| Age | Commit message (Collapse) | Author |
|---|
|
Prepare 0.64.0 Release
---------
Co-authored-by: Christian Poveda <christian.poveda@ferrous-systems.com>
|
|
* Generate extern wrappers for inlined functions
If bindgen finds an inlined function and the
`--generate-extern-functions` options is enabled, then:
- It will generate two new source and header files with external
functions that wrap the inlined functions.
- Rerun `Bindings::generate` using the new header file to include these
wrappers in the generated bindings.
The following additional options were added:
- `--extern-function-suffix=<suffix>`: Adds <suffix> to the name of each
external wrapper function (`__extern` is used by default).
- `--extern-functions-file-name=<name>`: Uses <name> as the file name
for the header and source files (`extern` is used by default).
- `--extern-function-directory=<dir>`: Creates the source and header
files inside <dir> (`/tmp/bindgen` is used by default).
The C code serialization is experimental and only supports a very
limited set of C functions.
Fixes #1090.
---------
Co-authored-by: Amanjeev Sethi <aj@amanjeev.com>
|
|
|
|
Fixes #2396.
This makes it possible to workaround cc/bindgen issues with esp-rs
projects by using only environment varaibles (TARGET_CC, CLANG_PATH,
etc). Without this, it requires modifying each crate's build.rs that
you try to depend on to add a target option passed along to clang.
|
|
* custom derives after DeriveInfo
* Introduce `TypeKind` instead of `CompKind`
* Add tests
* Emit CLI flags for callbacks
* update changelog
* run rustfmt
* fix tests
* fix features
Co-authored-by: Christian Poveda <christian.poveda@ferrous-systems.com>
|
|
* Migrate CLI argument parsing to the derive API
* Fix docs
* Fix tests
* Use `override_usage`
* Update changelog
|
|
|
|
* gh action for merging main to master on pr merge
* remove pull_request as trigger. only use push, that is after merging the pull_request
|
|
* Update to clap 4.
* Bump MSRV to minimum required by clap 4.
* Mention clap update in CHANGELOG.
|
|
|
|
Signed-off-by: Amanjeev Sethi <aj@amanjeev.com>
|
|
* codegen: Look through typedefs to detect void return type.
And reuse a bit more code.
Should fix #2377, but needs a test (can't run tests atm).
* Add tests
* Run rustfmt
* Update changelog
Co-authored-by: Christian Poveda <christian.poveda@ferrous-systems.com>
|
|
|
|
* Document how to format code
* Fix typo
* Fix formatting
* Ignore code snippet
|
|
|
|
|
|
* Bump `clang-sys` version
* add libtinfo5 package to fix CI
|
|
The bug only affects virtual inheritance, so instead disable layout
tests in the test that we know is broken. Not generating layout tests is
wrong anyways, because the offset would be wrong.
|
|
|
|
* This change updates `ParseCallbacks::generated_name_override` to accept a second parameter indicating the kind of item the name applies to (currently, either `Function` or `Var`).
* A `CallbackItemKind` enum was added to serve as the type for this second parameter.
* Tests have been updated to verify that the names of both function and variable can be updated by this callback.
|
|
|
|
Fixes #2008.
Example:
```c
enum Enum { Variant };
typedef int16_t Enum;
```
This is valid and idiomatic C (though not valid C++). `cbindgen` uses this idiom as the default C translation of Rust enums, the equivalent of what would be `enum Enum : int16_t { Variant };` in C++.
`bindgen header.h` before:
```rust
pub const Enum_Variant: Enum = 0;
pub type Enum = ::std::os::raw::c_uint;
pub type Enum = i16;
```
```console
error[E0428]: the name `Enum` is defined multiple times
--> generated.rs:3:1
|
2 | pub type Enum = ::std::os::raw::c_uint;
| --------------------------------------- previous definition of the type `Enum` here
3 | pub type Enum = i16;
| ^^^^^^^^^^^^^^^^^^^^ `Enum` redefined here
|
= note: `Enum` must be defined only once in the type namespace of this module
```
After:
```rust
pub const Enum_Variant: Enum = 0;
pub type Enum = i16;
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
This reverts commit e8ffb42ab66405ac56d04494a30e54b584f2d4dd and adds a new `--wrap-unsafe-ops` option as a workaround.
|
|
This method can be used to process comments and replace them with whatever the user wants.
|
|
This PR introduces a new non-exhaustive `struct` called `DeriveInfo` to be used as the sole argument of `ParseCallbacks::add_derives` with the purpose of being able to extend the information passed to this method in a backwards-compatible manner, meaning that adding new fields to `DeriveInfo` won't be a breaking change when releasing a new version.
|
|
|
|
|
|
|
|
|
|
|
|
Even though this change does name deduplication in a slower way, it
avoids name collisions without any breaking changes in the test suite.
Fixes #2202
|
|
Given that C keeps a different namespace for `struct`/`enum`/`union` and `typedef` aliases. The
following patterns
```c
typedef const struct foo {
void *inner;
} *foo;
typedef struct bar {
void *inner;
} *bar;
```
are valid C code and produces both a `struct` and a pointer called `foo`
and `bar` in different namespaces. Given that Rust does not make this
distinction, we add the `_ptr` prefix to the pointer type aliases to
avoid any name collisions.
|
|
|
|
This fixes #2325.
The issue is that `__bf16` is not exposed at all by libclang, which
causes us to crash. It's a bit of a shame libclang doesn't expose it but
there's no rust equivalent I think, so this should be ok for now.
Unfortunately no test because the header crashes older clang versions.
|
|
|
|
This guarantees that bindings generated by `bindgen` compile even if the `unsafe_op_in_unsafe_fn` lint is denied.
|
|
* Clean the implementation of `Default` for `BindgenOptions`
|
|
* Add support for the `"C-unwind"` ABI
This allows using `"C-unwind"` as an ABI override if the rust target is
nightly.
|
|
* Add the `--override-abi` option.
This option can be used from the CLI with the <abi>:<regex> syntax and
it overrides the ABI of a function if it matches <regex>.
Fixes #2257
|
|
* Allow callback composition
Store all the callbacks added to the builder in a `Vec` so bindgen
invokes each one of them in a last-to-first manner.
|
|
|
|
* Run `Bindings::generate` again if required.
This adds a mechanism so `bindgen` is able to run `Bindings::generate`
multiple times with the same user input if the `generate_static_inline`
option is enabled and `GenerateResult::ShouldRestart` is returned by
`Bindings::generate`.
This is done to eventually solve #1090 which would require to check for
any static inline functions and generate a new header file to be used as
an extra input and run `Bindings::generate` again.
|
|
* add tutorial about non-system libraries
* fix broken comment
* ignore code snippet
|
|
The original commit adding this note referenced #778, but then went on to talk about how the problem was RVO and that the problem is not possible to solve in bindgen because bindgen can't know if RVO happened or not. However this is incorrect in several ways. The problem in #778 has nothing whatsoever to do with RVO but rather with bindgen simply not understanding the calling convention for passing or returning types that are "non trivial for the purposes of calls". This is completely consistent and does not depend on what the optimizer decided to do (after all, if callsite side calling-convention did depend on what the optimizer happened to do inside the function it would be impossible for C++ compilers to emit correct calls to external C++ functions!).
You can see this quite clearly [here](https://godbolt.org/#z:OYLghAFBqd5QCxAYwPYBMCmBRdBLAF1QCcAaPECAMzwBtMA7AQwFtMQByARg9KtQYEAysib0QXACx8BBAKoBnTAAUAHpwAMvAFYTStJg1DIApACYAQuYukl9ZATwDKjdAGFUtAK4sGIM6SuADJ4DJgAcj4ARpjEEgDspAAOqAqETgwe3r7%2ByanpAiFhkSwxcVyJdpgOGUIETMQEWT5%2BAVU1AnUNBEUR0bEJtvWNzTltwz2hfaUDFQCUtqhexMjsHAD06wDUACoAnkmYW3tLxFtoWFsIsZikWyRbtKhM6FuGW5iqrEn0AHQmGgAggoCMQvA4tgRMCCokwlFsTPErPEACImADMViBILBEKhIIRSIBgK2pMh0IIEDmGKxJLJ%2BMpaAYBIZ5gAbNTMcSyVsAG6oPDoABUkJpxMRaK5QIZWwUAEcvA1MFTCbSeTKCIjkZK1WTiJgCMsGKKpYCJeKgaECFsWExQiqtdz6RTIQj0SjZQqlVSxUCnaT9YbiMa0gAvTCoKgQBmc2kSjgLWicACsvD8HC0pFQnDc1msstOqwRZnRPFIBE0CYWAGsQOj0b9603my22fpOJI05Ws5xeAoQBpy5WFnBYEg0CwknRYuRKBOp/Q4sAFMwkgoEKgCKQsLy8KsAGp4TAAdwA8od02WaLQocR%2BxAot2oqEGntOGWJ2xBKeGLQ3xneCwW0jHEADtzwfUal5aFu0%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%2Bzpl5ny/O89sOFTWruxCsLIpiuKEqSlKvDSjKICy0nizygrhxK%2BtGwq8qVZqureAa2wmsKrRipVswgvq3tmo9hZoLvDIQEkIA) Notice how the body of `main` is identical no matter if copy elision is enabled or not. To get `test` passed in a register you must remove the copy constructor.
I spent a few hours being really confused by this note before I tracked down the original PR and realized what it was trying to talk about. Hopefully this saves the next person to come across it the same trouble.
As an aside I suspect the clang c bindings can already give you information on special member functions and base classes, making it fairly easy to do this correctly. AFAICT bindgen already does not rely on llvm's code generation facilities to figure out the ABI of calls for C (except perhaps indirectly, via rustc), so this seems reasonable to just do in bindgen, explicitly.
|
