Make eldoc-mode fully subsume haskell-doc-mode

[gracjan]

In light of issues mulled over in #649 and #820 haskell-doc-mode should be removed, functionality should be put in eldoc-mode. eldoc actually works already with haskell-mode:

  (set (make-local-variable 'eldoc-documentation-function)
       'haskell-doc-current-info)

• find out if the above snippet is all that is needed for eldoc-mode
• implement what is missing
• write some coverage tests
• remove functionality that duplicates eldoc-mode from haskell-doc.el
• remove mentions of haskell-doc-mode from texi
• remove mentions of haskell-doc-mode from wiki
• remove mentions of haskell-doc-mode from all the random wikis and tutorials out there
• remove mentions of haskell-doc-mode from spacemacs (if needed)

[fice-t]

Few things I noticed:

• haskell-doc-mode already seems to be absent from the wiki and .texi
• spacemacs doesn't appear to have any mention of it either; they use haddock
• haskell-doc-prelude-types and its generator haskell-doc-extract-and-insert-types appear to be Haskell 98-specific, which I don't think is terribly useful. That should be changed. The same goes for haskell-doc-strategy-ids.
• There's a bug where the output of the type isn't prettified (unicode, etc) unless a process is active as the prettifying is done in haskell-process-get-type. This also affects eldoc in its current state.
• haskell-doc is really old, I even saw some xemacs-specific stuff in there 😧

[gracjan]

Hmm, is there anything of value to salvage from haskell-doc-mode?

[fice-t]

The unicode output of function type signatures seems helpful. For example, haskell-process-do-type could use that.

Otherwise I don't immediately see a whole lot. There's a lot of code that seems like it would have to be updated for this century. For example, there's an option to use inf-haskell to do completions, but a new version might as well use interactive-haskell-mode instead.

[gracjan]

So now we need a brave decision to kill it wholesale. Lets do like this:

1. In Monthly Report we announce the killing.
2. Wait one month so that people can chime in.
3. Do the kill after a month.

It seems to be a good time to do stable release before this happens.

[gracjan]

Reference: bbatsov/prelude#989

[fice-t]

@gracjan I'm not sure what the best way to do it is, but since Emacs 25 uses global-eldoc-mode by default, I'd say something should be done before 25.1 releases (~1 month I believe).

[gracjan]

Just yesterday we have fixed the 'async' message from haskell-doc-mode so there are people using it.

[fice-t]

True (though it did take a bit for a report to come in). I was thinking the least that could be done is to make-obsolete haskell-doc-mode and perhaps make it an obsolete alias of eldoc-mode, which would give people warnings when byte-compiling files using it.

The issue that might want to be avoided though is having both (global)eldoc-mode and haskell-doc-mode on at the same time.

[gracjan]

People do not enable haskell-doc-mode because they know what they are doing, they just copy some config from somewhere and run with it. For example Emacs Prelude stopped using haskell-doc-mode only just recently on my request.

[fice-t]

Well, it doesn't seem like the eldoc functionality is given much spotlight (it's not in the manual). If one didn't read the source code, read the 1.45 NEWS section, or already use eldoc mode in their prog-mode hook then it's not surprising they don't know about it.

Are you saying that a byte-compilation warning might not be good enough (i.e. they wouldn't byte-compile their init file)? If so, then one option is to straight up:

 (defun haskell-doc-mode (&optional arg)
  (interactive)
  (warn "Please use `eldoc-mode' instead.")
  (eldoc-mode arg))

I don't care either way, but as yesterday reminded me, it might be best to just get it out of the way as with Emacs 25 there's going to be a lot more eldoc-mode users.