Change mangling of extern(C++) ctor.

[redstar]

This is a part of issue #1312.

[redstar]

@kinke Does this help for ctor mangling?

[kinke]

Thx Kai, I'll check it over the weekend.
So this should allow using C++ ctors from D by just declaring them in D. But this doesn't fix the actual mangling bug in #1312.

[yebblies]

Without also matching the ABI, matching mangling is a little dangerous.

[kinke]

That's what I thought too. You mean the different destruction rules, right? Does the front-end support argument destruction by the caller for extern(C++) functions?

[yebblies]

I'm not sure. The issue I've had trouble with is that D will blit init[] over before calling the constructor, while C++ won't. We should (hopefully) be matching argument destruction correctly for normal C++ calls, or is it different for ctors?

[kinke]

D will blit init[] over before calling the constructor, while C++ won't.

I'd think that a C++ ctor cannot make any assumptions about the initial payload, so I'd consider blitting to be useless (and affecting runtime a bit when calling it from D) but not troublesome.

We should (hopefully) be matching argument destruction correctly for normal C++ calls

That's what I wanted to hear - very good.

[yebblies]

I'd think that a C++ ctor cannot make any assumptions about the initial payload, so I'd consider blitting to be useless (and affecting runtime a bit when calling it from D) but not troublesome.

When calling a ctor written in C++ from D, yes. But when going the other way the ctor may not init all fields.

[kinke]

But when going the other way the ctor may not init all fields.

True. That could be compensated for when generating an extern(C++) ctor implemented in D by having the front-end inject the blitting at the beginning of the ctor (instead of blitting before calling it from D). We could also simply define the semantics for extern(C++) ctors implemented in D as fully C++ compatible, i.e., no implicit pre-init - but that would break ddmd I guess.

[redstar]

Closing in favor of dlang/dmd#5884.