Difficult-Rocket/libs/pyglet/image/__init__.py

"""Image load, capture and high-level texture functions.

Only basic functionality is described here; for full reference see the
accompanying documentation.

To load an image::

    from pyglet import image
    pic = image.load('picture.png')

The supported image file types include PNG, BMP, GIF, JPG, and many more,
somewhat depending on the operating system.  To load an image from a file-like
object instead of a filename::

    pic = image.load('hint.jpg', file=fileobj)

The hint helps the module locate an appropriate decoder to use based on the
file extension.  It is optional.

Once loaded, images can be used directly by most other modules of pyglet.  All
images have a width and height you can access::

    width, height = pic.width, pic.height

You can extract a region of an image (this keeps the original image intact;
the memory is shared efficiently)::

    subimage = pic.get_region(x, y, width, height)

Remember that y-coordinates are always increasing upwards.

Drawing images
--------------

To draw an image at some point on the screen::

    pic.blit(x, y, z)

This assumes an appropriate view transform and projection have been applied.

Some images have an intrinsic "anchor point": this is the point which will be
aligned to the ``x`` and ``y`` coordinates when the image is drawn.  By
default the anchor point is the lower-left corner of the image.  You can use
the anchor point to center an image at a given point, for example::

    pic.anchor_x = pic.width // 2
    pic.anchor_y = pic.height // 2
    pic.blit(x, y, z)

Texture access
--------------

If you are using OpenGL directly, you can access the image as a texture::

    texture = pic.get_texture()

(This is the most efficient way to obtain a texture; some images are
immediately loaded as textures, whereas others go through an intermediate
form).  To use a texture with pyglet.gl::

    from pyglet.gl import *
    glEnable(texture.target)        # typically target is GL_TEXTURE_2D
    glBindTexture(texture.target, texture.id)
    # ... draw with the texture

Pixel access
------------

To access raw pixel data of an image::

    rawimage = pic.get_image_data()

(If the image has just been loaded this will be a very quick operation;
however if the image is a texture a relatively expensive readback operation
will occur).  The pixels can be accessed as a string::

    format = 'RGBA'
    pitch = rawimage.width * len(format)
    pixels = rawimage.get_data(format, pitch)

"format" strings consist of characters that give the byte order of each color
component.  For example, if rawimage.format is 'RGBA', there are four color
components: red, green, blue and alpha, in that order.  Other common format
strings are 'RGB', 'LA' (luminance, alpha) and 'I' (intensity).

The "pitch" of an image is the number of bytes in a row (this may validly be
more than the number required to make up the width of the image, it is common
to see this for word alignment).  If "pitch" is negative the rows of the image
are ordered from top to bottom, otherwise they are ordered from bottom to top.

Retrieving data with the format and pitch given in `ImageData.format` and
`ImageData.pitch` avoids the need for data conversion (assuming you can make
use of the data in this arbitrary format).

"""
import re
import weakref

from ctypes import *
from io import open, BytesIO

import pyglet

from pyglet.gl import *
from pyglet.gl import gl_info
from pyglet.util import asbytes

from .codecs import ImageEncodeException, ImageDecodeException
from .codecs import registry as _codec_registry
from .codecs import add_default_codecs as _add_default_codecs

from .animation import Animation, AnimationFrame
from .buffer import *
from . import atlas


class ImageException(Exception):
    pass


class TextureArraySizeExceeded(Exception):
    """Exception occurs ImageData dimensions are larger than the array supports."""
    pass


class TextureArrayDepthExceeded(Exception):
    """Exception occurs when depth has hit the maximum supported of the array."""
    pass


def load(filename, file=None, decoder=None):
    """Load an image from a file.

    :Parameters:
        `filename` : str
            Used to guess the image format, and to load the file if `file` is
            unspecified.
        `file` : file-like object or None
            Source of image data in any supported format.
        `decoder` : ImageDecoder or None
            If unspecified, all decoders that are registered for the filename
            extension are tried.  If none succeed, the exception from the
            first decoder is raised.

    :rtype: AbstractImage

    .. note:: You can make no assumptions about the return type; usually it will
        be ImageData or CompressedImageData, but decoders are free to return
        any subclass of AbstractImage.

    """
    if decoder:
        return decoder.decode(filename, file)
    else:
        return _codec_registry.decode(filename, file)


def load_animation(filename, file=None, decoder=None):
    """Load an animation from a file.

    Currently, the only supported format is GIF.

    :Parameters:
        `filename` : str
            Used to guess the animation format, and to load the file if `file`
            is unspecified.
        `file` : file-like object or None
            File object containing the animation stream.
        `decoder` : ImageDecoder or None
            If unspecified, all decoders that are registered for the filename
            extension are tried.  If none succeed, the exception from the
            first decoder is raised.

    :rtype: Animation
    """
    if decoder:
        return decoder.decode_animation(filename, file)
    else:
        return _codec_registry.decode_animation(filename, file)


def create(width, height, pattern=None):
    """Create an image optionally filled with the given pattern.

    :Parameters:
        `width` : int
            Width of image to create
        `height` : int
            Height of image to create
        `pattern` : ImagePattern or None
            Pattern to fill image with.  If unspecified, the image will
            initially be transparent.

    :rtype: AbstractImage

    .. note:: You can make no assumptions about the return type; usually it will
              be ImageData or CompressedImageData, but patterns are free to return
              any subclass of AbstractImage.
    """
    if not pattern:
        pattern = SolidColorImagePattern()
    return pattern.create_image(width, height)


def get_max_texture_size():
    """Query the maximum texture size available"""
    size = c_int()
    glGetIntegerv(GL_MAX_TEXTURE_SIZE, size)
    return size.value


def get_max_array_texture_layers():
    """Query the maximum TextureArray depth"""
    max_layers = c_int()
    glGetIntegerv(GL_MAX_ARRAY_TEXTURE_LAYERS, max_layers)
    return max_layers.value


def _color_as_bytes(color):
    if len(color) != 4:
        raise TypeError("color is expected to have 4 components")
    return bytes(color)


class ImagePattern:
    """Abstract image creation class."""

    def create_image(self, width, height):
        """Create an image of the given size.

        :Parameters:
            `width` : int
                Width of image to create
            `height` : int
                Height of image to create

        :rtype: AbstractImage
        """
        raise NotImplementedError('abstract')


class SolidColorImagePattern(ImagePattern):
    """Creates an image filled with a solid color."""

    def __init__(self, color=(0, 0, 0, 0)):
        """Create a solid image pattern with the given color.

        :Parameters:
            `color` : (int, int, int, int)
                4-tuple of ints in range [0,255] giving RGBA components of
                color to fill with.

        """
        self.color = _color_as_bytes(color)

    def create_image(self, width, height):
        data = self.color * width * height
        return ImageData(width, height, 'RGBA', data)


class CheckerImagePattern(ImagePattern):
    """Create an image with a tileable checker image.
    """

    def __init__(self, color1=(150, 150, 150, 255), color2=(200, 200, 200, 255)):
        """Initialise with the given colors.

        :Parameters:
            `color1` : (int, int, int, int)
                4-tuple of ints in range [0,255] giving RGBA components of
                color to fill with.  This color appears in the top-left and
                bottom-right corners of the image.
            `color2` : (int, int, int, int)
                4-tuple of ints in range [0,255] giving RGBA components of
                color to fill with.  This color appears in the top-right and
                bottom-left corners of the image.

        """
        self.color1 = _color_as_bytes(color1)
        self.color2 = _color_as_bytes(color2)

    def create_image(self, width, height):
        hw = width // 2
        hh = height // 2
        row1 = self.color1 * hw + self.color2 * hw
        row2 = self.color2 * hw + self.color1 * hw
        data = row1 * hh + row2 * hh
        return ImageData(width, height, 'RGBA', data)


class AbstractImage:
    """Abstract class representing an image.

    :Parameters:
        `width` : int
            Width of image
        `height` : int
            Height of image
        `anchor_x` : int
            X coordinate of anchor, relative to left edge of image data
        `anchor_y` : int
            Y coordinate of anchor, relative to bottom edge of image data
    """
    anchor_x = 0
    anchor_y = 0

    def __init__(self, width, height):
        self.width = width
        self.height = height

    def __repr__(self):
        return "{}(size={}x{})".format(self.__class__.__name__, self.width, self.height)

    def get_image_data(self):
        """Get an ImageData view of this image.

        Changes to the returned instance may or may not be reflected in this
        image.

        :rtype: :py:class:`~pyglet.image.ImageData`

        .. versionadded:: 1.1
        """
        raise ImageException('Cannot retrieve image data for %r' % self)

    def get_texture(self, rectangle=False):
        """A :py:class:`~pyglet.image.Texture` view of this image.

        :Parameters:
            `rectangle` : bool
                Unused. Kept for compatibility.

                .. versionadded:: 1.1.4.
        :rtype: :py:class:`~pyglet.image.Texture`

        .. versionadded:: 1.1
        """
        raise ImageException('Cannot retrieve texture for %r' % self)

    def get_mipmapped_texture(self):
        """Retrieve a :py:class:`~pyglet.image.Texture` instance with all mipmap levels filled in.

        :rtype: :py:class:`~pyglet.image.Texture`

        .. versionadded:: 1.1
        """
        raise ImageException('Cannot retrieve mipmapped texture for %r' % self)

    def get_region(self, x, y, width, height):
        """Retrieve a rectangular region of this image.

        :Parameters:
            `x` : int
                Left edge of region.
            `y` : int
                Bottom edge of region.
            `width` : int
                Width of region.
            `height` : int
                Height of region.

        :rtype: AbstractImage
        """
        raise ImageException('Cannot get region for %r' % self)

    def save(self, filename=None, file=None, encoder=None):
        """Save this image to a file.

        :Parameters:
            `filename` : str
                Used to set the image file format, and to open the output file
                if `file` is unspecified.
            `file` : file-like object or None
                File to write image data to.
            `encoder` : ImageEncoder or None
                If unspecified, all encoders matching the filename extension
                are tried.  If all fail, the exception from the first one
                attempted is raised.

        """
        if not file:
            file = open(filename, 'wb')

        if encoder:
            encoder.encode(self, filename, file)
        else:
            first_exception = None
            for encoder in _codec_registry.get_encoders(filename):
                try:
                    return encoder.encode(self, filename, file)
                except ImageEncodeException as e:
                    first_exception = first_exception or e
                    file.seek(0)

            if not first_exception:
                raise ImageEncodeException('No image encoders are available')
            raise first_exception

    def blit(self, x, y, z=0):
        """Draw this image to the active framebuffers.

        The image will be drawn with the lower-left corner at
        (``x -`` `anchor_x`, ``y -`` `anchor_y`, ``z``).
        """
        raise ImageException('Cannot blit %r.' % self)

    def blit_into(self, source, x, y, z):
        """Draw `source` on this image.

        `source` will be copied into this image such that its anchor point
        is aligned with the `x` and `y` parameters.  If this image is a 3D
        texture, the `z` coordinate gives the image slice to copy into.

        Note that if `source` is larger than this image (or the positioning
        would cause the copy to go out of bounds) then you must pass a
        region of `source` to this method, typically using get_region().
        """
        raise ImageException('Cannot blit images onto %r.' % self)

    def blit_to_texture(self, target, level, x, y, z=0):
        """Draw this image on the currently bound texture at `target`.

        This image is copied into the texture such that this image's anchor
        point is aligned with the given `x` and `y` coordinates of the
        destination texture.  If the currently bound texture is a 3D texture,
        the `z` coordinate gives the image slice to blit into.
        """
        raise ImageException('Cannot blit %r to a texture.' % self)


class AbstractImageSequence:
    """Abstract sequence of images.

    The sequence is useful for storing image animations or slices of a volume.
    For efficient access, use the `texture_sequence` member.  The class
    also implements the sequence interface (`__len__`, `__getitem__`,
    `__setitem__`).
    """

    def get_texture_sequence(self):
        """Get a TextureSequence.

        :rtype: `TextureSequence`

        .. versionadded:: 1.1
        """
        raise NotImplementedError('abstract')

    def get_animation(self, period, loop=True):
        """Create an animation over this image sequence for the given constant
        framerate.

        :Parameters
            `period` : float
                Number of seconds to display each frame.
            `loop` : bool
                If True, the animation will loop continuously.

        :rtype: :py:class:`~pyglet.image.Animation`

        .. versionadded:: 1.1
        """
        return Animation.from_image_sequence(self, period, loop)

    def __getitem__(self, slice):
        """Retrieve a (list of) image.

        :rtype: AbstractImage
        """
        raise NotImplementedError('abstract')

    def __setitem__(self, slice, image):
        """Replace one or more images in the sequence.

        :Parameters:
            `image` : `~pyglet.image.AbstractImage`
                The replacement image.  The actual instance may not be used,
                depending on this implementation.

        """
        raise NotImplementedError('abstract')

    def __len__(self):
        raise NotImplementedError('abstract')

    def __iter__(self):
        """Iterate over the images in sequence.

        :rtype: Iterator

        .. versionadded:: 1.1
        """
        raise NotImplementedError('abstract')


class TextureSequence(AbstractImageSequence):
    """Interface for a sequence of textures.

    Typical implementations store multiple :py:class:`~pyglet.image.TextureRegion` s within one
    :py:class:`~pyglet.image.Texture` so as to minimise state changes.
    """

    def get_texture_sequence(self):
        return self


class UniformTextureSequence(TextureSequence):
    """Interface for a sequence of textures, each with the same dimensions.

    :Parameters:
        `item_width` : int
            Width of each texture in the sequence.
        `item_height` : int
            Height of each texture in the sequence.

    """

    def _get_item_width(self):
        raise NotImplementedError('abstract')

    def _get_item_height(self):
        raise NotImplementedError('abstract')

    @property
    def item_width(self):
        return self._get_item_width()

    @property
    def item_height(self):
        return self._get_item_height()


class ImageData(AbstractImage):
    """An image represented as a string of unsigned bytes.

    :Parameters:
        `data` : str
            Pixel data, encoded according to `format` and `pitch`.
        `format` : str
            The format string to use when reading or writing `data`.
        `pitch` : int
            Number of bytes per row.  Negative values indicate a top-to-bottom
            arrangement.

    """
    _swap1_pattern = re.compile(asbytes('(.)'), re.DOTALL)
    _swap2_pattern = re.compile(asbytes('(.)(.)'), re.DOTALL)
    _swap3_pattern = re.compile(asbytes('(.)(.)(.)'), re.DOTALL)
    _swap4_pattern = re.compile(asbytes('(.)(.)(.)(.)'), re.DOTALL)

    _current_texture = None
    _current_mipmap_texture = None

    def __init__(self, width, height, fmt, data, pitch=None):
        """Initialise image data.

        :Parameters:
            `width` : int
                Width of image data
            `height` : int
                Height of image data
            `fmt` : str
                A valid format string, such as 'RGB', 'RGBA', 'ARGB', etc.
            `data` : sequence
                String or array/list of bytes giving the decoded data.
            `pitch` : int or None
                If specified, the number of bytes per row.  Negative values
                indicate a top-to-bottom arrangement.  Defaults to
                ``width * len(format)``.

        """
        super().__init__(width, height)

        self._current_format = self._desired_format = fmt.upper()
        self._current_data = data
        self.pitch = pitch or width * len(fmt)
        self._current_pitch = self.pitch
        self.mipmap_images = []

    def __getstate__(self):
        return {
            'width': self.width,
            'height': self.height,
            '_current_data': self.get_data(self._current_format, self._current_pitch),
            '_current_format': self._current_format,
            '_desired_format': self._desired_format,
            '_current_pitch': self._current_pitch,
            'pitch': self.pitch,
            'mipmap_images': self.mipmap_images
        }

    def get_image_data(self):
        return self

    @property
    def format(self):
        """Format string of the data.  Read-write.

        :type: str
        """
        return self._desired_format

    @format.setter
    def format(self, fmt):
        self._desired_format = fmt.upper()
        self._current_texture = None

    def get_data(self, fmt=None, pitch=None):
        """Get the byte data of the image.

        :Parameters:
            `fmt` : str
                Format string of the return data.
            `pitch` : int
                Number of bytes per row.  Negative values indicate a
                top-to-bottom arrangement.

        .. versionadded:: 1.1

        :rtype: sequence of bytes, or str
        """
        fmt = fmt or self._desired_format
        pitch = pitch or self._current_pitch

        if fmt == self._current_format and pitch == self._current_pitch:
            return self._current_data
        return self._convert(fmt, pitch)

    def set_data(self, fmt, pitch, data):
        """Set the byte data of the image.

        :Parameters:
            `fmt` : str
                Format string of the return data.
            `pitch` : int
                Number of bytes per row.  Negative values indicate a
                top-to-bottom arrangement.
            `data` : str or sequence of bytes
                Image data.

        .. versionadded:: 1.1
        """
        self._current_format = fmt
        self._current_pitch = pitch
        self._current_data = data
        self._current_texture = None
        self._current_mipmap_texture = None

    def set_mipmap_image(self, level, image):
        """Set a mipmap image for a particular level.

        The mipmap image will be applied to textures obtained via
        `get_mipmapped_texture`.

        :Parameters:
            `level` : int
                Mipmap level to set image at, must be >= 1.
            `image` : AbstractImage
                Image to set.  Must have correct dimensions for that mipmap
                level (i.e., width >> level, height >> level)
        """

        if level == 0:
            raise ImageException('Cannot set mipmap image at level 0 (it is this image)')

        # Check dimensions of mipmap
        width, height = self.width, self.height
        for i in range(level):
            width >>= 1
            height >>= 1
        if width != image.width or height != image.height:
            raise ImageException('Mipmap image has wrong dimensions for level %d' % level)

        # Extend mipmap_images list to required level
        self.mipmap_images += [None] * (level - len(self.mipmap_images))
        self.mipmap_images[level - 1] = image

    def create_texture(self, cls, rectangle=False):
        """Create a texture containing this image.

        :Parameters:
            `cls` : class (subclass of Texture)
                Class to construct.
            `rectangle` : bool
                Unused. kept for compatibility.

                .. versionadded:: 1.1

        :rtype: cls or cls.region_class
        """
        internalformat = self._get_internalformat(self._desired_format)
        texture = cls.create(self.width, self.height, GL_TEXTURE_2D, internalformat, False, blank_data=False)
        if self.anchor_x or self.anchor_y:
            texture.anchor_x = self.anchor_x
            texture.anchor_y = self.anchor_y

        self.blit_to_texture(texture.target, texture.level, self.anchor_x, self.anchor_y, 0, None)

        return texture

    def get_texture(self, rectangle=False):
        if not self._current_texture:
            self._current_texture = self.create_texture(Texture, rectangle)
        return self._current_texture

    def get_mipmapped_texture(self):
        """Return a Texture with mipmaps.

        If :py:class:`~pyglet.image.set_mipmap_Image` has been called with at least one image, the set
        of images defined will be used.  Otherwise, mipmaps will be
        automatically generated.

        :rtype: :py:class:`~pyglet.image.Texture`

        .. versionadded:: 1.1
        """
        if self._current_mipmap_texture:
            return self._current_mipmap_texture

        texture = Texture.create(self.width, self.height, GL_TEXTURE_2D, None, blank_data=False)
        if self.anchor_x or self.anchor_y:
            texture.anchor_x = self.anchor_x
            texture.anchor_y = self.anchor_y

        internalformat = self._get_internalformat(self.format)

        glBindTexture(texture.target, texture.id)
        glTexParameteri(texture.target, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR)

        if self.mipmap_images:
            self.blit_to_texture(texture.target, texture.level, self.anchor_x, self.anchor_y, 0, internalformat)
            level = 0
            for image in self.mipmap_images:
                level += 1
                if image:
                    image.blit_to_texture(texture.target, level, self.anchor_x, self.anchor_y, 0, internalformat)
                    # TODO: should set base and max mipmap level if some mipmaps are missing.
        else:
            glGenerateMipmap(texture.target)
            self.blit_to_texture(texture.target, texture.level, self.anchor_x, self.anchor_y, 0, internalformat)

        self._current_mipmap_texture = texture
        return texture

    def get_region(self, x, y, width, height):
        """Retrieve a rectangular region of this image data.

        :Parameters:
            `x` : int
                Left edge of region.
            `y` : int
                Bottom edge of region.
            `width` : int
                Width of region.
            `height` : int
                Height of region.

        :rtype: ImageDataRegion
        """
        return ImageDataRegion(x, y, width, height, self)

    def blit(self, x, y, z=0, width=None, height=None):
        self.get_texture().blit(x, y, z, width, height)

    def blit_to_texture(self, target, level, x, y, z, internalformat=None):
        """Draw this image to to the currently bound texture at `target`.

        This image's anchor point will be aligned to the given `x` and `y`
        coordinates.  If the currently bound texture is a 3D texture, the `z`
        parameter gives the image slice to blit into.

        If `internalformat` is specified, glTexImage is used to initialise
        the texture; otherwise, glTexSubImage is used to update a region.
        """
        x -= self.anchor_x
        y -= self.anchor_y

        data_format = self.format
        data_pitch = abs(self._current_pitch)

        # Determine pixel format from format string
        fmt, gl_type = self._get_gl_format_and_type(data_format)

        if fmt is None:
            # Need to convert data to a standard form
            data_format = {
                1: 'R',
                2: 'RG',
                3: 'RGB',
                4: 'RGBA'}.get(len(data_format))
            fmt, gl_type = self._get_gl_format_and_type(data_format)

        # Get data in required format (hopefully will be the same format it's already
        # in, unless that's an obscure format, upside-down or the driver is old).
        data = self._convert(data_format, data_pitch)

        if data_pitch & 0x1:
            align = 1
        elif data_pitch & 0x2:
            align = 2
        else:
            align = 4
        row_length = data_pitch // len(data_format)

        glPixelStorei(GL_UNPACK_ALIGNMENT, align)
        glPixelStorei(GL_UNPACK_ROW_LENGTH, row_length)
        self._apply_region_unpack()

        if target == GL_TEXTURE_3D or target == GL_TEXTURE_2D_ARRAY:
            assert not internalformat
            glTexSubImage3D(target, level,
                            x, y, z,
                            self.width, self.height, 1,
                            fmt, gl_type,
                            data)
        elif internalformat:
            glTexImage2D(target, level,
                         internalformat,
                         self.width, self.height,
                         0,
                         fmt, gl_type,
                         data)
        else:
            glTexSubImage2D(target, level,
                            x, y,
                            self.width, self.height,
                            fmt, gl_type,
                            data)

        # Unset GL_UNPACK_ROW_LENGTH:
        glPixelStorei(GL_UNPACK_ROW_LENGTH, 0)
        self._default_region_unpack()

        # Flush image upload before data get GC'd:
        glFlush()

    def _apply_region_unpack(self):
        pass

    def _default_region_unpack(self):
        pass

    def _convert(self, fmt, pitch):
        """Return data in the desired format; does not alter this instance's
        current format or pitch.
        """
        if fmt == self._current_format and pitch == self._current_pitch:
            if type(self._current_data) is str:
                return asbytes(self._current_data)
            return self._current_data

        self._ensure_bytes()
        data = self._current_data
        current_pitch = self._current_pitch
        current_format = self._current_format
        sign_pitch = current_pitch // abs(current_pitch)
        if fmt != self._current_format:
            # Create replacement string, e.g. r'\4\1\2\3' to convert RGBA to ARGB
            repl = asbytes('')
            for c in fmt:
                try:
                    idx = current_format.index(c) + 1
                except ValueError:
                    idx = 1
                repl += asbytes(r'\%d' % idx)

            if len(current_format) == 1:
                swap_pattern = self._swap1_pattern
            elif len(current_format) == 2:
                swap_pattern = self._swap2_pattern
            elif len(current_format) == 3:
                swap_pattern = self._swap3_pattern
            elif len(current_format) == 4:
                swap_pattern = self._swap4_pattern
            else:
                raise ImageException('Current image format is wider than 32 bits.')

            packed_pitch = self.width * len(current_format)
            if abs(self._current_pitch) != packed_pitch:
                # Pitch is wider than pixel data, need to go row-by-row.
                new_pitch = abs(self._current_pitch)
                rows = [data[i:i+new_pitch] for i in range(0, len(data), new_pitch)]
                rows = [swap_pattern.sub(repl, r[:packed_pitch]) for r in rows]
                data = asbytes('').join(rows)
            else:
                # Rows are tightly packed, apply regex over whole image.
                data = swap_pattern.sub(repl, data)

            # After conversion, rows will always be tightly packed
            current_pitch = sign_pitch * (len(fmt) * self.width)

        if pitch != current_pitch:
            diff = abs(current_pitch) - abs(pitch)
            if diff > 0:
                # New pitch is shorter than old pitch, chop bytes off each row
                new_pitch = abs(pitch)
                rows = [data[i:i+new_pitch-diff] for i in range(0, len(data), new_pitch)]
                data = asbytes('').join(rows)

            elif diff < 0:
                # New pitch is longer than old pitch, add '0' bytes to each row
                new_pitch = abs(current_pitch)
                padding = asbytes(1) * -diff
                rows = [data[i:i+new_pitch] + padding for i in range(0, len(data), new_pitch)]
                data = asbytes('').join(rows)

            if current_pitch * pitch < 0:
                # Pitch differs in sign, swap row order
                new_pitch = abs(pitch)
                rows = [data[i:i+new_pitch] for i in range(0, len(data), new_pitch)]
                rows.reverse()
                data = asbytes('').join(rows)

        return asbytes(data)

    def _ensure_bytes(self):
        if type(self._current_data) is not bytes:
            self._current_data = asbytes(self._current_data)

    @staticmethod
    def _get_gl_format_and_type(fmt):
        if fmt == 'R':
            return GL_RED, GL_UNSIGNED_BYTE
        elif fmt == 'RG':
            return GL_RG, GL_UNSIGNED_BYTE
        elif fmt == 'RGB':
            return GL_RGB, GL_UNSIGNED_BYTE
        elif fmt == 'BGR':
            return GL_BGR, GL_UNSIGNED_BYTE
        elif fmt == 'RGBA':
            return GL_RGBA, GL_UNSIGNED_BYTE
        elif fmt == 'BGRA':
            return GL_BGRA, GL_UNSIGNED_BYTE

        elif fmt == 'L':
            return GL_LUMINANCE, GL_UNSIGNED_BYTE
        elif fmt == 'A':
            return GL_ALPHA, GL_UNSIGNED_BYTE

        return None, None

    @staticmethod
    def _get_internalformat(fmt):
        if fmt == 'R':
            return GL_RED
        elif fmt == 'RG':
            return GL_RG
        elif fmt == 'RGB':
            return GL_RGB
        elif fmt == 'RGBA':
            return GL_RGBA
        elif fmt == 'D':
            return GL_DEPTH_COMPONENT
        elif fmt == 'DS':
            return GL_DEPTH_STENCIL

        elif fmt == 'L':
            return GL_LUMINANCE
        elif fmt == 'A':
            return GL_ALPHA

        return GL_RGBA


class ImageDataRegion(ImageData):
    def __init__(self, x, y, width, height, image_data):
        super().__init__(width, height,
                         image_data._current_format,
                         image_data._current_data,
                         image_data._current_pitch)
        self.x = x
        self.y = y

    def __getstate__(self):
        return {
            'width': self.width,
            'height': self.height,
            '_current_data': self.get_data(self._current_format, self._current_pitch),
            '_current_format': self._current_format,
            '_desired_format': self._desired_format,
            '_current_pitch': self._current_pitch,
            'pitch': self.pitch,
            'mipmap_images': self.mipmap_images,
            'x': self.x,
            'y': self.y
        }

    def get_data(self, fmt=None, pitch=None):
        x1 = len(self._current_format) * self.x
        x2 = len(self._current_format) * (self.x + self.width)

        self._ensure_bytes()
        data = self._convert(self._current_format, abs(self._current_pitch))
        new_pitch = abs(self._current_pitch)
        rows = [data[i:i+new_pitch] for i in range(0, len(data), new_pitch)]
        rows = [row[x1:x2] for row in rows[self.y:self.y + self.height]]
        self._current_data = b''.join(rows)
        self._current_pitch = self.width * len(self._current_format)
        self._current_texture = None
        self.x = 0
        self.y = 0

        fmt = fmt or self._desired_format
        pitch = pitch or self._current_pitch
        return super().get_data(fmt, pitch)

    def set_data(self, fmt, pitch, data):
        self.x = 0
        self.y = 0
        super().set_data(fmt, pitch, data)

    def _apply_region_unpack(self):
        glPixelStorei(GL_UNPACK_SKIP_PIXELS, self.x)
        glPixelStorei(GL_UNPACK_SKIP_ROWS, self.y)

    def _default_region_unpack(self):
        glPixelStorei(GL_UNPACK_SKIP_PIXELS, 0)
        glPixelStorei(GL_UNPACK_SKIP_ROWS, 0)

    def get_region(self, x, y, width, height):
        x += self.x
        y += self.y
        return super().get_region(x, y, width, height)


class CompressedImageData(AbstractImage):
    """Image representing some compressed data suitable for direct uploading
    to driver.
    """

    _current_texture = None
    _current_mipmap_texture = None

    def __init__(self, width, height, gl_format, data, extension=None, decoder=None):
        """Construct a CompressedImageData with the given compressed data.

        :Parameters:
            `width` : int
                Width of image
            `height` : int
                Height of image
            `gl_format` : int
                GL constant giving format of compressed data; for example,
                ``GL_COMPRESSED_RGBA_S3TC_DXT5_EXT``.
            `data` : sequence
                String or array/list of bytes giving compressed image data.
            `extension` : str or None
                If specified, gives the name of a GL extension to check for
                before creating a texture.
            `decoder` : function(data, width, height) -> AbstractImage
                A function to decode the compressed data, to be used if the
                required extension is not present.

        """
        super().__init__(width, height)
        self.data = data
        self.gl_format = gl_format
        self.extension = extension
        self.decoder = decoder
        self.mipmap_data = []

    def set_mipmap_data(self, level, data):
        """Set data for a mipmap level.

        Supplied data gives a compressed image for the given mipmap level.
        The image must be of the correct dimensions for the level
        (i.e., width >> level, height >> level); but this is not checked.  If
        any mipmap levels are specified, they are used; otherwise, mipmaps for
        `mipmapped_texture` are generated automatically.

        :Parameters:
            `level` : int
                Level of mipmap image to set.
            `data` : sequence
                String or array/list of bytes giving compressed image data.
                Data must be in same format as specified in constructor.

        """
        # Extend mipmap_data list to required level
        self.mipmap_data += [None] * (level - len(self.mipmap_data))
        self.mipmap_data[level - 1] = data

    def _have_extension(self):
        return self.extension is None or gl_info.have_extension(self.extension)

    def _verify_driver_supported(self):
        """Assert that the extension required for this image data is
        supported.

        Raises `ImageException` if not.
        """

        if not self._have_extension():
            raise ImageException('%s is required to decode %r' % (self.extension, self))

    def get_texture(self, rectangle=False):
        if rectangle:
            raise ImageException('Compressed texture rectangles not supported')

        if self._current_texture:
            return self._current_texture

        texture = Texture.create(self.width, self.height, GL_TEXTURE_2D, None)

        if self.anchor_x or self.anchor_y:
            texture.anchor_x = self.anchor_x
            texture.anchor_y = self.anchor_y

        glBindTexture(texture.target, texture.id)
        glTexParameteri(texture.target, GL_TEXTURE_MIN_FILTER, texture.min_filter)
        glTexParameteri(texture.target, GL_TEXTURE_MAG_FILTER, texture.mag_filter)

        if self._have_extension():
            glCompressedTexImage2D(texture.target, texture.level,
                                   self.gl_format,
                                   self.width, self.height, 0,
                                   len(self.data), self.data)
        else:
            image = self.decoder(self.data, self.width, self.height)
            texture = image.get_texture()
            assert texture.width == self.width
            assert texture.height == self.height

        glFlush()
        self._current_texture = texture
        return texture

    def get_mipmapped_texture(self):
        if self._current_mipmap_texture:
            return self._current_mipmap_texture

        if not self._have_extension():
            # TODO mip-mapped software decoded compressed textures.  For now,
            # just return a non-mipmapped texture.
            return self.get_texture()

        texture = Texture.create(self.width, self.height, GL_TEXTURE_2D, None)

        if self.anchor_x or self.anchor_y:
            texture.anchor_x = self.anchor_x
            texture.anchor_y = self.anchor_y

        glBindTexture(texture.target, texture.id)

        glTexParameteri(texture.target, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR)

        if not self.mipmap_data:
            glGenerateMipmap(texture.target)

        glCompressedTexImage2D(texture.target, texture.level,
                               self.gl_format,
                               self.width, self.height, 0,
                               len(self.data), self.data)

        width, height = self.width, self.height
        level = 0
        for data in self.mipmap_data:
            width >>= 1
            height >>= 1
            level += 1
            glCompressedTexImage2D(texture.target, level, self.gl_format, width, height, 0, len(data), data)

        glFlush()

        self._current_mipmap_texture = texture
        return texture

    def blit_to_texture(self, target, level, x, y, z):
        self._verify_driver_supported()

        if target == GL_TEXTURE_3D:
            glCompressedTexSubImage3D(target, level,
                                      x - self.anchor_x, y - self.anchor_y, z,
                                      self.width, self.height, 1,
                                      self.gl_format,
                                      len(self.data), self.data)
        else:
            glCompressedTexSubImage2D(target, level,
                                      x - self.anchor_x, y - self.anchor_y,
                                      self.width, self.height,
                                      self.gl_format,
                                      len(self.data), self.data)


class Texture(AbstractImage):
    """An image loaded into video memory that can be efficiently drawn
    to the framebuffer.

    Typically, you will get an instance of Texture by accessing the `texture`
    member of any other AbstractImage.

    :Parameters:
        `region_class` : class (subclass of TextureRegion)
            Class to use when constructing regions of this texture.
        `tex_coords` : tuple
            12-tuple of float, named (u1, v1, r1, u2, v2, r2, ...).  u, v, r
            give the 3D texture coordinates for vertices 1-4.  The vertices
            are specified in the order bottom-left, bottom-right, top-right
            and top-left.
        `target` : int
            The GL texture target (e.g., ``GL_TEXTURE_2D``).
        `level` : int
            The mipmap level of this texture.

    """

    region_class = None  # Set to TextureRegion after it's defined
    tex_coords = (0, 0, 0, 1, 0, 0, 1, 1, 0, 0, 1, 0)
    tex_coords_order = (0, 1, 2, 3)
    colors = (0, 0, 0, 0) * 4
    level = 0
    images = 1
    x = y = z = 0
    default_min_filter = GL_LINEAR
    default_mag_filter = GL_LINEAR

    def __init__(self, width, height, target, tex_id):
        super().__init__(width, height)
        self.target = target
        self.id = tex_id
        self._context = pyglet.gl.current_context

    def delete(self):
        """Delete this texture and the memory it occupies.
        After this, it may not be used anymore.
        """
        glDeleteTextures(1, self.id)
        self.id = None

    def __del__(self):
        if self.id is not None:
            try:
                self._context.delete_texture(self.id)
                self.id = None
            except (AttributeError, ImportError):
                pass  # Interpreter is shutting down

    def bind(self, texture_unit: int = 0):
        """Bind to a specific Texture Unit by number."""
        glActiveTexture(GL_TEXTURE0 + texture_unit)
        glBindTexture(self.target, self.id)

    def bind_image_texture(self, unit, level=0, layered=False, layer=0, access=GL_READ_WRITE, fmt=GL_RGBA32F):
        """Bind as an ImageTexture for use with a :py:class:`~pyglet.shader.ComputeShaderProgram`.

        .. note:: OpenGL 4.3, or 4.2 with the GL_ARB_compute_shader extention is required.
        """
        glBindImageTexture(unit, self.id, level, layered, layer, access, fmt)

    @classmethod
    def create(cls, width, height, target=GL_TEXTURE_2D, internalformat=GL_RGBA8, min_filter=None, mag_filter=None, fmt=GL_RGBA, blank_data=True):
        """Create a Texture

        Create a Texture with the specified dimentions, target and format.
        On return, the texture will be bound.

        :Parameters:
            `width` : int
                Width of texture in pixels.
            `height` : int
                Height of texture in pixels.
            `target` : int
                GL constant giving texture target to use, typically ``GL_TEXTURE_2D``.
            `internalformat` : int
                GL constant giving internal format of texture; for example, ``GL_RGBA``.
                The internal format decides how the texture data will be stored internally.
                If ``None``, the texture will be created but not initialized.
            `min_filter` : int
                The minifaction filter used for this texture, commonly ``GL_LINEAR`` or ``GL_NEAREST``
            `mag_filter` : int
                The magnification filter used for this texture, commonly ``GL_LINEAR`` or ``GL_NEAREST``
            `fmt` : int
                GL constant giving format of texture; for example, ``GL_RGBA``.
                The format specifies what format the pixel data we're expecting to write
                to the texture and should ideally be the same as for internal format.
            `blank_data` : bool
                Setting to True will initialize the texture data with all zeros. Setting False, will initialize Texture
                with no data.

        :rtype: :py:class:`~pyglet.image.Texture`
        """
        min_filter = min_filter or cls.default_min_filter
        mag_filter = mag_filter or cls.default_mag_filter

        tex_id = GLuint()
        glGenTextures(1, byref(tex_id))
        glBindTexture(target, tex_id.value)
        glTexParameteri(target, GL_TEXTURE_MIN_FILTER, min_filter)
        glTexParameteri(target, GL_TEXTURE_MAG_FILTER, mag_filter)

        if internalformat is not None:
            blank = (GLubyte * (width * height * 4))() if blank_data else None
            glTexImage2D(target, 0,
                         internalformat,
                         width, height,
                         0,
                         fmt,
                         GL_UNSIGNED_BYTE,
                         blank)
            glFlush()

        texture = cls(width, height, target, tex_id.value)
        texture.min_filter = min_filter
        texture.mag_filter = mag_filter
        texture.tex_coords = cls.tex_coords

        return texture

    def get_image_data(self, z=0):
        """Get the image data of this texture.

        Changes to the returned instance will not be reflected in this
        texture.

        :Parameters:
            `z` : int
                For 3D textures, the image slice to retrieve.

        :rtype: :py:class:`~pyglet.image.ImageData`
        """
        glBindTexture(self.target, self.id)

        # Always extract complete RGBA data.  Could check internalformat
        # to only extract used channels. XXX
        fmt = 'RGBA'
        gl_format = GL_RGBA

        buf = (GLubyte * (self.width * self.height * self.images * len(fmt)))()

        # TODO: Clean up this temporary hack
        if pyglet.gl.current_context.get_info().get_opengl_api() == "gles":
            fbo = c_uint()
            glGenFramebuffers(1, fbo)
            glBindFramebuffer(GL_FRAMEBUFFER, fbo.value)
            glPixelStorei(GL_PACK_ALIGNMENT, 1)
            glCheckFramebufferStatus(GL_FRAMEBUFFER)
            glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, self.id, self.level)
            glReadPixels(0, 0, self.width, self.height, gl_format, GL_UNSIGNED_BYTE, buf)
            glBindFramebuffer(GL_FRAMEBUFFER, 0)
            glDeleteFramebuffers(1, fbo)
        else:
            glPixelStorei(GL_PACK_ALIGNMENT, 1)
            glGetTexImage(self.target, self.level, gl_format, GL_UNSIGNED_BYTE, buf)

        data = ImageData(self.width, self.height, fmt, buf)
        if self.images > 1:
            data = data.get_region(0, z * self.height, self.width, self.height)
        return data

    def get_texture(self, rectangle=False):
        return self

    # no implementation of blit_to_texture yet

    def blit(self, x, y, z=0, width=None, height=None):
        x1 = x - self.anchor_x
        y1 = y - self.anchor_y
        x2 = x1 + (width is None and self.width or width)
        y2 = y1 + (height is None and self.height or height)
        vertices = x1, y1, z,  x2, y1, z,  x2, y2, z,  x1, y2, z

        glActiveTexture(GL_TEXTURE0)
        glBindTexture(self.target, self.id)

        pyglet.graphics.draw_indexed(4, GL_TRIANGLES, [0, 1, 2, 0, 2, 3],
                                     position=('f', vertices),
                                     tex_coords=('f', self.tex_coords),
                                     colors=('Bn', self.colors))

        glBindTexture(self.target, 0)

    def blit_into(self, source, x, y, z):
        glBindTexture(self.target, self.id)
        source.blit_to_texture(self.target, self.level, x, y, z)

    def get_region(self, x, y, width, height):
        return self.region_class(x, y, 0, width, height, self)

    def get_transform(self, flip_x=False, flip_y=False, rotate=0):
        """Create a copy of this image applying a simple transformation.

        The transformation is applied to the texture coordinates only;
        :py:meth:`~pyglet.image.ImageData.get_image_data` will return the untransformed data.  The
        transformation is applied around the anchor point.

        :Parameters:
            `flip_x` : bool
                If True, the returned image will be flipped horizontally.
            `flip_y` : bool
                If True, the returned image will be flipped vertically.
            `rotate` : int
                Degrees of clockwise rotation of the returned image.  Only
                90-degree increments are supported.

        :rtype: :py:class:`~pyglet.image.TextureRegion`
        """
        transform = self.get_region(0, 0, self.width, self.height)
        bl, br, tr, tl = 0, 1, 2, 3
        transform.anchor_x = self.anchor_x
        transform.anchor_y = self.anchor_y
        if flip_x:
            bl, br, tl, tr = br, bl, tr, tl
            transform.anchor_x = self.width - self.anchor_x
        if flip_y:
            bl, br, tl, tr = tl, tr, bl, br
            transform.anchor_y = self.height - self.anchor_y
        rotate %= 360
        if rotate < 0:
            rotate += 360
        if rotate == 0:
            pass
        elif rotate == 90:
            bl, br, tr, tl = br, tr, tl, bl
            transform.anchor_x, transform.anchor_y = transform.anchor_y, transform.width - transform.anchor_x
        elif rotate == 180:
            bl, br, tr, tl = tr, tl, bl, br
            transform.anchor_x = transform.width - transform.anchor_x
            transform.anchor_y = transform.height - transform.anchor_y
        elif rotate == 270:
            bl, br, tr, tl = tl, bl, br, tr
            transform.anchor_x, transform.anchor_y = transform.height - transform.anchor_y, transform.anchor_x
        else:
            assert False, 'Only 90 degree rotations are supported.'
        if rotate in (90, 270):
            transform.width, transform.height = transform.height, transform.width
        transform._set_tex_coords_order(bl, br, tr, tl)
        return transform

    def _set_tex_coords_order(self, bl, br, tr, tl):
        tex_coords = (self.tex_coords[:3],
                      self.tex_coords[3:6],
                      self.tex_coords[6:9],
                      self.tex_coords[9:])
        self.tex_coords = tex_coords[bl] + tex_coords[br] + tex_coords[tr] + tex_coords[tl]

        order = self.tex_coords_order
        self.tex_coords_order = (order[bl], order[br], order[tr], order[tl])

    @property
    def uv(self):
        """Tuple containing the left, bottom, right, top 2D texture coordinates."""
        tex_coords = self.tex_coords
        return tex_coords[0], tex_coords[1], tex_coords[3], tex_coords[7]

    def __repr__(self):
        return "{}(id={}, size={}x{})".format(self.__class__.__name__, self.id, self.width, self.height)


class TextureRegion(Texture):
    """A rectangular region of a texture, presented as if it were a separate texture.
    """

    def __init__(self, x, y, z, width, height, owner):
        super().__init__(width, height, owner.target, owner.id)

        self.x = x
        self.y = y
        self.z = z
        self.owner = owner
        owner_u1 = owner.tex_coords[0]
        owner_v1 = owner.tex_coords[1]
        owner_u2 = owner.tex_coords[3]
        owner_v2 = owner.tex_coords[7]
        scale_u = owner_u2 - owner_u1
        scale_v = owner_v2 - owner_v1
        u1 = x / owner.width * scale_u + owner_u1
        v1 = y / owner.height * scale_v + owner_v1
        u2 = (x + width) / owner.width * scale_u + owner_u1
        v2 = (y + height) / owner.height * scale_v + owner_v1
        r = z / owner.images + owner.tex_coords[2]
        self.tex_coords = (u1, v1, r, u2, v1, r, u2, v2, r, u1, v2, r)

    def get_image_data(self):
        image_data = self.owner.get_image_data(self.z)
        return image_data.get_region(self.x, self.y, self.width, self.height)

    def get_region(self, x, y, width, height):
        x += self.x
        y += self.y
        region = self.region_class(x, y, self.z, width, height, self.owner)
        region._set_tex_coords_order(*self.tex_coords_order)
        return region

    def blit_into(self, source, x, y, z):
        self.owner.blit_into(source, x + self.x, y + self.y, z + self.z)

    def __repr__(self):
        return "{}(id={}, size={}x{}, owner={}x{})".format(self.__class__.__name__, self.id, self.width, self.height,
                                                           self.owner.width, self.owner.height)

    def delete(self):
        """Deleting a TextureRegion has no effect. Operate on the owning
        texture instead.
        """
        pass

    def __del__(self):
        pass


Texture.region_class = TextureRegion


class Texture3D(Texture, UniformTextureSequence):
    """A texture with more than one image slice.

    Use `create_for_images` or `create_for_image_grid` classmethod to
    construct.
    """
    item_width = 0
    item_height = 0
    items = ()

    @classmethod
    def create_for_images(cls, images, internalformat=GL_RGBA, blank_data=True):
        item_width = images[0].width
        item_height = images[0].height
        for image in images:
            if image.width != item_width or image.height != item_height:
                raise ImageException('Images do not have same dimensions.')

        depth = len(images)

        texture = cls.create(item_width, item_height, GL_TEXTURE_3D, None)
        if images[0].anchor_x or images[0].anchor_y:
            texture.anchor_x = images[0].anchor_x
            texture.anchor_y = images[0].anchor_y

        texture.images = depth

        blank = (GLubyte * (texture.width * texture.height * texture.images))() if blank_data else None
        glBindTexture(texture.target, texture.id)
        glTexImage3D(texture.target, texture.level,
                     internalformat,
                     texture.width, texture.height, texture.images, 0,
                     GL_ALPHA, GL_UNSIGNED_BYTE,
                     blank)

        items = []
        for i, image in enumerate(images):
            item = cls.region_class(0, 0, i, item_width, item_height, texture)
            items.append(item)
            image.blit_to_texture(texture.target, texture.level, image.anchor_x, image.anchor_y, i)

        glFlush()

        texture.items = items
        texture.item_width = item_width
        texture.item_height = item_height
        return texture

    @classmethod
    def create_for_image_grid(cls, grid, internalformat=GL_RGBA):
        return cls.create_for_images(grid[:], internalformat)

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

    def __getitem__(self, index):
        return self.items[index]

    def __setitem__(self, index, value):
        if type(index) is slice:
            for item, image in zip(self[index], value):
                image.blit_to_texture(self.target, self.level, image.anchor_x, image.anchor_y, item.z)
        else:
            value.blit_to_texture(self.target, self.level, value.anchor_x, value.anchor_y, self[index].z)

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


class TextureArrayRegion(TextureRegion):
    """A region of a TextureArray, presented as if it were a separate texture.
    """

    def __repr__(self):
        return "{}(id={}, size={}x{}, layer={})".format(self.__class__.__name__, self.id, self.width, self.height,
                                                        self.z)


class TextureArray(Texture, UniformTextureSequence):
    def __init__(self, width, height, target, tex_id, max_depth):
        super().__init__(width, height, target, tex_id)
        self.max_depth = max_depth
        self.items = []

    @classmethod
    def create(cls, width, height, internalformat=GL_RGBA, min_filter=None, mag_filter=None, max_depth=256):
        """Create an empty TextureArray.

        You may specify the maximum depth, or layers, the Texture Array should have. This defaults
        to 256, but will be hardware and driver dependent.

        :Parameters:
            `width` : int
                Width of the texture.
            `height` : int
                Height of the texture.
            `internalformat` : int
                GL constant giving the internal format of the texture array; for example, ``GL_RGBA``.
            `min_filter` : int
                The minifaction filter used for this texture array, commonly ``GL_LINEAR`` or ``GL_NEAREST``
            `mag_filter` : int
                The magnification filter used for this texture array, commonly ``GL_LINEAR`` or ``GL_NEAREST``
            `max_depth` : int
                The number of layers in the texture array.

        :rtype: :py:class:`~pyglet.image.TextureArray`

        .. versionadded:: 2.0
        """
        min_filter = min_filter or cls.default_min_filter
        mag_filter = mag_filter or cls.default_mag_filter

        max_depth_limit = get_max_array_texture_layers()
        assert max_depth <= max_depth_limit, "TextureArray max_depth supported is {}.".format(max_depth_limit)

        tex_id = GLuint()
        glGenTextures(1, byref(tex_id))
        glBindTexture(GL_TEXTURE_2D_ARRAY, tex_id.value)
        glTexParameteri(GL_TEXTURE_2D_ARRAY, GL_TEXTURE_MIN_FILTER, min_filter)
        glTexParameteri(GL_TEXTURE_2D_ARRAY, GL_TEXTURE_MAG_FILTER, mag_filter)

        glTexImage3D(GL_TEXTURE_2D_ARRAY, 0,
                     internalformat,
                     width, height, max_depth,
                     0,
                     internalformat, GL_UNSIGNED_BYTE,
                     0)
        glFlush()

        texture = cls(width, height, GL_TEXTURE_2D_ARRAY, tex_id.value, max_depth)
        texture.min_filter = min_filter
        texture.mag_filter = mag_filter

        return texture

    def _verify_size(self, image):
        if image.width > self.width or image.height > self.height:
            raise TextureArraySizeExceeded(f'Image ({image.width}x{image.height}) exceeds the size of the TextureArray ({self.width}x{self.height})')

    def add(self, image: pyglet.image.ImageData):
        if len(self.items) >= self.max_depth:
            raise TextureArrayDepthExceeded(f"TextureArray is full.")

        self._verify_size(image)
        start_length = len(self.items)
        item = self.region_class(0, 0, start_length, image.width, image.height, self)
        image.blit_to_texture(self.target, self.level, image.anchor_x, image.anchor_y, start_length)
        self.items.append(item)
        return item

    def allocate(self, *images):
        """Allocates multiple images at once."""
        if len(self.items) + len(images) > self.max_depth:
            raise TextureArrayDepthExceeded("The amount of images being added exceeds the depth of this TextureArray.")

        start_length = len(self.items)
        for i, image in enumerate(images):
            self._verify_size(image)
            item = self.region_class(0, 0, start_length + i, image.width, image.height, self)
            self.items.append(item)
            image.blit_to_texture(self.target, self.level, image.anchor_x, image.anchor_y, start_length + i)

        return self.items[start_length:]

    @classmethod
    def create_for_image_grid(cls, grid, internalformat=GL_RGBA):
        texture_array = cls.create(grid[0].width, grid[0].height, internalformat, max_depth=len(grid))
        texture_array.allocate(*grid[:])
        return texture_array

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

    def __getitem__(self, index):
        return self.items[index]

    def __setitem__(self, index, value):
        if type(index) is slice:
            for old_item, image in zip(self[index], value):
                self._verify_size(image)
                item = self.region_class(0, 0, old_item.z, image.width, image.height, self)
                image.blit_to_texture(self.target, self.level, image.anchor_x, image.anchor_y, old_item.z)
                self.items[old_item.z] = item
        else:
            self._verify_size(value)
            item = self.region_class(0, 0, index, value.width, value.height, self)
            value.blit_to_texture(self.target, self.level, value.anchor_x, value.anchor_y, index)
            self.items[index] = item

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


TextureArray.region_class = TextureArrayRegion
TextureArrayRegion.region_class = TextureArrayRegion


class TileableTexture(Texture):
    """A texture that can be tiled efficiently.

    Use :py:class:`~pyglet.image.create_for_image` classmethod to construct.
    """

    def get_region(self, x, y, width, height):
        raise ImageException('Cannot get region of %r' % self)

    def blit_tiled(self, x, y, z, width, height):
        """Blit this texture tiled over the given area.

        The image will be tiled with the bottom-left corner of the destination
        rectangle aligned with the anchor point of this texture.
        """
        u1 = self.anchor_x / self.width
        v1 = self.anchor_y / self.height
        u2 = u1 + width / self.width
        v2 = v1 + height / self.height
        w, h = width, height
        t = self.tex_coords

        vertices = (x, y, z,
                    x + w, y, z,
                    x + w, y + h, z,
                    x, y + h, z)

        tex_coords = (u1, v1, t[2],
                      u2, v1, t[5],
                      u2, v2, t[8],
                      u1, v2, t[11],)

        glActiveTexture(GL_TEXTURE0)
        glBindTexture(self.target, self.id)
        pyglet.graphics.draw_indexed(4, GL_TRIANGLES, [0, 1, 2, 0, 2, 3],
                                     position=('f', vertices),
                                     tex_coords=('f', tex_coords))
        glBindTexture(self.target, 0)

    @classmethod
    def create_for_image(cls, image):
        image = image.get_image_data()
        return image.create_texture(cls)


class DepthTexture(Texture):
    """A texture with depth samples (typically 24-bit)."""

    def blit_into(self, source, x, y, z):
        glBindTexture(self.target, self.id)
        source.blit_to_texture(self.level, x, y, z)


class ImageGrid(AbstractImage, AbstractImageSequence):
    """An imaginary grid placed over an image allowing easy access to
    regular regions of that image.

    The grid can be accessed either as a complete image, or as a sequence
    of images.  The most useful applications are to access the grid
    as a :py:class:`~pyglet.image.TextureGrid`::

        image_grid = ImageGrid(...)
        texture_grid = image_grid.get_texture_sequence()

    or as a :py:class:`~pyglet.image.Texture3D`::

        image_grid = ImageGrid(...)
        texture_3d = Texture3D.create_for_image_grid(image_grid)

    """
    _items = ()
    _texture_grid = None

    def __init__(self, image, rows, columns, item_width=None, item_height=None, row_padding=0, column_padding=0):
        """Construct a grid for the given image.

        You can specify parameters for the grid, for example setting
        the padding between cells.  Grids are always aligned to the
        bottom-left corner of the image.

        :Parameters:
            `image` : AbstractImage
                Image over which to construct the grid.
            `rows` : int
                Number of rows in the grid.
            `columns` : int
                Number of columns in the grid.
            `item_width` : int
                Width of each column.  If unspecified, is calculated such
                that the entire image width is used.
            `item_height` : int
                Height of each row.  If unspecified, is calculated such that
                the entire image height is used.
            `row_padding` : int
                Pixels separating adjacent rows.  The padding is only
                inserted between rows, not at the edges of the grid.
            `column_padding` : int
                Pixels separating adjacent columns.  The padding is only
                inserted between columns, not at the edges of the grid.
        """
        super().__init__(image.width, image.height)
        self.image = image
        self.rows = rows
        self.columns = columns
        self.item_width = item_width or (image.width - column_padding * (columns - 1)) // columns
        self.item_height = item_height or (image.height - row_padding * (rows - 1)) // rows
        self.row_padding = row_padding
        self.column_padding = column_padding

    def get_texture(self, rectangle=False):
        return self.image.get_texture(rectangle)

    def get_image_data(self):
        return self.image.get_image_data()

    def get_texture_sequence(self):
        if not self._texture_grid:
            self._texture_grid = TextureGrid(self)
        return self._texture_grid

    def __len__(self):
        return self.rows * self.columns

    def _update_items(self):
        if not self._items:
            self._items = []
            y = 0
            for row in range(self.rows):
                x = 0
                for col in range(self.columns):
                    self._items.append(self.image.get_region(x, y, self.item_width, self.item_height))
                    x += self.item_width + self.column_padding
                y += self.item_height + self.row_padding

    def __getitem__(self, index):
        self._update_items()
        if type(index) is tuple:
            row, column = index
            assert 0 <= row < self.rows and 0 <= column < self.columns
            return self._items[row * self.columns + column]
        else:
            return self._items[index]

    def __iter__(self):
        self._update_items()
        return iter(self._items)


class TextureGrid(TextureRegion, UniformTextureSequence):
    """A texture containing a regular grid of texture regions.

    To construct, create an :py:class:`~pyglet.image.ImageGrid` first::

        image_grid = ImageGrid(...)
        texture_grid = TextureGrid(image_grid)

    The texture grid can be accessed as a single texture, or as a sequence
    of :py:class:`~pyglet.image.TextureRegion`.  When accessing as a sequence, you can specify
    integer indexes, in which the images are arranged in rows from the
    bottom-left to the top-right::

        # assume the texture_grid is 3x3:
        current_texture = texture_grid[3] # get the middle-left image

    You can also specify tuples in the sequence methods, which are addressed
    as ``row, column``::

        # equivalent to the previous example:
        current_texture = texture_grid[1, 0]

    When using tuples in a slice, the returned sequence is over the
    rectangular region defined by the slice::

        # returns center, center-right, center-top, top-right images in that
        # order:
        images = texture_grid[(1,1):]
        # equivalent to
        images = texture_grid[(1,1):(3,3)]

    """
    items = ()
    rows = 1
    columns = 1
    item_width = 0
    item_height = 0

    def __init__(self, grid):
        image = grid.get_texture()
        if isinstance(image, TextureRegion):
            owner = image.owner
        else:
            owner = image

        super().__init__(image.x, image.y, image.z, image.width, image.height, owner)

        items = []
        y = 0
        for row in range(grid.rows):
            x = 0
            for col in range(grid.columns):
                items.append(self.get_region(x, y, grid.item_width, grid.item_height))
                x += grid.item_width + grid.column_padding
            y += grid.item_height + grid.row_padding

        self.items = items
        self.rows = grid.rows
        self.columns = grid.columns
        self.item_width = grid.item_width
        self.item_height = grid.item_height

    def get(self, row, column):
        return self[(row, column)]

    def __getitem__(self, index):
        if type(index) is slice:
            if type(index.start) is not tuple and type(index.stop) is not tuple:
                return self.items[index]
            else:
                row1 = 0
                col1 = 0
                row2 = self.rows
                col2 = self.columns
                if type(index.start) is tuple:
                    row1, col1 = index.start
                elif type(index.start) is int:
                    row1 = index.start // self.columns
                    col1 = index.start % self.columns
                assert 0 <= row1 < self.rows and 0 <= col1 < self.columns

                if type(index.stop) is tuple:
                    row2, col2 = index.stop
                elif type(index.stop) is int:
                    row2 = index.stop // self.columns
                    col2 = index.stop % self.columns
                assert 0 <= row2 <= self.rows and 0 <= col2 <= self.columns

                result = []
                i = row1 * self.columns
                for row in range(row1, row2):
                    result += self.items[i + col1:i + col2]
                    i += self.columns
                return result
        else:
            if type(index) is tuple:
                row, column = index
                assert 0 <= row < self.rows and 0 <= column < self.columns
                return self.items[row * self.columns + column]
            elif type(index) is int:
                return self.items[index]

    def __setitem__(self, index, value):
        if type(index) is slice:
            for region, image in zip(self[index], value):
                if image.width != self.item_width or image.height != self.item_height:
                    raise ImageException('Image has incorrect dimensions')
                image.blit_into(region, image.anchor_x, image.anchor_y, 0)
        else:
            image = value
            if image.width != self.item_width or image.height != self.item_height:
                raise ImageException('Image has incorrect dimensions')
            image.blit_into(self[index], image.anchor_x, image.anchor_y, 0)

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

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


# Initialise default codecs
_add_default_codecs()

# Default Framebuffer classes:
###############################################################


class BufferManager:
    """Manages the set of framebuffers for a context.

    Use :py:func:`~pyglet.image.get_buffer_manager` to obtain the instance of this class for the
    current context.
    """

    def __init__(self):
        self.color_buffer = None
        self.depth_buffer = None
        self.free_stencil_bits = None
        self.refs = []

    @staticmethod
    def get_viewport():
        """Get the current OpenGL viewport dimensions.

        :rtype: 4-tuple of float.
        :return: Left, top, right and bottom dimensions.
        """
        viewport = (GLint * 4)()
        glGetIntegerv(GL_VIEWPORT, viewport)
        return viewport

    def get_color_buffer(self):
        """Get the color buffer.

        :rtype: :py:class:`~pyglet.image.ColorBufferImage`
        """
        viewport = self.get_viewport()
        viewport_width = viewport[2]
        viewport_height = viewport[3]
        if (not self.color_buffer or
                viewport_width != self.color_buffer.width or
                viewport_height != self.color_buffer.height):
            self.color_buffer = ColorBufferImage(*viewport)
        return self.color_buffer

    def get_depth_buffer(self):
        """Get the depth buffer.

        :rtype: :py:class:`~pyglet.image.DepthBufferImage`
        """
        viewport = self.get_viewport()
        viewport_width = viewport[2]
        viewport_height = viewport[3]
        if (not self.depth_buffer or
                viewport_width != self.depth_buffer.width or
                viewport_height != self.depth_buffer.height):
            self.depth_buffer = DepthBufferImage(*viewport)
        return self.depth_buffer

    def get_buffer_mask(self):
        """Get a free bitmask buffer.

        A bitmask buffer is a buffer referencing a single bit in the stencil
        buffer.  If no bits are free, `ImageException` is raised.  Bits are
        released when the bitmask buffer is garbage collected.

        :rtype: :py:class:`~pyglet.image.BufferImageMask`
        """
        if self.free_stencil_bits is None:
            try:
                stencil_bits = GLint()
                glGetFramebufferAttachmentParameteriv(GL_DRAW_FRAMEBUFFER,
                                                      GL_STENCIL,
                                                      GL_FRAMEBUFFER_ATTACHMENT_STENCIL_SIZE,
                                                      stencil_bits)
                self.free_stencil_bits = list(range(stencil_bits.value))
            except GLException:
                pass

        if not self.free_stencil_bits:
            raise ImageException('No free stencil bits are available.')

        stencil_bit = self.free_stencil_bits.pop(0)
        x, y, width, height = self.get_viewport()
        bufimg = BufferImageMask(x, y, width, height)
        bufimg.stencil_bit = stencil_bit

        def release_buffer(ref, owner=self):
            owner.free_stencil_bits.insert(0, stencil_bit)

        self.refs.append(weakref.ref(bufimg, release_buffer))

        return bufimg


def get_buffer_manager():
    """Get the buffer manager for the current OpenGL context.

    :rtype: :py:class:`~pyglet.image.BufferManager`
    """
    context = pyglet.gl.current_context
    if not hasattr(context, 'image_buffer_manager'):
        context.image_buffer_manager = BufferManager()
    return context.image_buffer_manager


class BufferImage(AbstractImage):
    """An abstract framebuffer.
    """
    #: The OpenGL read and write target for this buffer.
    gl_buffer = GL_BACK

    #: The OpenGL format constant for image data.
    gl_format = 0

    #: The format string used for image data.
    format = ''

    owner = None

    # TODO: enable methods

    def __init__(self, x, y, width, height):
        super().__init__(width, height)
        self.x = x
        self.y = y
        self.width = width
        self.height = height

    def get_image_data(self):
        buf = (GLubyte * (len(self.format) * self.width * self.height))()

        x = self.x
        y = self.y
        if self.owner:
            x += self.owner.x
            y += self.owner.y

        glReadBuffer(self.gl_buffer)
        glPixelStorei(GL_PACK_ALIGNMENT, 1)
        glReadPixels(x, y, self.width, self.height, self.gl_format, GL_UNSIGNED_BYTE, buf)
        return ImageData(self.width, self.height, self.format, buf)

    def get_region(self, x, y, width, height):
        if self.owner:
            return self.owner.get_region(x + self.x, y + self.y, width, height)

        region = self.__class__(x + self.x, y + self.y, width, height)
        region.gl_buffer = self.gl_buffer
        region.owner = self
        return region


class ColorBufferImage(BufferImage):
    """A color framebuffer.

    This class is used to wrap the primary color buffer (i.e., the back
    buffer)
    """
    gl_format = GL_RGBA
    format = 'RGBA'

    def get_texture(self, rectangle=False):
        texture = Texture.create(self.width, self.height, GL_TEXTURE_2D, GL_RGBA, blank_data=False)
        self.blit_to_texture(texture.target, texture.level, self.anchor_x, self.anchor_y, 0)
        return texture

    def blit_to_texture(self, target, level, x, y, z):
        glReadBuffer(self.gl_buffer)
        glCopyTexSubImage2D(target, level, x-self.anchor_x, y-self.anchor_y, self.x, self.y, self.width, self.height)


class DepthBufferImage(BufferImage):
    """The depth buffer.
    """
    gl_format = GL_DEPTH_COMPONENT
    format = 'L'

    def get_texture(self, rectangle=False):
        assert rectangle is False, 'Depth textures cannot be rectangular'

        texture = DepthTexture.create(self.width, self.height, GL_TEXTURE_2D, None)
        if self.anchor_x or self.anchor_y:
            texture.anchor_x = self.anchor_x
            texture.anchor_y = self.anchor_y

        glReadBuffer(self.gl_buffer)
        glCopyTexImage2D(texture.target, 0,
                         GL_DEPTH_COMPONENT,
                         self.x, self.y, self.width, self.height,
                         0)
        return texture

    def blit_to_texture(self, target, level, x, y, z):
        glReadBuffer(self.gl_buffer)
        glCopyTexSubImage2D(target, level, x-self.anchor_x, y-self.anchor_y, self.x, self.y, self.width, self.height)


class BufferImageMask(BufferImage):
    """A single bit of the stencil buffer.
    """
    gl_format = GL_STENCIL_INDEX
    format = 'L'

    # TODO mask methods