Compliment-like features

[bbatsov]

In order to provide a similar experience for users of both Clojure and ClojureScript it'd be nice if the completion handling code supported several compliment features:

• context aware completion
• fuzzy completion

I guess those should be straightforward to port?

[cichli]

We can definitely mirror the contextual completion in ns declarations (see ns_mappings.clj in compliment). It'll take a bit of tweaking since ns is a little different in Clojurescript, but it's fundamentally the same.

Fuzzy matching should be easy; we just need to switch out the following line in complete.clj:

(filter #(.startsWith % prefix))

[bbatsov]

Well, I'd love to see this for cljs and I know many other people would love to see it as well. :-)

[cichli]

I actually have a branch with flx-like fuzzy completion (using Damerau-Levenshtein and a QuickSilver-like algorithm) - the problem I encountered is that emacs has completion-styles now, so a lot of the fuzzy matches get filtered out. We'd need a client option to tell CIDER to use its own fuzzy matching and ignore completion-styles.

These algorithms are also inherently tunable - should we let users pick the scoring thresholds? If so it might be best to return all potential candidates to the middleware and have the fuzzy-matching logic there (so the same client settings apply to both clj and cljs).

The other option is to return all potential candidates to emacs and let completion-styles filter them, but some caching would need to occur. It's noticeably sluggish if the backend returns all potential candidates that often. Even with caching, doing flx-like matching in emacs is probably too slow regardless - see company-mode/company-mode#47 - so I think this option is out (unless you think the built-in initials completion style is good enough).

[cichli]

I will work on the ns-form / locals contextual completion, though :-).

[bbatsov]

We'd need a client option to tell CIDER to use its own fuzzy matching and ignore completion-styles.

Makes sense to me.

These algorithms are also inherently tunable - should we let users pick the scoring thresholds? If so it might be best to return all potential candidates to the middleware and have the fuzzy-matching logic there (so the same client settings apply to both clj and cljs).

I'd suggest keeping things simple and just returning whatever compliment and cljs-tooling are computing. Perhaps we can adjust this down the road.

The other option is to return all potential candidates to emacs and let completion-styles filter them, but some caching would need to occur. It's noticeably sluggish if the backend returns all potential candidates that often. Even with caching, doing flx-like matching in emacs is probably too slow regardless - see company-mode/company-mode#47 - so I think this option is out (unless you think the built-in initials completion style is good enough).

I'm definitely not a fan of this approach.

[bbatsov]

Btw, a related ticket clojure-emacs/cider#467

[cichli]

@bbatsov I'll rebase the fuzzy-completion branch once #16 is merged.

[bbatsov]

Any progress here?

[bbatsov]

I guess not. :-)

[dpsutton]

@cichli is there any chance you still have this branch?

[bbatsov]

Likely this is the only commit in this branch - cichli@352f42c

[MalloZup]

autogenerated with https://github.com/MalloZup/doghub: issue inactive since 450 days. Please update the issue or close it