We Removed C++

Fish 4.0 has shipped with 0% C++ and an almost entirely Rust codebase, marking a rare “rewrite-while-staying-shippable” modernization for a widely used Unix shell. The move matters because it targets the pain points that made C++ a poor fit for Fish’s day-to-day needs—especially tooling, safety, and concurrency—while keeping the shell usable throughout the transition.

The port was framed as a component-by-component replacement that preserved a working Fish at every stage, avoiding months-long “disappear into a hole” rewrites that would have stalled development and broken end-to-end testing. Built-ins—small programs with their own argument streams and exit codes—became the first targets because they could be ported independently. The team used a “strangler pattern” approach: Rust and C++ coexisted while calls crossed the boundary through FFI glue, with C++ versions kept around until the Rust replacements were complete. To keep comparisons meaningful, Fish largely retained the existing C++ subsystem structure instead of doing a full redesign.

The C++ critique was practical rather than theoretical. The transition highlighted recurring friction in short tools and compilers, platform differences, and ergonomics—especially around thread safety and ownership. Developers described how C++ changes often triggered cascading debates over pointer types and ownership conventions, and how tooling like autocomplete and definition jumping could fail unpredictably in large codebases. Safety concerns also came up repeatedly: verbose and error-prone string handling, unsafe char-pointer temptations, and modern C++ features like string_view being easy to misuse into use-after-free bugs.

Rust was chosen for its tooling and its concurrency model. The team emphasized rustup for easy setup, strong compiler errors, and the “send”/“sync” type system as a guardrail for cross-thread correctness. While Rust isn’t portrayed as perfect—portability still requires careful target enumeration and can devolve into “ifdef life”—the overall development experience was considered better aligned with Fish’s needs.

Key technical outcomes included replacing the curses dependency with a Rust-based terminfo approach, eliminating a major source of build grief and reducing global-state awkwardness. The port also enabled packaging improvements: Fish can be distributed as self-installable packages that bundle scripts, completions, and assets, and the team produced statically linked Linux binaries as single-file downloads for users without root access.

The project’s timeline stretched from an initial Rust rewrite proposal in 2023 to a beta in 2024, with the full release arriving in February. Along the way, the team hit dead ends and false starts—especially around auto-generated C++ bindings (autocxx), portability edge cases, and translation mismatches that sometimes surfaced as crashes or panics. Even so, performance landed slightly better in time-to-run, with memory showing a modest tradeoff.

Fish 4.0 remains “an odd duck” in Rust terms, retaining some low-level C-like patterns (such as direct file descriptor handling and UTF-32 strings), but the release is positioned as a foundation for further modernization—particularly around unlocking fully multi-threaded execution for features like asynchronous prompts and non-blocking completions. C++ is declared effectively dead for Fish, not as a slogan, but as a shipped, testable reality.