neovim-remote

This package provides an executable called nvr which solves these cases:

• Controlling nvim processes from the shell. E.g. opening files in another terminal window.
• Opening files from within :terminal without starting a nested nvim process.

---

• Installation
• Theory
• First steps
• Typical use cases
• Demos
• FAQ

---

Installation

pip3 install neovim-remote

If you encounter any issues, e.g. permission denied errors or you can't find the nvr executable, read INSTALLATION.md.

Theory

Nvim always starts a server. Get its address via :echo $NVIM_LISTEN_ADDRESS or :echo v:servername. Or specify an address at startup: NVIM_LISTEN_ADDRESS=/tmp/nvimsocket nvim.

nvr will use $NVIM_LISTEN_ADDRESS or any address given to it via --servername.

If the targeted address does not exist, nvr starts a new process by running "nvim". You can change the command by setting $NVR_CMD. (This requires forking, so it won't work on Windows.)

First steps

Start a nvim process (which acts as a server) in one shell:

NVIM_LISTEN_ADDRESS=/tmp/nvimsocket nvim

And do this in another shell:

# nvr uses /tmp/nvimsocket by default, so we're good.

# Open two files:
nvr --remote file1 file2

# Send keys to the current buffer:
nvr --remote-send 'iabc<esc>'
# Enter insert mode, insert 'abc', and go back to normal mode again.

# Evaluate any VimL expression, e.g. get the current buffer:
nvr --remote-expr 'bufname("")'
README.md

click here to see all nvr options

$ nvr -h
usage: nvr [arguments]

Remote control Neovim processes.

If no process is found, a new one will be started.

    $ nvr --remote-send 'iabc<cr><esc>'
    $ nvr --remote-expr 'map([1,2,3], "v:val + 1")'

Any arguments not consumed by options will be fed to --remote-silent:

    $ nvr --remote-silent file1 file2
    $ nvr file1 file2

All --remote options take optional commands.
Exception: --remote-expr, --remote-send.

    $ nvr +10 file
    $ nvr +'echomsg "foo" | echomsg "bar"' file
    $ nvr --remote-tab-wait +'set bufhidden=delete' file

Open files in a new window from a terminal buffer:

    $ nvr -cc split file1 file2

Use nvr from git to edit commit messages:

    $ git config --global core.editor 'nvr --remote-wait-silent'

optional arguments:
  -h, --help            show this help message and exit
  --remote [<file> [<file> ...]]
                        Use :edit to open files. If no process is found, throw
                        an error and start a new one.
  --remote-wait [<file> [<file> ...]]
                        Like --remote, but block until all buffers opened by
                        this option get deleted or the process exits.
  --remote-silent [<file> [<file> ...]]
                        Like --remote, but throw no error if no process is
                        found.
  --remote-wait-silent [<file> [<file> ...]]
                        Combines --remote-wait and --remote-silent.
  --remote-tab [<file> [<file> ...]]
                        Like --remote, but use :tabedit.
  --remote-tab-wait [<file> [<file> ...]]
                        Like --remote-wait, but use :tabedit.
  --remote-tab-silent [<file> [<file> ...]]
                        Like --remote-silent, but use :tabedit.
  --remote-tab-wait-silent [<file> [<file> ...]]
                        Like --remote-wait-silent, but use :tabedit.
  --remote-send <keys>  Send key presses.
  --remote-expr <expr>  Evaluate expression and print result in shell.
  --servername <addr>   Set the address to be used. This overrides the default
                        "/tmp/nvimsocket" and $NVIM_LISTEN_ADDRESS.
  --serverlist          Print the TCPv4 and Unix domain socket addresses of
                        all nvim processes.
  -cc <cmd>             Execute a command before every other option.
  -c <cmd>              Execute a command after every other option.
  -d                    Diff mode. Use :diffthis on all to be opened buffers.
  -l                    Change to previous window via ":wincmd p".
  -o <file> [<file> ...]
                        Open files via ":split".
  -O <file> [<file> ...]
                        Open files via ":vsplit".
  -p <file> [<file> ...]
                        Open files via ":tabedit".
  -q <errorfile>        Read errorfile into quickfix list and display first
                        error.
  -s                    Silence "no server found" message.
  -t <tag>              Jump to file and position of given tag.
  --nostart             If no process is found, do not start a new one.
  --version             Show the nvr version.

Development: https://github.com/mhinz/neovim-remote

Happy hacking!

Typical use cases

• Use nvr as git mergetool.

  If you want to use nvr for git difftool and git mergetool, put this in your gitconfig:

  [diff]
      tool = nvr
  [difftool "nvr"]
      cmd = nvr -s -d $LOCAL $REMOTE
  [merge]
      tool = nvr
  [mergetool "nvr"]
      cmd = nvr -s -d $LOCAL $BASE $REMOTE $MERGED -c 'wincmd J | wincmd ='
  

  nvr -d is a shortcut for nvr -d -O and acts like vim -d, thus it uses :vsplit to open the buffers. If you want them to be opened via :split instead, use nvr -d -o.

  When used as mergetool and all four buffers got opened, the cursor is in the window containing the $MERGED buffer. We move it to the bottom via :wincmd J and then equalize the size of all windows via :wincmd =.

• Open files from within :terminal without starting a nested nvim.

  Easy-peasy! Just nvr file.

  This works without any prior setup, because $NVIM_LISTEN_ADDRESS is always set within Nvim. And nvr will default to that address.

  I often work with two windows next to each other. If one contains the terminal, I can use nvr -l foo to open the file in the other window.

• Open files always in the same nvim no matter which terminal you're in.

  If you just run nvr -s, a new nvim process will start and set its address to /tmp/nvimsocket automatically.

  Now, no matter in which terminal you are, nvr file will always work on that nvim process. That is akin to emacsclient from Emacs.

• Use nvr as temporary editor.

  Imagine Neovim is set as your default editor: VISUAL=nvim.

  Now run git commit. In a regular shell, a new nvim process starts. That's exactly what you want.

  But in a terminal buffer (:terminal), a new nvim process starts as well. Now you have one nvim nested within another. You don't want that. Put this in your vimrc:

  if has('nvim')
    let $VISUAL = 'nvr -cc split --remote-wait'
  endif
  

  That way, you get a new window for entering the commit message instead of a nested nvim process.

  Alternatively, you can make git always using nvr. In a regular shell, nvr will create a new nvim process. In a terminal buffer, nvr will open a new buffer (use :w | bd to commit).

    $ git config --global core.editor 'nvr --remote-wait-silent'
  

• Use nvr in plugins.

  Some plugins rely on the --remote family of options from Vim. Nvim had to remove those when they switched to outsource a lot of manual code to libuv. These options are planned to be added back, though.

  In these cases nvr can be used as a drop-in replacement. E.g. vimtex can be configured to use nvr to jump to a certain file and line: read.

Demos

(Click on the GIFs to watch them full-size.)

Using nvr from another shell:

Using nvr from within :terminal:

FAQ

• How to open directories?

  :e /tmp opens a directory view via netrw. Netrw works by hooking into certain events, BufEnter in this case (see :au FileExplorer for all of them).

  Unfortunately Neovim's API doesn't trigger any autocmds on its own, so simply nvr /tmp won't work. Meanwhile you can work around it like this:

    $ nvr /tmp -c 'doautocmd BufEnter'
  

• Reading from stdin?

  Yes! E.g. echo "foo\nbar" | nvr -o - and cat file | nvr --remote - work just as you would expect them to work.

• Exit code?

  If you use a recent enough Neovim, nvr will use the same exit code as the linked nvim.

  E.g. nvr --remote-wait <file> and then :cquit in the linked nvim will make nvr return with 1.

• How to send a message to all waiting clients?

  If you open a buffer with any of the wait options, that buffer will get a variable b:nvr. The variable contains a list of channels wheres each channel is a waiting nvr client.

  Currently nvr only understands the Exit message. You could use it to disconnect all waiting nvr clients at once:

  command! DisconnectClients
      \  if exists('b:nvr')
      \|   for client in b:nvr
      \|     silent! call rpcnotify(client, 'Exit', 1)
      \|   endfor
      \| endif