Python Trending Weekly
ISSUE 98

CUDA Finally Adds Native Python Support!

2025-04-19 Season 4

TL;DR

Template strings are officially accepted, with PEP 787 making subprocess calls safer; a hardcore experiment writes a TCP stack in Python, and the A2A protocol for agent interoperability debuts.

Articles 12 Projects 12 Media 2
🦄

Articles & Tutorials

× 12
01 PEP 750 – Template Strings peps.python.org

PEP-750 has been accepted and will be merged into Python 3.14! It proposes introducing t-strings that return a Template type instead of a string, suitable for custom string processing, web templates, DSLs, and other scenarios while avoiding security issues like SQL injection and XSS attacks. (Also: PEP750 Template Strings Chinese Translation)

02 PEP 787 – Safer subprocess usage using t-strings peps.python.org

This PEP proposes extending the subprocess and shlex modules to support template strings (t-strings), enabling safer and more convenient execution of shell commands with interpolation.

03 Python at the Speed of Rust blog.fxn.ai

Introduces how to compile Python code to native code, achieving performance close to Rust. Uses CPython's frame evaluation API to implement symbolic tracing, builds IR graphs, and then progressively lowers IR graphs to platform-specific native code.

2025-04-18-rust-speed
04 What happens if you write a TCP stack in Python? jvns.ca

How to develop a simple TCP protocol stack using Python and scapy? The article analyzes the TCP three-way handshake process, implements its interaction mechanism through code, and demonstrates Python's speed limitations in network communication.

05 14 Advanced Python Features blog.edward-li.com

The article shares 14 advanced Python features with examples, including type overloading, generics, pattern matching, advanced f-strings, proxy properties, metaclasses, and more.

06 Is Python Code Sensitive to CPU Caching? lukasatkinson.de

Python doesn't support controlling memory layout of data structures - does this mean CPU caching has no impact on Python performance? After testing, the author found that random list access is 1.2-1.5 times slower than sequential access, indicating that cache optimization applies to interpreted languages as well.

07 Use polars in 2025 lrt.one

The author recommends using polars in data analysis projects because it offers better performance, more intuitive API design, and a more unified type system. The author also mentions two shortcomings of polars.

08 Mastering DuckDB when you're used to pandas or Polars: part 2 labs.quansight.org

The second part of a series introducing how to translate dataframe operations into SQL, including filtering with window functions, expression composition, and nested data types.

09 Stop Writing `__init__` Methods blog.glyph.im

Recommends avoiding writing __init__ methods to prevent redundancy, testing difficulties, and inflexibility issues. The author suggests using dataclass to auto-generate __init__, using classmethod as object factories, and strengthening type constraints through NewType.

10 Maps with Django⁽³⁾: GeoDjango, Pillow & GPS paulox.net

Introduces how to extract GPS information from images using the Django web framework, GeoDjango module, and Pillow image library, then create web maps with images.

11 Quick Primer on Model Context Protocol (MCP) polarsparc.com

Provides an accessible introduction to Model Context Protocol (MCP), demonstrating environment setup, practical operations using Python for MCP, and integration applications with other tools.

12 From DeepSeek LLM to DeepSeek R1 oilbeater.com

The first part of a series attempting to trace DeepSeek's step-by-step exploration journey (DeepSeek LLM -> DeepSeek MoE -> DeepSeek V2 -> DeepSeek V3 -> DeepSeek R1), showcasing DeepSeek's unique approach to technical route selection and model optimization.

🐿️

Projects & Resources

× 12
01 A2A: Protocol for agent-to-agent communication and interoperability github.com

Google's latest open-source Agent2Agent protocol, enabling seamless communication and collaboration between agents implemented in different frameworks. (12K stars)

02 FunctionTrace: Graphical Python performance profiler gitlab.com

Quickly see everything your program is doing without modifying any code. Integrates Firefox Profiler and supports multiple analysis views including stack charts, flame graphs, and call trees.

FunctionTrace: Graphical Python performance profiler
03 Spider: Web Crawler built using asynchronous Python and distributed task management github.com

A web crawler designed specifically for efficient distributed crawling and data extraction, built with asynchronous I/O, robust logging, plugin architecture, and Celery's distributed task processing.

04 pdf-craft: Convert scanned PDFs to other formats github.com

Focuses on converting scanned book PDF files to various other formats. Extracts main text while filtering out headers, footers, footnotes, page numbers, and other elements, using algorithms to handle cross-page continuity issues. (2.1K stars)

05 mineworld: A Real-time interactive world model on Minecraft github.com

Microsoft's newly open-sourced interactive world model based on Minecraft, featuring high generation quality, powerful controllability, and fast inference speed.

06 intentkit: An open and fair framework for everyone to build AI agents equipped with powerful skills github.com

Supports multi-agent systems, autonomous agent management, blockchain integration, social media integration (Twitter and Telegram), extensible skill systems, and extensible plugin systems. (6.1K stars)

07 airunner: Graphical tool for running large models locally github.com

Local AI chat tool requiring no complex configuration, usable for AI conversations and text generation, image generation and editing, text-to-speech, and speech-to-text.

08 pypacktrends: Compare Python package download volumes github.com

A website project that can search and view Python download trend charts, supporting comparative viewing of multiple packages' download trends.

09 cuda-python: Performance meets Productivity github.com

A toolkit for Python operations with CUDA, containing multiple related component packages. (2.3K stars)

10 convert-markdown: Convert Markdown to HTML/PDF and other formats github.com

Supports converting Markdown to HTML, PDF, DOCX, and PPTX, supports executing code blocks and rendering plots. Uses LLM output to create beautiful reports with charts.

convert-markdown: Convert Markdown to HTML/PDF and other formats
11 xlwings: Python and Excel calling each other github.com

Supports calling Python from Excel and vice versa. Supports Excel on Windows and MacOS, as well as Google Sheets and web-based Excel. (3.1K stars)

12 tariff: Make import great again... github.com

A purely satirical library mocking America's arbitrary import tariffs. It can set "tariffs" required for importing certain libraries (like numpy), with the effect of extending import time when importing these libraries... (2.2K stars)

🐢

Podcasts & Videos

× 2
01 Conf42 Python 2025: Video Collection youtube.com

A video collection from "Conf42 Python" with 50+ videos available.

02 Jupyter's Architecture Unpacked youtube.com

Video guests are two Jupyter contributors and project leaders, discussing topics including: How does Jupyter work? How did they develop an interactive language execution engine?

ESC
↑↓ select · ↵ open175 issues