[libunwind] [libunwind][cmake] Compile _Unwind* routines with -fexceptions (PR #121819)

via cfe-commits cfe-commits at lists.llvm.org
Thu Jan 9 03:20:34 PST 2025


Andarwinux wrote:

> It does add `-fno-exceptions` even on the first bootstrap build for me. See e.g. https://github.com/mstorsjo/llvm-mingw/actions/runs/12662020869/job/35286224252#step:3:23134:
> 
> ```
> -- The C compiler identification is Clang 20.0.0
> -- The CXX compiler identification is Clang 20.0.0
> -- The ASM compiler identification is Clang with GNU-like command-line
> -- Found assembler: /home/runner/work/llvm-mingw/llvm-mingw/install/llvm-mingw/bin/i686-w64-mingw32-clang
> -- Detecting C compiler ABI info
> -- Detecting C compiler ABI info - failed
> -- Detecting C compile features
> -- Detecting C compile features - done
> -- Detecting CXX compiler ABI info
> -- Detecting CXX compiler ABI info - failed
> -- Detecting CXX compile features
> -- Detecting CXX compile features - done
> [...]
> -- Performing Test CXX_SUPPORTS_FUNWIND_TABLES_FLAG
> -- Performing Test CXX_SUPPORTS_FUNWIND_TABLES_FLAG - Success
> -- Performing Test CXX_SUPPORTS_FNO_EXCEPTIONS_FLAG
> -- Performing Test CXX_SUPPORTS_FNO_EXCEPTIONS_FLAG - Success
> ```

I'm sorry, but this was caused by my own mistake: my modified clang++ wrapper has a -lc++abi, which I forgot what it was previously introduced to solve.



> > If you build MinGW LTO libunwind without `-fno-exceptions` or with -`fexceptions`, libunwind references `__gcc_personality_seh0`, which results in undefined symbols.
> 
> Perhaps this is a symbol which we need to implicitly mark as needed in the linker? Because before the LTO, I think it only sees a definition of this symbol, but nothing referencing it, so the LTO logic thinks it doesn't need to retain/expose this symbol - while after the LTO, the generated code does add new references to it.
> 
> I don't find any precedence for doing this for LTO so far, but we have a couple cases of doing something similar for the regular GC in LLD.

Should I open a new issue for this? Since it won't actually happen in reality unless libunwind does switch to `-fexceptions` one day to compile the whole library, I'm not sure what to do.

https://github.com/llvm/llvm-project/pull/121819


More information about the cfe-commits mailing list