Class Methods in Kernels

The problem

If you try to call a standard Python class method, such as self.get_value(), from inside a function decorated with @basic or @structural, the compiler will fail. Historically, before Kirin v0.16.8, this could appear as an AttributeError about arg_names.

Why this happens

A kernel is a piece of code that describes instructions intended to run on the target device, not on your host machine. Because of this, the Kirin compiler cannot evaluate arbitrary Python runtime expressions, such as object attribute lookups or dynamically calling class methods, during device execution.

The compiler only parses global constants and already-resolved kernel functions.

The solution

Resolve the Python method on the host before defining the kernel. By assigning the result to a local variable, the inner kernel can cleanly capture that variable as a constant.

Incorrect

This tries to evaluate the method inside the device kernel.

from kirin.prelude import basic


class Offset:
    def __init__(self, value: int):
        self.value = value

    def get_value(self) -> int:
        return self.value

    def bad_method(self):
        @basic
        def add_offset(x: int) -> int:
            # ERROR: Kirin cannot evaluate arbitrary Python class methods.
            return x + self.get_value()

        return add_offset

Correct

Resolve the method on the host side, then capture it in the kernel closure.

from kirin.prelude import basic


class Offset:
    def __init__(self, value: int):
        self.value = value

    def get_value(self) -> int:
        return self.value

    def method(self):
        # 1. Resolve the method in host Python code.
        offset = self.get_value()

        # 2. Define the device kernel.
        @basic
        def add_offset(x: int) -> int:
            # 3. Use the captured constant inside the kernel.
            return x + offset

        return add_offset