            float absL = std::fabs(left[i]);
            float absR = std::fabs(right[i]);
            peak = std::max(peak, std::max(absL, absR));
        }
        float peakDb = peak > 0.0f ? 20.0f * std::log10(peak) : -100.0f;
        truePeak.store(peakDb);

        // --- Silence Gate ---
        if (peakDb < -60.0f)
        {
            isSilent.store(true);
            return;
        }
        isSilent.store(false);

        // --- LUFS (simplified momentary — sum of squares over 400ms) ---
        for (int i = 0; i < numSamples; ++i)
        {
            float mono = (left[i] + right[i]) * 0.5f;
            lufsAccumulator += (double)(mono * mono);
            lufsSampleCount++;

            if (lufsSampleCount >= lufsWindowSamples)
            {
                double meanSquare = lufsAccumulator / (double)lufsSampleCount;
                float lufs = (meanSquare > 0.0) ? (float)(-0.691 + 10.0 * std::log10(meanSquare)) : -100.0f;
                momentaryLUFS.store(lufs);
                lufsAccumulator = 0.0;
                lufsSampleCount = 0;
            }
        }

        // --- Stereo Width (correlation-based) ---
        float sumMid = 0.0f, sumSide = 0.0f;
        for (int i = 0; i < numSamples; ++i)
        {
            float mid = (left[i] + right[i]) * 0.5f;
            float side = (left[i] - right[i]) * 0.5f;
            sumMid += mid * mid;
            sumSide += side * side;
        }
        float totalEnergy = sumMid + sumSide;
        float width = (totalEnergy > 1e-10f) ? sumSide / totalEnergy : 0.0f;
        // Smooth
        float prevWidth = stereoWidth.load();
        stereoWidth.store(prevWidth * 0.9f + width * 0.1f);

        // --- FFT accumulator (ring buffer → run FFT when full) ---
        for (int i = 0; i < numSamples; ++i)
        {
            float mono = (left[i] + right[i]) * 0.5f;
            fftBuffer[fftWritePos] = mono;
            fftWritePos++;

            if (fftWritePos >= kFFTSize)
            {
                runFFT();
                fftWritePos = 0;
            }
        }
    }

    // Called from HTTP/MCP thread — thread-safe read of atomic values
    std::string getCompactAnalysis() const
    {
        if (isSilent.load())
            return "silent";

        char buf[256];
        float lufs = momentaryLUFS.load();
        float tp = truePeak.load();
        float width = stereoWidth.load();
        float centroid = spectralCentroid.load();

        // Band deviations from flat (0 = flat reference)
        // Only report bands that deviate > ±2dB
        std::string bandStr;
        static const char* bandNames[] = {"sub", "bass", "low", "mid", "hi", "pres", "brill"};
        for (int i = 0; i < 7; ++i)
        {
            float db = bands[i].load();
            if (db > 2.0f)
                bandStr += std::string(bandNames[i]) + ":+" + std::to_string((int)db) + " ";
            else if (db < -2.0f)
                bandStr += std::string(bandNames[i]) + ":" + std::to_string((int)db) + " ";
        }
        if (bandStr.empty())
            bandStr = "flat ";

        // Brightness descriptor from centroid
        const char* brightnessDesc = "balanced";
        if (centroid > 4000.0f) brightnessDesc = "bright";
        else if (centroid < 1500.0f) brightnessDesc = "dark";

        snprintf(buf, sizeof(buf), "%.1f LUFS | TP:%.1f | %s| W:%.2f | %s",
                 lufs, tp, bandStr.c_str(), width, brightnessDesc);

        return std::string(buf);
    }

private:
    static constexpr int kFFTOrder = 10;
    static constexpr int kFFTSize = 1 << kFFTOrder; // 1024

    void runFFT()
    {
        // Apply window
        std::array<float, kFFTSize> windowed;
        for (int i = 0; i < kFFTSize; ++i)
            windowed[i] = fftBuffer[i] * hanningWindow(i, kFFTSize);

        // JUCE FFT requires 2x buffer (real + imaginary interleaved)
        std::array<float, kFFTSize * 2> fftData{};
        for (int i = 0; i < kFFTSize; ++i)
            fftData[i] = windowed[i];

        // fft is a stored member — no heap allocation in audio thread
        fft.performFrequencyOnlyForwardTransform(fftData.data());

        // Now fftData[0..kFFTSize/2] contains magnitudes
        int halfSize = kFFTSize / 2;
        float binWidth = (float)currentSampleRate / (float)kFFTSize;

        // --- 7 Band Energy ---
        // sub: 20-60, bass: 60-250, low: 250-500, mid: 500-2k, hi: 2k-6k, pres: 6k-12k, brill: 12k-20k
        static const float bandEdges[] = {20, 60, 250, 500, 2000, 6000, 12000, 20000};

        float bandEnergy[7] = {};
        int bandCount[7] = {};

        for (int bin = 1; bin < halfSize; ++bin)
        {
            float freq = bin * binWidth;
            float mag = fftData[bin];

            for (int b = 0; b < 7; ++b)
            {
                if (freq >= bandEdges[b] && freq < bandEdges[b + 1])
                {
                    bandEnergy[b] += mag * mag;
                    bandCount[b]++;
                    break;
                }
            }
        }

        // Convert to dB relative to average
        float totalEnergy = 0.0f;
        for (int b = 0; b < 7; ++b)
            totalEnergy += bandEnergy[b];

        float avgEnergy = totalEnergy / 7.0f;

        for (int b = 0; b < 7; ++b)
        {
            float bandDb = 0.0f;
            if (avgEnergy > 1e-10f && bandEnergy[b] > 1e-10f)
                bandDb = 10.0f * std::log10(bandEnergy[b] / avgEnergy);
            bands[b].store(bandDb);
        }

        // --- Spectral Centroid ---
        float weightedSum = 0.0f;
        float magnitudeSum = 0.0f;
        for (int bin = 1; bin < halfSize; ++bin)
        {
            float freq = bin * binWidth;
            float mag = fftData[bin];
            weightedSum += freq * mag;
            magnitudeSum += mag;
        }
        float centroid = (magnitudeSum > 1e-10f) ? weightedSum / magnitudeSum : 0.0f;
        // Smooth
        float prevCentroid = spectralCentroid.load();
        spectralCentroid.store(prevCentroid * 0.8f + centroid * 0.2f);
    }

    static float hanningWindow(int i, int size)
    {
        return 0.5f * (1.0f - std::cos(2.0f * 3.14159265358979f * i / (float)(size - 1)));
    }

    // State
    double currentSampleRate = 44100.0;
    int currentBlockSize = 512;

    // FFT — constructed once, reused per block (no allocation in audio thread)
    juce::dsp::FFT fft{kFFTOrder};

    // FFT ring buffer
    std::array<float, kFFTSize> fftBuffer{};
    int fftWritePos = 0;

    // LUFS accumulator
    int lufsWindowSamples = 17640; // 400ms at 44.1kHz
    double lufsAccumulator = 0.0;
    int lufsSampleCount = 0;

    // Thread-safe outputs (written in audio thread, read from HTTP thread)
    std::atomic<float> momentaryLUFS{-100.0f};
    std::atomic<float> truePeak{-100.0f};
    std::atomic<float> stereoWidth{0.0f};
    std::atomic<float> spectralCentroid{0.0f};
    std::atomic<bool> isSilent{true};
    std::array<std::atomic<float>, 7> bands{}; // sub, bass, low, mid, hi, pres, brill
};