no call to Base.rest if vararg is all-underscore

[simeonschaub]

I just noticed that a, _... = b still contains a call to Base.rest after lowering, which we probably don't want, especially if we eventually decide to go with #37132. This just skips all-underscore varargs during destructuring.

[simeonschaub]

() = a actually currently produces a dangling NewVarNode, which I guess is not a big deal, especially since it's a pretty contrived example, but I guess avoiding this doesn't hurt?

[JeffBezanson]

Yeah it's worth removing the unused NewvarNode.

[JeffBezanson]

This is perfectly fine, but I remain opposed to #37132 in the strongest possible terms. It's pure user harassment. In _..., the tail is allowed to be empty, so that change amounts to telling people that the syntax a, b = f(x) now has to be a, b, _... = f(x) for no reason. That's because the latter will always work, but the former will now sometimes error, so to be safe (especially when updating your code to placate the awakened beast) you have to write the latter one. And it just seems strange to have to announce that you are not using some values. Do I have to go to amazon.com right now and tell them that I am not ordering a shower head?

[StefanKarpinski]

The real question is why aren't you ordering a new shower head?