Inconsistency between setlocal and vim.opt_local for signcolumn

[krady21]

TLDR

vim.opt_local.signcolumn seems to behave as window local
setlocal signcolumn seems to behave as buffer local

• nvim --version:

NVIM v0.5.0-dev
Build type: RelWithDebInfo
LuaJIT 2.1.0-beta3

• Operating system/version: Ubuntu 18.04.5 LTS
• Terminal name/version: GNOME Shell 3.28.4
• $TERM: xterm-256color

Steps to reproduce using nvim -u mini_init.lua

mkdir ~/.config/nvim/pack/test/start
git clone https://github.com/neovim/nvim-lspconfig.git ~/.config/nvim/pack/test/start/nvim-lspconfig

mini_init.lua:

local on_attach = function(client, bufnr)
     vim.opt_local.signcolumn = 'yes'
end
require'lspconfig'.rust_analyzer.setup{ on_attach = on_attach } -- or any lsp server for that matter

:e file.rs
:e Makefile

Actual behaviour

Neovim sets signcolumn to 'yes' in all the buffers that are opened afterwards in the window where the lsp attach function was first called, even if those buffers are not associated to any lsp server.

Expected behaviour

Set signcolumn to 'yes' only in buffers where lsp is used. Using vim.cmd('setlocal signcolumn=yes')in the on_attach function produces the desired behavior and makes the signcolumn setting buffer local (only applied to lsp attached buffers). This inconsistency probably comes from the fact the signcolumn is exposed as a window local option in lua, but I was hoping that adding opt_local in #13479 would fix it.

[mnarrell]

This behavior seems to apply to spell as well. I've done vim.opt_local.spell = true for the markdown filetype. After opening a markdown file, each subsequent buffer has spell checking enabled.

[tjdevries]

Hmmm, this is how window local options usually work. They are inherited from the window that you were in before, so when you split and do other things, I thought they should keep the same value.

For example, signcolumn is a window option, as mentioned in the help:

						*'signcolumn'* *'scl'*
'signcolumn' 'scl'	string	(default "auto")
			local to window
``

cc @bfredl am I crazy :)

[tjdevries]

Oops, accidentally tagged @bfredl in the code block 😆

[ii14]

@tjdevries If you do :setl scl=yes, then open a new buffer, scl on the new buffer is back to default value. It's basically as if it was local to buffer. See #14620

[tjdevries]

How are you opening a new buffer? Via a split or via edit or what? Is the original buffer a listed, non-scratch buffer?

[ii14]

Just :enew. Previous buffer has to be modified to open a new one of course.

[francisco285]

@tjdevries Try this:

1. Use vim.wo.signcolumn = 'yes' in the on_attach function
2. Open neovim
3. :e init.lua (assuming this file will use lsp)
4. It will set signcolumn to yes on the current window
5. Using :new the new window also have signcolumn set to yes

Using vim.cmd([[setlocal signcolumn=yes]]) in 1 results in 5 having signcolumn set to no (the default)

[ii14]

I don't really know what I'm talking about so this might make no sense, but could it be that vim docs are just misleading when they say that these options are local to window? Usually you don't want methods like signcolumn, foldmethod etc to be local to window. You want them to be local to buffer.

[tjdevries]

OK, Yeah, this option is confusing me 😆 I'll have a chat w/ bfredl about it. Perhaps it's an issue with nvim_win_set_option (which is what vim.opt, and all the other ones are using under the hood).

It is a window option. That is how it's defined in the code. I don't know if there is special casing for this or if other things come into play. I'll have to look into that a bit more.

[bfredl]

The API is simply just incomplete for window options. Extending it properly will be a goal for 0.6.

[Shatur]

Also noticed the same issue when transferred my ftplugin settings that use setlocal to lua. vim.wo also doesn't behave like setlocal.

[zeertzjq]

This issue also seems to exist for conceallevel.

[andrewferrier]

This issue also seems to exist for number and relativenumber. Using vim.opt_local.number = false turns off number for all buffers opened in a window and is not buffer-local (which is not desirable, at least to me!). setlocal nonumber in a .vim file has the desired effect.

[andrewferrier]

Curiously, number is local to window according to the documentation. So maybe vim's original implementation of setlocal is wrong, as it definitely appears to be local to a buffer (which makes a lot more sense to me).

[TheBlob42]

Since I just stumbled across this issue as well I did a bit of research and found :h local-options

[...]
It's possible to set a local window option specifically for a type of buffer.
When you edit another buffer in the same window, you don't want to keep
using these local window options.  Therefore Vim keeps a global value of the
local window options, which is used when editing another buffer.  Each window
has its own copy of these values.  Thus these are local to the window, but
global to all buffers in the window.  With this you can do: >
	:e one
	:set list
	:e two
Now the 'list' option will also be set in "two", since with the ":set list"
command you have also set the global value. >
	:set nolist
	:e one
	:setlocal list
	:e two
Now the 'list' option is not set, because ":set nolist" resets the global
value, ":setlocal list" only changes the local value and ":e two" gets the
global value.  Note that if you do this next: >
	:e one
You will get back the 'list' value as it was the last time you edited "one".
The options local to a window are remembered for each buffer.  This also
happens when the buffer is not loaded, but they are lost when the buffer is
wiped out |:bwipe|.

I guess this is the part which is not correctly reflected by vim.opt_local and would explain the difference regarding setlocal

[lewis6991]

This issue also seems to exist for number and relativenumber. Using vim.opt_local.number = false turns off number for all buffers opened in a window and is not buffer-local (which is not desirable, at least to me!). setlocal nonumber in a .vim file has the desired effect.

This doesn't seem to be true (anymore). vim.opt_local.number = false only disables number for the current buffer. I believe #5996 fixed this.

In fact, I think that PR addresses this entire issue. Can someone confirm?

[ii14]

@lewis6991 Yes, I believe this was fixed with the introduction of nvim_set_option_value.

[lewis6991]

Ok, will close.

Note, this issue still exists for vim.wo and nvim_win_set_option, but this is by design (#9110).

[andrewferrier]

Agreed with @lewis6991, not sure when this changed, but at least since 0.7, this seems to be working as expected and vim.opt_local.number = false (as an example) works as I expected - buffer-local, not window-local.