Template alias full and partial specialization improvements. #255
Conversation
| @@ -630,15 +634,29 @@ impl<'ctx> BindgenContext<'ctx> { | |||
| // argument names don't matter in the global context. | |||
| if (declaration.kind() == CXCursor_ClassTemplate || | |||
| declaration.kind() == | |||
| CXCursor_ClassTemplatePartialSpecialization) && | |||
| CXCursor_ClassTemplatePartialSpecialization || | |||
| declaration.kind() == CXCursor_TypeAliasTemplateDecl) && | |||
There was a problem hiding this comment.
It seems we would benefit from a Cursor::is_template() method here, that we would use like declaration.is_template() and replace all these kind comparisons.
As it is right now, this condition is super intense.
There was a problem hiding this comment.
Yup, I agree, but we special-case ClassTemplate/ClassTemplatePartialSpecialization in a few places, but TypeAliasTemplateDecl only here. I'll try to organize it better.
| // This is _tricky_, I know :( | ||
| if declaration.kind() == CXCursor_TypeAliasTemplateDecl && | ||
| ty.canonical_type().is_valid_and_exposed() { | ||
| println!("{:?}", ty.canonical_type()); |
There was a problem hiding this comment.
Use warn! or debug! instead of println!. Also, a diagnostic message is needed.
There was a problem hiding this comment.
Whoops, that should've gone away, sorry, will clean it up when I'm in front of my computer.
| @@ -7,7 +7,8 @@ | |||
| #[repr(C)] | |||
| #[derive(Debug, Copy, Clone)] | |||
| pub struct Rooted<T> { | |||
| pub ptr: MaybeWrapped<T>, | |||
| pub ptr: ::std::os::raw::c_int, | |||
There was a problem hiding this comment.
This is a regression. We used to correctly have MaybeWrapped<T>, but now we have c_int. I don't think we should land this patch until this regression is fixed.
There was a problem hiding this comment.
Not quite a regression since it's really an int. Unfortunately there's no way to grab the specialized members from clang. I might take some time to make numTemplateArgs work for this, but until that this is the best we can do.
This was working just by chance, because MaybeWrapped's template argument just happened to be T, and we grabbed the top-level type. If instead of that it was Rooted<U> -> MaybeWrapped<U>, it'd failed (there's a test-case for that).
I guess I can special-case if the spellings are equal to keep the original type, but that's... not really ideal.
|
Ok, I fixed the regression explicitly looking for |
|
Libclang patch to eventually fix this hackery: https://reviews.llvm.org/D26663 |
|
@bors-servo r=fitzgen |
|
📌 Commit 2003f9c has been approved by |
|
⚡ Test exempted - status |
Instead of generating a enum like this:
```
pub enum Enum_CUipcMem_flags_enum {
CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS = 1,
}
```
Which is a compiler error because repr can't be used with univariant enums.
See rust-lang/rust#10292
We generate a two variant enum with a dummy variant.
```
pub enum Enum_CUipcMem_flags_enum {
CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS = 1,
__DUMMY,
}
```
Which is the recommend workaround to the previous compiler error.
closes rust-lang#255
This doesn't completely fix #251, but fixes other related problems.
r? @fitzgen