v7.x Target-typed "default" literal

[RexJaeschke]

@gafter, I'm assigning this to you in the hope you can give me early feedback on my initial thoughts on how to write a proposal for this feature, so I can get started. Thanks in advance! Of course, input from other members is also welcome.

Proposal https://github.com/dotnet/csharplang/blob/standard-proposals/proposals/csharp-7.1/target-typed-default.md allows the use of default on its own (rather than default(T)) in contexts where the type can be inferred.

The "Detailed design" part of the proposal suggests introducing a new kind of expression, the default literal. At a first/second read, that suggests a separate section in expressions.md. However, the "Summary" part of the proposal states

The target-typed default feature is a shorter form variation of the default(T) operator, which allows the type to be omitted. Its type is inferred by target-typing instead. Aside from that, it behaves like default(T).

Here's my thinking:

Rather than add a new section, what we already have in 12.7.15 Default value expressions can be extended to handle this new spelling of a default value. For example, the existing grammar could be extended thus:

default_value_expression
    : 'default' '(' type ')'
    | default_literal
    ;

default_literal
    : 'default'
    ;

rather than the more succinct

default_value_expression
    : 'default' ('(' type ')')?
    ;

[By adding the grammatically unnecessary rule default_literal, we give a name to that alternative that we can refer to in the narrative.]

That said, I see an advantage to also having a name for the existing alternative; perhaps we could do something like the following instead:

default_value_expression
    : explictly_typed_default
    | default_literal
    ;

explictly_typed_default
    : 'default' '(' type ')'
    ;

default_literal
    : 'default'
    ;

Using this approach, I think the changes to this section are minimal.

The only other edits I see are

1. To conversion.md, add a new section default literal conversion
2. To expressions.md deal with the possibility of default == default (if that is not already covered)

I'll write some tests to figure out what support is given by the is and as operators.

[jskeet]

I think I'm most in favour of the last idea, giving them different names. But it's a weakly-held position, from someone who's really not a grammar expert.

[gafter]

I think the idea of using different names (your last grammar in the original post) makes sense to me.