kikairoya wrote:

> How so? Can you be more specific how this would break dylib builds, and what 
> exactly is in the progress?
>
> As mentioned in the comment dllimport attribute is not strictly required, 
> it's safe to remove it.

For code symbols, the attribute isn't strictly required, just suboptimal; 
lacking of `dllimport` causes additional jumps. However, for data symbols, it's 
mandatory (except for MinGW auto-import feature). See also: 
https://github.com/llvm/llvm-project/issues/149639#issuecomment-3114257062.
I'm not sure about the current progress of dylib builds for Windows.

`CLANG_ABI` is supposed to annotate `clang-cpp.dll` (which exposes C++ 
interfaces), but not `libclang.dll` (which exposes C interfaces). I think the 
problem is that `libclang.dll` is exporting C++ interfaces; that looks 
unintended.

The issue report uses a GNU style command line for the windows-msvc target, but 
such that configuration doesn't seem to have been tested well. Perhaps some of 
`if (MSVC)` blocks should be changed to `(WIN32 AND NOT MINGW)`.

https://github.com/llvm/llvm-project/pull/171054
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to