# Licensed under a 3-clause BSD style license - see LICENSE.rst
"""
This package contains functions for reading and writing HDF5 tables that are
not meant to be used directly, but instead are available as readers/writers in
`astropy.table`. See :ref:`astropy:table_io` for more details.
"""
import os
import warnings
import numpy as np
# NOTE: Do not import anything from astropy.table here.
# https://github.com/astropy/astropy/issues/6604
from astropy.utils.exceptions import AstropyUserWarning
from astropy.utils.misc import NOT_OVERWRITING_MSG
HDF5_SIGNATURE = b'\x89HDF\r\n\x1a\n'
META_KEY = '__table_column_meta__'
__all__ = ['read_table_hdf5', 'write_table_hdf5']
def meta_path(path):
return path + '.' + META_KEY
def _find_all_structured_arrays(handle):
"""
Find all structured arrays in an HDF5 file
"""
import h5py
structured_arrays = []
def append_structured_arrays(name, obj):
if isinstance(obj, h5py.Dataset) and obj.dtype.kind == 'V':
structured_arrays.append(name)
handle.visititems(append_structured_arrays)
return structured_arrays
def is_hdf5(origin, filepath, fileobj, *args, **kwargs):
if fileobj is not None:
loc = fileobj.tell()
try:
signature = fileobj.read(8)
finally:
fileobj.seek(loc)
return signature == HDF5_SIGNATURE
elif filepath is not None:
return filepath.endswith(('.hdf5', '.h5'))
try:
import h5py
except ImportError:
return False
else:
return isinstance(args[0], (h5py.File, h5py.Group, h5py.Dataset))
def read_table_hdf5(input, path=None, character_as_bytes=True):
"""
Read a Table object from an HDF5 file
This requires `h5py `_ to be installed. If more than one
table is present in the HDF5 file or group, the first table is read in and
a warning is displayed.
Parameters
----------
input : str or :class:`h5py.File` or :class:`h5py.Group` or
:class:`h5py.Dataset` If a string, the filename to read the table from.
If an h5py object, either the file or the group object to read the
table from.
path : str
The path from which to read the table inside the HDF5 file.
This should be relative to the input file or group.
character_as_bytes: bool
If `True` then Table columns are left as bytes.
If `False` then Table columns are converted to unicode.
"""
try:
import h5py
except ImportError:
raise Exception("h5py is required to read and write HDF5 files")
# This function is iterative, and only gets to writing the file when
# the input is an hdf5 Group. Moreover, the input variable is changed in
# place.
# Here, we save its value to be used at the end when the conditions are
# right.
input_save = input
if isinstance(input, (h5py.File, h5py.Group)):
# If a path was specified, follow the path
if path is not None:
try:
input = input[path]
except (KeyError, ValueError):
raise OSError(f"Path {path} does not exist")
# `input` is now either a group or a dataset. If it is a group, we
# will search for all structured arrays inside the group, and if there
# is one we can proceed otherwise an error is raised. If it is a
# dataset, we just proceed with the reading.
if isinstance(input, h5py.Group):
# Find all structured arrays in group
arrays = _find_all_structured_arrays(input)
if len(arrays) == 0:
raise ValueError(f"no table found in HDF5 group {path}")
elif len(arrays) > 0:
path = arrays[0] if path is None else path + '/' + arrays[0]
if len(arrays) > 1:
warnings.warn("path= was not specified but multiple tables"
" are present, reading in first available"
" table (path={})".format(path),
AstropyUserWarning)
return read_table_hdf5(input, path=path)
elif not isinstance(input, h5py.Dataset):
# If a file object was passed, then we need to extract the filename
# because h5py cannot properly read in file objects.
if hasattr(input, 'read'):
try:
input = input.name
except AttributeError:
raise TypeError("h5py can only open regular files")
# Open the file for reading, and recursively call read_table_hdf5 with
# the file object and the path.
f = h5py.File(input, 'r')
try:
return read_table_hdf5(f, path=path, character_as_bytes=character_as_bytes)
finally:
f.close()
# If we are here, `input` should be a Dataset object, which we can now
# convert to a Table.
# Create a Table object
from astropy.table import Table, meta, serialize
table = Table(np.array(input))
# Read the meta-data from the file. For back-compatibility, we can read
# the old file format where the serialized metadata were saved in the
# attributes of the HDF5 dataset.
# In the new format, instead, metadata are stored in a new dataset in the
# same file. This is introduced in Astropy 3.0
old_version_meta = META_KEY in input.attrs
new_version_meta = path is not None and meta_path(path) in input_save
if old_version_meta or new_version_meta:
if new_version_meta:
header = meta.get_header_from_yaml(
h.decode('utf-8') for h in input_save[meta_path(path)])
else:
# Must be old_version_meta is True. if (A or B) and not A then B is True
header = meta.get_header_from_yaml(
h.decode('utf-8') for h in input.attrs[META_KEY])
if 'meta' in list(header.keys()):
table.meta = header['meta']
header_cols = dict((x['name'], x) for x in header['datatype'])
for col in table.columns.values():
for attr in ('description', 'format', 'unit', 'meta'):
if attr in header_cols[col.name]:
setattr(col, attr, header_cols[col.name][attr])
# Construct new table with mixins, using tbl.meta['__serialized_columns__']
# as guidance.
table = serialize._construct_mixins_from_columns(table)
else:
# Read the meta-data from the file
table.meta.update(input.attrs)
if not character_as_bytes:
table.convert_bytestring_to_unicode()
return table
def _encode_mixins(tbl):
"""Encode a Table ``tbl`` that may have mixin columns to a Table with only
astropy Columns + appropriate meta-data to allow subsequent decoding.
"""
from astropy.table import serialize
from astropy import units as u
from astropy.utils.data_info import serialize_context_as
# Convert the table to one with no mixins, only Column objects. This adds
# meta data which is extracted with meta.get_yaml_from_table.
with serialize_context_as('hdf5'):
encode_tbl = serialize.represent_mixins_as_columns(tbl)
return encode_tbl
def write_table_hdf5(table, output, path=None, compression=False,
append=False, overwrite=False, serialize_meta=False,
**create_dataset_kwargs):
"""
Write a Table object to an HDF5 file
This requires `h5py `_ to be installed.
Parameters
----------
table : `~astropy.table.Table`
Data table that is to be written to file.
output : str or :class:`h5py.File` or :class:`h5py.Group`
If a string, the filename to write the table to. If an h5py object,
either the file or the group object to write the table to.
path : str
The path to which to write the table inside the HDF5 file.
This should be relative to the input file or group.
If not specified, defaults to ``__astropy_table__``.
compression : bool or str or int
Whether to compress the table inside the HDF5 file. If set to `True`,
``'gzip'`` compression is used. If a string is specified, it should be
one of ``'gzip'``, ``'szip'``, or ``'lzf'``. If an integer is
specified (in the range 0-9), ``'gzip'`` compression is used, and the
integer denotes the compression level.
append : bool
Whether to append the table to an existing HDF5 file.
overwrite : bool
Whether to overwrite any existing file without warning.
If ``append=True`` and ``overwrite=True`` then only the dataset will be
replaced; the file/group will not be overwritten.
**create_dataset_kwargs
Additional keyword arguments are passed to
``h5py.File.create_dataset()`` or ``h5py.Group.create_dataset()``.
"""
from astropy.table import meta
try:
import h5py
except ImportError:
raise Exception("h5py is required to read and write HDF5 files")
if path is None:
# table is just an arbitrary, hardcoded string here.
path = '__astropy_table__'
elif path.endswith('/'):
raise ValueError("table path should end with table name, not /")
if '/' in path:
group, name = path.rsplit('/', 1)
else:
group, name = None, path
if isinstance(output, (h5py.File, h5py.Group)):
if len(list(output.keys())) > 0 and name == '__astropy_table__':
raise ValueError("table path should always be set via the "
"path= argument when writing to existing "
"files")
elif name == '__astropy_table__':
warnings.warn("table path was not set via the path= argument; "
"using default path {}".format(path))
if group:
try:
output_group = output[group]
except (KeyError, ValueError):
output_group = output.create_group(group)
else:
output_group = output
elif isinstance(output, str):
if os.path.exists(output) and not append:
if overwrite and not append:
os.remove(output)
else:
raise OSError(NOT_OVERWRITING_MSG.format(output))
# Open the file for appending or writing
f = h5py.File(output, 'a' if append else 'w')
# Recursively call the write function
try:
return write_table_hdf5(table, f, path=path,
compression=compression, append=append,
overwrite=overwrite,
serialize_meta=serialize_meta)
finally:
f.close()
else:
raise TypeError('output should be a string or an h5py File or '
'Group object')
# Check whether table already exists
if name in output_group:
if append and overwrite:
# Delete only the dataset itself
del output_group[name]
if serialize_meta and name + '.__table_column_meta__' in output_group:
del output_group[name + '.__table_column_meta__']
else:
raise OSError(f"Table {path} already exists")
# Encode any mixin columns as plain columns + appropriate metadata
table = _encode_mixins(table)
# Table with numpy unicode strings can't be written in HDF5 so
# to write such a table a copy of table is made containing columns as
# bytestrings. Now this copy of the table can be written in HDF5.
if any(col.info.dtype.kind == 'U' for col in table.itercols()):
table = table.copy(copy_data=False)
table.convert_unicode_to_bytestring()
# Warn if information will be lost when serialize_meta=False. This is
# hardcoded to the set difference between column info attributes and what
# HDF5 can store natively (name, dtype) with no meta.
if serialize_meta is False:
for col in table.itercols():
for attr in ('unit', 'format', 'description', 'meta'):
if getattr(col.info, attr, None) not in (None, {}):
warnings.warn("table contains column(s) with defined 'unit', 'format',"
" 'description', or 'meta' info attributes. These will"
" be dropped since serialize_meta=False.",
AstropyUserWarning)
# Write the table to the file
if compression:
if compression is True:
compression = 'gzip'
dset = output_group.create_dataset(name, data=table.as_array(),
compression=compression,
**create_dataset_kwargs)
else:
dset = output_group.create_dataset(name, data=table.as_array(),
**create_dataset_kwargs)
if serialize_meta:
header_yaml = meta.get_yaml_from_table(table)
header_encoded = np.array([h.encode('utf-8') for h in header_yaml])
output_group.create_dataset(meta_path(name),
data=header_encoded)
else:
# Write the Table meta dict key:value pairs to the file as HDF5
# attributes. This works only for a limited set of scalar data types
# like numbers, strings, etc., but not any complex types. This path
# also ignores column meta like unit or format.
for key in table.meta:
val = table.meta[key]
try:
dset.attrs[key] = val
except TypeError:
warnings.warn("Attribute `{}` of type {} cannot be written to "
"HDF5 files - skipping. (Consider specifying "
"serialize_meta=True to write all meta data)".format(key, type(val)),
AstropyUserWarning)
def register_hdf5():
"""
Register HDF5 with Unified I/O.
"""
from astropy.io import registry as io_registry
from astropy.table import Table
io_registry.register_reader('hdf5', Table, read_table_hdf5)
io_registry.register_writer('hdf5', Table, write_table_hdf5)
io_registry.register_identifier('hdf5', Table, is_hdf5)