# pylint: disable-msg=E1101,W0613,W0603
High level interface to PyTables for reading and writing pandas data structures
to disk

import copy
from datetime import date, datetime
from distutils.version import LooseVersion
import itertools
import os
import re
import time
import warnings

import numpy as np

from pandas._libs import algos, lib, writers as libwriters
from pandas._libs.tslibs import timezones
from pandas.compat import PY3, filter, lrange, range, string_types
from pandas.errors import PerformanceWarning

from pandas.core.dtypes.common import (
    ensure_int64, ensure_object, ensure_platform_int, is_categorical_dtype,
    is_datetime64_dtype, is_datetime64tz_dtype, is_list_like,
from pandas.core.dtypes.missing import array_equivalent

from pandas import (
    DataFrame, DatetimeIndex, Index, Int64Index, MultiIndex, Panel,
    PeriodIndex, Series, SparseDataFrame, SparseSeries, TimedeltaIndex, compat,
    concat, isna, to_datetime)
from pandas.core import config
from pandas.core.algorithms import match, unique
from pandas.core.arrays.categorical import (
    Categorical, _factorize_from_iterables)
from pandas.core.arrays.sparse import BlockIndex, IntIndex
from pandas.core.base import StringMixin
import pandas.core.common as com
from pandas.core.computation.pytables import Expr, maybe_expression
from pandas.core.config import get_option
from pandas.core.index import ensure_index
from pandas.core.internals import (
    BlockManager, _block2d_to_blocknd, _block_shape, _factor_indexer,

from pandas.io.common import _stringify_path
from pandas.io.formats.printing import adjoin, pprint_thing

# versioning attribute
_version = '0.15.2'

# encoding
# PY3 encoding if we don't specify
_default_encoding = 'UTF-8'

def _ensure_decoded(s):
    """ if we have bytes, decode them to unicode """
    if isinstance(s, np.bytes_):
        s = s.decode('UTF-8')
    return s

def _ensure_encoding(encoding):
    # set the encoding if we need
    if encoding is None:
        if PY3:
            encoding = _default_encoding
    return encoding

def _ensure_str(name):
    """Ensure that an index / column name is a str (python 3) or
    unicode (python 2); otherwise they may be np.string dtype.
    Non-string dtypes are passed through unchanged.

    if isinstance(name, compat.string_types):
        name = compat.text_type(name)
    return name

Term = Expr

def _ensure_term(where, scope_level):
    ensure that the where is a Term or a list of Term
    this makes sure that we are capturing the scope of variables
    that are passed
    create the terms here with a frame_level=2 (we are 2 levels down)

    # only consider list/tuple here as an ndarray is automatically a coordinate
    # list
    level = scope_level + 1
    if isinstance(where, (list, tuple)):
        wlist = []
        for w in filter(lambda x: x is not None, where):
            if not maybe_expression(w):
                wlist.append(Term(w, scope_level=level))
        where = wlist
    elif maybe_expression(where):
        where = Term(where, scope_level=level)
    return where

class PossibleDataLossError(Exception):

class ClosedFileError(Exception):

class IncompatibilityWarning(Warning):

incompatibility_doc = """
where criteria is being ignored as this version [%s] is too old (or
not-defined), read the file in and write it out to a new file to upgrade (with
the copy_to method)

class AttributeConflictWarning(Warning):

attribute_conflict_doc = """
the [%s] attribute of the existing index is [%s] which conflicts with the new
[%s], resetting the attribute to None

class DuplicateWarning(Warning):

duplicate_doc = """
duplicate entries in table, taking most recently appended

performance_doc = """
your performance may suffer as PyTables will pickle object types that it cannot
map directly to c-types [inferred_type->%s,key->%s] [items->%s]

# formats
    u'f': 'fixed',
    u'fixed': 'fixed',
    u't': 'table',
    u'table': 'table',

format_deprecate_doc = """
the table keyword has been deprecated
use the format='fixed(f)|table(t)' keyword instead
  fixed(f) : specifies the Fixed format
             and is the default for put operations
  table(t) : specifies the Table format
             and is the default for append operations

# map object types

    Series: u'series',
    SparseSeries: u'sparse_series',
    DataFrame: u'frame',
    SparseDataFrame: u'sparse_frame',
    Panel: u'wide',

# storer class map
    u'Series': 'LegacySeriesFixed',
    u'DataFrame': 'LegacyFrameFixed',
    u'DataMatrix': 'LegacyFrameFixed',
    u'series': 'SeriesFixed',
    u'sparse_series': 'SparseSeriesFixed',
    u'frame': 'FrameFixed',
    u'sparse_frame': 'SparseFrameFixed',
    u'wide': 'PanelFixed',

# table class map
    u'generic_table': 'GenericTable',
    u'appendable_series': 'AppendableSeriesTable',
    u'appendable_multiseries': 'AppendableMultiSeriesTable',
    u'appendable_frame': 'AppendableFrameTable',
    u'appendable_multiframe': 'AppendableMultiFrameTable',
    u'appendable_panel': 'AppendablePanelTable',
    u'worm': 'WORMTable',
    u'legacy_frame': 'LegacyFrameTable',
    u'legacy_panel': 'LegacyPanelTable',

# axes map
    DataFrame: [0],
    Panel: [1, 2]

# register our configuration options
dropna_doc = """
: boolean
    drop ALL nan rows when appending to a table
format_doc = """
: format
    default format writing format, if None, then
    put will default to 'fixed' and append will default to 'table'

with config.config_prefix('io.hdf'):
    config.register_option('dropna_table', False, dropna_doc,
        'default_format', None, format_doc,
        validator=config.is_one_of_factory(['fixed', 'table', None])

# oh the troubles to reduce import time
_table_mod = None
_table_file_open_policy_is_strict = False

def _tables():
    global _table_mod
    global _table_file_open_policy_is_strict
    if _table_mod is None:
        import tables
        _table_mod = tables

        # version requirements
        if LooseVersion(tables.__version__) < LooseVersion('3.0.0'):
            raise ImportError("PyTables version >= 3.0.0 is required")

        # set the file open policy
        # return the file open policy; this changes as of pytables 3.1
        # depending on the HDF5 version
            _table_file_open_policy_is_strict = (
                tables.file._FILE_OPEN_POLICY == 'strict')
        except AttributeError:

    return _table_mod

# interface to/from ###

def to_hdf(path_or_buf, key, value, mode=None, complevel=None, complib=None,
           append=None, **kwargs):
    """ store this object, close it if we opened it """

    if append:
        f = lambda store: store.append(key, value, **kwargs)
        f = lambda store: store.put(key, value, **kwargs)

    path_or_buf = _stringify_path(path_or_buf)
    if isinstance(path_or_buf, string_types):
        with HDFStore(path_or_buf, mode=mode, complevel=complevel,
                      complib=complib) as store:

def read_hdf(path_or_buf, key=None, mode='r', **kwargs):
    Read from the store, close it if we opened it.

    Retrieve pandas object stored in file, optionally based on where

    path_or_buf : string, buffer or path object
        Path to the file to open, or an open :class:`pandas.HDFStore` object.
        Supports any object implementing the ``__fspath__`` protocol.
        This includes :class:`pathlib.Path` and py._path.local.LocalPath

        .. versionadded:: 0.19.0 support for pathlib, py.path.
        .. versionadded:: 0.21.0 support for __fspath__ protocol.

    key : object, optional
        The group identifier in the store. Can be omitted if the HDF file
        contains a single pandas object.
    mode : {'r', 'r+', 'a'}, optional
        Mode to use when opening the file. Ignored if path_or_buf is a
        :class:`pandas.HDFStore`. Default is 'r'.
    where : list, optional
        A list of Term (or convertible) objects.
    start : int, optional
        Row number to start selection.
    stop  : int, optional
        Row number to stop selection.
    columns : list, optional
        A list of columns names to return.
    iterator : bool, optional
        Return an iterator object.
    chunksize : int, optional
        Number of rows to include in an iteration when using an iterator.
    errors : str, default 'strict'
        Specifies how encoding and decoding errors are to be handled.
        See the errors argument for :func:`open` for a full list
        of options.
        Additional keyword arguments passed to HDFStore.

    item : object
        The selected object. Return type depends on the object stored.

    See Also
    pandas.DataFrame.to_hdf : Write a HDF file from a DataFrame.
    pandas.HDFStore : Low-level access to HDF files.

    >>> df = pd.DataFrame([[1, 1.0, 'a']], columns=['x', 'y', 'z'])
    >>> df.to_hdf('./store.h5', 'data')
    >>> reread = pd.read_hdf('./store.h5')

    if mode not in ['r', 'r+', 'a']:
        raise ValueError('mode {0} is not allowed while performing a read. '
                         'Allowed modes are r, r+ and a.'.format(mode))
    # grab the scope
    if 'where' in kwargs:
        kwargs['where'] = _ensure_term(kwargs['where'], scope_level=1)

    if isinstance(path_or_buf, HDFStore):
        if not path_or_buf.is_open:
            raise IOError('The HDFStore must be open for reading.')

        store = path_or_buf
        auto_close = False
        path_or_buf = _stringify_path(path_or_buf)
        if not isinstance(path_or_buf, string_types):
            raise NotImplementedError('Support for generic buffers has not '
                                      'been implemented.')
            exists = os.path.exists(path_or_buf)

        # if filepath is too long
        except (TypeError, ValueError):
            exists = False

        if not exists:
            raise compat.FileNotFoundError(
                'File {path} does not exist'.format(path=path_or_buf))

        store = HDFStore(path_or_buf, mode=mode, **kwargs)
        # can't auto open/close if we are using an iterator
        # so delegate to the iterator
        auto_close = True

        if key is None:
            groups = store.groups()
            if len(groups) == 0:
                raise ValueError('No dataset in HDF5 file.')
            candidate_only_group = groups[0]

            # For the HDF file to have only one dataset, all other groups
            # should then be metadata groups for that candidate group. (This
            # assumes that the groups() method enumerates parent groups
            # before their children.)
            for group_to_check in groups[1:]:
                if not _is_metadata_of(group_to_check, candidate_only_group):
                    raise ValueError('key must be provided when HDF5 file '
                                     'contains multiple datasets.')
            key = candidate_only_group._v_pathname
        return store.select(key, auto_close=auto_close, **kwargs)
    except (ValueError, TypeError):
        # if there is an error, close the store
        except AttributeError:


def _is_metadata_of(group, parent_group):
    """Check if a given group is a metadata group for a given parent_group."""
    if group._v_depth <= parent_group._v_depth:
        return False

    current = group
    while current._v_depth > 1:
        parent = current._v_parent
        if parent == parent_group and current._v_name == 'meta':
            return True
        current = current._v_parent
    return False

class HDFStore(StringMixin):

    Dict-like IO interface for storing pandas objects in PyTables
    either Fixed or Table format.

    path : string
        File path to HDF5 file
    mode : {'a', 'w', 'r', 'r+'}, default 'a'

            Read-only; no data can be modified.
            Write; a new file is created (an existing file with the same
            name would be deleted).
            Append; an existing file is opened for reading and writing,
            and if the file does not exist it is created.
            It is similar to ``'a'``, but the file must already exist.
    complevel : int, 0-9, default None
            Specifies a compression level for data.
            A value of 0 disables compression.
    complib : {'zlib', 'lzo', 'bzip2', 'blosc'}, default 'zlib'
            Specifies the compression library to be used.
            As of v0.20.2 these additional compressors for Blosc are supported
            (default if no compressor specified: 'blosc:blosclz'):
            {'blosc:blosclz', 'blosc:lz4', 'blosc:lz4hc', 'blosc:snappy',
             'blosc:zlib', 'blosc:zstd'}.
            Specifying a compression library which is not available issues
            a ValueError.
    fletcher32 : bool, default False
            If applying compression use the fletcher32 checksum

    >>> bar = pd.DataFrame(np.random.randn(10, 4))
    >>> store = pd.HDFStore('test.h5')
    >>> store['foo'] = bar   # write to HDF5
    >>> bar = store['foo']   # retrieve
    >>> store.close()

    def __init__(self, path, mode=None, complevel=None, complib=None,
                 fletcher32=False, **kwargs):

        if 'format' in kwargs:
            raise ValueError('format is not a defined argument for HDFStore')

            import tables  # noqa
        except ImportError as ex:  # pragma: no cover
            raise ImportError('HDFStore requires PyTables, "{ex!s}" problem '

        if complib is not None and complib not in tables.filters.all_complibs:
            raise ValueError(
                "complib only supports {libs} compression.".format(

        if complib is None and complevel is not None:
            complib = tables.filters.default_complib

        self._path = _stringify_path(path)
        if mode is None:
            mode = 'a'
        self._mode = mode
        self._handle = None
        self._complevel = complevel if complevel else 0
        self._complib = complib
        self._fletcher32 = fletcher32
        self._filters = None
        self.open(mode=mode, **kwargs)

    def __fspath__(self):
        return self._path

    def root(self):
        """ return the root node """
        return self._handle.root

    def filename(self):
        return self._path

    def __getitem__(self, key):
        return self.get(key)

    def __setitem__(self, key, value):
        self.put(key, value)

    def __delitem__(self, key):
        return self.remove(key)

    def __getattr__(self, name):
        """ allow attribute access to get stores """
            return self.get(name)
        except (KeyError, ClosedFileError):
        raise AttributeError(
            "'{object}' object has no attribute '{name}'".format(
                object=type(self).__name__, name=name))

    def __contains__(self, key):
        """ check for existence of this key
              can match the exact pathname or the pathnm w/o the leading '/'
        node = self.get_node(key)
        if node is not None:
            name = node._v_pathname
            if name == key or name[1:] == key:
                return True
        return False

    def __len__(self):
        return len(self.groups())

    def __unicode__(self):
        return '{type}\nFile path: {path}\n'.format(
            type=type(self), path=pprint_thing(self._path))

    def __enter__(self):
        return self

    def __exit__(self, exc_type, exc_value, traceback):

    def keys(self):
        Return a (potentially unordered) list of the keys corresponding to the
        objects stored in the HDFStore. These are ABSOLUTE path-names (e.g.
        have the leading '/'
        return [n._v_pathname for n in self.groups()]

    def __iter__(self):
        return iter(self.keys())

    def items(self):
        iterate on key->group
        for g in self.groups():
            yield g._v_pathname, g

    iteritems = items

    def open(self, mode='a', **kwargs):
        Open the file in the specified mode

        mode : {'a', 'w', 'r', 'r+'}, default 'a'
            See HDFStore docstring or tables.open_file for info about modes
        tables = _tables()

        if self._mode != mode:

            # if we are changing a write mode to read, ok
            if self._mode in ['a', 'w'] and mode in ['r', 'r+']:
            elif mode in ['w']:

                # this would truncate, raise here
                if self.is_open:
                    raise PossibleDataLossError(
                        "Re-opening the file [{0}] with mode [{1}] "
                        "will delete the current file!"
                        .format(self._path, self._mode)

            self._mode = mode

        # close and reopen the handle
        if self.is_open:

        if self._complevel and self._complevel > 0:
            self._filters = _tables().Filters(self._complevel, self._complib,

            self._handle = tables.open_file(self._path, self._mode, **kwargs)
        except (IOError) as e:  # pragma: no cover
            if 'can not be written' in str(e):
                    'Opening {path} in read-only mode'.format(path=self._path))
                self._handle = tables.open_file(self._path, 'r', **kwargs)

        except (ValueError) as e:

            # trap PyTables >= 3.1 FILE_OPEN_POLICY exception
            # to provide an updated message
            if 'FILE_OPEN_POLICY' in str(e):
                e = ValueError(
                    "PyTables [{version}] no longer supports opening multiple "
                    "even in read-only mode on this HDF5 version "
                    "[{hdf_version}]. You can accept this\n"
                    "and not open the same file multiple times at once,\n"
                    "upgrade the HDF5 version, or downgrade to PyTables 3.0.0 "
                    "which allows\n"
                    "files to be opened multiple times at once\n"

            raise e

        except (Exception) as e:

            # trying to read from a non-existent file causes an error which
            # is not part of IOError, make it one
            if self._mode == 'r' and 'Unable to open/create file' in str(e):
                raise IOError(str(e))

    def close(self):
        Close the PyTables file handle
        if self._handle is not None:
        self._handle = None

    def is_open(self):
        return a boolean indicating whether the file is open
        if self._handle is None:
            return False
        return bool(self._handle.isopen)

    def flush(self, fsync=False):
        Force all buffered modifications to be written to disk.

        fsync : bool (default False)
          call ``os.fsync()`` on the file handle to force writing to disk.

        Without ``fsync=True``, flushing may not guarantee that the OS writes
        to disk. With fsync, the operation will block until the OS claims the
        file has been written; however, other caching layers may still
        if self._handle is not None:
            if fsync:
                except OSError:

    def get(self, key):
        Retrieve pandas object stored in file

        key : object

        obj : same type as object stored in file
        group = self.get_node(key)
        if group is None:
            raise KeyError('No object named {key} in the file'.format(key=key))
        return self._read_group(group)

    def select(self, key, where=None, start=None, stop=None, columns=None,
               iterator=False, chunksize=None, auto_close=False, **kwargs):
        Retrieve pandas object stored in file, optionally based on where

        key : object
        where : list of Term (or convertible) objects, optional
        start : integer (defaults to None), row number to start selection
        stop  : integer (defaults to None), row number to stop selection
        columns : a list of columns that if not None, will limit the return
        iterator : boolean, return an iterator, default False
        chunksize : nrows to include in iteration, return an iterator
        auto_close : boolean, should automatically close the store when
            finished, default is False

        The selected object
        group = self.get_node(key)
        if group is None:
            raise KeyError('No object named {key} in the file'.format(key=key))

        # create the storer and axes
        where = _ensure_term(where, scope_level=1)
        s = self._create_storer(group)

        # function to call on iteration
        def func(_start, _stop, _where):
            return s.read(start=_start, stop=_stop,

        # create the iterator
        it = TableIterator(self, s, func, where=where, nrows=s.nrows,
                           start=start, stop=stop, iterator=iterator,
                           chunksize=chunksize, auto_close=auto_close)

        return it.get_result()

    def select_as_coordinates(
            self, key, where=None, start=None, stop=None, **kwargs):
        return the selection as an Index

        key : object
        where : list of Term (or convertible) objects, optional
        start : integer (defaults to None), row number to start selection
        stop  : integer (defaults to None), row number to stop selection
        where = _ensure_term(where, scope_level=1)
        return self.get_storer(key).read_coordinates(where=where, start=start,
                                                     stop=stop, **kwargs)

    def select_column(self, key, column, **kwargs):
        return a single column from the table. This is generally only useful to
        select an indexable

        key : object
        column: the column of interest

        raises KeyError if the column is not found (or key is not a valid
        raises ValueError if the column can not be extracted individually (it
            is part of a data block)

        return self.get_storer(key).read_column(column=column, **kwargs)

    def select_as_multiple(self, keys, where=None, selector=None, columns=None,
                           start=None, stop=None, iterator=False,
                           chunksize=None, auto_close=False, **kwargs):
        """ Retrieve pandas objects from multiple tables

        keys : a list of the tables
        selector : the table to apply the where criteria (defaults to keys[0]
            if not supplied)
        columns : the columns I want back
        start : integer (defaults to None), row number to start selection
        stop  : integer (defaults to None), row number to stop selection
        iterator : boolean, return an iterator, default False
        chunksize : nrows to include in iteration, return an iterator

        raises KeyError if keys or selector is not found or keys is empty
        raises TypeError if keys is not a list or tuple
        raises ValueError if the tables are not ALL THE SAME DIMENSIONS

        # default to single select
        where = _ensure_term(where, scope_level=1)
        if isinstance(keys, (list, tuple)) and len(keys) == 1:
            keys = keys[0]
        if isinstance(keys, string_types):
            return self.select(key=keys, where=where, columns=columns,
                               start=start, stop=stop, iterator=iterator,
                               chunksize=chunksize, **kwargs)

        if not isinstance(keys, (list, tuple)):
            raise TypeError("keys must be a list/tuple")

        if not len(keys):
            raise ValueError("keys must have a non-zero length")

        if selector is None:
            selector = keys[0]

        # collect the tables
        tbls = [self.get_storer(k) for k in keys]
        s = self.get_storer(selector)

        # validate rows
        nrows = None
        for t, k in itertools.chain([(s, selector)], zip(tbls, keys)):
            if t is None:
                raise KeyError("Invalid table [{key}]".format(key=k))
            if not t.is_table:
                raise TypeError(
                    "object [{obj}] is not a table, and cannot be used in all "
                    "select as multiple".format(obj=t.pathname)

            if nrows is None:
                nrows = t.nrows
            elif t.nrows != nrows:
                raise ValueError(
                    "all tables must have exactly the same nrows!")

        # axis is the concentation axes
        axis = list({t.non_index_axes[0][0] for t in tbls})[0]

        def func(_start, _stop, _where):

            # retrieve the objs, _where is always passed as a set of
            # coordinates here
            objs = [t.read(where=_where, columns=columns, start=_start,
                           stop=_stop, **kwargs) for t in tbls]

            # concat and return
            return concat(objs, axis=axis,

        # create the iterator
        it = TableIterator(self, s, func, where=where, nrows=nrows,
                           start=start, stop=stop, iterator=iterator,
                           chunksize=chunksize, auto_close=auto_close)

        return it.get_result(coordinates=True)

    def put(self, key, value, format=None, append=False, **kwargs):
        Store object in HDFStore

        key      : object
        value    : {Series, DataFrame, Panel}
        format   : 'fixed(f)|table(t)', default is 'fixed'
            fixed(f) : Fixed format
                       Fast writing/reading. Not-appendable, nor searchable
            table(t) : Table format
                       Write as a PyTables Table structure which may perform
                       worse but allow more flexible operations like searching
                       / selecting subsets of the data
        append   : boolean, default False
            This will force Table format, append the input data to the
        data_columns : list of columns to create as data columns, or True to
            use all columns. See
            `here <http://pandas.pydata.org/pandas-docs/stable/io.html#query-via-data-columns>`__ # noqa
        encoding : default None, provide an encoding for strings
        dropna   : boolean, default False, do not write an ALL nan row to
            the store settable by the option 'io.hdf.dropna_table'
        if format is None:
            format = get_option("io.hdf.default_format") or 'fixed'
        kwargs = self._validate_format(format, kwargs)
        self._write_to_group(key, value, append=append, **kwargs)

    def remove(self, key, where=None, start=None, stop=None):
        Remove pandas object partially by specifying the where condition

        key : string
            Node to remove or delete rows from
        where : list of Term (or convertible) objects, optional
        start : integer (defaults to None), row number to start selection
        stop  : integer (defaults to None), row number to stop selection

        number of rows removed (or None if not a Table)

        raises KeyError if key is not a valid store

        where = _ensure_term(where, scope_level=1)
            s = self.get_storer(key)
        except KeyError:
            # the key is not a valid store, re-raising KeyError
        except Exception:

            if where is not None:
                raise ValueError(
                    "trying to remove a node with a non-None where clause!")

            # we are actually trying to remove a node (with children)
            s = self.get_node(key)
            if s is not None:
                return None

        # remove the node
        if com._all_none(where, start, stop):

        # delete from the table
            if not s.is_table:
                raise ValueError(
                    'can only remove with where on objects written as tables')
            return s.delete(where=where, start=start, stop=stop)

    def append(self, key, value, format=None, append=True, columns=None,
               dropna=None, **kwargs):
        Append to Table in file. Node must already exist and be Table

        key : object
        value : {Series, DataFrame, Panel}
        format : 'table' is the default
            table(t) : table format
                       Write as a PyTables Table structure which may perform
                       worse but allow more flexible operations like searching
                       / selecting subsets of the data
        append       : boolean, default True, append the input data to the
        data_columns :  list of columns, or True, default None
            List of columns to create as indexed data columns for on-disk
            queries, or True to use all columns. By default only the axes
            of the object are indexed. See `here
        min_itemsize : dict of columns that specify minimum string sizes
        nan_rep      : string to use as string nan represenation
        chunksize    : size to chunk the writing
        expectedrows : expected TOTAL row size of this table
        encoding     : default None, provide an encoding for strings
        dropna       : boolean, default False, do not write an ALL nan row to
            the store settable by the option 'io.hdf.dropna_table'

        Does *not* check if data being appended overlaps with existing
        data in the table, so be careful
        if columns is not None:
            raise TypeError("columns is not a supported keyword in append, "
                            "try data_columns")

        if dropna is None:
            dropna = get_option("io.hdf.dropna_table")
        if format is None:
            format = get_option("io.hdf.default_format") or 'table'
        kwargs = self._validate_format(format, kwargs)
        self._write_to_group(key, value, append=append, dropna=dropna,

    def append_to_multiple(self, d, value, selector, data_columns=None,
                           axes=None, dropna=False, **kwargs):
        Append to multiple tables

        d : a dict of table_name to table_columns, None is acceptable as the
            values of one node (this will get all the remaining columns)
        value : a pandas object
        selector : a string that designates the indexable table; all of its
            columns will be designed as data_columns, unless data_columns is
            passed, in which case these are used
        data_columns : list of columns to create as data columns, or True to
            use all columns
        dropna : if evaluates to True, drop rows from all tables if any single
                 row in each table has all NaN. Default False.

        axes parameter is currently not accepted

        if axes is not None:
            raise TypeError("axes is currently not accepted as a parameter to"
                            " append_to_multiple; you can create the "
                            "tables independently instead")

        if not isinstance(d, dict):
            raise ValueError(
                "append_to_multiple must have a dictionary specified as the "
                "way to split the value"

        if selector not in d:
            raise ValueError(
                "append_to_multiple requires a selector that is in passed dict"

        # figure out the splitting axis (the non_index_axis)
        axis = list(set(range(value.ndim)) - set(_AXES_MAP[type(value)]))[0]

        # figure out how to split the value
        remain_key = None
        remain_values = []
        for k, v in d.items():
            if v is None:
                if remain_key is not None:
                    raise ValueError(
                        "append_to_multiple can only have one value in d that "
                        "is None"
                remain_key = k
        if remain_key is not None:
            ordered = value.axes[axis]
            ordd = ordered.difference(Index(remain_values))
            ordd = sorted(ordered.get_indexer(ordd))
            d[remain_key] = ordered.take(ordd)

        # data_columns
        if data_columns is None:
            data_columns = d[selector]

        # ensure rows are synchronized across the tables
        if dropna:
            idxs = (value[cols].dropna(how='all').index for cols in d.values())
            valid_index = next(idxs)
            for index in idxs:
                valid_index = valid_index.intersection(index)
            value = value.loc[valid_index]

        # append
        for k, v in d.items():
            dc = data_columns if k == selector else None

            # compute the val
            val = value.reindex(v, axis=axis)

            self.append(k, val, data_columns=dc, **kwargs)

    def create_table_index(self, key, **kwargs):
        """ Create a pytables index on the table
        key : object (the node to index)

        raises if the node is not a table


        # version requirements
        s = self.get_storer(key)
        if s is None:

        if not s.is_table:
            raise TypeError(
                "cannot create table index on a Fixed format store")

    def groups(self):
        """return a list of all the top-level nodes (that are not themselves a
        pandas storage object)
        return [
            g for g in self._handle.walk_groups()
            if (not isinstance(g, _table_mod.link.Link) and
                (getattr(g._v_attrs, 'pandas_type', None) or
                 getattr(g, 'table', None) or
                (isinstance(g, _table_mod.table.Table) and
                 g._v_name != u'table')))

    def walk(self, where="/"):
        """ Walk the pytables group hierarchy for pandas objects

        This generator will yield the group path, subgroups and pandas object
        names for each group.
        Any non-pandas PyTables objects that are not a group will be ignored.

        The `where` group itself is listed first (preorder), then each of its
        child groups (following an alphanumerical order) is also traversed,
        following the same procedure.

        .. versionadded:: 0.24.0

        where : str, optional
            Group where to start walking.
            If not supplied, the root group is used.

        path : str
            Full path to a group (without trailing '/')
        groups : list of str
            names of the groups contained in `path`
        leaves : list of str
            names of the pandas objects contained in `path`
        for g in self._handle.walk_groups(where):
            if getattr(g._v_attrs, 'pandas_type', None) is not None:

            groups = []
            leaves = []
            for child in g._v_children.values():
                pandas_type = getattr(child._v_attrs, 'pandas_type', None)
                if pandas_type is None:
                    if isinstance(child, _table_mod.group.Group):

            yield (g._v_pathname.rstrip('/'), groups, leaves)

    def get_node(self, key):
        """ return the node with the key or None if it does not exist """
            if not key.startswith('/'):
                key = '/' + key
            return self._handle.get_node(self.root, key)
        except _table_mod.exceptions.NoSuchNodeError:
            return None

    def get_storer(self, key):
        """ return the storer object for a key, raise if not in the file """
        group = self.get_node(key)
        if group is None:
            raise KeyError('No object named {key} in the file'.format(key=key))

        s = self._create_storer(group)
        return s

    def copy(self, file, mode='w', propindexes=True, keys=None, complib=None,
             complevel=None, fletcher32=False, overwrite=True):
        """ copy the existing store to a new file, upgrading in place

            propindexes: restore indexes in copied file (defaults to True)
            keys       : list of keys to include in the copy (defaults to all)
            overwrite  : overwrite (remove and replace) existing nodes in the
                new store (default is True)
            mode, complib, complevel, fletcher32 same as in HDFStore.__init__

            open file handle of the new store

        new_store = HDFStore(
        if keys is None:
            keys = list(self.keys())
        if not isinstance(keys, (tuple, list)):
            keys = [keys]
        for k in keys:
            s = self.get_storer(k)
            if s is not None:

                if k in new_store:
                    if overwrite:

                data = self.select(k)
                if s.is_table:

                    index = False
                    if propindexes:
                        index = [a.name for a in s.axes if a.is_indexed]
                        k, data, index=index,
                        data_columns=getattr(s, 'data_columns', None),
                    new_store.put(k, data, encoding=s.encoding)

        return new_store

    def info(self):
        Print detailed information on the store.

        .. versionadded:: 0.21.0
        output = '{type}\nFile path: {path}\n'.format(
            type=type(self), path=pprint_thing(self._path))
        if self.is_open:
            lkeys = sorted(list(self.keys()))
            if len(lkeys):
                keys = []
                values = []

                for k in lkeys:
                        s = self.get_storer(k)
                        if s is not None:
                            keys.append(pprint_thing(s.pathname or k))
                                pprint_thing(s or 'invalid_HDFStore node'))
                    except Exception as detail:
                            "[invalid_HDFStore node: {detail}]".format(

                output += adjoin(12, keys, values)
                output += 'Empty'
            output += "File is CLOSED"

        return output

    # private methods ######
    def _check_if_open(self):
        if not self.is_open:
            raise ClosedFileError("{0} file is not open!".format(self._path))

    def _validate_format(self, format, kwargs):
        """ validate / deprecate formats; return the new kwargs """
        kwargs = kwargs.copy()

        # validate
            kwargs['format'] = _FORMAT_MAP[format.lower()]
        except KeyError:
            raise TypeError("invalid HDFStore format specified [{0}]"

        return kwargs

    def _create_storer(self, group, format=None, value=None, append=False,
        """ return a suitable class to operate """

        def error(t):
            raise TypeError(
                "cannot properly create the storer for: [{t}] [group->"
                "kwargs->{kwargs}]".format(t=t, group=group,
                                           value=type(value), format=format,
                                           append=append, kwargs=kwargs))

        pt = _ensure_decoded(getattr(group._v_attrs, 'pandas_type', None))
        tt = _ensure_decoded(getattr(group._v_attrs, 'table_type', None))

        # infer the pt from the passed value
        if pt is None:
            if value is None:

                if (getattr(group, 'table', None) or
                        isinstance(group, _table_mod.table.Table)):
                    pt = u'frame_table'
                    tt = u'generic_table'
                    raise TypeError(
                        "cannot create a storer if the object is not existing "
                        "nor a value are passed")

                    pt = _TYPE_MAP[type(value)]
                except KeyError:

                # we are actually a table
                if format == 'table':
                    pt += u'_table'

        # a storer node
        if u'table' not in pt:
                return globals()[_STORER_MAP[pt]](self, group, **kwargs)
            except KeyError:

        # existing node (and must be a table)
        if tt is None:

            # if we are a writer, determine the tt
            if value is not None:

                if pt == u'series_table':
                    index = getattr(value, 'index', None)
                    if index is not None:
                        if index.nlevels == 1:
                            tt = u'appendable_series'
                        elif index.nlevels > 1:
                            tt = u'appendable_multiseries'
                elif pt == u'frame_table':
                    index = getattr(value, 'index', None)
                    if index is not None:
                        if index.nlevels == 1:
                            tt = u'appendable_frame'
                        elif index.nlevels > 1:
                            tt = u'appendable_multiframe'
                elif pt == u'wide_table':
                    tt = u'appendable_panel'
                elif pt == u'ndim_table':
                    tt = u'appendable_ndim'


                # distiguish between a frame/table
                tt = u'legacy_panel'
                    fields = group.table._v_attrs.fields
                    if len(fields) == 1 and fields[0] == u'value':
                        tt = u'legacy_frame'
                except IndexError:

            return globals()[_TABLE_MAP[tt]](self, group, **kwargs)
        except KeyError:

    def _write_to_group(self, key, value, format, index=True, append=False,
                        complib=None, encoding=None, **kwargs):
        group = self.get_node(key)

        # remove the node if we are not appending
        if group is not None and not append:
            self._handle.remove_node(group, recursive=True)
            group = None

        # we don't want to store a table node at all if are object is 0-len
        # as there are not dtypes
        if getattr(value, 'empty', None) and (format == 'table' or append):

        if group is None:
            paths = key.split('/')

            # recursively create the groups
            path = '/'
            for p in paths:
                if not len(p):
                new_path = path
                if not path.endswith('/'):
                    new_path += '/'
                new_path += p
                group = self.get_node(new_path)
                if group is None:
                    group = self._handle.create_group(path, p)
                path = new_path

        s = self._create_storer(group, format, value, append=append,
                                encoding=encoding, **kwargs)
        if append:
            # raise if we are trying to append to a Fixed format,
            #       or a table that exists (and we are putting)
            if (not s.is_table or
                    (s.is_table and format == 'fixed' and s.is_exists)):
                raise ValueError('Can only append to Tables')
            if not s.is_exists:

        if not s.is_table and complib:
            raise ValueError(
                'Compression not supported on Fixed format stores'

        # write the object
        s.write(obj=value, append=append, complib=complib, **kwargs)

        if s.is_table and index:

    def _read_group(self, group, **kwargs):
        s = self._create_storer(group)
        return s.read(**kwargs)

class TableIterator(object):

    """ define the iteration interface on a table


        store : the reference store
        s     : the referred storer
        func  : the function to execute the query
        where : the where of the query
        nrows : the rows to iterate on
        start : the passed start value (default is None)
        stop  : the passed stop value (default is None)
        iterator : boolean, whether to use the default iterator
        chunksize : the passed chunking value (default is 50000)
        auto_close : boolean, automatically close the store at the end of
            iteration, default is False
        kwargs : the passed kwargs

    def __init__(self, store, s, func, where, nrows, start=None, stop=None,
                 iterator=False, chunksize=None, auto_close=False):
        self.store = store
        self.s = s
        self.func = func
        self.where = where

        # set start/stop if they are not set if we are a table
        if self.s.is_table:
            if nrows is None:
                nrows = 0
            if start is None:
                start = 0
            if stop is None:
                stop = nrows
            stop = min(nrows, stop)

        self.nrows = nrows
        self.start = start
        self.stop = stop

        self.coordinates = None
        if iterator or chunksize is not None:
            if chunksize is None:
                chunksize = 100000
            self.chunksize = int(chunksize)
            self.chunksize = None

        self.auto_close = auto_close

    def __iter__(self):

        # iterate
        current = self.start
        while current < self.stop:

            stop = min(current + self.chunksize, self.stop)
            value = self.func(None, None, self.coordinates[current:stop])
            current = stop
            if value is None or not len(value):

            yield value


    def close(self):
        if self.auto_close:

    def get_result(self, coordinates=False):

        #  return the actual iterator
        if self.chunksize is not None:
            if not self.s.is_table:
                raise TypeError(
                    "can only use an iterator or chunksize on a table")

            self.coordinates = self.s.read_coordinates(where=self.where)

            return self

        # if specified read via coordinates (necessary for multiple selections
        if coordinates:
            where = self.s.read_coordinates(where=self.where, start=self.start,
            where = self.where

        # directly return the result
        results = self.func(self.start, self.stop, where)
        return results

class IndexCol(StringMixin):

    """ an index column description class


        axis   : axis which I reference
        values : the ndarray like converted values
        kind   : a string description of this type
        typ    : the pytables type
        pos    : the position in the pytables

    is_an_indexable = True
    is_data_indexable = True
    _info_fields = ['freq', 'tz', 'index_name']

    def __init__(self, values=None, kind=None, typ=None, cname=None,
                 itemsize=None, name=None, axis=None, kind_attr=None,
                 pos=None, freq=None, tz=None, index_name=None, **kwargs):
        self.values = values
        self.kind = kind
        self.typ = typ
        self.itemsize = itemsize
        self.name = name
        self.cname = cname
        self.kind_attr = kind_attr
        self.axis = axis
        self.pos = pos
        self.freq = freq
        self.tz = tz
        self.index_name = index_name
        self.table = None
        self.meta = None
        self.metadata = None

        if name is not None:
            self.set_name(name, kind_attr)
        if pos is not None:

    def set_name(self, name, kind_attr=None):
        """ set the name of this indexer """
        self.name = name
        self.kind_attr = kind_attr or "{name}_kind".format(name=name)
        if self.cname is None:
            self.cname = name

        return self

    def set_axis(self, axis):
        """ set the axis over which I index """
        self.axis = axis

        return self

    def set_pos(self, pos):
        """ set the position of this column in the Table """
        self.pos = pos
        if pos is not None and self.typ is not None:
            self.typ._v_pos = pos
        return self

    def set_table(self, table):
        self.table = table
        return self

    def __unicode__(self):
        temp = tuple(
        return ','.join(("{key}->{value}".format(key=key, value=value)
                         for key, value in zip(
            ['name', 'cname', 'axis', 'pos', 'kind'], temp)))

    def __eq__(self, other):
        """ compare 2 col items """
        return all(getattr(self, a, None) == getattr(other, a, None)
                   for a in ['name', 'cname', 'axis', 'pos'])

    def __ne__(self, other):
        return not self.__eq__(other)

    def is_indexed(self):
        """ return whether I am an indexed column """
            return getattr(self.table.cols, self.cname).is_indexed
        except AttributeError:

    def copy(self):
        new_self = copy.copy(self)
        return new_self

    def infer(self, handler):
        """infer this column from the table: create and return a new object"""
        table = handler.table
        new_self = self.copy()
        return new_self

    def convert(self, values, nan_rep, encoding, errors):
        """ set the values from this selection: take = take ownership """

        # values is a recarray
        if values.dtype.fields is not None:
            values = values[self.cname]

        values = _maybe_convert(values, self.kind, encoding, errors)

        kwargs = dict()
        if self.freq is not None:
            kwargs['freq'] = _ensure_decoded(self.freq)
        if self.index_name is not None:
            kwargs['name'] = _ensure_decoded(self.index_name)
        # making an Index instance could throw a number of different errors
            self.values = Index(values, **kwargs)
        except Exception:  # noqa: E722

            # if the output freq is different that what we recorded,
            # it should be None (see also 'doc example part 2')
            if 'freq' in kwargs:
                kwargs['freq'] = None
            self.values = Index(values, **kwargs)

        self.values = _set_tz(self.values, self.tz)

        return self

    def take_data(self):
        """ return the values & release the memory """
        self.values, values = None, self.values
        return values

    def attrs(self):
        return self.table._v_attrs

    def description(self):
        return self.table.description

    def col(self):
        """ return my current col description """
        return getattr(self.description, self.cname, None)

    def cvalues(self):
        """ return my cython values """
        return self.values

    def __iter__(self):
        return iter(self.values)

    def maybe_set_size(self, min_itemsize=None):
        """ maybe set a string col itemsize:
               min_itemsize can be an integer or a dict with this columns name
               with an integer size """
        if _ensure_decoded(self.kind) == u'string':

            if isinstance(min_itemsize, dict):
                min_itemsize = min_itemsize.get(self.name)

            if min_itemsize is not None and self.typ.itemsize < min_itemsize:
                self.typ = _tables(
                ).StringCol(itemsize=min_itemsize, pos=self.pos)

    def validate(self, handler, append):

    def validate_names(self):

    def validate_and_set(self, handler, append):

    def validate_col(self, itemsize=None):
        """ validate this column: return the compared against itemsize """

        # validate this column for string truncation (or reset to the max size)
        if _ensure_decoded(self.kind) == u'string':
            c = self.col
            if c is not None:
                if itemsize is None:
                    itemsize = self.itemsize
                if c.itemsize < itemsize:
                    raise ValueError(
                        "Trying to store a string with len [{itemsize}] in "
                        "[{cname}] column but\nthis column has a limit of "
                        "[{c_itemsize}]!\nConsider using min_itemsize to "
                        "preset the sizes on these columns".format(
                            itemsize=itemsize, cname=self.cname,
                return c.itemsize

        return None

    def validate_attr(self, append):
        # check for backwards incompatibility
        if append:
            existing_kind = getattr(self.attrs, self.kind_attr, None)
            if existing_kind is not None and existing_kind != self.kind:
                raise TypeError(
                    "incompatible kind in col [{existing} - "
                        existing=existing_kind, self_kind=self.kind))

    def update_info(self, info):
        """ set/update the info for this indexable with the key/value
            if there is a conflict raise/warn as needed """

        for key in self._info_fields:

            value = getattr(self, key, None)
            idx = _get_info(info, self.name)

            existing_value = idx.get(key)
            if key in idx and value is not None and existing_value != value:

                # frequency/name just warn
                if key in ['freq', 'index_name']:
                    ws = attribute_conflict_doc % (key, existing_value, value)
                    warnings.warn(ws, AttributeConflictWarning, stacklevel=6)

                    # reset
                    idx[key] = None
                    setattr(self, key, None)

                    raise ValueError(
                        "invalid info for [{name}] for [{key}], "
                        "existing_value [{existing_value}] conflicts with "
                        "new value [{value}]".format(
                            name=self.name, key=key,
                            existing_value=existing_value, value=value))
                if value is not None or existing_value is not None:
                    idx[key] = value

        return self

    def set_info(self, info):
        """ set my state from the passed info """
        idx = info.get(self.name)
        if idx is not None:

    def get_attr(self):
        """ set the kind for this column """
        self.kind = getattr(self.attrs, self.kind_attr, None)

    def set_attr(self):
        """ set the kind for this column """
        setattr(self.attrs, self.kind_attr, self.kind)

    def read_metadata(self, handler):
        """ retrieve the metadata for this columns """
        self.metadata = handler.read_metadata(self.cname)

    def validate_metadata(self, handler):
        """ validate that kind=category does not change the categories """
        if self.meta == 'category':
            new_metadata = self.metadata
            cur_metadata = handler.read_metadata(self.cname)
            if (new_metadata is not None and cur_metadata is not None and
                    not array_equivalent(new_metadata, cur_metadata)):
                raise ValueError("cannot append a categorical with "
                                 "different categories to the existing")

    def write_metadata(self, handler):
        """ set the meta data """
        if self.metadata is not None:
            handler.write_metadata(self.cname, self.metadata)

class GenericIndexCol(IndexCol):

    """ an index which is not represented in the data of the table """

    def is_indexed(self):
        return False

    def convert(self, values, nan_rep, encoding, errors):
        """ set the values from this selection: take = take ownership """

        self.values = Int64Index(np.arange(self.table.nrows))
        return self

    def get_attr(self):

    def set_attr(self):

class DataCol(IndexCol):

    """ a data holding column, by definition this is not indexable


        data   : the actual data
        cname  : the column name in the table to hold the data (typically
        meta   : a string description of the metadata
        metadata : the actual metadata
    is_an_indexable = False
    is_data_indexable = False
    _info_fields = ['tz', 'ordered']

    def create_for_block(
            cls, i=None, name=None, cname=None, version=None, **kwargs):
        """ return a new datacol with the block i """

        if cname is None:
            cname = name or 'values_block_{idx}'.format(idx=i)
        if name is None:
            name = cname

        # prior to 0.10.1, we named values blocks like: values_block_0 an the
        # name values_0
            if version[0] == 0 and version[1] <= 10 and version[2] == 0:
                m = re.search(r"values_block_(\d+)", name)
                if m:
                    name = "values_{group}".format(group=m.groups()[0])
        except IndexError:

        return cls(name=name, cname=cname, **kwargs)

    def __init__(self, values=None, kind=None, typ=None,
                 cname=None, data=None, meta=None, metadata=None,
                 block=None, **kwargs):
        super(DataCol, self).__init__(values=values, kind=kind, typ=typ,
                                      cname=cname, **kwargs)
        self.dtype = None
        self.dtype_attr = u'{name}_dtype'.format(name=self.name)
        self.meta = meta
        self.meta_attr = u'{name}_meta'.format(name=self.name)

    def __unicode__(self):
        temp = tuple(
        return ','.join(("{key}->{value}".format(key=key, value=value)
                         for key, value in zip(
            ['name', 'cname', 'dtype', 'kind', 'shape'], temp)))

    def __eq__(self, other):
        """ compare 2 col items """
        return all(getattr(self, a, None) == getattr(other, a, None)
                   for a in ['name', 'cname', 'dtype', 'pos'])

    def set_data(self, data, dtype=None):
        self.data = data
        if data is not None:
            if dtype is not None:
                self.dtype = dtype
            elif self.dtype is None:
                self.dtype = data.dtype.name

    def take_data(self):
        """ return the data & release the memory """
        self.data, data = None, self.data
        return data

    def set_metadata(self, metadata):
        """ record the metadata """
        if metadata is not None:
            metadata = np.array(metadata, copy=False).ravel()
        self.metadata = metadata

    def set_kind(self):
        # set my kind if we can

        if self.dtype is not None:
            dtype = _ensure_decoded(self.dtype)

            if dtype.startswith(u'string') or dtype.startswith(u'bytes'):
                self.kind = 'string'
            elif dtype.startswith(u'float'):
                self.kind = 'float'
            elif dtype.startswith(u'complex'):
                self.kind = 'complex'
            elif dtype.startswith(u'int') or dtype.startswith(u'uint'):
                self.kind = 'integer'
            elif dtype.startswith(u'date'):
                self.kind = 'datetime'
            elif dtype.startswith(u'timedelta'):
                self.kind = 'timedelta'
            elif dtype.startswith(u'bool'):
                self.kind = 'bool'
                raise AssertionError(
                    "cannot interpret dtype of [{dtype}] in [{obj}]".format(
                        dtype=dtype, obj=self))

            # set my typ if we need
            if self.typ is None:
                self.typ = getattr(self.description, self.cname, None)

    def set_atom(self, block, block_items, existing_col, min_itemsize,
                 nan_rep, info, encoding=None, errors='strict'):
        """ create and setup my atom from the block b """

        self.values = list(block_items)

        # short-cut certain block types
        if block.is_categorical:
            return self.set_atom_categorical(block, items=block_items,
        elif block.is_datetimetz:
            return self.set_atom_datetime64tz(block, info=info)
        elif block.is_datetime:
            return self.set_atom_datetime64(block)
        elif block.is_timedelta:
            return self.set_atom_timedelta64(block)
        elif block.is_complex:
            return self.set_atom_complex(block)

        dtype = block.dtype.name
        inferred_type = lib.infer_dtype(block.values, skipna=False)

        if inferred_type == 'date':
            raise TypeError(
                "[date] is not implemented as a table column")
        elif inferred_type == 'datetime':
            # after 8260
            # this only would be hit for a mutli-timezone dtype
            # which is an error

            raise TypeError(
                "too many timezones in this block, create separate "
                "data columns"
        elif inferred_type == 'unicode':
            raise TypeError(
                "[unicode] is not implemented as a table column")

        # this is basically a catchall; if say a datetime64 has nans then will
        # end up here ###
        elif inferred_type == 'string' or dtype == 'object':
                block, block_items,

        # set as a data block

    def get_atom_string(self, block, itemsize):
        return _tables().StringCol(itemsize=itemsize, shape=block.shape[0])

    def set_atom_string(self, block, block_items, existing_col, min_itemsize,
                        nan_rep, encoding, errors):
        # fill nan items with myself, don't disturb the blocks by
        # trying to downcast
        block = block.fillna(nan_rep, downcast=False)
        if isinstance(block, list):
            block = block[0]
        data = block.values

        # see if we have a valid string type
        inferred_type = lib.infer_dtype(data.ravel(), skipna=False)
        if inferred_type != 'string':

            # we cannot serialize this data, so report an exception on a column
            # by column basis
            for i, item in enumerate(block_items):

                col = block.iget(i)
                inferred_type = lib.infer_dtype(col.ravel(), skipna=False)
                if inferred_type != 'string':
                    raise TypeError(
                        "Cannot serialize the column [{item}] because\n"
                        "its data contents are [{type}] object dtype".format(
                            item=item, type=inferred_type)

        # itemsize is the maximum length of a string (along any dimension)
        data_converted = _convert_string_array(data, encoding, errors)
        itemsize = data_converted.itemsize

        # specified min_itemsize?
        if isinstance(min_itemsize, dict):
            min_itemsize = int(min_itemsize.get(
                self.name) or min_itemsize.get('values') or 0)
        itemsize = max(min_itemsize or 0, itemsize)

        # check for column in the values conflicts
        if existing_col is not None:
            eci = existing_col.validate_col(itemsize)
            if eci > itemsize:
                itemsize = eci

        self.itemsize = itemsize
        self.kind = 'string'
        self.typ = self.get_atom_string(block, itemsize)
            '|S{size}'.format(size=itemsize), copy=False))

    def get_atom_coltype(self, kind=None):
        """ return the PyTables column class for this column """
        if kind is None:
            kind = self.kind
        if self.kind.startswith('uint'):
            col_name = "UInt{name}Col".format(name=kind[4:])
            col_name = "{name}Col".format(name=kind.capitalize())

        return getattr(_tables(), col_name)

    def get_atom_data(self, block, kind=None):
        return self.get_atom_coltype(kind=kind)(shape=block.shape[0])

    def set_atom_complex(self, block):
        self.kind = block.dtype.name
        itemsize = int(self.kind.split('complex')[-1]) // 8
        self.typ = _tables().ComplexCol(
            itemsize=itemsize, shape=block.shape[0])
        self.set_data(block.values.astype(self.typ.type, copy=False))

    def set_atom_data(self, block):
        self.kind = block.dtype.name
        self.typ = self.get_atom_data(block)
        self.set_data(block.values.astype(self.typ.type, copy=False))

    def set_atom_categorical(self, block, items, info=None, values=None):
        # currently only supports a 1-D categorical
        # in a 1-D block

        values = block.values
        codes = values.codes
        self.kind = 'integer'
        self.dtype = codes.dtype.name
        if values.ndim > 1:
            raise NotImplementedError("only support 1-d categoricals")
        if len(items) > 1:
            raise NotImplementedError("only support single block categoricals")

        # write the codes; must be in a block shape
        self.ordered = values.ordered
        self.typ = self.get_atom_data(block, kind=codes.dtype.name)

        # write the categories
        self.meta = 'category'

        # update the info

    def get_atom_datetime64(self, block):
        return _tables().Int64Col(shape=block.shape[0])

    def set_atom_datetime64(self, block, values=None):
        self.kind = 'datetime64'
        self.typ = self.get_atom_datetime64(block)
        if values is None:
            values = block.values.view('i8')
        self.set_data(values, 'datetime64')

    def set_atom_datetime64tz(self, block, info, values=None):

        if values is None:
            values = block.values

        # convert this column to i8 in UTC, and save the tz
        values = values.asi8.reshape(block.shape)

        # store a converted timezone
        self.tz = _get_tz(block.values.tz)

        self.kind = 'datetime64'
        self.typ = self.get_atom_datetime64(block)
        self.set_data(values, 'datetime64')

    def get_atom_timedelta64(self, block):
        return _tables().Int64Col(shape=block.shape[0])

    def set_atom_timedelta64(self, block, values=None):
        self.kind = 'timedelta64'
        self.typ = self.get_atom_timedelta64(block)
        if values is None:
            values = block.values.view('i8')
        self.set_data(values, 'timedelta64')

    def shape(self):
        return getattr(self.data, 'shape', None)

    def cvalues(self):
        """ return my cython values """
        return self.data

    def validate_attr(self, append):
        """validate that we have the same order as the existing & same dtype"""
        if append:
            existing_fields = getattr(self.attrs, self.kind_attr, None)
            if (existing_fields is not None and
                    existing_fields != list(self.values)):
                raise ValueError("appended items do not match existing items"
                                 " in table!")

            existing_dtype = getattr(self.attrs, self.dtype_attr, None)
            if (existing_dtype is not None and
                    existing_dtype != self.dtype):
                raise ValueError("appended items dtype do not match existing "
                                 "items dtype in table!")

    def convert(self, values, nan_rep, encoding, errors):
        """set the data from this selection (and convert to the correct dtype
        if we can)

        # values is a recarray
        if values.dtype.fields is not None:
            values = values[self.cname]


        # use the meta if needed
        meta = _ensure_decoded(self.meta)

        # convert to the correct dtype
        if self.dtype is not None:
            dtype = _ensure_decoded(self.dtype)

            # reverse converts
            if dtype == u'datetime64':

                # recreate with tz if indicated
                self.data = _set_tz(self.data, self.tz, coerce=True)

            elif dtype == u'timedelta64':
                self.data = np.asarray(self.data, dtype='m8[ns]')
            elif dtype == u'date':
                    self.data = np.asarray(
                        [date.fromordinal(v) for v in self.data], dtype=object)
                except ValueError:
                    self.data = np.asarray(
                        [date.fromtimestamp(v) for v in self.data],
            elif dtype == u'datetime':
                self.data = np.asarray(
                    [datetime.fromtimestamp(v) for v in self.data],

            elif meta == u'category':

                # we have a categorical
                categories = self.metadata
                codes = self.data.ravel()

                # if we have stored a NaN in the categories
                # then strip it; in theory we could have BOTH
                # -1s in the codes and nulls :<
                if categories is None:
                    # Handle case of NaN-only categorical columns in which case
                    # the categories are an empty array; when this is stored,
                    # pytables cannot write a zero-len array, so on readback
                    # the categories would be None and `read_hdf()` would fail.
                    categories = Index([], dtype=np.float64)
                    mask = isna(categories)
                    if mask.any():
                        categories = categories[~mask]
                        codes[codes != -1] -= mask.astype(int).cumsum().values

                self.data = Categorical.from_codes(codes,


                    self.data = self.data.astype(dtype, copy=False)
                except TypeError:
                    self.data = self.data.astype('O', copy=False)

        # convert nans / decode
        if _ensure_decoded(self.kind) == u'string':
            self.data = _unconvert_string_array(
                self.data, nan_rep=nan_rep, encoding=encoding, errors=errors)

        return self

    def get_attr(self):
        """ get the data for this column """
        self.values = getattr(self.attrs, self.kind_attr, None)
        self.dtype = getattr(self.attrs, self.dtype_attr, None)
        self.meta = getattr(self.attrs, self.meta_attr, None)

    def set_attr(self):
        """ set the data for this column """
        setattr(self.attrs, self.kind_attr, self.values)
        setattr(self.attrs, self.meta_attr, self.meta)
        if self.dtype is not None:
            setattr(self.attrs, self.dtype_attr, self.dtype)

class DataIndexableCol(DataCol):

    """ represent a data column that can be indexed """
    is_data_indexable = True

    def validate_names(self):
        if not Index(self.values).is_object():
            raise ValueError("cannot have non-object label DataIndexableCol")

    def get_atom_string(self, block, itemsize):
        return _tables().StringCol(itemsize=itemsize)

    def get_atom_data(self, block, kind=None):
        return self.get_atom_coltype(kind=kind)()

    def get_atom_datetime64(self, block):
        return _tables().Int64Col()

    def get_atom_timedelta64(self, block):
        return _tables().Int64Col()

class GenericDataIndexableCol(DataIndexableCol):

    """ represent a generic pytables data column """

    def get_attr(self):

class Fixed(StringMixin):

    """ represent an object in my store
        facilitate read/write of various types of objects
        this is an abstract base class


        parent : my parent HDFStore
        group  : the group node where the table resides
    pandas_kind = None
    obj_type = None
    ndim = None
    is_table = False

    def __init__(self, parent, group, encoding=None, errors='strict',
        self.parent = parent
        self.group = group
        self.encoding = _ensure_encoding(encoding)
        self.errors = errors

    def is_old_version(self):
        return (self.version[0] <= 0 and self.version[1] <= 10 and
                self.version[2] < 1)

    def set_version(self):
        """ compute and set our version """
        version = _ensure_decoded(
            getattr(self.group._v_attrs, 'pandas_version', None))
            self.version = tuple(int(x) for x in version.split('.'))
            if len(self.version) == 2:
                self.version = self.version + (0,)
        except AttributeError:
            self.version = (0, 0, 0)

    def pandas_type(self):
        return _ensure_decoded(getattr(self.group._v_attrs,
                                       'pandas_type', None))

    def format_type(self):
        return 'fixed'

    def __unicode__(self):
        """ return a pretty representation of myself """
        s = self.shape
        if s is not None:
            if isinstance(s, (list, tuple)):
                s = "[{shape}]".format(
                    shape=','.join(pprint_thing(x) for x in s))
            return "{type:12.12} (shape->{shape})".format(
                type=self.pandas_type, shape=s)
        return self.pandas_type

    def set_object_info(self):
        """ set my pandas type & version """
        self.attrs.pandas_type = str(self.pandas_kind)
        self.attrs.pandas_version = str(_version)

    def copy(self):
        new_self = copy.copy(self)
        return new_self

    def storage_obj_type(self):
        return self.obj_type

    def shape(self):
        return self.nrows

    def pathname(self):
        return self.group._v_pathname

    def _handle(self):
        return self.parent._handle

    def _filters(self):
        return self.parent._filters

    def _complevel(self):
        return self.parent._complevel

    def _fletcher32(self):
        return self.parent._fletcher32

    def _complib(self):
        return self.parent._complib

    def attrs(self):
        return self.group._v_attrs

    def set_attrs(self):
        """ set our object attributes """

    def get_attrs(self):
        """ get our object attributes """

    def storable(self):
        """ return my storable """
        return self.group

    def is_exists(self):
        return False

    def nrows(self):
        return getattr(self.storable, 'nrows', None)

    def validate(self, other):
        """ validate against an existing storable """
        if other is None:
        return True

    def validate_version(self, where=None):
        """ are we trying to operate on an old version? """
        return True

    def infer_axes(self):
        """ infer the axes of my storer
              return a boolean indicating if we have a valid storer or not """

        s = self.storable
        if s is None:
            return False
        return True

    def read(self, **kwargs):
        raise NotImplementedError(
            "cannot read on an abstract storer: subclasses should implement")

    def write(self, **kwargs):
        raise NotImplementedError(
            "cannot write on an abstract storer: sublcasses should implement")

    def delete(self, where=None, start=None, stop=None, **kwargs):
        support fully deleting the node in its entirety (only) - where
        specification must be None
        if com._all_none(where, start, stop):
            self._handle.remove_node(self.group, recursive=True)
            return None

        raise TypeError("cannot delete on an abstract storer")

class GenericFixed(Fixed):

    """ a generified fixed version """
    _index_type_map = {DatetimeIndex: 'datetime', PeriodIndex: 'period'}
    _reverse_index_map = {v: k for k, v in compat.iteritems(_index_type_map)}
    attributes = []

    # indexer helpders
    def _class_to_alias(self, cls):
        return self._index_type_map.get(cls, '')

    def _alias_to_class(self, alias):
        if isinstance(alias, type):  # pragma: no cover
            # compat: for a short period of time master stored types
            return alias
        return self._reverse_index_map.get(alias, Index)

    def _get_index_factory(self, klass):
        if klass == DatetimeIndex:
            def f(values, freq=None, tz=None):
                # data are already in UTC, localize and convert if tz present
                result = DatetimeIndex._simple_new(values.values, name=None,
                if tz is not None:
                    result = result.tz_localize('UTC').tz_convert(tz)
                return result
            return f
        elif klass == PeriodIndex:
            def f(values, freq=None, tz=None):
                return PeriodIndex._simple_new(values, name=None, freq=freq)
            return f

        return klass

    def validate_read(self, kwargs):
        remove table keywords from kwargs and return
        raise if any keywords are passed which are not-None
        kwargs = copy.copy(kwargs)

        columns = kwargs.pop('columns', None)
        if columns is not None:
            raise TypeError("cannot pass a column specification when reading "
                            "a Fixed format store. this store must be "
                            "selected in its entirety")
        where = kwargs.pop('where', None)
        if where is not None:
            raise TypeError("cannot pass a where specification when reading "
                            "from a Fixed format store. this store must be "
                            "selected in its entirety")
        return kwargs

    def is_exists(self):
        return True

    def set_attrs(self):
        """ set our object attributes """
        self.attrs.encoding = self.encoding
        self.attrs.errors = self.errors

    def get_attrs(self):
        """ retrieve our attributes """
        self.encoding = _ensure_encoding(getattr(self.attrs, 'encoding', None))
        self.errors = _ensure_decoded(getattr(self.attrs, 'errors', 'strict'))
        for n in self.attributes:
            setattr(self, n, _ensure_decoded(getattr(self.attrs, n, None)))

    def write(self, obj, **kwargs):

    def read_array(self, key, start=None, stop=None):
        """ read an array for the specified node (off of group """
        import tables
        node = getattr(self.group, key)
        attrs = node._v_attrs

        transposed = getattr(attrs, 'transposed', False)

        if isinstance(node, tables.VLArray):
            ret = node[0][start:stop]
            dtype = getattr(attrs, 'value_type', None)
            shape = getattr(attrs, 'shape', None)

            if shape is not None:
                # length 0 axis
                ret = np.empty(shape, dtype=dtype)
                ret = node[start:stop]

            if dtype == u'datetime64':

                # reconstruct a timezone if indicated
                ret = _set_tz(ret, getattr(attrs, 'tz', None), coerce=True)

            elif dtype == u'timedelta64':
                ret = np.asarray(ret, dtype='m8[ns]')

        if transposed:
            return ret.T
            return ret

    def read_index(self, key, **kwargs):
        variety = _ensure_decoded(
            getattr(self.attrs, '{key}_variety'.format(key=key)))

        if variety == u'multi':
            return self.read_multi_index(key, **kwargs)
        elif variety == u'block':
            return self.read_block_index(key, **kwargs)
        elif variety == u'sparseint':
            return self.read_sparse_intindex(key, **kwargs)
        elif variety == u'regular':
            _, index = self.read_index_node(getattr(self.group, key), **kwargs)
            return index
        else:  # pragma: no cover
            raise TypeError(
                'unrecognized index variety: {variety}'.format(

    def write_index(self, key, index):
        if isinstance(index, MultiIndex):
            setattr(self.attrs, '{key}_variety'.format(key=key), 'multi')
            self.write_multi_index(key, index)
        elif isinstance(index, BlockIndex):
            setattr(self.attrs, '{key}_variety'.format(key=key), 'block')
            self.write_block_index(key, index)
        elif isinstance(index, IntIndex):
            setattr(self.attrs, '{key}_variety'.format(key=key), 'sparseint')
            self.write_sparse_intindex(key, index)
            setattr(self.attrs, '{key}_variety'.format(key=key), 'regular')
            converted = _convert_index(index, self.encoding, self.errors,

            self.write_array(key, converted.values)

            node = getattr(self.group, key)
            node._v_attrs.kind = converted.kind
            node._v_attrs.name = index.name

            if isinstance(index, (DatetimeIndex, PeriodIndex)):
                node._v_attrs.index_class = self._class_to_alias(type(index))

            if hasattr(index, 'freq'):
                node._v_attrs.freq = index.freq

            if hasattr(index, 'tz') and index.tz is not None:
                node._v_attrs.tz = _get_tz(index.tz)

    def write_block_index(self, key, index):
        self.write_array('{key}_blocs'.format(key=key), index.blocs)
        self.write_array('{key}_blengths'.format(key=key), index.blengths)
        setattr(self.attrs, '{key}_length'.format(key=key), index.length)

    def read_block_index(self, key, **kwargs):
        length = getattr(self.attrs, '{key}_length'.format(key=key))
        blocs = self.read_array('{key}_blocs'.format(key=key), **kwargs)
        blengths = self.read_array('{key}_blengths'.format(key=key), **kwargs)
        return BlockIndex(length, blocs, blengths)

    def write_sparse_intindex(self, key, index):
        self.write_array('{key}_indices'.format(key=key), index.indices)
        setattr(self.attrs, '{key}_length'.format(key=key), index.length)

    def read_sparse_intindex(self, key, **kwargs):
        length = getattr(self.attrs, '{key}_length'.format(key=key))
        indices = self.read_array('{key}_indices'.format(key=key), **kwargs)
        return IntIndex(length, indices)

    def write_multi_index(self, key, index):
        setattr(self.attrs, '{key}_nlevels'.format(key=key), index.nlevels)

        for i, (lev, level_codes, name) in enumerate(zip(index.levels,
            # write the level
            level_key = '{key}_level{idx}'.format(key=key, idx=i)
            conv_level = _convert_index(lev, self.encoding, self.errors,
            self.write_array(level_key, conv_level.values)
            node = getattr(self.group, level_key)
            node._v_attrs.kind = conv_level.kind
            node._v_attrs.name = name

            # write the name
            setattr(node._v_attrs, '{key}_name{name}'.format(
                key=key, name=name), name)

            # write the labels
            label_key = '{key}_label{idx}'.format(key=key, idx=i)
            self.write_array(label_key, level_codes)

    def read_multi_index(self, key, **kwargs):
        nlevels = getattr(self.attrs, '{key}_nlevels'.format(key=key))

        levels = []
        codes = []
        names = []
        for i in range(nlevels):
            level_key = '{key}_level{idx}'.format(key=key, idx=i)
            name, lev = self.read_index_node(getattr(self.group, level_key),

            label_key = '{key}_label{idx}'.format(key=key, idx=i)
            level_codes = self.read_array(label_key, **kwargs)

        return MultiIndex(levels=levels, codes=codes, names=names,

    def read_index_node(self, node, start=None, stop=None):
        data = node[start:stop]
        # If the index was an empty array write_array_empty() will
        # have written a sentinel. Here we relace it with the original.
        if ('shape' in node._v_attrs and
                self._is_empty_array(getattr(node._v_attrs, 'shape'))):
            data = np.empty(getattr(node._v_attrs, 'shape'),
                            dtype=getattr(node._v_attrs, 'value_type'))
        kind = _ensure_decoded(node._v_attrs.kind)
        name = None

        if 'name' in node._v_attrs:
            name = _ensure_str(node._v_attrs.name)
            name = _ensure_decoded(name)

        index_class = self._alias_to_class(_ensure_decoded(
            getattr(node._v_attrs, 'index_class', '')))
        factory = self._get_index_factory(index_class)

        kwargs = {}
        if u'freq' in node._v_attrs:
            kwargs['freq'] = node._v_attrs['freq']

        if u'tz' in node._v_attrs:
            kwargs['tz'] = node._v_attrs['tz']

        if kind in (u'date', u'datetime'):
            index = factory(_unconvert_index(data, kind,
                            dtype=object, **kwargs)
            index = factory(_unconvert_index(data, kind,
                                             errors=self.errors), **kwargs)

        index.name = name

        return name, index

    def write_array_empty(self, key, value):
        """ write a 0-len array """

        # ugly hack for length 0 axes
        arr = np.empty((1,) * value.ndim)
        self._handle.create_array(self.group, key, arr)
        getattr(self.group, key)._v_attrs.value_type = str(value.dtype)
        getattr(self.group, key)._v_attrs.shape = value.shape

    def _is_empty_array(self, shape):
        """Returns true if any axis is zero length."""
        return any(x == 0 for x in shape)

    def write_array(self, key, value, items=None):
        if key in self.group:
            self._handle.remove_node(self.group, key)

        # Transform needed to interface with pytables row/col notation
        empty_array = self._is_empty_array(value.shape)
        transposed = False

        if is_categorical_dtype(value):
            raise NotImplementedError('Cannot store a category dtype in '
                                      'a HDF5 dataset that uses format='
                                      '"fixed". Use format="table".')
        if not empty_array:
            if hasattr(value, 'T'):
                # ExtensionArrays (1d) may not have transpose.
                value = value.T
                transposed = True

        if self._filters is not None:
            atom = None
                # get the atom for this datatype
                atom = _tables().Atom.from_dtype(value.dtype)
            except ValueError:

            if atom is not None:
                # create an empty chunked array and fill it from value
                if not empty_array:
                    ca = self._handle.create_carray(self.group, key, atom,
                    ca[:] = value
                    getattr(self.group, key)._v_attrs.transposed = transposed

                    self.write_array_empty(key, value)


        if value.dtype.type == np.object_:

            # infer the type, warn if we have a non-string type here (for
            # performance)
            inferred_type = lib.infer_dtype(value.ravel(), skipna=False)
            if empty_array:
            elif inferred_type == 'string':
                    items = list(items)
                except TypeError:
                ws = performance_doc % (inferred_type, key, items)
                warnings.warn(ws, PerformanceWarning, stacklevel=7)

            vlarr = self._handle.create_vlarray(self.group, key,
            if empty_array:
                self.write_array_empty(key, value)
                if is_datetime64_dtype(value.dtype):
                        self.group, key, value.view('i8'))
                        self.group, key)._v_attrs.value_type = 'datetime64'
                elif is_datetime64tz_dtype(value.dtype):
                    # store as UTC
                    # with a zone
                    self._handle.create_array(self.group, key,

                    node = getattr(self.group, key)
                    node._v_attrs.tz = _get_tz(value.tz)
                    node._v_attrs.value_type = 'datetime64'
                elif is_timedelta64_dtype(value.dtype):
                        self.group, key, value.view('i8'))
                        self.group, key)._v_attrs.value_type = 'timedelta64'
                    self._handle.create_array(self.group, key, value)

        getattr(self.group, key)._v_attrs.transposed = transposed

class LegacyFixed(GenericFixed):

    def read_index_legacy(self, key, start=None, stop=None):
        node = getattr(self.group, key)
        data = node[start:stop]
        kind = node._v_attrs.kind
        return _unconvert_index_legacy(data, kind, encoding=self.encoding,

class LegacySeriesFixed(LegacyFixed):

    def read(self, **kwargs):
        kwargs = self.validate_read(kwargs)
        index = self.read_index_legacy('index')
        values = self.read_array('values')
        return Series(values, index=index)

class LegacyFrameFixed(LegacyFixed):

    def read(self, **kwargs):
        kwargs = self.validate_read(kwargs)
        index = self.read_index_legacy('index')
        columns = self.read_index_legacy('columns')
        values = self.read_array('values')
        return DataFrame(values, index=index, columns=columns)

class SeriesFixed(GenericFixed):
    pandas_kind = u'series'
    attributes = ['name']

    def shape(self):
            return len(getattr(self.group, 'values')),
        except (TypeError, AttributeError):
            return None

    def read(self, **kwargs):
        kwargs = self.validate_read(kwargs)
        index = self.read_index('index', **kwargs)
        values = self.read_array('values', **kwargs)
        return Series(values, index=index, name=self.name)

    def write(self, obj, **kwargs):
        super(SeriesFixed, self).write(obj, **kwargs)
        self.write_index('index', obj.index)
        self.write_array('values', obj.values)
        self.attrs.name = obj.name

class SparseFixed(GenericFixed):

    def validate_read(self, kwargs):
        we don't support start, stop kwds in Sparse
        kwargs = super(SparseFixed, self).validate_read(kwargs)
        if 'start' in kwargs or 'stop' in kwargs:
            raise NotImplementedError("start and/or stop are not supported "
                                      "in fixed Sparse reading")
        return kwargs

class SparseSeriesFixed(SparseFixed):
    pandas_kind = u'sparse_series'
    attributes = ['name', 'fill_value', 'kind']

    def read(self, **kwargs):
        kwargs = self.validate_read(kwargs)
        index = self.read_index('index')
        sp_values = self.read_array('sp_values')
        sp_index = self.read_index('sp_index')
        return SparseSeries(sp_values, index=index, sparse_index=sp_index,
                            kind=self.kind or u'block',

    def write(self, obj, **kwargs):
        super(SparseSeriesFixed, self).write(obj, **kwargs)
        self.write_index('index', obj.index)
        self.write_index('sp_index', obj.sp_index)
        self.write_array('sp_values', obj.sp_values)
        self.attrs.name = obj.name
        self.attrs.fill_value = obj.fill_value
        self.attrs.kind = obj.kind

class SparseFrameFixed(SparseFixed):
    pandas_kind = u'sparse_frame'
    attributes = ['default_kind', 'default_fill_value']

    def read(self, **kwargs):
        kwargs = self.validate_read(kwargs)
        columns = self.read_index('columns')
        sdict = {}
        for c in columns:
            key = 'sparse_series_{columns}'.format(columns=c)
            s = SparseSeriesFixed(self.parent, getattr(self.group, key))
            sdict[c] = s.read()
        return SparseDataFrame(sdict, columns=columns,

    def write(self, obj, **kwargs):
        """ write it as a collection of individual sparse series """
        super(SparseFrameFixed, self).write(obj, **kwargs)
        for name, ss in compat.iteritems(obj):
            key = 'sparse_series_{name}'.format(name=name)
            if key not in self.group._v_children:
                node = self._handle.create_group(self.group, key)
                node = getattr(self.group, key)
            s = SparseSeriesFixed(self.parent, node)
        self.attrs.default_fill_value = obj.default_fill_value
        self.attrs.default_kind = obj.default_kind
        self.write_index('columns', obj.columns)

class BlockManagerFixed(GenericFixed):
    attributes = ['ndim', 'nblocks']
    is_shape_reversed = False

    def shape(self):
            ndim = self.ndim

            # items
            items = 0
            for i in range(self.nblocks):
                node = getattr(self.group, 'block{idx}_items'.format(idx=i))
                shape = getattr(node, 'shape', None)
                if shape is not None:
                    items += shape[0]

            # data shape
            node = getattr(self.group, 'block0_values')
            shape = getattr(node, 'shape', None)
            if shape is not None:
                shape = list(shape[0:(ndim - 1)])
                shape = []


            # hacky - this works for frames, but is reversed for panels
            if self.is_shape_reversed:
                shape = shape[::-1]

            return shape
        except AttributeError:
            return None

    def read(self, start=None, stop=None, **kwargs):
        # start, stop applied to rows, so 0th axis only

        kwargs = self.validate_read(kwargs)
        select_axis = self.obj_type()._get_block_manager_axis(0)

        axes = []
        for i in range(self.ndim):

            _start, _stop = (start, stop) if i == select_axis else (None, None)
            ax = self.read_index('axis{idx}'.format(
                idx=i), start=_start, stop=_stop)

        items = axes[0]
        blocks = []
        for i in range(self.nblocks):

            blk_items = self.read_index('block{idx}_items'.format(idx=i))
            values = self.read_array('block{idx}_values'.format(idx=i),
                                     start=_start, stop=_stop)
            blk = make_block(values,

        return self.obj_type(BlockManager(blocks, axes))

    def write(self, obj, **kwargs):
        super(BlockManagerFixed, self).write(obj, **kwargs)
        data = obj._data
        if not data.is_consolidated():
            data = data.consolidate()

        self.attrs.ndim = data.ndim
        for i, ax in enumerate(data.axes):
            if i == 0:
                if not ax.is_unique:
                    raise ValueError(
                        "Columns index has to be unique for fixed format")
            self.write_index('axis{idx}'.format(idx=i), ax)

        # Supporting mixed-type DataFrame objects...nontrivial
        self.attrs.nblocks = len(data.blocks)
        for i, blk in enumerate(data.blocks):
            # I have no idea why, but writing values before items fixed #2299
            blk_items = data.items.take(blk.mgr_locs)
                             blk.values, items=blk_items)
            self.write_index('block{idx}_items'.format(idx=i), blk_items)

class FrameFixed(BlockManagerFixed):
    pandas_kind = u'frame'
    obj_type = DataFrame

class PanelFixed(BlockManagerFixed):
    pandas_kind = u'wide'
    obj_type = Panel
    is_shape_reversed = True

    def write(self, obj, **kwargs):
        return super(PanelFixed, self).write(obj, **kwargs)

class Table(Fixed):

    """ represent a table:
          facilitate read/write of various types of tables

        Attrs in Table Node
        These are attributes that are store in the main table node, they are
        necessary to recreate these tables when read back in.

        index_axes    : a list of tuples of the (original indexing axis and
            index column)
        non_index_axes: a list of tuples of the (original index axis and
            columns on a non-indexing axis)
        values_axes   : a list of the columns which comprise the data of this
        data_columns  : a list of the columns that we are allowing indexing
            (these become single columns in values_axes), or True to force all
        nan_rep       : the string to use for nan representations for string
        levels        : the names of levels
        metadata      : the names of the metadata columns

    pandas_kind = u'wide_table'
    table_type = None
    levels = 1
    is_table = True
    is_shape_reversed = False

    def __init__(self, *args, **kwargs):
        super(Table, self).__init__(*args, **kwargs)
        self.index_axes = []
        self.non_index_axes = []
        self.values_axes = []
        self.data_columns = []
        self.metadata = []
        self.info = dict()
        self.nan_rep = None
        self.selection = None

    def table_type_short(self):
        return self.table_type.split('_')[0]

    def format_type(self):
        return 'table'

    def __unicode__(self):
        """ return a pretty representatgion of myself """
        dc = ",dc->[{columns}]".format(columns=(','.join(
            self.data_columns) if len(self.data_columns) else ''))

        ver = ''
        if self.is_old_version:
            ver = "[{version}]".format(
                version='.'.join(str(x) for x in self.version))

        return (
            "{pandas_type:12.12}{ver} (typ->{table_type},nrows->{nrows},"
                pandas_type=self.pandas_type, ver=ver,
                table_type=self.table_type_short, nrows=self.nrows,
                index_axes=(','.join(a.name for a in self.index_axes)), dc=dc

    def __getitem__(self, c):
        """ return the axis for c """
        for a in self.axes:
            if c == a.name:
                return a
        return None

    def validate(self, other):
        """ validate against an existing table """
        if other is None:

        if other.table_type != self.table_type:
            raise TypeError(
                "incompatible table_type with existing "
                "[{other} - {self}]".format(
                    other=other.table_type, self=self.table_type))

        for c in ['index_axes', 'non_index_axes', 'values_axes']:
            sv = getattr(self, c, None)
            ov = getattr(other, c, None)
            if sv != ov:

                # show the error for the specific axes
                for i, sax in enumerate(sv):
                    oax = ov[i]
                    if sax != oax:
                        raise ValueError(
                            "invalid combinate of [{c}] on appending data "
                            "[{sax}] vs current table [{oax}]".format(
                                c=c, sax=sax, oax=oax))

                # should never get here
                raise Exception(
                    "invalid combinate of [{c}] on appending data [{sv}] vs "
                    "current table [{ov}]".format(c=c, sv=sv, ov=ov))

    def is_multi_index(self):
        """the levels attribute is 1 or a list in the case of a multi-index"""
        return isinstance(self.levels, list)

    def validate_metadata(self, existing):
        """ create / validate metadata """
        self.metadata = [
            c.name for c in self.values_axes if c.metadata is not None]

    def validate_multiindex(self, obj):
        """validate that we can store the multi-index; reset and return the
        new object
        levels = [l if l is not None else "level_{0}".format(i)
                  for i, l in enumerate(obj.index.names)]
            return obj.reset_index(), levels
        except ValueError:
            raise ValueError("duplicate names/columns in the multi-index when "
                             "storing as a table")

    def nrows_expected(self):
        """ based on our axes, compute the expected nrows """
        return np.prod([i.cvalues.shape[0] for i in self.index_axes])

    def is_exists(self):
        """ has this table been created """
        return u'table' in self.group

    def storable(self):
        return getattr(self.group, 'table', None)

    def table(self):
        """ return the table group (this is my storable) """
        return self.storable

    def dtype(self):
        return self.table.dtype

    def description(self):
        return self.table.description

    def axes(self):
        return itertools.chain(self.index_axes, self.values_axes)

    def ncols(self):
        """ the number of total columns in the values axes """
        return sum(len(a.values) for a in self.values_axes)

    def is_transposed(self):
        return False

    def data_orientation(self):
        """return a tuple of my permutated axes, non_indexable at the front"""
        return tuple(itertools.chain([int(a[0]) for a in self.non_index_axes],
                                     [int(a.axis) for a in self.index_axes]))

    def queryables(self):
        """ return a dict of the kinds allowable columns for this object """

        # compute the values_axes queryables
        return dict(
            [(a.cname, a) for a in self.index_axes] +
            [(self.storage_obj_type._AXIS_NAMES[axis], None)
             for axis, values in self.non_index_axes] +
            [(v.cname, v) for v in self.values_axes
             if v.name in set(self.data_columns)]

    def index_cols(self):
        """ return a list of my index cols """
        return [(i.axis, i.cname) for i in self.index_axes]

    def values_cols(self):
        """ return a list of my values cols """
        return [i.cname for i in self.values_axes]

    def _get_metadata_path(self, key):
        """ return the metadata pathname for this key """
        return "{group}/meta/{key}/meta".format(group=self.group._v_pathname,

    def write_metadata(self, key, values):
        write out a meta data array to the key as a fixed-format Series

        key : string
        values : ndarray

        values = Series(values)
        self.parent.put(self._get_metadata_path(key), values, format='table',
                        encoding=self.encoding, errors=self.errors,

    def read_metadata(self, key):
        """ return the meta data array for this key """
        if getattr(getattr(self.group, 'meta', None), key, None) is not None:
            return self.parent.select(self._get_metadata_path(key))
        return None

    def set_info(self):
        """ update our table index info """
        self.attrs.info = self.info

    def set_attrs(self):
        """ set our table type & indexables """
        self.attrs.table_type = str(self.table_type)
        self.attrs.index_cols = self.index_cols()
        self.attrs.values_cols = self.values_cols()
        self.attrs.non_index_axes = self.non_index_axes
        self.attrs.data_columns = self.data_columns
        self.attrs.nan_rep = self.nan_rep
        self.attrs.encoding = self.encoding
        self.attrs.errors = self.errors
        self.attrs.levels = self.levels
        self.attrs.metadata = self.metadata

    def get_attrs(self):
        """ retrieve our attributes """
        self.non_index_axes = getattr(
            self.attrs, 'non_index_axes', None) or []
        self.data_columns = getattr(
            self.attrs, 'data_columns', None) or []
        self.info = getattr(
            self.attrs, 'info', None) or dict()
        self.nan_rep = getattr(self.attrs, 'nan_rep', None)
        self.encoding = _ensure_encoding(
            getattr(self.attrs, 'encoding', None))
        self.errors = _ensure_decoded(getattr(self.attrs, 'errors', 'strict'))
        self.levels = getattr(
            self.attrs, 'levels', None) or []
        self.index_axes = [
            a.infer(self) for a in self.indexables if a.is_an_indexable
        self.values_axes = [
            a.infer(self) for a in self.indexables if not a.is_an_indexable
        self.metadata = getattr(
            self.attrs, 'metadata', None) or []

    def validate_version(self, where=None):
        """ are we trying to operate on an old version? """
        if where is not None:
            if (self.version[0] <= 0 and self.version[1] <= 10 and
                    self.version[2] < 1):
                ws = incompatibility_doc % '.'.join(
                    [str(x) for x in self.version])
                warnings.warn(ws, IncompatibilityWarning)

    def validate_min_itemsize(self, min_itemsize):
        """validate the min_itemisze doesn't contain items that are not in the
        axes this needs data_columns to be defined
        if min_itemsize is None:
        if not isinstance(min_itemsize, dict):

        q = self.queryables()
        for k, v in min_itemsize.items():

            # ok, apply generally
            if k == 'values':
            if k not in q:
                raise ValueError(
                    "min_itemsize has the key [{key}] which is not an axis or "

    def indexables(self):
        """ create/cache the indexables if they don't exist """
        if self._indexables is None:

            self._indexables = []

            # index columns
                IndexCol(name=name, axis=axis, pos=i)
                for i, (axis, name) in enumerate(self.attrs.index_cols)

            # values columns
            dc = set(self.data_columns)
            base_pos = len(self._indexables)

            def f(i, c):
                klass = DataCol
                if c in dc:
                    klass = DataIndexableCol
                return klass.create_for_block(i=i, name=c, pos=base_pos + i,

                [f(i, c) for i, c in enumerate(self.attrs.values_cols)])

        return self._indexables

    def create_index(self, columns=None, optlevel=None, kind=None):
        Create a pytables index on the specified columns
          note: cannot index Time64Col() or ComplexCol currently;
          PyTables must be >= 3.0

        columns : False (don't create an index), True (create all columns
            index), None or list_like (the indexers to index)
        optlevel: optimization level (defaults to 6)
        kind    : kind of index (defaults to 'medium')

        raises if the node is not a table


        if not self.infer_axes():
        if columns is False:

        # index all indexables and data_columns
        if columns is None or columns is True:
            columns = [a.cname for a in self.axes if a.is_data_indexable]
        if not isinstance(columns, (tuple, list)):
            columns = [columns]

        kw = dict()
        if optlevel is not None:
            kw['optlevel'] = optlevel
        if kind is not None:
            kw['kind'] = kind

        table = self.table
        for c in columns:
            v = getattr(table.cols, c, None)
            if v is not None:

                # remove the index if the kind/optlevel have changed
                if v.is_indexed:
                    index = v.index
                    cur_optlevel = index.optlevel
                    cur_kind = index.kind

                    if kind is not None and cur_kind != kind:
                        kw['kind'] = cur_kind

                    if optlevel is not None and cur_optlevel != optlevel:
                        kw['optlevel'] = cur_optlevel

                # create the index
                if not v.is_indexed:
                    if v.type.startswith('complex'):
                        raise TypeError(
                            'Columns containing complex values can be stored '
                            'but cannot'
                            ' be indexed when using table format. Either use '
                            'fixed format, set index=False, or do not include '
                            'the columns containing complex values to '
                            'data_columns when initializing the table.')

    def read_axes(self, where, **kwargs):
        """create and return the axes sniffed from the table: return boolean
        for success

        # validate the version

        # infer the data kind
        if not self.infer_axes():
            return False

        # create the selection
        self.selection = Selection(self, where=where, **kwargs)
        values = self.selection.select()

        # convert the data
        for a in self.axes:
            a.convert(values, nan_rep=self.nan_rep, encoding=self.encoding,

        return True

    def get_object(self, obj):
        """ return the data for this obj """
        return obj

    def validate_data_columns(self, data_columns, min_itemsize):
        """take the input data_columns and min_itemize and create a data
        columns spec

        if not len(self.non_index_axes):
            return []

        axis, axis_labels = self.non_index_axes[0]
        info = self.info.get(axis, dict())
        if info.get('type') == 'MultiIndex' and data_columns:
            raise ValueError("cannot use a multi-index on axis [{0}] with "
                             "data_columns {1}".format(axis, data_columns))

        # evaluate the passed data_columns, True == use all columns
        # take only valide axis labels
        if data_columns is True:
            data_columns = list(axis_labels)
        elif data_columns is None:
            data_columns = []

        # if min_itemsize is a dict, add the keys (exclude 'values')
        if isinstance(min_itemsize, dict):

            existing_data_columns = set(data_columns)
                k for k in min_itemsize.keys()
                if k != 'values' and k not in existing_data_columns

        # return valid columns in the order of our axis
        return [c for c in data_columns if c in axis_labels]

    def create_axes(self, axes, obj, validate=True, nan_rep=None,
                    data_columns=None, min_itemsize=None, **kwargs):
        """ create and return the axes
        leagcy tables create an indexable column, indexable index,
        non-indexable fields

            axes: a list of the axes in order to create (names or numbers of
                the axes)
            obj : the object to create axes on
            validate: validate the obj against an existing object already
            min_itemsize: a dict of the min size for a column in bytes
            nan_rep : a values to use for string column nan_rep
            encoding : the encoding for string values
            data_columns : a list of columns that we want to create separate to
                allow indexing (or True will force all columns)


        # set the default axes if needed
        if axes is None:
                axes = _AXES_MAP[type(obj)]
            except KeyError:
                raise TypeError(
                    "cannot properly create the storer for: [group->{group},"
                        group=self.group._v_name, value=type(obj)))

        # map axes to numbers
        axes = [obj._get_axis_number(a) for a in axes]

        # do we have an existing table (if so, use its axes & data_columns)
        if self.infer_axes():
            existing_table = self.copy()
            axes = [a.axis for a in existing_table.index_axes]
            data_columns = existing_table.data_columns
            nan_rep = existing_table.nan_rep
            self.encoding = existing_table.encoding
            self.errors = existing_table.errors
            self.info = copy.copy(existing_table.info)
            existing_table = None

        # currently support on ndim-1 axes
        if len(axes) != self.ndim - 1:
            raise ValueError(
                "currently only support ndim-1 indexers in an AppendableTable")

        # create according to the new data
        self.non_index_axes = []
        self.data_columns = []

        # nan_representation
        if nan_rep is None:
            nan_rep = 'nan'

        self.nan_rep = nan_rep

        # create axes to index and non_index
        index_axes_map = dict()
        for i, a in enumerate(obj.axes):

            if i in axes:
                name = obj._AXIS_NAMES[i]
                index_axes_map[i] = _convert_index(
                    a, self.encoding, self.errors, self.format_type

                # we might be able to change the axes on the appending data if
                # necessary
                append_axis = list(a)
                if existing_table is not None:
                    indexer = len(self.non_index_axes)
                    exist_axis = existing_table.non_index_axes[indexer][1]
                    if not array_equivalent(np.array(append_axis),

                        # ahah! -> reindex
                        if array_equivalent(np.array(sorted(append_axis)),
                            append_axis = exist_axis

                # the non_index_axes info
                info = _get_info(self.info, i)
                info['names'] = list(a.names)
                info['type'] = a.__class__.__name__

                self.non_index_axes.append((i, append_axis))

        # set axis positions (based on the axes)
        self.index_axes = [
            for j, a in enumerate(axes)
        j = len(self.index_axes)

        # check for column conflicts
        for a in self.axes:

        # reindex by our non_index_axes & compute data_columns
        for a in self.non_index_axes:
            obj = _reindex_axis(obj, a[0], a[1])

        def get_blk_items(mgr, blocks):
            return [mgr.items.take(blk.mgr_locs) for blk in blocks]

        # figure out data_columns and get out blocks
        block_obj = self.get_object(obj)._consolidate()
        blocks = block_obj._data.blocks
        blk_items = get_blk_items(block_obj._data, blocks)
        if len(self.non_index_axes):
            axis, axis_labels = self.non_index_axes[0]
            data_columns = self.validate_data_columns(
                data_columns, min_itemsize)
            if len(data_columns):
                mgr = block_obj.reindex(

                blocks = list(mgr.blocks)
                blk_items = get_blk_items(mgr, blocks)
                for c in data_columns:
                    mgr = block_obj.reindex([c], axis=axis)._data
                    blk_items.extend(get_blk_items(mgr, mgr.blocks))

        # reorder the blocks in the same order as the existing_table if we can
        if existing_table is not None:
            by_items = {tuple(b_items.tolist()): (b, b_items)
                        for b, b_items in zip(blocks, blk_items)}
            new_blocks = []
            new_blk_items = []
            for ea in existing_table.values_axes:
                items = tuple(ea.values)
                    b, b_items = by_items.pop(items)
                except (IndexError, KeyError):
                    raise ValueError(
                        "cannot match existing table structure for [{items}] "
                        "on appending data".format(
                            items=(','.join(pprint_thing(item) for
                                            item in items))))
            blocks = new_blocks
            blk_items = new_blk_items

        # add my values
        self.values_axes = []
        for i, (b, b_items) in enumerate(zip(blocks, blk_items)):

            # shape of the data column are the indexable axes
            klass = DataCol
            name = None

            # we have a data_column
            if (data_columns and len(b_items) == 1 and
                    b_items[0] in data_columns):
                klass = DataIndexableCol
                name = b_items[0]

            # make sure that we match up the existing columns
            # if we have an existing table
            if existing_table is not None and validate:
                    existing_col = existing_table.values_axes[i]
                except (IndexError, KeyError):
                    raise ValueError(
                        "Incompatible appended table [{blocks}]"
                        "with existing table [{table}]".format(
                existing_col = None

                col = klass.create_for_block(
                    i=i, name=name, version=self.version)
                col.set_atom(block=b, block_items=b_items,

            except (NotImplementedError, ValueError, TypeError) as e:
                raise e
            except Exception as detail:
                raise Exception(
                    "cannot find the correct atom type -> "
                    "[dtype->{name},items->{items}] {detail!s}".format(
                        name=b.dtype.name, items=b_items, detail=detail))
            j += 1

        # validate our min_itemsize

        # validate our metadata

        # validate the axes if we have an existing table
        if validate:

    def process_axes(self, obj, columns=None):
        """ process axes filters """

        # make a copy to avoid side effects
        if columns is not None:
            columns = list(columns)

        # make sure to include levels if we have them
        if columns is not None and self.is_multi_index:
            for n in se