Array literal shouldn't cause GC allocation for void element types

[ljmf00-wekaio]

This compiles:

@nogc
void main()
{
	ubyte[1] bar = [0x00];
}

Although, this doesn't, but should:

@nogc
void main()
{
	void[1] bar = [0x00];
}

[Geod24]

Although, this doesn't, but should:

Why does a void static array make sense ? It's size can't be known.

[ljmf00-wekaio]

Although, this doesn't, but should:

Why does a void static array make sense ? It's size can't be known.

It does make sense.

The size is known, the length is in bytes of the original slice. See https://dlang.org/spec/arrays.html#void_arrays

13.17.2.1. [...] The .length of a void array is the length of the data in bytes, rather than the number of elements in its original type.

The main difference from ubyte[] to void[] is that GC shouldn't assume no indirections on void[] arrays (on ubyte[] arrays, there's no pointers, on void[] there might be).

[dkorpel]

I think @nogc is the least of your problems, it looks like void[] arrays combined with array literals is a mess:

void main()
{
    ubyte[] b = [0x00]; // [cast(ubyte) 0]
    writeln(b.length);  // 1
    
    void[] v = [0x00]; // cast(void[]) [0]
    writeln(v.length); // 4
    
    void[1] bar = [0x00]; // Range violation (4 bytes assigned to 1 byte)
    // void[4] bar = [0x00]; // Error: Mismatched array lengths, 4 and 1
}

We need to decide whether integers in array literals converted to void[] represent 4 bytes or 1.

[ljmf00-wekaio]

I think @nogc is the least of your problems, it looks like void[] arrays combined with array literals is a mess:

void main()
{
    ubyte[] b = [0x00]; // [cast(ubyte) 0]
    writeln(b.length);  // 1
    
    void[] v = [0x00]; // cast(void[]) [0]
    writeln(v.length); // 4
    
    void[1] bar = [0x00]; // Range violation (4 bytes assigned to 1 byte)
    // void[4] bar = [0x00]; // Error: Mismatched array lengths, 4 and 1
}

We need to decide whether integers in array literals converted to void[] represent 4 bytes or 1.

Luckily, that's dmd specific, on LDC that is ok. Perhaps, you should create another bug report :D

[ljmf00-wekaio]

Btw, there is no problem regarding GC with D optimisations in LDC, as we benefit from the lowering from GC to stack transformation passes, see https://godbolt.org/z/1c77P8xY6 .