blog/content/software/nvim-latex.md

Title	Date	Lang	Author	Category	Tags	Slug	table-of-contents	Header_Cover	Summary
Neovim as a LaTex Development Environment	2023-10-14 12:00:00+0200	en	Fabrice	software	vim, neovim, latex, zathura	nvim-latex	true	../images/covers/fern-forest.jpg	How to turn Neovim into a full-fledged latex development environment

Introduction

LaTeX is a typesetting software for producing typographically sound printable documents that is mostly used by the scientific community (but not only) as it allows writing mathematics formulae in a somewhat not-that-much painful way, is shipped with bibliography engines, allows easy cross-referencing and automatically generates table of contents.

It is based on a markup language that allows the writers to focus on the content of the document and leaves the typesetting to the software (at least most of the time). It moreover enjoys many libraries that span from enabling new features to simpler version of more complete tools.

In this blog post we will see how to setup Neovim to manipulate LaTeX document while enabling modern features such as language server protocol and what you want from any LaTeX IDEs: backward and forward searches.

As a PDF reader, we will use zathura to show how to setup backward search (search from the document toward the source). It is a highly configurable, lightweight document viewer which natively enjoys vim-like shortcuts.

Ingredients

Before starting we will need several components to achieve this lofty goal of painlessly writing LaTeX documents with the best text editor.

• A configurable text editor to be able to write the document: Neovim. For that we will also need some plugins to unleash its full capability:
  • nvim-lspconfig: a plugin to facilitate the configuration LSP for Neovim.
  • nvim-cmp: a completion engine for Neovim.
  • vimtex: a language specific plugin for LaTeX files that supports many features such as accurate syntactic coloration, support of multi-files, add LaTeX-specific text objects, improved foldings and so on.
• texlab: to enable LSP features, you also a LSP server for vim to communicate with, which is exactly what texlab is.
• zathura: finally a PDF viewer, we will use zathura here, but vimtex supports many others with predefined setups. However you will have to look for the specific documentation of your pdf reader to enable reverse search if it is possible.

Setting Neovim up

Now that we have prepared everything, we need to setup Neovim to be up to the task. We will assume a blank configuration and starts from scratch. I got inspired by a blogpost about snippets in Neovim and used NVIM_APPNAME environment variables for testing this configuration. Please let me know if anything is not working as intended.

Being Lazy

Anyhow, we first need to install the different plugins that we need. For this purpose, I used the lazy plugin manager, but you can use whichever you see fit for the task.

-- Lazy Package Manager
local lazypath = vim.fn.stdpath("data") .. "/lazy/lazy.nvim"
if not vim.loop.fs_stat(lazypath) then
  vim.fn.system({
    "git",
    "clone",
    "--filter=blob:none",
    "https://github.com/folke/lazy.nvim.git",
    "--branch=stable", -- latest stable release
    lazypath,
  })
end
vim.opt.rtp:prepend(lazypath)

-- Packages
require("lazy").setup({
  "lervag/vimtex",
  "neovim/nvim-lspconfig",
  "hrsh7th/cmp-nvim-lsp",
  "hrsh7th/nvim-cmp",
})

In the code block above —in $NVIM_CONFIG/init.lua— the first part is to bootstrap lazy (so it can install itself if not already there) and the last block describe the installation of the following plugins : vimtex, nvim-lspconfig, nvim-cmp and finally cmp-nvim-lsp to glue the completion engine and lspconfig.

Now it is all good and done, but nothing is configured yet, and if you open a LaTeX file in this state, you will only enjoy the benefits of an unconfigured vimtex, which is already nice as is it, but not enough to achieve our goal. And it's a bit sad to have install three other plugins for nothing.

vimtex

It will be a bit anti-climatic after the previous teasing, but we will use vimtex as vanilla as possible… However, we still need, to tell it to use zathura as a pdf viewer:

vim.g.vimtex_view_method = "zathura"

This will allow vimtex to automatically open zathura after compilation, which is by default bound to <LocalLeader>ll. We now have to define <LocalLeader>, which I usually set to “,”:

vim.g.maplocalleader = ","

Now, you can use ,lv to view the current line in zathura, yay.

More can be then done, such as using vimtex folds, which are not enabled by default (contrary to what vim-latex was doing, which is the former plugin I used):

-- From: https://github.com/lervag/vimtex/blob/master/doc/vimtex.txt#L4671-L4713
vim.o.foldmethod = "expr"
vim.o.foldexpr="vimtex#fold#level(v:lnum)"
vim.o.foldtext="vimtex#fold#text()"
-- I like to see at least the content of the sections upon opening
vim.o.foldlevel=2

Now the sky is your limit, but to start with, here follows a quick list of what is possible now:

• Compile the document: ,ll
  • This also automatically generates a quickfix buffer which is quite complete… a bit too much sometimes. I used it as is to hunt for over/underfull hboxes, but you can filter them out by setting the vim.g.vimtex_quickfix_ignore_filters variable.
• View the current location in the document: ,lv
• Show table of content navigation: ,lt

• Using latex-specific text objects such as $ for math or e for environment (defined by \begin{…} and \end{…}).

• Insert command/environment : <F6>/<F7> (in normal and visual mode; these are not very accessible, but can be remapped).
• Support for TeX directives (which are common with others LaTeX' IDEs), such as %! TeX program = xelatex to specify a latex compiler.

• For machine-aided proofreading, you can also enable grammar checking tools, such as languagetool. I didn't check for grammalecte support for French yet, but it may prove to be an interesting endeavour.

Remark: vimtex claims that their coloration is more accurate than what tree-sitter, then if you are using nvim-treesitter, you may want to disable it for vimtex (it raises a warning otherwise):

require("nvim-treesitter.configs").setup({
	highlight = {
		enable = true,
		disable = { "latex", },
	},
})

Okay, that's all and good, but to quote wikipedia:

The goal of the protocol is to allow programming language support to be implemented and distributed independently of any given editor or IDE. In the early 2020s LSP quickly became a "norm" for language intelligence tools providers.

Source: https://en.wikipedia.org/wiki/Language_Server_Protocol

We are not early 2020s-ready for LaTeX yet, and even if we can send our current location to zathura, the contrary is not possible yet. Let us now address these two issues.

Language Server Protocol

Setting up language server protocol with Vim is a big morsel, and have been the topic of some tuppervim sessions at some point.

I'll present here a minimal configuration that should work with texlab:

-- Minimal lsp config
local lspconfig = require("lspconfig")
lspconfig.texlab.setup {}

Okay, that's all and good, we can see errors and warnings decorating the file like Christmas decorations, but we can't use any of the LSP tools such as obtaining information on a bibliography key, or rename a macro.

However, let us just remark that texlab is a pretty minimal LSP server, and don't implement the myriads of possible functionalities. Henceforth, I just copy-pasted the default example from the nvim-lspconfig Readme, tried the shortcuts one by one, and remove these which raised an error for “not implemented functionality”:

-- Use LspAttach autocommand to only map the following keys
-- after the language server attaches to the current buffer
vim.api.nvim_create_autocmd("LspAttach", {
	group = vim.api.nvim_create_augroup("UserLspConfig", {}),
	callback = function(ev)
		-- Enable completion triggered by <c-x><c-o>
		vim.bo[ev.buf].omnifunc = "v:lua.vim.lsp.omnifunc"

		-- Buffer local mappings.
		-- See `:help vim.lsp.*` for documentation on any of the below functions
		local opts = { buffer = ev.buf }
		vim.keymap.set("n", "gd", vim.lsp.buf.definition, opts)
		vim.keymap.set("n", "K", vim.lsp.buf.hover, opts)
		vim.keymap.set("n", "gR", vim.lsp.buf.rename, opts)
		vim.keymap.set("n", "gr", vim.lsp.buf.references, opts)
	end,
})

Which enables:

• Omnicompletion using LSP (I won't elaborate on this point, either you use it or not, but if you're using it, it may be useful to leave. I personally don't).
• Go to a definition, with gd, which can be a macro, a reference, or even a bibliography reference.
• Show the information about the element under the cursor using K, it can be useful to quickly check a reference. Note that pressing K twice jumps into the floating window, which can be useful to copy an article title to search for it somewhere else for instance.
• Rename a macro/variable among all files in the current working documents using gR. It's a lifesaver when renaming macros as it avoids writing regular expressions.
• Show each places where a reference appears with gr in a quickfix window. It allows checking where a formula is referenced or verifying if you cited yourself enough. I personally use telescope.nvim for that purpose as it is more readable, but it goes beyond the scope of this blogpost.

And that's it, we now simply have to enable the completion engine getting the configuration from the nvim-cmp readme file and the vimtex documentation, then pruning it.

-- nvim-cmp
local cmp = require("cmp")
cmp.setup({
	sources = cmp.config.sources({
			{ name = "nvim_lsp" },
			{ name = "buffer" },
		}),
	mapping = cmp.mapping.preset.insert({
			["<C-Space>"] = cmp.mapping.complete(),
			["<C-u>"] = cmp.mapping.scroll_docs(-4),
			["<C-d>"] = cmp.mapping.scroll_docs(4),
			["<C-l>"] = cmp.mapping.confirm({ select = true }),
		}),
})

And we're all good from Neovim's side. You can of course start fine-tuning it but it's not the purpose of this blogpost.

Plug it into zathura

Now that you tweaked your Neovim configuration so much that it now consumes 10GB of memory and takes 12s to launch using all your CPU cores, we can move to zathura.

One of the reasons I moved from vim-latex to vimtex is reverse search: to enable it with vim-latex, I was using nvim-remote which is a wrapper for nvim --listen with a lot of constraints, while the most annoying one is that if I used reverse search from a detached¹ zathura window without starting nvr first… then it is spawns the process which I cannot recover. Which usually happens when I'm in a rush to fix something quickly.

Fortunately, this is a thing of the past as it is possible to directly send a directive to vimtex which will look for the corresponding buffer and then open the file at the right location while following its state (which can be viewed with ,li). To do so, the documentation states that you have to launch the following command, where %l is the line in the file and %f is the name of the file:

nvim --headless -c "VimtexInverseSearch %l '%f'"

That's all and good, now we just have to tell Zathura which command to launch when doing backward search, which by default is done with Ctrl + left mouse button on the portion of the text you want to view in the code. To do that, the following configuration that you can put in $HOME/.config/zathura/zathurarc should do the trick:

set synctex true
set synctex-editor-command "nvim --headless -c \"VimtexInverseSearch %{line} '%{input}'\""

And… that's it! You can now go to the location you want in your file document, compile it on the fly, scrutinise the warnings to look for overfull hboxes!

Conclusion

In this blogpost, we saw how to minimally set up Neovim to work with latex using modern toolchains. You can use it as a base to then improve your workflow and write your documents in a breeze with neovim.

To summarise the configuration we used, it can be done in an init.lua file in your vim configuration directory:

-- Lazy Package Manager
local lazypath = vim.fn.stdpath("data") .. "/lazy/lazy.nvim"
if not vim.loop.fs_stat(lazypath) then
  vim.fn.system({
    "git",
    "clone",
    "--filter=blob:none",
    "https://github.com/folke/lazy.nvim.git",
    "--branch=stable", -- latest stable release
    lazypath,
  })
end
vim.opt.rtp:prepend(lazypath)

require("lazy").setup({
  "lervag/vimtex",
  "neovim/nvim-lspconfig",
  "hrsh7th/cmp-nvim-lsp",
  "hrsh7th/nvim-cmp",
})

-- vimtex
vim.g.vimtex_view_method = "zathura"
vim.g.maplocalleader = ","

vim.o.foldmethod = "expr"
vim.o.foldexpr="vimtex#fold#level(v:lnum)"
vim.o.foldtext="vimtex#fold#text()"
vim.o.foldlevel=2

-- Minimal lsp config
local lspconfig = require("lspconfig")
lspconfig.texlab.setup {}

-- Use LspAttach autocommand to only map the following keys
-- after the language server attaches to the current buffer
vim.api.nvim_create_autocmd("LspAttach", {
  group = vim.api.nvim_create_augroup("UserLspConfig", {}),
  callback = function(ev)
    -- Enable completion triggered by <c-x><c-o>
    vim.bo[ev.buf].omnifunc = "v:lua.vim.lsp.omnifunc"

    -- Buffer local mappings.
    -- See `:help vim.lsp.*` for documentation on any of the below functions
    local opts = { buffer = ev.buf }
    vim.keymap.set("n", "gd", vim.lsp.buf.definition, opts)
    vim.keymap.set("n", "K", vim.lsp.buf.hover, opts)
    vim.keymap.set("n", "gR", vim.lsp.buf.rename, opts)
    vim.keymap.set("n", "gr", vim.lsp.buf.references, opts)
  end,
})

-- nvim-cmp
local cmp = require("cmp")
cmp.setup({
  sources = cmp.config.sources({
    { name = "buffer" },
    { name = "nvim_lsp" },
  }),
  mapping = cmp.mapping.preset.insert({
    ["<C-Space>"] = cmp.mapping.complete(),
    ["<C-u>"] = cmp.mapping.scroll_docs(-4),
    ["<C-d>"] = cmp.mapping.scroll_docs(4),
    ["<C-l>"] = cmp.mapping.confirm({ select = true }),
  }),
})

and the following in your zathurarc file:

set synctex true
set synctex-editor-command "nvim --headless -c \"VimtexInverseSearch %{line} '%{input}'\""

Note that due to some technical limitations, it's not fully perfect. For instance, synctex is not fully accurate with beamer slides, and just select the whole slide instead of the selected text. It is still better than nothing in my opinion, and it's a drawback that every LaTeX IDE is subject to.

Bonus: Key bindings for bépo users

As a bépo user, I have some remapping done in Neovim, and especially direction keys:

-- Some shortcuts
local keymap = vim.keymap.set
local opts = {noremap = true, silent = true}

-- [HJKL] <-> {CTSR}
local map_list = {
  ['c'] = 'h', ['r'] = 'l', ['t'] = 'j', ['s'] = 'k', ['C'] = 'H', ['R'] = 'L', ['T'] = 'J', ['S'] = 'K', -- [HJKL] -> [CTSR]
  ['j'] = 't', ['J'] = 'T', ['l'] = 'c', ['L'] = 'C', ['h'] = 'r', ['H'] = 'R', ['k'] = 's', ['K'] = 'S', -- [CTSR] -> [HJKL]: J = until, L = change, h = replace, k = substitute
}
for key, binding in pairs(map_list) do
  keymap({'n', 'x'}, key, binding, opts)
end

That's nice and all but… it conflicts with the vimtex default mappings such as cse to rename an environment which can be useful to replace an align with align* for instance. Meaning that going back one char would trigger vim to wait for the next input, which is kind of annoying.

Hence the need to remap the vimtex default shortcuts starting with c, t, s or r. Fortunately, it's only the case for c and t. I first just add the remapping to $NVIMDIR/after/ftplugin/tex.lua, however I soon noticed that it's not sufficient as vimtex is also used for .tikz, .cls and .bib files,² thus we will use autocommand for that:

-- Some BÉPO mappings for vimtex
vim.api.nvim_create_autocmd({"BufEnter", "BufWinEnter"}, {
  pattern = {"*.tex", "*.bib", "*.cls", "*.tikz",},
  group = vim.api.nvim_create_augroup("latex", { clear = true }),
  callback = function()
    local vimtex_remaps = {
      -- c <-> t
      { mode = "n", source = "csd", target = "tsd", command = "<Plug>(vimtex-delim-change-math)"},
      { mode = "n", source = "csc", target = "lsc", command = "<Plug>(vimtex-cmd-change)"},
      { mode = "n", source = "cse", target = "lse", command = "<Plug>(vimtex-env-change)"},
      { mode = "n", source = "cs$", target = "ls$", command = "<Plug>(vimtex-env-change-math))"},
      -- t <-> j
      { mode = {"x", "n"}, source = "tsD", target = "jsD", command = "<Plug>(vimtex-delim-toggle-modifier-reverse)"},
      { mode = {"x", "n"}, source = "tsd", target = "jsd", command = "<Plug>(vimtex-delim-toggle-modifier)"},
      { mode = {"x", "n"}, source = "tsf", target = "jsf", command = "<Plug>(vimtex-cmd-toggle-frac)"},
      { mode = "n", source = "tsc", target = "jsc", command = "<Plug>(vimtex-cmd-toggle-star)"},
      { mode = "n", source = "ts$", target = "js$", command = "<Plug>(vimtex-env-toggle-math)"},
      { mode = "n", source = "tse", target = "jse", command = "<Plug>(vimtex-env-toggle-star)"},
    }

    for _,remap in pairs(vimtex_remaps) do
      if vim.fn.maparg(remap.source) ~= "" then
        vim.keymap.del(remap.mode, remap.source, { buffer = true })
        vim.keymap.set(remap.mode, remap.target, remap.command, { silent = true, noremap = true, buffer = true})
      end
    end
  end,
})

The sanity check with maparg(·) is done to avoid unmapping a mapping that already doesn't exist, which will raise an error (as I have the (bad?) habit to type :e to reload the current file when thinking, that what triggered this behaviour in my case).

To finish and for the sake of completeness, here follows the bépo-bindings for zathura, to put in your zathurarc file:

## BEPO
# hjkl → ctsr
map t scroll down
map s scroll up
map c scroll left
map r scroll right

# JK → TS
map T navigate next
map S navigate previous

# r → p
map p rotate rotate-cw

# R → u
map u reload

# Mode Index
map [index] t navigate_index down
map [index] s navigate_index up
map [index] r navigate_index expand
map [index] c navigate_index collapse

map [index] R navigate_index expand-all
map [index] C navigate_index collapse-all

---

1. meaning that it is not owned by any terminal I have opened, I can otherwise still recover it somehow. ↩︎

2. Actually .cls and .tikz are detected as tex files, so the ftplugin approach works but .bib is detected as a bibtex file and enjoys its own filetype. ↩︎