Fuzzing The Gleam Compiler

A reporter examines the process of uncovering potential bugs in the Gleam compiler through a technique called structure-aware fuzzing. This method involves generating valid code or abstract syntax trees (ASTs) to test the compiler's behavior across both JavaScript and Erlang targets. The reporter notes that fuzzing, which typically relies on randomized inputs, can be expensive, non-deterministic, and somewhat unpredictable.

However, structure-aware fuzzing offers a more targeted approach by focusing on the compiler's code generation, potentially surfacing edge cases that might otherwise go unnoticed. The Gleam compiler's minimalistic syntax and support for static types make it an ideal candidate for fuzzing. The reporter also highlights the ease with which the Rust-written compiler can be integrated with existing fuzzing tools.

The project's fuzzer is designed to be generation-based rather than mutation-based, aiming to generate a diverse set of valid inputs to evaluate the compiler's output. The reporter emphasizes the importance of using the public compiler API to ensure compatibility with future versions of Gleam and minimize the risk of false positives or negatives.

The reporter shares examples of how the parser categorizes the compiler's outputs, providing insight into the nature of the generated test cases.