for this function:

def f(a: List[int]) -> int:
    return len(a)

originally:

CPyTagged CPyDef_f(PyObject *cpy_r_a) {
    CPyTagged cpy_r_r0;
    Py_ssize_t __tmp1;
CPyL0: ;
    __tmp1 = PyList_GET_SIZE(cpy_r_a);
    cpy_r_r0 = CPyTagged_ShortFromSsize_t(__tmp1);
    return cpy_r_r0;
}

now:

CPyTagged CPyDef_f(PyObject *cpy_r_a) {
    CPyPtr cpy_r_r0;
    int64_t cpy_r_r1;
    CPyTagged cpy_r_r2;
CPyL0: ;
    cpy_r_r0 = (CPyPtr)&((PyVarObject *)cpy_r_a)->ob_size;
    cpy_r_r1 = *(int64_t *)cpy_r_r0;
    cpy_r_r2 = CPyTagged_ShortFromSsize_t(cpy_r_r1);
    return cpy_r_r2;
}

IR:

L0:
    r0 = get_element_ptr a ob_size :: PyVarObject
    r1 = load_mem r0 :: int64*
    r2 = CPyTagged_ShortFromSsize_t(r1)
    return r2

@JukkaL I find a potential problem when fixing the IR tests: although we implement builtins.len for list in the specializers, list_len_op is exported and used in some places directly. We should overcome that as well. I am thinking about adding a list_len to LowLevelBuilder so that both specializer and other places can use it. What do you think?

I am thinking about adding a list_len to LowLevelBuilder so that both specializer and other places can use it. What do you think?

That's a good idea.

I just updated that, I'll merge this once all tests go green.