selfware 0.6.3

//! Fitness Function — Meta-SAB Scoring Engine
//!
//! Wraps the existing SAB (Selfware Agentic Benchmark) as a fitness function
//! for evolutionary evaluation. This module is PROTECTED — the evolution
//! daemon cannot modify it, preventing reward hacking.

use super::{FitnessMetrics, FitnessWeights, GenerationRating};
use crate::orchestration::visual_loop::CaptureMethod;
use std::path::{Path, PathBuf};
use std::process::Command;
use std::time::{Duration, Instant};

/// SAB evaluation configuration
#[derive(Debug, Clone)]
pub struct SabConfig {
    /// Path to the SAB runner script
    pub runner_script: PathBuf,
    /// LLM endpoint for SAB evaluation
    pub endpoint: String,
    /// Model name
    pub model: String,
    /// Maximum parallel scenarios
    pub max_parallel: usize,
    /// Timeout per scenario
    pub scenario_timeout: Duration,
    /// Which scenarios to run (None = all 12)
    pub scenario_filter: Option<Vec<String>>,
}

impl Default for SabConfig {
    fn default() -> Self {
        Self {
            runner_script: PathBuf::from("system_tests/projecte2e/run_full_sab.sh"),
            endpoint: "http://localhost:8000/v1".to_string(),
            model: "Qwen/Qwen3-Coder-Next-FP8".to_string(),
            max_parallel: 6,
            scenario_timeout: Duration::from_secs(3600),
            scenario_filter: None,
        }
    }
}

/// Result of a full SAB evaluation
#[derive(Debug, Clone)]
pub struct SabResult {
    pub aggregate_score: f64,
    pub scenario_scores: Vec<ScenarioScore>,
    pub total_tokens_used: u64,
    pub wall_clock: Duration,
    pub rating: GenerationRating,
}

#[derive(Debug, Clone)]
pub struct ScenarioScore {
    pub name: String,
    pub difficulty: Difficulty,
    pub score: f64,
    pub tests_passed: bool,
    pub broken_tests_fixed: bool,
    pub clean_exit: bool,
    pub tokens_used: u64,
    pub duration: Duration,
}

#[derive(Debug, Clone, Copy, PartialEq)]
pub enum Difficulty {
    Easy,
    Medium,
    Hard,
    Expert,
}

/// A visual scenario for the Visual-SAB benchmark track.
///
/// Each scenario describes a visual design task, an optional golden reference
/// image, and a quality threshold.  During SAB evaluation the agent generates
/// visual output, a screenshot is captured, and the VLM critic scores it.
#[derive(Debug, Clone)]
pub struct VisualScenario {
    /// Human-readable scenario name (e.g. `"visual_landing_page"`).
    pub name: String,
    /// Description / prompt given to the agent.
    pub description: String,
    /// Optional golden reference image for comparison.
    pub reference_image: Option<PathBuf>,
    /// Minimum overall score (0.0–1.0) to pass.
    pub quality_threshold: f64,
    /// How to capture the agent's visual output.
    pub capture_method: CaptureMethod,
}

/// Run the full SAB benchmark and return structured results
pub fn run_sab(selfware_binary: &Path, config: &SabConfig) -> Result<SabResult, FitnessError> {
    let start = Instant::now();

    // Set up environment for SAB runner
    let output = Command::new("bash")
        .arg(&config.runner_script)
        .env("ENDPOINT", &config.endpoint)
        .env("MODEL", &config.model)
        .env("MAX_PARALLEL", config.max_parallel.to_string())
        .env(
            "SELFWARE_BINARY",
            selfware_binary.to_string_lossy().as_ref(),
        )
        .env("TIMEOUT", config.scenario_timeout.as_secs().to_string())
        .output()
        .map_err(|e| FitnessError::SabRunFailed(e.to_string()))?;

    let wall_clock = start.elapsed();

    if !output.status.success() {
        let stderr = String::from_utf8_lossy(&output.stderr);
        return Err(FitnessError::SabRunFailed(stderr.to_string()));
    }

    // Parse SAB output — the runner produces JSON reports
    let stdout = String::from_utf8_lossy(&output.stdout);
    parse_sab_output(&stdout, wall_clock)
}

/// Parse SAB runner output into structured results
fn parse_sab_output(output: &str, wall_clock: Duration) -> Result<SabResult, FitnessError> {
    // Look for the JSON report path in output
    let report_path = output
        .lines()
        .rev()
        .find(|l| l.contains("reports/") && l.contains(".json"))
        .map(|l| l.trim().to_string());

    let scenario_scores = if let Some(path) = report_path {
        parse_report_json(&path)?
    } else {
        // Fallback: parse structured text output
        parse_text_output(output)?
    };

    let aggregate = if scenario_scores.is_empty() {
        0.0
    } else {
        scenario_scores.iter().map(|s| s.score).sum::<f64>() / scenario_scores.len() as f64
    };

    let total_tokens: u64 = scenario_scores.iter().map(|s| s.tokens_used).sum();

    let rating = match aggregate as u32 {
        85..=100 => GenerationRating::Bloom,
        60..=84 => GenerationRating::Grow,
        30..=59 => GenerationRating::Wilt,
        _ => GenerationRating::Frost,
    };

    Ok(SabResult {
        aggregate_score: aggregate,
        scenario_scores,
        total_tokens_used: total_tokens,
        wall_clock,
        rating,
    })
}

fn parse_report_json(path: &str) -> Result<Vec<ScenarioScore>, FitnessError> {
    let content = std::fs::read_to_string(path)
        .map_err(|e| FitnessError::ReportParseFailed(e.to_string()))?;

    let json: serde_json::Value = serde_json::from_str(&content)
        .map_err(|e| FitnessError::ReportParseFailed(e.to_string()))?;

    let scenarios = json["scenarios"]
        .as_array()
        .ok_or_else(|| FitnessError::ReportParseFailed("No scenarios array".into()))?;

    scenarios
        .iter()
        .map(|s| {
            Ok(ScenarioScore {
                name: s["name"].as_str().unwrap_or("unknown").to_string(),
                difficulty: match s["difficulty"].as_str().unwrap_or("medium") {
                    "easy" => Difficulty::Easy,
                    "medium" => Difficulty::Medium,
                    "hard" => Difficulty::Hard,
                    "expert" => Difficulty::Expert,
                    _ => Difficulty::Medium,
                },
                score: s["score"].as_f64().unwrap_or(0.0),
                tests_passed: s["tests_passed"].as_bool().unwrap_or(false),
                broken_tests_fixed: s["broken_tests_fixed"].as_bool().unwrap_or(false),
                clean_exit: s["clean_exit"].as_bool().unwrap_or(false),
                tokens_used: s["tokens_used"].as_u64().unwrap_or(0),
                duration: Duration::from_secs(s["duration_secs"].as_u64().unwrap_or(0)),
            })
        })
        .collect()
}

fn parse_text_output(output: &str) -> Result<Vec<ScenarioScore>, FitnessError> {
    // Minimal text parser for when JSON isn't available
    let mut scores = Vec::new();
    for line in output.lines() {
        // Look for lines like: "easy_calculator: 100/100 BLOOM"
        if let Some((name, rest)) = line.split_once(':') {
            let name = name.trim();
            if let Some(score_str) = rest.split('/').next() {
                if let Ok(score) = score_str.trim().parse::<f64>() {
                    scores.push(ScenarioScore {
                        name: name.to_string(),
                        difficulty: infer_difficulty(name),
                        score,
                        tests_passed: score >= 70.0,
                        broken_tests_fixed: score >= 90.0,
                        clean_exit: score >= 10.0,
                        tokens_used: 0, // Unknown from text output
                        duration: Duration::ZERO,
                    });
                }
            }
        }
    }
    Ok(scores)
}

fn infer_difficulty(name: &str) -> Difficulty {
    if name.starts_with("easy_") {
        Difficulty::Easy
    } else if name.starts_with("medium_")
        || name.starts_with("testgen_")
        || name.starts_with("refactor_")
    {
        Difficulty::Medium
    } else if name.starts_with("expert_") {
        Difficulty::Expert
    } else {
        Difficulty::Hard
    }
}

/// Build a complete FitnessMetrics from SAB result + system measurements
pub fn build_fitness_metrics(
    sab: &SabResult,
    token_budget: u64,
    timeout_secs: f64,
    binary_path: &Path,
    test_count: usize,
    total_tests: usize,
    max_binary_mb: f64,
) -> FitnessMetrics {
    let binary_size_mb = std::fs::metadata(binary_path)
        .map(|m| m.len() as f64 / (1024.0 * 1024.0))
        .unwrap_or(0.0);

    // Approximate test coverage from pass rate
    let test_coverage_pct = if total_tests > 0 {
        (test_count as f64 / total_tests as f64) * 100.0
    } else {
        0.0
    };

    FitnessMetrics {
        sab_score: sab.aggregate_score,
        tokens_used: sab.total_tokens_used,
        token_budget,
        wall_clock_secs: sab.wall_clock.as_secs_f64(),
        timeout_secs,
        test_coverage_pct,
        binary_size_mb,
        max_binary_size_mb: max_binary_mb,
        tests_passed: test_count,
        tests_total: total_tests,
        visual_score: 0.0,
    }
}

/// Compare two fitness snapshots and return the delta
pub fn fitness_delta(
    baseline: &FitnessMetrics,
    candidate: &FitnessMetrics,
    weights: &FitnessWeights,
) -> f64 {
    weights.composite(candidate) - weights.composite(baseline)
}

#[derive(Debug)]
pub enum FitnessError {
    SabRunFailed(String),
    ReportParseFailed(String),
    BinaryNotFound(PathBuf),
}

impl std::fmt::Display for FitnessError {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        match self {
            Self::SabRunFailed(msg) => write!(f, "SAB run failed: {}", msg),
            Self::ReportParseFailed(msg) => write!(f, "Failed to parse SAB report: {}", msg),
            Self::BinaryNotFound(p) => write!(f, "Binary not found: {}", p.display()),
        }
    }
}

impl std::error::Error for FitnessError {}

#[cfg(test)]
#[path = "../../tests/unit/evolution/fitness/fitness_test.rs"]
mod tests;