Martin Kinkelin

> Not sure what version that is in relation to what's being used by ldc 1.33 DMD/Phobos v2.103. The nightly DMD version on godbolt is v2.091.0-beta.2-master-ec39fe5... I've tested it here...

What's the error that you are seeing? And only with `-link-internally`, the MS linker works?

Just a guess, but as you've mentioned `/WHOLEARCHIVE` in the other issue, I guess you are aware that it's required exactly for such static libs compiled with `-fvisibility=public`, which happens...

`/WHOLEARCHIVE` works for me with LDC v1.31. `dub build -v` with original linking cmdline: ``` C:\users\packer\dlang\ldc-1.31.0-sym3\bin\ldc2.exe -ofC:\Users\packer\AppData\Local\dub\cache\test_shared\~master\build\library-debug-hDW8YLE1ZVPJ7wLXHHNnIw\test_shared.dll C:\Users\packer\AppData\Local\dub\cache\test_shared\~master\build\library-debug-hDW8YLE1ZVPJ7wLXHHNnIw\test_shared.obj C:\Users\packer\AppData\Local\dub\cache\hipengine_api\~master\build\script-debug-XLe1SoMpflO1lIo_d2vvrg\hipengine_api.lib C:\Users\packer\AppData\Local\dub\cache\util\~master\build\library-debug-bz9ubPFLpKbYegIlMvik0w\util.lib --link-internally -shared -g lld-link: error: undefined symbol: __declspec(dllimport) Interface for...

I don't think we can sensibly handle this in the compiler, without making working with Windows DLLs a similar pain as with C++. So I think this should be handled...

AFAIK, a ModuleInfo A only references another imported module B if B has a module ctor directly, or imports (directly or indirectly) another module with a module ctor (as said,...

> Is .di files supposed to generate ModuleInfo at all? Nope. They aren't compiled, so `pragma(LDC_no_moduleinfo);` currently doesn't affect them at all. The pragma is ignored for the referencing part...

Yeah well doing this for all .di headers is probably not desirable. E.g., some vendor might distribute prebuilt libs and .di stubs for the interface; in that case, the .di...

With what LDC version were these problems encountered? At work, we hit one such instance with D v2.101, but D v2.102 (LDC v1.32) fixed it.

[Working as intended, i.e., the failures are legit without hackily disabling these remaining ones, as we do for the *Cirrus* aarch64 job.]