How do you feel about the lib[std]c++ dependency? The upstream PR is dlang/druntime#2788.

Got back to this and reworked it to avoid C++ runtime dependencies. A few non-virtual functions are still declared only, but the D vtable should now be populatable (no more linker errors for the druntime test runners) and the D implementations (extern(D) mangling) shouldn't conflict with the C++ ones anymore.

Note that populating the D vtable with non-conflicting extern(D) implementations only works here because the virtual function signatures are trivial and should be ABI-compatible across C++ and D (edit: probably not for 32-bit x86, this passed in different registers IIRC...). An extern(D) virtual function with 2 params would already suffice to break this, as D callers would call the C++ implementation (if the vptr points to the C++ vtable, i.e., if the instance has been created on the C++ side) with reversed args.

Reworked with @weak extern(C++) for the 2 problematic core.stdcpp.{exception,typeinfo} modules, seems to work fine.

I've also ported druntime's stdcpp integration tests on Windows from DigitalMars make to GNU make, so that they are tested with LDC too. They take about 37 seconds on my box (not parallelized) and cover more modules than Posix hosts (e.g., core.stdcpp.vector is currently only supported for MSVC targets) and also test all 4 MSVC runtime flavors (static/dynamic, release/debug).