Graydon Hoare: The Rust I Wanted Had No Future
In a 2023 post, Rust creator Graydon Hoare lists the design choices he disagreed with or lost, from async/await and traits to explicit lifetimes, and explains how his priorities differed from the community's focus on zero-cost abstractions.
In June 2023, Graydon Hoare published a post titled “The Rust I Wanted Had No Future.” It responded to a podcast remark by Jeremy Soller, who said he believed Hoare “would have said no to some things we all like now.” Hoare agreed emphatically: “the Rust We Got is many, many miles away from The Rust I Wanted.” He also wrote that he likes the result and is “very happy to use it in preference to C++.”
Hoare catalogued two kinds of departures. Some early design choices he now considers half-baked and improved by the community—such as opt-in move semantics, a broken effect system, first-class modules that didn’t work, and Erlang-style actors without threads. But he listed a much longer set of “disagreeable divergences,” where he still believes the community chose the wrong direction.
Where the language went off his path
He grouped the disagreements into four patterns: his idea lost to a more persistent or influential advocate; someone’s prototype temporarily replaced his idea and never got revisited; porting to LLVM forced a compromise that stuck; and his design was removed to avoid bad foundations, leaving a vacuum filled later by the ecosystem.
Among the choices he still dislikes: cross-crate inlining and monomorphization, which he blames for Rust’s slow compile times and lack of a stable ABI. He wanted crates to expose stable entry points while inlining internally, similar to Swift. He also wanted containers, iteration, and smart pointers as compiler built-ins rather than library-defined code that requires aggressive inlining.
For iteration, Hoare preferred stack-based, non-escaping coroutines (interior iteration) over today’s exterior iterators. He noted that LLVM now supports such coroutines and that they appear in languages like BLISS and Modula-2. Async/await is another major split: he originally wanted a standard green-thread runtime with growable stacks, closer to Go’s model. He admits a “grudging respect” for where Rust landed, especially heterogeneous selects, but says he would never have agreed to the direction as BDFL.
On types, Hoare rejects first-class references and explicit lifetimes. He wanted & to be a second-class parameter-passing mode only, not returnable or storable; he believed lifetimes should be fully inferred. He also dislikes traits/typeclasses, calling them too type-directed and global, and says they create too much coupling between libraries. He wanted a first-class module system in the ML tradition instead.
The post lists several missing features: quasiquotes, reflection, a proper error-handling system before the 2018 edition, automatic big integers, decimal floating-point, and tail calls. He also regrets the grammar’s complexity: he wanted to keep Rust LL(1), but today it is hard to parse and auto-format.
Simplicity over zero-cost abstractions
Hoare says these differences reflect a thematic divergence. “I would have traded performance and expressivity away for simplicity — both end-user cognitive load and implementation simplicity in the compiler.” He rejects the idea that zero-cost abstraction is a core promise of Rust, calling it a C++ idea that constrains the design space. His version would have been slower, likely in the performance band of Ada and Pascal, and less expressive, feeling “clunky and bureaucratic.”
He closes by saying the point isn’t to lobby the project with direction suggestions or grind axes, but to indicate thematic divergence. “Maybe you’ve nodded your head about one or two of the things above and think they’re good ideas — ‘hey, maybe we should have had a BDFL!’ — though more likely you think they’re terrible.” But his priorities, he writes, were never the priorities of the community that grew around Rust.