Between July 29 and September 28, 2026, the Rust compiler’s mean wall‑time dropped by 4.57 %, with 555 of 629 benchmark measurements improving and only 74 regressing, as reported by Nicholas Nethercote in his September 30 blog post. This overall gain came from a series of targeted improvements across the compiler ecosystem rather than a single change.
Notable contributions included Noah Lev’s work on rustdoc that delivered large speed wins; Jakub Beránek’s PR #159642 that enabled profile‑guided optimization for Clippy, yielding up to an 18 % wall‑time reduction in some benchmarks; and Nikita Popov’s upgrade of LLVM to version 23 (PR #158734) that gave a 1.2 % mean improvement. Nethercote himself changed the control‑flow‑graph traversal algorithm in dataflow analyses (PR #160193), cutting the number of apply_effects_in_block calls from 1.5 million to 90 thousand for the EverInitializedPlaces analysis in the cranelift‑codegen crate and achieving roughly a 30 % wall‑time reduction for that workload. Additional PRs from xmakro, Chris Denton, and others trimmed instruction counts in various hot paths.
While the new Polonius Alpha borrow checker and the Pineapple Häagen‑Dazs trait solver (initially nicknamed Penelope Hammertime) were enabled on Nightly and accept more valid programs, they initially caused compile‑time regressions in some cases. Jack Huey’s lazy liveness‑computation PRs (#161938, #163027) reduced instruction counts for serde by 3‑5 %, and Jana Dönszelmann’s detailed post outlines ongoing work to improve the new trait solver’s performance. Nethercote notes that despite these regressions, the “sea of green” in benchmark results shows that the many improvements have outweighed the costs in most scenarios, and he will soon join Hexcat to continue compiler performance optimizations.