Scalable Polonius support on nightly (original) (raw)

Summary

Keep working on implementing a native rustc version of the Polonius next generation borrow checking algorithm, that would scale better than the previous datalog implementation, continuing from the 2024h2 goal.

Motivation

Polonius is an improved version of the borrow checker that resolves common limitations of the borrow checker and which is needed to support future patterns such as "lending iterators" (see #92985). Its model also prepares us for further improvements in the future.

Some support exists on nightly, but this older prototype has no path to stabilization due to scalability issues. We need an improved architecture and implementation to fix these issues.

The next six months

• Complete the ongoing work to land polonius on nightly

The "shiny future" we are working towards

Stable support for Polonius.

Ownership and team asks

Owner: lqd

Other support provided by Amanda Stjerna as part of her PhD.

Support needed from the project

We expect most support to be needed from the types team, for design, reviews, interactions with the trait solver, and so on. We expect Niko Matsakis, leading the polonius working group and design, to provide guidance and design time, and Michael Goulet and Matthew Jasper to help with reviews.

Outputs and milestones

Outputs

Nightly implementation of polonius that passes NLL problem case #3 and accepts lending iterators (#92985).

Performance should be reasonable enough that we can run the full test suite, do crater runs, and test it on CI, without significant slowdowns. We do not expect to be production-ready yet by then, and therefore the implementation would still be gated under a nightly -Z feature flag.

As our model is a superset of NLLs, we expect little to no diagnostics regressions, but improvements would probably still be needed for the new errors.

Milestones

Note: some of these are currently being worked on as part of the 2024h2 goal, and could be completed before the 2025h1 period.