[Bug]: aiter() raises `async for`-specific TypeError message instead of CPython's

[jseop-lim]

Description

When aiter() is called with an object that lacks __aiter__, GraalPy raises TypeError with the message 'async for' requires object with __aiter__ method, got NoneType. CPython raises the same exception type with a different, builtin-specific message: 'NoneType' object is not an async iterable. This is a CPython compatibility issue.

Root Cause

The aiter() builtin in BuiltinFunctions.AIter delegates to GetAIterNode, which is the node used by the async for statement. GetAIterNode raises the async for-specific message ASYNC_FOR_NO_AITER ('async for' requires object with __aiter__ method, got %N), so the builtin leaks an async for-themed message to users who never wrote async for.

CPython keeps the two paths separate: async for goes through the GET_AITER opcode while builtin_aiter calls PyObject_GetAIter, which produces a distinct, object-centric message.

CPython reference: Objects/abstract.c#L2801-L2817

// PyObject_GetAIter raises a builtin-specific message (distinct from the
// GET_AITER opcode used by `async for`) when am_aiter is not implemented.
PyObject *
PyObject_GetAIter(PyObject *o) {
    PyTypeObject *t = Py_TYPE(o);
    if (t->tp_as_async == NULL || t->tp_as_async->am_aiter == NULL) {
        return type_error("'%.200s' object is not an async iterable", o);
    }
    ...
}

Note that GraalPy's own C-API port at com.oracle.graal.python.cext/src/abstract.c already uses the CPython message — only the Java-side builtin entry point reuses the async for node and inherits its message.

Fix

Make AIter replicate PyObject_GetAIter inline (check am_aiter slot, call it, verify the result with PyAIterCheckNode) and add a dedicated NOT_AN_ASYNC_ITERABLE error message, instead of delegating to GetAIterNode.

Reproduction

aiter(None)

Output

GraalPy:

TypeError: 'async for' requires object with __aiter__ method, got NoneType

CPython:

TypeError: 'NoneType' object is not an async iterable

Environment

• GraalPy 25.0.2 (Python 3.12.8)
• CPython 3.12.13
• OS: Debian 12