It is an open question whether L2 and L4 should be included, or
conversely whether L1 should be eliminated

Also clarify description of ColorType::Unknown and fix a few warnings

Refactor the ColorType enum

Continue refactoring ColorType

Fixes #901

A bunch more ColorType improvements for image:next

Make copy_from return an ImageResult instead of a boolean

Update next with latest changes to master

Unlike the other formats this is not a single acronym, doesn't contain
too many capitals and the website and documentation strongly suggests
this way.

Convert ImageFormat variants to CamelCase

Proper capitalization for decoder struct names

Co-Authored-By: est31 <est31@users.noreply.github.com>

Co-Authored-By: Yuki Okushi <huyuumi.dev@gmail.com>

Opinionated changes to ImageDecoder

Make read_rect take u32 for {x, y, width, height}

The latter is part of the a publicly visible crate interface. This makes
it impossible to upgrade the image format crates used for these specific
format over major versions. Using this pattern is problematic in several
aspects:

* less mature formats are harder to adopt as they are a huge commitment
* bugs and issues that require breaking changes require a coordinate
  effort across the whole dependency chain and ecosystem
* dependent crates we control do not evolve as independently as I would
  wish

It should arguably be avoided to have an unecessary public interface
dependency *unless a format reaches a stable 1.0 interface itself* where
it is no large liability. For consistency this should be uniform across
formats without regards to whether they are internally implemented, a
crate in image-rs, or otherwise.

Convert error with private methods, not From

Less in root

Fixes #1058. Migrated to ImageError from std::io::Error

The encoder used to hardcode a pixel aspect ratio of 1x1
It allows setting a custom DPI value.

Fixes #1077

…nsity

requested by @fintelia in
#1078 (review)

Allow setting a custom pixel density in the JPEG encoder

Bump tiff version for `next`

These images are now discouraged for use by the Nature Publishing Group,
Optical Society, and SIAM, as well as having been removed from other
major image processing libraries including Matlab and FIJI/ImageJ.

Instead the Mandrill image (SIPI image 4.2.03) is used for tiff
integration testing, benchmarking, and examples, and a reference image
from PNGSuite is used for interlaced tests.

Fix benchmarks and ensure compilation on Travis

Replace Lenna test images

This makes it possible to add some members in the future. Note that only
this library can create an instance which means it is currently unusable
by user-defined decoders. This is a less pressing concern though, these
can simply never call the callback.

Document and expose all new pub structures

Update release and notes for 0.23 preview

…tures"

Simplify cargo toml

Add 16 bit buffers for all 16 bit types in ColorType. Include these as
supported types in DynamicImage. These 16 bit images are currently only
supported with u16 representation; changing this would require
specialization and macroing of color conversions.

Load 16 bit PNGs at full bit depth rather than squashing to 8 bit.

Note that tests will not pass until upgrading to a release of image-png
including image-rs/image-png#174.

Closes #560.
Closes #665.
Closes #940.

This ImageEncoder trait is currently minimal, but establishes a uniform
interface so that all formats have native endian encoding interfaces,
which is necessary for 16 bit per channel encoding.

Closes #1076.

This fixes 16bpc tests.

Per review comments, this assumes equal channel bit depths, which will
not be the case in the future.

Because the return type depends on the buffer's bits per channel, make
the return optional and provide separate methods for 8 and 16 bit
images.

Instead create a crate-local AsBytes wrapper trait. This trait is not
sealed as Primitive was not sealed, so that users can delegate it for
newtypes if needed.

Bytemuck is more feature constrained and builds on Rust 1.34.2.

16 bit per channel image buffers and dynamic images

Adds checked_mul when computing img size.

Checked also for the last add.

Fixes #1090.

Fix potential overflow in GIF decoding

Update to Edition 2018

Move error types to seperate module

It has become quite outdated at this point but luckily is maintained
with a commitment to MSRV of 1.30. The main point of the update are some
reported issues with rand and the dependency rdrand. Not that any of
those are actually relevant but this reduces the burden of proof for
verifications (crev, etc.) of code bases using the image crate.

Update the quickcheck dev-dependency

Update `next` from master

Also note the curious and complicated but clever way of statically
eliminating the actual branch in matchings, and avoid the need for
unreachable markers within the crate. This has been ported to the other
uses of non-exhaustive enumerations.

Add support for DDS files using DXT1, DXT3 or DXT5 compression

Make ImageFormat related enums non-exhaustive

Add missing feature requirement dxt for dds

Moves away from a single, open representation of errors. This considers
a previously identified few issues:

* The new variants are somewhat actionable since they at least identify the
  cause more closely.
* They are extensible as most state is opaque to some degree and key
  enumerations are marked as non-exhaustive.
* They are more precise as the opaque state allows storing specialized
  variants for supported formats while still maintaining a uniform
  interface and permitting extension formats outside the library.
* Handling of 'stringification' is more principled as error values can
  be stored as-is instead of converting them on the spot. This might
  also improve the performance when the conversion is not required.

This is part of a patch series. The commit merely adds the new type and
a rough interface but does not yet convert the existing error instances
and return types.

Emulates the old enum variants with the new type through constructor
methods. Note that some variants have been slightly abused and do not
map cleanly to a single new kind. These should be adjusted over time.

The constructors are private to the crate. While exposing them would
help transition, this also allows us to selectively and incrementally
rework our own error usage by deprecating them and fixing occurring
warnings. In particular, we should expose underlying errors as causes
and so on.

This commit is part of a patch series reworking error types. The next
patch switches the old error type for the new one.

* The format detection is switch to clean variants
* The pnm decoder was the only to utilize NotEnoughData and has been
  switched to a generic format error. It should not be an IO error.
* The external test matches on the renamed unsupported variant

Note that these are structural and thus also asserted on the interior
subtypes. (Unless an unsafe impl were to be added but that should never
happen in any case.)

All `new` constructors accept a detailed underlying description while
the shorthands and alternatives are renamed to `from_*_hint` and
`from_kind` respectively. This is supposed to encourage utilizing a
detailed error representation in dependencies with the new constructor.

Note that the UnsupportedError and LimitError have no `new` constructor
but utilize the naming convention regardless. This is to avoid blocking
the name if we introduce an underlying error representation for them.

These are not modified while storing within the error and should be
faded out over time. Storing the Box<str> saves some space in the
representation with the size of a pointer. Note that DecodingError is
actually the largest struct right now due to it supporting some legacy
representations.

Fix benchmarks to edition 2018

Error rework

Hide some public interface dependencies on gif and jpeg. Note that frame
oriented functions are now unused and will likely be brought back in a
(slightly less generic) manner that is more uniform over other multi
frame formats as well.

Note that some use within the library itself would use a buffer with
wrong buffer size or disregard width and height errors.

Interface restrictions for gif and jpeg

See the nightly feature `public-dependency`, tracking issue: #44663

Add travis script and public dependency reference

Two major reasons:

* This avoids the public dependency on `num_ratio`.
* The internal representation can be changed. This already proved
  necessary once (from u16 rational) and in this manner we are free to
  do it again by adding an inner enum representation.

Note that we already have rounding semantics for the meaning of Delay as
gif rounds to the nearest 10ms when writing frames. Any more or less
precise representation can be made to round to the closest u32 ratio.

With the original available as a millisecond ratio this can easily be
done to the accuracy of nanoseconds, which is the finest available.

Given the prior usage, it does not make much sense to provide an exact
conversion. However, it should be clear that precision is not finite and
the representation is not exactly that of a Duration.

The implementation turned out to be fairly involved as the num-fraction
crate does not provide approximation algorithms for fractions. No fear,
we code them ourselves with extensive testing to cover much more than
the required accuracy tests.

Encapsulate the delay ratio

Upstream from master