[Issue 23337] New: Wrongly elided postblit/copy ctor for array construction (_d_arrayctor lowering)

[d-bugmail at puremagic.com]

https://issues.dlang.org/show_bug.cgi?id=23337

          Issue ID: 23337
           Summary: Wrongly elided postblit/copy ctor for array
                    construction (_d_arrayctor lowering)
           Product: D
           Version: D2
          Hardware: All
                OS: All
            Status: NEW
          Severity: regression
          Priority: P1
         Component: dmd
          Assignee: nobody at puremagic.com
          Reporter: kinke at gmx.net

The following code worked for DMD v2.085-2.090, but regressed with v2.091 and
still fails with current DMD master, after the templated `_d_arrayctor`
introduction:

```
void main() {
    int copies;
    struct S
    {
        int i;
        this(this) { ++copies; }
        //this(inout ref S) inout { ++copies; }
    }

    import core.lifetime;
    S[2] ss1 = void;
    S[2] ss2;

    emplace(&ss1, ss2);
    assert(copies == 2);

    copies = 0;
    emplace(&ss1, ss2[]);
    assert(copies == 2); // fails (is 0)
}
```

[It works just fine with LDC though.] This came up when bumping LDC's frontend,
as the glue layer would still lower this to old untemplated `_d_arrayctor`. So
the frontend misses the new templated `_d_arrayctor` lowering in this case; the
according buggy logic is probably here:
https://github.com/dlang/dmd/blob/638de4022d1b2d7a8f529d8b81c7916a958bbe9d/compiler/src/dmd/expressionsem.d#L9913-L9923

--