repotoire 0.8.2

//! AI Duplicate Block Detector
//!
//! Detects near-identical code blocks that AI coding assistants tend to create
//! (copy-paste patterns). Uses AST-based similarity analysis per ICSE 2025 research.
//!
//! AI assistants often generate repetitive code with minor variations like:
//! - Different variable names but same logic
//! - Same structure with different literals
//! - Copy-paste patterns with slight modifications
//!
//! This detector uses normalized identifier hashing and Jaccard similarity
//! to find these near-duplicates. Threshold: ≥70% similarity.

#![allow(dead_code)] // Module under development - structs/helpers used in tests only

use crate::detectors::base::{Detector, DetectorConfig, DetectorScope};
use crate::graph::GraphQueryExt;
use crate::models::{Finding, Severity};
use anyhow::Result;
use std::collections::HashSet;
use std::path::PathBuf;
use tracing::{debug, info};

/// Default thresholds (based on ICSE 2025 research)
const DEFAULT_SIMILARITY_THRESHOLD: f64 = 0.70; // 70% Jaccard similarity
const DEFAULT_GENERIC_NAME_THRESHOLD: f64 = 0.40; // 40% generic identifiers
const DEFAULT_MIN_LOC: usize = 5;
const DEFAULT_MAX_FINDINGS: usize = 50;

/// Generic identifier patterns commonly produced by AI assistants
const GENERIC_IDENTIFIERS: &[&str] = &[
    "result",
    "res",
    "ret",
    "return_value",
    "rv",
    "temp",
    "tmp",
    "t",
    "data",
    "d",
    "value",
    "val",
    "v",
    "item",
    "items",
    "i",
    "obj",
    "object",
    "o",
    "x",
    "y",
    "z",
    "a",
    "b",
    "c",
    "arr",
    "array",
    "list",
    "lst",
    "dict",
    "dictionary",
    "map",
    "mapping",
    "str",
    "string",
    "s",
    "num",
    "number",
    "n",
    "count",
    "cnt",
    "index",
    "idx",
    "key",
    "k",
    "var",
    "variable",
    "input",
    "output",
    "out",
    "response",
    "resp",
    "request",
    "req",
    "config",
    "cfg",
    "args",
    "kwargs",
    "params",
    "parameters",
    "options",
    "opts",
    "settings",
    "handler",
    "callback",
    "cb",
    "func",
    "fn",
    "function",
    "elem",
    "element",
    "node",
    "current",
    "curr",
    "cur",
    "previous",
    "prev",
    "next",
];

/// Processed function data for similarity comparison.
/// Uses an index into a shared signatures array instead of storing full HashSets.
#[derive(Debug, Clone)]
pub struct FunctionData {
    pub qualified_name: String,
    pub name: String,
    pub file_path: String,
    pub line_start: u32,
    pub line_end: u32,
    pub loc: usize,
    /// Index into the shared MinHash signatures array.
    pub sig_idx: usize,
    pub generic_ratio: f64,
    pub ast_size: usize,
    /// String literals extracted from source (sorted, deduplicated).
    /// Used by Tier 4 to distinguish leaf functions with same AST but different constants.
    pub string_literals: Vec<String>,
}

/// Pre-built set for O(1) generic identifier lookup
static GENERIC_SET: std::sync::LazyLock<HashSet<&'static str>> =
    std::sync::LazyLock::new(|| GENERIC_IDENTIFIERS.iter().copied().collect());

/// Calculate ratio of generic identifiers in a function
fn calculate_generic_ratio(identifiers: &[String]) -> f64 {
    if identifiers.is_empty() {
        return 0.0;
    }

    let generic_count = identifiers
        .iter()
        .filter(|id| GENERIC_SET.contains(id.to_lowercase().as_str()))
        .count();

    generic_count as f64 / identifiers.len() as f64
}

/// Extract string literals from a function's source lines.
/// Returns sorted, deduplicated set of strings found between double quotes.
/// Used by Tier 4 to distinguish leaf functions that match on the same enum
/// but return different string constants.
fn extract_string_literals(content: &str, line_start: u32, line_end: u32) -> Vec<String> {
    let mut literals = HashSet::new();
    for (idx, line) in content.lines().enumerate() {
        let line_num = idx as u32 + 1;
        if line_num < line_start || line_num > line_end {
            continue;
        }
        // Simple extraction: scan for "..." substrings
        let bytes = line.as_bytes();
        let mut i = 0;
        while i < bytes.len() {
            if bytes[i] == b'"' {
                let start = i + 1;
                i = start;
                while i < bytes.len() {
                    if bytes[i] == b'\\' {
                        i += 2; // skip escaped char
                        continue;
                    }
                    if bytes[i] == b'"' {
                        let lit = &line[start..i];
                        if !lit.is_empty() && lit.len() < 100 {
                            literals.insert(lit.to_string());
                        }
                        break;
                    }
                    i += 1;
                }
            }
            i += 1;
        }
    }
    let mut sorted: Vec<String> = literals.into_iter().collect();
    sorted.sort();
    sorted
}

/// Detect near-identical code blocks typical of AI-generated code
pub struct AIDuplicateBlockDetector {
    config: DetectorConfig,
    similarity_threshold: f64,
    generic_name_threshold: f64,
    min_loc: usize,
    max_findings: usize,
}

impl AIDuplicateBlockDetector {
    /// Create a new detector with default settings
    pub fn new() -> Self {
        Self {
            config: DetectorConfig::new(),
            similarity_threshold: DEFAULT_SIMILARITY_THRESHOLD,
            generic_name_threshold: DEFAULT_GENERIC_NAME_THRESHOLD,
            min_loc: DEFAULT_MIN_LOC,
            max_findings: DEFAULT_MAX_FINDINGS,
        }
    }

    /// Create with custom config
    pub fn with_config(config: DetectorConfig) -> Self {
        Self {
            similarity_threshold: config
                .get_option_or("similarity_threshold", DEFAULT_SIMILARITY_THRESHOLD),
            generic_name_threshold: config
                .get_option_or("generic_name_threshold", DEFAULT_GENERIC_NAME_THRESHOLD),
            min_loc: config.get_option_or("min_loc", DEFAULT_MIN_LOC),
            max_findings: config.get_option_or("max_findings", DEFAULT_MAX_FINDINGS),
            config,
        }
    }

    /// Compute similarity threshold based on average function size.
    /// Small functions need higher thresholds because they have fewer
    /// distinguishing bigrams (higher chance of coincidental similarity).
    fn size_adaptive_threshold(&self, loc1: usize, loc2: usize) -> f64 {
        let avg_loc = (loc1 + loc2) / 2;
        if avg_loc <= 10 {
            0.90
        } else if avg_loc <= 20 {
            // Linear interpolation: 0.90 at 10 LOC → base threshold at 21+ LOC
            let base = self.similarity_threshold;
            0.90 - (avg_loc as f64 - 10.0) * (0.90 - base) / 10.0
        } else {
            self.similarity_threshold
        }
    }

    /// Find duplicate pairs using MinHash/LSH + MinHash-estimated Jaccard.
    ///
    /// Uses LSH banding on pre-computed MinHash signatures for candidate generation,
    /// then verifies candidates with MinHash-estimated Jaccard (±0.1 at k=100).
    /// Drops the full `HashSet<String>` bigram sets — only signatures are needed.
    fn find_duplicates(
        &self,
        functions: &[FunctionData],
        signatures: &[[u64; 100]],
    ) -> Vec<(FunctionData, FunctionData, f64)> {
        use rayon::prelude::*;

        if functions.len() < 2 {
            return Vec::new();
        }

        let candidates =
            crate::detectors::ast_fingerprint::lsh_candidate_pairs_from_sigs(signatures);

        debug!(
            "LSH: {} candidates from {} functions ({:.1}% of {:.0} total pairs)",
            candidates.len(),
            functions.len(),
            candidates.len() as f64 / (functions.len() as f64 * (functions.len() - 1) as f64 / 2.0)
                * 100.0,
            functions.len() as f64 * (functions.len() - 1) as f64 / 2.0,
        );

        // Parallel MinHash Jaccard verification on LSH candidates
        let threshold = self.similarity_threshold;
        let candidate_vec: Vec<_> = candidates.into_iter().collect();
        let mut duplicates: Vec<(FunctionData, FunctionData, f64)> = candidate_vec
            .par_iter()
            .filter_map(|&(i, j)| {
                let func1 = &functions[i];
                let func2 = &functions[j];

                // Skip same-file comparisons
                if func1.file_path == func2.file_path {
                    return None;
                }

                // Size-ratio pre-filter (Jaccard upper bound)
                if func1.ast_size > 0 && func2.ast_size > 0 {
                    let size_ratio = func1.ast_size.min(func2.ast_size) as f64
                        / func1.ast_size.max(func2.ast_size) as f64;
                    if size_ratio < threshold {
                        return None;
                    }
                }

                // MinHash-estimated Jaccard (±0.1 at k=100)
                let similarity = crate::detectors::ast_fingerprint::minhash_jaccard(
                    &signatures[func1.sig_idx],
                    &signatures[func2.sig_idx],
                );
                // Size-adaptive threshold: small functions need higher similarity
                let adaptive_threshold = self.size_adaptive_threshold(func1.loc, func2.loc);
                if similarity >= adaptive_threshold {
                    Some((func1.clone(), func2.clone(), similarity))
                } else {
                    None
                }
            })
            .collect();

        // Sort by similarity (highest first)
        duplicates.sort_by(|a, b| b.2.partial_cmp(&a.2).unwrap_or(std::cmp::Ordering::Equal));

        duplicates
    }

    /// Check if a qualified name represents a trait implementation.
    fn is_trait_impl(qn: &str) -> bool {
        qn.contains("impl<") && qn.contains(" for ")
    }

    /// Extract the trait name from a trait impl qualified name.
    /// "src/foo.rs::impl<Display for MyStruct>::fmt:30" -> Some("Display")
    fn extract_trait_name(qn: &str) -> Option<&str> {
        let impl_start = qn.find("impl<")? + 5;
        let for_pos = qn[impl_start..].find(" for ")?;
        Some(&qn[impl_start..impl_start + for_pos])
    }

    /// Extract the implementing type from a qualified name.
    /// "src/foo.rs::impl<Display for MyStruct>::fmt:30" -> Some("MyStruct")
    /// "src/foo.rs::impl<MyStruct>::new:10" -> Some("MyStruct")
    fn extract_impl_type(qn: &str) -> Option<&str> {
        let impl_start = qn.find("impl<")? + 5;
        let close = qn[impl_start..].find('>')?;
        let inner = &qn[impl_start..impl_start + close];
        if let Some(for_pos) = inner.find(" for ") {
            Some(&inner[for_pos + 5..])
        } else {
            Some(inner)
        }
    }

    /// Resolve a FunctionData to its graph QN via find_function_at.
    /// Falls back to the detector's simplified QN if not found in graph.
    fn resolve_graph_qn(func: &FunctionData, graph: &dyn crate::graph::GraphQuery) -> String {
        let i = graph.interner();
        graph
            .find_function_at(&func.file_path, func.line_start)
            .map(|n| n.qn(i).to_string())
            .unwrap_or_else(|| func.qualified_name.clone())
    }

    /// Verify that a candidate duplicate pair is semantically real, not coincidental.
    /// Returns false if the pair should be rejected as a false positive.
    fn verify_semantic_overlap(
        func1: &FunctionData,
        func2: &FunctionData,
        _similarity: f64,
        graph: &dyn crate::graph::GraphQuery,
    ) -> bool {
        // Resolve to graph QNs (which include impl<Type> info)
        let gqn1 = Self::resolve_graph_qn(func1, graph);
        let gqn2 = Self::resolve_graph_qn(func2, graph);

        // Tier 1: Trait impl filter
        // Same trait on different types -> not a real clone (idiomatic pattern)
        if Self::is_trait_impl(&gqn1) && Self::is_trait_impl(&gqn2) {
            let trait1 = Self::extract_trait_name(&gqn1);
            let trait2 = Self::extract_trait_name(&gqn2);
            let type1 = Self::extract_impl_type(&gqn1);
            let type2 = Self::extract_impl_type(&gqn2);
            if trait1 == trait2 && type1 != type2 {
                return false;
            }
        }

        // Tier 2: Callee overlap check (for functions with callees)
        let i = graph.interner();
        let callees1: HashSet<String> = graph
            .get_callees(&gqn1)
            .into_iter()
            .map(|n| n.qn(i).to_string())
            .collect();
        let callees2: HashSet<String> = graph
            .get_callees(&gqn2)
            .into_iter()
            .map(|n| n.qn(i).to_string())
            .collect();

        if !callees1.is_empty() || !callees2.is_empty() {
            let intersection = callees1.intersection(&callees2).count();
            let union = callees1.union(&callees2).count();
            let overlap = if union > 0 {
                intersection as f64 / union as f64
            } else {
                0.0
            };
            if overlap < 0.3 {
                return false;
            }
        }

        // Tier 3: Leaf function context (no callees on either side)
        // Leaf functions in different impl types are likely coincidental
        if callees1.is_empty() && callees2.is_empty() {
            let type1 = Self::extract_impl_type(&gqn1);
            let type2 = Self::extract_impl_type(&gqn2);
            if type1.is_some() && type2.is_some() && type1 != type2 {
                return false;
            }

            // Tier 4: String literal divergence for leaf functions.
            // Functions with identical AST structure but different string constants
            // (e.g. match on same enum returning different strings) are template
            // siblings, not true duplicates.
            if !func1.string_literals.is_empty() || !func2.string_literals.is_empty() {
                let set1: HashSet<&str> =
                    func1.string_literals.iter().map(|s| s.as_str()).collect();
                let set2: HashSet<&str> =
                    func2.string_literals.iter().map(|s| s.as_str()).collect();
                let intersection = set1.intersection(&set2).count();
                let union = set1.union(&set2).count();
                let literal_overlap = if union > 0 {
                    intersection as f64 / union as f64
                } else {
                    1.0
                };
                if literal_overlap < 0.2 {
                    return false;
                }
            }
        }

        true
    }

    /// Create a finding from a duplicate pair
    fn create_finding(
        &self,
        func1: &FunctionData,
        func2: &FunctionData,
        similarity: f64,
    ) -> Finding {
        let similarity_pct = (similarity * 100.0) as u32;

        // Check for generic naming pattern
        let has_generic_naming = func1.generic_ratio >= self.generic_name_threshold
            || func2.generic_ratio >= self.generic_name_threshold;

        // Determine severity based on similarity and generic naming.
        // Duplicate code is a refactoring smell, never a security/data-loss issue.
        // Cap at Medium (exact/near-exact matches) or Low (partial matches).
        let severity = if similarity >= 0.85 {
            Severity::Medium
        } else {
            Severity::Low
        };

        // Build description
        let mut description = format!(
            "Functions '{}' and '{}' have {}% AST similarity, \
             indicating structural duplication.\n\n\
             **{}** ({} LOC): `{}`\n\
             **{}** ({} LOC): `{}`\n\n",
            func1.name,
            func2.name,
            similarity_pct,
            func1.name,
            func1.loc,
            func1.file_path,
            func2.name,
            func2.loc,
            func2.file_path,
        );

        if has_generic_naming {
            let avg_generic = (func1.generic_ratio + func2.generic_ratio) / 2.0;
            description.push_str(&format!(
                "⚠️ **Generic naming detected**: {:.0}% of identifiers \
                 use generic names (result, temp, data, etc.), suggesting low-effort duplication.\n\n",
                avg_generic * 100.0
            ));
        }

        description.push_str(
            "Near-identical functions increase maintenance burden and \
             can lead to inconsistent bug fixes.",
        );

        let suggestion = "Consider one of the following approaches:\n\
             1. **Extract common logic** into a shared helper function\n\
             2. **Use a template/factory pattern** if variations are intentional\n\
             3. **Consolidate** into a single implementation if truly duplicates\n\
             4. **Add documentation** explaining why similar implementations exist"
            .to_string();

        let mut affected_files = Vec::new();
        if func1.file_path != "unknown" {
            affected_files.push(PathBuf::from(&func1.file_path));
        }
        if func2.file_path != "unknown" {
            affected_files.push(PathBuf::from(&func2.file_path));
        }

        Finding {
            id: String::new(),
            detector: "AIDuplicateBlockDetector".to_string(),
            severity,
            title: format!(
                "Structural duplicate: {} ≈ {} ({}% AST)",
                func1.name, func2.name, similarity_pct
            ),
            description,
            affected_files,
            line_start: Some(func1.line_start),
            line_end: Some(func1.line_end),
            suggested_fix: Some(suggestion),
            estimated_effort: Some("Medium (1-2 hours)".to_string()),
            category: Some("duplication".to_string()),
            cwe_id: None,
            why_it_matters: Some(
                "Near-identical code duplicates increase maintenance burden. \
                 When bugs are found, they must be fixed in multiple places. \
                 When requirements change, all copies must be updated consistently."
                    .to_string(),
            ),
            ..Default::default()
        }
    }
}

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

impl Detector for AIDuplicateBlockDetector {
    fn name(&self) -> &'static str {
        "AIDuplicateBlockDetector"
    }

    fn description(&self) -> &'static str {
        "Detects near-identical code blocks using AST similarity (≥70%)"
    }

    fn category(&self) -> &'static str {
        "ai_generated"
    }

    fn requires_graph(&self) -> bool {
        true
    }

    fn detector_scope(&self) -> DetectorScope {
        // Produces cross-file findings (compares code blocks across files).
        DetectorScope::FileScopedGraph
    }

    fn config(&self) -> Option<&DetectorConfig> {
        Some(&self.config)
    }

    fn file_extensions(&self) -> &'static [&'static str] {
        &[
            "py", "js", "ts", "jsx", "tsx", "java", "go", "rs", "c", "cpp", "cs",
        ]
    }

    fn detect(
        &self,
        ctx: &crate::detectors::analysis_context::AnalysisContext,
    ) -> Result<Vec<Finding>> {
        let files = &ctx.as_file_provider();
        use rayon::prelude::*;

        let source_exts = &["py", "js", "ts", "jsx", "tsx", "java", "go", "rs"];

        // Collect file paths + content upfront (FileProvider requires &self borrows)
        let file_data: Vec<_> = files
            .files_with_extensions(source_exts)
            .into_iter()
            .filter(|path| !crate::detectors::base::is_test_path(&path.to_string_lossy()))
            .filter_map(|path| {
                let content = files.content(path)?;
                let ext = path.extension().and_then(|e| e.to_str()).unwrap_or("");
                let has_functions = match ext {
                    "py" => content.contains("\ndef ") || content.starts_with("def "),
                    "js" | "jsx" => {
                        content.contains("function ")
                            || content.contains("=> ")
                            || content.contains("=>{")
                    }
                    "ts" | "tsx" => {
                        content.contains("function ")
                            || content.contains("=> ")
                            || content.contains("=>{")
                    }
                    "go" => content.contains("\nfunc ") || content.starts_with("func "),
                    "rs" => {
                        content.contains("\nfn ")
                            || content.starts_with("fn ")
                            || content.contains(" fn ")
                    }
                    "java" => {
                        content.contains("void ")
                            || content.contains("public ")
                            || content.contains("private ")
                    }
                    _ => true,
                };
                if !has_functions {
                    return None;
                }
                Some((path.to_path_buf(), content, ext.to_string()))
            })
            .collect();

        // Collect functions AND their MinHash signatures in parallel.
        // FunctionData stores only a sig_idx (index into signatures vec).
        let min_loc = self.min_loc;
        let mut all_functions: Vec<FunctionData> = Vec::new();
        let mut all_signatures: Vec<[u64; 100]> = Vec::new();

        // Parallel phase: collect (FunctionData-minus-sig_idx, signature) tuples
        struct FuncWithSig {
            qualified_name: String,
            name: String,
            file_path: String,
            line_start: u32,
            line_end: u32,
            string_literals: Vec<String>,
            loc: usize,
            generic_ratio: f64,
            ast_size: usize,
            sig: [u64; 100],
        }

        let func_sigs: Vec<FuncWithSig> = file_data
            .par_iter()
            .flat_map_iter(|(path, content, ext)| {
                let path_str = path.to_string_lossy().to_string();

                // Fast path: use cached fingerprints + pre-computed sigs from parse phase
                if let Some(cached) = crate::parsers::get_cached_fps(&path_str) {
                    return cached
                        .into_iter()
                        .filter_map(|fp| {
                            let loc = (fp.line_end - fp.line_start + 1) as usize;
                            if loc < min_loc || fp.normalized_bigrams.is_empty() {
                                return None;
                            }
                            let generic_ratio = calculate_generic_ratio(&fp.identifiers);
                            let ast_size = fp.normalized_bigrams.len();
                            // Use pre-computed sig or compute from bigrams
                            let sig = fp.minhash_sig.unwrap_or_else(|| {
                                crate::detectors::ast_fingerprint::compute_minhash_signature(
                                    &fp.normalized_bigrams,
                                )
                            });
                            let string_literals =
                                extract_string_literals(content, fp.line_start, fp.line_end);
                            Some(FuncWithSig {
                                qualified_name: format!("{}::{}", path_str, fp.name),
                                name: fp.name,
                                file_path: path_str.clone(),
                                line_start: fp.line_start,
                                line_end: fp.line_end,
                                string_literals,
                                loc,
                                generic_ratio,
                                ast_size,
                                sig,
                            })
                        })
                        .collect::<Vec<_>>();
                }

                // Slow path: fallback to re-parsing (e.g. tests with MockFileProvider)
                let lang = crate::parsers::lightweight::Language::from_extension(ext);
                let functions =
                    crate::detectors::ast_fingerprint::parse_functions_with_fingerprints(
                        content, lang,
                    );

                functions
                    .into_iter()
                    .filter_map(move |(func, fp)| {
                        let loc = (func.line_end - func.line_start + 1) as usize;
                        if loc < min_loc || fp.normalized_bigrams.is_empty() {
                            return None;
                        }
                        let generic_ratio = calculate_generic_ratio(&fp.identifiers);
                        let ast_size = fp.normalized_bigrams.len();
                        let sig = crate::detectors::ast_fingerprint::compute_minhash_signature(
                            &fp.normalized_bigrams,
                        );
                        let string_literals =
                            extract_string_literals(content, func.line_start, func.line_end);
                        Some(FuncWithSig {
                            qualified_name: format!("{}::{}", path_str, func.name),
                            name: func.name,
                            file_path: path_str.clone(),
                            line_start: func.line_start,
                            line_end: func.line_end,
                            string_literals,
                            loc,
                            generic_ratio,
                            ast_size,
                            sig,
                        })
                    })
                    .collect::<Vec<_>>()
            })
            .collect();

        // Filter out test functions BEFORE building parallel arrays.
        // Inline #[cfg(test)] modules live in production source files, so the
        // file-level is_test_path filter doesn't catch them. Use graph context
        // when available, fall back to name heuristic.
        let func_sigs: Vec<FuncWithSig> = func_sigs
            .into_iter()
            .filter(|fs| {
                if let Some(graph_func) = ctx.graph.find_function_at(&fs.file_path, fs.line_start) {
                    let i = ctx.graph.interner();
                    let qn = graph_func.qn(i);
                    if ctx.is_test_function(qn) {
                        return false;
                    }
                    let decos = ctx.decorators(qn);
                    for d in decos {
                        if d == "test" || d.starts_with("cfg(test") {
                            return false;
                        }
                    }
                    true
                } else {
                    if fs.name.starts_with("test_") {
                        info!(
                            "AIDupBlock filter: find_function_at MISS for test fn {}:{} (file={})",
                            fs.name, fs.line_start, fs.file_path
                        );
                    }
                    !fs.name.starts_with("test_")
                }
            })
            .collect();

        // Assign sig_idx sequentially (signatures must be contiguous)
        for fs in func_sigs {
            let sig_idx = all_signatures.len();
            all_signatures.push(fs.sig);
            all_functions.push(FunctionData {
                qualified_name: fs.qualified_name,
                name: fs.name,
                file_path: fs.file_path,
                line_start: fs.line_start,
                line_end: fs.line_end,
                loc: fs.loc,
                sig_idx,
                generic_ratio: fs.generic_ratio,
                ast_size: fs.ast_size,
                string_literals: fs.string_literals,
            });
        }

        let duplicates = self.find_duplicates(&all_functions, &all_signatures);

        // Graph-verified semantic overlap: reject coincidental matches
        let duplicates: Vec<_> = duplicates
            .into_iter()
            .filter(|(func1, func2, similarity)| {
                Self::verify_semantic_overlap(func1, func2, *similarity, ctx.graph)
            })
            .collect();

        let mut findings = Vec::new();
        for (func1, func2, similarity) in &duplicates {
            findings.push(self.create_finding(func1, func2, *similarity));
            if findings.len() >= self.max_findings {
                break;
            }
        }

        info!("AIDuplicateBlockDetector found {} findings", findings.len());
        Ok(findings)
    }
}

impl crate::detectors::RegisteredDetector for AIDuplicateBlockDetector {
    fn create(_init: &crate::detectors::DetectorInit) -> std::sync::Arc<dyn Detector> {
        std::sync::Arc::new(Self::new())
    }
}

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

    #[test]
    fn test_minhash_jaccard_identical() {
        let set: HashSet<String> = ["a", "b", "c", "d"].iter().map(|s| s.to_string()).collect();
        let sig = crate::detectors::ast_fingerprint::compute_minhash_signature(&set);
        let sim = crate::detectors::ast_fingerprint::minhash_jaccard(&sig, &sig);
        assert_eq!(sim, 1.0, "Identical sets should have Jaccard 1.0");
    }

    #[test]
    fn test_minhash_jaccard_disjoint() {
        let set1: HashSet<String> = (0..50).map(|i| format!("a_{}", i)).collect();
        let set2: HashSet<String> = (50..100).map(|i| format!("b_{}", i)).collect();
        let sig1 = crate::detectors::ast_fingerprint::compute_minhash_signature(&set1);
        let sig2 = crate::detectors::ast_fingerprint::compute_minhash_signature(&set2);
        let sim = crate::detectors::ast_fingerprint::minhash_jaccard(&sig1, &sig2);
        assert!(
            sim < 0.2,
            "Disjoint sets should have low Jaccard, got {}",
            sim
        );
    }

    #[test]
    fn test_generic_ratio() {
        let identifiers = vec![
            "result".to_string(),
            "temp".to_string(),
            "user_id".to_string(),
            "data".to_string(),
        ];
        let ratio = calculate_generic_ratio(&identifiers);
        // result, temp, data are generic (3/4 = 0.75)
        assert!((ratio - 0.75).abs() < 0.01);

        let no_generic = vec!["user_id".to_string(), "order_amount".to_string()];
        assert_eq!(calculate_generic_ratio(&no_generic), 0.0);

        let empty: Vec<String> = vec![];
        assert_eq!(calculate_generic_ratio(&empty), 0.0);
    }

    #[test]
    fn test_detector_defaults() {
        let detector = AIDuplicateBlockDetector::new();
        assert!((detector.similarity_threshold - 0.70).abs() < 0.01);
        assert!((detector.generic_name_threshold - 0.40).abs() < 0.01);
        assert_eq!(detector.min_loc, 5);
    }

    #[test]
    fn test_detects_near_duplicates() {
        // Two functions with identical structure but different variable names
        // across different files — classic Type-2 clone.
        let store = crate::graph::GraphBuilder::new().freeze();
        let detector = AIDuplicateBlockDetector::new();
        let ctx = crate::detectors::analysis_context::AnalysisContext::test_with_mock_files(&store, vec![
            (
                "services/user_service.py",
                "def process_user(data):\n    result = validate(data)\n    if result is None:\n        raise ValueError('invalid')\n    output = transform(result)\n    return output\n",
            ),
            (
                "services/order_service.py",
                "def process_order(info):\n    value = validate(info)\n    if value is None:\n        raise ValueError('invalid')\n    output = transform(value)\n    return output\n",
            ),
        ]);
        let findings = detector
            .detect(&ctx)
            .expect("should detect duplicate blocks");
        assert!(
            !findings.is_empty(),
            "Should detect near-duplicate functions with same structure but different variable names"
        );
        assert_eq!(findings[0].detector, "AIDuplicateBlockDetector");
        assert!(
            findings[0].title.contains("process_user")
                || findings[0].title.contains("process_order"),
            "Finding title should reference the duplicate function names. Got: {}",
            findings[0].title
        );
        // Duplicate code is a refactoring smell, never Critical or High
        assert!(
            findings[0].severity == Severity::Medium || findings[0].severity == Severity::Low,
            "Duplicate-block severity must be at most Medium, got {:?}",
            findings[0].severity
        );
    }

    #[test]
    fn test_severity_cap_exact_match() {
        // create_finding with similarity = 1.0 (100% AST) must not exceed Medium
        let detector = AIDuplicateBlockDetector::new();
        let func1 = FunctionData {
            qualified_name: "a.rs::escapeHtml".to_string(),
            name: "escapeHtml".to_string(),
            file_path: "src/a.rs".to_string(),
            line_start: 1,
            line_end: 10,
            loc: 10,
            sig_idx: 0,
            generic_ratio: 0.0,
            ast_size: 20,
            string_literals: vec![],
        };
        let func2 = FunctionData {
            qualified_name: "b.rs::escapeHtml".to_string(),
            name: "escapeHtml".to_string(),
            file_path: "src/b.rs".to_string(),
            line_start: 1,
            line_end: 10,
            loc: 10,
            sig_idx: 1,
            generic_ratio: 0.0,
            ast_size: 20,
            string_literals: vec![],
        };
        // 100% AST match — must be Medium, never Critical or High
        let finding = detector.create_finding(&func1, &func2, 1.0);
        assert_eq!(
            finding.severity,
            Severity::Medium,
            "100% AST-match duplicate must be Severity::Medium, got {:?}",
            finding.severity
        );
    }

    #[test]
    fn test_severity_cap_partial_match() {
        // create_finding with similarity = 0.75 (partial match) must be Low
        let detector = AIDuplicateBlockDetector::new();
        let func1 = FunctionData {
            qualified_name: "a.rs::parseJsonOrText".to_string(),
            name: "parseJsonOrText".to_string(),
            file_path: "src/a.rs".to_string(),
            line_start: 1,
            line_end: 8,
            loc: 8,
            sig_idx: 0,
            generic_ratio: 0.0,
            ast_size: 15,
            string_literals: vec![],
        };
        let func2 = FunctionData {
            qualified_name: "b.rs::parseJsonOrText".to_string(),
            name: "parseJsonOrText".to_string(),
            file_path: "src/b.rs".to_string(),
            line_start: 1,
            line_end: 8,
            loc: 8,
            sig_idx: 1,
            generic_ratio: 0.0,
            ast_size: 15,
            string_literals: vec![],
        };
        // 75% match — below the 0.85 threshold, must be Low
        let finding = detector.create_finding(&func1, &func2, 0.75);
        assert_eq!(
            finding.severity,
            Severity::Low,
            "75% AST-match duplicate must be Severity::Low, got {:?}",
            finding.severity
        );
    }

    #[test]
    fn test_no_finding_for_different_functions() {
        // Two structurally different functions — should produce no duplicates.
        let store = crate::graph::GraphBuilder::new().freeze();
        let detector = AIDuplicateBlockDetector::new();
        let ctx = crate::detectors::analysis_context::AnalysisContext::test_with_mock_files(&store, vec![
            (
                "auth.py",
                "def login(username, password):\n    user = authenticate(username, password)\n    if user is None:\n        raise AuthError('Invalid credentials')\n    token = create_token(user)\n    return token\n",
            ),
            (
                "export.py",
                "def export_csv(data, output_path):\n    with open(output_path, 'w') as f:\n        writer = csv.writer(f)\n        writer.writerow(data[0].keys())\n        for row in data:\n            writer.writerow(row.values())\n",
            ),
        ]);
        let findings = detector
            .detect(&ctx)
            .expect("should detect different functions");
        assert!(
            findings.is_empty(),
            "Should not flag structurally different functions. Found: {:?}",
            findings.iter().map(|f| &f.title).collect::<Vec<_>>()
        );
    }
}