Unexpected behavior when passing np.ndarray to PauliOp constructor

[jamesETsmith]

General

@stand-by noticed some strange ctor behavior outlined below

Detail

      .def("__init__",
           [](fp::PauliOp<float_type> *new_obj, nb::ndarray<cfloat_t> coeffs,
              std::vector<fp::PauliString> const &pauli_strings) {
             fmt::println("ctor 1");
             auto [coeffs_vec, _] =
                 fp::__detail::ndarray_to_raw<cfloat_t, 1>(coeffs);
             new (new_obj) fp::PauliOp<float_type>(coeffs_vec, pauli_strings);
           })
      .def("__init__",
           [](fp::PauliOp<float_type> *new_obj,
              std::vector<cfloat_t> coeffs_vec,
              std::vector<fp::PauliString> const &pauli_strings) {
             fmt::println("ctor 2");
             new (new_obj) fp::PauliOp<float_type>(coeffs_vec, pauli_strings);
           })
      .def("__init__",
           [](fp::PauliOp<float_type> *new_obj,
              std::vector<cfloat_t> coeffs_vec,
              std::vector<std::string> const &strings) {
             fmt::println("ctor 3");
             std::vector<fp::PauliString> pauli_strings;
             std::transform(strings.begin(), strings.end(),
                            std::back_inserter(pauli_strings),
                            [](std::string const &pauli) {
                              return fp::PauliString(pauli);
                            });
             new (new_obj) fp::PauliOp<float_type>(coeffs_vec, pauli_strings);
           })

import fast_pauli._fast_pauli as fp
import numpy as np

n_strings = 10
coeffs = np.random.rand(n_strings).astype(np.complex128)
strings = ["".join(np.random.choice(["I", "X", "Y", "Z"], 4)) for _ in range(n_strings)]
print(strings)

print("should be in ctor 1")
op = fp.PauliOp(coeffs, [fp.PauliString(s) for s in strings])


print("should be in ctor 2")
op = fp.PauliOp(coeffs.tolist(), [fp.PauliString(s) for s in strings])

print("should be in ctor 3")
op = fp.PauliOp(coeffs.tolist(), strings)
op = fp.PauliOp(coeffs, strings)  # Why does this not error at runtime?