Markus =?utf-8?q?Gschoßmann?= <markus.gschossm...@tum.de>, Markus =?utf-8?q?Gschoßmann?= <markus.gschossm...@tum.de> Message-ID: In-Reply-To: <llvm.org/llvm/llvm-project/pull/130...@github.com>
================ @@ -0,0 +1,16 @@ +// RUN: %clang_cc1 -triple x86_64-none-linux-gnu -emit-llvm -debug-info-kind=limited %s -o - | FileCheck %s ---------------- mgschossmann wrote: Thanks for your comment. Indeed, my code changes behavior when compiling for i686-pc-windows-msvc, which I did not expect. However, there seems to be the same problem: There is an implicit parameter `should_call_delete`, whose `DILocalVariable` has `arg: 2` even though there is no second argument in the function type's `types` array. Similarly to `vtt`, the patched code adds this parameter to the array. Additionally, it changes the return type of the destructor to a void pointer. I am unsure, why this happens, since that does not happen on ItaniumABI. Does the MSVC ABI feature an implicit void return? Unfortunately, I am lacking knowledge of the MSVC ABI to be able to judge what the correct behavior should be. Following are clang outputs for `debug-info-vtt.cpp`: [clang_outputs.zip](https://github.com/user-attachments/files/19509545/clang_outputs.zip) PS: I am unsure whether the bug is that the AST is not correctly updated in `clang/lib/CodeGen/ItaniumCXXABI.cpp` (and `clang/lib/CodeGen/MicrosoftCXXABI.cpp`) or that the AST is not correctly read in `clang/lib/CodeGen/CGDebugInfo.cpp`. In the former case, my patch feels like a workaround. https://github.com/llvm/llvm-project/pull/130674 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
