git_bayesect: Bayesian Git Bisection Tool

git_bayesect is a powerful tool designed to identify changes in the likelihood of events within a codebase using Bayesian inference. This tool is particularly useful for pinpointing the exact commit where a test, which was previously somewhat unreliable, becomes significantly more unstable. By leveraging Bayesian inference, git_bayesect employs a method of greedy minimization of expected entropy to select commits, using a Beta-Bernoulli conjugacy trick to manage unknown failure rates effectively.

Installation is straightforward, with options to use pip or the uv tool. Once installed, users can initiate a Bayesian bisection with commands like `git bayesect start --old $COMMIT` and record observations with `git bayesect fail` or `git bayesect pass --commit $COMMIT`. The tool also allows users to set priors for commits based on filenames or commit messages, enhancing the precision of the bisection process.

For those interested in experimenting, the repository includes a demo with a script that simulates a flaky test scenario. Users can create a fake repository, run the script, and use git_bayesect to identify the commit where the failure rate changed. This hands-on approach helps users understand the practical application of Bayesian bisection in real-world scenarios.