Bug fixes

• #[repr] attributes containing lists like align(..) are now properly handled. Thanks, @minty99! (#132)

Breaking changes

• The MSRV is now 1.76. This is not currently a hard requirement of modular-bitfield itself but is instead the path of least resistance due to changes to underlying dev-dependencies, so please open a ticket if you really need support for an even older version of Rust.
• From<[u8; {n}]> (when conversions are guaranteed) and TryFrom<[u8; {n}]> traits are now automatically derived by #[bitfield]. This will conflict with any manual implementations of the same trait, but you wrote those implementations to do this thing anyway, right? (#120)
• The BitfieldSpecifier alias, which was deprecated in v0.12, is removed.

Enhancements

• Diagnostic messages around invalid bit counts have been improved.
• Minor documentation improvements for the Specifier trait.

Bug fixes

• Unnecessary brackets in the generated output which were triggering new Clippy lints in Rust 1.90+ have been fixed. (#130)
• Unnecessary identity operations in the generated output which were triggering Clippy lints in some cases have been fixed. (#131)

Breaking changes

• The MSRV is now 1.70. This is not currently a hard requirement of modular-bitfield itself but is instead the path of least resistance due to changes to underlying dev-dependencies, so please open a ticket if you really need support for an even older version of Rust.
• The InvalidBitPattern::invalid_bytes field is now private. Use the InvalidBitPattern::invalid_bytes() method to access the invalid bytes (and InvalidBitPattern::new() to create new errors).

Deprecations

• To conform with idiomatic Rust code, the derive macro #[derive(BitfieldSpecifier)] has been renamed to #[derive(Specifier)] because it derives the Specifier trait. The deprecated alias will be removed in the next release of modular-bitfield.

Enhancements

• The syn dependency has been updated to version 2.
• Generated constructors and getters are now annotated with #[must_use]. (#79)
• Structs with const generics can now use #[bitfield]. (#87)

Bug fixes

• Enumeration variant names will no longer conflict with code generated by #[derive(Specifier)]. (#82)
• Using #[bitfield] on an empty struct now emits an appropriate diagnostic instead of causing a compile-time arithmetic error. (#104)
• Various fixes to lints. (#62, #66, #93, #95. Thanks, @dimpolo, @diseraluca, @nilfit, and @LDVSOFT!)
• Various fixes to documentation formatting. (#91, #101, #103. Thanks, @00xc, @SebastianJL, and @jruderman!)
• When automatically deriving Debug, tuple-like structs are now formatted normally instead of like structs with numeric fields.

0.11.2 (2020-11-07)

• Fixed a bug that all but the first #[skip(..)] attribute for a bitfield were ignored despite proper error handling.
  E.g. when flagging a field with #[skip(getters)] #[skip(setters)] the #[skip(setters)] was ignored.

0.11.1 (2020-11-07)

• Allow non-overlapping #[skip(getters)] and #[skip(setters)] attributes. This might be useful for conditional compilation.
• Fixed a bug with #[skip] skipped fields not properly bumping the bit offset
• Fixed a minor bug where duplicate #[skip] attributes would yield a confusing error.

0.11.0 (2020-11-06)

• Add bits: int parameter to the #[bitfield] macro which allows to precisely control the resulting bit width
  of the generated bitfield struct. Use it like: #[bitfield(bits = 5)].
• Replace #[bitfield(specifier = true)] syntax with #[bitfield] #[derive(BitfieldSpecifier)].
  Semantics of the new syntax is the same as the old.
• It is now possible to flag fields of #[bitfield] structs with #[skip(..)] in order to skip code generation for them.
  There are possibilities to skip code generation of only setters using #[skip(setters)], only getters #[skip(getters)]
  or both. Having no arguments (e.g. just #[skip]) defaults to skipping both setters and getters.
  A neat trick is to specify double wildcards as identifiers for skipped fields to avoid having the need to come up with
  an identifier for them: For example: #[skip]: __: B10
• Attributes applied to #[bitfield] fields are now properly propagated to their generated getters and setters.
  Note thought that it might be confusing that an attribute applied to a struct field is actually applied to a function
  through macro expansion.
• Fixed several bugs and significantly improved error reporting for #[bits = N] field attributes for #[bitfield] fields.
• Minor fixes and improvements to code generation:
  • The automatically implemented #[derive(Debug)] now produces code that the Rust 2015 edition will accept.
  • Some minor macro hygiene improvements.

0.10.0 (2020-10-29)

• (Thanks @jam1garner): The #[bitfield] macro now looks for #[derive(Debug)] annotations and if found will implement
  one that formats the struct with its fields as a user would expect them. Previously having a #[derive(Debug)]
  annotation on a #[bitfield] struct would simply print the contents of the generated internal byte array.
• Implement #[bitfield(bytes = N)] parameter to ensure at compile time that the bitfield struct
  requires N bytes of memory.
• Implement #[bitfield(filled: bool)] parameter to control whether the bitfield struct ensures that either
  all bits are well defined (filled = true) or if there are going to be some bits intentionally undefined (filled = false).
  The default is filled = true.
• Implement #[repr(uN)] for #[bitfield] structs where uN is one of u8, u16, u32, u64 or u128 with which it is
  possible to control whether a bitfield allow conversions between uN and the bitfield. Also it ensures at compile time that
  the bitfield requires the exact same amount of bits. This is in conflict with filled = false.
• Bitfield structs are no longer implicitly #[repr(transparent)]. If a user wants their bitfield struct to remain transparent
  they have to add #[repr(transparent)] manually to their struct definition.
• The default behaviour for #[bitfield(specifier = bool)] got changed so that it no longer by default allows for
  unfilled (filled = false) or undefined bits. Instead users now have to additionally add filled = false if they
  explicitly want this behaviour.
• Renamed the generated as_bytes method for #[bitfield] structs to into_bytes. It now takes self instead of &self
  and returns the byte array by value instead of reference. This change was necessary for working properly with the new
  #[repr(uN)] feature.
• Fixed a bug with #[bitfield(specifier = true)] bitfields that were not aligned to have a power-of-two bytes.

0.9.0 (2020-10-26)

• Add #[bitfield(specifier = bool)] parameter with which it now is possible to have bitfield structs automatically also
  implement the modular_bitfield::Specifier trait which makes it possible to have bitfields as fields of bitfields.
• No longer generates an unsafe fn from_bytes_unchecked. Now generates a safe fn from_bytes that is basically identical.
  The difference is that we no longer consider bitfields containing invalid bit patterns as invalid since generated getters
  will protect their access anyways.
• Update crate documentation and README.

Release Notes: 0.8.0 (2020-10-25)

• The #[derive(BitfieldSpecifier)] now allows an amount of variants that is not a power of two via
  the new attribute #[bits = N] where N is the bit width of the deriving enum. (Thanks @lkolbly)
• Add Specifier implementations for u8, u16, u32, u64 and u128.
• The #[bitfield] macro now additionally generates getters for expected failures just as it already
  does for the various setters. For a field a the new fail-safe getter is called a_or_err and returns
  a Result.
• Silence repetitive dead_code warnings originating from generated #[bitfield] getters and setters.
• Improve error span information in a few use cases.
• Cleaned up backend code for modular_bitfield crate and its generated code.

Release Notes: 0.7.0 (2020-10-18)

• Tons of macro generated code hygiene improvements for both #[bitfield] and #[derive(BitfieldSpecifier)].
• #[bitfield] now produces with_{} and with_{}_checked methods that take self and return Self. (Thanks @lkolbly)
  • This is useful for building up your bitfields without the need of a mutable reference.
• #[bitfield] generated getters now are the same identifiers as their fields. (Thanks @Qyriad)
  • Previously for a field a it generated fn get_a(..) but now it generates fn a(..) instead.
• #[bitfield] now properly inherits visibility modifiers for all fields onto their generated getters and setters. (Thanks @WorldSEnder)
• Remove TryFrom implementation from #[bitfield] generated structs since it never was really safe.
  • It didn't check for invalid bit patterns and thus could trigger undefined behaviour.
• The #[bitfield] generated new constructor is now a const fn.
• Rename the #[bitfield] generated to_bytes to as_bytes.
  • Also it now returns a refernce to an array, e.g. &[u8; N] instead of a slice.
• The #[bitfield] macro now generates an unsafe fn from_bytes_unchecked that allows constructing from raw bytes.
  • This is meant as an unsafe replacement for the removed TryFrom implementation.
  • We plan to create safer alternatives in future updates.
• Updated all benchmarks to use the criterion crate.
• Fixed a parsing bug that enum variant names sometimes could clash with Rust keywords. (Thanks @Qyriad)
• Fixed a bug that caused getters to sometimes read with extra invalid bits. (Thanks @crzysdrs)
• Fixed an issue that was caused by single bit manipulations. (Thanks @crzysdrs)