Align

Align is an AOT-compiled, data-oriented programming language designed to align human intent, AI generation, compiler optimization, and modern hardware.

Less code. Predictable performance.

This is an early-stage project. The design lives in draft.md + docs/; the compiler (alignc) is being implemented in Rust under crates/.

Status

Milestones M0–M3 are complete: programs flow end to end through lexer → parser → sema → MIR → LLVM → executable. M1 adds functions and calls, if/comparisons/bool, mut reassignment, structs (declaration, literal, field access), the full primitive set (i8..u64, f32/f64, char), and a builtin print wired to the runtime. M2 adds Option<T> with else-unwrap, Result<T, E> with ?, and a Result-returning main (mapped to an exit code). M3 adds the memory model: arena {} regions with bulk free, the heap box box<T> (heap.new/.get()/.clone()), move / use-after-move checking, and arena escape checking. M4 (core slice) adds fixed-length arrays and the fused pipeline [...].map(f).where(p).sum() — lowered to a single loop with no intermediate arrays.

cargo build
cargo test
cargo run --bin alignc -- run examples/arena.align     # arena + heap box; exits 42
cargo run --bin alignc -- run examples/pipeline.align  # fused map/where/sum; exits 24

alignc subcommands: check, emit-mir, emit-llvm, build, run.

Performance & portability

Align targets the cloud/container reality of build once, run on a varied fleet. The default build uses a safe, portable per-architecture baseline (x86-64-v2 on amd64, armv8-a/NEON on arm64), so one binary runs across mixed Intel/AMD/Graviton and feature-masked hosts. More aggressive targets are opt-in, never the default (--target-cpu native for a source build on the host you run on, or --target-cpu x86-64-v3 — a portable AVX2/FMA tier, the recommended "fast" build for a server fleet you control). Wide SIMD on a varied fleet comes from runtime CPU-feature dispatch in the library (one binary picks AVX2/NEON at run time, falling back safely) — not from a fixed high baseline. See draft.md §3.4 and docs/open-questions.md ("Build targets & portability"). The default build now targets that portable baseline (x86-64-v2 on amd64, the armv8-a floor on arm64); alignc build|run … --target-cpu native opts into the build host's exact CPU. (The codegen baseline + opt-in is in place; library runtime-dispatched SIMD still lands with the std/runtime layer — the current backend builds scalar IR and leans on LLVM -O2 autovectorization.)

Requirements

• Rust (stable)
• LLVM 19 (llvm-config on PATH), a C compiler (cc) for linking

Layout

• draft.md — authoritative language specification
• docs/ — design rationale, history, non-goals, open questions
• docs/impl/ — compiler implementation plan (stages, MIR, backend, roadmap)
• crates/ — the alignc compiler workspace

License

MIT