struct S( Element )
{
    Element[] data;

    this( in uint len )
    {
        data.length = len;
    }

    ref Element opIndex( in uint n )
    {
        return data[n];
    }

    ref Element opIndexAssign( in Element a, in uint n )
    {
        data[n] += a;
        return data[n];
    }

    invariant()
    {
	// If I comment out next line compilation goes smoothly:
        assert( Element.sizeof > 0 );
    }
}

void main()
{
    alias S!(double) st;
    st test = st(20);
    auto a = test[2];
    test[2] = 3;
}

compilation error:
$ dmd demostruct
Error: __result = this is not an lvalue
demostruct.d(12): Error: __result = (this.data[n]) is not an lvalue
demostruct.d(18): Error: __result = (this.data[n]) is not an lvalue
demostruct.d(29): template instance demostruct.S!(double) error instantiating

Reply to Weed,


> ref Element opIndex( in uint n )
> {
> return data[n];
> }

> ref Element opIndexAssign( in Element a, in uint n )
> {
> data[n] += a;
> return data[n];
> }

I'm guessing as I don't use 2.0 but I think that this is a bug. DMD is trying to say that the above returns are trying to return something that can't be referenced (like a math expression result).

> invariant()
> {
> // If I comment out next line compilation goes smoothly:
> assert( Element.sizeof > 0 );
> }


OTOH that assert is wrong. Element.sizeof will always return 8, the size of an array reference. What you want is Element.length.

On Fri, 12 Dec 2008 02:21:11 +0300, BCS <ao@pathlink.com> wrote:

> Reply to Weed,
>
>
>> ref Element opIndex( in uint n )
>> {
>> return data[n];
>> }
>
>> ref Element opIndexAssign( in Element a, in uint n )
>> {
>> data[n] += a;
>> return data[n];
>> }
>
> I'm guessing as I don't use 2.0 but I think that this is a bug. DMD is trying to say that the above returns are trying to return something that can't be referenced (like a math expression result).
>

It is a bug, indeed. The struct 'invariant' prevents proper template instantiation somehow... Removing it makes code work as intended.
I'll submit a bug report.

>> invariant()
>> {
>> // If I comment out next line compilation goes smoothly:
>> assert( Element.sizeof > 0 );
>> }
>
>
> OTOH that assert is wrong. Element.sizeof will always return 8, the size of an array reference. What you want is Element.length.
>
>

No, I don't think so. Don't confuse 'Element' with 'data'. Element is a type, it doesn't have a length property. But it does have sizeof (an example of Element is an 'int', which is 4 bytes long).

>>> ref Element opIndexAssign( in Element a, in uint n )
>>> {
>>> data[n] += a;
>>> return data[n];
>>> }
>>
>> I'm guessing as I don't use 2.0 but I think that this is a bug. DMD is trying to say that the above returns are trying to return something that can't be referenced (like a math expression result).
>>
> 
> It is a bug, indeed. The struct 'invariant' prevents proper template instantiation somehow... Removing it makes code work as intended.
> I'll submit a bug report.

Thank you!

Give a link to the bugreport later?

Reply to Denis,

> On Fri, 12 Dec 2008 02:21:11 +0300, BCS <ao@pathlink.com> wrote:
> 
>> Reply to Weed,
>>
>>> invariant()
>>> {
>>> // If I comment out next line compilation goes smoothly:
>>> assert( Element.sizeof > 0 );
>>> }
>> OTOH that assert is wrong. Element.sizeof will always return 8, the
>> size  of an array reference. What you want is Element.length.
>> 
> No, I don't think so. Don't confuse 'Element' with 'data'. Element is
> a type, it doesn't have a length property. But it does have sizeof (an
> example of Element is an 'int', which is 4 bytes long).
> 

Oh, fud. I did get that wrong. I still think it's wrong as I'm not sure anything has 0 sizeof and even then that should be static assert.

BCS пишет:
> Reply to Denis,
> 
>> On Fri, 12 Dec 2008 02:21:11 +0300, BCS <ao@pathlink.com> wrote:
>>
>>> Reply to Weed,
>>>
>>>> invariant()
>>>> {
>>>> // If I comment out next line compilation goes smoothly:
>>>> assert( Element.sizeof > 0 );
>>>> }
>>> OTOH that assert is wrong. Element.sizeof will always return 8, the
>>> size  of an array reference. What you want is Element.length.
>>>
>> No, I don't think so. Don't confuse 'Element' with 'data'. Element is
>> a type, it doesn't have a length property. But it does have sizeof (an
>> example of Element is an 'int', which is 4 bytes long).
>>
> 
> Oh, fud. I did get that wrong. I still think it's wrong as I'm not sure anything has 0 sizeof and even then that should be static assert.


You are right.

It happens that the verification Element not null. static assert would be better.

On Fri, 12 Dec 2008 02:47:42 +0300, BCS <ao@pathlink.com> wrote:

> Reply to Denis,
>
>> On Fri, 12 Dec 2008 02:21:11 +0300, BCS <ao@pathlink.com> wrote:
>>
>>> Reply to Weed,
>>>
>>>> invariant()
>>>> {
>>>> // If I comment out next line compilation goes smoothly:
>>>> assert( Element.sizeof > 0 );
>>>> }
>>> OTOH that assert is wrong. Element.sizeof will always return 8, the
>>> size  of an array reference. What you want is Element.length.
>>>
>> No, I don't think so. Don't confuse 'Element' with 'data'. Element is
>> a type, it doesn't have a length property. But it does have sizeof (an
>> example of Element is an 'int', which is 4 bytes long).
>>
>
> Oh, fud. I did get that wrong. I still think it's wrong as I'm not sure anything has 0 sizeof and even then that should be static assert.
>
>

No, zero-sized arrays are allowed in D:

struct Palette
{
  uint    size;
  ARGB[0] colors
}

Palette* createPalette(int numColors)
{
   return cast(Palette*)GC.malloc(uint.sizeof + numColors * ARGB.sizeof);
}