New dabbrev completions too eager

[geraldus]

Currently I'm writing some GHCJS code and didn't setup interactive version yet, so I always have dabbrev completions. With 10-20 opened buffers dabbrev completion method becomes too eager already, i.e. completions takes too long. It is possible to limit dabbrev list of buffers to search completions. For example, try to lookup nearest folder contains .git, or .cabal file, set some haskell-completions-top-folder local var, and when searching dynamic abbrevs take this var in account. When session is available, search for d-abbrevs in buffers assigned to current session.

What do you think?

[gracjan]

I use dabbrev by M-/ command key. I would love to have M-/ do dabbrev, completion kept only for more context-sensitive semantic information only.

[geraldus]

If you don't use company-mode I think you does not encounter this issue. Here is a little demo. Company mode, no REPL (hence completions are always comes from dynamic abbrevs), 50 buffers opened.

[geraldus]

If someone want to fix this before I find a time this might be helpful dabbrev.el.

[gracjan]

@geraldus: Why do you use dabbrev--reset-global-variables? It is here:

haskell-mode/haskell-completions.el

Line 272 in a814671

(dabbrev--reset-global-variables)

As I understand dabbrev has some caching mechanism to speed up expansions, resetting kills this optimization. Or am I mistaken?

[gracjan]

@geraldus: My point about M-/ was that company-mode should NOT use dabbrev as source. I envision it like this:

• company-mode uses only contextual sources for completions: :complete, PRAGMA, import
• M-/ uses non-contextual sources for completions: abbrev, dabbrev

So 'smart' stuff is for company/completion, non-smart for dabbrev. Am I mistaken?

[cocreature]

@gracjan I agree if people want to use dabbrev via company mode they can still add it themselves but I am not sure what we gain by doing that from haskell mode.

[geraldus]

Sometimes REPL is not running for some reason (in my case there is no GHCJSi available yet) or user just does not want to run it, that is, :complete command is not available too. So how company should look up completions?

[gracjan]

@geraldus: My (original) company setup has this part:

...
Company Backends:
INS DEL Back-end: Value Menu completion-at-point-functions
INS DEL Back-end: Value Menu Merged Back-ends:
            INS DEL Back-end: Value Menu dabbrev for code
            INS DEL Back-end: Value Menu GNU Global
            INS DEL Back-end: Value Menu etags
            INS DEL Back-end: Value Menu Programming language keywords
            INS
INS DEL Back-end: Value Menu Oddmuse
INS DEL Back-end: Value Menu Files
INS DEL Back-end: Value Menu dabbrev for plain text

Does this mean dabbrev is just handled by company-mode and there is no need to do anything about it inside haskell-mode?

[gracjan]

@geraldus: If interpreter is not running then :complete is not an available source. What is left is of the semantic sources, currently PRAGMA, LANGUAGE and that is enough. Someday we might add more sources

dabbrev is available always as M-/, so nothing is lost.

[gracjan]

@cocreature: How do they 'add it themselves'? It is of vital importance to me that stuff works out of the box as it should.

[cocreature]

@gracjan IIRC there is a company-dabbrev backend built into company mode so you can just add that to your backend. Looking at it we don't seem to provide a company backend atm, so I am not quite sure how to easily combine them. Of course it is important that stuff works out of the box, but to be frank the current dabbrev integration doesn't work (well enough). Mostly it hangs up my emacs.

[gracjan]

In that case we should make sure that `company-dabbrev` works with `haskell-mode` in case somebody decides to use it. `haskell-mode` should NOT do anything special with dabbrev because this is code duplication and is problematic to get right.

[geraldus]

Well, I think we should have some overrides for company's dabbrev source, for example case sensitivity. And if it is possible we should have some notes about completions itself and company-mode particularly.

[gracjan]

We have some configuration for dabbrev in haskell-mode already.

[geraldus]

Ok, to test this I need to rewrite some code in completion function itself, I'll update pull request soon.

[cocreature]

Could we just disable dabbrev until this is fixed? It hangs up my emacs for multiple seconds which is for obvious reasons pretty annoying. I'll be happy to send a pullrequest for that if you agree.

[geraldus]

Please excuse me my delay and inconveniences brought by last changes. I have some troubles here and can do nothing at the moment. I will continue as soon as possible! Go ahead!

[geraldus]

@cocreature by the way haskell-mode have some configuration for dabbrev already. When you'll disable my stuff, can you please test how it works and share your feelings? The one thing I've already mentioned is that it ignores case and making completions useless in most cases. In the end, I think it is possible to remove dabbrev-stuff which I provided replacing it with a better configuration to make everybody happy out-of-box (when REPL completions are no available).

[gracjan]

Pull request disabling dabbrev welcome.

[gracjan]

Fixed by #838.