ZNPY

Write extensions modules for Python 3.7+ in Zig instead of C.

Example

Simple and contrived:

const std = @import("std");
const znpy = @import("znpy");

comptime {
    znpy.defineExtensionModule();
}

pub fn divide_f32_default_1(
    args: struct {
        a: f32,
        b: f32 = 1,
    },
) f32 {
    return args.a / args.b;
}

Exhaustive feature list

• Tested on Python versions: 3.7, 3.8, 3.9, 3.10, 3.11, 3.12, 3.13

  • on Alpine Linux
  • on Arch Linux
  • on Windows 11

• Test on Numpy versions: [1.21 to 2.3] (API seems stable enough)

• Supported argument types:

  • 'bool'
    • auto-conversion from object.__bool__()
  • signed/unsigned integers - u8, u16, u32, u64, i8, i16, i32, i64
    • auto-conversion from number-like things
  • single/double precision floats - f32, f64
    • auto-conversion from number-like things
  • 'str' (immutable)
  • 'bytes' (immutable)
  • 'bytearray' (mutable)
  • 'numpy.ndarray' (mutable)
  • 'list' (mutable)
  • 'dict' (mutable)

• Supported return types:

  • void
  • 'bool'
  • signed/unsigned integers - u8, u16, u32, u64, i8, i16, i32, i64
  • single/double precision floats - f32, f64

• Supports Optional[T] on most parameters and all return types (except void, duh).

• Supports converting Zig errors to Python Exceptions.

• Supports keyword arguments

Not supported yet (WIP)

• Arguments:
  • 'iterable'
  • 'memoryview' (mutable/immutable)
  • 'file-descriptor' (file, socket, etc.)
  • streams (io.BytesIO, etc.)
  • custom data types
• Return values:
  • 'numpy.ndarray'
  • 'bytes' / 'bytearray'
  • 'list' / 'tuple' / 'dict'
  • custom data types
• QoL:
  • auto-conversion from array-like to numpy.ndarray