"""File formats.

This module implements the read(), iread() and write() functions in ase.io.
For each file format there is an IOFormat object.

There is a dict, ioformats, which stores the objects.

Example
=======

The xyz format is implemented in the ase/io/xyz.py file which has a
read_xyz() generator and a write_xyz() function.  This and other
information can be obtained from ioformats['xyz'].
"""

import io
import re
import functools
import inspect
import os
import sys
import numbers
import warnings
from pathlib import Path, PurePath
from typing import (
    IO, List, Any, Iterable, Tuple, Union, Sequence, Dict, Optional)

from cmmde_atoms import Atoms
from importlib import import_module
from cmmde_parallel import parallel_function, parallel_generator


PEEK_BYTES = 50000


class UnknownFileTypeError(Exception):
    pass


class IOFormat:
    def __init__(self, name: str, desc: str, code: str, module_name: str,
                 encoding: str = None) -> None:
        self.name = name
        self.description = desc
        assert len(code) == 2
        assert code[0] in list('+1')
        assert code[1] in list('BFS')
        self.code = code
        self.module_name = module_name
        self.encoding = encoding

        # (To be set by define_io_format())
        self.extensions: List[str] = []
        self.globs: List[str] = []
        self.magic: List[str] = []
        self.magic_regex: Optional[bytes] = None

    def open(self, fname, mode: str = 'r') -> IO:
        # We might want append mode, too
        # We can allow more flags as needed (buffering etc.)
        if mode not in list('rwa'):
            raise ValueError("Only modes allowed are 'r', 'w', and 'a'")
        if mode == 'r' and not self.can_read:
            raise NotImplementedError('No reader implemented for {} format'
                                      .format(self.name))
        if mode == 'w' and not self.can_write:
            raise NotImplementedError('No writer implemented for {} format'
                                      .format(self.name))
        if mode == 'a' and not self.can_append:
            raise NotImplementedError('Appending not supported by {} format'
                                      .format(self.name))

        if self.isbinary:
            mode += 'b'

        path = Path(fname)
        return path.open(mode, encoding=self.encoding)

    def _buf_as_filelike(self, data: Union[str, bytes]) -> IO:
        encoding = self.encoding
        if encoding is None:
            encoding = 'utf-8'  # Best hacky guess.

        if self.isbinary:
            if isinstance(data, str):
                data = data.encode(encoding)
        else:
            if isinstance(data, bytes):
                data = data.decode(encoding)

        return self._ioclass(data)

    @property
    def _ioclass(self):
        if self.isbinary:
            return io.BytesIO
        else:
            return io.StringIO

    def parse_images(self, data: Union[str, bytes],
                     **kwargs) -> Sequence[Atoms]:
        with self._buf_as_filelike(data) as fd:
            outputs = self.read(fd, **kwargs)
            if self.single:
                assert isinstance(outputs, Atoms)
                return [outputs]
            else:
                return list(self.read(fd, **kwargs))

    def parse_atoms(self, data: Union[str, bytes], **kwargs) -> Atoms:
        images = self.parse_images(data, **kwargs)
        return images[-1]

    @property
    def can_read(self) -> bool:
        return self._readfunc() is not None

    @property
    def can_write(self) -> bool:
        return self._writefunc() is not None

    @property
    def can_append(self) -> bool:
        writefunc = self._writefunc()
        return self.can_write and 'append' in writefunc.__code__.co_varnames

    def __repr__(self) -> str:
        tokens = ['{}={}'.format(name, repr(value))
                  for name, value in vars(self).items()]
        return 'IOFormat({})'.format(', '.join(tokens))

    def __getitem__(self, i):
        # For compatibility.
        #
        # Historically, the ioformats were listed as tuples
        # with (description, code).  We look like such a tuple.
        return (self.description, self.code)[i]

    @property
    def single(self) -> bool:
        """Whether this format is for a single Atoms object."""
        return self.code[0] == '1'

    @property
    def _formatname(self) -> str:
        return self.name.replace('-', '_')

    def _readfunc(self):
        return getattr(self.module, 'read_' + self._formatname, None)

    def _writefunc(self):
        return getattr(self.module, 'write_' + self._formatname, None)

    @property
    def read(self):
        if not self.can_read:
            self._warn_none('read')
            return None

        return self._read_wrapper

    def _read_wrapper(self, *args, **kwargs):
        function = self._readfunc()
        if function is None:
            self._warn_none('read')
            return None
        if not inspect.isgeneratorfunction(function):
            function = functools.partial(wrap_read_function, function)
        return function(*args, **kwargs)

    def _warn_none(self, action):
        msg = ('Accessing the IOFormat.{action} property on a format '
               'without {action} support will change behaviour in the '
               'future and return a callable instead of None.  '
               'Use IOFormat.can_{action} to check whether {action} '
               'is supported.')
        warnings.warn(msg.format(action=action), FutureWarning)

    @property
    def write(self):
        if not self.can_write:
            self._warn_none('write')
            return None

        return self._write_wrapper

    def _write_wrapper(self, *args, **kwargs):
        function = self._writefunc()
        if function is None:
            raise ValueError(f'Cannot write to {self.name}-format')
        return function(*args, **kwargs)

    @property
    def modes(self) -> str:
        modes = ''
        if self.can_read:
            modes += 'r'
        if self.can_write:
            modes += 'w'
        return modes

    def full_description(self) -> str:
        lines = [f'Name:        {self.name}',
                 f'Description: {self.description}',
                 f'Modes:       {self.modes}',
                 f'Encoding:    {self.encoding}',
                 f'Module:      {self.module_name}',
                 f'Code:        {self.code}',
                 f'Extensions:  {self.extensions}',
                 f'Globs:       {self.globs}',
                 f'Magic:       {self.magic}']
        return '\n'.join(lines)

    @property
    def acceptsfd(self) -> bool:
        return self.code[1] != 'S'

    @property
    def isbinary(self) -> bool:
        return self.code[1] == 'B'

    @property
    def module(self):
        if not self.module_name.startswith('ase.io.'):
            raise ValueError('Will only import modules from ase.io, '
                             'not {}'.format(self.module_name))
        try:
            return import_module(self.module_name)
        except ImportError as err:
            raise UnknownFileTypeError(
                f'File format not recognized: {self.name}.  Error: {err}')

    def match_name(self, basename: str) -> bool:
        from fnmatch import fnmatch
        return any(fnmatch(basename, pattern)
                   for pattern in self.globs)

    def match_magic(self, data: bytes) -> bool:
        if self.magic_regex:
            assert not self.magic, 'Define only one of magic and magic_regex'
            match = re.match(self.magic_regex, data, re.M | re.S)
            return match is not None

        from fnmatch import fnmatchcase
        return any(fnmatchcase(data, magic + b'*')  # type: ignore
                   for magic in self.magic)


ioformats: Dict[str, IOFormat] = {}  # These will be filled at run-time.
extension2format = {}


all_formats = ioformats  # Aliased for compatibility only.  Please do not use.
format2modulename = {}  # Left for compatibility only.


def define_io_format(name, desc, code, *, module=None, ext=None,
                     glob=None, magic=None, encoding=None,
                     magic_regex=None):
    if module is None:
        module = name.replace('-', '_')
        format2modulename[name] = module

    def normalize_patterns(strings):
        if strings is None:
            strings = []
        elif isinstance(strings, (str, bytes)):
            strings = [strings]
        else:
            strings = list(strings)
        return strings

    fmt = IOFormat(name, desc, code, module_name='ase.io.' + module,
                   encoding=encoding)
    fmt.extensions = normalize_patterns(ext)
    fmt.globs = normalize_patterns(glob)
    fmt.magic = normalize_patterns(magic)

    if magic_regex is not None:
        fmt.magic_regex = magic_regex

    for ext in fmt.extensions:
        if ext in extension2format:
            raise ValueError('extension "{}" already registered'.format(ext))
        extension2format[ext] = fmt

    ioformats[name] = fmt
    return fmt


def get_ioformat(name: str) -> IOFormat:
    """Return ioformat object or raise appropriate error."""
    if name not in ioformats:
        raise UnknownFileTypeError(name)
    fmt = ioformats[name]
    # Make sure module is importable, since this could also raise an error.
    fmt.module
    return ioformats[name]


# We define all the IO formats below.  Each IO format has a code,
# such as '1F', which defines some of the format's properties:
#
# 1=single atoms object
# +=multiple atoms objects
# F=accepts a file-descriptor
# S=needs a file-name str
# B=like F, but opens in binary mode

F = define_io_format
F('abinit-in', 'ABINIT input file', '1F',
  module='abinit', magic=b'*znucl *')
F('abinit-out', 'ABINIT output file', '1F',
  module='abinit', magic=b'*.Version * of ABINIT')
F('aims', 'FHI-aims geometry file', '1S', ext='in')
F('aims-output', 'FHI-aims output', '+S',
  module='aims', magic=b'*Invoking FHI-aims ...')
F('bundletrajectory', 'ASE bundle trajectory', '+S')
F('castep-castep', 'CASTEP output file', '+F',
  module='castep', ext='castep')
F('castep-cell', 'CASTEP geom file', '1F',
  module='castep', ext='cell')
F('castep-geom', 'CASTEP trajectory file', '+F',
  module='castep', ext='geom')
F('castep-md', 'CASTEP molecular dynamics file', '+F',
  module='castep', ext='md')
F('castep-phonon', 'CASTEP phonon file', '1F',
  module='castep', ext='phonon')
F('cfg', 'AtomEye configuration', '1F')
F('cif', 'CIF-file', '+B', ext='cif')
F('cmdft', 'CMDFT-file', '1F', glob='*I_info')
F('cml', 'Chemical json file', '1F', ext='cml')
F('cp2k-dcd', 'CP2K DCD file', '+B',
  module='cp2k', ext='dcd')
F('cp2k-restart', 'CP2K restart file', '1F',
  module='cp2k', ext='restart')
F('crystal', 'Crystal fort.34 format', '1F',
  ext=['f34', '34'], glob=['f34', '34'])
F('cube', 'CUBE file', '1F', ext='cube')
F('dacapo-text', 'Dacapo text output', '1F',
  module='dacapo', magic=b'*&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&\n')
F('db', 'ASE SQLite database file', '+S')
F('dftb', 'DftbPlus input file', '1S', magic=b'Geometry')
F('dlp4', 'DL_POLY_4 CONFIG file', '1F',
  module='dlp4', ext='config', glob=['*CONFIG*'])
F('dlp-history', 'DL_POLY HISTORY file', '+F',
  module='dlp4', glob='HISTORY')
F('dmol-arc', 'DMol3 arc file', '+S',
  module='dmol')
F('dmol-car', 'DMol3 structure file', '1S',
  module='dmol', ext='car')
F('dmol-incoor', 'DMol3 structure file', '1S',
  module='dmol')
F('elk', 'ELK atoms definition from GEOMETRY.OUT', '1F',
  glob=['GEOMETRY.OUT'])
F('elk-in', 'ELK input file', '1F', module='elk')
F('eon', 'EON CON file', '+F',
  ext='con')
F('eps', 'Encapsulated Postscript', '1S')
F('espresso-in', 'Quantum espresso in file', '1F',
  module='espresso', ext='pwi', magic=[b'*\n&system', b'*\n&SYSTEM'])
F('espresso-out', 'Quantum espresso out file', '+F',
  module='espresso', ext=['out', 'pwo'], magic=b'*Program PWSCF')
F('exciting', 'exciting input', '1F', glob='input.xml')
F('extxyz', 'Extended XYZ file', '+F', ext='xyz')
F('findsym', 'FINDSYM-format', '+F')
F('gamess-us-out', 'GAMESS-US output file', '1F',
  module='gamess_us', magic=b'*GAMESS')
F('gamess-us-in', 'GAMESS-US input file', '1F',
  module='gamess_us')
F('gamess-us-punch', 'GAMESS-US punchcard file', '1F',
  module='gamess_us', magic=b' $DATA', ext='dat')
F('gaussian-in', 'Gaussian com (input) file', '1F',
  module='gaussian', ext=['com', 'gjf'])
F('gaussian-out', 'Gaussian output file', '+F',
  module='gaussian', ext='log', magic=b'*Entering Gaussian System')
F('acemolecule-out', 'ACE output file', '1S',
  module='acemolecule')
F('acemolecule-input', 'ACE input file', '1S',
  module='acemolecule')
F('gen', 'DFTBPlus GEN format', '1F')
F('gif', 'Graphics interchange format', '+S',
  module='animation')
F('gpaw-out', 'GPAW text output', '+F',
  magic=b'*  ___ ___ ___ _ _ _')
F('gpumd', 'GPUMD input file', '1F', glob='xyz.in')
F('gpw', 'GPAW restart-file', '1S',
  magic=[b'- of UlmGPAW', b'AFFormatGPAW'])
F('gromacs', 'Gromacs coordinates', '1F',
  ext='gro')
F('gromos', 'Gromos96 geometry file', '1F', ext='g96')
F('html', 'X3DOM HTML', '1F', module='x3d')
F('json', 'ASE JSON database file', '+F', ext='json', module='db')
F('jsv', 'JSV file format', '1F')
F('lammps-dump-text', 'LAMMPS text dump file', '+F',
  module='lammpsrun', magic_regex=b'.*?^ITEM: TIMESTEP$')
F('lammps-dump-binary', 'LAMMPS binary dump file', '+B',
  module='lammpsrun')
F('lammps-data', 'LAMMPS data file', '1F', module='lammpsdata',
  encoding='ascii')
F('magres', 'MAGRES ab initio NMR data file', '1F')
F('mol', 'MDL Molfile', '1F')
F('mp4', 'MP4 animation', '+S',
  module='animation')
F('mustem', 'muSTEM xtl file', '1F',
  ext='xtl')
F('mysql', 'ASE MySQL database file', '+S',
  module='db')
F('netcdftrajectory', 'AMBER NetCDF trajectory file', '+S',
  magic=b'CDF')
F('nomad-json', 'JSON from Nomad archive', '+F',
  ext='nomad-json')
F('nwchem-in', 'NWChem input file', '1F',
  module='nwchem', ext='nwi')
F('nwchem-out', 'NWChem output file', '+F',
  module='nwchem', ext='nwo',
  magic=b'*Northwest Computational Chemistry Package')
F('octopus-in', 'Octopus input file', '1F',
  module='octopus', glob='inp')
F('proteindatabank', 'Protein Data Bank', '+F',
  ext='pdb')
F('png', 'Portable Network Graphics', '1B')
F('postgresql', 'ASE PostgreSQL database file', '+S', module='db')
F('pov', 'Persistance of Vision', '1S')
# prismatic: Should have ext='xyz' if/when multiple formats can have the same
# extension
F('prismatic', 'prismatic and computem XYZ-file', '1F')
F('py', 'Python file', '+F')
F('sys', 'qball sys file', '1F')
F('qbox', 'QBOX output file', '+F',
  magic=b'*:simulation xmlns:')
F('res', 'SHELX format', '1S', ext='shelx')
F('rmc6f', 'RMCProfile', '1S', ext='rmc6f')
F('sdf', 'SDF format', '1F')
F('siesta-xv', 'Siesta .XV file', '1F',
  glob='*.XV', module='siesta')
F('struct', 'WIEN2k structure file', '1S', module='wien2k')
F('struct_out', 'SIESTA STRUCT file', '1F', module='siesta')
F('traj', 'ASE trajectory', '+B', module='trajectory', ext='traj',
  magic=[b'- of UlmASE-Trajectory', b'AFFormatASE-Trajectory'])
F('turbomole', 'TURBOMOLE coord file', '1F', glob='coord',
  magic=b'$coord')
F('turbomole-gradient', 'TURBOMOLE gradient file', '+F',
  module='turbomole', glob='gradient', magic=b'$grad')
F('v-sim', 'V_Sim ascii file', '1F', ext='ascii')
F('vasp', 'VASP POSCAR/CONTCAR', '1F',
  ext='poscar', glob=['*POSCAR*', '*CONTCAR*'])
F('vasp-out', 'VASP OUTCAR file', '+F',
  module='vasp', glob='*OUTCAR*')
F('vasp-xdatcar', 'VASP XDATCAR file', '+F',
  module='vasp', glob='*XDATCAR*')
F('vasp-xml', 'VASP vasprun.xml file', '+F',
  module='vasp', glob='*vasp*.xml')
F('vti', 'VTK XML Image Data', '1F', module='vtkxml')
F('vtu', 'VTK XML Unstructured Grid', '1F', module='vtkxml', ext='vtu')
F('wout', 'Wannier90 output', '1F', module='wannier90')
F('x3d', 'X3D', '1S')
F('xsd', 'Materials Studio file', '1F')
F('xsf', 'XCrySDen Structure File', '+F',
  magic=[b'*\nANIMSTEPS', b'*\nCRYSTAL', b'*\nSLAB', b'*\nPOLYMER',
         b'*\nMOLECULE', b'*\nATOMS'])
F('xtd', 'Materials Studio file', '+F')
# xyz: No `ext='xyz'` in the definition below.
#      The .xyz files are handled by the extxyz module by default.
F('xyz', 'XYZ-file', '+F')


def get_compression(filename: str) -> Tuple[str, Optional[str]]:
    """
    Parse any expected file compression from the extension of a filename.
    Return the filename without the extension, and the extension. Recognises
    ``.gz``, ``.bz2``, ``.xz``.

    >>> get_compression('H2O.pdb.gz')
    ('H2O.pdb', 'gz')
    >>> get_compression('crystal.cif')
    ('crystal.cif', None)

    Parameters
    ==========
    filename: str
        Full filename including extension.

    Returns
    =======
    (root, extension): (str, str or None)
        Filename split into root without extension, and the extension
        indicating compression format. Will not split if compression
        is not recognised.
    """
    # Update if anything is added
    valid_compression = ['gz', 'bz2', 'xz']

    # Use stdlib as it handles most edge cases
    root, compression = os.path.splitext(filename)

    # extension keeps the '.' so remember to remove it
    if compression.strip('.') in valid_compression:
        return root, compression.strip('.')
    else:
        return filename, None


def open_with_compression(filename: str, mode: str = 'r') -> IO:
    """
    Wrapper around builtin `open` that will guess compression of a file
    from the filename and open it for reading or writing as if it were
    a standard file.

    Implemented for ``gz``(gzip), ``bz2``(bzip2) and ``xz``(lzma).

    Supported modes are:
       * 'r', 'rt', 'w', 'wt' for text mode read and write.
       * 'rb, 'wb' for binary read and write.

    Parameters
    ==========
    filename: str
        Path to the file to open, including any extensions that indicate
        the compression used.
    mode: str
        Mode to open the file, same as for builtin ``open``, e.g 'r', 'w'.

    Returns
    =======
    fd: file
        File-like object open with the specified mode.
    """

    # Compressed formats sometimes default to binary, so force text mode.
    if mode == 'r':
        mode = 'rt'
    elif mode == 'w':
        mode = 'wt'
    elif mode == 'a':
        mode = 'at'

    root, compression = get_compression(filename)

    if compression == 'gz':
        import gzip
        return gzip.open(filename, mode=mode)  # type: ignore
    elif compression == 'bz2':
        import bz2
        return bz2.open(filename, mode=mode)
    elif compression == 'xz':
        import lzma
        return lzma.open(filename, mode)
    else:
        # Either None or unknown string
        return open(filename, mode)


def wrap_read_function(read, filename, index=None, **kwargs):
    """Convert read-function to generator."""
    if index is None:
        yield read(filename, **kwargs)
    else:
        for atoms in read(filename, index, **kwargs):
            yield atoms


NameOrFile = Union[str, PurePath, IO]


def write(
        filename: NameOrFile,
        images: Union[Atoms, Sequence[Atoms]],
        format: str = None,
        parallel: bool = True,
        append: bool = False,
        **kwargs: dict
) -> None:
    """Write Atoms object(s) to file.

    filename: str or file
        Name of the file to write to or a file descriptor.  The name '-'
        means standard output.
    images: Atoms object or list of Atoms objects
        A single Atoms object or a list of Atoms objects.
    format: str
        Used to specify the file-format.  If not given, the
        file-format will be taken from suffix of the filename.
    parallel: bool
        Default is to write on master only.  Use parallel=False to write
        from all slaves.
    append: bool
        Default is to open files in 'w' or 'wb' mode, overwriting
        existing files.  In some cases opening the file in 'a' or 'ab'
        mode (appending) is useful,
        e.g. writing trajectories or saving multiple Atoms objects in one file.
        WARNING: If the file format does not support multiple entries without
        additional keywords/headers, files created using 'append=True'
        might not be readable by any program! They will nevertheless be
        written without error message.

    The use of additional keywords is format specific. write() may
    return an object after writing certain formats, but this behaviour
    may change in the future.

    """

    if isinstance(filename, PurePath):
        filename = str(filename)

    if isinstance(filename, str):
        fd = None
        if filename == '-':
            fd = sys.stdout
            filename = None  # type: ignore
        elif format is None:
            format = filetype(filename, read=False)
            assert isinstance(format, str)
    else:
        fd = filename  # type: ignore
        if format is None:
            try:
                format = filetype(filename, read=False)
                assert isinstance(format, str)
            except UnknownFileTypeError:
                format = None
        filename = None  # type: ignore

    format = format or 'json'  # default is json

    io = get_ioformat(format)

    return _write(filename, fd, format, io, images,
                  parallel=parallel, append=append, **kwargs)


@parallel_function
def _write(filename, fd, format, io, images, parallel=None, append=False,
           **kwargs):
    if isinstance(images, Atoms):
        images = [images]

    if io.single:
        if len(images) > 1:
            raise ValueError('{}-format can only store 1 Atoms object.'
                             .format(format))
        images = images[0]

    if not io.can_write:
        raise ValueError("Can't write to {}-format".format(format))

    # Special case for json-format:
    if format == 'json' and (len(images) > 1 or append):
        if filename is not None:
            return io.write(filename, images, append=append, **kwargs)
        raise ValueError("Can't write more than one image to file-descriptor "
                         'using json-format.')

    if io.acceptsfd:
        open_new = (fd is None)
        try:
            if open_new:
                mode = 'wb' if io.isbinary else 'w'
                if append:
                    mode = mode.replace('w', 'a')
                fd = open_with_compression(filename, mode)
                # XXX remember to re-enable compressed open
                # fd = io.open(filename, mode)
            return io.write(fd, images, **kwargs)
        finally:
            if open_new and fd is not None:
                fd.close()
    else:
        if fd is not None:
            raise ValueError("Can't write {}-format to file-descriptor"
                             .format(format))
        if io.can_append:
            return io.write(filename, images, append=append, **kwargs)
        elif append:
            raise ValueError("Cannot append to {}-format, write-function "
                             "does not support the append keyword."
                             .format(format))
        else:
            return io.write(filename, images, **kwargs)


def read(
        filename: NameOrFile,
        index: Any = None,
        format: str = None,
        parallel: bool = True,
        do_not_split_by_at_sign: bool = False,
        **kwargs
) -> Union[Atoms, List[Atoms]]:
    """Read Atoms object(s) from file.

    filename: str or file
        Name of the file to read from or a file descriptor.
    index: int, slice or str
        The last configuration will be returned by default.  Examples:

            * ``index=0``: first configuration
            * ``index=-2``: second to last
            * ``index=':'`` or ``index=slice(None)``: all
            * ``index='-3:'`` or ``index=slice(-3, None)``: three last
            * ``index='::2'`` or ``index=slice(0, None, 2)``: even
            * ``index='1::2'`` or ``index=slice(1, None, 2)``: odd
    format: str
        Used to specify the file-format.  If not given, the
        file-format will be guessed by the *filetype* function.
    parallel: bool
        Default is to read on master and broadcast to slaves.  Use
        parallel=False to read on all slaves.
    do_not_split_by_at_sign: bool
        If False (default) ``filename`` is splited by at sign ``@``

    Many formats allow on open file-like object to be passed instead
    of ``filename``. In this case the format cannot be auto-decected,
    so the ``format`` argument should be explicitly given."""

    if isinstance(filename, PurePath):
        filename = str(filename)
    if filename == '-':
        filename = sys.stdin
    if isinstance(index, str):
        try:
            index = string2index(index)
        except ValueError:
            pass

    filename, index = parse_filename(filename, index, do_not_split_by_at_sign)
    if index is None:
        index = -1
    format = format or filetype(filename, read=isinstance(filename, str))

    io = get_ioformat(format)
    if isinstance(index, (slice, str)):
        return list(_iread(filename, index, format, io, parallel=parallel,
                           **kwargs))
    else:
        return next(_iread(filename, slice(index, None), format, io,
                           parallel=parallel, **kwargs))


def iread(
        filename: NameOrFile,
        index: Any = None,
        format: str = None,
        parallel: bool = True,
        do_not_split_by_at_sign: bool = False,
        **kwargs
) -> Iterable[Atoms]:
    """Iterator for reading Atoms objects from file.

    Works as the `read` function, but yields one Atoms object at a time
    instead of all at once."""

    if isinstance(filename, PurePath):
        filename = str(filename)

    if isinstance(index, str):
        index = string2index(index)

    filename, index = parse_filename(filename, index, do_not_split_by_at_sign)

    if index is None or index == ':':
        index = slice(None, None, None)

    if not isinstance(index, (slice, str)):
        index = slice(index, (index + 1) or None)

    format = format or filetype(filename, read=isinstance(filename, str))
    io = get_ioformat(format)

    for atoms in _iread(filename, index, format, io, parallel=parallel,
                        **kwargs):
        yield atoms


@parallel_generator
def _iread(filename, index, format, io, parallel=None, full_output=False,
           **kwargs):

    if not io.can_read:
        raise ValueError("Can't read from {}-format".format(format))

    if io.single:
        start = index.start
        assert start is None or start == 0 or start == -1
        args = ()
    else:
        args = (index,)

    must_close_fd = False
    if isinstance(filename, str):
        if io.acceptsfd:
            mode = 'rb' if io.isbinary else 'r'
            fd = open_with_compression(filename, mode)
            must_close_fd = True
        else:
            fd = filename
    else:
        assert io.acceptsfd
        fd = filename

    # Make sure fd is closed in case loop doesn't finish:
    try:
        for dct in io.read(fd, *args, **kwargs):
            if not isinstance(dct, dict):
                dct = {'atoms': dct}
            if full_output:
                yield dct
            else:
                yield dct['atoms']
    finally:
        if must_close_fd:
            fd.close()


def parse_filename(filename, index=None, do_not_split_by_at_sign=False):
    if not isinstance(filename, str):
        return filename, index

    basename = os.path.basename(filename)
    if do_not_split_by_at_sign or '@' not in basename:
        return filename, index

    newindex = None
    newfilename, newindex = filename.rsplit('@', 1)

    if isinstance(index, slice):
        return newfilename, index
    try:
        newindex = string2index(newindex)
    except ValueError:
        warnings.warn('Can not parse index for path \n'
                      ' "%s" \nConsider set '
                      'do_not_split_by_at_sign=True \nif '
                      'there is no index.' % filename)
    return newfilename, newindex


def match_magic(data: bytes) -> IOFormat:
    data = data[:PEEK_BYTES]
    for ioformat in ioformats.values():
        if ioformat.match_magic(data):
            return ioformat
    raise UnknownFileTypeError('Cannot guess file type from contents')


def string2index(string: str) -> Union[int, slice, str]:
    """Convert index string to either int or slice"""
    if ':' not in string:
        # may contain database accessor
        try:
            return int(string)
        except ValueError:
            return string
    i: List[Optional[int]] = []
    for s in string.split(':'):
        if s == '':
            i.append(None)
        else:
            i.append(int(s))
    i += (3 - len(i)) * [None]
    return slice(*i)


def filetype(
        filename: NameOrFile,
        read: bool = True,
        guess: bool = True,
) -> str:
    """Try to guess the type of the file.

    First, special signatures in the filename will be checked for.  If that
    does not identify the file type, then the first 2000 bytes of the file
    will be read and analysed.  Turn off this second part by using
    read=False.

    Can be used from the command-line also::

        $ ase info filename ...
    """

    orig_filename = filename
    if hasattr(filename, 'name'):
        filename = filename.name  # type: ignore

    ext = None
    if isinstance(filename, str):
        if os.path.isdir(filename):
            if os.path.basename(os.path.normpath(filename)) == 'states':
                return 'eon'
            return 'bundletrajectory'

        if filename.startswith('postgres'):
            return 'postgresql'

        if filename.startswith('mysql') or filename.startswith('mariadb'):
            return 'mysql'

        # strip any compression extensions that can be read
        root, compression = get_compression(filename)
        basename = os.path.basename(root)

        if '.' in basename:
            ext = os.path.splitext(basename)[1].strip('.').lower()

        for fmt in ioformats.values():
            if fmt.match_name(basename):
                return fmt.name

        if not read:
            if ext is None:
                raise UnknownFileTypeError('Could not guess file type')
            ioformat = extension2format.get(ext)
            if ioformat:
                return ioformat.name

            # askhl: This is strange, we don't know if ext is a format:
            return ext

        if orig_filename == filename:
            fd = open_with_compression(filename, 'rb')
        else:
            fd = orig_filename  # type: ignore
    else:
        fd = filename    # type: ignore
        if fd is sys.stdin:
            return 'json'

    data = fd.read(PEEK_BYTES)
    if fd is not filename:
        fd.close()
    else:
        fd.seek(0)

    if len(data) == 0:
        raise UnknownFileTypeError('Empty file: ' + filename)    # type: ignore

    try:
        return match_magic(data).name
    except UnknownFileTypeError:
        pass

    format = None
    if ext in extension2format:
        format = extension2format[ext].name

    if format is None and guess:
        format = ext
    if format is None:
        # Do quick xyz check:
        lines = data.splitlines()
        if lines and lines[0].strip().isdigit():
            return extension2format['xyz'].name

        raise UnknownFileTypeError('Could not guess file type')
    assert isinstance(format, str)
    return format


def index2range(index, length):
    """Convert slice or integer to range.

    If index is an integer, range will contain only that integer."""
    obj = range(length)[index]
    if isinstance(obj, numbers.Integral):
        obj = range(obj, obj + 1)
    return obj