zclaw_openfang/desktop/src-tauri/src/intelligence/pain_aggregator.rs

//! Pain Point Aggregator — detect, merge, and score recurring user difficulties.
//!
//! Collects pain signals from conversations, merges similar ones with increasing
//! confidence, and surfaces high-confidence pain points for solution generation.

use std::sync::Arc;

use chrono::{DateTime, Utc};
use serde::{Deserialize, Serialize};
use tokio::sync::RwLock;
use tracing::debug;
use uuid::Uuid;
use zclaw_types::Result;

use super::pain_storage::PainStorage;

// ---------------------------------------------------------------------------
// Data structures
// ---------------------------------------------------------------------------

/// A recurring difficulty or systemic problem the user encounters.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct PainPoint {
    pub id: String,
    pub agent_id: String,
    pub user_id: String,
    pub summary: String,
    pub category: String,
    pub severity: PainSeverity,
    pub evidence: Vec<PainEvidence>,
    pub occurrence_count: u32,
    pub first_seen: DateTime<Utc>,
    pub last_seen: DateTime<Utc>,
    pub confidence: f64,
    pub status: PainStatus,
}

/// A single piece of evidence supporting a pain point.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct PainEvidence {
    pub when: DateTime<Utc>,
    pub user_said: String,
    pub why_flagged: String,
}

/// How severe the pain point is.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize, Default)]
#[serde(rename_all = "lowercase")]
pub enum PainSeverity {
    #[default]
    Low,
    Medium,
    High,
}

/// Lifecycle status of a pain point.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum PainStatus {
    Detected,
    Confirmed,
    Solving,
    Solved,
    Dismissed,
}

impl PainPoint {
    /// Create a new pain point with a single evidence entry.
    pub fn new(
        agent_id: &str,
        user_id: &str,
        summary: &str,
        category: &str,
        severity: PainSeverity,
        user_said: &str,
        why_flagged: &str,
    ) -> Self {
        let now = Utc::now();
        Self {
            id: Uuid::new_v4().to_string(),
            agent_id: agent_id.to_string(),
            user_id: user_id.to_string(),
            summary: summary.to_string(),
            category: category.to_string(),
            severity,
            evidence: vec![PainEvidence {
                when: now,
                user_said: user_said.to_string(),
                why_flagged: why_flagged.to_string(),
            }],
            occurrence_count: 1,
            first_seen: now,
            last_seen: now,
            confidence: 0.25,
            status: PainStatus::Detected,
        }
    }
}

// ---------------------------------------------------------------------------
// LLM analysis result
// ---------------------------------------------------------------------------

/// Result of LLM-based pain point analysis on a conversation.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct PainAnalysisResult {
    pub found: bool,
    #[serde(default)]
    pub summary: String,
    #[serde(default)]
    pub category: String,
    #[serde(default)]
    pub severity: PainSeverity,
    #[serde(default)]
    pub evidence: String,
}

// ---------------------------------------------------------------------------
// PainAggregator — cross-session merge + confidence scoring
// ---------------------------------------------------------------------------

/// Aggregates pain points across conversations, merging similar ones
/// and escalating confidence as evidence accumulates.
///
/// When the global `PAIN_STORAGE` is initialized (via `init_pain_storage`),
/// writes are dual: memory Vec (hot cache) + SQLite (durable).
pub struct PainAggregator {
    pain_points: Arc<RwLock<Vec<PainPoint>>>,
}

impl PainAggregator {
    pub fn new() -> Self {
        Self {
            pain_points: Arc::new(RwLock::new(Vec::new())),
        }
    }

    /// Get the global pain storage, if initialized.
    fn get_storage() -> Option<Arc<PainStorage>> {
        PAIN_STORAGE.get().cloned()
    }

    /// Load all persisted pain points from storage into the in-memory cache.
    /// Call this once during app startup after `init_pain_storage()`.
    pub async fn load_from_storage(&self) -> Result<()> {
        if let Some(ref storage) = Self::get_storage() {
            let persisted = storage.load_all_pain_points().await?;
            let mut points = self.pain_points.write().await;
            *points = persisted;
            debug!("[PainAggregator] Loaded {} pain points from storage", points.len());
        }
        Ok(())
    }

    /// Merge a new pain point with an existing similar one, or create a new entry.
    /// Persists to SQLite if storage is configured.
    pub async fn merge_or_create(&self, new_pain: PainPoint) -> Result<PainPoint> {
        let mut points = self.pain_points.write().await;

        let similar_idx = points.iter().position(|p| {
            p.agent_id == new_pain.agent_id
                && p.category == new_pain.category
                && Self::summaries_similar(&p.summary, &new_pain.summary)
        });

        let result = if let Some(idx) = similar_idx {
            let existing = &mut points[idx];
            existing.evidence.extend(new_pain.evidence);
            existing.occurrence_count += 1;
            existing.last_seen = Utc::now();
            existing.confidence = Self::calculate_confidence(existing);
            if new_pain.severity as u8 > existing.severity as u8 {
                existing.severity = new_pain.severity;
            }
            if existing.occurrence_count >= 2 && existing.status == PainStatus::Detected {
                existing.status = PainStatus::Confirmed;
            }
            existing.clone()
        } else {
            let result = new_pain.clone();
            points.push(new_pain);
            result
        };

        // Dual-write: persist to SQLite
        if let Some(storage) = Self::get_storage() {
            if let Err(e) = storage.store_pain_point(&result).await {
                debug!("[PainAggregator] Failed to persist pain point: {}", e);
            }
        }

        Ok(result)
    }

    /// Get all high-confidence pain points for an agent.
    pub async fn get_high_confidence_pains(
        &self,
        agent_id: &str,
        min_confidence: f64,
    ) -> Vec<PainPoint> {
        let points = self.pain_points.read().await;
        points
            .iter()
            .filter(|p| {
                p.agent_id == agent_id
                    && p.confidence >= min_confidence
                    && p.status != PainStatus::Solved
                    && p.status != PainStatus::Dismissed
            })
            .cloned()
            .collect()
    }

    /// List all pain points for an agent.
    pub async fn list_pain_points(&self, agent_id: &str) -> Vec<PainPoint> {
        let points = self.pain_points.read().await;
        points
            .iter()
            .filter(|p| p.agent_id == agent_id)
            .cloned()
            .collect()
    }

    /// List all pain points across all agents (for internal use).
    pub async fn list_all_pain_points(&self) -> Vec<PainPoint> {
        let points = self.pain_points.read().await;
        points.clone()
    }

    /// Update the status of a pain point. Persists to SQLite if configured.
    pub async fn update_status(&self, pain_id: &str, status: PainStatus) -> Result<()> {
        let mut points = self.pain_points.write().await;
        if let Some(p) = points.iter_mut().find(|p| p.id == pain_id) {
            p.status = status;
            // Dual-write
            if let Some(storage) = Self::get_storage() {
                if let Err(e) = storage.store_pain_point(p).await {
                    debug!("[PainAggregator] Failed to persist status update: {}", e);
                }
            }
        }
        Ok(())
    }

    // -- internal helpers ----------------------------------------------------

    fn summaries_similar(a: &str, b: &str) -> bool {
        let a_lower = a.to_lowercase();
        let b_lower = b.to_lowercase();

        if a_lower.contains(&b_lower) || b_lower.contains(&a_lower) {
            return true;
        }

        let a_ngrams = Self::char_ngrams(&a_lower, 2);
        let b_ngrams = Self::char_ngrams(&b_lower, 2);
        if a_ngrams.is_empty() || b_ngrams.is_empty() {
            return false;
        }
        let shared = a_ngrams.iter().filter(|ng| b_ngrams.contains(ng)).count();
        let overlap_ratio = shared as f64 / a_ngrams.len().min(b_ngrams.len()) as f64;
        overlap_ratio >= 0.4
    }

    fn char_ngrams(text: &str, n: usize) -> Vec<String> {
        let chars: Vec<char> = text.chars().collect();
        if chars.len() < n {
            return Vec::new();
        }
        chars.windows(n).map(|w| w.iter().collect()).collect()
    }

    fn calculate_confidence(pain: &PainPoint) -> f64 {
        let base = (pain.occurrence_count as f64 * 0.25).min(0.75);
        let days_since_last = (Utc::now() - pain.last_seen).num_days().max(0) as f64;
        let decay = 1.0 - (days_since_last / 90.0);
        let severity_boost = match pain.severity {
            PainSeverity::High => 0.1,
            PainSeverity::Medium => 0.05,
            PainSeverity::Low => 0.0,
        };
        (base * decay.max(0.0) + severity_boost).min(1.0)
    }
}

impl Default for PainAggregator {
    fn default() -> Self {
        Self::new()
    }
}

// ---------------------------------------------------------------------------
// Rule-based pain signal detection
// ---------------------------------------------------------------------------

const FRUSTRATION_SIGNALS: &[&str] = &[
    "烦死了", "又来了", "每次都", "受不了", "还是不行",
    "又出错", "第三次", "第四次", "第五次", "怎么又",
    "老是", "一直", "还是没解决", "退了", "被退",
    "又要改", "搞不定", "没办法",
];

/// Analyze messages for pain signals using rule-based detection.
pub fn analyze_for_pain_signals(messages: &[zclaw_types::Message]) -> Option<PainAnalysisResult> {
    let mut user_frustrations: Vec<String> = Vec::new();

    for msg in messages {
        if let zclaw_types::Message::User { content } = msg {
            for signal in FRUSTRATION_SIGNALS {
                if content.contains(signal) {
                    user_frustrations.push(content.clone());
                    break;
                }
            }
        }
    }

    if user_frustrations.is_empty() {
        return None;
    }

    let combined = user_frustrations.join(" ");
    let severity = if combined.contains("烦死了")
        || combined.contains("受不了")
        || user_frustrations.len() >= 3
    {
        PainSeverity::High
    } else if user_frustrations.len() >= 2 {
        PainSeverity::Medium
    } else {
        PainSeverity::Low
    };

    let summary = extract_summary(&user_frustrations[0]);
    let category = classify_category(&combined);

    Some(PainAnalysisResult {
        found: true,
        summary,
        category,
        severity,
        evidence: user_frustrations.join("; "),
    })
}

fn extract_summary(text: &str) -> String {
    let trimmed = text.trim();
    if trimmed.chars().count() <= 50 {
        trimmed.to_string()
    } else {
        let end = trimmed.char_indices().take_while(|(i, _)| *i < 50).last();
        match end {
            Some((i, c)) => trimmed[..i + c.len_utf8()].to_string() + "…",
            None => trimmed.chars().take(50).collect::<String>() + "…",
        }
    }
}

fn classify_category(text: &str) -> String {
    let lower = text.to_lowercase();
    if ["包", "货", "出口", "退"].iter().any(|k| lower.contains(k)) {
        "logistics".to_string()
    } else if ["合规", "法规", "标准"].iter().any(|k| lower.contains(k)) {
        "compliance".to_string()
    } else if ["客户", "投诉", "反馈"].iter().any(|k| lower.contains(k)) {
        "customer".to_string()
    } else if ["报价", "价格", "成本"].iter().any(|k| lower.contains(k)) {
        "pricing".to_string()
    } else if ["系统", "软件", "电脑"].iter().any(|k| lower.contains(k)) {
        "technology".to_string()
    } else {
        "general".to_string()
    }
}

// ---------------------------------------------------------------------------
// Tauri commands
// ---------------------------------------------------------------------------

use std::sync::OnceLock;

use super::solution_generator::{Proposal, ProposalStatus, SolutionGenerator};

static PAIN_AGGREGATOR: OnceLock<Arc<PainAggregator>> = OnceLock::new();
static SOLUTION_GENERATOR: OnceLock<Arc<SolutionGenerator>> = OnceLock::new();
pub(crate) static PAIN_STORAGE: OnceLock<Arc<PainStorage>> = OnceLock::new();

fn pain_store() -> Arc<PainAggregator> {
    PAIN_AGGREGATOR.get_or_init(|| Arc::new(PainAggregator::new())).clone()
}

fn solution_store() -> Arc<SolutionGenerator> {
    SOLUTION_GENERATOR.get_or_init(|| Arc::new(SolutionGenerator::new())).clone()
}

/// Initialize pain point persistence with a SQLite pool.
///
/// Creates the schema, sets the global storage, and loads any previously
/// persisted data into the in-memory caches.
///
/// Should be called once during app startup, before any pain-related operations.
pub async fn init_pain_storage(pool: sqlx::SqlitePool) -> Result<()> {
    let storage = Arc::new(PainStorage::new(pool));
    storage.initialize_schema().await?;

    // Set global storage (must succeed on first call)
    PAIN_STORAGE.set(storage.clone()).map_err(|_| zclaw_types::ZclawError::StorageError("PainStorage already initialized".into()))?;

    // Warm the in-memory caches from SQLite
    let aggregator = pain_store();
    aggregator.load_from_storage().await?;

    let generator = solution_store();
    generator.load_from_storage().await?;

    debug!("[init_pain_storage] Pain storage initialized successfully");
    Ok(())
}

/// List all pain points for an agent.
// @reserved: no frontend UI yet
#[tauri::command]
pub async fn butler_list_pain_points(agent_id: String) -> std::result::Result<Vec<PainPoint>, String> {
    Ok(pain_store().list_pain_points(&agent_id).await)
}

/// Record a new pain point from conversation analysis.
// @reserved: no frontend UI yet
#[tauri::command]
pub async fn butler_record_pain_point(
    agent_id: String,
    user_id: String,
    summary: String,
    category: String,
    severity: String,
    user_said: String,
    why_flagged: String,
) -> std::result::Result<PainPoint, String> {
    let sev = match severity.as_str() {
        "high" => PainSeverity::High,
        "medium" => PainSeverity::Medium,
        _ => PainSeverity::Low,
    };
    let pain = PainPoint::new(&agent_id, &user_id, &summary, &category, sev, &user_said, &why_flagged);
    pain_store().merge_or_create(pain).await.map_err(|e| e.to_string())
}

/// List all proposals for an agent.
// @reserved: no frontend UI yet
#[tauri::command]
pub async fn butler_list_proposals(agent_id: String) -> std::result::Result<Vec<Proposal>, String> {
    let store = pain_store();
    let gen = solution_store();
    let pains = store.list_pain_points(&agent_id).await;
    let mut all_proposals = Vec::new();
    for pain in &pains {
        let proposals = gen.list_proposals(&[pain.id.clone()]).await;
        all_proposals.extend(proposals);
    }
    Ok(all_proposals)
}

/// Generate a solution for a high-confidence pain point.
// @reserved: no frontend UI yet
#[tauri::command]
pub async fn butler_generate_solution(pain_id: String) -> std::result::Result<Proposal, String> {
    let store = pain_store();
    let gen = solution_store();
    let all_pains = store.list_all_pain_points().await;
    let pain = all_pains
        .into_iter()
        .find(|p| p.id == pain_id)
        .ok_or_else(|| format!("Pain point {} not found", pain_id))?;

    if pain.confidence < 0.7 {
        return Err("Pain point confidence too low to generate solution".into());
    }

    let proposal = gen.generate_solution(&pain).await;
    store
        .update_status(&pain_id, PainStatus::Solving)
        .await
        .map_err(|e| e.to_string())?;
    Ok(proposal)
}

/// Update the status of a proposal.
// @reserved: no frontend UI yet
#[tauri::command]
pub async fn butler_update_proposal_status(
    proposal_id: String,
    status: String,
) -> std::result::Result<(), String> {
    let new_status = match status.as_str() {
        "accepted" => ProposalStatus::Accepted,
        "rejected" => ProposalStatus::Rejected,
        "completed" => ProposalStatus::Completed,
        _ => ProposalStatus::Pending,
    };
    solution_store()
        .update_status(&proposal_id, new_status)
        .await
        .ok_or_else(|| format!("Proposal {} not found", proposal_id))?;
    Ok(())
}

// ---------------------------------------------------------------------------
// Tests
// ---------------------------------------------------------------------------

#[cfg(test)]
mod tests {
    use super::*;
    use zclaw_types::Message;

    #[test]
    fn test_pain_point_serialization() {
        let pp = PainPoint::new(
            "agent-1", "user-1", "出口包装合规反复出错",
            "compliance", PainSeverity::High, "又要改包装", "用户第三次提到包装问题",
        );
        assert_eq!(pp.summary, "出口包装合规反复出错");
        assert_eq!(pp.occurrence_count, 1);
        assert_eq!(pp.status, PainStatus::Detected);

        let json = serde_json::to_string(&pp).unwrap();
        let decoded: PainPoint = serde_json::from_str(&json).unwrap();
        assert_eq!(decoded.summary, "出口包装合规反复出错");
        assert_eq!(decoded.severity, PainSeverity::High);
    }

    #[test]
    fn test_merge_similar_pain_points() {
        let aggregator = PainAggregator::new();
        let rt = tokio::runtime::Runtime::new().unwrap();

        let p1 = PainPoint::new("agent-1", "user-1", "出口包装不合格",
            "compliance", PainSeverity::Medium, "包装被退了", "用户首次提到");
        let merged1 = rt.block_on(aggregator.merge_or_create(p1)).unwrap();
        assert_eq!(merged1.occurrence_count, 1);

        let p2 = PainPoint::new("agent-1", "user-1", "包装不合格又要改",
            "compliance", PainSeverity::High, "又要改包装", "用户第二次提到");
        let merged2 = rt.block_on(aggregator.merge_or_create(p2)).unwrap();
        assert_eq!(merged2.occurrence_count, 2);
        assert!(merged2.confidence > 0.3);
        assert_eq!(merged2.status, PainStatus::Confirmed);
    }

    #[test]
    fn test_different_categories_not_merged() {
        let aggregator = PainAggregator::new();
        let rt = tokio::runtime::Runtime::new().unwrap();

        let p1 = PainPoint::new("agent-1", "user-1", "出口包装不合格",
            "compliance", PainSeverity::Medium, "包装被退", "包装问题");
        let p2 = PainPoint::new("agent-1", "user-1", "客户投诉服务态度",
            "customer", PainSeverity::High, "客户又投诉了", "客户问题");

        rt.block_on(aggregator.merge_or_create(p1)).unwrap();
        let merged2 = rt.block_on(aggregator.merge_or_create(p2)).unwrap();
        assert_eq!(merged2.occurrence_count, 1);
    }

    #[test]
    fn test_confidence_calculation() {
        let mut pain = PainPoint::new("agent-1", "user-1", "test",
            "general", PainSeverity::High, "test", "test");
        pain.occurrence_count = 3;
        pain.last_seen = Utc::now();

        let conf = PainAggregator::calculate_confidence(&pain);
        assert!(conf > 0.5, "3 occurrences + High severity should be > 0.5, got {}", conf);
        assert!(conf <= 1.0);
    }

    #[test]
    fn test_analyze_frustration_signals() {
        let messages = vec![
            Message::user("这批货又被退了"),
            Message::assistant("很抱歉听到这个消息"),
            Message::user("又要改包装，烦死了，第三次了"),
        ];
        let result = analyze_for_pain_signals(&messages);
        assert!(result.is_some());
        let analysis = result.unwrap();
        assert!(analysis.found);
        assert_eq!(analysis.severity, PainSeverity::High);
        assert!(!analysis.summary.is_empty());
    }

    #[test]
    fn test_analyze_no_frustration() {
        let messages = vec![
            Message::user("今天天气不错"),
            Message::assistant("是的，天气很好"),
        ];
        assert!(analyze_for_pain_signals(&messages).is_none());
    }

    #[test]
    fn test_classify_category() {
        assert_eq!(classify_category("出口包装被退了"), "logistics");
        assert_eq!(classify_category("合规标准变了"), "compliance");
        assert_eq!(classify_category("客户投诉很多"), "customer");
        assert_eq!(classify_category("报价太低了"), "pricing");
        assert_eq!(classify_category("系统又崩了"), "technology");
        assert_eq!(classify_category("随便聊聊天"), "general");
    }

    #[test]
    fn test_severity_ordering() {
        let messages = vec![
            Message::user("这又来了"),
            Message::user("还是不行"),
        ];
        let result = analyze_for_pain_signals(&messages);
        assert!(result.is_some());
        assert_eq!(result.unwrap().severity, PainSeverity::Medium);
    }

    #[test]
    fn test_get_high_confidence_filters() {
        let aggregator = PainAggregator::new();
        let rt = tokio::runtime::Runtime::new().unwrap();

        for i in 0..3 {
            let p = PainPoint::new("agent-1", "user-1", "包装不合格反复出错",
                "compliance", PainSeverity::High,
                &format!("第{}次出现", i + 1), "recurring");
            rt.block_on(aggregator.merge_or_create(p)).unwrap();
        }

        let high = rt.block_on(aggregator.get_high_confidence_pains("agent-1", 0.5));
        assert!(!high.is_empty());
        assert!(high[0].confidence >= 0.5);
        assert_eq!(high[0].occurrence_count, 3);
    }
}