Add resolve timeout for accepting

[datwaft]

Make sure you have done the following

• I have updated to the latest version of blink.cmp
• I have read the README

Bug Description

I have a big (around 40k lines of code) Playwright project which heavily uses test fixtures which have a noticeable effect on the performance of the LSP server for TypeScript.

I have noticed lag after accepting a completion item, it takes around 1 or 2 seconds and any text written after accepting the completion item is removed when the completion item is inserted.

The wait for the completion items to show up in the completion list is expected (I also have the same issue with nvim-cmp which is probably the fault of the LSP) but the lag after accepting a completion item in which I cannot write anything because it will be removed after the completion item is inserted seems like an issue with blink.cmp (I haven't seen it when using nvim-cmp).

Here is a recording of the issue:

Screen.Recording.2024-12-17.at.14.09.55.mov

Relevant configuration

You can see my configuration here: https://github.com/datwaft/nvim.conf/blob/d834a1b01457ddcb34f6644efd1caa345d1182e4/lua/packages/completion.lua

If it is relevant I am accepting with <space> and have the following completion configuration:

completion = {
  list = {
    selection = "manual",
  },
  accept = {
    auto_brackets = {
      enabled = false,
    },
  },
},

neovim version

NVIM v0.11.0-dev-1369+gb03e790cdd

blink.cmp version: branch, tag, or commit

main (c3ef922)

[Saghen]

When you get the chance, open ~/.local/share/nvim/lazy/blink.cmp/lua/completion/accept/init.lua and check how long the .resolve call takes. You can use vim.uv.hrtime() to get the current time (in nanoseconds)

[datwaft]

I modified ~/.local/share/nvim/lazy/blink.cmp/lua/blink/cmp/completion/accept/init.lua like this:

Here are some of the results:

It took 17583 nanoseconds for `accept`
It took 19333 nanoseconds for `accept`
It took 52500 nanoseconds for `accept`
It took 834666 nanoseconds for `accept`
It took 834750 nanoseconds for `accept`
It took 751750 nanoseconds for `accept`
It took 1170625 nanoseconds for `accept`

Because of how the printed time is growing I think I am doing something wrong, any suggestions?

[datwaft]

Btw, I haven't tried the new commit. I am going to use it and see if it helps.

[Saghen]

Sorry I should've clarified, put the print inside the :map. It seems likely that new commit will help reduce the delay, although it may not fully get rid of it

[datwaft]

Here are some results (with the new commit):

It took 2109791 nanoseconds for `accept`
It took 1920208 nanoseconds for `accept`
It took 1444042 nanoseconds for `accept`
It took 1561542 nanoseconds for `accept`
It took 2257792 nanoseconds for `accept`
It took 2500791 nanoseconds for `accept`
It took 1195458 nanoseconds for `accept`

[datwaft]

NOTE: I am still seeing the lag sometimes.

[Saghen]

Haha, once again I should've clarified 😂 The start_time should be before the .resolve and the end_time should be inside the :map so we can see how long that async call takes. Sorry for the back and forth on this

[datwaft]

😅 🤣

Here are the new results:

It took 408560458 nanoseconds for `accept`
It took 824164041 nanoseconds for `accept`
It took 346669333 nanoseconds for `accept`
It took 25417 nanoseconds for `accept`
It took 1018236167 nanoseconds for `accept`
It took 363493625 nanoseconds for `accept`

[Saghen]

It seems like nvim-cmp skips the resolve if it takes more than 80ms by default, whereas yours are taking up to 1s