Note that the ergonomics are still not great, since `NoUninit` is also required to use `derive(ByteEq, ByteHash)`, and `derive(NoUninit)` doesn't support generics,
but that *explicitly* doesn't support generics, instead of silently ignoring generics like `derive(ByteEq, ByteHash)` used to, so this is strictly an improvement.
* simplify `ToTokens` impl for `Representation`
Instead of collecting the representation and modifier into `Option`s
and determining whether a comma is needed manually, we can use the
`Puncutuated` struct which handles commas automatically.
This will also make emitting the `align` modifier in the future easier.
* emit alignment modifier
This is required for correctly implementing `CheckedBitPattern` because
we need the layout of the type and its `Bits` type to have the same
layout.
* add unit test for `#[repr]` parsing
* allow multiple alignment modifiers
According to RFC #1358, if multiple alignment modifiers are specified,
the resulting alignment is the maximum of all alignment modifiers.
* actually return the error we just created
* factor out the integer Repr's into their own type
This is a preparation step for adding support for `#[repr(C, int)]`.
* allow parsing `#[repr(C, int)]`
This can be used on enums with fields.
* derive `CheckedBitPattern` for enums with fields
The implementation mostly mirrors the desugaring described at
https://doc.rust-lang.org/reference/type-layout.html
* add comments and rename some idents
* update error message
* update docs for `CheckedBitPattern` derive
* add new nested test case, change generated type naming scheme
* fix wrong comment
* small nit
---------
Co-authored-by: Gray Olson <gray@grayolson.com>
`from_integer` and `into_integer` are usually provided by the trait's
default implementation. We override this implementation because it goes
through `transmute_copy`, which can lead to inefficient assembly as seen
in https://github.com/Lokathor/bytemuck/issues/175 .
Doctests only run in the crate proper; doctests inside of integration tests are not run.
Move the existing doctests from derive/tests/basic.rs to derive/lib.rs, expanding them to make them
better for documentation as well.
* Add space because rust-analyzer thinks it's a weird literal otherwise.
* Add custom bounds for Zeroable.
* Cleanup custom bounds code.
* Only parse explicit bounds for Zeroable.
* Qualify syn types.
* If no explicit bounds are given, apply the default bounds.
* Add perfect derive semantics to `#[zeroable(bound = "...")]`.
This upgrades the derives from `syn` 1.0 to `syn` 2.0. While the MSRV
for `syn` 2.0 is not Rust 1.56 which is higher than the MSRV of
`bytemuck`, the derives don't fall under the MSRV policy.
Uses the compiler to check that all non-wrapped fields are actually 1-ZSTs,
and uses Zeroable to check that all non-wrapped fields are "conjurable".
Additionally, relaxes the bound of `PhantomData<T: Zeroable>: Zeroable` to all `T: ?Sized`.
* parse `repr(usize)` and `repr(isize)`
This can be used with enums.
* emit packed repr without type suffix
Previously we emitted the packed attribute with a type suffix for the
value `packed(4u32)`. This is not allowed. Instead we should emit
`packed(4)`.
* read bits instead of creating reference
This is required for packed structs where creating a reference to
fields is not allowed. We can make a copy of the bits because the bits
are `AnyBitPattern` which implies `Copy`.
* emit generics on implied traits
* Implement `ByteEq` and `ByteHash` derives
This adds the derives `ByteEq` and `ByteHash` that can be used as an
alternative to the `Eq` / `PartialEq` and `Hash` derives from the
standard library. The difference is that these variants use `bytemuck`
to convert their values to byte slices before comparing / hashing them.
This allows the comparisons to turn into a simple `memcmp` / `bcmp` (or
completely inlined as a few vector instructions) and allows hashers to
process all bytes at once, possibly allowing for some vector operations
as well.
Here's a quick comparison of the generated assembly:
* Address review comments
* allow deriving traits on unions in some cases
* basic union tests for nopadding and anybitpattern
* implement derives for unions for more traits
* remove Pod and AnyBitPattern derives for unions due to possible unsoundness
* add MaybePod and NoPadding traits
* MaybePod and NoPadding derive macros
* fix doctest
* fmt
* fix bad doc link
* move new casting functions into separate modules
* fmt
* fix doctest and derive test
* remove relaxed module, add anybitpattern
* rename MaybePod to CheckedCastFromPod
* rename checked casting functions
* rework CheckedCastFromPod into CheckedBitPattern
* add anybitpattern derive, fix up other derives
* fix doctest
* fix derive trait of bits type
* export AnyBitPattern derive
* export anybitpattern from traits
* actually export derive macro for AnyBitPattern
* make bits struct pub because of type leaking rules
* allow clippy lint in derive
* add copy bound to CheckedBitPattern
* - replace Pod bounds with NoPadding+AnyBitPattern
- add try and panic versions of checked cast functions
- slightly update docs
* fix derive tests
* - adapt the allocation module cast functions as well
- as part of that, make AnyBitPattern a subtrait of Zeroable
- AnyBitPattern derive also derives Zeroable
* @JakobDegen and @zakarumych nits
* superset -> subset on CheckedBitPattern and NoPadding docs
* derive Debug on generated `Bits` structs, which can be useful for debugging failures
* don't derive debug on spirv target arch
* make it work on 1.34
* merge conflicts
* fix erroneous behavior in doctest
* add basic derive macro for Pod and Zeroable for structs
* add derive macro for TransparentWrapper
* use core::mem::size_of instead of std::mem::size_of in generated code
* cleanup error handling a bit
* remove unneeded iter logic
* remove unneeded clone and order impl
* fix generics
* fix doc typo
Co-authored-by: Lucien Greathouse <me@lpghatguy.com>
* remove unneeded lifetime anotation
* use unreachable for already rejected patch
Co-authored-by: Lucien Greathouse <me@lpghatguy.com>
