fix Issue 15869 - RVO can overwrite argument

[WalterBright]

No description provided.

[dlang-bot]

Thanks for your pull request, @WalterBright!

Bugzilla references

Auto-close	Bugzilla	Severity	Description
✓	15869	normal	RVO can overwrite argument

Testing this PR locally

If you don't have a local development environment setup, you can use Digger to test this PR:

dub fetch digger
dub run digger -- build "master + dmd#8200"

[ibuclaw]

If (f != vd) continue? As its quite a nest you have here of statements.

[RazvanN7]

what if the value isn't actually initialized? Maybe it should be checked that vd._init !is null otherwise issue an error that the address cannot be taken if the variable isn't initialized and the default constructor is disabled

[WalterBright]

It not being initialized should only be possible in @System code, where anything goes.

[RazvanN7]

I find it odd that the compiler forces you to initialize the field but lets you use it before initialization. I agree that if the struct does have a default constructor it should be allowed, but when it's missing why am I forced to initialize it if I can use the struct anyway?
For example, this code compiles:

struct A
{
    @disable this();
    this(string) {}
    int c;
}

struct B
{
    A a;
    int c;
    this(int)
    {   
        c = a.c; 
        a = A("string"); // if this is commented, error : field `a` must be initialized in constructor
       // why? if I can use the fields anyway...
    }   
}

void main() {}

[WalterBright]

By having a constructor, and default construction disabled, you have to initialize it explicitly rather than it being default initialized. It's the whole point of disabling default construction. The initial value, however, is still defined and is still useful.

[RazvanN7]

This needs to be documented somewhere, otherwise you may end up with surprising error messages. For example, this code:

immutable(int)* g;

struct X { 
    int a = 10; 
    immutable this(int x)
    {   
            g = &a; 
            a = 42;    // error
    }   
}

void main()
{
    auto x = immutable X();
}

Leads to : test.d(7): Error: immutable field a initialized multiple times which is correct but not obvious

[yshui]

I think we should point to the first initialization in the error message. And maybe a special message explaining why taking address is initialization.