absl::Status ExampleToNumpyDict()

in tfx_bsl/cc/coders/example_numpy_decoder.cc [33:108]
63 lines of code
10 McCabe index (conditional complexity)

absl::Status ExampleToNumpyDict(absl::string_view serialized_proto,
                                PyObject** result) {
  // Import numpy. (This is actually a macro, and "ret" is the return value
  // if import fails.)
  import_array1(/*ret=*/absl::InternalError("Unable to import numpy."));
  Example example;
  if (!example.ParseFromArray(serialized_proto.data(),
                              serialized_proto.size())) {
    return absl::DataLossError("Failed to parse input proto.");
  }

  // Initialize Python result dict.
  *result = PyDict_New();

  // Iterate over the features and add it to the dict.
  for (const auto& p : example.features().feature()) {
    const std::string& feature_name = p.first;
    const Feature& feature = p.second;

    PyObject* feature_values_ndarray;

    switch (feature.kind_case()) {
      case Feature::kBytesList: {
        const auto& values = feature.bytes_list().value();
        // Creating ndarray.
        npy_intp values_dims[] = {static_cast<npy_intp>(values.size())};
        feature_values_ndarray =
            PyArray_SimpleNew(1, values_dims, PyArray_OBJECT);
        PyObject** buffer =
            reinterpret_cast<PyObject**>(PyArray_DATA(feature_values_ndarray));
        for (int i = 0; i < values.size(); ++i) {
          const std::string& v = values[i];
          buffer[i] = PyBytes_FromStringAndSize(v.data(), v.size());
        }
        break;
      }
      case Feature::kFloatList: {
        const auto& values = feature.float_list().value();
        // Creating ndarray.
        npy_intp values_dims[] = {static_cast<npy_intp>(values.size())};
        feature_values_ndarray =
            PyArray_SimpleNew(1, values_dims, PyArray_FLOAT32);
        memcpy(reinterpret_cast<void*>(PyArray_DATA(feature_values_ndarray)),
               values.data(), values.size() * sizeof(float));
        break;
      }
      case Feature::kInt64List: {
        const auto& values = feature.int64_list().value();
        // Creating ndarray.
        npy_intp values_dims[] = {static_cast<npy_intp>(values.size())};
        feature_values_ndarray =
            PyArray_SimpleNew(1, values_dims, PyArray_INT64);
        memcpy(reinterpret_cast<void*>(PyArray_DATA(feature_values_ndarray)),
               values.data(), values.size() * sizeof(int64_t));
        break;
      }
      case Feature::KIND_NOT_SET: {
        // If we have a feature with no value list, we consider it to be a
        // missing value.
        feature_values_ndarray = Py_None;
        Py_INCREF(Py_None);
        break;
      }
      default: {
        return absl::DataLossError("Invalid value list in input proto.");
      }
    }
    const int err = PyDict_SetItemString(
        *result, feature_name.data(), feature_values_ndarray);
    Py_XDECREF(feature_values_ndarray);
    if (err == -1) {
      return absl::InternalError("Failed to insert item into Dict.");
    }
  }
  return absl::OkStatus();
}