Rust's Two Biggest Compiler Projects Carry No Dates; the Measured Speed-Up Came From LLVM 23

Rust's new trait solver and the Polonius borrow checker both appear in the Rust project's 2026 goal list as accepted, and neither page gives a stabilization date. The one speed change a contributor calls "unambiguous" arrived from outside the project: LLVM 23.

That contrast comes from a post dated October 19 by Matyáš Racek, a compiler performance contributor who writes at Yoyo Code and says he now plans to work on Rust more seriously. His list is checked here against the project's own pages for the next solver and Polonius.

What the goal pages promise

The next-solver goal is to stabilize -Znext-solver=globally, which replaces the existing trait solver entirely. Its owners are lcnr and Niko Matsakis, and the page says the solver has been used for coherence checking since Rust 1.84 and replaced chalk in rust-analyzer. The remaining plan is triaging crater results, reaching performance parity with the old solver, writing an RFC on cycle semantics and a stabilization report.

Polonius is described as "an improved borrow checker that accepts conditional borrows and lending iterators". Rémy Rakic is the contact for the goal. The page accepts a compile-time cost of 10 to 20 percent and says worst cases are still unknown.

Project	Status	Date or size
Next trait solver	Accepted, in use for coherence since Rust 1.84	No date
Polonius Alpha	Accepted, nightly preview planned	No date
no-embed-metadata default	Announced August 18, 2026	5 to 30 percent smaller target directory
Maintainers in Residence	Announced August 26, 2026	No size

The table's last two rows are taken from Racek's post, not from the goal pages.

What is measured

Racek reports three performance data points. An LLVM 10 upgrade caused a regression of roughly 10 percent in rustc. LLVM 22 was green on instruction-count benchmarks but mostly red on wall time. LLVM 23, in his words, was an "unambiguous and unexpectedly large positive change in performance".

He gives no percentage for LLVM 23, and neither does any page fetched for this story, so the size of the win is his characterization alone.

The other number is disk, not speed. Making no-embed-metadata the default, announced on August 18, 2026, shrinks the target directory by 5 to 30 percent depending on the workload. A project with a large target folder in its build cache is the person who notices first.

1. Low end5 %
2. High end30 %

Source: Matyas Racek, Yoyo Code, October 19, 2026

What is still in flight

Racek says the parallel frontend is close to stabilization, restarted by a contributor named SparrowLi and driven by petrochenkov. He lists a proposed MIR pass by Amanieu to remove copies LLVM struggles to optimize, and a polymorphization experiment by camelid inspired by Swift. A shared cache for cargo check and cargo build is a project goal led by blyxyas.

A new cargo build-directory layout had its call for testing on March 13, 2026, and is meant to allow finer-grained locking of the target directory and caching across workspaces.

None of those items carries a benchmark in the post. Racek's closing observation is about organization rather than code: "Turns out that Conway's law is still a law."

What would change the picture

The two goal pages are explicit about funding and support. For the next solver, the core contribution is listed as fully funded and the acceleration ask as not yet funded. Polonius needs the types team at a "Large" support level for design review and the stabilization decision.

For readers tracking the other Rust compiler story this week, Cloudflare's tsrs port of TypeScript is a Rust program, not a change to rustc, and its 8x figure is the vendor's. A different compiler project, Futhark, has its own fix for integer division on this site.

The next checkpoint is the Polonius nightly preview, which its goal page says will ship behind a feature gate with a call for testing and a blog post. No date has been published for it.