stemedb/applications/aphoria/src/community/pattern_syncer.rs

//! Pattern syncer for cross-project learning.
//!
//! Handles uploading learned patterns to the hosted server after anonymization.

use tracing::{info, instrument};

use crate::community::{SharedClaimTemplate, SharedPattern};
use crate::config::CrossProjectConfig;
use crate::error::AphoriaError;
use crate::hosted::{HostedClient, PushPatternsResponse};
use crate::learning::{LearnedPattern, PatternStore};

/// Syncs learned patterns to the hosted server.
///
/// Filters patterns by eligibility criteria, converts them to the
/// anonymized `SharedPattern` format, and pushes to the server.
pub struct PatternSyncer<'a> {
    client: &'a HostedClient,
    config: &'a CrossProjectConfig,
}

impl<'a> PatternSyncer<'a> {
    /// Create a new pattern syncer.
    pub fn new(client: &'a HostedClient, config: &'a CrossProjectConfig) -> Self {
        Self { client, config }
    }

    /// Get patterns eligible for sharing from the store.
    ///
    /// Filters by:
    /// - Not already promoted
    /// - Meets minimum local project count
    /// - Meets minimum local confidence
    /// - Not in exclude list
    pub fn get_shareable_patterns<S: PatternStore>(&self, store: &S) -> Vec<SharedPattern> {
        store
            .get_promotion_candidates(
                self.config.min_local_projects,
                self.config.min_local_confidence,
            )
            .into_iter()
            .filter(|p| !p.promoted)
            .filter(|p| self.passes_subject_filters(p))
            .map(|p| self.to_shared_pattern(&p))
            .collect()
    }

    /// Check if a pattern passes subject exclusion filters.
    fn passes_subject_filters(&self, pattern: &LearnedPattern) -> bool {
        let subject = &pattern.claim_template.subject_template;
        !self.config.is_subject_excluded(subject)
    }

    /// Convert a LearnedPattern to an anonymized SharedPattern.
    ///
    /// Privacy: Does NOT include `example_code` or `project_hashes`.
    fn to_shared_pattern(&self, pattern: &LearnedPattern) -> SharedPattern {
        SharedPattern {
            pattern_hash: compute_pattern_hash(&pattern.normalized_pattern, &pattern.language),
            normalized_pattern: pattern.normalized_pattern.clone(),
            claim_template: SharedClaimTemplate::new(
                &pattern.claim_template.subject_template,
                &pattern.claim_template.predicate,
                pattern.claim_template.value_type.to_string(),
            ),
            language: pattern.language.to_string(),
            project_count: pattern.project_count(),
            occurrences: pattern.occurrences,
            avg_confidence: pattern.avg_confidence,
        }
    }

    /// Sync all eligible patterns to the hosted server.
    ///
    /// Returns the server response with counts of accepted, merged, and deduplicated patterns.
    #[instrument(skip(self, store), fields(project = %self.client.project_id()))]
    pub fn sync<S: PatternStore>(&self, store: &S) -> Result<PushPatternsResponse, AphoriaError> {
        let patterns = self.get_shareable_patterns(store);

        if patterns.is_empty() {
            info!("No patterns eligible for sharing");
            return Ok(PushPatternsResponse::default());
        }

        info!(count = patterns.len(), "Syncing patterns to hosted server");
        self.client.push_patterns(patterns)
    }

    /// Get the count of patterns that would be synced (for preview).
    pub fn preview_count<S: PatternStore>(&self, store: &S) -> usize {
        self.get_shareable_patterns(store).len()
    }
}

/// Compute BLAKE3 hash of (normalized_pattern, language) for deduplication.
///
/// This hash uniquely identifies a pattern across organizations,
/// enabling server-side deduplication without revealing source code.
pub fn compute_pattern_hash(pattern: &str, language: &crate::types::Language) -> String {
    let mut hasher = blake3::Hasher::new();
    hasher.update(pattern.as_bytes());
    hasher.update(b":");
    hasher.update(language.to_string().as_bytes());
    hex::encode(hasher.finalize().as_bytes())
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::learning::{ClaimTemplate, ValueType};
    use crate::types::Language;

    /// Mock pattern store for testing
    struct MockPatternStore {
        patterns: Vec<LearnedPattern>,
    }

    impl MockPatternStore {
        fn new(patterns: Vec<LearnedPattern>) -> Self {
            Self { patterns }
        }
    }

    impl PatternStore for MockPatternStore {
        fn record_pattern(
            &self,
            _pattern: &LearnedPattern,
            _max_patterns: Option<usize>,
        ) -> Result<(), AphoriaError> {
            Ok(())
        }

        fn find_similar(
            &self,
            _normalized: &str,
            _language: Language,
            _threshold: f32,
        ) -> Option<LearnedPattern> {
            None
        }

        fn get_promotion_candidates(
            &self,
            min_projects: usize,
            min_confidence: f32,
        ) -> Vec<LearnedPattern> {
            self.patterns
                .iter()
                .filter(|p| p.is_promotion_candidate(min_projects, min_confidence))
                .cloned()
                .collect()
        }

        fn mark_promoted(
            &self,
            _id: &uuid::Uuid,
            _extractor_name: &str,
        ) -> Result<(), AphoriaError> {
            Ok(())
        }

        fn prune_stale(&self, _max_age_days: u32) -> Result<usize, AphoriaError> {
            Ok(0)
        }

        fn pattern_count(&self) -> usize {
            self.patterns.len()
        }

        fn get_all_patterns(&self) -> Vec<LearnedPattern> {
            self.patterns.clone()
        }

        fn get_pattern_by_id(&self, id: &uuid::Uuid) -> Option<LearnedPattern> {
            self.patterns.iter().find(|p| p.id == *id).cloned()
        }
    }

    fn create_test_pattern(
        subject: &str,
        project_count: usize,
        confidence: f32,
        promoted: bool,
    ) -> LearnedPattern {
        let template = ClaimTemplate::new(subject, "version", ValueType::Text, "Test pattern");

        let mut pattern = LearnedPattern::new(
            "test code",
            "const X = <string>",
            template,
            Language::Rust,
            "project1",
            confidence,
        );

        // Add more projects
        for i in 1..project_count {
            pattern.project_hashes.insert(format!("project{}", i));
        }
        pattern.promoted = promoted;

        pattern
    }

    #[test]
    fn test_compute_pattern_hash() {
        let hash1 = compute_pattern_hash("const X = <string>", &Language::Rust);
        let hash2 = compute_pattern_hash("const X = <string>", &Language::Rust);
        let hash3 = compute_pattern_hash("const X = <string>", &Language::Python);
        let hash4 = compute_pattern_hash("const Y = <number>", &Language::Rust);

        // Same input = same hash
        assert_eq!(hash1, hash2);
        // Different language = different hash
        assert_ne!(hash1, hash3);
        // Different pattern = different hash
        assert_ne!(hash1, hash4);
        // Hash should be 64 hex characters
        assert_eq!(hash1.len(), 64);
    }

    #[test]
    fn test_subject_exclusion() {
        // Note: is_subject_excluded uses simple prefix matching with starts_with
        let config = CrossProjectConfig {
            exclude_subjects: vec![
                "code://rust/internal/".to_string(),
                "vendor://acme/".to_string(),
            ],
            min_local_projects: 1,
            min_local_confidence: 0.5,
            ..Default::default()
        };

        // Create patterns (unused but kept for documentation of intent)
        let _internal = create_test_pattern("code://rust/internal/auth", 5, 0.9, false);
        let _vendor = create_test_pattern("vendor://acme/secret", 5, 0.9, false);
        let _public = create_test_pattern("code://rust/tls/version", 5, 0.9, false);

        // We need a hosted client to create the syncer - use a test fixture approach
        // Since we can't easily create a HostedClient without actual config,
        // we test the filter logic directly
        assert!(config.is_subject_excluded("code://rust/internal/auth"));
        assert!(config.is_subject_excluded("vendor://acme/secret"));
        assert!(!config.is_subject_excluded("code://rust/tls/version"));
    }

    #[test]
    fn test_promoted_patterns_excluded() {
        let promoted = create_test_pattern("tls/version", 5, 0.9, true);
        let not_promoted = create_test_pattern("db/pool_size", 5, 0.9, false);

        let store = MockPatternStore::new(vec![promoted, not_promoted]);

        // Get candidates (promoted should be filtered by the store itself)
        let candidates = store.get_promotion_candidates(3, 0.8);

        // Promoted pattern should be filtered out by is_promotion_candidate
        assert_eq!(candidates.len(), 1);
        assert!(!candidates[0].promoted);
    }

    #[test]
    fn test_to_shared_pattern_anonymization() {
        let template =
            ClaimTemplate::new("tls/min_version", "version", ValueType::Text, "TLS version");

        let mut pattern = LearnedPattern::new(
            "const TLS_MIN_VERSION = \"1.2\"", // This should NOT be shared
            "const TLS_MIN_VERSION = <string>",
            template,
            Language::Rust,
            "secret-project-hash", // This should NOT be shared
            0.9,
        );
        pattern.project_hashes.insert("another-secret-hash".to_string());

        // Create syncer with a mock - testing the conversion logic directly
        // Since we need a HostedClient, we test the SharedPattern structure
        let shared = SharedPattern {
            pattern_hash: compute_pattern_hash(&pattern.normalized_pattern, &pattern.language),
            normalized_pattern: pattern.normalized_pattern.clone(),
            claim_template: SharedClaimTemplate::new(
                &pattern.claim_template.subject_template,
                &pattern.claim_template.predicate,
                pattern.claim_template.value_type.to_string(),
            ),
            language: pattern.language.to_string(),
            project_count: pattern.project_count(),
            occurrences: pattern.occurrences,
            avg_confidence: pattern.avg_confidence,
        };

        // Verify anonymization - no example_code or project_hashes
        assert_eq!(shared.normalized_pattern, "const TLS_MIN_VERSION = <string>");
        assert_eq!(shared.project_count, 2);
        assert_eq!(shared.occurrences, 1);
        assert!((shared.avg_confidence - 0.9).abs() < 0.001);

        // Verify the pattern_hash computation
        assert_eq!(shared.pattern_hash.len(), 64);
    }
}