https://github.com/quansight-labs/mtypes

Memory Types for Python
https://github.com/quansight-labs/mtypes

Last synced: 5 months ago
JSON representation

Memory Types for Python

• Host: GitHub
• URL: https://github.com/quansight-labs/mtypes
• Owner: Quansight-Labs
• License: bsd-3-clause
• Created: 2018-09-05T00:48:37.000Z (about 7 years ago)
• Default Branch: master
• Last Pushed: 2021-02-19T18:30:08.000Z (over 4 years ago)
• Last Synced: 2025-04-02T08:05:21.088Z (6 months ago)
• Language: C
• Size: 219 KB
• Stars: 5
• Watchers: 4
• Forks: 5
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# `mtypes`

Memory Types for Python

## Table of Contents

- [`mtypes`](#mtypes)

  - [Table of Contents](#table-of-contents)

  - [Abstract](#abstract)

  - [Use Cases](#use-cases)

  - [Current Status](#current-status)

  - [Roadmap](#roadmap)

## Abstract

`mtypes` aims to tie Python's type system to other C level type hierarchies.

It provide two critical things that are useful for C extension types in Python:

- Easy marshalling of python types between Python and C.

- A universal mechanism for easily calling out to C code from Python.

`mtypes` does not aim to be another monolithic Python-to-C library, instead it focuses on extending the Python type system to make wrapping C libraries easier. It aims to do this by making it possible to associate C information with a type, and C information with objects, in addition to mechanisms that allow calling out to C natively from Python (the `__cdict__` protocol).

## Use Cases

`mtypes` intends to let objects work seamlessly between C and Python, meaning that there are several use cases. Common ones include:

- Wrapping numerical libraries.

  - So it could be of use to NumPy, PyTorch, Tensorflow, etc.

- Wrapping C utility libraries or device libraries.

  - So it can can be used to, for example, wrap common CUDA calls.

- Application Migration

  - So you can wrap existing applications or libraries, and give them an easy Python API.

Since C has a well-defined ABI, you can wrap any kind of library that's compatible with the C ABI. Examples include Fortran numerical libraries, C++ libraries that have a C-compatible API, and more.

## Current Status

Currently, `mtypes` can marshall a single C `long` to and from C, more to come.

## Roadmap

 - [ ] Specify API interface

   - [x] Specify struct API.

   - [ ] Specify `__cdict__` API.

 - [ ] Build out the `__cdict__` protocol.

   - [x] Build out basic structure.

   - [x] Build out the `box` and `unbox` methods.

   - [ ] Build out implementation-selection mechanism.

   - [ ] Build out marshalling for more complex types

   - [ ] Documentation