Upstream https://bugs.kde.org claims that building Plasma 6 with LLVM and libc++ causing the plasmashell to crash is a Clang/libc++ bug and references the following ancient github issue from 2018 as the reason: https://github.com/llvm/llvm-project/issues/36746 The temporary suggestion is to build libc++ with DLIBCXX_TYPEINFO_COMPARISON_IMPLEMENTATION=2 to avoid these crashes, but: 1) As an average user, I don't know what unfortunate butterfly-effect issues the rest of the system would experience if I just applied a patch willy-nilly to the toolchain 2) The upstream commits related to this issue are similarly ancient (Not sure if they would even apply anyway) That said, getting this sorted out would make Plasma 6 much more usable under LLVM, as Plasma 5 is. Reproducible: Always Steps to Reproduce: 1. Use the LLVM profile 2. Set libc++ as the alternative to stdlibc++ for the system (rebuild subsequently, as needed, if this is not already the case) 3. Install Plasma 6 and SDDM (If not already) 4. Login to a Plasma 6 wayland session via SDDM (This is what I did) 5. Right click anywhere on the Plasma desktop Actual Results: Plasma 6 desktop crashes upon right-clicking any natural part of the Plasma desktop when built with LLVM. Expected Results: For Plasmashell to not crash upon right-clicks.
Also not, since I forgot to mention, that upstream bugs.kde.org already closed the issue as not their problem; therefore, I'm writing this to keep the issue catalogued since most distros don't support building systems with LLVM at all and because the upstream LLVM issue has been basically dead since 6 years ago.
Oh my. Yeah, I don't think we should workaround this on our end. Besides it looking like we'll end up having to use this "non-standard" code path forever (or at least until upstream removes support for it), I suspect it may change ABI, effectively breaking random packages until they are rebuilt.
My hope is that havibg written this issue will help able people realize, if they haven't already, who are willing to help to fix the issue upatream somehow in a way that wouldn't be hacky (such as attempting to fix it here, as you say).
