Feature request: more general support for external auto-completion

[echasnovski]

Right now plugin's vim.ui.input defines its own completion. It seems to conflict with other external (auto-)completion functionality. This is vaguely confirmed by hard-coded 'nvim-cmp' disabling. I understand that 'nvim-cmp' is the completion engine right now, but it would be nice to at least have notice somewhere that external completion engine should be disabled in specified buffer. For example, this caused a hard-to-find behavior in 'mini.completion' (echasnovski/mini.nvim#68).

[stevearc]

Happy to disable mini completion by default, but there should be nothing wrong with having it enabled. I disabled cmp completion because it's usually undesirable in the input buffer, not because it was causing errors. I believe the issues that were reported are coming from some interaction between mini completion and "stopinsert". I can reproduce the issue with the following code:

require("mini.completion").setup()
vim.defer_fn(function()
  vim.api.nvim_feedkeys("Gohelloworld", "t", true)
end, 100)
vim.defer_fn(function()
  vim.cmd([[stopinsert]])
end, 1000)
vim.defer_fn(function()
  -- Now any attempt to insert triggers the error
  vim.api.nvim_feedkeys("i", "t", true)
end, 1500)

[echasnovski]

Thanks for automated disabling. I've added the note in 'mini.completion' docs, so that is not that necessary, but still nice.

Your example left me really puzzled, though. Not really a 'mini.completion' issue, but issue with InsertCharPre autocommand. I think I managed to narrow it down to some sort of weird behavior of Neovim itself. For example, that code works fine if vim.cmd([[stopinsert]]) is replaced with vim.api.nvim_feedkeys(vim.api.nvim_replace_termcodes('<Esc>', true, true, true), "t", true). Which is weird because :h stopinsert quite literally says "Works like typing <Esc> in Insert mode".

I'll ponder a bit until submitting an issue and probably close this after. Does this sound good?

[stevearc]

Works for me. I wouldn't be surprised if it were some edge case within Neovim itself; I've noticed some odd behavior for startinsert in the past.

[echasnovski]

So from neovim/neovim#18395 it does indeed seem like a Neovim issue. Thanks for the code example! Closing as irrelevant to the plugin itself.