Activate specific source using a prefix character, thus remove the prefix from the inserted text

[DrKGD]

I'd like to display snippets only by prefixing them using ";", so far i've found someone who already dealt with that

snippets = {
	name = "SNP",
	module = "blink.cmp.sources.snippets",
	score_offset = 95,

	-- THANKS: https://github.com/linkarzu/dotfiles-latest/blob/main/neovim/neobean/lua/plugins/blink-cmp.lua
	should_show_items = function()
		local col = vim.api.nvim_win_get_cursor(0)[2]
		local before_cursor = vim.api.nvim_get_current_line():sub(1, col)
		return before_cursor:match(";%w*$") ~= nil
	end,
}

He also provides a way to "delete" the special prefix character soon after accepting the snippet, but I feel like it is more of a hack and could easily break.

I tried experimenting on my own, but I don't think it is possibile to hijack the item itself to drop the prefix character, all I managed to do is prefixing the displayed label with the character itself

transform_items = function(a, items)
	return vim.iter(items)
		:map(function(item)
			item.label = (";%s"):format(item.label)
			return item
		end)
		:totable()
end

Note that I am not using luasnip as of now (as it is duplicating results) but the builtin snippet source, having friendly-snippets installed. Thanks!

[LamprosPitsillos]

I am also interested in this , any news on this?

[soifou]

Seems related to #1340

[oedotme]

@DrKGD @LamprosPitsillos I got triggering snippets by a semicolon ; working with a combination of should_show_items + get_trigger_characters for the initial trigger, resolve appends the ; to the current completion item - so that the ; is considered part of the snippet prefix and it's replaced automatically once selected/accepted:

opts = {
  sources = {
    providers = {
      snippy = {
        name = 'snippy',
        module = 'utils.snippy',
        min_keyword_length = 1,
        max_items = 8,
        should_show_items = function(ctx) return ctx.trigger.initial_character == ';' end,
        override = {
          get_trigger_characters = function(_) return { ';' } end,
          resolve = function(_, item) item.insertText = ';' .. item.insertText end,
        },
      },
      buffer = {
        min_keyword_length = 2,
        max_items = 8,
        should_show_items = function(ctx) return ctx.trigger.initial_character ~= ';' end,
      },
    },
  },
}

[DrKGD]

Mh, also tried that, no luck, here's a summary of what I've observed

1. If I were to configure a resolve method, the given prefix is kept in the buffer
2. If I were to configure transform_items it also prepends a second ;, I suppose the method is called more than once?
3. Not sure if the source is the culprit here, i've tried disabling luasnip and use the default engine only

snippets = {
	preset = 'default',
},

-- ...

providers = {
	snippets = {
		name = "SNP",
	module = "blink.cmp.sources.snippets",
	score_offset = 100,
	min_keyword_length = 1,
	max_items = 10,

        -- I want to be absolutely sure I don't load friendly snippets by mistake
	opts = {
		friendly_snippets = false
	},

	should_show_items = function(ctx)
		return ctx.trigger.initial_character == ';'
	end,

	override = {
		resolve = function(_, item)
			item.insertText = ';' .. item.insertText
		end,

		get_trigger_characters = function(_)
			return { ';' }
		end,
	},
},

Thus I've configured a test snippet only as suggested in the guide

In .config/nvim/snippets/package.json

{
  "name": "personal-snippets",
  "contributes": {
    "snippets": [
      { "language": "c", "path": "./c.json" }
    ]
  }
}

In .config/nvim/snippets/c.json

{
  "foo": {
    "prefix": "foo",
    "body": [
      "test"
    ]
  }
}

I am only getting that one snippet result, but this time the behavior is consistent, thus the snippet always has the ; prefix

4. If I were to update the test snippet by including the prefix, it works just fine

{
  "foo": {
    "prefix": ";foo",
    "body": [
      "test"
    ]
  }
}

I am on version v1.4.1 if that matters

[oedotme]

If I were to configure transform_items it also prepends a second ;, I suppose the method is called more than once?

That's strange, did you keep override.resolve around when you tried transform_items?

[oedotme]

Nevermind, I could reproduce what you described.

[DrKGD]

If I really have to guess, the problem lies somewhere around here...

https://github.com/Saghen/blink.cmp/blob/586ee87534f5bf65f1c8dea2d1da2a57e8cddd36/lua/blink/cmp/sources/snippets/luasnip.lua#L190-L202

By adding a couple of prints right before line 190...

print('item', vim.inspect(item))
print('region_before', vim.inspect(clear_region))
print('params', vim.inspect(expand_params))
print('expand_params', vim.inspect(expand_params))

And this one after line 202...

print('region_before', vim.inspect(clear_region))

And if I were to type out the entire suggestion ";if" (notice the presence of exact = true)

item {
  cursor_column = 1,
  data = {
    show_condition = <function 1>,
    snip_id = 18
  },
  exact = true,
  insertText = ";if",
  insertTextFormat = 1,
  kind = 15,
  label = "if",
  score = 153,
  score_offset = 100,
  sortText = "00000000|||if",
  source_id = "snippets",
  source_name = "SNP",
  textEdit = {
    newText = ";if",
    range = {
      ["end"] = {
        character = 3,
        line = 7
      },
      start = {
        character = 0,
        line = 7
      }
    }
  }
}

region_before {
  from = { 7, 0 },
  to = { 7, 3 }
}

params {
  captures = {},
  trigger = "if"
}

expand {
  captures = {},
  trigger = "if"
}

region_after {
  from = { 7, 1 },
  to = { 7, 3 }
}

The expand.trigger is populated, thus this section updates the clear_region, which range is now wrong

    elseif expand_params.trigger ~= nil then
      clear_region = {
        from = { cursor[1], cursor[2] - #expand_params.trigger },
        to = cursor,
      }
    end

But whenever I have a partial match (again, exact = false this time)

item {
  cursor_column = 1,
  data = {
    show_condition = <function 1>,
    snip_id = 30
  },
  exact = false,
  insertText = ";if",
  insertTextFormat = 1,
  kind = 15,
  label = "if",
  score = 129,
  score_offset = 100,
  sortText = "00000000|||if",
  source_id = "snippets",
  source_name = "SNP",
  textEdit = {
    newText = ";if",
    range = {
      ["end"] = {
        character = 2,
        line = 7
      },
      start = {
        character = 0,
        line = 7
      }
    }
  }
}
region_before {
  from = { 7, 0 },
  to = { 7, 2 }
}
params nil
expand nil
region_after {
  from = { 7, 0 },
  to = { 7, 2 }
}

The clear_region is not getting updated, thus the insertText is adeguately being replaced

I don't really know what really was the intent here and whether or not commenting that code would break something else in the future 🤔

[tgharib]

I ran into the same issue as @DrKGD. I ended up abandoning this approach. I disabled snippets entirely in blink.cmp and now use the expand hotkey in mini.snippets instead. It has a dedicated Ctrl+J keybind to trigger a snippet.

[DrKGD]

Good news everyone, I've found a somewhat reliable solution with luasnip (at least, it seems reliable with my limited testing)

Basically it consists of monkeypatching the luasnip expand function such that it behaves the way that blink.cmp intended it to behave, let's get straight to the point

1. Key in the get_trigger_characters and should_show_items as suggest by @oedotme, discard the resolve function as it won't be required.

{ 'snippets', true, {
			-- This is not needed, kept it in just for the demo
			-- name = "SNP",
			score_offset = 100,
			module = "blink.cmp.sources.snippets",
			min_keyword_length = 0,
			max_items = 10,

			override = {
				get_trigger_characters = function(_)
					return { ';' }
				end,
			},

			should_show_items = function(ctx)
				return ctx.trigger.initial_character == ';'
			end,
		}},

2. Apply the monkey patch soon after the luasnip plugin has been loaded, basically we just need to force luasnip.snip_expand to use our version of the function instead, which consists of expanding the region to the selected prefix.

   I don't think package.preloaded is needed, but it basically prevents luasnip from ever being unloaded and always return the version we hijacked.

   (You don't need to copy the whole configuration, just the patch itself)

-- Additional snippet engine
return {
	'L3MON4D3/LuaSnip',
		name = 'blink-cmp.dep.snip',
		build = "make install_jsregexp",
		dependencies = { 'blink-cmp.dep.friendly-snippets' },
		version = "v2.*",
		config = function()
			-- From runtime
			require('luasnip.loaders.from_vscode').lazy_load { }

			-- From scissors
			require("luasnip.loaders.from_vscode").lazy_load {
				paths = { config.dirs.scissors },
			}

			-- HIJACK: Hijack expand function to work with blink_cmp
			-- https://github.com/Saghen/blink.cmp/discussions/1104
			local ls					= require('luasnip')
			local expand_orig = ls.snip_expand
			ls.snip_expand		= function(snip, opts)
				if opts and opts.clear_region then
					opts.clear_region.from[2] = opts.clear_region.from[2] - (';'):len() end
				return expand_orig(snip, opts)
			end

			-- Assert that if luasnip were to be unloaded
			--  it'd load our patched version instead
			package.preload['luasnip'] = function() return ls end
		end
}

Let me know if that works out for you as well!

AhVj7iUhIzpLl6CC.mp4