stemedb/applications/aphoria/src/promotion/pipeline.rs

//! Promotion pipeline for converting learned patterns to declarative extractors.
//!
//! Orchestrates the full promotion flow: candidates → regex generation → validation → YAML output.

use std::path::PathBuf;

use tracing::{debug, info, warn};
use uuid::Uuid;

use super::regex_gen::RegexGenerator;
use super::types::{PromotionCandidate, PromotionStats, ValidationResult};
use super::validator::ExtractorValidator;
use super::writer::YamlWriter;
use crate::config::PromotionConfig;
use crate::learning::{LearnedPattern, PatternStore};
use crate::llm::GeminiClient;
use crate::AphoriaError;

/// The promotion pipeline orchestrates pattern-to-extractor conversion.
pub struct PromotionPipeline<'a, S: PatternStore> {
    /// Pattern store for fetching candidates.
    store: &'a S,

    /// LLM client for regex generation.
    client: Option<&'a GeminiClient>,

    /// Configuration for promotion thresholds.
    config: &'a PromotionConfig,

    /// Validator for testing generated extractors.
    validator: ExtractorValidator,

    /// YAML writer for output.
    writer: Option<YamlWriter>,
}

impl<'a, S: PatternStore> PromotionPipeline<'a, S> {
    /// Create a new promotion pipeline.
    ///
    /// If `output_dir` is None, uses the default `.aphoria/extractors/learned/`.
    pub fn new(
        store: &'a S,
        client: Option<&'a GeminiClient>,
        config: &'a PromotionConfig,
        output_dir: Option<PathBuf>,
    ) -> Result<Self, AphoriaError> {
        let writer = if let Some(dir) = output_dir { Some(YamlWriter::new(dir)?) } else { None };

        Ok(Self { store, client, config, validator: ExtractorValidator::default(), writer })
    }

    /// Get patterns eligible for promotion.
    ///
    /// Returns patterns that meet the configured thresholds for project count
    /// and confidence.
    pub fn get_candidates(&self) -> Vec<LearnedPattern> {
        self.store.get_promotion_candidates(self.config.min_projects, self.config.min_confidence)
    }

    /// Generate a promotion candidate from a learned pattern.
    ///
    /// Uses the LLM to generate a regex pattern and validates it.
    pub fn generate_candidate(
        &self,
        pattern: &LearnedPattern,
    ) -> Result<PromotionCandidate, AphoriaError> {
        let client = self.client.ok_or_else(|| {
            AphoriaError::Promotion("LLM client not configured for regex generation".to_string())
        })?;

        // Generate extractor definition using LLM
        let generator = RegexGenerator::new(client);
        let extractor_def = generator.generate(pattern)?;

        // Validate the generated extractor
        let validation = self.validator.validate(&extractor_def, pattern)?;

        Ok(PromotionCandidate::new(pattern.clone(), extractor_def, validation))
    }

    /// Promote a candidate by writing it to YAML and marking the pattern as promoted.
    ///
    /// Returns the path to the written YAML file.
    pub fn promote(&self, candidate: &PromotionCandidate) -> Result<PathBuf, AphoriaError> {
        // Check if candidate is ready
        if !candidate.is_ready() {
            return Err(AphoriaError::Promotion(format!(
                "Candidate {} is not ready for promotion: validation={}, performance={}",
                candidate.pattern_id(),
                candidate.validation.passed,
                candidate.validation.performance_ok
            )));
        }

        // Get or create writer
        let writer = if let Some(ref w) = self.writer {
            w
        } else {
            return Err(AphoriaError::Promotion("YAML writer not configured".to_string()));
        };

        // Check if already exists
        if writer.exists(candidate.extractor_name()) {
            return Err(AphoriaError::Promotion(format!(
                "Extractor '{}' already exists",
                candidate.extractor_name()
            )));
        }

        // Write YAML file
        let path = writer.write(&candidate.extractor_def, &candidate.pattern)?;

        // Mark pattern as promoted
        self.store.mark_promoted(&candidate.pattern_id(), candidate.extractor_name())?;

        info!(
            pattern_id = %candidate.pattern_id(),
            extractor = %candidate.extractor_name(),
            path = %path.display(),
            "Pattern promoted to extractor"
        );

        Ok(path)
    }

    /// Process all eligible patterns and return promotion candidates.
    ///
    /// Generates and validates extractors for each eligible pattern.
    /// Does not actually promote (write YAML) - use `promote()` for that.
    pub fn process_all(&self) -> Vec<Result<PromotionCandidate, AphoriaError>> {
        let patterns = self.get_candidates();

        debug!(count = patterns.len(), "Processing promotion candidates");

        patterns.iter().map(|pattern| self.generate_candidate(pattern)).collect()
    }

    /// Auto-promote all ready candidates.
    ///
    /// Only runs if `auto_promote` is enabled in config.
    /// Returns the number of patterns promoted and any errors.
    pub fn auto_promote_all(&self) -> (usize, Vec<AphoriaError>) {
        if !self.config.auto_promote {
            warn!("auto_promote is disabled in config");
            return (0, vec![]);
        }

        let candidates = self.process_all();
        let mut promoted = 0;
        let mut errors = Vec::new();

        for result in candidates {
            match result {
                Ok(candidate) if candidate.is_ready() => match self.promote(&candidate) {
                    Ok(_) => promoted += 1,
                    Err(e) => errors.push(e),
                },
                Ok(candidate) => {
                    debug!(
                        pattern_id = %candidate.pattern_id(),
                        "Candidate not ready for auto-promotion"
                    );
                }
                Err(e) => errors.push(e),
            }
        }

        (promoted, errors)
    }

    /// Get statistics about the promotion pipeline.
    pub fn stats(&self) -> PromotionStats {
        let all_patterns: Vec<LearnedPattern> = self.store.get_promotion_candidates(0, 0.0); // Get all patterns

        let eligible = self.get_candidates();
        let promoted: Vec<_> = all_patterns.iter().filter(|p| p.promoted).collect();

        let avg_confidence = if eligible.is_empty() {
            0.0
        } else {
            eligible.iter().map(|p| p.avg_confidence).sum::<f32>() / eligible.len() as f32
        };

        let avg_projects = if eligible.is_empty() {
            0.0
        } else {
            eligible.iter().map(|p| p.project_count() as f32).sum::<f32>() / eligible.len() as f32
        };

        PromotionStats {
            total_patterns: all_patterns.len(),
            eligible_patterns: eligible.len(),
            promoted_patterns: promoted.len(),
            pending_review: eligible.len().saturating_sub(promoted.len()),
            avg_confidence,
            avg_projects,
        }
    }

    /// Promote a specific pattern by ID.
    pub fn promote_by_id(&self, pattern_id: &Uuid) -> Result<PathBuf, AphoriaError> {
        // Find the pattern
        let candidates = self.get_candidates();
        let pattern = candidates.iter().find(|p| &p.id == pattern_id).ok_or_else(|| {
            AphoriaError::Promotion(format!("Pattern {} not found in candidates", pattern_id))
        })?;

        // Generate and validate
        let candidate = self.generate_candidate(pattern)?;

        // Promote
        self.promote(&candidate)
    }

    /// Validate a pattern without promoting it.
    ///
    /// Returns the validation result for inspection.
    pub fn validate_pattern(
        &self,
        pattern: &LearnedPattern,
    ) -> Result<ValidationResult, AphoriaError> {
        let client = self.client.ok_or_else(|| {
            AphoriaError::Promotion("LLM client not configured for regex generation".to_string())
        })?;

        let generator = RegexGenerator::new(client);
        let extractor_def = generator.generate(pattern)?;

        self.validator.validate(&extractor_def, pattern)
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::config::PromotionConfig;
    use crate::learning::{ClaimTemplate, LocalPatternStore, ValueType};
    use crate::types::Language;
    use chrono::Utc;
    use tempfile::TempDir;

    fn create_test_store(temp: &TempDir) -> LocalPatternStore {
        LocalPatternStore::new(temp.path()).expect("create store")
    }

    fn create_eligible_pattern() -> LearnedPattern {
        let mut pattern = LearnedPattern::new(
            "verify_ssl = false",
            "verify_ssl = <boolean>",
            ClaimTemplate::new("ssl/verify", "enabled", ValueType::Boolean, "SSL verification"),
            Language::Python,
            "project1",
            0.9,
        );

        // Add enough projects to meet threshold
        for i in 2..=6 {
            pattern.record_observation(format!("project{}", i), 0.85, Utc::now());
        }

        pattern
    }

    #[test]
    fn test_pipeline_creation() {
        let temp = TempDir::new().expect("temp dir");
        let store = create_test_store(&temp);
        let config = PromotionConfig::default();

        let pipeline =
            PromotionPipeline::new(&store, None, &config, Some(temp.path().to_path_buf()));
        assert!(pipeline.is_ok());
    }

    #[test]
    fn test_get_candidates_empty() {
        let temp = TempDir::new().expect("temp dir");
        let store = create_test_store(&temp);
        let config = PromotionConfig::default();

        let pipeline =
            PromotionPipeline::new(&store, None, &config, None).expect("create pipeline");

        let candidates = pipeline.get_candidates();
        assert!(candidates.is_empty());
    }

    #[test]
    fn test_get_candidates_with_eligible() {
        let temp = TempDir::new().expect("temp dir");
        let store = create_test_store(&temp);
        let config = PromotionConfig::default();

        // Add eligible pattern
        let pattern = create_eligible_pattern();
        store.record_pattern(&pattern, None).expect("record");

        let pipeline =
            PromotionPipeline::new(&store, None, &config, None).expect("create pipeline");

        let candidates = pipeline.get_candidates();
        assert_eq!(candidates.len(), 1);
    }

    #[test]
    fn test_stats_empty_store() {
        let temp = TempDir::new().expect("temp dir");
        let store = create_test_store(&temp);
        let config = PromotionConfig::default();

        let pipeline =
            PromotionPipeline::new(&store, None, &config, None).expect("create pipeline");

        let stats = pipeline.stats();
        assert_eq!(stats.total_patterns, 0);
        assert_eq!(stats.eligible_patterns, 0);
        assert_eq!(stats.promoted_patterns, 0);
    }

    #[test]
    fn test_stats_with_patterns() {
        let temp = TempDir::new().expect("temp dir");
        let store = create_test_store(&temp);
        let config = PromotionConfig::default();

        // Add eligible pattern
        let pattern = create_eligible_pattern();
        store.record_pattern(&pattern, None).expect("record");

        // Add non-eligible pattern (not enough projects)
        let small_pattern = LearnedPattern::new(
            "test = true",
            "test = <boolean>",
            ClaimTemplate::new("test", "value", ValueType::Boolean, "Test"),
            Language::Rust,
            "project1",
            0.9,
        );
        store.record_pattern(&small_pattern, None).expect("record");

        let pipeline =
            PromotionPipeline::new(&store, None, &config, None).expect("create pipeline");

        let stats = pipeline.stats();
        assert_eq!(stats.eligible_patterns, 1);
        assert_eq!(stats.pending_review, 1);
    }

    #[test]
    fn test_generate_candidate_requires_client() {
        let temp = TempDir::new().expect("temp dir");
        let store = create_test_store(&temp);
        let config = PromotionConfig::default();
        let pattern = create_eligible_pattern();

        let pipeline =
            PromotionPipeline::new(&store, None, &config, None).expect("create pipeline");

        let result = pipeline.generate_candidate(&pattern);
        assert!(result.is_err());
        assert!(result.unwrap_err().to_string().contains("LLM client not configured"));
    }

    #[test]
    fn test_auto_promote_disabled() {
        let temp = TempDir::new().expect("temp dir");
        let store = create_test_store(&temp);
        let config = PromotionConfig { auto_promote: false, ..Default::default() };

        let pipeline =
            PromotionPipeline::new(&store, None, &config, None).expect("create pipeline");

        let (promoted, errors) = pipeline.auto_promote_all();
        assert_eq!(promoted, 0);
        assert!(errors.is_empty());
    }
}