tiny-skia-path

Update version numbers and changelog for v0.12.0 release.

cargo-semver-checks says that tiny-skia-path also requires a
semver-breaking release (and it looks like historically it's version has
been locked to that of tiny-skia anyway), so I've also updated it's
version number.

Signed-off-by: Nico Burns [email protected]

Fixed

• Fixed a panic when drawing small anti-aliased rectangles. Thanks to @nwhitehead

Added

• 2-point conical gradient support for (RadialGradient). Thanks to @wmedrano
• Sweep gradient support (SweepGradient). Thanks to @wmedrano
• Support for a small number of color spaces. Thanks to @danieldg
• Added the is_empty and join methods to Rect. Thanks to @danieldg
• Implement Hash for PathVerb, Rect and Size and IntSize. Thanks to @LaurenzV
• Add support for relaxed SIMD in WebAssembly. Thanks to @CryZe
• Add the from_rgba_unchecked method to Color so it can be created in const contexts. Thanks to @jeffreyrosenbluth
• Add the scale_by and scale_to method to Size. Thanks to @jermy
• Add the take_demultiplied method to Pixmap. Thanks to @JustForFun88
• Add the take method to Mask. Thanks to @valadaptive
• Add the round_out method to Rect. Thanks to @valadaptive

Changed

• The RadialGradient::new requires a start radius now. Set the second argument to 0.0 to preserve the old behavior.

Fixed

• Path::compute_tight_bounds calculation. Thanks to @qbcbyb

Added

• Path::compute_tight_bounds

Changed

• MSRV bumped to stable, because of the flate2 crate.

Fixed

• Transform::is_valid was treating 1/4096 as zero. We need a higher precision.
• Build failure on AVX2 with --no-default-features. Thanks to @linkmauve

Changed

• MSRV bumped to 1.60, because of the log crate.

Fixed

• LineJoin::MiterClip handling with small miter_limit. Thanks to @torokati44

Added

• LineJoin::MiterClip. Thanks to @torokati44

Changed

• Rect::inset and Rect::outset are no longer require a mutable self. Thanks to @wezm

Added

• PathBuilder::push_path
• NonZeroRect
• Size
• Rect::transform
• Rect::bbox_transform
• Transform::from_bbox
• Transform::is_valid
• Transform::get_scale
• Transform::pre_rotate
• Transform::post_rotate
• Transform::pre_rotate_at
• Transform::post_rotate_at
• Transform::map_point

Changed

• ColorU8 and PremultipliedColorU8 are stored as [u8; 4] instead of u32. This fixes potential alignment issues and make the code easier to understand. Thanks to @e00E
• PathBuilder::push_rect accepts Rect and not f32 numbers now.

Removed

• tiny_skia_path::ScreenIntRect. It become private.
• Path::is_empty. Path cannot be empty by design.

Removed

• ColorU8::get and PremultipliedColorU8::get. Use the getters instead.

Added

• Mask::from_vec
• Mask::from_pixmap to convert Pixmap into Mask by extracting alpha or luminance.
• Mask::width
• Mask::height
• Mask::data
• Mask::data_mut
• Mask::fill_path
• Mask::decode_png
• Mask::load_png
• Mask::encode_png
• Mask::save_png
• Mask::invert
• MaskType
• Pixmap::apply_mask

Changed

• Rename ClipMask into Mask.
• Mask is closer to a 8bit (A8) Pixmap now, rather than being its own thing.
• Mask::new requires width and height arguments now.
• Drawing on Mask using Mask::fill_path uses our SIMD pipeline now instead of a scalar code that should make it a bit faster.
• Painting API no longer returns Option<()>, but simply adds a warning to the log.
• Paint::anti_alias is set to true by default.

Removed

• Mask::set_path. Use Mask::fill_path instead.
• Mask::default(). Mask cannot be empty anymore.

Added

• Implement PartialEq for Paint and subtypes. Thanks to @hecrj

Changed

• MSRV bumped to 1.57, mainly because of the png crate.

Fixed

• ClipMasks larger than 8191x8191 pixels. Previously, the creation of a large mask via ClipMask::set_path would have created an empty mask.

Fixed

• Performance regression, probably due to LLVM update in Rust. Thanks to @mostafa-khaled775
• Big-endian targets support. Thanks to @ids1024

Added

• Pixmap::from_vec.

Fixed

• Increase Conic to Quad conversion precision. This allows us to produce nicer round caps. Previously, they were not as round as needed.

Fixed

• Conditional compilation of FasterMinMax on fallback platforms. Thanks to @CryZe

Added

• AArch64 Neon SIMD support. Up to 3x faster on Apple M1. Thanks to @CryZe

Changed

• FiniteF32, NormalizedF32 and NonZeroPositiveF32 types have been moved to the strict-num crate.
• Rename NormalizedF32::from_u8 into NormalizedF32::new_u8.
• Rename NormalizedF32::new_bounded into NormalizedF32::new_clamped.
• Use explicit SIMD intrinsic instead of relying on safe_arch.
• MSRV bumped to 1.51

Added

• tiny-skia-path dependency that can be used independently from tiny-skia. It contains the tiny-skia Bezier path implementation, including stroking and dashing. As well as all the geometry primitives (like Point and Rect).

Changed

• When disabling the std feature, one have to enable no-std-float feature instead of libm now.