Marrow-Lang

Overview

Marrow-Lang is my compiled language project — a from-scratch compiler pipeline written in C, targeting Cortex-VM as the backend. The entire toolchain is hand-rolled: no parser generators, no LLVM, no runtime borrowed from another language. Every component from lexer to code emission is designed and implemented directly.

Building Cortex-VM first was intentional. The VM's flat calling convention, word-addressed memory model, and 64-register file were all designed with Marrow's code generation needs in mind. Now that the backend is stable and well-tested, the compiler frontend is the active work.

Current Status

• Tokenizer — complete; parses source and prints token streams
• Number subsystem — complete; arbitrary-precision decimal digit representation with basic arithmetic
• Parser / AST — in progress; building AST from token stream
• Bytecode emission — next; will emit Cortex-VM assembly via the cortexAssemble library API
• VM execution — pending; Cortex-VM is ready and waiting at v0.5.0, 201 tests passing

Pipeline

• Lexer — tokenises source into a flat token stream
• Parser — hand-rolled recursive-descent, produces an AST; AST nodes allocated through the arena library
• Semantic analysis — type checking and name resolution
• Code generation — walks the AST and emits Cortex-VM assembly, assembled via cortexAssemble()

Why a Custom Backend?

Targeting LLVM means inheriting its complexity. Targeting x86 directly means writing a platform-specific backend and wrestling with a calling convention I didn't design. Cortex-VM gives me a backend I understand completely — every instruction, every calling convention rule, every binary format detail is something I wrote and can change. That control makes it much easier to experiment with language semantics without getting stuck on backend mysteries.