Support discards in deconstructions, out var and patterns

[jcouv]

(int i, int j, *) = M();
var (i, j, *) = M();

Even if wildcards don't make it into C# 7, we need to figure out the syntax model and design, so they can be added smoothly later.

[HaloFour]

Has the wildcard syntax been settled? My vote is definitely for * (rather than trying to make _ do double duty contextually).

[jcouv]

@HaloFour The syntax is not finalized yet, but * has some advantages as it is not ambiguous with any identifier. We're leaning that way at the moment.

[jcouv]

@HaloFour One more thing to set expectations, it is possible that wildcards may not make it in C# 7, but only 7.1 or 8, due to time constraint. I intend to figure out the design so that it can be added later without introducing compat issues.

[jaredpar]

Moving the actual language feature out to 2.1 for now. Issue #12664 is tracking the syntax work in RC.

[gafter]

This will be easier to address once #13270 has been done.

[jcouv]

This will be easier to address once #13270 has been done.

@gafter I didn't understand that comment. The way I'd implement wildcards in the current design would be make a special DeconstructionVariable for wildcard and not produce the final assignment step to such a variable.
We'd still produce conversion steps (since they are observable) as normal.

[gafter]

@jcouv The initial binder should not be producing any "steps" at all, as those are not part of the static semantics of the construct.

[jcouv]

This issue was superseded by the more detailed issue #14794