nd::array nd::view(const nd::array& arr, const ndt::type& tp)
{
// If the types match exactly, simply return 'arr'
if (arr.get_type() == tp) {
return arr;
} else if (arr.get_ndim() == tp.get_ndim()) {
// Allocate a result array to attempt the view in it
array result(make_array_memory_block(tp.get_metadata_size()));
// Copy the fields
result.get_ndo()->m_data_pointer = arr.get_ndo()->m_data_pointer;
if (arr.get_ndo()->m_data_reference == NULL) {
// Embedded data, need reference to the array
result.get_ndo()->m_data_reference = arr.get_memblock().release();
} else {
// Use the same data reference, avoid producing a chain
result.get_ndo()->m_data_reference = arr.get_data_memblock().release();
}
result.get_ndo()->m_type = ndt::type(tp).release();
result.get_ndo()->m_flags = arr.get_ndo()->m_flags;
// Now try to copy the metadata as a view
if (try_view(arr.get_type(), arr.get_ndo_meta(), tp,
result.get_ndo_meta(), arr.get_memblock().get())) {
// If it succeeded, return it
return result;
}
// Otherwise fall through, let it get destructed, and raise an error
}
stringstream ss;
ss << "Unable to view nd::array of type " << arr.get_type();
ss << "as type " << tp;
throw type_error(ss.str());
}
nd::array dynd::format_json(const nd::array& n)
{
// Create a UTF-8 string
nd::array result = nd::empty(ndt::make_string());
// Initialize the output with some memory
output_data out;
out.blockref = reinterpret_cast<const string_type_metadata *>(result.get_ndo_meta())->blockref;
out.api = get_memory_block_pod_allocator_api(out.blockref);
out.api->allocate(out.blockref, 1024, 1, &out.out_begin, &out.out_capacity_end);
out.out_end = out.out_begin;
if (!n.get_type().is_expression()) {
::format_json(out, n.get_type(), n.get_ndo_meta(), n.get_readonly_originptr());
} else {
nd::array tmp = n.eval();
::format_json(out, tmp.get_type(), tmp.get_ndo_meta(), tmp.get_readonly_originptr());
}
// Shrink the memory to fit, and set the pointers in the output
string_type_data *d = reinterpret_cast<string_type_data *>(result.get_readwrite_originptr());
d->begin = out.out_begin;
d->end = out.out_capacity_end;
out.api->resize(out.blockref, out.out_end - out.out_begin, &d->begin, &d->end);
// Finalize processing and mark the result as immutable
result.get_type().extended()->metadata_finalize_buffers(result.get_ndo_meta());
result.flag_as_immutable();
return result;
}
void dynd::parse_json(nd::array &out, const char *json_begin,
const char *json_end, const eval::eval_context *ectx)
{
try {
const char *begin = json_begin, *end = json_end;
ndt::type tp = out.get_type();
::parse_json(tp, out.get_ndo_meta(), out.get_readwrite_originptr(), begin, end, ectx);
begin = skip_whitespace(begin, end);
if (begin != end) {
throw json_parse_error(begin, "unexpected trailing JSON text", tp);
}
} catch (const json_parse_error& e) {
stringstream ss;
string line_prev, line_cur;
int line, column;
get_error_line_column(json_begin, json_end, e.get_position(),
line_prev, line_cur, line, column);
ss << "Error parsing JSON at line " << line << ", column " << column << "\n";
if (e.get_type().get_type_id() != uninitialized_type_id) {
ss << "DType: " << e.get_type() << "\n";
}
ss << "Message: " << e.get_message() << "\n";
print_json_parse_error_marker(ss, line_prev, line_cur, line, column);
throw runtime_error(ss.str());
}
}
static void set(const ndt::type& paramtype, char *metadata, char *data, const nd::array& value) {
if (paramtype.get_type_id() == void_pointer_type_id) {
// TODO: switch to a better mechanism for passing nd::array references
*reinterpret_cast<const array_preamble **>(data) = value.get_ndo();
} else {
typed_data_assign(paramtype, metadata, data, value.get_type(), value.get_ndo_meta(), value.get_ndo()->m_data_pointer);
}
}
static void json_as_buffer(const nd::array& json, nd::array& out_tmp_ref, const char *&begin, const char *&end)
{
// Check the type of 'json', and get pointers to the begin/end of a UTF-8 buffer
ndt::type json_type = json.get_type().value_type();
switch (json_type.get_kind()) {
case string_kind: {
const base_string_type *sdt = static_cast<const base_string_type *>(json_type.extended());
switch (sdt->get_encoding()) {
case string_encoding_ascii:
case string_encoding_utf_8:
out_tmp_ref = json.eval();
// The data is already UTF-8, so use the buffer directly
sdt->get_string_range(&begin, &end,
out_tmp_ref.get_ndo_meta(), out_tmp_ref.get_readonly_originptr());
break;
default: {
// The data needs to be converted to UTF-8 before parsing
ndt::type utf8_tp = ndt::make_string(string_encoding_utf_8);
out_tmp_ref = json.ucast(utf8_tp).eval();
sdt = static_cast<const base_string_type *>(utf8_tp.extended());
sdt->get_string_range(&begin, &end,
out_tmp_ref.get_ndo_meta(), out_tmp_ref.get_readonly_originptr());
break;
}
}
break;
}
case bytes_kind: {
out_tmp_ref = json.eval();
const base_bytes_type *bdt = static_cast<const base_bytes_type *>(json_type.extended());
bdt->get_bytes_range(&begin, &end,
out_tmp_ref.get_ndo_meta(), out_tmp_ref.get_readonly_originptr());
break;
}
default: {
stringstream ss;
ss << "Input for JSON parsing must be either bytes (interpreted as UTF-8) or a string, not ";
ss << json_type;
throw runtime_error(ss.str());
break;
}
}
}
uint32_t categorical_type::get_value_from_category(const nd::array& category) const
{
if (category.get_type() == m_category_tp) {
// If the type is right, get the category value directly
return get_value_from_category(category.get_ndo_meta(), category.get_readonly_originptr());
} else {
// Otherwise convert to the correct type, then get the category value
nd::array c = nd::empty(m_category_tp);
c.val_assign(category);
return get_value_from_category(c.get_ndo_meta(), c.get_readonly_originptr());
}
}
// Constructor which creates the output based on the input's broadcast shape
array_iter(const ndt::type& op0_dtype, nd::array& out_op0, const nd::array& op1, const nd::array& op2, const nd::array& op3) {
create_broadcast_result(op0_dtype, op1, op2, op3, out_op0, m_iter_ndim[0], m_itershape);
nd::array ops[4] = {out_op0, op1, op2, op3};
m_array_tp[0] = out_op0.get_type();
m_array_tp[1] = op1.get_type();
m_array_tp[2] = op2.get_type();
m_array_tp[3] = op3.get_type();
m_itersize = 1;
m_iter_ndim[1] = m_array_tp[1].get_ndim();
m_iter_ndim[2] = m_array_tp[2].get_ndim();
m_iter_ndim[3] = m_array_tp[3].get_ndim();
// Allocate and initialize the iterdata
if (m_iter_ndim[0] != 0) {
m_iterindex.init(m_iter_ndim[0]);
memset(m_iterindex.get(), 0, sizeof(intptr_t) * m_iter_ndim[0]);
// The destination iterdata
size_t iterdata_size = m_array_tp[0].get_iterdata_size(m_iter_ndim[0]);
m_iterdata[0] = reinterpret_cast<iterdata_common *>(malloc(iterdata_size));
if (!m_iterdata[0]) {
throw std::bad_alloc();
}
m_metadata[0] = out_op0.get_ndo_meta();
m_array_tp[0].iterdata_construct(m_iterdata[0],
&m_metadata[0], m_iter_ndim[0], m_itershape.get(), m_uniform_tp[0]);
m_data[0] = m_iterdata[0]->reset(m_iterdata[0], out_op0.get_readwrite_originptr(), m_iter_ndim[0]);
// The op iterdata
for (int i = 1; i < 4; ++i) {
iterdata_size = m_array_tp[i].get_broadcasted_iterdata_size(m_iter_ndim[i]);
m_iterdata[i] = reinterpret_cast<iterdata_common *>(malloc(iterdata_size));
if (!m_iterdata[i]) {
throw std::bad_alloc();
}
m_metadata[i] = ops[i].get_ndo_meta();
m_array_tp[i].broadcasted_iterdata_construct(m_iterdata[i],
&m_metadata[i], m_iter_ndim[i],
m_itershape.get() + (m_iter_ndim[0] - m_iter_ndim[i]), m_uniform_tp[i]);
m_data[i] = m_iterdata[i]->reset(m_iterdata[i], ops[i].get_ndo()->m_data_pointer, m_iter_ndim[0]);
}
for (size_t i = 0, i_end = m_iter_ndim[0]; i != i_end; ++i) {
m_itersize *= m_itershape[i];
}
} else {
for (size_t i = 0; i < 4; ++i) {
m_iterdata[i] = NULL;
m_uniform_tp[i] = m_array_tp[i];
m_data[i] = ops[i].get_ndo()->m_data_pointer;
m_metadata[i] = ops[i].get_ndo_meta();
}
}
}
categorical_type::categorical_type(const nd::array& categories, bool presorted)
: base_type(categorical_type_id, custom_kind, 4, 4, type_flag_scalar, 0, 0)
{
intptr_t category_count;
if (presorted) {
// This is construction shortcut, for the case when the categories are already
// sorted. No validation of this is done, the caller should have ensured it
// was correct already, typically by construction.
m_categories = categories.eval_immutable();
m_category_tp = m_categories.get_type().at(0);
category_count = categories.get_dim_size();
m_value_to_category_index.resize(category_count);
m_category_index_to_value.resize(category_count);
for (size_t i = 0; i != (size_t)category_count; ++i) {
m_value_to_category_index[i] = i;
m_category_index_to_value[i] = i;
}
} else {
// Process the categories array to make sure it's valid
const ndt::type& cdt = categories.get_type();
if (cdt.get_type_id() != strided_dim_type_id) {
throw runtime_error("categorical_type only supports construction from a strided array of categories");
}
m_category_tp = categories.get_type().at(0);
if (!m_category_tp.is_scalar()) {
throw runtime_error("categorical_type only supports construction from a 1-dimensional strided array of categories");
}
category_count = categories.get_dim_size();
intptr_t categories_stride = reinterpret_cast<const strided_dim_type_metadata *>(categories.get_ndo_meta())->stride;
const char *categories_element_metadata = categories.get_ndo_meta() + sizeof(strided_dim_type_metadata);
comparison_ckernel_builder k;
::make_comparison_kernel(&k, 0,
m_category_tp, categories_element_metadata,
m_category_tp, categories_element_metadata,
comparison_type_sorting_less, &eval::default_eval_context);
cmp less(k.get_function(), k.get());
set<const char *, cmp> uniques(less);
m_value_to_category_index.resize(category_count);
m_category_index_to_value.resize(category_count);
// create the mapping from indices of (to be lexicographically sorted) categories to values
for (size_t i = 0; i != (size_t)category_count; ++i) {
m_category_index_to_value[i] = i;
const char *category_value = categories.get_readonly_originptr() +
i * categories_stride;
if (uniques.find(category_value) == uniques.end()) {
uniques.insert(category_value);
} else {
stringstream ss;
ss << "categories must be unique: category value ";
m_category_tp.print_data(ss, categories_element_metadata, category_value);
ss << " appears more than once";
throw std::runtime_error(ss.str());
}
}
// TODO: Putting everything in a set already caused a sort operation to occur,
// there's no reason we should need a second sort.
std::sort(m_category_index_to_value.begin(), m_category_index_to_value.end(),
sorter(categories.get_readonly_originptr(), categories_stride,
k.get_function(), k.get()));
// invert the m_category_index_to_value permutation
for (uint32_t i = 0; i < m_category_index_to_value.size(); ++i) {
m_value_to_category_index[m_category_index_to_value[i]] = i;
}
m_categories = make_sorted_categories(uniques, m_category_tp,
categories_element_metadata);
}
// Use the number of categories to set which underlying integer storage to use
if (category_count <= 256) {
m_storage_type = ndt::make_type<uint8_t>();
} else if (category_count <= 65536) {
m_storage_type = ndt::make_type<uint16_t>();
} else {
m_storage_type = ndt::make_type<uint32_t>();
}
m_members.data_size = m_storage_type.get_data_size();
m_members.data_alignment = (uint8_t)m_storage_type.get_data_alignment();
}
array_iter(const nd::array& op0) {
init(op0.get_type(), op0.get_ndo_meta(), op0.get_readwrite_originptr());
}
dynd::nd::array pydynd::nd_fields(const nd::array& n, PyObject *field_list)
{
vector<string> selected_fields;
pyobject_as_vector_string(field_list, selected_fields);
// TODO: Move this implementation into dynd
ndt::type fdt = n.get_dtype();
if (fdt.get_kind() != struct_kind) {
stringstream ss;
ss << "nd.fields must be given a dynd array of 'struct' kind, not ";
ss << fdt;
throw runtime_error(ss.str());
}
const base_struct_type *bsd = static_cast<const base_struct_type *>(fdt.extended());
const ndt::type *field_types = bsd->get_field_types();
if (selected_fields.empty()) {
throw runtime_error("nd.fields requires at least one field name to be specified");
}
// Construct the field mapping and output field types
vector<intptr_t> selected_index(selected_fields.size());
vector<ndt::type> selected_ndt_types(selected_fields.size());
for (size_t i = 0; i != selected_fields.size(); ++i) {
selected_index[i] = bsd->get_field_index(selected_fields[i]);
if (selected_index[i] < 0) {
stringstream ss;
ss << "field name ";
print_escaped_utf8_string(ss, selected_fields[i]);
ss << " does not exist in dynd type " << fdt;
throw runtime_error(ss.str());
}
selected_ndt_types[i] = field_types[selected_index[i]];
}
// Create the result udt
ndt::type rudt = ndt::make_struct(selected_ndt_types, selected_fields);
ndt::type result_tp = n.get_type().with_replaced_dtype(rudt);
const base_struct_type *rudt_bsd = static_cast<const base_struct_type *>(rudt.extended());
// Allocate the new memory block.
size_t metadata_size = result_tp.get_metadata_size();
nd::array result(make_array_memory_block(metadata_size));
// Clone the data pointer
result.get_ndo()->m_data_pointer = n.get_ndo()->m_data_pointer;
result.get_ndo()->m_data_reference = n.get_ndo()->m_data_reference;
if (result.get_ndo()->m_data_reference == NULL) {
result.get_ndo()->m_data_reference = n.get_memblock().get();
}
memory_block_incref(result.get_ndo()->m_data_reference);
// Copy the flags
result.get_ndo()->m_flags = n.get_ndo()->m_flags;
// Set the type and transform the metadata
result.get_ndo()->m_type = ndt::type(result_tp).release();
// First copy all the array data type metadata
ndt::type tmp_dt = result_tp;
char *dst_metadata = result.get_ndo_meta();
const char *src_metadata = n.get_ndo_meta();
while (tmp_dt.get_ndim() > 0) {
if (tmp_dt.get_kind() != uniform_dim_kind) {
throw runtime_error("nd.fields doesn't support dimensions with pointers yet");
}
const base_uniform_dim_type *budd = static_cast<const base_uniform_dim_type *>(
tmp_dt.extended());
size_t offset = budd->metadata_copy_construct_onedim(dst_metadata, src_metadata,
n.get_memblock().get());
dst_metadata += offset;
src_metadata += offset;
tmp_dt = budd->get_element_type();
}
// Then create the metadata for the new struct
const size_t *metadata_offsets = bsd->get_metadata_offsets();
const size_t *result_metadata_offsets = rudt_bsd->get_metadata_offsets();
const size_t *data_offsets = bsd->get_data_offsets(src_metadata);
size_t *result_data_offsets = reinterpret_cast<size_t *>(dst_metadata);
for (size_t i = 0; i != selected_fields.size(); ++i) {
const ndt::type& dt = selected_ndt_types[i];
// Copy the data offset
result_data_offsets[i] = data_offsets[selected_index[i]];
// Copy the metadata for this field
if (dt.get_metadata_size() > 0) {
dt.extended()->metadata_copy_construct(dst_metadata + result_metadata_offsets[i],
src_metadata + metadata_offsets[selected_index[i]],
n.get_memblock().get());
}
}
return result;
}
