Salt

Salt is a systems programming language designed to provide mathematically verified safety without incurring runtime costs. Its core innovation is compile-time verification of program correctness, using Z3-proven contracts to ensure properties such as array bounds and preconditions are satisfied before code is executed. This enables developers to achieve mathematical certainty in program behavior, eliminating the need for garbage collection or borrow checking, and delivering performance on par with C compiled at high optimization levels.

Arenas are used for memory management in Salt, offering an alternative to borrow checking and garbage collection, and supporting efficient allocation patterns suitable for systems-level tasks. The language includes a standard library with data structures such as hash maps and offers features like optional chaining, f-strings, and direct manipulation of pointers and slices. Salt’s code generation leverages MLIR’s affine optimizer, and the language has been used to implement complex systems such as microkernels, neural network trainers, Redis-compatible data stores, and GPU-accelerated graphics engines. In each of these projects, Salt’s contracts and compile-time proofs are used to verify safety properties, such as array bounds in compute kernels and pixel writes in rendering engines.

The platform demonstrates performance parity with C in multiple benchmarks, including context switching, IPC, memory allocation, and networking throughput. Salt programs can be written and tested live, and the language is built alongside real systems to validate its compiler guarantees in practice. Notably, projects like Basalt (a Llama 2 inference engine), KeuOS (a microkernel), Lettuce (an in-memory data store), and FACET (a 2D graphics engine) have been implemented in Salt to showcase its capabilities and safety guarantees.

Salt is aimed at systems programmers seeking verified safety and high performance without external dependencies or runtime overhead.