Incorrect indentation with `use`

[mattdeboard]

elixir-mode seems to think a . in a use expression is the beginning of a new scope and indents accordingly:

defmodule Banana do
  use GenServer.Behaviour

               def elixir_mode_indents_to_here do
                 1 + 1
               end  

In other words, it aligns the next line to the period.

[mattdeboard]

So I added a couple of tests to my fork:

(elixir-def-indentation-test indents-use-dot-module-newline
    (:expected-result :failed)
  "
defmodule Foo do
use GenServer.Behaviour

def foobar do
if true, do: IO.puts \"yay\"
end
end"
  "
defmodule Foo do
  use GenServer.Behaviour

  def foobar do
    if true, do: IO.puts \"yay\"
  end
end")

(elixir-def-indentation-test indents-use-dot-module ()
  "
defmodule Foo do
use GenServer.Behaviour
def foobar do
if true, do: IO.puts \"yay\"
end
end"
  "
defmodule Foo do
  use GenServer.Behaviour
  def foobar do
    if true, do: IO.puts \"yay\"
  end
end")

The first test (the expected failure) fails. The second test does not. The only difference is the extra newline after use GenServer.Behaviour.

[antifuchs]

Huh, I've run into this also - I believe the elixir-smie tokenizer returns the right values, but smie itself (or emacs's syntax support) might interfere here. I think most of my state on this got lost in time, but if you can find the root cause of this, I'd be so very happy.

[mattdeboard]

Yes it does return the correct tokens, but when it gets two newlines in a row for whatever reason it treats the previous DOT as the start of a list or something.

I can reproduce it with more trivial examples:

x = 1.2

      3
      4

x = 1.2
3
4

[mattdeboard]

FWIW It appears ruby-mode has the same or similar problem:

http://lists.gnu.org/archive/html/emacs-devel/2014-01/msg01889.html

Was elixir-mode adapted from ruby-mode? It appears to have inherited its warts, if so.

[mattdeboard]

@antifuchs @tomterl So I think I've found where in SMIE things are going awry for this ticket specifically. It's on line 1703 of smie.el.

For whatever reason, smie-backward-sexp is returning nil here (I think in my test case above, smie-backward-sexp is being called when the cursor is at the end of the use line, which takes it back to the beginning of the STRING "Behavior", which in turn returns the nil I mentioned above). Because of this, the DOT preceding it is passed to smie-rules-function (I think this is actually elixir-smie-rules, no?) with the :list-intro method.

THEN, because of THAT, this block of code is executed, complete with helpful remark:

       (arg
        ;; There's a previous element, and it's not special (it's not
        ;; the function), so let's just align with that one.
        (goto-char (car positions))
        (current-column))

If I'm understanding correctly, that comment in the code about sums it up. Because the first element in "STRING" "DOT" "STRING", i.e. GenServer.Behavior, is not special, the next line gets aligned with the last "special" token: DOT.

I would love any input possible on the best way to fix this. Might it be in elixir-smie-rules, add a matching pattern? Thanks.

[mattdeboard]

Another relevant thing I just found:

                         ;; FIXME?  `smie-indent-calculate' can (and often
                         ;; does) return a result that actually depends on the
                         ;; presence/absence of a newline, so the gain computed
                         ;; here may not be accurate, but in practice it seems
                         ;; to work well enough.

edit:

So after tinkering with this and manually examining the output of each of the smie-indent-functions, I've found the one that's causing the issues. In my test case I pasted in my March comment, smie-indent-after-keyword is returning 16, which is the number of spaces needed to achieve that wonky alignment.

edit 2:

I have a fix ready for this now. I'm thrilled.