Location data for functions should include the ending semicolon

[alangpierce]

This is kind of a subtle issue involving location data, so it probably only affects tools for CoffeeScript rather than ever causing incorrect CoffeeScript output.

With code like this:

->
  ;

the location for the function ends at one character before the semicolon rather than it including the semicolon. This caused a bug in decaffeinate, and I ended up working around it with a decaffeinate-specific change, but figure it would be good to leave a note here.

More context here: decaffeinate/decaffeinate-parser#229

I think this is related to my change involving OUTDENT tokens here: #4296

I spent a little time digging and the issue is that semicolon tokens are removed in the normalizeLines step of the rewriter, which means that fixOutdentLocationData pushes them left to the INDENT token. I tried just reordering the phases, but that broke tests. I think the right behavior is to include the TERMINATOR token in the function range if it's a semicolon, but not if it's a newline, so the ideal solution might be to distinguish those in the rewriter.

Here's how someone can repro it more precisely. In a project with CoffeeScript installed, run this in the node repl:

> require('coffee-script').nodes('->\n  ;').expressions[0]
Code {
  params: [],
  body: 
   Block {
     expressions: [],
     locationData: { first_line: 1, first_column: 0, last_line: 1, last_column: 1 } },
  bound: false,
  isGenerator: false,
  locationData: { first_line: 0, first_column: 0, last_line: 1, last_column: 1 } }

locationData in CoffeeScript is 0-indexed with an inclusive end, so both the Code node and the Block node end at the space just before the semicolon.

[GeoffreyBooth]

I would welcome a fix for this, but it’s not really a “bug” as far as CoffeeScript is concerned, so I’m closing the issue. I don’t see any of the contributors spending time on this, but a PR that addresses it would be welcome.