Martin Kinkelin

How come this doesn't show up as a CI problem?

This has landed from upstream now.

I've tried to add the package in #4931, but hit a blocker while trying to link `ldc2.exe` (using the prebuilt LLVM from https://github.com/ldc-developers/llvm-project/releases/tag/ldc-v20.1.5). And I'm pretty sure we'd hit failures...

Yeah thanks for the report, we're hitting assertions earlier, related to the dark corners around `DFuncValue` (doesn't handle `Tdelegate`, `DtoRVal()` yields the function pointer only). I'll try to fix that.

This can hardly be the general case, as macOS CI includes backtrace tests and is happy on both x86_64 and arm64. So at least some aspects of this seem specific...

As said above, you need an additional `-link-defaultlib-debug` to make sure druntime & Phobos keep the frame pointers. Then my understanding is that Apple's ld64 linker at one point decided...

Btw guys, what is that .dSYM bundle? From the linked stackoverflow post: > dsymutil can be thought of as a debug info linker. It does this same process -- read...

> At least on macOS, there is the backtrace/backtrace_symbols API that AFAIR worked fine in my tests a while ago, could that be an alternative? No, `atos` is specifically only...

One thing I could easily do is restoring the `-preserve-dwarf-line-section` workaround, by re-[hacking](https://github.com/ldc-developers/llvm-project/commit/de3d950f5c8b6ab89c4d71eebb628077b419fabe) the upcoming LLVM 20 binaries used for the *official* LDC binaries from GitHub. > However, the x86-64...

Okay as expected, and if you had tested `-link-defaultlib-debug` as suggested earlier, we should have known in advance. ;)