Martin Kinkelin

Oh right, I think I've simplified it (this is a v1.31 regression) because the user doing this explicitly has been an error for a while: ```D enum U bla =...

`U.init.setAll(1)` is lowered to a `U(0, [1])`... and using `arr[] = x;` in `setAll()`, instead of setting the single element explicitly, makes the problematic `x` initializer vanish.

This *might* be enough: https://github.com/ldc-developers/ldc/pull/4299

Yeah this is an unfortunate side effect of the prebuilt package now being generated on a CI runner with VS2022 image.

This sounds like it might be the problem of the .obj from the static lib containing the `__InterfaceZ` definition and export linker directive not being pulled in during linking -...

After a superficial glance, it looks like your data layout diverges in the .di - no fields at all, and no initializers for the fields. So the class size is...

As odd as it sounds, this LDC-specific 'optimization' probably spared you from even-harder-to-track-down bugs later on with code relying on `__traits(classInstanceSize)` etc. :) Just out of interest, what's your .di...

The `{d,c,linker}Flags` are all supposed to be separated by semicolons, not spaces (in order to allow flag values with spaces). That's shown on the Wiki page and part of the...

With that `armv7a-none-newlibeabi` triple, you're targeting bare-metal, not Linux. [The `--targetSystem` stuff is just for our CMake script.] So the compiler won't predefine `linux`, and then druntime most likely doesn't...

https://github.com/ldc-developers/ldc/pull/3946 added support for a predefined `CRuntime_Newlib`. druntime needs extensions to work with new targets - ISAs, OSs, C runtimes - so yeah, there's definitely contribution opportunity. :) - Upstream...