cisv 0.2.3

High-performance CSV parser with SIMD optimizations

CISV Python Bindings (nanobind)

High-performance Python bindings for the CISV CSV parser using nanobind.

Performance

These bindings are 10-100x faster than the ctypes-based bindings because they:

1. Use the batch API: All data is parsed in C and returned at once, eliminating millions of per-field callbacks
2. Use nanobind: Much lower overhead than ctypes or pybind11
3. Release the GIL: Parallel parsing runs without holding the Python GIL

File Size	ctypes	nanobind	Speedup
142MB (1M rows × 10 cols)	~20s	<0.8s	25x+

Installation

From PyPI (recommended)

pip install cisv

From source

cd bindings/python-nanobind
pip install .

Development install

cd bindings/python-nanobind
pip install -e .

Usage

import cisv

# Parse a file
rows = cisv.parse_file('data.csv')

# Parse with options
rows = cisv.parse_file(
    'data.csv',
    delimiter=';',
    quote="'",
    trim=True,
    skip_empty_lines=True
)

# Parse large files in parallel (faster on multi-core systems)
rows = cisv.parse_file('large.csv', parallel=True)

# Parse a string
rows = cisv.parse_string("a,b,c\n1,2,3")

# Count rows quickly (SIMD-accelerated)
count = cisv.count_rows('data.csv')

API Reference

parse_file(path, delimiter=',', quote='"', *, trim=False, skip_empty_lines=False, parallel=False, num_threads=0)

Parse a CSV file and return all rows.

Parameters:

• path: Path to the CSV file
• delimiter: Field delimiter character (default: ',')
• quote: Quote character (default: '"')
• trim: Whether to trim whitespace from fields
• skip_empty_lines: Whether to skip empty lines
• parallel: Use multi-threaded parsing (faster for large files)
• num_threads: Number of threads for parallel parsing (0 = auto-detect)

Returns: List of rows, where each row is a list of field values.

parse_string(content, delimiter=',', quote='"', *, trim=False, skip_empty_lines=False)

Parse a CSV string and return all rows.

Parameters:

• content: CSV content as a string
• delimiter: Field delimiter character (default: ',')
• quote: Quote character (default: '"')
• trim: Whether to trim whitespace from fields
• skip_empty_lines: Whether to skip empty lines

Returns: List of rows, where each row is a list of field values.

count_rows(path)

Count the number of rows in a CSV file without full parsing.

This is very fast as it only scans for newlines using SIMD instructions.

Parameters:

• path: Path to the CSV file

Returns: Number of rows in the file.

Running Tests

cd bindings/python-nanobind
pip install -e ".[test]"
pytest

Benchmarking

pip install -e ".[benchmark]"
python -c "
import cisv
import time

# Create test file
with open('/tmp/test.csv', 'w') as f:
    f.write('col1,col2,col3\n')
    for i in range(100000):
        f.write(f'value{i}_1,value{i}_2,value{i}_3\n')

# Benchmark
start = time.time()
rows = cisv.parse_file('/tmp/test.csv')
print(f'Parsed {len(rows)} rows in {time.time()-start:.3f}s')
"