proposal: reflect: allow Value.Set with zero Value for interface values

[dsnet]

Consider the following Go code:

type I = interface { ... }
var x I = ...
var y I = ...
x = y

Consider the seemingly equivalent Go reflection code:

vx := reflect.ValueOf(&x).Elem()
vy := reflect.ValueOf(y)
vx.Set(vy) // analogous to x = y

However, the reflection code about is not quite equivalent since it panics if y is nil.
This occurs because reflect.ValueOf loses track of the fact that y may be a nil interface value
and simply constructs a zero reflect.Value.

The workaround is somewhat convoluted and not immediately obvious as necessary:

vx := reflect.ValueOf(&x).Elem()
vy := reflect.ValueOf(y)
if vy.IsValid() {
    vx.Set(vy)
} else {
    vx.Set(reflect.Zero(vx.Type()))
}

I propose changing the behavior of func (v Value) Set(x Value) such that x may be a zero Value only if v is an interface value, in which case it simply stores a nil value.

This should be a generally safe change since switching from a panic to a non-panic is not likely to break existing code.

[dsnet]

It just occurred to me that I could do:

vy := reflect.ValueOf(&y).Elem()
vx.Set(vy) // analogous to x = y

but that always causes y to escape be allocated on the heap.

[ianlancetaylor]

In the same general space as #51649. CC @robpike

[rsc]

This is very similar to #51649 but I think we need to figure out what the scope of "treat zero Value as untyped nil" would be. Where do we stop? If someone can come up with a clear answer for that question, it would help us move forward.

[rsc]

This proposal has been added to the active column of the proposals project
and will now be reviewed at the weekly proposal review meetings.
— rsc for the proposal review group

[dsnet]

Where do we stop? If someone can come up with a clear answer for that question, it would help us move forward.

In #51649 (comment), I came up with a concrete proposal for comprehensive changes to reflect if we were to consistently treat a zero Value as equivalent to an untyped nil.

[rsc]

Let's merge this into #51649. Closing as duplicate.