Since this is the first post, I'm including some items that aren't strictly new changes in v0.29.0, but are recent changes nonetheless.
Added
"Constified enum modules" translating C/C++ enums into constants within a module for namespacing, rather than mangling the name of the generated constants.
/* automatically generated by rust-bindgen */
pub mod PetKind {
pub type Type = ::std::os::raw::c_uint;
pub const Doggo: Type = 0;
pub const Kitty: Type = 1;
pub const Hamster: Type = 2;
}
#[repr(C)]
#[derive(Debug, Copy)]
pub struct Pet {
pub kind: PetKind::Type,
pub noise: *mut ::std::os::raw::c_char,
}
The default translation strategy for enums will generate constants with names like PetKind_Hamster instead.
Use bindgen::Builder::constified_enum_module or --constified-enum-module.
You can now mark particular template instantiations as "opaque", so that bindgen emits a blob of bytes with the correct size and alignment rather than creating generic Rust types. This is useful as a workaround for when a template has a specialization for the given type arguments, which bindgen does not yet support. Previously, it was all of a templates' instantiations would be opaque or none of them would be. Use bindgen::Builder::opaque_type("SomeTemplate<Foo, Bar>") or --opaque-type "SomeTemplate<Foo, Bar>".
Added the ability to preprocess and dump the input headers given to bindgen to a file. This should make creating reproducible, system independent, standalone test cases much easier! Bring on the new issues! Use bindgen::Builder::dump_preprocessed_input or --dump-preprocessed-input.
We now use a fix-point analysis to determine whether any given type can derive Debug, or whether it has an explicit virtual table pointer. Previously we were using an ad-hoc algorithm that had at various times suffered from things like going into infinite loops when coming across cycles. Hopefully those kinds of bugs are a thing of the past! #767#765
No longer generating layout tests for template instantiations using type arguments that we didn't generate bindings for (which then caused compilation errors). #679
Fixed function name mangling when cross compiling bindings for iOS. #776
Don't include parent inline namespaces' names in types' names. Names of types from some STLs were showing up like std___cxx11_basic_string when they should have been std_basic_string. #789
Fixed a bug where we wouldn't generate type definitions for some types referenced by an opaque type's methods, causing compilation errors. #807
Fixed function name mangling issues for win32 targets. #819
Fixed a bug where bindgen was generating a generic type alias that didn't use its type parameter, which is illegal Rust code and caused compilation errors. #820
The generated size, alignment, and field offset unit tests now have stable names rather than sometimes including an internal identifier which is inherently unstable. This was causing unnecessary diffs when folks were checking in new versions of bindings into their VCS. #394
Fixed a bug where we would try and derive(Debug, Default) on structs that had padding like [u8; 33], which is larger than the largest array length for which Rust will derive traits. This would cause compilation errors when compiling the emitted bindings. #648
Friends
Thanks to everyone who contributed to these last couple releases!
Dylan McKay
Emilio Cobos Ălvarez
Holger Rapp
Kevin Lefevre
Manish Goregaokar
Nick Fitzgerald
Omar Akkila
Ralph Giles
Robin Lambertz
Shing Lyu
Travis Finkenauer
tz70s
UK992
Xidorn Quan
zzhu
Contributing
Want to join us? Check out our CONTRIBUTING.md and take a look at some of these issues:
Explicit control over choosing which Rust version (specific stable versions or
nightly Rust) to target. This defaults to the latest stable Rust version. #832
bindgen::Builder::default()
.rust_target(bindgen::RustTarget::Stable_1_19)
// or `.rust_target(bindgen::RustTarget::Nightly)` to use unstable features
or
$ bindgen --rust-target 1.19
# or `--rust-target nightly` to use unstable features
Started adding derive(Copy) for large arrays of Copy things, even when the
array is too large to derive(Clone) because Rust doesn't implement Clone
for arrays of length greater than 32. #874
bindgen can now determine which types are hashable and add derive(Hash) to
those types that support it. This is disabled by default, but can be enabled
via bindgen::Builder::derive_hash or --with-derive-hash. #876
bindgen can now generate impl Debug for Blah trait implementations for
types that contain non-Debug types, and therefore cannot derive(Debug). This behavior can be enabled with bindgen::Builder::impl_debug and --impl-debug. #875
bindgen can now invoke rustfmt on the generated bindings. The bindings
have historically been fairly pretty printed, but sometimes this is not the
case, especially with the new impl Debug for Blah feature. Have bindgen
run rustfmt with bindgen::Builder::rustfmt_bindings and --rustfmt-bindings, and use non-default rustfmt configuration files with bindgen::Builder::rustfmt_configuration_file and --rustfmt-configuration-file. #900
bindgen can now determine which types can be compared with == and add derive(PartialEq) to those types that support it. This is disabled by
default, but can be enabled via bindgen::Builder::derive_partialeq or --with-derive-partialeq. #878
Additionally, bindgen can also add derive(Eq) to those types which we
determined we could derive(PartialEq) and do not transitively contain any
floats. Enable this behavior with bindgen::Builder::derive_eq or --with-derive-eq. #880
Changed
Started emitting Rust unions when targeting stable Rust >= 1.19, not just
unstable nightly Rust. #832
Emitted layout #[test]s no longer contain internal IDs for template
instantiations including pointers and arrays. This should make generated
bindings more stable across updates to unrelated parts of the input
headers. #871
Determining whether a type can derive Copy or not was ported from an ad-hoc
algorithm to our fix-point framework. #766
Determining whether a type has a destructor or not was also ported from an
ad-hoc algorithm to our fix-point framework. #927
Deprecated
bindgen::Builder::unstable_rust/--unstable-rust is deprecated, in favor of
targeting explicit Rust versions with bindgen::Builder::rust_target/--rust-target instead. #832
Fixed
Fixed a regression in the derive(Default) analysis that resulted in some
opaque types deriving Default when they shouldn't have. #889
Fixed a regression where template instantiation layout #[test]s were being
generated with invalid Rust identifiers. #906
Friends
Thanks to everyone who contributed to this release!
some people could provide feedback: does this give you the information you feel like you'd need to dive head first into bindgen's code base? If not, what's missing or what would you change?
A new bindgen reviewer: @pepyakin You can ask @pepyakin to review all your future pull requests with r? @pepyakin from now on
Timers for seeing which phases bindgen is spending its time in. On the command line, use the --time-phases flag. From a builder, use the bindgen::Builder::time_phases(true) method. #938
You can now disable #[derive(Copy)] for all types with --no-derive-copy and bindgen::Builder::derive_copy(false). #948
We now have an overview of bindgen's code base and architecture for newcomers in CONTRIBUTING.md. #988
Derive PartialOrd with the --with-derive-partialord CLI flag or bindgen::Builder::derive_partialord(true) builder method. #882
Derive Ord with the --with-derive-ord CLI flag or bindgen::Builder::derive_ord(true) builder method. #884
When PartialEq cannot be derived because of an array larger than Rust's array-derive limit, bindgen can emit an impl PartialEq for ... block. Enable this behavior with the --impl-partialeq CLI flag or the bindgen::Builder::impl_partialeq(true) method. #1012
When deriving PartialEq for all types, you can now specify particular types that shouldn't derive(PartialEq) with the --no-partialeq <regex> CLI flag or bindgen::Builder::no_partialeq("<regex>") builder method. #996
Specify types that should not derive Copy with the --no-copy <regex> CLI flag or bindgen::Builder::no_copy("<regex>") builder method. This functionality was previously only available via comment annotations in the header sources. #1099
When deriving Hash for all types, you can now specify particular types that shouldn't derive(Hash) with the --no-hash <regex> CLI flag or bindgen::Builder::no_hash("<regex>") builder method. #1105
The bindgen users guide now has an FAQ section! If you have any FAQ suggestions to put up there, please open a pull request. [#1020]
Added csmith fuzzing infrastructure. csmith generates random C and C++ programs, we feed those into bindgen as headers to generate bindings to, then test that the generated bindings compile and that their layout tests pass. This infrastructure landed in many small bits.
Added experimental support for the thiscall ABI when targetting Rust nightly. #1065
Changed
If the user does not explicitly pass a --target argument for libclang, bindgen will insert such an argument itself. See #942, #947, and #953 for details.
C/C++ enums are now translated into constants by default, rather than Rust enums. The old behavior was a big footgun because rustc assumes that the only values of an enum are its variants, whereas a lot of C/C++ code uses random values as enums. Put these two things and it leads to undefined behavior. Translating C/C++ enums into Rust enums is still available with the --rustified-enum <regex> CLI flag and bindgen::Builder::rustified_enum("<regex>") builder method. #758
Generated bindings are now pretty printed with rustfmt by default. Previously, this option existed, but was off by default because syntex did an OK job at pretty printing the bindings. Now that we are using quote! { ... } instead of syntex, we lost that pretty printing, and now rely on rustfmt. You can disable rustfmting with --no-rustfmt-bindings or bindgen::Builder::rustfmt_bindings(false). See #925 and #1022 for details.
Deprecated
bindgen::Builder::hide_type is deprecated in favor of bindgen::Builder::blacklist_type. #987
bindgen::Builder::whitelisted_type is deprecated in favor of bindgen::Builder::whitelist_type. #987
bindgen::Builder::whitelisted_function is deprecated in favor of bindgen::Builder::whitelist_function. #985
bindgen::Builder::whitelisted_var is deprecated in favor of bindgen::Builder::whitelist_var. #989
Removed
Removed the dependency on (unmaintained) syntex, and build times are cut in half!
Before:
$ cargo clean; cargo build
<snip>
Finished dev [unoptimized + debuginfo] target(s) in 98.75 secs
After:
$ cargo clean; cargo build
<snip>
Finished dev [unoptimized + debuginfo] target(s) in 46.26 secs
The BindgenOptions type is no longer public. It had been deprecated in previous releases. Use bindgen::Builder instead. #1000
Fixed
Under certain conditions, a globally scoped enum could end up with bindings in the wrong namespace module. #888
Blacklisted types were incorrectly assumed to always be Copyable (and assumed to implement other traits as well). bindgen is now conservatively pessimistic about the traits that blacklisted types implement. #944
When bitfields have a ridiculously large number of bits (for example, unsigned : 632;) then bindgen was incorrectly deriving traits that couldn't be derived, resulting in errors when compiling the bindings, and was also generating structs with an incorrect layout. Both issues have been fixed. #982
_ is a valid identifier in some C++ contexts, but can't be referenced in Rust, as it is the "throwaway identifier" (a term I just made up, if you use it now, then you owe me money). bindgen will now translate _ into __ so that it can be used on the Rust side. #1008
Nested class definitions were sometimes being emitted in the wrong namespace module in the generated bindings. #1048
bindgen was mis-handling unions that contained bitfield members. This has been fixed. #744
Unsigned constants that were greater than u32::MAX were being mis-translated by bindgen. This is now fixed. #1040
When given a directory as an input file, or a file to which we don't have read permissions, then bindgen will print a more useful error message now. #1029
bindgen previously attempted to derive Hash for structures with flexibly-sized array members, but knowing how many elements exist in such arrays requires program-specific knowledge that bindgen cannot have. #1094
Friends
Thanks to everyone who contributed to this release!
Added support for bit-field allocation units that are larger than 64 bits
wide. Note that individual bit-fields within such units are still restricted
to being no wider than 64 bits. #1158
We can now generate random C header files and test that bindgen can process
them with the quickcheck crate. Initial support landed in #1159 with a
few more additions in follow up pull requests.
Changed
The bindgen::Builder::{constified_enum_module,{bitfield,rustified}_enum}
builder methods and their corresponding CLI flags now compare their argument
to the C/C++ enum's "canonical path", which includes leading namespaces,
rather than its "canonical name", which does not. This is a breaking change
that requires callers which target a namespaced C++ enum to call e.g. bitfield_enum("<namespace>::<enum_name>") rather than e.g. bitfield_enum("<enum_name>"). #1162
When a struct is packed to a smaller alignment that is still greater than one, bindgen cannot emit Rust bindings that match the input source. Before, it
would emit #[repr(packed)] anyways, which packs to an alignment of one, but
this can lead to misalignment and UB. Now, bindgen will detect these
situations and convert the struct into an opaque blob of bytes with the proper
alignment. We are eagerly awaiting support for #[repr(packed(N))] in
Rust. #1136
Fixed
There was a perfect storm of conditions that could cause bindgen not to emit
any bindings if spawning rustfmt to format the bindings failed. This is now
fixed. #1112
In some circumstances, bindgen would emit type parameters twice for
references to template instantiations. This is now fixed. #1113
When a C/C++ struct had a field named with a Rust keyword, and impl_debug
was enabled, the generated impl Debug for ... blocks could reference the
field by the Rust keyword name, rather than the non-keyword field name we
actually end up generating. This is now fixed. #1123
There was a regression in 0.31.0 where C++ template aliases to opaque types
would sometimes not treat the aliased type as opaque. This is now
fixed. #1118
There was a regression in 0.31.0 that could cause bindgen to panic when
parsing nested template classes. This is now fixed. #1127
Unnamed bit-fields do not affect alignment of their struct or class in C/C++,
however bindgen interpreted them as doing so, which could generate #[repr(C)] structs expecting to have an incorrect alignment. This is now
fixed. #1076
When a zero-sized type was used in a bit-field, bindgen could
divide-by-zero. This is now fixed. #1137
When a template parameter is used in a bit-field, bindgen would panic. This
is now fixed. #1140
There was a regression in 0.31.0 where if bindgen was given a header file
that did not exist, it would panic. This is now fixed, and it will instead
properly report the error. #1146
In some cases, generated bit-field getters and setters could access memory
beyond self. This is now fixed. #954
Friends
Thanks to everyone who contributed to the 0.31.1, 0.31.2, 0.31.3, and 0.32.0
releases!
Anthony Ramine
Bastien Orivel
Emilio Cobos Ălvarez
Glyn Normington
Nick Fitzgerald
roblabla
seemyvest
Sergey Pepyakin
Shea Newton
Tamir Duberstein
Trevor Reiff
Zhiting Zhu
Contributing
Want to join us? Check out our CONTRIBUTING.md and take a look
at some of these issues:
