Avoid indentation errors after 'end'.

[tomterl]

This is, if not a fix, a workaround for half of #18. In my elixir files
this produces no indentation errors; maybe you can test this against
your setup and see if it doesn't raise any red flags for you.

I don't know enough smie (yet), to look after the statement indentation
error, that is the other error reported in #18.

[tomterl]

I'm currently digging into smie and elixir-smie.el to get the second problem solved (wrong indentation after virtual ; ).

The solution with :after . "end" doesn't look totally silly anymore after a look at the smie manual;

@tonini I saw that there are a couple of test cases regarding indentation, but a target test is not defined in the rakefile -- would a maintainer who knows about the tests be able to add a test target to the rakefile, so that I could use canonical tests for my changes without having to dig into ert ATM?

[antifuchs]

Just sneaking in to say I love that you're taking this on, Tom! (:

I'm happy to walk you through running tests on emacs-elixir! They're ~undocumented at the moment (so sorry for that!). To run the full elixir-mode test suite, you can run elixir-mode-run-tests - it'll load the test files, and run the appropriate suite with ert. Here's the definition of the function, if you want to know the basic structure of what it does (:

[antifuchs]

Also, I should add - feel free to make more tests (even if they're expected-failing atm)! The suite is very very basic, and the reason it doesn't catch most of the errors is that I just haven't written enough code with elixir-mode yet. S-:

[tomterl]

Cool, Thanks tests are running (I had to add the emacs-elixir directory to load-path, other than that all set now.

More tests when I have something to test :)

[mattdeboard]

So while I was messing with this same code region myself I found that replacing this line with the followingmade an additional test pass (up to 19 now, with the two I've added):

("def" non-block-expr "do" statements "end")

It also fixed the specific example of the problem described in #18 . However it's still broken if you try to add a declaration in module scope (use, require, alias, etc.). Don't know if this helps.

[mattdeboard]

With that change and with a module-level declaration...

Good:

defmodule FooBar do
  use Baz
  def foo do
    if true, do: IO.puts "yay"
    20
  end

  def bar do
    if true, do: IO.puts "yay"
    20
  end
end

Bad:

defmodule FooBar do
  use Baz

  def foo do
    if true, do: IO.puts "yay"
    20
  end

    def bar do
      if true, do: IO.puts "yay"
      20
    end
end

And it's not just the blank line that is creating the issue. If I remove the declaration, it's really fine, even with extra blanks:

defmodule FooBar do

  def foo do
    if true, do: IO.puts "yay"
    20
  end

  def bar do
    if true, do: IO.puts "yay"
    20
  end
end

I suspect but don't know that use, require, import, and alias have a distinct syntax, a function call that's a sibling of a method definition. There's no end to a declaration. And since the declarations aren't defined in the grammar specifically there's a precedence conflict. Note here that the four declarations are basically meaningless to the grammar:

defmodule FooBar do
  garbage
  things
  that
  mean
  nothing


    def foo do
      if true, do: IO.puts "yay"
      20
    end

      def bar do
        if true, do: IO.puts "yay"
        20
      end
end

So maybe having like declarations being a distinct section of the grammar would help?

[mattdeboard]

One final behavior:

defmodule FooBar do

  def foo do
    if true, do: IO.puts "yay"
    20
  end

  def bar do
    if true, do: IO.puts "yay"
    20
  end

    def ban do
      1 + 1
    end

end

So weird. I don't grok why the third method in line breaks indentation.

[tomterl]

Cool.
I'll try to debug that further, As far as I can tell right now, the problem is that smie as setup handles lines it doesn't know the speficics about as hanging - the virtual ; doesn't really help, albeit it should somehow - that includes arithmetic expressions and use, etc. statements.

I'm in the process of groking the process and hope that solving/understanding the virtual ; handling will result in grammar/rules adjustments that make special handling of end and the likes of use unnecessary.

I so hope this isn't a wild goose chase :)

[tomterl]

With the (:after . "end") rule in place, the indentation error after use etc isn't triggered.

The problem with indentation errors after newlines/comments is that consecutive newlines/comments aren't really ignored resp. treated as one - consider the following case

 a = 1 + 2
|b = 2 + 3

(| beeing the cursor ) the last token considered here is the injected ; as (:before . ";"), as it should and the indentation therefore is correct

 a = 1 + 2

              |b = 2 + 3

The last token considered here is basic (:elem . basic) with hanging:t - I'm trying to understand now, why the to \n characters aren't treated as one injected ;, as they should per my understanding.

Or am I going astray somehow (digging vertically in in something unknown leaves much out of focus :).

[mattdeboard]

@tonini @antifuchs Are there any plans for progress on this PR? It's been almost 3 weeks and this syntactic breakage prevents me from using my primary editor for Elixir. I don't mind vim but obviously I'd greatly prefer to use Emacs.

[mattdeboard]

Is there anything I can do to help progress on this PR? Please let me know, I will be quite happy to help.

[tomterl]

I don't have time to play with elixir and the tools ATM. I'm pretty sure the problem is the implicit ';' -- the smie grammar looks correct, but nevertheless produces the described (#41 (comment)) discrepancy. If someone with a bit more knowledge could investigate that angle to see if it's nonsense or not, that would reduce the (perceived) costs for me to make time to look further into it.

[mattdeboard]

@tomterl I understand. I have spent several evenings trying to get to the bottom of this issue myself. The code here looks like it fits the profile of these double-newline indentation errors, but I have a lot of trouble forming a good picture of the logic in my head.

I hope my previous comment didn't come across as nagging.

[mattdeboard]

Fixed in #48