It may be desired for the exact matches to show first. Though it may need some moderate refactoring.

It may be desired for the exact matches to show first. Though it may need some moderate refactoring.

I don't know how other editors handle the order, vscode already does it by choosing the exact match first.

The matches are case-insensitive. The first few are matches from the start, then some matches not from the start, then some fuzzy matches (these exist because the completion list is computed on G).

If the completion list is computed from GLM, then it looks like

How about the following situation?

Abc <- 1
aBc <- 2

What do Ab and aB trigger?

Thank you for working on it. Didn't have the time to test it. By the way, am I correct that for

abc <- 1
bca <- 1

ab will trigger both abc and bca?

ab will trigger both abc and bca?

Currently, no. I only switch from startWith(x, token) to grepl(token, x, ignore.case = TRUE). The completions are case-insensitive and should include the token (not necessarily from the start). For example,

abc <- 1
cab <- 2

Ab should return both abc and cab, but Ca should only return cab.

Given the returned completion items, it is then up to the editor to determine how to present these items.

I am a bit confused why for example as.table is there when ab is typed. The a the b in as.table are not consecutive.

Perhaps it was because the list was generated by the server when a was typed. And when b was typed, VSCode does not query the server and ,in order to speed up things, only do simple filtering?

When a is pressed, the completion request is sent, and it got all results that contain a or A, and isIncomplete = false. Then the following typing does not trigger completion request and then vscode does the filtering and sorting according to the token under cursor.

/**
  * This list it not complete. Further typing should result in recomputing
  * this list.
  */
  isIncomplete: boolean;

If the token at cursor is already ab and then c is pressed, then a completion request is sent with token abc, and we only get results that case-insensitively contains abc.

Thank you for the explanation. Totally forgot this setting. It makes total sense.