Permit new with trivial end

[som-snytt]

Fixes #24250

Make trivial end new useful by supplying the empty template in parser.

This makes

new C {
}

equivalent to

new C:
end new

[sjrd]

Then

new C
end new

is just new C.

Why would we want that? It doesn't seem to follow the rule that end closes a block.

[som-snytt]

I thought the rule was that you can remove the marker without changing semantics, but that does not hold.

if b then e
end if

I'll try that out. Why doesn't github offer embedded REPL?

This obeys the Odersky blank line rule:

def f(b: Boolean) =
  if b then println("""hello,







    world""")
  end if
@main def test = f(true)

[som-snytt]

Why would we want that?

I remember the argument for

class C:
end C

was that I'm in the middle of editing, though I don't know if that's why it was finally allowed. But the same applies.

new C 
  // todo maybe override something later?
  // do we need an extra class? maybe we can avoid it.

  // override def toString = "Test C"
end new

The least github could do is test my late-night snippets in the browser.

[sjrd]

The thing is that new C and new C {} mean completely different things. The latter, because of the block, creates an anonymous subclass of C. The former doesn't. You can't just "add overrides" to a new C. There's a first, very signifiant step where you have to introduce an anonymous class. That's when you add the : and possibly the end, not before.

[som-snytt]

Yeah, maybe. I'm not wedded to any of this. Is colon enough of an indicator?

if I can write

  1
+ 2
end +

and I'm not sure they will yet, but my point is new gets equal protection. That is SIP 77. Or a similar argument for a named method, but I'm exaggerating for rhetorical effect.

[som-snytt]

it makes it hard to discern when a large indentation region ends

I'll update to read "or a very small region." Joke.

But this is clear:

If the statement defines an anonymous class, then s must be new.

end is intended only for when there was an anon class. I'll remove the even-more-trivial case of a named class instance.

That is at https://dotty.epfl.ch/docs/reference/other-new-features/indentation.html#the-end-marker

[som-snytt]

isNew was never used

[som-snytt]

ignores the boolean value

[som-snytt]

preserves existing behavior, but what is the span of the result?

[som-snytt]

comment inlined from the previous PR

[som-snytt]

Is it possible to break the habit of inserting an initial blank line in test sources?