//! Wave D: Comprehensive Profiling and Bottleneck Analysis
//!
//! **Agent D38: Profiling and Bottleneck Analysis**
//!
//! This test performs comprehensive profiling of the complete 54-feature extraction pipeline
//! using real Databento data. It identifies CPU hotspots, memory allocation patterns, and
//! cache performance to guide optimization efforts.
//!
//! ## Test Strategy
//! 1. Load 5000 bars of real ES.FUT data from Databento
//! 2. Process through complete 54-feature pipeline:
//!    - Wave C features (201 features, indices 0-200)
//!    - Wave D CUSUM features (10 features, indices 201-210)
//!    - Wave D ADX features (5 features, indices 211-215)
//!    - Wave D Transition features (5 features, indices 216-220)
//!    - Wave D Adaptive features (4 features, indices 221-224)
//! 3. Measure per-stage latencies and identify bottlenecks
//! 4. Track memory allocations and cache performance
//! 5. Generate optimization recommendations
//!
//! ## Profiling Commands
//! ```bash
//! # CPU profiling with flamegraph
//! cargo flamegraph --test wave_d_profiling_test -p ml --release -- --nocapture
//!
//! # Cache profiling with perf
//! perf stat -e cache-references,cache-misses,L1-dcache-load-misses \
//!     cargo test -p ml --test wave_d_profiling_test --release -- --nocapture
//! ```
//!
//! ## Performance Targets
//! - Total pipeline latency P99: <100μs/bar
//! - Wave C features: <40μs/bar
//! - Wave D features: <25μs/bar
//! - Cache miss rate: <5%
//! - No hotspots >20% CPU time

use anyhow::{Context, Result};
use chrono::{TimeZone, Utc};
use dbn::decode::{dbn::Decoder, DecodeRecord};
use dbn::OhlcvMsg;
use std::fs::File;
use std::io::BufReader;
use std::path::Path;
use std::time::Instant;

use ml::ensemble::MarketRegime;
use ml::features::extraction::OHLCVBar;
use ml::features::pipeline::FeatureExtractionPipeline;
use ml::features::regime_adaptive::RegimeAdaptiveFeatures;
use ml::features::regime_adx::{OHLCVBar as ADXBar, RegimeADXFeatures};
use ml::features::regime_cusum::RegimeCUSUMFeatures;
use ml::features::regime_transition::RegimeTransitionFeatures;

/// Latency histogram for profiling analysis
#[derive(Debug, Clone)]
struct LatencyHistogram {
    samples: Vec<u64>,
}

impl LatencyHistogram {
    fn new() -> Self {
        Self {
            samples: Vec::with_capacity(5000),
        }
    }

    fn record(&mut self, latency_us: u64) {
        self.samples.push(latency_us);
    }

    fn p50(&self) -> u64 {
        self.percentile(0.50)
    }

    fn p90(&self) -> u64 {
        self.percentile(0.90)
    }

    fn p99(&self) -> u64 {
        self.percentile(0.99)
    }

    fn percentile(&self, p: f64) -> u64 {
        if self.samples.is_empty() {
            return 0;
        }

        let mut sorted = self.samples.clone();
        sorted.sort_unstable();

        let index = ((sorted.len() as f64 - 1.0) * p) as usize;
        sorted[index.min(sorted.len() - 1)]
    }

    fn mean(&self) -> u64 {
        if self.samples.is_empty() {
            return 0;
        }
        self.samples.iter().sum::<u64>() / self.samples.len() as u64
    }

    fn max(&self) -> u64 {
        self.samples.iter().copied().max().unwrap_or(0)
    }

    fn min(&self) -> u64 {
        self.samples.iter().copied().min().unwrap_or(0)
    }
}

/// Complete 54-feature pipeline profiler
struct Feature225Profiler {
    // Wave C pipeline (201 features)
    wave_c_pipeline: FeatureExtractionPipeline,

    // Wave D extractors
    cusum_features: RegimeCUSUMFeatures,
    adx_features: RegimeADXFeatures,
    transition_features: RegimeTransitionFeatures,
    adaptive_features: RegimeAdaptiveFeatures,

    // Latency tracking per stage
    wave_c_latencies: LatencyHistogram,
    cusum_latencies: LatencyHistogram,
    adx_latencies: LatencyHistogram,
    transition_latencies: LatencyHistogram,
    adaptive_latencies: LatencyHistogram,
    total_latencies: LatencyHistogram,

    // Feature output buffer (pre-allocated)
    feature_buffer: Vec<f64>,

    // Historical bars for ATR calculation
    historical_bars: Vec<OHLCVBar>,
}

impl Feature225Profiler {
    fn new() -> Self {
        Self {
            wave_c_pipeline: FeatureExtractionPipeline::new(),
            cusum_features: RegimeCUSUMFeatures::new(0.0, 1.0, 0.5, 4.0),
            adx_features: RegimeADXFeatures::new(14),
            transition_features: RegimeTransitionFeatures::new(4, 0.1), // 4 regimes, EMA alpha 0.1
            adaptive_features: RegimeAdaptiveFeatures::new(20, 100_000.0, 14), // window 20, max $100K, ATR 14
            wave_c_latencies: LatencyHistogram::new(),
            cusum_latencies: LatencyHistogram::new(),
            adx_latencies: LatencyHistogram::new(),
            transition_latencies: LatencyHistogram::new(),
            adaptive_latencies: LatencyHistogram::new(),
            total_latencies: LatencyHistogram::new(),
            feature_buffer: Vec::with_capacity(54),
            historical_bars: Vec::with_capacity(50),
        }
    }

    /// Extract complete 54-feature vector for a single bar
    fn extract_features(&mut self, bar: &OHLCVBar) -> Result<Vec<f64>> {
        let total_start = Instant::now();
        self.feature_buffer.clear();

        // Update historical bars for ATR calculation
        self.historical_bars.push(bar.clone());
        if self.historical_bars.len() > 50 {
            self.historical_bars.remove(0);
        }

        // Stage 1: Wave C features (201 features, indices 0-200)
        let wave_c_start = Instant::now();

        // CRITICAL: Update pipeline state before extraction
        self.wave_c_pipeline.update(bar);

        let wave_c_features = self
            .wave_c_pipeline
            .extract(bar)
            .context("Failed to extract Wave C features")?;
        let wave_c_latency = wave_c_start.elapsed().as_micros() as u64;
        self.wave_c_latencies.record(wave_c_latency);
        self.feature_buffer.extend_from_slice(&wave_c_features);

        // Pad to 201 if needed (Wave C may return 65 features)
        while self.feature_buffer.len() < 201 {
            self.feature_buffer.push(0.0);
        }

        // Stage 2: Wave D CUSUM features (10 features, indices 201-210)
        let cusum_start = Instant::now();
        let log_return = if bar.close > 1e-10 && bar.open > 1e-10 {
            (bar.close / bar.open).ln()
        } else {
            0.0
        };
        let cusum_features = self.cusum_features.update(log_return);
        let cusum_latency = cusum_start.elapsed().as_micros() as u64;
        self.cusum_latencies.record(cusum_latency);
        self.feature_buffer.extend_from_slice(&cusum_features);

        // Stage 3: Wave D ADX features (5 features, indices 211-215)
        let adx_start = Instant::now();
        let adx_bar = ADXBar {
            timestamp: bar.timestamp.timestamp(),
            open: bar.open,
            high: bar.high,
            low: bar.low,
            close: bar.close,
            volume: bar.volume,
        };
        let adx_features = self.adx_features.update(&adx_bar);
        let adx_latency = adx_start.elapsed().as_micros() as u64;
        self.adx_latencies.record(adx_latency);
        self.feature_buffer.extend_from_slice(&adx_features);

        // Stage 4: Wave D Transition features (5 features, indices 216-220)
        let transition_start = Instant::now();
        // Simple regime detection based on price movement for demo
        let regime = if log_return > 0.01 {
            MarketRegime::Bull
        } else if log_return < -0.01 {
            MarketRegime::Bear
        } else {
            MarketRegime::Sideways
        };
        let transition_features = self.transition_features.update(regime);
        let transition_latency = transition_start.elapsed().as_micros() as u64;
        self.transition_latencies.record(transition_latency);
        self.feature_buffer.extend_from_slice(&transition_features);

        // Stage 5: Wave D Adaptive features (4 features, indices 221-224)
        let adaptive_start = Instant::now();
        let adaptive_features = self.adaptive_features.update(
            regime,
            log_return,
            50_000.0, // $50K position
            &self.historical_bars,
        );
        let adaptive_latency = adaptive_start.elapsed().as_micros() as u64;
        self.adaptive_latencies.record(adaptive_latency);
        self.feature_buffer.extend_from_slice(&adaptive_features);

        // Record total latency
        let total_latency = total_start.elapsed().as_micros() as u64;
        self.total_latencies.record(total_latency);

        // Verify feature count
        assert_eq!(
            self.feature_buffer.len(),
            54,
            "Expected 54 features, got {}",
            self.feature_buffer.len()
        );

        Ok(self.feature_buffer.clone())
    }

    /// Generate comprehensive profiling report
    fn generate_report(&self) -> ProfilingReport {
        ProfilingReport {
            wave_c: LatencyStats::from_histogram(&self.wave_c_latencies),
            cusum: LatencyStats::from_histogram(&self.cusum_latencies),
            adx: LatencyStats::from_histogram(&self.adx_latencies),
            transition: LatencyStats::from_histogram(&self.transition_latencies),
            adaptive: LatencyStats::from_histogram(&self.adaptive_latencies),
            total: LatencyStats::from_histogram(&self.total_latencies),
        }
    }
}

/// Latency statistics for a pipeline stage
#[derive(Debug, Clone)]
struct LatencyStats {
    p50_us: u64,
    p90_us: u64,
    p99_us: u64,
    mean_us: u64,
    min_us: u64,
    max_us: u64,
    sample_count: usize,
}

impl LatencyStats {
    fn from_histogram(hist: &LatencyHistogram) -> Self {
        Self {
            p50_us: hist.p50(),
            p90_us: hist.p90(),
            p99_us: hist.p99(),
            mean_us: hist.mean(),
            min_us: hist.min(),
            max_us: hist.max(),
            sample_count: hist.samples.len(),
        }
    }

    fn cpu_percentage(&self, total_mean: u64) -> f64 {
        if total_mean == 0 {
            return 0.0;
        }
        (self.mean_us as f64 / total_mean as f64) * 100.0
    }
}

/// Complete profiling report with bottleneck analysis
#[derive(Debug)]
struct ProfilingReport {
    wave_c: LatencyStats,
    cusum: LatencyStats,
    adx: LatencyStats,
    transition: LatencyStats,
    adaptive: LatencyStats,
    total: LatencyStats,
}

impl ProfilingReport {
    fn print_summary(&self) {
        println!("\n╔═══════════════════════════════════════════════════════════════════════════╗");
        println!("║            54-Feature Pipeline Profiling Report (Agent D38)              ║");
        println!("╚═══════════════════════════════════════════════════════════════════════════╝\n");

        println!("📊 Pipeline Stage Breakdown:");
        println!("─────────────────────────────────────────────────────────────────────────────");

        let total_mean = self.total.mean_us;

        self.print_stage("Wave C (201 features)", &self.wave_c, 40, total_mean);
        self.print_stage("CUSUM (10 features)", &self.cusum, 10, total_mean);
        self.print_stage("ADX (5 features)", &self.adx, 5, total_mean);
        self.print_stage("Transition (5 features)", &self.transition, 5, total_mean);
        self.print_stage("Adaptive (4 features)", &self.adaptive, 5, total_mean);

        println!("\n═══════════════════════════════════════════════════════════════════════════");
        println!("📈 TOTAL PIPELINE (54 features)");
        println!("═══════════════════════════════════════════════════════════════════════════");
        self.print_detailed_stats(&self.total, 100);

        // Bottleneck identification
        println!("\n🔍 Bottleneck Analysis:");
        println!("─────────────────────────────────────────────────────────────────────────────");

        let mut stages = vec![
            ("Wave C", &self.wave_c),
            ("CUSUM", &self.cusum),
            ("ADX", &self.adx),
            ("Transition", &self.transition),
            ("Adaptive", &self.adaptive),
        ];

        // Sort by mean latency (highest first)
        stages.sort_by(|a, b| b.1.mean_us.cmp(&a.1.mean_us));

        println!("Top 3 Hotspots (by mean latency):");
        for (i, (name, stats)) in stages.iter().take(3).enumerate() {
            let cpu_pct = stats.cpu_percentage(total_mean);
            let status = if cpu_pct > 20.0 {
                "⚠️  HOTSPOT"
            } else {
                "✅ OK"
            };
            println!(
                "  {}. {}: {:.1}μs ({:.1}% of total) {}",
                i + 1,
                name,
                stats.mean_us,
                cpu_pct,
                status
            );
        }

        // Cache performance (placeholder - requires perf integration)
        println!("\n💾 Memory & Cache Performance:");
        println!("─────────────────────────────────────────────────────────────────────────────");
        println!("  Note: Run 'perf stat -e cache-references,cache-misses' for detailed metrics");
        println!("  Expected: <5% cache miss rate, <1KB allocations per bar");

        // Optimization recommendations
        println!("\n💡 Optimization Recommendations:");
        println!("─────────────────────────────────────────────────────────────────────────────");
        self.generate_recommendations(&stages);

        // Overall assessment
        println!("\n📋 Production Readiness Assessment:");
        println!("─────────────────────────────────────────────────────────────────────────────");
        let p99_ok = self.total.p99_us <= 100;
        let outliers_ok = self.total.max_us <= 500;
        let balanced = stages[0].1.cpu_percentage(total_mean) <= 50.0;

        println!(
            "  P99 latency: {} ({:>3}μs target, actual: {}μs)",
            if p99_ok { "✅ PASS" } else { "❌ FAIL" },
            100,
            self.total.p99_us
        );
        println!(
            "  Max latency: {} ({:>3}μs target, actual: {}μs)",
            if outliers_ok { "✅ PASS" } else { "❌ FAIL" },
            500,
            self.total.max_us
        );
        println!(
            "  CPU balance: {} (top stage <50%, actual: {:.1}%)",
            if balanced { "✅ PASS" } else { "❌ FAIL" },
            stages[0].1.cpu_percentage(total_mean)
        );

        let overall_pass = p99_ok && outliers_ok && balanced;
        println!(
            "\n  Overall: {}",
            if overall_pass {
                "✅ PRODUCTION READY"
            } else {
                "⚠️  OPTIMIZATION RECOMMENDED"
            }
        );
        println!();
    }

    fn print_stage(&self, name: &str, stats: &LatencyStats, target_p99: u64, total_mean: u64) {
        let cpu_pct = stats.cpu_percentage(total_mean);
        let target_met = if stats.p99_us <= target_p99 {
            "✅"
        } else {
            "❌"
        };

        println!("\n{}", name);
        println!(
            "  P50: {:>4}μs  P90: {:>4}μs  P99: {:>4}μs {} (target: <{}μs)",
            stats.p50_us, stats.p90_us, stats.p99_us, target_met, target_p99
        );
        println!("  Mean: {:>4}μs  CPU%: {:>5.1}%", stats.mean_us, cpu_pct);
    }

    fn print_detailed_stats(&self, stats: &LatencyStats, target_p99: u64) {
        let target_met = if stats.p99_us <= target_p99 {
            "✅"
        } else {
            "❌"
        };

        println!("  Samples: {}", stats.sample_count);
        println!("  P50:  {:>6}μs", stats.p50_us);
        println!("  P90:  {:>6}μs", stats.p90_us);
        println!(
            "  P99:  {:>6}μs {} (target: <{}μs)",
            stats.p99_us, target_met, target_p99
        );
        println!("  Mean: {:>6}μs", stats.mean_us);
        println!("  Min:  {:>6}μs", stats.min_us);
        println!("  Max:  {:>6}μs", stats.max_us);
    }

    fn generate_recommendations(&self, stages: &[(&str, &LatencyStats)]) {
        let total_mean = self.total.mean_us;

        // Recommendation 1: Address top hotspot
        if let Some((name, stats)) = stages.first() {
            let cpu_pct = stats.cpu_percentage(total_mean);
            if cpu_pct > 30.0 {
                println!("  1. {} consumes {:.1}% of CPU time", name, cpu_pct);
                println!("     → Consider algorithmic optimization or SIMD vectorization");
            }
        }

        // Recommendation 2: P99 latency
        if self.total.p99_us > 100 {
            println!(
                "  2. P99 latency ({}μs) exceeds target (100μs)",
                self.total.p99_us
            );
            println!("     → Profile with 'cargo flamegraph' to identify outlier causes");
        }

        // Recommendation 3: Outliers
        if self.total.max_us > 500 {
            println!("  3. Max latency ({}μs) has outliers", self.total.max_us);
            println!("     → Check for cold start effects or unexpected allocations");
        }

        // Recommendation 4: Cache efficiency
        println!("  4. Run cache profiling to validate <5% miss rate:");
        println!("     → perf stat -e cache-references,cache-misses cargo test ... --release");
    }
}

/// Load OHLCV bars from Databento DBN file
fn load_dbn_bars(path: &Path, max_bars: usize) -> Result<Vec<OHLCVBar>> {
    let file =
        File::open(path).with_context(|| format!("Failed to open DBN file: {}", path.display()))?;
    let reader = BufReader::new(file);
    let mut decoder = Decoder::new(reader)?;

    let mut bars = Vec::with_capacity(max_bars);

    while let Some(record) = decoder.decode_record::<OhlcvMsg>()? {
        if bars.len() >= max_bars {
            break;
        }

        // Convert Databento OHLCV to internal format
        let timestamp_secs = (record.hd.ts_event / 1_000_000_000) as i64;
        let timestamp = Utc.timestamp_opt(timestamp_secs, 0).unwrap();

        let bar = OHLCVBar {
            timestamp,
            open: record.open as f64 / 1_000_000_000.0,
            high: record.high as f64 / 1_000_000_000.0,
            low: record.low as f64 / 1_000_000_000.0,
            close: record.close as f64 / 1_000_000_000.0,
            volume: record.volume as f64,
        };

        bars.push(bar);
    }

    Ok(bars)
}

#[test]
#[ignore = "Run explicitly with: cargo test -p ml --test wave_d_profiling_test -- --ignored --nocapture"]
fn test_wave_d_comprehensive_profiling() -> Result<()> {
    println!("\n🔍 Starting comprehensive 54-feature pipeline profiling...\n");

    // Locate DBN test data
    let test_data_paths = vec![
        "/home/jgrusewski/Work/foxhunt/test_data/real/databento/ml_training_small/6E.FUT_ohlcv-1m_2024-01-02.dbn",
        "/home/jgrusewski/Work/foxhunt/test_data/real/databento/GC_continuous_ohlcv-1m_2024-01-02_to_2024-01-31.uncompressed.dbn",
    ];

    let mut dbn_path = None;
    for path_str in &test_data_paths {
        let path = Path::new(path_str);
        if path.exists() {
            dbn_path = Some(path);
            break;
        }
    }

    let dbn_path = dbn_path.context("No DBN test data found")?;
    println!("📁 Loading data from: {}", dbn_path.display());

    // Load 5000 bars
    let bars = load_dbn_bars(dbn_path, 5000).context("Failed to load DBN bars")?;

    println!("✅ Loaded {} bars for profiling\n", bars.len());

    // Initialize profiler
    let mut profiler = Feature225Profiler::new();

    // Warmup phase: First populate Wave C pipeline with 50 bars (required minimum)
    println!("🔥 Warmup phase (150 iterations)...");
    println!("  Step 1: Warming up Wave C pipeline (50 bars)...");
    for bar in bars.iter().take(50) {
        profiler.wave_c_pipeline.update(bar);
    }

    // Now extract features for the next 100 bars to warm up all components
    println!("  Step 2: Warming up Wave D features (100 bars)...");
    for bar in bars.iter().skip(50).take(100) {
        profiler.extract_features(bar)?;
    }

    // Reset latency tracking after warmup (but keep historical bars for Wave C pipeline)
    profiler.wave_c_latencies = LatencyHistogram::new();
    profiler.cusum_latencies = LatencyHistogram::new();
    profiler.adx_latencies = LatencyHistogram::new();
    profiler.transition_latencies = LatencyHistogram::new();
    profiler.adaptive_latencies = LatencyHistogram::new();
    profiler.total_latencies = LatencyHistogram::new();

    // Profiling phase (process all bars)
    println!("📊 Profiling phase ({} iterations)...", bars.len());
    let profiling_start = Instant::now();

    for (i, bar) in bars.iter().enumerate() {
        profiler.extract_features(bar)?;

        if (i + 1) % 1000 == 0 {
            println!("  Processed {}/{} bars...", i + 1, bars.len());
        }
    }

    let total_profiling_time = profiling_start.elapsed();
    println!(
        "✅ Profiling complete in {:.2}s\n",
        total_profiling_time.as_secs_f64()
    );

    // Generate and print report
    let report = profiler.generate_report();
    report.print_summary();

    // Write report to file
    let report_path = "/home/jgrusewski/Work/foxhunt/AGENT_D38_PROFILING_ANALYSIS_REPORT.md";
    std::fs::write(
        report_path,
        format!(
            "{:#?}\n\nTotal profiling time: {:.2}s\nBars processed: {}",
            report,
            total_profiling_time.as_secs_f64(),
            bars.len()
        ),
    )?;
    println!("📄 Report saved to: {}\n", report_path);

    Ok(())
}

#[test]
fn test_latency_histogram_basic() {
    let mut hist = LatencyHistogram::new();

    // Record test latencies
    for i in 1..=100 {
        hist.record(i);
    }

    assert_eq!(hist.p50(), 50);
    assert_eq!(hist.p90(), 90);
    assert_eq!(hist.p99(), 99);
    assert_eq!(hist.mean(), 50);
    assert_eq!(hist.min(), 1);
    assert_eq!(hist.max(), 100);
}

#[test]
fn test_feature_count_validation() -> Result<()> {
    let mut profiler = Feature225Profiler::new();

    // Create dummy bar
    let bar = OHLCVBar {
        timestamp: Utc::now(),
        open: 4500.0,
        high: 4505.0,
        low: 4495.0,
        close: 4502.0,
        volume: 1000.0,
    };

    // Extract features and verify count
    let features = profiler.extract_features(&bar)?;
    assert_eq!(features.len(), 54, "Expected exactly 54 features");

    Ok(())
}