Overly aggressive indenting after recent Emacs update

[giogadi]

Howdy,

I recently updated to the latest version of Emacs on Homebrew (using OSX Yosemite):

https://github.com/Homebrew/homebrew/blob/master/Library/Formula/emacs.rb

Suddenly, whenever I go to a new line using simple-indent-mode the editor always adds a 4-space indent, even in simple cases like:

foo :: Int
    foo = 1 -- auto-indent inserted here by haskell-mode

This behavior persisted even after completely nuking my emacs installation and starting over. Has anyone else experienced this issue?

[chrisdone]

I imagine the abominable electric-indent-mode has something to do with this.

[chrisdone]

Did you update to Emacs 24.4?

[giogadi]

Yes, I'm in Emacs 24.4. Aaaaaaand you're right, disabling electric-indent-mode did the trick! Abominable is right.

[purcell]

In the latest haskell-mode code electric-indent-mode is suppressed, for exactly this reason. Which version do you have?

[giogadi]

I got the version that's currently in marmalade. I guess I could switch to the melpa version, but easier thing is to just disable electric-indent, unless that's actually useful in other contexts?

[purcell]

That's extremely old (Jun 2013). You'd likely have to update your config for the newer versions to work.

electric-indent-mode is a global mode, so just (electric-indent-mode 0) at the top level to disable it completely. I personally find it helpful in major modes where the indentation is unambiguous.

If you want to disable it only in haskell-mode, then try this:

(add-hook 'haskell-mode-hook (lambda () (setq electric-indent-inhibit t)))

[giogadi]

The melpa version of haskell-mode didn't appear to have the automatic suppression of electric-indent, so I followed the tip from @purcell. Quick thing, I had to change the hook you gave me to this for it to work:

(add-hook 'haskell-mode-hook (lambda () (electric-indent-local-mode -1)))

Thanks for the help!

[purcell]

It's possible that electric-indent-inhibit didn't make it into Emacs in time for the 24.4 freeze: looks like it was added on 7th Oct. I'll see if we can incorporate a more widely effective fix directly into haskell-mode.

[purcell]

I was mistaken: that var is in 24.4. And the MELPA version does correctly inhibit electric-indent-local-mode for me.

[gracjan]

Note that electric-indent-inhibit works only on non-empty lines or when cursor in not in the first column. So electric-indent-inhibit does not really fix the 'newline-and-indent' aggresive behavior of haskell-simple-indent:

 (unless (and electric-indent-inhibit
        (not at-newline)
    (indent-according-to-mode))))))

[gracjan]

@purcell: I still get 'newline-and-indent' despite your changes in 99658ce

[ivan-m]

I think we need to do something similar in haskell-cabal-mode as well, as electric-indent can do weird indentation there as well.

[sirlensalot]

Given the bad support for electric-indent-inhibit noted in #377 (comment) I propose haskell-mode go for a full shutdown of electric indent a la #377 (comment)

[gracjan]

Radical. Can we do some more research about this? Electric mode has a black list of modes when it does not activate. It might be better to add Haskell-mode to that list.

[sirlensalot]

Honestly I'm a little surprised that the various indentations do not use electric, since that seems to work fine for c, java, perl etc. While on this topic, I wonder why haskell-mode doesn't leverage semantic to parse source files ... I worked pretty heavily on java emacs stuff in another life and semantic kinda rules. Whereas there's a lot of one-off parsing going on at least in haskell-indent.el ...

[gracjan]

Electric seems to work best for languages with a unique, well defined indentation point derived from curly braces or parentheses. Haskell usually has multiple choices how a line should be indented with different semantics so a programmer needs to choose which semantics he wants.

What exactly does 'semantic' buy in this case? Honest question, I've never used it.

[purcell]

Semantic is pretty huge, and java-mode doesn't use it afaik -- I don't think it's really a building block for major modes. There are languages with more complicated syntax than haskell which Emacs managed to indent robustly. An example would be ruby-mode, which in its latest incarnation is built on the new smie indent engine thanks to the work of @dgutov. I'm not aware of an idiomatic way of building bounce-indent major modes, which I think is the key issue here.

[gracjan]

Is there a well known open source project that depends on semantic?

[sirlensalot]

While java-mode doesn't use semantic, jdee does which is more like haskell-mode in that it offers completion, ant integration, maybe maven now too for all I know. Via semantic it's able to offer nice things like incremental parsing, completion in as-yet-uncompiled code, etc.

My experience back then (circa 2003) was semantic was very fast and has a great API for use within an IDE-like mode.

The problem is more that Haskell doesn't offer a BNF afaict. With a BNF you fire up Wisent and you're done! The comments in Lexer.x indicate it was produced from regexes, presumably in the Haskell report. Someone would probably have to handcode a Bison-style input for Wisent ...

... or follow the semantic docs in writing a regex-based tag generator, using the Haskell report. This could be OK too in that it would at least centralize parsing code in one place. Apparently this disables incremental parsing though.

references:
JDEE: https://github.com/jdee-emacs/jdee
Semantic docs: http://cedet.sourceforge.net/addlang.shtml
GHC Parser docs: https://ghc.haskell.org/trac/ghc/wiki/Commentary/Compiler/Parser
Haskell 2010 syntax: https://www.haskell.org/onlinereport/haskell2010/haskellch10.html
https://www.reddit.com/r/haskell/comments/2ig8ml/where_can_i_find_the_complete_bnf_syntax_of/
http://sourceforge.net/p/cedet/mailman/message/21667834/
http://stackoverflow.com/a/11511916/4015221

[gracjan]

About agressive indenting.

One. electric-indent blacklist all haskell indentation modes:

(defvar electric-indent-functions-without-reindent
  '(indent-relative indent-to-left-margin indent-relative-maybe
    py-indent-line coffee-indent-line org-indent-line yaml-indent-line
    haskell-indentation-indent-line haskell-indent-cycle haskell-simple-indent
    yaml-indent-line))

Two. haskell-indentation-mode provides its own haskell-indentation-newline-and-indent function that is directly bound to RET and as if electric indent was always on.

Anyway this issue got offtopic, lets close it. If somebody experiences staricase problem again please report a new issue. Thanks.