[java] not enough indentation when using builder pattern (redux)

[mtaran-google]

On some detailed inspection, I noticed the patch for #124 didn't actually fix the problem with the initial example:

class Clazz {
  void foo() {
    return new Foo.newBlah(
        new Bar().)
  }
}

pressing enter before the last . should indent +4, but currently indents +0

[marijnh]

The problem appears to be that the grammar doesn't understand new Foo.newBlah(...) (it assumes the target for the new operator is a class name). Is that new syntax I'm not aware of? What is it called (so I can look it up)?

[mtaran-google]

Ah, my bad!

A more representative example is

class Clazz {
  void foo() {
    return Foo.builder(
        Bar.builder().);
  }
}

That one does work (gives a +4 indent before the last period), but only when that last period is already there. When you're writing new code, that's generally not the case:

class Clazz {
  void foo() {
    return Foo.builder(
        Bar.builder());
  }
}

In this version, when pressing enter before the last close paren the indentation goes +0 where it should go +4.

[marijnh]

Right, but at that point there isn't really enough information yet to determine whether the expression will continue, so the +0 is reasonable. Does adding the case where the line starts with a period to electricInput in src/cpp.js, so that the editor will reindent the line when you type a period at the start of a line, produce the effect you are looking for?

[mtaran-google]

at that point there isn't really enough information yet to determine whether the expression will continue, so the +0 is reasonable.

I'm not sure I follow you. There's an expression in the outer set of parens, it's had one non-empty line in it (Bar.builder()) and if there's another line break following that part, it's almost certainly going to be more of the expression. The other options I see are

1. A blank line (allowed per https://google.github.io/styleguide/javaguide.html#s4.6.1-vertical-whitespace)
2. A comment (followed by no other code, only the closing paren)

Option 1 is theoretically possible, but I wouldn't consider it at all normal (since grouping whitespace is typcially useful between parallel pieces of code). Similarly option 2 would be quite strange since comments are usually either above some code to the right of it on the same line.

Or am I missing something?

[marijnh]

The mode acts on a parse of the actual code, and doesn't use whitespace-based heuristics to predict what is likely to be added to that code next. Also, the same indentation logic is used to block-indent stuff, so it doesn't really assume the cursor is on/near the start of the line that's being indented.

Again, would setting things up so that lines are reindented when you type a period at their start (after whitespace) help address this?