Fixes backslash unicode autocompletion

[VarLad]

Fixes #1807

[VarLad]

Currently, the purpose of this PR is to catch all the issues which occur because of this fix, and try solving it here before merging

[archseer]

This seems to be the reverse of what you want: you want to scan backwards while char is not whitespace (stop on first whitespace)

[VarLad]

@archseer doing !chars::char_is_whitespace seems to break everything
Another solution proposed @ChrHorn chars::char_is_word(*ch) || *ch == '\\' which solves this particular problem.
I guess we can go along with that for now?
It would be nice if we could define this logic for the Julia LSP specifically so as to not break other langauges

[ChrHorn]

As mentioned, using only whitespace as a separator will break ., ::, etc.

An alternative proposal

I am not really sure that we actually need the filter by prefix logic. Just send all the results and let the UI handle the filerting/sorting. Would then be able to match expressions that don't start with the same prefix. That's how it's done in VSCode.

VSCode Example:

Sorting is currently alphabetical in Helix and would have to be improved.

[VarLad]

@archseer For now I've used the same workaround as @ChrHorn
It would be nice if we were able to do this in a "more general manner" (this one seems like a hack for Julia)

[archseer]

The more general way would be to use the trigger_character list from the LSP if available:

https://github.com/helix-editor/helix/blob/8493b5fca68097efd258ff0ffbcb6863e0c362c9/helix-term/src/commands.rs#L2676-L2682=

So you keep scanning back until you hit one of these characters (if the list is empty, just use is_word like before)

[VarLad]

@archseer Does this make sense? 😅

[the-mikedavis]

I think these are a bit more idiomatic

[VarLad]

It still doesn't trigger on . (for example, Base.)
That worked before (without this PR), so something is wrong with the logic here.
On the other hand, \ does trigger now

[VarLad]

Here's a summary of discussion on Matrix:

I talked to the folks who implemented the LS in Julia, and according to them, cases like :: and \\ are "same as completions after any whitespace". I went back to try implementing stopping at whitespaces.
I found the fact that let offset = iter.take_while(|ch| chars::char_is_whitespace(*ch)).count(); gives a problematic (although mostly working) result, while let offset = iter.take_while(|ch| !chars::char_is_whitespace(*ch)).count(); as weird.
Since the logic says that we should stop at the first whitespace, which didn't work, while its opposite seems to work.
An explanation by @ChrHorn is,

The good results with char_is_whitespace, are because empty prefix. The prefix is only used to prefilter the results. So empty prefix means not filtering by prefix. You send the whole completion responds to the UI which then does another round of filtering, that's why it works. The sorting is currently broken, that's why the results are a bit unordered.

Awaiting further input from @archseer on how to approach this

[maxtremblay]

Hello, I just want to say that I would really like this feature to be merged. Anything I can do to help?

[ChrHorn]

Best way to fix this issue is getting #3215 merged. Then we could get rid off the prefix filter which would also fix the backslash issue.