"""Utilities for fast persistence of big data, with optional compression."""

# Author: Gael Varoquaux <gael dot varoquaux at normalesup dot org>
# Copyright (c) 2009 Gael Varoquaux
# License: BSD Style, 3 clauses.

import pickle
import os
import sys
import warnings
try:
    from pathlib import Path
except ImportError:
    Path = None

from .compressor import lz4, LZ4_NOT_INSTALLED_ERROR
from .compressor import _COMPRESSORS, register_compressor, BinaryZlibFile
from .compressor import (ZlibCompressorWrapper, GzipCompressorWrapper,
                         BZ2CompressorWrapper, LZMACompressorWrapper,
                         XZCompressorWrapper, LZ4CompressorWrapper)
from .numpy_pickle_utils import Unpickler, Pickler
from .numpy_pickle_utils import _read_fileobject, _write_fileobject
from .numpy_pickle_utils import _read_bytes, BUFFER_SIZE
from .numpy_pickle_compat import load_compatibility
from .numpy_pickle_compat import NDArrayWrapper
# For compatibility with old versions of joblib, we need ZNDArrayWrapper
# to be visible in the current namespace.
# Explicitly skipping next line from flake8 as it triggers an F401 warning
# which we don't care.
from .numpy_pickle_compat import ZNDArrayWrapper  # noqa
from ._compat import _basestring, PY3_OR_LATER
from .backports import make_memmap

# Register supported compressors
register_compressor('zlib', ZlibCompressorWrapper())
register_compressor('gzip', GzipCompressorWrapper())
register_compressor('bz2', BZ2CompressorWrapper())
register_compressor('lzma', LZMACompressorWrapper())
register_compressor('xz', XZCompressorWrapper())
register_compressor('lz4', LZ4CompressorWrapper())

###############################################################################
# Utility objects for persistence.


class NumpyArrayWrapper(object):
    """An object to be persisted instead of numpy arrays.

    This object is used to hack into the pickle machinery and read numpy
    array data from our custom persistence format.
    More precisely, this object is used for:
    * carrying the information of the persisted array: subclass, shape, order,
    dtype. Those ndarray metadata are used to correctly reconstruct the array
    with low level numpy functions.
    * determining if memmap is allowed on the array.
    * reading the array bytes from a file.
    * reading the array using memorymap from a file.
    * writing the array bytes to a file.

    Attributes
    ----------
    subclass: numpy.ndarray subclass
        Determine the subclass of the wrapped array.
    shape: numpy.ndarray shape
        Determine the shape of the wrapped array.
    order: {'C', 'F'}
        Determine the order of wrapped array data. 'C' is for C order, 'F' is
        for fortran order.
    dtype: numpy.ndarray dtype
        Determine the data type of the wrapped array.
    allow_mmap: bool
        Determine if memory mapping is allowed on the wrapped array.
        Default: False.
    """

    def __init__(self, subclass, shape, order, dtype, allow_mmap=False):
        """Constructor. Store the useful information for later."""
        self.subclass = subclass
        self.shape = shape
        self.order = order
        self.dtype = dtype
        self.allow_mmap = allow_mmap

    def write_array(self, array, pickler):
        """Write array bytes to pickler file handle.

        This function is an adaptation of the numpy write_array function
        available in version 1.10.1 in numpy/lib/format.py.
        """
        # Set buffer size to 16 MiB to hide the Python loop overhead.
        buffersize = max(16 * 1024 ** 2 // array.itemsize, 1)
        if array.dtype.hasobject:
            # We contain Python objects so we cannot write out the data
            # directly. Instead, we will pickle it out with version 2 of the
            # pickle protocol.
            pickle.dump(array, pickler.file_handle, protocol=2)
        else:
            for chunk in pickler.np.nditer(array,
                                           flags=['external_loop',
                                                  'buffered',
                                                  'zerosize_ok'],
                                           buffersize=buffersize,
                                           order=self.order):
                pickler.file_handle.write(chunk.tostring('C'))

    def read_array(self, unpickler):
        """Read array from unpickler file handle.

        This function is an adaptation of the numpy read_array function
        available in version 1.10.1 in numpy/lib/format.py.
        """
        if len(self.shape) == 0:
            count = 1
        else:
            # joblib issue #859: we cast the elements of self.shape to int64 to
            # prevent a potential overflow when computing their product.
            shape_int64 = [unpickler.np.int64(x) for x in self.shape]
            count = unpickler.np.multiply.reduce(shape_int64)
        # Now read the actual data.
        if self.dtype.hasobject:
            # The array contained Python objects. We need to unpickle the data.
            array = pickle.load(unpickler.file_handle)
        else:
            if (not PY3_OR_LATER and
                    unpickler.np.compat.isfileobj(unpickler.file_handle)):
                # In python 2, gzip.GzipFile is considered as a file so one
                # can use numpy.fromfile().
                # For file objects, use np.fromfile function.
                # This function is faster than the memory-intensive
                # method below.
                array = unpickler.np.fromfile(unpickler.file_handle,
                                              dtype=self.dtype, count=count)
            else:
                # This is not a real file. We have to read it the
                # memory-intensive way.
                # crc32 module fails on reads greater than 2 ** 32 bytes,
                # breaking large reads from gzip streams. Chunk reads to
                # BUFFER_SIZE bytes to avoid issue and reduce memory overhead
                # of the read. In non-chunked case count < max_read_count, so
                # only one read is performed.
                max_read_count = BUFFER_SIZE // min(BUFFER_SIZE,
                                                    self.dtype.itemsize)

                array = unpickler.np.empty(count, dtype=self.dtype)
                for i in range(0, count, max_read_count):
                    read_count = min(max_read_count, count - i)
                    read_size = int(read_count * self.dtype.itemsize)
                    data = _read_bytes(unpickler.file_handle,
                                       read_size, "array data")
                    array[i:i + read_count] = \
                        unpickler.np.frombuffer(data, dtype=self.dtype,
                                                count=read_count)
                    del data

            if self.order == 'F':
                array.shape = self.shape[::-1]
                array = array.transpose()
            else:
                array.shape = self.shape

        return array

    def read_mmap(self, unpickler):
        """Read an array using numpy memmap."""
        offset = unpickler.file_handle.tell()
        if unpickler.mmap_mode == 'w+':
            unpickler.mmap_mode = 'r+'

        marray = make_memmap(unpickler.filename,
                             dtype=self.dtype,
                             shape=self.shape,
                             order=self.order,
                             mode=unpickler.mmap_mode,
                             offset=offset)
        # update the offset so that it corresponds to the end of the read array
        unpickler.file_handle.seek(offset + marray.nbytes)

        return marray

    def read(self, unpickler):
        """Read the array corresponding to this wrapper.

        Use the unpickler to get all information to correctly read the array.

        Parameters
        ----------
        unpickler: NumpyUnpickler

        Returns
        -------
        array: numpy.ndarray

        """
        # When requested, only use memmap mode if allowed.
        if unpickler.mmap_mode is not None and self.allow_mmap:
            array = self.read_mmap(unpickler)
        else:
            array = self.read_array(unpickler)

        # Manage array subclass case
        if (hasattr(array, '__array_prepare__') and
            self.subclass not in (unpickler.np.ndarray,
                                  unpickler.np.memmap)):
            # We need to reconstruct another subclass
            new_array = unpickler.np.core.multiarray._reconstruct(
                self.subclass, (0,), 'b')
            return new_array.__array_prepare__(array)
        else:
            return array

###############################################################################
# Pickler classes


class NumpyPickler(Pickler):
    """A pickler to persist big data efficiently.

    The main features of this object are:
    * persistence of numpy arrays in a single file.
    * optional compression with a special care on avoiding memory copies.

    Attributes
    ----------
    fp: file
        File object handle used for serializing the input object.
    protocol: int, optional
        Pickle protocol used. Default is pickle.DEFAULT_PROTOCOL under
        python 3, pickle.HIGHEST_PROTOCOL otherwise.
    """

    dispatch = Pickler.dispatch.copy()

    def __init__(self, fp, protocol=None):
        self.file_handle = fp
        self.buffered = isinstance(self.file_handle, BinaryZlibFile)

        # By default we want a pickle protocol that only changes with
        # the major python version and not the minor one
        if protocol is None:
            protocol = (pickle.DEFAULT_PROTOCOL if PY3_OR_LATER
                        else pickle.HIGHEST_PROTOCOL)

        Pickler.__init__(self, self.file_handle, protocol=protocol)
        # delayed import of numpy, to avoid tight coupling
        try:
            import numpy as np
        except ImportError:
            np = None
        self.np = np

    def _create_array_wrapper(self, array):
        """Create and returns a numpy array wrapper from a numpy array."""
        order = 'F' if (array.flags.f_contiguous and
                        not array.flags.c_contiguous) else 'C'
        allow_mmap = not self.buffered and not array.dtype.hasobject
        wrapper = NumpyArrayWrapper(type(array),
                                    array.shape, order, array.dtype,
                                    allow_mmap=allow_mmap)

        return wrapper

    def save(self, obj):
        """Subclass the Pickler `save` method.

        This is a total abuse of the Pickler class in order to use the numpy
        persistence function `save` instead of the default pickle
        implementation. The numpy array is replaced by a custom wrapper in the
        pickle persistence stack and the serialized array is written right
        after in the file. Warning: the file produced does not follow the
        pickle format. As such it can not be read with `pickle.load`.
        """
        if self.np is not None and type(obj) in (self.np.ndarray,
                                                 self.np.matrix,
                                                 self.np.memmap):
            if type(obj) is self.np.memmap:
                # Pickling doesn't work with memmapped arrays
                obj = self.np.asanyarray(obj)

            # The array wrapper is pickled instead of the real array.
            wrapper = self._create_array_wrapper(obj)
            Pickler.save(self, wrapper)

            # A framer was introduced with pickle protocol 4 and we want to
            # ensure the wrapper object is written before the numpy array
            # buffer in the pickle file.
            # See https://www.python.org/dev/peps/pep-3154/#framing to get
            # more information on the framer behavior.
            if self.proto >= 4:
                self.framer.commit_frame(force=True)

            # And then array bytes are written right after the wrapper.
            wrapper.write_array(obj, self)
            return

        return Pickler.save(self, obj)


class NumpyUnpickler(Unpickler):
    """A subclass of the Unpickler to unpickle our numpy pickles.

    Attributes
    ----------
    mmap_mode: str
        The memorymap mode to use for reading numpy arrays.
    file_handle: file_like
        File object to unpickle from.
    filename: str
        Name of the file to unpickle from. It should correspond to file_handle.
        This parameter is required when using mmap_mode.
    np: module
        Reference to numpy module if numpy is installed else None.

    """

    dispatch = Unpickler.dispatch.copy()

    def __init__(self, filename, file_handle, mmap_mode=None):
        # The next line is for backward compatibility with pickle generated
        # with joblib versions less than 0.10.
        self._dirname = os.path.dirname(filename)

        self.mmap_mode = mmap_mode
        self.file_handle = file_handle
        # filename is required for numpy mmap mode.
        self.filename = filename
        self.compat_mode = False
        Unpickler.__init__(self, self.file_handle)
        try:
            import numpy as np
        except ImportError:
            np = None
        self.np = np

    def load_build(self):
        """Called to set the state of a newly created object.

        We capture it to replace our place-holder objects, NDArrayWrapper or
        NumpyArrayWrapper, by the array we are interested in. We
        replace them directly in the stack of pickler.
        NDArrayWrapper is used for backward compatibility with joblib <= 0.9.
        """
        Unpickler.load_build(self)

        # For backward compatibility, we support NDArrayWrapper objects.
        if isinstance(self.stack[-1], (NDArrayWrapper, NumpyArrayWrapper)):