Typify: A Lightweight Usage-driven Static Analyzer for Precise Python Type Inference

A team of researchers has unveiled Typify, a novel static analysis engine designed to solve a persistent problem in Python development: inferring accurate type information in large, dynamic codebases. Unlike current state-of-the-art tools like Type4Py and HiTyper, which rely on deep learning models trained on massive datasets, Typify takes a fundamentally different approach. It combines symbolic execution with iterative fixpoint analysis and a context-matching retrieval system to propagate type information by analyzing how functions and variables are actually used throughout a project's dependency graph. This method allows it to connect calls to definitions and infer semantics without any prior type annotations or statistical models.

In rigorous testing on real-world Python repositories from the ManyTypes4Py and Typilus datasets, Typify demonstrated its competitive edge. The tool consistently matched or surpassed the accuracy of both data-driven AI systems and industry-standard static analyzers like Facebook's Pyre. The research, accepted at the IEEE/ACM International Conference on Program Comprehension (ICPC 2026), proves that a precise, usage-driven logic-based approach can achieve top-tier results. This offers a practical alternative that is inherently more interpretable—developers can trace the reasoning behind each inferred type—and far more computationally efficient, as it eliminates the need for GPU-intensive model training or large corpora.

The significance of Typify lies in its potential to streamline the modernization of legacy Python code. For teams managing large, evolving, and sparsely annotated projects, it provides a lightweight, scalable tool to automatically add precise type hints. This enhances code quality, enables better IDE support and static error checking, and facilitates maintenance, all while being transparent and resource-friendly compared to opaque AI black boxes.