C# Language Design Notes for Jan. 9th, 2019

Agenda

1. GetAsyncEnumerator signature
2. Ambiguities in nullable array type syntax
3. Recursive Patterns Open Language Issues https://github.com/dotnet/csharplang/issues/2095

Discussion

Async streams GetAsyncEnumerator revisited

Discussed over email as well.

Proposal: bind e.GetAsyncEnumerator() and use the result, including methods with optional parameters or params and extension methods.

We like the simple bind accepting optional parameters and params, but extension methods are a problem. We have a number of different features that have backwards compatibility constraints preventing them from preferring extension methods over interface implementation. In addition, the design of extension methods is always as a last resort and we prefer code directly from the type (including interface implementation).

Conclusion

Bind a lookup, but without extension methods. After looking for the interface, consider extension methods. Follow-up issue to deal with the difference with simple foreach, which always looks for a zero-parameter method and doesn't consider extension methods.

Q: What about [Caller...] attributes?

A: Let's do what we did for LINQ, which has to do roughly the same thing.

Obsolete

Do we respect the obsolete attribute in lookups in foreach, async foreach, and pattern dispose?

Conclusion

Yes.

Ambiguities in nullable array types (again)

Example:

if (o is string[][]?(var e0, var e1, var e2) s)

This is now ambiguous because the token after the ? is not helpful in resolving the ambiguity.

1. Keep the order, but address using lookahead
2. ? is a type constructor (reverse the order of brackets and question marks)
3. reverse the order of ?'s only
4. Introduce Array<T>

Array<int[]?>
new Array<int[]?>(3)
new Array<int[]?> { new[] {1}, null, new [] {2, 3 }}

5. Don't allow nullability in jagged arrays

Conclusion

Let's try (2) and see how it works.

Should SwitchExpressionException capture an unreified tuple input?

Yes, we think so.

Rename deconstruct pattern to positional pattern?

Yes. The extra developer work required is worth it.

Permit trailing comma in a switch expression?

e switch {
    1 => true,
    2 => false,
}

Yes, permit optional commas anywhere there are curly braces, including switch expressions, property patterns, etc.

Warnings for non-nullable fields that are not explicitly initialized

Scenarios that result in unnecessary warnings:

• Initialization helpers called from other constructors
• Factory methods that call constructors
• Object initializers
• Chained constructors
• Set-up/tear-down methods in test frameworks
• Reflection

Options:

0. Do nothing, keep the current behavior
1. Don't track initialization.
2. No warnings for private constructors
3. No warnings for constructors that call methods that might modify this
4. Track initialization debt with the class only, reporting warnings at public entry points.
5. Track initialization debt across classes (and assemblies), reporting warnings where the object is constructed

Conclusion

Let's keep option 0 for now. We're worried about silently hiding legitimate warnings or that options 4 and 5 are too complicated or expensive to implement.