Static initialization of static arrays is weird

[kinke]

On Sunday, 19 August 2018 at 11:20:41 UTC, Dennis wrote:
> I have a two dimensional static array in a game board struct 
> and want to explicitly set the default value for each cell. Now 
> typing the whole 9x9 array out would be cumbersome and I can't 
> change the default constructor of a struct, so I played around 
> with initializers and found some... strange behavior.
>
> Demo:
> ```
> import std.stdio: writeln;
>
> struct T {
>     int[3][3] arr = [2, 1];
>     this(int stub) {
>         arr[0][0] = 9;
>     }
> }
>
> void main() {
>     T.init.writeln;
>     T(0).writeln;
> }
> ```
> Output:
> ```
> T([[2, 1, 0], [0, 0, 118033674], [723976, 0, 4100]])
> T([[9, 2, 2], [1, 1, 1], [0, 0, 0]])
> ```
> So it seems that in the .init value, it puts [2, 1] in the 
> first array and the rest is garbage. But in the constructor 
> case, it filled the first two static arrays with 2 and 1 and 
> the other one is 0 / garbage (can't tell). I turned to the spec:
>
> https://dlang.org/spec/arrays.html#static-init-static
>
> But that doesn't really help specify this case. Should I submit 
> a bugzilla issue? I don't know what's supposed to happen here.

I think the spec is pretty clear; the elements of the 
right-hand-side initializer array are interpreted as per-element 
initializer, i.e., `result[0] = 2, result[1] = 1` (rest: 
default-init). So (latest) LDC outputs

```
T([[2, 2, 2], [1, 1, 1], [0, 0, 0]])
T([[9, 2, 2], [1, 1, 1], [0, 0, 0]])
```

DMD's T.init featuring garbage is clearly a bug, as is the 
divergence wrt. T.init and the initialization of the struct 
literal `T(0)` (which seems to be correct, initializing like LDC, 
but only starting with v2.072). With a non-literal, `auto t = 
T(0); t.writeln();`, the result is `[[9, 1, 0], <garbage from 
T.init>]` again.

So please do file a bugzilla issue.