Discards, tuples and out variables

[MadsTorgersen]

This is a very rough cut of discards, tuples and out variables, and is here for discussion/sharing purposes. It lacks examples and cross references throughout and doesn't touch on patterns at all. It's also missing treatment of additional declaration spaces at statement/expression level, as well as the syntactic short form var (a, b) in deconstruction.

The intent is to test out an alternative approach to #44, #63 and #596 along the lines suggested in #561, and to fill out several gaps in those PRs.

The approach differs in several ways:

• Discards are proper variables, only without names. Otherwise, the treatment of out vars cannot explain where the variable comes from that is passed to the out parameter.
• The entities on the left hand side of a deconstructing assignment are proper tuple expressions. Otherwise we cannot adequately account for when their elements are full-blown expressions.
• A new expression classification for tuples helps describe when, whether and how tuples should be evaluated.
• The variable-declaring capability of tuples and out variables is factored out into a new kind of expression, a "declaration expression" instead of adding new special cases to each of their productions.

I believe this approach is more correct now, and more resilient to future language additions. But there is still a ways to go to clean it up.

[jskeet]

Submitting the comments as far as I've got. (More reading and thinking to do.)

[BillWagner]

I like this approach. I likely went into details that were beyond where you've already speced things out, but I wanted to make sure I noted them, even if just to extend the TODO list.

This does make a number of concepts more cohesive.

[MadsTorgersen]

Current status - the PR is taking shape but the following work is still TODO:

• Examples of Declaration expressions, Discards, Tuple expressions and Deconstructing assignments
• Add declaration_expression to Declaration Spaces where necessary
• Generalize scope to not just be blocks where necessary (to allow expression bodies with declaration expressions)

The scope of this PR will not include patterns.

[RexJaeschke]

Re discards, I propose adding the following text to lexical-structure.md, near the end, just before the mention of identifiers starting with __:

The semantics of an identifier named _ depends on the context in which it appears:

• It can denote a named program element, such as a variable, class, or method, or
• It can denote a discard (§discards-new-clause)

[RexJaeschke]

Some editorial feedback:

1. Replace all "legal" with "valid."
2. For grammar blocks, replace all "[backticks][space]ANTLR" with "[backticks]ANTLR", so the tools can find them.
3. For C# code blocks, replace "[backticks][space]c#" with "[backticks]csharp", so the tools can find them.

[MadsTorgersen]

This is missing several special behaviors for tuples, e.g. the behavior of ==.

The == and != behavior is now specified.

[jskeet]

Agreed we're good to merge this - there are still tiny bits to do, but getting this in and then raising more bits will be better than trying to get it all the way.