polydup 0.4.0

Cross-language duplicate code detector

PolyDup Python Bindings

Python bindings for PolyDup, a cross-language duplicate code detector powered by Tree-sitter and Rabin-Karp hashing.

Features

• Multi-language support: Detect duplicates across Rust, Python, and JavaScript/TypeScript
• Type-2 clone detection: Finds structurally similar code (normalized identifiers/literals)
• GIL-free scanning: Releases Python's Global Interpreter Lock during CPU-intensive operations
• Parallel processing: Built on Rayon for multi-core performance
• Zero-copy architecture: Direct FFI to Rust core for minimal overhead

Installation

From Source (Development)

cd crates/polydup-py
maturin develop --release

From PyPI (Future)

pip install polydup

Usage

Basic Example

import polydup

# Scan a directory for duplicates
report = polydup.find_duplicates(
    paths=['./src', './lib'],
    min_block_size=50,    # Minimum tokens per code block
    threshold=0.85        # 85% similarity threshold
)

print(f"Scanned {report.files_scanned} files")
print(f"Analyzed {report.functions_analyzed} functions")
print(f"Found {len(report.duplicates)} duplicates")
print(f"Took {report.stats.duration_ms}ms")

# Iterate through duplicates
for dup in report.duplicates:
    print(f"\n{dup.file1}:{dup.start_line1} ↔️ {dup.file2}:{dup.start_line2}")
    print(f"  Similarity: {dup.similarity * 100:.1f}%")
    print(f"  Length: {dup.length} tokens")

Dictionary Output

For JSON serialization or dict-based workflows:

import polydup
import json

report_dict = polydup.find_duplicates_dict(
    paths=['./src'],
    min_block_size=30,
    threshold=0.9
)

# Serialize to JSON
print(json.dumps(report_dict, indent=2))

Concurrent Execution

Critical: PolyDup releases the GIL during scanning, allowing concurrent Python code:

import polydup
import concurrent.futures

def scan_project(path):
    return polydup.find_duplicates([path])

with concurrent.futures.ThreadPoolExecutor(max_workers=4) as executor:
    # These scans run in parallel thanks to GIL release
    futures = [
        executor.submit(scan_project, './project1'),
        executor.submit(scan_project, './project2'),
        executor.submit(scan_project, './project3'),
    ]

    for future in concurrent.futures.as_completed(futures):
        report = future.result()
        print(f"Found {len(report.duplicates)} duplicates")

API Reference

find_duplicates(paths, min_block_size=50, threshold=0.85)

Scan files for duplicate code and return a Report object.

Parameters:

• paths (list[str]): List of file or directory paths to scan
• min_block_size (int, optional): Minimum code block size in tokens. Default: 50
• threshold (float, optional): Similarity threshold (0.0-1.0). Default: 0.85

Returns: Report object with scan results

Raises: RuntimeError if scanning fails

---

find_duplicates_dict(paths, min_block_size=50, threshold=0.85)

Same as find_duplicates() but returns a Python dictionary.

Returns: dict with keys:

• files_scanned (int)
• functions_analyzed (int)
• duplicates (list[dict])
• stats (dict)

---

version()

Get the PolyDup library version.

Returns: str (e.g., "0.1.0")

---

Class: Report

Attributes:

• files_scanned (int): Number of files processed
• functions_analyzed (int): Number of functions extracted
• duplicates (list[DuplicateMatch]): List of detected duplicates
• stats (ScanStats): Performance metrics

Methods:

• to_dict(): Convert to Python dictionary
• __len__(): Returns number of duplicates

---

Class: DuplicateMatch

Attributes:

• file1 (str): First file path
• file2 (str): Second file path
• start_line1 (int): Starting line in first file
• start_line2 (int): Starting line in second file
• length (int): Length in tokens
• similarity (float): Similarity score (0.0-1.0)
• hash (str): Rolling hash value (hex string)

Methods:

• to_dict(): Convert to Python dictionary

---

Class: ScanStats

Attributes:

• total_lines (int): Total lines of code processed
• total_tokens (int): Total tokens analyzed
• unique_hashes (int): Number of unique code blocks
• duration_ms (int): Scan duration in milliseconds

Methods:

• to_dict(): Convert to Python dictionary

Performance

PolyDup's Python bindings use py.allow_threads() to release the Global Interpreter Lock during scanning. This enables:

1. Concurrent Python execution: Other Python threads continue running
2. True parallelism: Rust's Rayon uses all CPU cores
3. Minimal overhead: Zero-copy FFI with direct Rust integration

Benchmark Example

import polydup
import time

start = time.time()
report = polydup.find_duplicates(['./large-project'], min_block_size=30)
elapsed = time.time() - start

print(f"Scanned {report.files_scanned} files in {elapsed:.2f}s")
print(f"Found {len(report.duplicates)} duplicates")
print(f"Throughput: {report.stats.total_tokens / elapsed:.0f} tokens/sec")

Algorithm

PolyDup uses:

• Tree-sitter for language-agnostic AST parsing
• Token normalization for Type-2 clone detection (e.g., userId → $$ID)
• Rabin-Karp rolling hash with window size 50 for efficient similarity detection
• Rayon for parallel processing across CPU cores

See architecture-research.md for detailed algorithm analysis.

Development

Build

cd crates/polydup-py
maturin develop  # Debug build
maturin develop --release  # Optimized build

Test

python test.py

Type Checking

pip install mypy
mypy test.py

License

MIT OR Apache-2.0

Links

• GitHub: https://github.com/wiesnerbernard/polydup
• Core Library: polydup-core
• CLI Tool: polydup-cli
• Node.js Bindings: polydup-node