Language Design Lab

• google/gemma4:26b-mlx delivers top speed at 42.8 tokens/s and sharp analytical output across logic, code, and summary tasks, outperforming larger...

AI agents successfully built and operated a semantic layer in Malloy, yet the experiment shows semantic layers become disposable when agents generate...

Chen Tianqi examined how AI compilers must adapt in the agent era and what future directions to pursue.

Key presentations explored:

• Triton language...

Rust 1.98 adds algebraic floating-point methods for f32/f64 that permit reordering and vectorization similar to -ffast-math.

• Performance boost:...

Qualcomm's acquisition of Modular accelerated Mojo's open-sourcing rather than delaying it, with the full compiler and toolchain released under Apache...

Odin repair benchmarks have been missing from LLM research, which has centered on Java and Python. OdinEval supplies 168 auditable instances drawn...

Background shapes language preference between Rust's abstractions and Zig's directness.

• Memory visibility: Functional chaining in Rust can hide...

SimDSL shows how a domain-specific language can embed an Entity Component System architecture to drive large-scale simulations directly on the GPU....

Go 1.27 introduces generic methods as a core language enhancement, alongside flexible struct field selectors and broader function type inference. This...

This taxonomy frames programming languages as combinations of recurring mechanisms—lexical rules, binding, type systems, concurrency primitives—that...

Mojo enables compiler optimizations through libraries rather than core compiler changes by building directly on MLIR. Designed by expert C++...

This paper investigates programs with continuations in the framework of classical linear logic, providing a theoretical basis for language features around control flow and resource management.

Rust's nightly experiment with the #[rustc_splat] attribute explores how far the existing trait system can support ergonomic function overloading for...

A template is not a function or a class but a pattern from which the compiler generates one per distinct argument set at the point of use. This structure turns templates into a separate language for compile-time computation in systems programming.

Grayscale emphasizes simplicity with guardrails to make systems programming approachable.

• Readable syntax like for_each, as_long_as, and or_return...

AI tools naturally favor established languages with abundant training data and proven ecosystems, according to TypeScript creator Anders Hejlsberg....

AI-driven code generation is scaling dramatically while also empowering automated attacks, making mechanically proven properties essential beyond...

• Software code becomes a throwaway artifact, routinely regenerated from evolved requirements via automation.
• Unambiguous specification languages...

Compiler calculation methods like Bahr and Hutton's monadic approach only consider converging source behavior, leaving compilers potentially unsound...

Erlang's set-theoretic type system trades the absence of principal types for annotation-driven refinement, enabling flexible typing in a dynamic...