C# Language Design Notes for January 7th, 2019

Agenda

Nullable:

1. Variance in overriding/interface implementation
2. Breaking change in parsing array specifiers

Discussion

variance in overriding/interface implementation

Issues: https://github.com/dotnet/roslyn/issues/23268, https://github.com/dotnet/roslyn/issues/30958

We currently don't support overriding or inheritance with co/contravariance in nullability.

Overriding: we mostly take the signature from the overriding member, not the base.

We could allow nullability variance and then use the most derived implementation.

The problem is if we don't provide a warning on further derivation, another deriver can "re-loosen the contract."

There doesn't seem to be any problem with an interface implementation, because the safety is provided in the actual code.

Note: if you override an object-returning method with a dynamic-returning method, we currently use the dynamic return on callsites, which is a parallel to how nullable works.

Options:

1. Keep stringent invariant rule
2. Allow safe variation on interface implicit implementation only
3. Allow on overrides also, check against base override

Conclusion

General agreement on (3). Possibly change tuple names to operate similarly (use the base's override as the authoritative signature). We will also use the same rules for explicit interface implementation.

Breaking change in parsing array specifiers

Issue: https://github.com/dotnet/roslyn/issues/32141

a ? x is A[][] ? b : c
            // ^ Is this a nullable annotation or part of a ?:

Options:

1. Keep decision find another solution for ambiguity a. always prefer ternary for back compat, use parens for new meaning in is/as
2. Change array decision a. ? is a type constructor that's applicable to arrays b. Alternate syntax: ? inside brackets: a[?][,] and a[][,?]

One of the main problems with changing the array is: new string[3][]; In this example, 3 refers to the outer array size, but if we make string[][]? mean "top-level nullable," there's either an inconsistency between the question mark location (new string[3]?[] means inner array is nullable) or we swap the location of the rank (new string[]?[3]).

Conclusion

Let's try to stick with our original design, and add tie breakers to parsing to prefer ternary in any ambiguity. Right now we know of problems with array types in is and as. There may be more expressions which can end with an array type, which we will address as we find them.

Update: This decision was revised in the Jan. 9th meeting.