Declaring ref locals in for loop initializer

[jnm2]

The ref returns and locals spec is not linked from https://github.com/dotnet/csharplang/blob/master/Language-Version-History.md#c-70-visual-studio-2017 and I'm failing to find it by searching. Because I can't find that trail, I have no idea if this is a limitation of the spec; it almost seems like a bug in Roslyn.

I expect to be able to write this:

for (ref var en = ref enumerator; en.MoveNext();)
{
   // en.Current
}

Since I can write this:
(I thought this didn't work because I mistakenly had enumerator as a readonly class field, but the point still remains.)

for (var en = enumerator; en.MoveNext();)
{
   // en.Current
}

The workaround is to stop scoping en to the loop:

ref var en = ref enumerator;
while (en.MoveNext())
{
   // en.Current
}

You may dispute the liberties I tend to take with for loops, but I could construct a similar example that uses iteration, condition and iterator.

The point is that it seems wrong that a valid local declaration statement cannot be used in the for_initializer:

for_initializer
    : local_variable_declaration
    | statement_expression_list
    ;

https://github.com/dotnet/csharplang/blob/master/spec/statements.md#the-for-statement

[HaloFour]

@jnm2

Since I can write this (though this doesn't work because the enumerator is a struct which mutates on MoveNext):

Why do you need a ref for that? Structs have always been allowed to mutate themselves via instance methods:

Point pt = new Point(2, 4);
pt.Offset(2, 2); // self-mutating method
Debug.Assert(pt.X == 4 && pt.Y == 6);

Any time you call a struct instance method the this argument is actually a ref.

[jnm2]

@HaloFour That's what I thought, but I observed this not being the case when I moved the struct enumerator to a class field. I'll create a code example.

[HaloFour]

@jnm2

Was the field readonly? If it is the C# compiler will try to protect you by assigning the field to a local first and passing the ref to the local to the instance method. That way if the method mutates the local would be affected and not the field.

[jnm2]

@HaloFour Yes, that must be what happened. Just tried it both ways. So I don't actually need ref locals for anything yet... I was so happy there for a moment 😜

But my dislike of the inconsistency in for_initializer still stands.