Meta Releases Pyrefly: A Python Language Server Written in Rust

Meta today announced the first public beta of its new Python language server Pyrefly, written in Rust.

This project was initially created as an internal tool for Meta’s Instagram developers, who faced significant performance issues while navigating codebases with existing Python tools.

Compared to other language servers, Pyrefly aims to be both fast and fully capable of type checking.

Main features include:

• Type Inference: Capable of inferring types in most locations (except function parameters)

• Flow Types: Understands the control flow of programs to optimize static types

• Incremental Processing: Designed for large-scale incremental processing at the module level, with optimized checks and parallelism

Pyrefly complies with the Language Server Protocol (LSP), allowing developers to experience fast autocompletion even in dense codebases.

It is important to note that as of the beta release in November 2025, Pyrefly complies with approximately 70% of Python type specifications, and may not yet be suitable for complex production codebases, but is appropriate for small personal projects.

CPython Considers Integrating the Rust Programming Language

CPython core developers have proposed a significant initiative: to integrate the Rust language into the Python core.

The preliminary plan is to support optional extension modules, with the long-term goal of making Rust a core dependency of CPython.

The proposal indicates that Rust can significantly enhance memory safety, reduce issues related to C language pointers, and bring modern language features and higher development efficiency to CPython. This move draws on the successful experiences of large projects like the Linux kernel and Android.

Currently, the proposal is still under discussion, with the community evaluating its potential impact on the build system, platform compatibility, and ecosystem. If successful, CPython may enter a C + Rust hybrid development phase.

New Compiler Backend Lamina Released

Today, a new compiler infrastructure Lamina has been released, which is a compiler backend that does not rely on LLVM or Cranelift.

Lamina generates native assembly for multiple architectures and is intended for building compilers for new languages, educational projects, and any project that can leverage custom code generation syntax.

Key features include:

• Direct Code Generation: From IR directly to assembly/machine code, without LLVM/Cranelift

• SSA-based IR: Single Static Assignment form, optimized for analysis and optimization passes

• MIR-based Code Generation (experimental): New intermediate representation with register allocation and advanced optimizations

• IRBuilder API: A fluent interface for building modules, functions, blocks, and control flow

• No External Backend Dependencies: Simplifies builds and provides a transparent pipeline for faster build speeds

Lamina is written in Rust (2024 edition), with the current version being 0.0.7.

New Modeling Language M and Toolchain Released

A new research paper published today on arXiv introduces a new modeling language and toolchain M, designed for reusable model compilation.

M is a textual, syntax-driven language based on participant models and extends discrete event scheduling semantics, providing constructs for modeling system entities, message-based interactions, and time or state-triggered reactions.

From these models, M can systematically generate diverse target artifacts, while maintaining semantic consistency with the original models. Additionally, M can serve as an intermediate language, allowing other modeling languages to anchor to it, thereby enabling them to benefit from its compilation framework.