This New Language Just Beat C++.

Summary

The video examines Goose, a new programming language whose published benchmark suite reports performance ahead of C++ and Rust.

Memory management

Goose avoids a traditional heap allocator, garbage collection, and reference counting. Instead, it groups data into contiguous memory regions. Allocation can often be as simple as advancing a pointer, while freeing data can release an entire region with little overhead.

Arena allocation is also available in C++ and Rust, but Goose builds this model into the language. Its compiler tracks data and reference lifetimes.

References and memory efficiency

References to array elements can remain valid as arrays grow because the underlying data does not need to move. Goose also supports compact relative references—potentially two or four bytes instead of the usual eight—which can reduce memory use and improve cache efficiency.

Benchmarks

Across 16 published tests involving strings, graphs, trees, and parsers, the project reports that Goose averages 3.54× the speed of regular C++ and is about 19% faster than Rust. Heavily optimized C++ was reported to be about 26% slower.

In one heterogeneous-record test, Goose was about twice as fast as Rust while using nearly four times less memory. These results come from the project’s benchmark suite; the narrator notes that real-world testing is still needed.

Compilation

Goose translates programs into C, which is then compiled with a conventional C compiler. The narrator notes that this allows the project to use existing compiler optimizations and could make Goose suitable for some embedded applications.

Potential uses and outlook

The narrator sees potential uses in compute-heavy services, such as JSON processing, and in embedded projects. They do not expect Goose to replace established languages such as C++ or Rust.

Development

According to the project’s README, AI helped implement the compiler. The language architecture had been designed earlier, and Claude helped the author build much of the implementation.

Main speakers and sources

• Speaker: The video’s narrator from the Nandaba channel.
• Sources discussed: Goose’s repository, README, and published benchmarks. The author’s identity is unclear in the auto-generated captions.