// native_bridge/src/RealtimeFxLoop.cpp
// fx_vst_bridge --realtime-fx <job.json> --shm <name> [--parent <pid>]
// Vòng lặp SHM realtime: engine (Python) tạo mapping FxRealtimeIPC, bridge mở,
// đọc block input ring (engine->bridge), chạy qua RealtimeFxChain (VST3 FX,
// SEH-guarded, in-place), ghi block output ring (bridge->engine). Heartbeat
// 100ms riêng (watchdog). Thoát khi running==0 hoặc parent chết.
//
// job.json: { "sample_rate": 48000, "block_size": 256, "fx_chain": [ ... ] }
//   fx_chain: mảng {type,path,name,preset_b64,bypass} — RealtimeFxChain::setChain
//   nhận JSON mảng {path,bypass,preset_b64} (type/name bỏ qua).
#include "RenderFxJob.h"
#include "FxRealtimeIPC.h"

#ifdef _WIN32
#ifndef NOMINMAX
#define NOMINMAX
#endif
#include <windows.h>
#include <mmsystem.h>
#endif

#include <algorithm>
#include <cctype>
#include <chrono>
#include <cstring>
#include <fstream>
#include <iostream>
#include <iterator>
#include <string>
#include <thread>

namespace {

struct ShmView {
    void* map = nullptr;
    void* view = nullptr;
};

ShmView* openShm(const std::string& name, size_t size) {
    int wlen = MultiByteToWideChar(CP_UTF8, 0, name.c_str(), -1, nullptr, 0);
    std::wstring wname(wlen, L'\0');
    MultiByteToWideChar(CP_UTF8, 0, name.c_str(), -1, &wname[0], wlen);
    HANDLE map = OpenFileMappingW(FILE_MAP_ALL_ACCESS, FALSE, wname.c_str());
    if (!map) return nullptr;
    void* view = MapViewOfFile(map, FILE_MAP_ALL_ACCESS, 0, 0, size);
    if (!view) { CloseHandle(map); return nullptr; }
    ShmView* v = new ShmView();
    v->map = map;
    v->view = view;
    return v;
}

void closeShm(ShmView* v) {
    if (!v) return;
    if (v->view) UnmapViewOfFile(v->view);
    if (v->map) CloseHandle((HANDLE)v->map);
    delete v;
}

static bool parentAlive(uint32_t pid) {
    if (pid == 0) return true;
    HANDLE h = OpenProcess(PROCESS_QUERY_LIMITED_INFORMATION, FALSE, pid);
    if (!h) return false;
    CloseHandle(h);
    return true;
}

static void sleepMs(uint32_t ms) { Sleep(ms); }

std::string readFile(const std::string& path) {
    std::ifstream f(path, std::ios::binary);
    return std::string((std::istreambuf_iterator<char>(f)),
                       std::istreambuf_iterator<char>());
}

// Tách giá trị số khỏi JSON flat: "key":<num> hoặc "key": <num>
double jsonNumber(const std::string& s, const std::string& key, double def) {
    std::string k = "\"" + key + "\"";
    size_t p = s.find(k);
    if (p == std::string::npos) return def;
    p += k.size();
    while (p < s.size() && (s[p] == ':' || s[p] == ' ' || s[p] == '\t' ||
                            s[p] == '\r' || s[p] == '\n'))
        ++p;
    size_t e = p;
    while (e < s.size() && (std::isdigit((unsigned char)s[e]) || s[e] == '.' ||
                            s[e] == '-' || s[e] == '+' || s[e] == 'e' ||
                            s[e] == 'E'))
        ++e;
    if (e == p) return def;
    try { return std::stod(s.substr(p, e - p)); } catch (...) { return def; }
}

// Cắt substring mảng JSON "fx_chain":[...] (base64 không chứa [ ] nên depth
// scan an toàn). Trả rỗng nếu không có.
std::string extractChainJson(const std::string& s) {
    std::string k = "\"fx_chain\"";
    size_t p = s.find(k);
    if (p == std::string::npos) return "";
    p = s.find('[', p);
    if (p == std::string::npos) return "";
    size_t depth = 0, i = p;
    for (; i < s.size(); ++i) {
        if (s[i] == '[') ++depth;
        else if (s[i] == ']') { --depth; if (depth == 0) break; }
    }
    if (i >= s.size()) return "";
    return s.substr(p, i - p + 1);
}

}  // namespace

int run_realtime_fx_loop(const std::string& jobPath, const std::string& shmName,
                         uint32_t parentPid) {
    std::cerr << "[RealtimeFxLoop] start job=" << jobPath << " shm=" << shmName
              << " parent=" << parentPid << std::endl;
    const std::string job = readFile(jobPath);
    if (job.empty()) {
        std::cerr << "[RealtimeFxLoop] cannot read job file" << std::endl;
        return 1;
    }
    const double srD = jsonNumber(job, "sample_rate", 44100.0);
    const uint32_t sampleRate = (uint32_t)srD;
    const double blkD = jsonNumber(job, "block_size", (double)FXRT_BLOCK);
    const uint32_t block = (uint32_t)std::max<double>(32.0, std::min<double>(blkD, (double)FXRT_BLOCK));
    const std::string chainJson = extractChainJson(job);
    if (chainJson.empty()) {
        std::cerr << "[RealtimeFxLoop] no fx_chain in job (empty chain = passthrough)" << std::endl;
    }

    // Batching (fix lag/giật/crackle): gom tối đa FXRT_IN_SLOTS block input
    // (256 mẫu) → 1 call chain.process(L,R,batchN). Overhead VST3 (memcpy,
    // lock param, setup processData, memset output) amortize 4 lần → engine
    // theo kịp 187.5 block/s. setChain với batchN để plugin buffer đủ cho
    // process n lớn (không overflow khi xử lí burst).
    const uint32_t batchN = block * FXRT_IN_SLOTS;
    RealtimeFxChain chain;
    chain.setChain(chainJson, sampleRate, (int32_t)batchN);

    ShmView* v = openShm(shmName, sizeof(FxRealtimeIPC));
    if (!v) {
        std::cerr << "[RealtimeFxLoop] cannot open SHM: " << shmName
                  << " (engine phải tạo trước)" << std::endl;
        return 2;
    }
    auto* ipc = static_cast<FxRealtimeIPC*>(v->view);

    // Chờ engine khởi tạo header (magic) — tối đa 2s.
    for (int i = 0; i < 200 && ipc->h.magic != FXRT_MAGIC; ++i) sleepMs(10);
    if (ipc->h.magic != FXRT_MAGIC) {
        std::cerr << "[RealtimeFxLoop] SHM magic mismatch (engine chưa init?)" << std::endl;
        closeShm(v);
        return 2;
    }
    if (ipc->h.inSlots != FXRT_IN_SLOTS || ipc->h.outSlots != FXRT_OUT_SLOTS) {
        std::cerr << "[RealtimeFxLoop] slot count mismatch" << std::endl;
        closeShm(v);
        return 2;
    }
    // Windows timer resolution 1ms — bỏ granularity Sleep 15.6ms cho toàn
    // process (idle sleep, chain wait, heartbeat). timeEndPeriod đối ứng ở
    // cuối. (winmm đã link sẵn — CMakeLists fx_vst_bridge.)
#ifdef _WIN32
    timeBeginPeriod(1);
    // Thread audio realtime ABOVE_NORMAL: khong bi tien trinh NORMAL khac
    // (fx-gui spawn + VST load instance thu 2, native VSTi GUI open...) preempt
    // -> underrun -> am bi "bop nghet"/ngat quan. ChannelWorker (main.cpp)
    // da dung THREAD_PRIORITY_ABOVE_NORMAL.
    SetThreadPriority(GetCurrentThread(), THREAD_PRIORITY_ABOVE_NORMAL);
#endif
    // Chờ chain ĐẦU TIÊN được worker build xong (gen>=1) trước khi READY.
    // setChain() là ASYNC — worker thread load VST3 + preset mất 1-4s; nếu set
    // READY ngay, engine mở WS và audio chảy qua chain RỖNG (passthrough), rồi
    // worker swap chain GIỮA STREAM khi load xong → jump dry→VST + latency đổi
    // đột ngột → crackle. Chờ swap xong: engine không gửi audio tới khi chain
    // thật sự sẵn sàng. (Engine wait_ready timeout = 30s > deadline 28s.)
    const auto chainDeadline = std::chrono::steady_clock::now() +
                               std::chrono::seconds(28);
    while (chain.chainGen() == 0 &&
           std::chrono::steady_clock::now() < chainDeadline)
        sleepMs(10);
    if (chain.chainGen() == 0)
        std::cerr << "[RealtimeFxLoop] chain load TIMEOUT 28s — READY anyway"
                  << std::endl;
    else
        std::cerr << "[RealtimeFxLoop] chain ready (gen=" << chain.chainGen()
                  << ")" << std::endl;
    ipc->h.state = FXRT_STATE_READY;
    std::cerr << "[RealtimeFxLoop] ready sr=" << sampleRate << " block=" << block
              << " chain=" << (chainJson.empty() ? "(empty)" : "loaded") << std::endl;

    // Heartbeat thread riêng — watchdog của engine (loop chậm không = chết).
    std::thread hb([&]() {
        while (ipc->h.running && ipc->h.state == FXRT_STATE_READY) {
            ipc->h.heartbeat++;
            sleepMs(100);
        }
    });

    // Chain update (set_chain): engine ghi lại job file seq++ — poll mỗi
    // 100ms; seq đổi → setChain ASYNC (worker load chain mới, swap atomic
    // dưới mutex — audio thread copy shared_ptr, không dừng) → load/đổi VST
    // FX + đổi preset GUI không ngắt âm/crackle.
    uint64_t lastSeq = (uint64_t)jsonNumber(job, "seq", 0);
    auto lastPoll = std::chrono::steady_clock::now();
    const uint32_t n = block;
    float L[FXRT_BLOCK * FXRT_IN_SLOTS], R[FXRT_BLOCK * FXRT_IN_SLOTS];
    uint32_t inMask = ipc->h.inSlots - 1;
    uint32_t outMask = ipc->h.outSlots - 1;
    uint64_t processed = 0;
    uint64_t lastGen = 0;  // chain generation đã báo latency
    uint32_t fadeLeft = 0;  // output fade-in còn lại sau chain swap (che gap latency jump)
    const uint32_t fadeTotal = 512;  // ~11.6ms @44.1k
    // ---- FXRT_PERF instrumentation (tạm, đo crackle) ----
    uint64_t perfIter = 0, perfProc = 0, perfIdle = 0, perfOutFull = 0;
    double perfProcSum = 0.0, perfProcMax = 0.0, perfProcMin = 1e9;
    double perfLoopSum = 0.0, perfLoopMax = 0.0;
    uint32_t perfTakeSum = 0;
    auto perfT0 = std::chrono::steady_clock::now();
    auto perfRunStart = perfT0;
    uint64_t perfLastOut = 0;
    while (ipc->h.running) {
        if (parentPid && !parentAlive(parentPid)) {
            std::cerr << "[RealtimeFxLoop] parent gone — exiting" << std::endl;
            break;
        }
        {
            const auto nowT = std::chrono::steady_clock::now();
            if (std::chrono::duration_cast<std::chrono::milliseconds>(nowT - lastPoll).count() >= 100) {
                lastPoll = nowT;
                const std::string j2 = readFile(jobPath);
                if (!j2.empty()) {
                    const uint64_t s2 = (uint64_t)jsonNumber(j2, "seq", 0);
                    if (s2 != lastSeq) {
                        const std::string c2 = extractChainJson(j2);
                        if (!c2.empty()) {
                            lastSeq = s2;
                            std::cerr << "[RealtimeFxLoop] chain update seq=" << s2 << std::endl;
                            chain.setChain(c2, sampleRate, (int32_t)batchN);
                        }
                    }
                }
            }
        }
        // SET_PARAM ring (engine -> bridge, Phase 2.8): drain mỗi iteration —
        // áp dụng vào chain TRƯỚC block kế tiếp. Ring đầy → drop lệnh cũ.
        while (ipc->h.ctrlRead < ipc->h.ctrlWrite) {
            const uint32_t cs = ipc->h.ctrlRead & (FXRT_CTRL_SLOTS - 1);
            const FxCtrlCmd& cmd = ipc->ctrl[cs];
            const char* kend = std::find(cmd.key, cmd.key + sizeof(cmd.key), '\0');
            chain.setParam((int)cmd.slot,
                           std::string(cmd.key, (size_t)(kend - cmd.key)),
                           (double)cmd.value);
            MemoryBarrier();
            ipc->h.ctrlRead++;
        }
        // REPORT_LATENCY (bridge -> engine): chain swap (gen đổi) → báo lại
        // latency từng slot (builtin=0; VST3=getLatencySamples). Ring 8 —
        // engine drain định kỳ, đầy thì overwrite (drop cũ).
        const uint64_t gen = chain.chainGen();
        if (gen != lastGen) {
            const std::vector<uint32_t> lats = chain.entryLatencies();
            for (size_t i = 0; i < lats.size(); ++i) {
                const uint32_t ls = ipc->h.latWrite & (FXRT_LAT_SLOTS - 1);
                ipc->lat[ls].slot = (uint32_t)i;
                ipc->lat[ls].samples = lats[i];
                MemoryBarrier();
                ipc->h.latWrite++;
            }
            lastGen = gen;
            fadeLeft = fadeTotal;
            std::cerr << "[RealtimeFxLoop] latency reported: " << lats.size()
                      << " slot(s)" << std::endl;
        }
        // OUT-ring backpressure (fix crackle "wobble" — pointer collision):
        // kiểm tra chỗ trống TRƯỚC khi consume input. Nếu out ring đầy (engine
        // pump chậm/stall: WS send bị block khi main-thread browser busy) mà
        // bridge vẫn ghi → ghi đè slot engine đang đọc = read/write pointer
        // giẫm nhau → mẫu torn → méo "wobble" kéo dài tới khi reset. Khi đầy:
        // KHÔNG xử lí batch này (nhường engine drain), input backlog dồn về
        // engine → engine drop batch cũ nhất (drop sạch, worklet hold+fade che
        // gap) — không bao giờ ghi đè out slot chưa đọc.
        const uint32_t outFree = ipc->h.outSlots - (ipc->h.outWrite - ipc->h.outRead);
        if (outFree < std::min<uint32_t>(ipc->h.inWrite - ipc->h.inRead, FXRT_IN_SLOTS)) {
            ++perfOutFull;
            std::this_thread::sleep_until(std::chrono::steady_clock::now() +
                                          std::chrono::milliseconds(1));
            continue;
        }
        uint32_t avail = ipc->h.inWrite - ipc->h.inRead;
        if (avail == 0) {
            ++perfIdle;
            // sleep_until: pacing chính xác theo steady_clock (Sleep(1) bị
            // granularity 15.6ms khi không có timeBeginPeriod → jitter loop
            // tới 32ms, phá headroom FILL của worklet). timeBeginPeriod(1)
            // ở đầu loop + sleep_until giữ loopMax ~2ms.
            std::this_thread::sleep_until(std::chrono::steady_clock::now() +
                                          std::chrono::milliseconds(1));
            continue;
        }
        // Batch: tiêu thụ tối đa FXRT_IN_SLOTS block, xử lí 1 call. Input về
        // đều → take=1 (latency thấp nhất); dồn burst → take>1 gom lại.
        const uint32_t take = std::min<uint32_t>(avail, FXRT_IN_SLOTS);
        uint32_t off = 0;
        for (uint32_t i = 0; i < take; ++i) {
            const uint32_t slot = ipc->h.inRead & inMask;
            std::memcpy(L + off, ipc->inL[slot], n * sizeof(float));
            std::memcpy(R + off, ipc->inR[slot], n * sizeof(float));
            ipc->h.inRead++;  // consume
            off += n;
        }
        const auto perfP0 = std::chrono::steady_clock::now();
        chain.process(L, R, off);  // SEH-guarded; chain rỗng = passthrough
        const auto perfP1 = std::chrono::steady_clock::now();
        const double perfProcMs = std::chrono::duration<double, std::milli>(perfP1 - perfP0).count();
        ++perfProc; perfProcSum += perfProcMs; perfTakeSum += take;
        if (perfProcMs > perfProcMax) perfProcMax = perfProcMs;
        if (perfProcMs < perfProcMin) perfProcMin = perfProcMs;
        for (uint32_t i = 0, o = 0; i < take; ++i, o += n) {
            if (fadeLeft > 0) {
                const uint32_t f = std::min<uint32_t>(fadeLeft, n);
                const float start = (float)(fadeTotal - fadeLeft);
                for (uint32_t k = 0; k < f; ++k) {
                    const float g = (start + (float)k) / (float)fadeTotal;
                    L[o + k] *= g;
                    R[o + k] *= g;
                }
                fadeLeft -= f;
            }
            const uint32_t oslot = ipc->h.outWrite & outMask;
            std::memcpy(ipc->outL[oslot], L + o, n * sizeof(float));
            std::memcpy(ipc->outR[oslot], R + o, n * sizeof(float));
            MemoryBarrier();  // dữ liệu trước index (đúng thứ tự trên ARM64)
            ipc->h.outWrite++;  // publish
            ++processed;
        }
        const auto perfIterT1 = std::chrono::steady_clock::now();
        const double perfLoopMs = std::chrono::duration<double, std::milli>(perfIterT1 - perfT0).count();
        perfT0 = perfIterT1; ++perfIter;
        perfLoopSum += perfLoopMs; if (perfLoopMs > perfLoopMax) perfLoopMax = perfLoopMs;
        if (perfIter % 100 == 0) {
            const double runSec = std::chrono::duration<double>(std::chrono::steady_clock::now() - perfRunStart).count();
            std::cerr << "[FxRTPerf] iter=" << perfIter
                      << " procN=" << perfProc
                      << " procAvg=" << (perfProc ? perfProcSum / perfProc : 0.0) << "ms"
                      << " procMax=" << perfProcMax << "ms"
                      << " procMin=" << (perfProc ? perfProcMin : 0.0) << "ms"
                      << " loopAvg=" << (perfIter ? perfLoopSum / perfIter : 0.0) << "ms"
                      << " loopMax=" << perfLoopMax << "ms"
                      << " idle=" << perfIdle
                      << " outFull=" << perfOutFull
                      << " takeAvg=" << (perfProc ? (double)perfTakeSum / perfProc : 0.0)
                      << " procBlocks=" << processed
                      << " rate=" << (runSec > 0 ? processed / runSec : 0.0) << "blk/s"
                      << " outDepth=" << (ipc->h.outWrite - ipc->h.outRead)
                      << std::endl;
            perfProc = 0; perfProcSum = 0.0; perfTakeSum = 0; perfIdle = 0; perfOutFull = 0;
        }
    }
    ipc->h.state = FXRT_STATE_STARTING;  // đã dừng
    hb.join();
    closeShm(v);
#ifdef _WIN32
    timeEndPeriod(1);
#endif
    std::cerr << "[RealtimeFxLoop] exit processed=" << processed << std::endl;
    return 0;
}