Using treesitter completions with this plugins makes the whole insert mode quite laggy. Can anything be done about that or is it unavoidable?

Hi,
I keep getting this error:

E5108: Error executing lua .../packer/start/cmp-treesitter/lua/cmp_treesitter/init.lua:51: bad argument #1 to 'ipairs' (table expected, got nil)```

Thank you for looking into it.
Regards,
Omar

i want to set the highlight groups in cmp but im not getting how to, i tried
vim.api.nvim_set_hl(0, "CmpItemKindVariable", { fg = "green", bg = "NONE" })
but it is not applying to treesitter variable, it applied to lsp variable idk why

Maybe add examples on how this is useful. Between all the sources that cmp can have, I can't legitimately tell what this does for me. I love both cmp and treesitter, so I'd love to see if I get benefit from this.

stack traceback:
        ...rapped-bad218c/share/nvim/runtime/lua/vim/treesitter.lua:28: in function '__index'                                            
        ...s/start/cmp-treesitter/lua/cmp_treesitter/treesitter.lua:43: in function <...s/start/cmp-treesitter/lua/cmp_treesitter/treesitter.lua:31
>

This appears briefly upon typing in insert mode and typing... I'm not sure how to get more context for what's happening here.

I occasionally will get this error. I've not really found it to be reproducible though:

Error executing vim.schedule lua callback: vim/shared.lua:0: after the second argument: expected table, got nil
stack traceback:
        [C]: in function 'error'
        vim/shared.lua: in function 'validate'
        vim/shared.lua: in function 'tbl_deep_extend'
        .../packer/start/cmp-treesitter/lua/cmp_treesitter/init.lua:19: in function 'get_keyword_pattern'
        .../nvim/site/pack/packer/start/nvim-cmp/lua/cmp/source.lua:163: in function 'callback'
        .../site/pack/packer/start/nvim-cmp/lua/cmp/utils/cache.lua:36: in function 'get_default_replace_range'
        ...e/nvim/site/pack/packer/start/nvim-cmp/lua/cmp/entry.lua:45: in function 'new'
        .../nvim/site/pack/packer/start/nvim-cmp/lua/cmp/source.lua:316: in function ''
        vim.lua: in function <vim.lua:0>

Maybe it makes sense to you?

Hi there!

I'm experiencing some strange issue with cmp-treesitter with [quotes] autopairing:

1. when I trying to complete some substring inside quotes (and it finds corresponding node for completion, and it is also inside quotes), cmp-treesitter suggest that quoted node as completion (dunno why TS doesn't cut quotes when collecting nodes 🤷).
   This causes situation that completed string have two closing quotes (which causes syntax errors).

This is demonstration:

Is it possible to maybe somehow ask cmp-treesitter to cut quotes before passing node to cmp?
Or, maybe, it is possible to somehow ask cmp (or cmp-treesitter?) to take closing quote into account (like it does with middle-word/fuzzy completions)?

2. something similar happens when I trying to comple some "common" comments (like luacheck (linter) directives).

a. I start comment block
b. I type lua as start of word-to-be-completed
c. I see -- luacheck: ... as treesitter completion
d. I select it and I see it inserted one more -- (so there is double commented string, which, by the way, doesn't count as valid luacheck directive).

Here is a demonstration:

So, it seems, in that situation it doesn't even count "left" part of completeion...

Can that things be fixed?

Or it is all on cmp itself behalf, and cmp-treesitter have nothing to do here?

async fn run(options: Opt) -> Result<()> {
    let (certs, key) = if let (Some(key_path), Some(cert_path)) = (&options.key, &options.cert) {
      key_log
        let key = fs::read(key_path).context("failed to read private key")?;
        let key = if key_path.extension().map_or(false, |x| x == "der") {
            rustls::PrivateKey(key)
        } else {
            let pkcs8 = rustls_pemfile::pkcs8_private_keys(&mut &*key)
                .context("malformed PKCS #8 private key")?;

is sample Rust code from quinn (server.rs in examples).

You see that Some(key_path) defines a var you want to complete.

That Some(key_path) get's understood as identifier (See HERE)

    let_declaration [62, 4] - [121, 6]
      pattern: tuple_pattern [62, 8] - [62, 20]
        identifier [62, 9] - [62, 14]
        identifier [62, 16] - [62, 19]
      value: if_expression [62, 23] - [121, 5]
        condition: let_condition [62, 26] - [62, 95]
          pattern: tuple_pattern [62, 30] - [62, 63]
            tuple_struct_pattern [62, 31] - [62, 45]
              type: identifier [62, 31] - [62, 35] << HERE
              identifier [62, 36] - [62, 44]
            tuple_struct_pattern [62, 47] - [62, 62]
              type: identifier [62, 47] - [62, 51]
              identifier [62, 52] - [62, 61]

Not all identifiers are declarations.

https://github.com/ms-jpq/coq_nvim also has a treesitter based completion implementation. So there might be yet another trace for joint efforts ?

What I am exactly interested in is

1. preserve order (above cursor, closest first) because I am most likely to be using them again and it means least typing effort
   You seem to have implemented caching, and tell that treesitter is slow. The playground indicated might be fast ?

I have uploaded my code here so that you get a feeling how it behaves and how much it saves.
https://github.com/MarcWeber/nvim-cmp-local-identifiers-by-regex-completion-quick-and-dirty

Could you please share what font is being used in the screenshots, especially for those gorgeous italics?