imageun/

image.rs

#![allow(clippy::too_many_arguments)]
use std::ffi::OsStr;
use std::io::{self, Write};
use std::mem::size_of;
use std::ops::{Deref, DerefMut};
use std::path::Path;

use crate::color::{ColorType, ExtendedColorType};
use crate::error::{
    ImageError, ImageFormatHint, ImageResult, LimitError, LimitErrorKind, ParameterError,
    ParameterErrorKind,
};
use crate::math::Rect;
use crate::traits::Pixel;
use crate::ImageBuffer;

use crate::animation::Frames;

/// An enumeration of supported image formats.
/// Not all formats support both encoding and decoding.
#[derive(Clone, Copy, PartialEq, Eq, Debug, Hash)]
#[non_exhaustive]
pub enum ImageFormat {
    /// An Image in PNG Format
    Png,

    /// An Image in JPEG Format
    Jpeg,

    /// An Image in GIF Format
    Gif,

    /// An Image in WEBP Format
    WebP,

    /// An Image in general PNM Format
    Pnm,

    /// An Image in TIFF Format
    Tiff,

    /// An Image in TGA Format
    Tga,

    /// An Image in DDS Format
    Dds,

    /// An Image in BMP Format
    Bmp,

    /// An Image in ICO Format
    Ico,

    /// An Image in Radiance HDR Format
    Hdr,

    /// An Image in OpenEXR Format
    OpenExr,

    /// An Image in farbfeld Format
    Farbfeld,

    /// An Image in AVIF Format
    Avif,

    /// An Image in QOI Format
    Qoi,
}

impl ImageFormat {
    /// Return the image format specified by a path's file extension.
    ///
    /// # Example
    ///
    /// ```
    /// use image::ImageFormat;
    ///
    /// let format = ImageFormat::from_extension("jpg");
    /// assert_eq!(format, Some(ImageFormat::Jpeg));
    /// ```
    #[inline]
    pub fn from_extension<S>(ext: S) -> Option<Self>
    where
        S: AsRef<OsStr>,
    {
        // thin wrapper function to strip generics
        fn inner(ext: &OsStr) -> Option<ImageFormat> {
            let ext = ext.to_str()?.to_ascii_lowercase();

            Some(match ext.as_str() {
                "avif" => ImageFormat::Avif,
                "jpg" | "jpeg" => ImageFormat::Jpeg,
                "png" | "apng" => ImageFormat::Png,
                "gif" => ImageFormat::Gif,
                "webp" => ImageFormat::WebP,
                "tif" | "tiff" => ImageFormat::Tiff,
                "tga" => ImageFormat::Tga,
                "dds" => ImageFormat::Dds,
                "bmp" => ImageFormat::Bmp,
                "ico" => ImageFormat::Ico,
                "hdr" => ImageFormat::Hdr,
                "exr" => ImageFormat::OpenExr,
                "pbm" | "pam" | "ppm" | "pgm" => ImageFormat::Pnm,
                "ff" => ImageFormat::Farbfeld,
                "qoi" => ImageFormat::Qoi,
                _ => return None,
            })
        }

        inner(ext.as_ref())
    }

    /// Return the image format specified by the path's file extension.
    ///
    /// # Example
    ///
    /// ```
    /// use image::ImageFormat;
    ///
    /// let format = ImageFormat::from_path("images/ferris.png")?;
    /// assert_eq!(format, ImageFormat::Png);
    ///
    /// # Ok::<(), image::error::ImageError>(())
    /// ```
    #[inline]
    pub fn from_path<P>(path: P) -> ImageResult<Self>
    where
        P: AsRef<Path>,
    {
        // thin wrapper function to strip generics
        fn inner(path: &Path) -> ImageResult<ImageFormat> {
            let exact_ext = path.extension();
            exact_ext
                .and_then(ImageFormat::from_extension)
                .ok_or_else(|| {
                    let format_hint = match exact_ext {
                        None => ImageFormatHint::Unknown,
                        Some(os) => ImageFormatHint::PathExtension(os.into()),
                    };
                    ImageError::Unsupported(format_hint.into())
                })
        }

        inner(path.as_ref())
    }

    /// Return the image format specified by a MIME type.
    ///
    /// # Example
    ///
    /// ```
    /// use image::ImageFormat;
    ///
    /// let format = ImageFormat::from_mime_type("image/png").unwrap();
    /// assert_eq!(format, ImageFormat::Png);
    /// ```
    pub fn from_mime_type<M>(mime_type: M) -> Option<Self>
    where
        M: AsRef<str>,
    {
        match mime_type.as_ref() {
            "image/avif" => Some(ImageFormat::Avif),
            "image/jpeg" => Some(ImageFormat::Jpeg),
            "image/png" => Some(ImageFormat::Png),
            "image/gif" => Some(ImageFormat::Gif),
            "image/webp" => Some(ImageFormat::WebP),
            "image/tiff" => Some(ImageFormat::Tiff),
            "image/x-targa" | "image/x-tga" => Some(ImageFormat::Tga),
            "image/vnd-ms.dds" => Some(ImageFormat::Dds),
            "image/bmp" => Some(ImageFormat::Bmp),
            "image/x-icon" => Some(ImageFormat::Ico),
            "image/vnd.radiance" => Some(ImageFormat::Hdr),
            "image/x-exr" => Some(ImageFormat::OpenExr),
            "image/x-portable-bitmap"
            | "image/x-portable-graymap"
            | "image/x-portable-pixmap"
            | "image/x-portable-anymap" => Some(ImageFormat::Pnm),
            // Qoi's MIME type is being worked on.
            // See: https://github.com/phoboslab/qoi/issues/167
            "image/x-qoi" => Some(ImageFormat::Qoi),
            _ => None,
        }
    }

    /// Return the MIME type for this image format or "application/octet-stream" if no MIME type
    /// exists for the format.
    ///
    /// Some notes on a few of the MIME types:
    ///
    /// - The portable anymap format has a separate MIME type for the pixmap, graymap and bitmap
    ///   formats, but this method returns the general "image/x-portable-anymap" MIME type.
    /// - The Targa format has two common MIME types, "image/x-targa"  and "image/x-tga"; this
    ///   method returns "image/x-targa" for that format.
    /// - The QOI MIME type is still a work in progress. This method returns "image/x-qoi" for
    ///   that format.
    ///
    /// # Example
    ///
    /// ```
    /// use image::ImageFormat;
    ///
    /// let mime_type = ImageFormat::Png.to_mime_type();
    /// assert_eq!(mime_type, "image/png");
    /// ```
    #[must_use]
    pub fn to_mime_type(&self) -> &'static str {
        match self {
            ImageFormat::Avif => "image/avif",
            ImageFormat::Jpeg => "image/jpeg",
            ImageFormat::Png => "image/png",
            ImageFormat::Gif => "image/gif",
            ImageFormat::WebP => "image/webp",
            ImageFormat::Tiff => "image/tiff",
            // the targa MIME type has two options, but this one seems to be used more
            ImageFormat::Tga => "image/x-targa",
            ImageFormat::Dds => "image/vnd-ms.dds",
            ImageFormat::Bmp => "image/bmp",
            ImageFormat::Ico => "image/x-icon",
            ImageFormat::Hdr => "image/vnd.radiance",
            ImageFormat::OpenExr => "image/x-exr",
            // return the most general MIME type
            ImageFormat::Pnm => "image/x-portable-anymap",
            // Qoi's MIME type is being worked on.
            // See: https://github.com/phoboslab/qoi/issues/167
            ImageFormat::Qoi => "image/x-qoi",
            // farbfeld's MIME type taken from https://www.wikidata.org/wiki/Q28206109
            ImageFormat::Farbfeld => "application/octet-stream",
        }
    }

    /// Return if the `ImageFormat` can be decoded by the lib.
    #[inline]
    #[must_use]
    pub fn can_read(&self) -> bool {
        // Needs to be updated once a new variant's decoder is added to free_functions.rs::load
        match self {
            ImageFormat::Png => true,
            ImageFormat::Gif => true,
            ImageFormat::Jpeg => true,
            ImageFormat::WebP => true,
            ImageFormat::Tiff => true,
            ImageFormat::Tga => true,
            ImageFormat::Dds => false,
            ImageFormat::Bmp => true,
            ImageFormat::Ico => true,
            ImageFormat::Hdr => true,
            ImageFormat::OpenExr => true,
            ImageFormat::Pnm => true,
            ImageFormat::Farbfeld => true,
            ImageFormat::Avif => true,
            ImageFormat::Qoi => true,
        }
    }

    /// Return if the `ImageFormat` can be encoded by the lib.
    #[inline]
    #[must_use]
    pub fn can_write(&self) -> bool {
        // Needs to be updated once a new variant's encoder is added to free_functions.rs::save_buffer_with_format_impl
        match self {
            ImageFormat::Gif => true,
            ImageFormat::Ico => true,
            ImageFormat::Jpeg => true,
            ImageFormat::Png => true,
            ImageFormat::Bmp => true,
            ImageFormat::Tiff => true,
            ImageFormat::Tga => true,
            ImageFormat::Pnm => true,
            ImageFormat::Farbfeld => true,
            ImageFormat::Avif => true,
            ImageFormat::WebP => true,
            ImageFormat::Hdr => true,
            ImageFormat::OpenExr => true,
            ImageFormat::Dds => false,
            ImageFormat::Qoi => true,
        }
    }

    /// Return a list of applicable extensions for this format.
    ///
    /// All currently recognized image formats specify at least on extension but for future
    /// compatibility you should not rely on this fact. The list may be empty if the format has no
    /// recognized file representation, for example in case it is used as a purely transient memory
    /// format.
    ///
    /// The method name `extensions` remains reserved for introducing another method in the future
    /// that yields a slice of `OsStr` which is blocked by several features of const evaluation.
    #[must_use]
    pub fn extensions_str(self) -> &'static [&'static str] {
        match self {
            ImageFormat::Png => &["png"],
            ImageFormat::Jpeg => &["jpg", "jpeg"],
            ImageFormat::Gif => &["gif"],
            ImageFormat::WebP => &["webp"],
            ImageFormat::Pnm => &["pbm", "pam", "ppm", "pgm"],
            ImageFormat::Tiff => &["tiff", "tif"],
            ImageFormat::Tga => &["tga"],
            ImageFormat::Dds => &["dds"],
            ImageFormat::Bmp => &["bmp"],
            ImageFormat::Ico => &["ico"],
            ImageFormat::Hdr => &["hdr"],
            ImageFormat::OpenExr => &["exr"],
            ImageFormat::Farbfeld => &["ff"],
            // According to: https://aomediacodec.github.io/av1-avif/#mime-registration
            ImageFormat::Avif => &["avif"],
            ImageFormat::Qoi => &["qoi"],
        }
    }

    /// Return the `ImageFormat`s which are enabled for reading.
    #[inline]
    #[must_use]
    pub fn reading_enabled(&self) -> bool {
        match self {
            ImageFormat::Png => cfg!(feature = "png"),
            ImageFormat::Gif => cfg!(feature = "gif"),
            ImageFormat::Jpeg => cfg!(feature = "jpeg"),
            ImageFormat::WebP => cfg!(feature = "webp"),
            ImageFormat::Tiff => cfg!(feature = "tiff"),
            ImageFormat::Tga => cfg!(feature = "tga"),
            ImageFormat::Bmp => cfg!(feature = "bmp"),
            ImageFormat::Ico => cfg!(feature = "ico"),
            ImageFormat::Hdr => cfg!(feature = "hdr"),
            ImageFormat::OpenExr => cfg!(feature = "exr"),
            ImageFormat::Pnm => cfg!(feature = "pnm"),
            ImageFormat::Farbfeld => cfg!(feature = "ff"),
            ImageFormat::Avif => cfg!(feature = "avif"),
            ImageFormat::Qoi => cfg!(feature = "qoi"),
            ImageFormat::Dds => false,
        }
    }

    /// Return the `ImageFormat`s which are enabled for writing.
    #[inline]
    #[must_use]
    pub fn writing_enabled(&self) -> bool {
        match self {
            ImageFormat::Gif => cfg!(feature = "gif"),
            ImageFormat::Ico => cfg!(feature = "ico"),
            ImageFormat::Jpeg => cfg!(feature = "jpeg"),
            ImageFormat::Png => cfg!(feature = "png"),
            ImageFormat::Bmp => cfg!(feature = "bmp"),
            ImageFormat::Tiff => cfg!(feature = "tiff"),
            ImageFormat::Tga => cfg!(feature = "tga"),
            ImageFormat::Pnm => cfg!(feature = "pnm"),
            ImageFormat::Farbfeld => cfg!(feature = "ff"),
            ImageFormat::Avif => cfg!(feature = "avif"),
            ImageFormat::WebP => cfg!(feature = "webp"),
            ImageFormat::OpenExr => cfg!(feature = "exr"),
            ImageFormat::Qoi => cfg!(feature = "qoi"),
            ImageFormat::Hdr => cfg!(feature = "hdr"),
            ImageFormat::Dds => false,
        }
    }

    /// Return all `ImageFormat`s
    pub fn all() -> impl Iterator<Item = ImageFormat> {
        [
            ImageFormat::Gif,
            ImageFormat::Ico,
            ImageFormat::Jpeg,
            ImageFormat::Png,
            ImageFormat::Bmp,
            ImageFormat::Tiff,
            ImageFormat::Tga,
            ImageFormat::Pnm,
            ImageFormat::Farbfeld,
            ImageFormat::Avif,
            ImageFormat::WebP,
            ImageFormat::OpenExr,
            ImageFormat::Qoi,
            ImageFormat::Dds,
            ImageFormat::Hdr,
        ]
        .iter()
        .copied()
    }
}

// This struct manages buffering associated with implementing `Read` and `Seek` on decoders that can
// must decode ranges of bytes at a time.
#[allow(dead_code)]
// When no image formats that use it are enabled
pub(crate) struct ImageReadBuffer {
    scanline_bytes: usize,
    buffer: Vec<u8>,
    consumed: usize,

    total_bytes: u64,
    offset: u64,
}
impl ImageReadBuffer {
    /// Create a new `ImageReadBuffer`.
    ///
    /// Panics if `scanline_bytes` doesn't fit into a usize, because that would mean reading anything
    /// from the image would take more RAM than the entire virtual address space. In other words,
    /// actually using this struct would instantly OOM so just get it out of the way now.
    #[allow(dead_code)]
    // When no image formats that use it are enabled
    pub(crate) fn new(scanline_bytes: u64, total_bytes: u64) -> Self {
        Self {
            scanline_bytes: usize::try_from(scanline_bytes).unwrap(),
            buffer: Vec::new(),
            consumed: 0,
            total_bytes,
            offset: 0,
        }
    }

    #[allow(dead_code)]
    // When no image formats that use it are enabled
    pub(crate) fn read<F>(&mut self, buf: &mut [u8], mut read_scanline: F) -> io::Result<usize>
    where
        F: FnMut(&mut [u8]) -> io::Result<usize>,
    {
        if self.buffer.len() == self.consumed {
            if self.offset == self.total_bytes {
                return Ok(0);
            } else if buf.len() >= self.scanline_bytes {
                // If there is nothing buffered and the user requested a full scanline worth of
                // data, skip buffering.
                let bytes_read = read_scanline(&mut buf[..self.scanline_bytes])?;
                self.offset += u64::try_from(bytes_read).unwrap();
                return Ok(bytes_read);
            } else {
                // Lazily allocate buffer the first time that read is called with a buffer smaller
                // than the scanline size.
                if self.buffer.is_empty() {
                    self.buffer.resize(self.scanline_bytes, 0);
                }

                self.consumed = 0;
                let bytes_read = read_scanline(&mut self.buffer[..])?;
                self.buffer.resize(bytes_read, 0);
                self.offset += u64::try_from(bytes_read).unwrap();

                assert!(bytes_read == self.scanline_bytes || self.offset == self.total_bytes);
            }
        }

        // Finally, copy bytes into output buffer.
        let bytes_buffered = self.buffer.len() - self.consumed;
        if bytes_buffered > buf.len() {
            buf.copy_from_slice(&self.buffer[self.consumed..][..buf.len()]);
            self.consumed += buf.len();
            Ok(buf.len())
        } else {
            buf[..bytes_buffered].copy_from_slice(&self.buffer[self.consumed..][..bytes_buffered]);
            self.consumed = self.buffer.len();
            Ok(bytes_buffered)
        }
    }
}

/// Decodes a specific region of the image, represented by the rectangle
/// starting from ```x``` and ```y``` and having ```length``` and ```width```
#[allow(dead_code)]
// When no image formats that use it are enabled
pub(crate) fn load_rect<D, F1, F2, E>(
    x: u32,
    y: u32,
    width: u32,
    height: u32,
    buf: &mut [u8],
    row_pitch: usize,
    decoder: &mut D,
    scanline_bytes: usize,
    mut seek_scanline: F1,
    mut read_scanline: F2,
) -> ImageResult<()>
where
    D: ImageDecoder,
    F1: FnMut(&mut D, u64) -> io::Result<()>,
    F2: FnMut(&mut D, &mut [u8]) -> Result<(), E>,
    ImageError: From<E>,
{
    let scanline_bytes = u64::try_from(scanline_bytes).unwrap();
    let row_pitch = u64::try_from(row_pitch).unwrap();

    let (x, y, width, height) = (
        u64::from(x),
        u64::from(y),
        u64::from(width),
        u64::from(height),
    );
    let dimensions = decoder.dimensions();
    let bytes_per_pixel = u64::from(decoder.color_type().bytes_per_pixel());
    let row_bytes = bytes_per_pixel * u64::from(dimensions.0);
    let total_bytes = width * height * bytes_per_pixel;

    assert!(
        buf.len() >= usize::try_from(total_bytes).unwrap_or(usize::MAX),
        "output buffer too short\n expected `{}`, provided `{}`",
        total_bytes,
        buf.len()
    );

    let mut current_scanline = 0;
    let mut tmp = Vec::new();
    let mut tmp_scanline = None;

    {
        // Read a range of the image starting from byte number `start` and continuing until byte
        // number `end`. Updates `current_scanline` and `bytes_read` appropriately.
        let mut read_image_range =
            |mut start: u64, end: u64, mut output: &mut [u8]| -> ImageResult<()> {
                // If the first scanline we need is already stored in the temporary buffer, then handle
                // it first.
                let target_scanline = start / scanline_bytes;
                if tmp_scanline == Some(target_scanline) {
                    let position = target_scanline * scanline_bytes;
                    let offset = start.saturating_sub(position);
                    let len = (end - start)
                        .min(scanline_bytes - offset)
                        .min(end - position);

                    output
                        .write_all(&tmp[offset as usize..][..len as usize])
                        .unwrap();
                    start += len;

                    if start == end {
                        return Ok(());
                    }
                }

                let target_scanline = start / scanline_bytes;
                if target_scanline != current_scanline {
                    seek_scanline(decoder, target_scanline)?;
                    current_scanline = target_scanline;
                }

                let mut position = current_scanline * scanline_bytes;
                while position < end {
                    if position >= start && end - position >= scanline_bytes {
                        read_scanline(decoder, &mut output[..(scanline_bytes as usize)])?;
                        output = &mut output[scanline_bytes as usize..];
                    } else {
                        tmp.resize(scanline_bytes as usize, 0u8);
                        read_scanline(decoder, &mut tmp)?;
                        tmp_scanline = Some(current_scanline);

                        let offset = start.saturating_sub(position);
                        let len = (end - start)
                            .min(scanline_bytes - offset)
                            .min(end - position);

                        output
                            .write_all(&tmp[offset as usize..][..len as usize])
                            .unwrap();
                    }

                    current_scanline += 1;
                    position += scanline_bytes;
                }
                Ok(())
            };

        if x + width > u64::from(dimensions.0)
            || y + height > u64::from(dimensions.1)
            || width == 0
            || height == 0
        {
            return Err(ImageError::Parameter(ParameterError::from_kind(
                ParameterErrorKind::DimensionMismatch,
            )));
        }
        if scanline_bytes > usize::MAX as u64 {
            return Err(ImageError::Limits(LimitError::from_kind(
                LimitErrorKind::InsufficientMemory,
            )));
        }

        if x == 0 && width == u64::from(dimensions.0) && row_pitch == row_bytes {
            let start = x * bytes_per_pixel + y * row_bytes;
            let end = (x + width) * bytes_per_pixel + (y + height - 1) * row_bytes;
            read_image_range(start, end, buf)?;
        } else {
            for (output_slice, row) in buf.chunks_mut(row_pitch as usize).zip(y..(y + height)) {
                let start = x * bytes_per_pixel + row * row_bytes;
                let end = (x + width) * bytes_per_pixel + row * row_bytes;
                read_image_range(start, end, output_slice)?;
            }
        }
    }

    // Seek back to the start
    Ok(seek_scanline(decoder, 0)?)
}

/// Reads all of the bytes of a decoder into a Vec<T>. No particular alignment
/// of the output buffer is guaranteed.
///
/// Panics if there isn't enough memory to decode the image.
pub(crate) fn decoder_to_vec<T>(decoder: impl ImageDecoder) -> ImageResult<Vec<T>>
where
    T: crate::traits::Primitive + bytemuck::Pod,
{
    let total_bytes = usize::try_from(decoder.total_bytes());
    if total_bytes.is_err() || total_bytes.unwrap() > isize::MAX as usize {
        return Err(ImageError::Limits(LimitError::from_kind(
            LimitErrorKind::InsufficientMemory,
        )));
    }

    let mut buf = vec![num_traits::Zero::zero(); total_bytes.unwrap() / size_of::<T>()];
    decoder.read_image(bytemuck::cast_slice_mut(buf.as_mut_slice()))?;
    Ok(buf)
}

/// The trait that all decoders implement
pub trait ImageDecoder {
    /// Returns a tuple containing the width and height of the image
    fn dimensions(&self) -> (u32, u32);

    /// Returns the color type of the image data produced by this decoder
    fn color_type(&self) -> ColorType;

    /// Returns the color type of the image file before decoding
    fn original_color_type(&self) -> ExtendedColorType {
        self.color_type().into()
    }

    /// Returns the ICC color profile embedded in the image, or `Ok(None)` if the image does not have one.
    ///
    /// For formats that don't support embedded profiles this function should always return `Ok(None)`.
    fn icc_profile(&mut self) -> ImageResult<Option<Vec<u8>>> {
        Ok(None)
    }

    /// Returns the raw [Exif](https://en.wikipedia.org/wiki/Exif) chunk, if it is present.
    /// A third-party crate such as [`kamadak-exif`](https://docs.rs/kamadak-exif/) is required to actually parse it.
    ///
    /// For formats that don't support embedded profiles this function should always return `Ok(None)`.
    fn exif_metadata(&mut self) -> ImageResult<Option<Vec<u8>>> {
        Ok(None)
    }

    /// Returns the total number of bytes in the decoded image.
    ///
    /// This is the size of the buffer that must be passed to `read_image` or
    /// `read_image_with_progress`. The returned value may exceed `usize::MAX`, in
    /// which case it isn't actually possible to construct a buffer to decode all the image data
    /// into. If, however, the size does not fit in a u64 then `u64::MAX` is returned.
    fn total_bytes(&self) -> u64 {
        let dimensions = self.dimensions();
        let total_pixels = u64::from(dimensions.0) * u64::from(dimensions.1);
        let bytes_per_pixel = u64::from(self.color_type().bytes_per_pixel());
        total_pixels.saturating_mul(bytes_per_pixel)
    }

    /// Returns all the bytes in the image.
    ///
    /// This function takes a slice of bytes and writes the pixel data of the image into it.
    /// Although not required, for certain color types callers may want to pass buffers which are
    /// aligned to 2 or 4 byte boundaries to the slice can be cast to a [u16] or [u32]. To accommodate
    /// such casts, the returned contents will always be in native endian.
    ///
    /// # Panics
    ///
    /// This function panics if `buf.len() != self.total_bytes()`.
    ///
    /// # Examples
    ///
    /// ```no_build
    /// use zerocopy::{AsBytes, FromBytes};
    /// fn read_16bit_image(decoder: impl ImageDecoder) -> Vec<16> {
    ///     let mut buf: Vec<u16> = vec![0; decoder.total_bytes()/2];
    ///     decoder.read_image(buf.as_bytes());
    ///     buf
    /// }
    /// ```
    fn read_image(self, buf: &mut [u8]) -> ImageResult<()>
    where
        Self: Sized;

    /// Set the decoder to have the specified limits. See [`Limits`] for the different kinds of
    /// limits that is possible to set.
    ///
    /// Note to implementors: make sure you call [`Limits::check_support`] so that
    /// decoding fails if any unsupported strict limits are set. Also make sure
    /// you call [`Limits::check_dimensions`] to check the `max_image_width` and
    /// `max_image_height` limits.
    ///
    /// [`Limits`]: ./io/struct.Limits.html
    /// [`Limits::check_support`]: ./io/struct.Limits.html#method.check_support
    /// [`Limits::check_dimensions`]: ./io/struct.Limits.html#method.check_dimensions
    fn set_limits(&mut self, limits: crate::Limits) -> ImageResult<()> {
        limits.check_support(&crate::LimitSupport::default())?;
        let (width, height) = self.dimensions();
        limits.check_dimensions(width, height)?;
        Ok(())
    }

    /// Use `read_image` instead; this method is an implementation detail needed so the trait can
    /// be object safe.
    ///
    /// Note to implementors: This method should be implemented by calling `read_image` on
    /// the boxed decoder...
    /// ```no_build
    ///     fn read_image_boxed(self: Box<Self>, buf: &mut [u8]) -> ImageResult<()> {
    ///        (*self).read_image(buf)
    ///    }
    /// ```
    fn read_image_boxed(self: Box<Self>, buf: &mut [u8]) -> ImageResult<()>;
}

impl<T: ?Sized + ImageDecoder> ImageDecoder for Box<T> {
    fn dimensions(&self) -> (u32, u32) {
        (**self).dimensions()
    }
    fn color_type(&self) -> ColorType {
        (**self).color_type()
    }
    fn original_color_type(&self) -> ExtendedColorType {
        (**self).original_color_type()
    }
    fn icc_profile(&mut self) -> ImageResult<Option<Vec<u8>>> {
        (**self).icc_profile()
    }
    fn exif_metadata(&mut self) -> ImageResult<Option<Vec<u8>>> {
        (**self).exif_metadata()
    }
    fn total_bytes(&self) -> u64 {
        (**self).total_bytes()
    }
    fn read_image(self, buf: &mut [u8]) -> ImageResult<()>
    where
        Self: Sized,
    {
        T::read_image_boxed(self, buf)
    }
    fn read_image_boxed(self: Box<Self>, buf: &mut [u8]) -> ImageResult<()> {
        T::read_image_boxed(*self, buf)
    }
    fn set_limits(&mut self, limits: crate::Limits) -> ImageResult<()> {
        (**self).set_limits(limits)
    }
}

/// Specialized image decoding not be supported by all formats
pub trait ImageDecoderRect: ImageDecoder {
    /// Decode a rectangular section of the image.
    ///
    /// This function takes a slice of bytes and writes the pixel data of the image into it.
    /// The rectangle is specified by the x and y coordinates of the top left corner, the width
    /// and height of the rectangle, and the row pitch of the buffer. The row pitch is the number
    /// of bytes between the start of one row and the start of the next row. The row pitch must be
    /// at least as large as the width of the rectangle in bytes.
    fn read_rect(
        &mut self,
        x: u32,
        y: u32,
        width: u32,
        height: u32,
        buf: &mut [u8],
        row_pitch: usize,
    ) -> ImageResult<()>;
}

/// `AnimationDecoder` trait
pub trait AnimationDecoder<'a> {
    /// Consume the decoder producing a series of frames.
    fn into_frames(self) -> Frames<'a>;
}

/// The trait all encoders implement
pub trait ImageEncoder {
    /// Writes all the bytes in an image to the encoder.
    ///
    /// This function takes a slice of bytes of the pixel data of the image
    /// and encodes them. Unlike particular format encoders inherent impl encode
    /// methods where endianness is not specified, here image data bytes should
    /// always be in native endian. The implementor will reorder the endianness
    /// as necessary for the target encoding format.
    ///
    /// See also `ImageDecoder::read_image` which reads byte buffers into
    /// native endian.
    ///
    /// # Panics
    ///
    /// Panics if `width * height * color_type.bytes_per_pixel() != buf.len()`.
    fn write_image(
        self,
        buf: &[u8],
        width: u32,
        height: u32,
        color_type: ExtendedColorType,
    ) -> ImageResult<()>;
}

/// Immutable pixel iterator
#[derive(Debug)]
pub struct Pixels<'a, I: ?Sized + 'a> {
    image: &'a I,
    x: u32,
    y: u32,
    width: u32,
    height: u32,
}

impl<'a, I: GenericImageView> Iterator for Pixels<'a, I> {
    type Item = (u32, u32, I::Pixel);

    fn next(&mut self) -> Option<(u32, u32, I::Pixel)> {
        if self.x >= self.width {
            self.x = 0;
            self.y += 1;
        }

        if self.y >= self.height {
            None
        } else {
            let pixel = self.image.get_pixel(self.x, self.y);
            let p = (self.x, self.y, pixel);

            self.x += 1;

            Some(p)
        }
    }
}

impl<I: ?Sized> Clone for Pixels<'_, I> {
    fn clone(&self) -> Self {
        Pixels { ..*self }
    }
}

/// Trait to inspect an image.
///
/// ```
/// use image::{GenericImageView, Rgb, RgbImage};
///
/// let buffer = RgbImage::new(10, 10);
/// let image: &dyn GenericImageView<Pixel=Rgb<u8>> = &buffer;
/// ```
pub trait GenericImageView {
    /// The type of pixel.
    type Pixel: Pixel;

    /// The width and height of this image.
    fn dimensions(&self) -> (u32, u32);

    /// The width of this image.
    fn width(&self) -> u32 {
        let (w, _) = self.dimensions();
        w
    }

    /// The height of this image.
    fn height(&self) -> u32 {
        let (_, h) = self.dimensions();
        h
    }

    /// Returns true if this x, y coordinate is contained inside the image.
    fn in_bounds(&self, x: u32, y: u32) -> bool {
        let (width, height) = self.dimensions();
        x < width && y < height
    }

    /// Returns the pixel located at (x, y). Indexed from top left.
    ///
    /// # Panics
    ///
    /// Panics if `(x, y)` is out of bounds.
    fn get_pixel(&self, x: u32, y: u32) -> Self::Pixel;

    /// Returns the pixel located at (x, y). Indexed from top left.
    ///
    /// This function can be implemented in a way that ignores bounds checking.
    /// # Safety
    ///
    /// The coordinates must be [`in_bounds`] of the image.
    ///
    /// [`in_bounds`]: #method.in_bounds
    unsafe fn unsafe_get_pixel(&self, x: u32, y: u32) -> Self::Pixel {
        self.get_pixel(x, y)
    }

    /// Returns an Iterator over the pixels of this image.
    /// The iterator yields the coordinates of each pixel
    /// along with their value
    fn pixels(&self) -> Pixels<Self>
    where
        Self: Sized,
    {
        let (width, height) = self.dimensions();

        Pixels {
            image: self,
            x: 0,
            y: 0,
            width,
            height,
        }
    }

    /// Returns a subimage that is an immutable view into this image.
    /// You can use [`GenericImage::sub_image`] if you need a mutable view instead.
    /// The coordinates set the position of the top left corner of the view.
    fn view(&self, x: u32, y: u32, width: u32, height: u32) -> SubImage<&Self>
    where
        Self: Sized,
    {
        assert!(u64::from(x) + u64::from(width) <= u64::from(self.width()));
        assert!(u64::from(y) + u64::from(height) <= u64::from(self.height()));
        SubImage::new(self, x, y, width, height)
    }
}

/// A trait for manipulating images.
pub trait GenericImage: GenericImageView {
    /// Gets a reference to the mutable pixel at location `(x, y)`. Indexed from top left.
    ///
    /// # Panics
    ///
    /// Panics if `(x, y)` is out of bounds.
    ///
    /// Panics for dynamic images (this method is deprecated and will be removed).
    ///
    /// ## Known issues
    ///
    /// This requires the buffer to contain a unique set of continuous channels in the exact order
    /// and byte representation that the pixel type requires. This is somewhat restrictive.
    ///
    /// TODO: Maybe use some kind of entry API? this would allow pixel type conversion on the fly
    /// while still doing only one array lookup:
    ///
    /// ```ignore
    /// let px = image.pixel_entry_at(x,y);
    /// px.set_from_rgba(rgba)
    /// ```
    #[deprecated(since = "0.24.0", note = "Use `get_pixel` and `put_pixel` instead.")]
    fn get_pixel_mut(&mut self, x: u32, y: u32) -> &mut Self::Pixel;

    /// Put a pixel at location (x, y). Indexed from top left.
    ///
    /// # Panics
    ///
    /// Panics if `(x, y)` is out of bounds.
    fn put_pixel(&mut self, x: u32, y: u32, pixel: Self::Pixel);

    /// Puts a pixel at location (x, y). Indexed from top left.
    ///
    /// This function can be implemented in a way that ignores bounds checking.
    /// # Safety
    ///
    /// The coordinates must be [`in_bounds`] of the image.
    ///
    /// [`in_bounds`]: traits.GenericImageView.html#method.in_bounds
    unsafe fn unsafe_put_pixel(&mut self, x: u32, y: u32, pixel: Self::Pixel) {
        self.put_pixel(x, y, pixel);
    }

    /// Put a pixel at location (x, y), taking into account alpha channels
    #[deprecated(
        since = "0.24.0",
        note = "Use iterator `pixels_mut` to blend the pixels directly"
    )]
    fn blend_pixel(&mut self, x: u32, y: u32, pixel: Self::Pixel);

    /// Copies all of the pixels from another image into this image.
    ///
    /// The other image is copied with the top-left corner of the
    /// other image placed at (x, y).
    ///
    /// In order to copy only a piece of the other image, use [`GenericImageView::view`].
    ///
    /// You can use [`FlatSamples`] to source pixels from an arbitrary regular raster of channel
    /// values, for example from a foreign interface or a fixed image.
    ///
    /// # Returns
    /// Returns an error if the image is too large to be copied at the given position
    ///
    /// [`GenericImageView::view`]: trait.GenericImageView.html#method.view
    /// [`FlatSamples`]: flat/struct.FlatSamples.html
    fn copy_from<O>(&mut self, other: &O, x: u32, y: u32) -> ImageResult<()>
    where
        O: GenericImageView<Pixel = Self::Pixel>,
    {
        // Do bounds checking here so we can use the non-bounds-checking
        // functions to copy pixels.
        if self.width() < other.width() + x || self.height() < other.height() + y {
            return Err(ImageError::Parameter(ParameterError::from_kind(
                ParameterErrorKind::DimensionMismatch,
            )));
        }

        for k in 0..other.height() {
            for i in 0..other.width() {
                let p = other.get_pixel(i, k);
                self.put_pixel(i + x, k + y, p);
            }
        }
        Ok(())
    }

    /// Copies all of the pixels from one part of this image to another part of this image.
    ///
    /// The destination rectangle of the copy is specified with the top-left corner placed at (x, y).
    ///
    /// # Returns
    /// `true` if the copy was successful, `false` if the image could not
    /// be copied due to size constraints.
    fn copy_within(&mut self, source: Rect, x: u32, y: u32) -> bool {
        let Rect {
            x: sx,
            y: sy,
            width,
            height,
        } = source;
        let dx = x;
        let dy = y;
        assert!(sx < self.width() && dx < self.width());
        assert!(sy < self.height() && dy < self.height());
        if self.width() - dx.max(sx) < width || self.height() - dy.max(sy) < height {
            return false;
        }
        // since `.rev()` creates a new dype we would either have to go with dynamic dispatch for the ranges
        // or have quite a lot of code bloat. A macro gives us static dispatch with less visible bloat.
        macro_rules! copy_within_impl_ {
            ($xiter:expr, $yiter:expr) => {
                for y in $yiter {
                    let sy = sy + y;
                    let dy = dy + y;
                    for x in $xiter {
                        let sx = sx + x;
                        let dx = dx + x;
                        let pixel = self.get_pixel(sx, sy);
                        self.put_pixel(dx, dy, pixel);
                    }
                }
            };
        }
        // check how target and source rectangles relate to each other so we dont overwrite data before we copied it.
        match (sx < dx, sy < dy) {
            (true, true) => copy_within_impl_!((0..width).rev(), (0..height).rev()),
            (true, false) => copy_within_impl_!((0..width).rev(), 0..height),
            (false, true) => copy_within_impl_!(0..width, (0..height).rev()),
            (false, false) => copy_within_impl_!(0..width, 0..height),
        }
        true
    }

    /// Returns a mutable subimage that is a view into this image.
    /// If you want an immutable subimage instead, use [`GenericImageView::view`]
    /// The coordinates set the position of the top left corner of the `SubImage`.
    fn sub_image(&mut self, x: u32, y: u32, width: u32, height: u32) -> SubImage<&mut Self>
    where
        Self: Sized,
    {
        assert!(u64::from(x) + u64::from(width) <= u64::from(self.width()));
        assert!(u64::from(y) + u64::from(height) <= u64::from(self.height()));
        SubImage::new(self, x, y, width, height)
    }
}

/// A View into another image
///
/// Instances of this struct can be created using:
///   - [`GenericImage::sub_image`] to create a mutable view,
///   - [`GenericImageView::view`] to create an immutable view,
///   - [`SubImage::new`] to instantiate the struct directly.
///
/// Note that this does _not_ implement `GenericImage`, but it dereferences to one which allows you
/// to use it as if it did. See [Design Considerations](#Design-Considerations) below for details.
///
/// # Design Considerations
///
/// For reasons relating to coherence, this is not itself a `GenericImage` or a `GenericImageView`.
/// In short, we want to reserve the ability of adding traits implemented for _all_ generic images
/// but in a different manner for `SubImage`. This may be required to ensure that stacking
/// sub-images comes at no double indirect cost.
///
/// If, ultimately, this is not needed then a directly implementation of `GenericImage` can and
/// will get added. This inconvenience may alternatively get resolved if Rust allows some forms of
/// specialization, which might make this trick unnecessary and thus also allows for a direct
/// implementation.
#[derive(Copy, Clone)]
pub struct SubImage<I> {
    inner: SubImageInner<I>,
}

/// The inner type of `SubImage` that implements `GenericImage{,View}`.
///
/// This type is _nominally_ `pub` but it is not exported from the crate. It should be regarded as
/// an existential type in any case.
#[derive(Copy, Clone)]
pub struct SubImageInner<I> {
    image: I,
    xoffset: u32,
    yoffset: u32,
    xstride: u32,
    ystride: u32,
}

/// Alias to access Pixel behind a reference
type DerefPixel<I> = <<I as Deref>::Target as GenericImageView>::Pixel;

/// Alias to access Subpixel behind a reference
type DerefSubpixel<I> = <DerefPixel<I> as Pixel>::Subpixel;

impl<I> SubImage<I> {
    /// Construct a new subimage
    /// The coordinates set the position of the top left corner of the `SubImage`.
    pub fn new(image: I, x: u32, y: u32, width: u32, height: u32) -> SubImage<I> {
        SubImage {
            inner: SubImageInner {
                image,
                xoffset: x,
                yoffset: y,
                xstride: width,
                ystride: height,
            },
        }
    }

    /// Change the coordinates of this subimage.
    pub fn change_bounds(&mut self, x: u32, y: u32, width: u32, height: u32) {
        self.inner.xoffset = x;
        self.inner.yoffset = y;
        self.inner.xstride = width;
        self.inner.ystride = height;
    }

    /// The offsets of this subimage relative to the underlying image.
    pub fn offsets(&self) -> (u32, u32) {
        (self.inner.xoffset, self.inner.yoffset)
    }

    /// Convert this subimage to an `ImageBuffer`
    pub fn to_image(&self) -> ImageBuffer<DerefPixel<I>, Vec<DerefSubpixel<I>>>
    where
        I: Deref,
        I::Target: GenericImageView + 'static,
    {
        let mut out = ImageBuffer::new(self.inner.xstride, self.inner.ystride);
        let borrowed = &*self.inner.image;

        for y in 0..self.inner.ystride {
            for x in 0..self.inner.xstride {
                let p = borrowed.get_pixel(x + self.inner.xoffset, y + self.inner.yoffset);
                out.put_pixel(x, y, p);
            }
        }

        out
    }
}

/// Methods for readable images.
impl<I> SubImage<I>
where
    I: Deref,
    I::Target: GenericImageView,
{
    /// Create a sub-view of the image.
    ///
    /// The coordinates given are relative to the current view on the underlying image.
    ///
    /// Note that this method is preferred to the one from `GenericImageView`. This is accessible
    /// with the explicit method call syntax but it should rarely be needed due to causing an
    /// extra level of indirection.
    ///
    /// ```
    /// use image::{GenericImageView, RgbImage, SubImage};
    /// let buffer = RgbImage::new(10, 10);
    ///
    /// let subimage: SubImage<&RgbImage> = buffer.view(0, 0, 10, 10);
    /// let subview: SubImage<&RgbImage> = subimage.view(0, 0, 10, 10);
    ///
    /// // Less efficient and NOT &RgbImage
    /// let _: SubImage<&_> = GenericImageView::view(&*subimage, 0, 0, 10, 10);
    /// ```
    pub fn view(&self, x: u32, y: u32, width: u32, height: u32) -> SubImage<&I::Target> {
        use crate::GenericImageView as _;
        assert!(u64::from(x) + u64::from(width) <= u64::from(self.inner.width()));
        assert!(u64::from(y) + u64::from(height) <= u64::from(self.inner.height()));
        let x = self.inner.xoffset.saturating_add(x);
        let y = self.inner.yoffset.saturating_add(y);
        SubImage::new(&*self.inner.image, x, y, width, height)
    }

    /// Get a reference to the underlying image.
    pub fn inner(&self) -> &I::Target {
        &self.inner.image
    }
}

impl<I> SubImage<I>
where
    I: DerefMut,
    I::Target: GenericImage,
{
    /// Create a mutable sub-view of the image.
    ///
    /// The coordinates given are relative to the current view on the underlying image.
    pub fn sub_image(
        &mut self,
        x: u32,
        y: u32,
        width: u32,
        height: u32,
    ) -> SubImage<&mut I::Target> {
        assert!(u64::from(x) + u64::from(width) <= u64::from(self.inner.width()));
        assert!(u64::from(y) + u64::from(height) <= u64::from(self.inner.height()));
        let x = self.inner.xoffset.saturating_add(x);
        let y = self.inner.yoffset.saturating_add(y);
        SubImage::new(&mut *self.inner.image, x, y, width, height)
    }

    /// Get a mutable reference to the underlying image.
    pub fn inner_mut(&mut self) -> &mut I::Target {
        &mut self.inner.image
    }
}

impl<I> Deref for SubImage<I>
where
    I: Deref,
{
    type Target = SubImageInner<I>;
    fn deref(&self) -> &Self::Target {
        &self.inner
    }
}

impl<I> DerefMut for SubImage<I>
where
    I: DerefMut,
{
    fn deref_mut(&mut self) -> &mut Self::Target {
        &mut self.inner
    }
}

#[allow(deprecated)]
impl<I> GenericImageView for SubImageInner<I>
where
    I: Deref,
    I::Target: GenericImageView,
{
    type Pixel = DerefPixel<I>;

    fn dimensions(&self) -> (u32, u32) {
        (self.xstride, self.ystride)
    }

    fn get_pixel(&self, x: u32, y: u32) -> Self::Pixel {
        self.image.get_pixel(x + self.xoffset, y + self.yoffset)
    }
}

#[allow(deprecated)]
impl<I> GenericImage for SubImageInner<I>
where
    I: DerefMut,
    I::Target: GenericImage + Sized,
{
    fn get_pixel_mut(&mut self, x: u32, y: u32) -> &mut Self::Pixel {
        self.image.get_pixel_mut(x + self.xoffset, y + self.yoffset)
    }

    fn put_pixel(&mut self, x: u32, y: u32, pixel: Self::Pixel) {
        self.image
            .put_pixel(x + self.xoffset, y + self.yoffset, pixel);
    }

    /// DEPRECATED: This method will be removed. Blend the pixel directly instead.
    fn blend_pixel(&mut self, x: u32, y: u32, pixel: Self::Pixel) {
        self.image
            .blend_pixel(x + self.xoffset, y + self.yoffset, pixel);
    }
}

#[cfg(test)]
mod tests {
    use std::collections::HashSet;
    use std::io;
    use std::path::Path;

    use super::{
        load_rect, ColorType, GenericImage, GenericImageView, ImageDecoder, ImageFormat,
        ImageResult,
    };
    use crate::color::Rgba;
    use crate::math::Rect;
    use crate::{GrayImage, ImageBuffer};

    #[test]
    #[allow(deprecated)]
    /// Test that alpha blending works as expected
    fn test_image_alpha_blending() {
        let mut target = ImageBuffer::new(1, 1);
        target.put_pixel(0, 0, Rgba([255u8, 0, 0, 255]));
        assert!(*target.get_pixel(0, 0) == Rgba([255, 0, 0, 255]));
        target.blend_pixel(0, 0, Rgba([0, 255, 0, 255]));
        assert!(*target.get_pixel(0, 0) == Rgba([0, 255, 0, 255]));

        // Blending an alpha channel onto a solid background
        target.blend_pixel(0, 0, Rgba([255, 0, 0, 127]));
        assert!(*target.get_pixel(0, 0) == Rgba([127, 127, 0, 255]));

        // Blending two alpha channels
        target.put_pixel(0, 0, Rgba([0, 255, 0, 127]));
        target.blend_pixel(0, 0, Rgba([255, 0, 0, 127]));
        assert!(*target.get_pixel(0, 0) == Rgba([169, 85, 0, 190]));
    }

    #[test]
    fn test_in_bounds() {
        let mut target = ImageBuffer::new(2, 2);
        target.put_pixel(0, 0, Rgba([255u8, 0, 0, 255]));

        assert!(target.in_bounds(0, 0));
        assert!(target.in_bounds(1, 0));
        assert!(target.in_bounds(0, 1));
        assert!(target.in_bounds(1, 1));

        assert!(!target.in_bounds(2, 0));
        assert!(!target.in_bounds(0, 2));
        assert!(!target.in_bounds(2, 2));
    }

    #[test]
    fn test_can_subimage_clone_nonmut() {
        let mut source = ImageBuffer::new(3, 3);
        source.put_pixel(1, 1, Rgba([255u8, 0, 0, 255]));

        // A non-mutable copy of the source image
        let source = source.clone();

        // Clone a view into non-mutable to a separate buffer
        let cloned = source.view(1, 1, 1, 1).to_image();

        assert!(cloned.get_pixel(0, 0) == source.get_pixel(1, 1));
    }

    #[test]
    fn test_can_nest_views() {
        let mut source = ImageBuffer::from_pixel(3, 3, Rgba([255u8, 0, 0, 255]));

        {
            let mut sub1 = source.sub_image(0, 0, 2, 2);
            let mut sub2 = sub1.sub_image(1, 1, 1, 1);
            sub2.put_pixel(0, 0, Rgba([0, 0, 0, 0]));
        }

        assert_eq!(*source.get_pixel(1, 1), Rgba([0, 0, 0, 0]));

        let view1 = source.view(0, 0, 2, 2);
        assert_eq!(*source.get_pixel(1, 1), view1.get_pixel(1, 1));

        let view2 = view1.view(1, 1, 1, 1);
        assert_eq!(*source.get_pixel(1, 1), view2.get_pixel(0, 0));
    }

    #[test]
    #[should_panic]
    fn test_view_out_of_bounds() {
        let source = ImageBuffer::from_pixel(3, 3, Rgba([255u8, 0, 0, 255]));
        source.view(1, 1, 3, 3);
    }

    #[test]
    #[should_panic]
    fn test_view_coordinates_out_of_bounds() {
        let source = ImageBuffer::from_pixel(3, 3, Rgba([255u8, 0, 0, 255]));
        source.view(3, 3, 3, 3);
    }

    #[test]
    #[should_panic]
    fn test_view_width_out_of_bounds() {
        let source = ImageBuffer::from_pixel(3, 3, Rgba([255u8, 0, 0, 255]));
        source.view(1, 1, 3, 2);
    }

    #[test]
    #[should_panic]
    fn test_view_height_out_of_bounds() {
        let source = ImageBuffer::from_pixel(3, 3, Rgba([255u8, 0, 0, 255]));
        source.view(1, 1, 2, 3);
    }

    #[test]
    #[should_panic]
    fn test_view_x_out_of_bounds() {
        let source = ImageBuffer::from_pixel(3, 3, Rgba([255u8, 0, 0, 255]));
        source.view(3, 1, 3, 3);
    }

    #[test]
    #[should_panic]
    fn test_view_y_out_of_bounds() {
        let source = ImageBuffer::from_pixel(3, 3, Rgba([255u8, 0, 0, 255]));
        source.view(1, 3, 3, 3);
    }

    #[test]
    fn test_view_in_bounds() {
        let source = ImageBuffer::from_pixel(3, 3, Rgba([255u8, 0, 0, 255]));
        source.view(0, 0, 3, 3);
        source.view(1, 1, 2, 2);
        source.view(2, 2, 0, 0);
    }

    #[test]
    fn test_copy_sub_image() {
        let source = ImageBuffer::from_pixel(3, 3, Rgba([255u8, 0, 0, 255]));
        let view = source.view(0, 0, 3, 3);
        let _view2 = view;
        view.to_image();
    }

    #[test]
    fn test_load_rect() {
        struct MockDecoder {
            scanline_number: u64,
            scanline_bytes: u64,
        }
        impl ImageDecoder for MockDecoder {
            fn dimensions(&self) -> (u32, u32) {
                (5, 5)
            }
            fn color_type(&self) -> ColorType {
                ColorType::L8
            }
            fn read_image(self, _buf: &mut [u8]) -> ImageResult<()> {
                unimplemented!()
            }
            fn read_image_boxed(self: Box<Self>, buf: &mut [u8]) -> ImageResult<()> {
                (*self).read_image(buf)
            }
        }

        const DATA: [u8; 25] = [
            0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23,
            24,
        ];

        fn seek_scanline(m: &mut MockDecoder, n: u64) -> io::Result<()> {
            m.scanline_number = n;
            Ok(())
        }
        fn read_scanline(m: &mut MockDecoder, buf: &mut [u8]) -> io::Result<()> {
            let bytes_read = m.scanline_number * m.scanline_bytes;
            if bytes_read >= 25 {
                return Ok(());
            }

            let len = m.scanline_bytes.min(25 - bytes_read);
            buf[..(len as usize)].copy_from_slice(&DATA[(bytes_read as usize)..][..(len as usize)]);
            m.scanline_number += 1;
            Ok(())
        }

        for scanline_bytes in 1..30 {
            let mut output = [0u8; 26];

            load_rect(
                0,
                0,
                5,
                5,
                &mut output,
                5,
                &mut MockDecoder {
                    scanline_number: 0,
                    scanline_bytes,
                },
                scanline_bytes as usize,
                seek_scanline,
                read_scanline,
            )
            .unwrap();
            assert_eq!(output[0..25], DATA);
            assert_eq!(output[25], 0);

            output = [0u8; 26];
            load_rect(
                3,
                2,
                1,
                1,
                &mut output,
                1,
                &mut MockDecoder {
                    scanline_number: 0,
                    scanline_bytes,
                },
                scanline_bytes as usize,
                seek_scanline,
                read_scanline,
            )
            .unwrap();
            assert_eq!(output[0..2], [13, 0]);

            output = [0u8; 26];
            load_rect(
                3,
                2,
                2,
                2,
                &mut output,
                2,
                &mut MockDecoder {
                    scanline_number: 0,
                    scanline_bytes,
                },
                scanline_bytes as usize,
                seek_scanline,
                read_scanline,
            )
            .unwrap();
            assert_eq!(output[0..5], [13, 14, 18, 19, 0]);

            output = [0u8; 26];
            load_rect(
                1,
                1,
                2,
                4,
                &mut output,
                2,
                &mut MockDecoder {
                    scanline_number: 0,
                    scanline_bytes,
                },
                scanline_bytes as usize,
                seek_scanline,
                read_scanline,
            )
            .unwrap();
            assert_eq!(output[0..9], [6, 7, 11, 12, 16, 17, 21, 22, 0]);
        }
    }

    #[test]
    fn test_load_rect_single_scanline() {
        const DATA: [u8; 25] = [
            0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23,
            24,
        ];

        struct MockDecoder;
        impl ImageDecoder for MockDecoder {
            fn dimensions(&self) -> (u32, u32) {
                (5, 5)
            }
            fn color_type(&self) -> ColorType {
                ColorType::L8
            }
            fn read_image(self, _buf: &mut [u8]) -> ImageResult<()> {
                unimplemented!()
            }
            fn read_image_boxed(self: Box<Self>, buf: &mut [u8]) -> ImageResult<()> {
                (*self).read_image(buf)
            }
        }

        // Ensure that seek scanline is called only once.
        let mut seeks = 0;
        let seek_scanline = |_d: &mut MockDecoder, n: u64| -> io::Result<()> {
            seeks += 1;
            assert_eq!(n, 0);
            assert_eq!(seeks, 1);
            Ok(())
        };

        fn read_scanline(_m: &mut MockDecoder, buf: &mut [u8]) -> io::Result<()> {
            buf.copy_from_slice(&DATA);
            Ok(())
        }

        let mut output = [0; 26];
        load_rect(
            1,
            1,
            2,
            4,
            &mut output,
            2,
            &mut MockDecoder,
            DATA.len(),
            seek_scanline,
            read_scanline,
        )
        .unwrap();
        assert_eq!(output[0..9], [6, 7, 11, 12, 16, 17, 21, 22, 0]);
    }

    #[test]
    fn test_image_format_from_path() {
        fn from_path(s: &str) -> ImageResult<ImageFormat> {
            ImageFormat::from_path(Path::new(s))
        }
        assert_eq!(from_path("./a.jpg").unwrap(), ImageFormat::Jpeg);
        assert_eq!(from_path("./a.jpeg").unwrap(), ImageFormat::Jpeg);
        assert_eq!(from_path("./a.JPEG").unwrap(), ImageFormat::Jpeg);
        assert_eq!(from_path("./a.pNg").unwrap(), ImageFormat::Png);
        assert_eq!(from_path("./a.gif").unwrap(), ImageFormat::Gif);
        assert_eq!(from_path("./a.webp").unwrap(), ImageFormat::WebP);
        assert_eq!(from_path("./a.tiFF").unwrap(), ImageFormat::Tiff);
        assert_eq!(from_path("./a.tif").unwrap(), ImageFormat::Tiff);
        assert_eq!(from_path("./a.tga").unwrap(), ImageFormat::Tga);
        assert_eq!(from_path("./a.dds").unwrap(), ImageFormat::Dds);
        assert_eq!(from_path("./a.bmp").unwrap(), ImageFormat::Bmp);
        assert_eq!(from_path("./a.Ico").unwrap(), ImageFormat::Ico);
        assert_eq!(from_path("./a.hdr").unwrap(), ImageFormat::Hdr);
        assert_eq!(from_path("./a.exr").unwrap(), ImageFormat::OpenExr);
        assert_eq!(from_path("./a.pbm").unwrap(), ImageFormat::Pnm);
        assert_eq!(from_path("./a.pAM").unwrap(), ImageFormat::Pnm);
        assert_eq!(from_path("./a.Ppm").unwrap(), ImageFormat::Pnm);
        assert_eq!(from_path("./a.pgm").unwrap(), ImageFormat::Pnm);
        assert_eq!(from_path("./a.AViF").unwrap(), ImageFormat::Avif);
        assert!(from_path("./a.txt").is_err());
        assert!(from_path("./a").is_err());
    }

    #[test]
    fn test_generic_image_copy_within_oob() {
        let mut image: GrayImage = ImageBuffer::from_raw(4, 4, vec![0u8; 16]).unwrap();
        assert!(!image.sub_image(0, 0, 4, 4).copy_within(
            Rect {
                x: 0,
                y: 0,
                width: 5,
                height: 4
            },
            0,
            0
        ));
        assert!(!image.sub_image(0, 0, 4, 4).copy_within(
            Rect {
                x: 0,
                y: 0,
                width: 4,
                height: 5
            },
            0,
            0
        ));
        assert!(!image.sub_image(0, 0, 4, 4).copy_within(
            Rect {
                x: 1,
                y: 0,
                width: 4,
                height: 4
            },
            0,
            0
        ));
        assert!(!image.sub_image(0, 0, 4, 4).copy_within(
            Rect {
                x: 0,
                y: 0,
                width: 4,
                height: 4
            },
            1,
            0
        ));
        assert!(!image.sub_image(0, 0, 4, 4).copy_within(
            Rect {
                x: 0,
                y: 1,
                width: 4,
                height: 4
            },
            0,
            0
        ));
        assert!(!image.sub_image(0, 0, 4, 4).copy_within(
            Rect {
                x: 0,
                y: 0,
                width: 4,
                height: 4
            },
            0,
            1
        ));
        assert!(!image.sub_image(0, 0, 4, 4).copy_within(
            Rect {
                x: 1,
                y: 1,
                width: 4,
                height: 4
            },
            0,
            0
        ));
    }

    #[test]
    fn test_generic_image_copy_within_tl() {
        let data = &[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15];
        let expected = [0, 1, 2, 3, 4, 0, 1, 2, 8, 4, 5, 6, 12, 8, 9, 10];
        let mut image: GrayImage = ImageBuffer::from_raw(4, 4, Vec::from(&data[..])).unwrap();
        assert!(image.sub_image(0, 0, 4, 4).copy_within(
            Rect {
                x: 0,
                y: 0,
                width: 3,
                height: 3
            },
            1,
            1
        ));
        assert_eq!(&image.into_raw(), &expected);
    }

    #[test]
    fn test_generic_image_copy_within_tr() {
        let data = &[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15];
        let expected = [0, 1, 2, 3, 1, 2, 3, 7, 5, 6, 7, 11, 9, 10, 11, 15];
        let mut image: GrayImage = ImageBuffer::from_raw(4, 4, Vec::from(&data[..])).unwrap();
        assert!(image.sub_image(0, 0, 4, 4).copy_within(
            Rect {
                x: 1,
                y: 0,
                width: 3,
                height: 3
            },
            0,
            1
        ));
        assert_eq!(&image.into_raw(), &expected);
    }

    #[test]
    fn test_generic_image_copy_within_bl() {
        let data = &[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15];
        let expected = [0, 4, 5, 6, 4, 8, 9, 10, 8, 12, 13, 14, 12, 13, 14, 15];
        let mut image: GrayImage = ImageBuffer::from_raw(4, 4, Vec::from(&data[..])).unwrap();
        assert!(image.sub_image(0, 0, 4, 4).copy_within(
            Rect {
                x: 0,
                y: 1,
                width: 3,
                height: 3
            },
            1,
            0
        ));
        assert_eq!(&image.into_raw(), &expected);
    }

    #[test]
    fn test_generic_image_copy_within_br() {
        let data = &[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15];
        let expected = [5, 6, 7, 3, 9, 10, 11, 7, 13, 14, 15, 11, 12, 13, 14, 15];
        let mut image: GrayImage = ImageBuffer::from_raw(4, 4, Vec::from(&data[..])).unwrap();
        assert!(image.sub_image(0, 0, 4, 4).copy_within(
            Rect {
                x: 1,
                y: 1,
                width: 3,
                height: 3
            },
            0,
            0
        ));
        assert_eq!(&image.into_raw(), &expected);
    }

    #[test]
    fn image_formats_are_recognized() {
        use ImageFormat::*;
        const ALL_FORMATS: &[ImageFormat] = &[
            Avif, Png, Jpeg, Gif, WebP, Pnm, Tiff, Tga, Dds, Bmp, Ico, Hdr, Farbfeld, OpenExr,
        ];
        for &format in ALL_FORMATS {
            let mut file = Path::new("file.nothing").to_owned();
            for ext in format.extensions_str() {
                assert!(file.set_extension(ext));
                match ImageFormat::from_path(&file) {
                    Err(_) => panic!("Path {} not recognized as {:?}", file.display(), format),
                    Ok(result) => assert_eq!(format, result),
                }
            }
        }
    }

    #[test]
    fn total_bytes_overflow() {
        struct D;
        impl ImageDecoder for D {
            fn color_type(&self) -> ColorType {
                ColorType::Rgb8
            }
            fn dimensions(&self) -> (u32, u32) {
                (0xffff_ffff, 0xffff_ffff)
            }
            fn read_image(self, _buf: &mut [u8]) -> ImageResult<()> {
                unimplemented!()
            }
            fn read_image_boxed(self: Box<Self>, buf: &mut [u8]) -> ImageResult<()> {
                (*self).read_image(buf)
            }
        }
        assert_eq!(D.total_bytes(), u64::MAX);

        let v: ImageResult<Vec<u8>> = super::decoder_to_vec(D);
        assert!(v.is_err());
    }

    #[test]
    fn all() {
        let all_formats: HashSet<ImageFormat> = ImageFormat::all().collect();
        assert!(all_formats.contains(&ImageFormat::Avif));
        assert!(all_formats.contains(&ImageFormat::Gif));
        assert!(all_formats.contains(&ImageFormat::Bmp));
        assert!(all_formats.contains(&ImageFormat::Farbfeld));
        assert!(all_formats.contains(&ImageFormat::Jpeg));
    }

    #[test]
    fn reading_enabled() {
        assert_eq!(cfg!(feature = "jpeg"), ImageFormat::Jpeg.reading_enabled());
        assert_eq!(
            cfg!(feature = "ff"),
            ImageFormat::Farbfeld.reading_enabled()
        );
        assert!(!ImageFormat::Dds.reading_enabled());
    }

    #[test]
    fn writing_enabled() {
        assert_eq!(cfg!(feature = "jpeg"), ImageFormat::Jpeg.writing_enabled());
        assert_eq!(
            cfg!(feature = "ff"),
            ImageFormat::Farbfeld.writing_enabled()
        );
        assert!(!ImageFormat::Dds.writing_enabled());
    }
}