fix: gitsigns not automatically attached due to async not returning

[yamgent]

Attempt to fix #903.

What's Going On?

This bug is triggered when opening neovim inside a directory, with both neo-tree and this plugin installed.

Opening the directory causes neovim to create a netrw buffer.

When the setup() function of this plugin is called, the plugin will eventually reach this particular part of the code:

gitsigns.nvim/lua/gitsigns.lua

Lines 122 to 128 in 37d26d7

	-- Attach to all open buffers
	for _, buf in ipairs(api.nvim_list_bufs()) do
	if api.nvim_buf_is_loaded(buf) and api.nvim_buf_get_name(buf) ~= '' then
	M.attach(buf, nil, 'setup')
	scheduler()
	end
	end

At this point in time, the netrw buffer exists, so M.attach() will be called with that buffer. M.attach() will eventually give up control here with async.scheduler_if_buf_valid():

gitsigns.nvim/lua/gitsigns/attach.lua

Lines 296 to 299 in 37d26d7

	if not git_obj and not ctx then
	git_obj = try_worktrees(cbuf, file, encoding)
	async.scheduler_if_buf_valid(cbuf)
	end

After giving up control, the neotree plugin will run the "hijack netrw" logic (NB: hijacking is the default behaviour of neotree and does not need to be configured manually.) The hijack logic can be found here:

https://github.com/nvim-neo-tree/neo-tree.nvim/blob/f86e871584bd3c5a00b4ff8344305889eb52ebff/lua/neo-tree/setup/netrw.lua#L53-L101

As part of the hijack, it deletes the netrw buffer:

    local remove_dir_buf = vim.schedule_wrap(function()
      log.trace("Deleting buffer in netrw hijack", dir_bufnr)
      pcall(vim.api.nvim_buf_delete, dir_bufnr, { force = true })
    end)

After the hijack is done, the gitsigns plugin regains control here:

gitsigns.nvim/lua/gitsigns/async.lua

Lines 183 to 189 in 37d26d7

	M.scheduler_if_buf_valid = M.wrap(function(buf, cb)
	vim.schedule(function()
	if not buf or vim.api.nvim_buf_is_valid(buf) then
	cb()
	end
	end)
	end, 2)

Unfortunately, buf is not nil (it is the ID of the now-deleted netrw buffer), and vim.api.nvim_buf_is_valid(buf) returns false, so cb() is never executed, and so we never continue executing.

This is problematic, because the code after the for loop that contains M.attach() is the one responsible for adding the attach autocmd, but this adding is never executed:

  for _, buf in ipairs(api.nvim_list_bufs()) do
    if api.nvim_buf_is_loaded(buf) and api.nvim_buf_get_name(buf) ~= '' then
      M.attach(buf, nil, 'setup')  ----- never returns
      scheduler()
    end
  end

  api.nvim_create_autocmd({ 'BufRead', 'BufNewFile', 'BufWritePost' }, {   ----- does not get executed as a result :(
    group = 'gitsigns',
    callback = function(data)
      M.attach(nil, nil, data.event)
    end,
  })

Solution

Moved the autocmd to before the for loop, instead of after the for loop.

This solution feels like a temporary band-aid, but it does fix the problem.

[lewis6991]

I think this should fix the issue too. Thanks!