Previously the function assumed that the version number appeared in the
third word. This PR adds a heuristic - take the first word that starts
with a number.
This way we can also parse: `debian clang version 11.0` that my clang
reports.
Instead of always generating the _bindgen_union_align method (which
shouldn't be needed at all for Rust structs, since the struct layout
tracker already deals with adding repr(align) as necessary) make sure to
visit all fields appropriately to generate the correct alignment.
Fixes#1947.
There are two separate issues here: First, the change in comp.rs ensures
that we don't round up the amount of storage to the alignment of the
bitfield. That generates the "expected" output in #1947
(`__BindgenBitfieldUnit<[u8; 3], u16>`).
But that's still not enough to fix that test-case because
__BindgenBitfieldUnit would be aligned and have padding, and Rust won't
put the extra field in the padding.
In order to ensure the bitfield starts at the right alignment, but that
Rust can put stuff in the extra field, we need to make a breaking change
and split the generated fields in two: One preceding that guarantees
alignment, and the actual storage, bit-aligned.
This keeps the existing behavior while fixing that test-case.
Add a `--fit-macro-constant-types` option to make bindgen try to fit
macro integer constants into types smaller than u32/i32.
Useful especially when dealing with 8/16-bit architectures.
Closes#1945
This can cause unnecessary padding to be computed otherwise at the end
of the struct.
With repr(packed(n)), a field can have padding to adjacent fields as
long as its alignment is less than n. So reuse the code we have to align
to a field layout, aligning to the struct layout instead.
Fixes#1934
Sometimes, we need customize the implement of `Default` for certain types,
In these cases, the `nodefault` annotation can be used to prevent bindgen
to autoderive the `Default` traits for a type.
Right now trying to generate a dynamic library with variadic functions
panics because we don't account for the extra `...` in the arguments.
Keeping the current interface for variadic functions is tricky, as we
cannot "wrap" a variadic function (VaList[1] is nightly-only).
However, we don't need to. We're already exposing the libloading error,
so exposing the function pointer field as public is just fine and allows
consumers to call the variadic function.
At that point the can_call() / CheckFoo libraries become pointless (you
can just do library.function.is_ok() or such), so we can simplify the
code as well removing those.
[1]: https://doc.rust-lang.org/std/ffi/struct.VaList.html
