Pueden leer la versión en español aquí.
Yes, we can. The complexity of the setup will depend on what you want to achieve in your workflow. But if you know your way around lua you should be able to understand the "boilerplate" needed to get a decent setup.
Let's figure out how to make it work.
The building blocks
It turns out we only need two functions. One to initialize the language server. One that notifies text changes to the server:
vim.lsp.start_client()
: This function creates a "client object" that handles all communications with a language server.vim.lsp.buf_attach_client()
: It notifies all text changes to the language server.
A minimal and inefficient example using typescript's language server would look like this.
local launch_tsserver = function()
local config = {
cmd = {'typescript-language-server', '--stdio'},
name = 'tsserver',
root_dir = vim.fn.getcwd(),
}
local client_id = vim.lsp.start_client(config)
if client_id then
vim.lsp.buf_attach_client(0, client_id)
end
end
vim.api.nvim_create_user_command(
'LaunchTsserver',
launch_tsserver,
{desc = 'Start tsserver'}
)
Now with the command LaunchTsserver
we would be able to start a typescript-language-server
process and get diagnostics in the current file.
Why is this inefficient? Because it would only work on a single file. It'll spawn a new process each time we use it. What you really want is to "share" a single process of typescript-language-server
for your project.
Project setup
So what's the missing piece that would make this usable in a project? An autocommand. We need to tell neovim we want to attach the server to a buffer everytime we open a new file of a specific filetype.
local filetypes = {
'typescript',
'javascript',
'typescriptreact',
'javascriptreact',
'typescript.tsx',
'javascript.jsx'
}
local buf_attach = function()
vim.lsp.buf_attach_client(0, client_id)
end
autocmd_id = vim.api.nvim_create_autocmd('FileType', {
desc = string.format('Attach LSP: %s', client_name),
pattern = filetypes,
callback = buf_attach
})
When we create an autocommand with nvim_create_autocmd
we get an id, which can be used to delete it later on. To delete the autocommand we would call nvim_del_autocmd
.
vim.api.nvim_del_autocmd(autocmd_id)
But what would be the best time to create this autocommand? After the server is ready, of course. We can use the on_init
hook to setup the autocommand and on_exit
to clean it up.
If we apply all this knowledge to our launch_tsserver
function we would have this.
local launch_tsserver = function()
local autocmd
local filetypes = {
'typescript',
'javascript',
'typescriptreact',
'javascriptreact',
'typescript.tsx',
'javascript.jsx'
}
local config = {
cmd = {'typescript-language-server', '--stdio'},
name = 'tsserver',
root_dir = vim.fn.getcwd(),
}
config.on_init = function(client, results)
local buf_attach = function()
vim.lsp.buf_attach_client(0, client.id)
end
autocmd = vim.api.nvim_create_autocmd('FileType', {
desc = string.format('Attach LSP: %s', client.name),
pattern = filetypes,
callback = buf_attach
})
if vim.v.vim_did_enter == 1 and
vim.tbl_contains(filetypes, vim.bo.filetype)
then
buf_attach()
end
end
config.on_exit = vim.schedule_wrap(function(code, signal, client_id)
vim.api.nvim_del_autocmd(autocmd)
end)
vim.lsp.start_client(config)
end
Sending the settings
Our example is not complete just yet. Most servers have unique options we can use to enable some feature or tweak some behaviour, and right now we are not sending that information to the server.
After the server has started we need to send a "notification" which will carry the data we want to send to the server. So in the on_init
hook we should add this.
if results.offsetEncoding then
client.offset_encoding = results.offsetEncoding
end
if client.config.settings then
client.notify('workspace/didChangeConfiguration', {
settings = client.config.settings
})
end
Neovim's documentation suggest we should set the encoding of the client before sending any request or notification. So if the server responds with offsetEncoding
we use that to configure the client.
Now if the client configuration contains an option called settings
we send that to the server. Is worth mention that client.config
is actually a copy of the arguments we pass to vim.lsp.start_client()
.
What about some keybindings?
Right? I mean, you probably want to use some of those LSP features with just a few keystrokes. How do we proceed? The documentation says we should use the on_attach
hook. We could hard code the keybindings in the launch_tsserver
function but it doesn't feel right. I suggest you trigger a "user event" in on_attach
so you can define the keybindings anywhere in your neovim configuration.
config.on_attach = function(client, bufnr)
vim.api.nvim_exec_autocmds('User', {pattern = 'LspAttached'})
end
Now you can define your autocommand some place else.
vim.api.nvim_create_autocmd('User', {
pattern = 'LspAttached',
desc = 'LSP actions',
callback = function()
local bufmap = function(mode, lhs, rhs)
local opts = {buffer = true}
vim.keymap.set(mode, lhs, rhs, opts)
end
-- Setup 'omnifunc' completion
-- It is triggered using Ctrl-x + Ctrl-o
vim.bo.omnifunc = 'v:lua.vim.lsp.omnifunc'
-- Use the language server as a source for "tags" whenever possible
vim.bo.tagfunc = 'v:lua.vim.lsp.tagfunc'
-- Displays hover information about the symbol under the cursor
bufmap('n', 'K', '<cmd>lua vim.lsp.buf.hover()<cr>')
-- Jump to the definition
bufmap('n', 'gd', '<cmd>lua vim.lsp.buf.definition()<cr>')
-- Jump to declaration
bufmap('n', 'gD', '<cmd>lua vim.lsp.buf.declaration()<cr>')
-- Lists all the implementations for the symbol under the cursor
bufmap('n', 'gi', '<cmd>lua vim.lsp.buf.implementation()<cr>')
-- Jumps to the definition of the type symbol
bufmap('n', 'go', '<cmd>lua vim.lsp.buf.type_definition()<cr>')
-- Lists all the references
bufmap('n', 'gr', '<cmd>lua vim.lsp.buf.references()<cr>')
-- Displays a function's signature information
bufmap('n', '<C-k>', '<cmd>lua vim.lsp.buf.signature_help()<cr>')
-- Renames all references to the symbol under the cursor
bufmap('n', '<F2>', '<cmd>lua vim.lsp.buf.rename()<cr>')
-- Selects a code action available at the current cursor position
bufmap('n', '<F4>', '<cmd>lua vim.lsp.buf.code_action()<cr>')
bufmap('x', '<F4>', '<cmd>lua vim.lsp.buf.range_code_action()<cr>')
-- Show diagnostics in a floating window
bufmap('n', 'gl', '<cmd>lua vim.diagnostic.open_float()<cr>')
-- Move to the previous diagnostic
bufmap('n', '[d', '<cmd>lua vim.diagnostic.goto_prev()<cr>')
-- Move to the next diagnostic
bufmap('n', ']d', '<cmd>lua vim.diagnostic.goto_next()<cr>')
end
})
Complete code
And now if we apply the new changes to our example this would be it.
local launch_tsserver = function()
local autocmd
local filetypes = {
'typescript',
'javascript',
'typescriptreact',
'javascriptreact',
'typescript.tsx',
'javascript.jsx'
}
local config = {
cmd = {'typescript-language-server', '--stdio'},
name = 'tsserver',
root_dir = vim.fn.getcwd(),
capabilities = vim.lsp.protocol.make_client_capabilities(),
}
config.on_attach = function(client, bufnr)
vim.api.nvim_exec_autocmds('User', {pattern = 'LspAttached'})
end
config.on_init = function(client, results)
if results.offsetEncoding then
client.offset_encoding = results.offsetEncoding
end
if client.config.settings then
client.notify('workspace/didChangeConfiguration', {
settings = client.config.settings
})
end
local buf_attach = function()
vim.lsp.buf_attach_client(0, client.id)
end
autocmd = vim.api.nvim_create_autocmd('FileType', {
desc = string.format('Attach LSP: %s', client.name),
pattern = filetypes,
callback = buf_attach
})
if vim.v.vim_did_enter == 1 and
vim.tbl_contains(filetypes, vim.bo.filetype)
then
buf_attach()
end
end
config.on_exit = vim.schedule_wrap(function(code, signal, client_id)
vim.api.nvim_del_autocmd(autocmd)
end)
vim.lsp.start_client(config)
end
vim.api.nvim_create_user_command(
'LaunchTsserver',
launch_tsserver,
{desc = 'Start tsserver'}
)
Can we do better?
Sure. I mean, you can hide almost all the boilerplate inside another function and reduce the noise.
Just imagine this.
local launch_tsserver = function()
local config = make_config({
cmd = {'typescript-language-server', '--stdio'},
name = 'tsserver',
filetypes = {
'typescript',
'javascript',
'typescriptreact',
'javascriptreact',
'typescript.tsx',
'javascript.jsx'
}
})
vim.lsp.start_client(config)
end
vim.api.nvim_create_user_command(
'LaunchTsserver',
launch_tsserver,
{desc = 'Start tsserver'}
)
How does make_config
look like? Well... that's homework, my friend. You can implement it anyway you want. I've already showed you all the code you need to make it possible.
You really want to know how I would do it? The answer is in this github repository: VonHeikemen/nvim-lsp-sans-plugins
Conclusion
We learned enough about neovim's builtin LSP client to create our own little setup. We know how to initialize the language server. We can "share" the same server across multiple files. In the process we worked with autocommands, now we know how to create and delete them. We could totally manage some LSP servers without plugins.
Sources
Thank you for your time. If you find this article useful and want to support my efforts, consider leaving a tip in ko-fi.com/vonheikemen.
Top comments (1)
I don't know why this post doesn't attract people. But for me, nothing explains better than yours.