Completion Improvements to LEM

[cianoc]

Lem currently has quite good completions functionality, but obviously it could be improved. Emacs out of the box offers more flexibility, and Emacs also has a bunch of other completion systems (Company, etc).

Which of their features do we wish to steal, and how can we improve completion so it is flexible, powerful and customizable?

[cianoc]

A pretty good overview of current Emacs completion features:
Mastering Emacs - Minibuffer Completion

Company Mode - what does this do well? Missing features?

Orderless - this is quite a powerful way to do completion. Some completion modes in Lem kind of do this already.

Vertico - this does a lot (too much maybe), and I think a lot of its behavior is typical emacs overkill. Still, there's some stuff here.

Consult - this allows you to perform actions on a selection in the minibuffer. There are other extensions in Emacs that also support this. Definitely not necessary, but very nice way to work when you get used to it.

Embark - similar to consult, only it operates as a popup. Again very cool, but very much an enhancement.

Marginalia - adds contextual information to a completion (e.g. the documentation for M-x commands). We have some of this, it would be nice if it could be improved. Ideally this would be extensible by the user.

[vindarel]

What I would appreciate:

• less needs to press TAB
• working completion with spaces in the search input (like typing "proj swi" would find "project-switch")
• fuzzy completion
  • see https://github.com/apr3vau/lw-plugins has fuzzy-matching
  • see https://github.com/vindarel/fuzzy-match the first algorithm used in the Next (now Nyxt) browser: "The parameters are hand-picked for the results to feel natural. A candidate that starts with the input substring should appear first. For example, we use the Damerau-Levenshtein distance thanks to the MK-STRING-METRICS library under the hood, but we don't obey to its result."

[cianoc]

I would add:

• Setting for completions showing automatically (perhaps by default).
  • Tab should probably work differently for automatic completions, so that will require a little thought.
• The ability to choose different algorithms for different completions (e.g. - files should probably work a little differently to command).
  • It would be nice if the user had control over this at some point, though I think that should be a long term goal. And I'd much rather it work well out of the gate for 90% of users.
• Possibly a mechanism for combining algorithms, and defining the matching order (similar to what you've done for history)
  • Fuzzy matches should probably come after exact matches - for example.
• Hard agree on space separation.

[cxxxr]

I am concerned that the references are too biased toward GNU Emacs.
I feel that other editors have better features.

[cianoc]

That's totally fair. I added references that I'm familiar with, but I'm mostly concerned with making LEM as good at completion as we (well me - as I'm planning to do a lot of this work) can make it.

If you add other references that would be super helpful, then I'll try to go through and collate a set of features/behaviors that hopefully we can all agree on.

[cianoc]

I think that Emacs today has some good completion features built in (though you have to customize it to make them work properly - out of the box it's terrible), and some of the extensions also have some features/ideas. There are also some terrible UI decisions, bad UX and these things are way too customizable.

I'd like to identify the stuff that people feel is good, anything that people think we should avoid and just build something that works well out of the box, has some reasonable customization and which feels cool (e.g. company/vertico can sometimes feel a bit like magic). But I have zero interest in replicating Emacs in LEM. There are reasons I want to stop using Emacs.

One huge advantage that LEM already has over Emacs is that it uses floating windows for completion!