ArxJIT

ArxJIT is the planned Numba-style route from a restricted subset of pure Python to ASTx, IRx, LLVM, and native callable code.

Important: native JIT compilation is not implemented yet. A decorated function currently executes the original Python function.

Current API

from arxjit import i64, jit


@jit(signature=i64(i64, i64), cache=True)
def add(left, right):
    return left + right


assert add(20, 22) == 42  # Python fallback today

Implemented foundations:

• @jit and the JitFunction wrapper
• scalar signatures: i32, i64, f32, f64, and bool_
• robust function-source extraction with real-file locations
• validation for the proposed scalar Python subset
• structured diagnostics and public error types

The validator accepts typed scalar arguments, arithmetic/comparison/Boolean expressions, single-target assignments, if/else, while, for over the builtin range, and return. It rejects unsupported constructs with one diagnostic per violation.

Not implemented:

• Python AST to ASTx lowering
• IRx compilation or native function calls
• runtime argument/result marshalling
• signature inference and artifact caching
• array, Tensor, or Apache Arrow signatures

See the staged design for the Python-to-ASTx pipeline.

License: Apache-2.0.