In the previous Polonius post, we formulated the original borrow checker in a Polonius-like style. In this post, we are going to explore how we can extend that formulation to be flow-sensitive. In so doing, we will enable the original Polonius goals, but also overcome some of its shortcomings. I believe this formulation is also more amenable to efficient implementation. As I’ll cover at the end, though, I do find myself wondering if there’s still more room for improvement.

lqd has been doing awesome work driving progress on polonius. He’s authoring an update for Inside Rust, but the TL;DR is that, with his latest PR, we’ve reimplemented the traditional Rust borrow checker in a more polonius-like style. We are working to iron out the last few performance hiccups and thinking about replacing the existing borrow checker with this new re-implementation, which is effectively a no-op from a user’s perspective (including from a performance perspective). This blog post walks through that work, describing how the new analysis works at a high-level. I plan to write some follow-up posts diving into how we can extend this analysis to be more precise (while hopefully remaining efficient).