Vim has built-in support for completions in insert or replace modes,
mostly based on the content of the current file and included files.
By default, it uses separate Emacs-style key bindings (<C-x C-key>)
for different kinds of completions.
See :help ins-completion for more information.
The modern approach is utilizing language server protocol, a.k.a. LSP, and machine learning.
The concept of LSP is pioneered by vscode, so a lot of language servers are written in Node.js and TypeScript. While mostly used for LSP, CoC brings the ecosystem of vscode extensions to vim, which makes porting them to vim easier.
Then Neovim 0.5+ introduces native LSP support, based on Lua instead of Node.js. However, as mentioned above, since a lot of language servers are written in Node.js, most likely Node.js is still needed.
I use LunarVim, a Neovim distribution, which uses native LSP and will install supported language servers automatically upon opening files.
LunarVim disables automatic support for some file types. I have to install and enable LSP for them manually.
For example, to install the marksman LSP for Markdown, under LunarVim, type:
:LSPInstall marksman
After installation, edit
~/.config/lvim/after/ftplugin/markdown.lua
to configure LSP for the Markdown file type:
require("lvim.lsp.manager").setup("marksman")
-- marksman has no formatting capability.
local formatters = require "lvim.lsp.null-ls.formatters"
formatters.setup {
{ name = "prettier", filetypes = { "markdown" } },
}
Two popular options of machine learning assisted completion are Tabnine and GitHub Copilot.
Compared to Copilot, Tabnine has two advantages:
Free plan available.
I have used the free plan for a while. It does not include whole line and full-function code completion. And Tabnine Pro is $12 per month, slightly higher than Copilot ($10 per month).
Local machine mode.
Tabnine's free plan is based on the cloud. Tabnine Pro's offline installation requires contacting sales.
Thus, I use Copilot with LunarVim.