Skip to content

Latest commit

 

History

History
1178 lines (968 loc) · 42.8 KB

DOC.md

File metadata and controls

1178 lines (968 loc) · 42.8 KB
            __                       ____
           /\ \                     /\  _`\           __
           \ \ \      __  __     __ \ \,\L\_\    ___ /\_\  _____
            \ \ \  __/\ \/\ \  /'__`\\/_\__ \  /' _ `\/\ \/\ '__`\
             \ \ \L\ \ \ \_\ \/\ \L\.\_/\ \L\ \/\ \/\ \ \ \ \ \L\ \
              \ \____/\ \____/\ \__/.\_\ `\____\ \_\ \_\ \_\ \ ,__/
               \/___/  \/___/  \/__/\/_/\/_____/\/_/\/_/\/_/\ \ \/
                                                             \ \_\
                                                              \/_/

Luasnip is a snippet-engine written entirely in lua. It has some great features like inserting text (luasnip-function-node) or nodes (luasnip-dynamic-node) based on user input, parsing LSP syntax and switching nodes (luasnip-choice-node). For basic setup like mappings and installing, check the README.

All code-snippets in this help assume that

local ls = require"luasnip"
local s = ls.snippet
local sn = ls.snippet_node
local isn = ls.indent_snippet_node
local t = ls.text_node
local i = ls.insert_node
local f = ls.function_node
local c = ls.choice_node
local d = ls.dynamic_node
local r = ls.restore_node
local events = require("luasnip.util.events")
local ai = require("luasnip.nodes.absolute_indexer")

BASICS

In LuaSnip, snippets are made up of nodes. These can contain either

  • static text (textNode)
  • text that can be edited (insertNode)
  • text that can be generated from the contents of other nodes (functionNode)
  • other nodes
    • choiceNode: allows choosing between two nodes (which might contain more nodes)
    • restoreNode: store and restore input to nodes
  • or nodes that can be generated based on input (dynamicNode).

Snippets are always created using the s(trigger:string, nodes:table)-function. It is explained in more detail in SNIPPETS, but the gist is that it creates a snippet that contains the nodes specified in nodes, which will be inserted into a buffer if the text before the cursor matches trigger when expand is called.
The snippets for a given filetype have to be appended to the ls.snippets.<filetype>-table. Snippets that should be accessible globally (in all filetypes) will be read from the ls.snippets.all-table:

ls.snippets = {
	all = {
		s("ternary", {
			-- equivalent to "${1:cond} ? ${2:then} : ${3:else}"
			i(1, "cond"), t(" ? "), i(2, "then"), t(" : "), i(3, "else")
		})
	}
}

It is possible to make snippets from one filetype available to another using ls.filetype_extend, more info on that here.

SNIPPETS

The most direct way to define snippets is s:

s({trig="trigger"}, {})

(This snippet is useless beyond being a minimal example)

s accepts, as the first argument, a table with the following possible entries:

  • trig: string, plain text by default. The only entry that must be given.
  • name: string, can be used by eg. nvim-compe to identify the snippet.
  • dscr: string, description of the snippet, \n-separated or table for multiple lines.
  • wordTrig: boolean, if true, the snippet is only expanded if the word ([%w_]+) before the cursor matches the trigger entirely. True by default.
  • regTrig: boolean, whether the trigger should be interpreted as a lua pattern. False by default.
  • docstring: string, textual representation of the snippet, specified like dscr. Overrides docstrings loaded from json.
  • docTrig: string, for snippets triggered using a lua pattern: define the trigger that is used during docstring-generation.
  • hidden: hint for completion-engines, if set, the snippet should not show up when querying snippets.

s can also be a single string, in which case it is used instead of trig, all other values being defaulted:

s("trigger", {})

The second argument to s is a table containing all nodes that belong to the snippet. If the table only has a single node, it can be passed directly without wrapping it in a table.

The third argument is a table with the following valid keys:

  • condition: the condition-function fn(line_to_cursor, matched_trigger, captures) -> bool.
    The snippet will be expanded only if it returns true (default is a function that just returns true).
    The function is called before the text is modified in any way.
    Some parameters are passed to the function: The line up to the cursor, the matched trigger and the captures (table).
  • show_condition: Function with signature f(line_to_cursor) -> bool.
    It is a hint for completion-engines, indicating when the snippet should be included in current completion candidates.
    Defaults to a function returning true.
    This is different than condition because condition is evaluated by LuaSnip on snippet expansion (and thus has access to the matched trigger and captures), while show_condition is evaluated by the completion-engine when scanning for available snippet candidates.
  • callbacks: Contains functions that are called upon enterin/leaving a node of this snippet.
    To print text upon entering the second node of a snippet, callbacks should be set as follows: 
    {
	[2] = {
		[events.enter] = function(node) print("2!") end
	}
}
    To register a callback for the snippets' own events, the key [-1] may be used.
    The callbacks are passed only one argument, the node that triggered it. This opts-table can also be passed to eg. snippetNode or indentSnippetNode, but only callbacks is used there.

Snippets contain some interesting tables, eg. snippet.env contains variables used in the LSP-protocol like TM_CURRENT_LINE or TM_FILENAME or snippet.captures, where capture-groups of regex-triggers are stored. Additionally, the string that was used to trigger the snippet is stored in snippet.trigger. These variables/tables are primarily useful in dynamic/functionNodes, where the snippet can be accessed through the immediate parent (parent.snippet), which is passed to the function.

TEXTNODE

The most simple kind of node; just text.

s("trigger", { t("Wow! Text!") })

This snippet expands to

    Wow! Text!⎵

Where ⎵ is the cursor.
Multiline-strings can be defined by passing a table of lines rather than a string:

s("trigger", {
	t({"Wow! Text!", "And another line."})
})

INSERTNODE

These Nodes contain editable text and can be jumped to- and from (eg. traditional placeholders, like $1 in textmate-snippets).

The functionality is best demonstrated with an example:

s("trigger", {
	t({"After expanding, the cursor is here ->"}), i(1),
	t({"", "After jumping forward once, cursor is here ->"}), i(2),
	t({"", "After jumping once more, the snippet is exited there ->"}), i(0),
})

The InsertNodes are jumped over in order from 1 to n.
0-th node is special as it's always the last one.
So the order of InsertNode jump is as follows:

  • After expansion, we will be at InsertNode 1.
  • After jumping forward, we will be at InsertNode 2.
  • After jumping forward again, we will be at InsertNode 0.

If no 0-th InsertNode is found in a snippet, one is automatically inserted after all other nodes.

The jumping-order doesn't have to follow the "textual" order of the nodes:

s("trigger", {
	t({"After jumping forward once, cursor is here ->"}), i(2),
	t({"", "After expanding, the cursor is here ->"}), i(1),
	t({"", "After jumping once more, the snippet is exited there ->"}), i(0),
})

The above snippet will behave as follows:

  • After expansion, we will be at InsertNode 1.
  • After jumping forward, we will be at InsertNode 2.
  • After jumping forward again, we will be at InsertNode 0.

An important (because here luasnip differs from other snippet-engines) detail is that the jump-order restarts at 1 in nested snippets:

s("trigger", {
	i(1, "First jump"),
	t(" :: "),
	sn(2, {
		i(1, "Second jump"),
		t" : ",
		i(2, "Third jump")
	})
})

as opposed to eg. the textmate-syntax, where tabstops are snippet-global:

${1:First jump} :: ${2: ${3:Second jump} : ${4:Third jump}}

(this is not exactly the same snippet of course, but as close as possible)
(the restart-rule only applies when defining snippets in lua, the above textmate-snippet will expand correctly).

It's possible to have initial text inside an InsertNode, which is comfortable for potentially keeping some default-value:

	s("trigger", i(1, "This text is SELECTed after expanding the snippet."))

This initial text is defined the same way as textNodes, eg. can be multiline.

i(0)s can have initial text, but do note that when the SELECTed text is replaced, its' replacement won't end up in the i(0), but behind it (for reasons, check out Luasnip#110).

FUNCTIONNODE

Function Nodes insert text based on the content of other nodes using a user-defined function:

 s("trig", {
 	i(1),
 	f(function(args, snip, user_arg_1) return args[1][1] .. user_arg_1 end,
 		{1},
 		"Will be appended to text from i(0)"),
 	i(0)
 })

The first parameter of f is the function. Its parameters are:

  1. A table of the text of currently contained in the argnodes. (eg. {{line1}, {line1, line2}}). The snippet-indent will be removed from all lines following the first.

  2. The immediate parent of the functionNode. It is included here as it allows easy access to anything that could be useful in functionNodes (ie. parent.snippet.env or parent.snippet.captures, which contains capture groups of regex-triggered snippets). In most cases parent.env works, but if a functionNode is nested within a snippetNode, the immediate parent (a snippetNode) will contain neither captures nor env. Those are only stored in the snippet, which can be accessed as parent.snippet.

  3. Any parameters passed to f behind the second (included to more easily reuse functions, ie. ternary if based on text in an insertNode).

The second parameter is a table of indices of jumpable nodes whose text is passed to the function.
The table may be empty, in this case the function is evaluated once upon snippet-expansion.
If the table only has a single node, it can be passed directly without wrapping it in a table.
The indices can be specified either as relative to the functionNodes' parent using numbers or as absolute, using the absolute_indexer.

The function shall return a string, which will be inserted as-is, or a table of strings for multiline-string, here all lines following the first will be prefixed with the snippets' indentation.

Examples: Use captures from the regex-trigger using a functionNode:

s({trig = "b(%d)", regTrig = true},
	f(function(args, snip) return
		"Captured Text: " .. snip.captures[1] .. "." end, {})
)

The table passed to functionNode:

s("trig", {
	i(1, "text_of_first"),
	i(2, {"first_line_of_second", "second_line_of_second"}),
	-- order is 2,1, not 1,2!!
	f(function(args, snip)
		--here
	end, {2, 1} )})

At --here, args would look as follows (provided no text was changed after expansion):

args = {
	{"first_line_of_second", "second_line_of_second"},
	{"text_of_first"}
}

One more example to show usage of absolute_indexer:

s("trig", {
	i(1, "text_of_first"),
	i(2, {"first_line_of_second", "second_line_of_second"}),
	f(function(args, snip)
		-- just concat first lines of both.
		return args[1][1] .. args[2][1]
	end, {ai[2], ai[1]} )})

If the function only performs simple operations on text, consider using the lambda from luasnip.extras

CHOICENODE

ChoiceNodes allow choosing between multiple nodes.

 s("trig", c(1, {
 	t("Ugh boring, a text node"),
 	i(nil, "At least I can edit something now..."),
 	f(function(args) return "Still only counts as text!!" end, {})
 }))

c() expects as its first arg, as with any jumpable node, its position in the jumplist, and as its second a table with nodes, the choices. This table can either contain a single node or a table of nodes. In the latter case the table will be converted into a snippetNode. The third parameter is a table of options with the following keys:

  • restore_cursor: false by default. If it is set and the node that was being edited also appears in the switched-to choice (can be the case if a restoreNode is present in both choice) the cursor is restored relative to that node. The default is false as enabling might lead to worse performance. It's possible to override the default by wrapping the choiceNode-constructor in another function that sets opts.restore_cursor to true and then using that to construct choiceNodes: 
     local function restore_cursor_choice(pos, choices, opts)
 	if opts then
 		opts.restore_cursor = true
 	else
 		opts = {restore_cursor = true}
 	end
 	return c(pos, choices, opts)
 end

Jumpable nodes that normally expect an index as their first parameter don't need one inside a choiceNode; their index is the same as the choiceNodes'.

SNIPPETNODE

SnippetNodes directly insert their contents into the surrounding snippet. This is useful for choiceNodes, which only accept one child, or dynamicNodes, where nodes are created at runtime and inserted as a snippetNode.

Syntax is similar to snippets, however, where snippets require a table specifying when to expand, snippetNodes, similar to insertNodes, expect a number, as they too are jumpable:

 s("trig", sn(1, {
 	t("basically just text "),
 	i(1, "And an insertNode.")
 }))

Note that snippetNodes don't expect an i(0).

INDENTSNIPPETNODE

By default, all nodes are indented at least as deep as the trigger. With these nodes it's possible to override that behaviour:

s("isn", {
	isn(1, {
		t({"This is indented as deep as the trigger",
		"and this is at the beginning of the next line"})
	}, "")
})

(Note the empty string passed to isn).

Indent is only applied after linebreaks, so it's not possible to remove indent on the line where the snippet was triggered using ISN (That is possible via regex-triggers where the entire line before the trigger is matched).

Another nice usecase for ISN is inserting text, eg. // or some other comment- string before the nodes of the snippet:

s("isn2", {
	isn(1, t({"//This is", "A multiline", "comment"}), "$PARENT_INDENT//")
})

Here the // before This is is important, once again, because indent is only applied after linebreaks. To enable such usage, $PARENT_INDENT in the indentstring is replaced by the parents' indent (duh).

DYNAMICNODE

Very similar to functionNode, but returns a snippetNode instead of just text, which makes them very powerful as parts of the snippet can be changed based on user-input.

The prototype for the dynamicNodes' constructor is d(position:int, function, argnodes:table of nodes, user_args1, ..., user_argsn):

  1. position: just like all jumpable nodes, when this node will be jumped into.
  2. function: fn(args, parent, old_state, user_args1, ..., user_argsn) -> snippetNode
    This function is called when the argnodes' text changes. It generates and returns (wrapped inside a snippetNode) the nodes that should be inserted at the dynamicNodes place.
    args, parent and user_args are also explained in functionNode
    • args: table of text ({{"node1line1", "node1line2"}, {"node2line1"}}) from nodes the dynamicNode depends on.
    • parent: the immediate parent of the dynamicNode).
    • old_state: a user-defined table. This table may contain anything, its intended usage is to preserve information from the previously generated snippetNode: If the dynamicNode depends on other nodes it may be reconstructed, which means all user input (text inserted in insertNodes, changed choices) to the previous dynamicNode is lost.
      The old_state table must be stored in snippetNode returned by the function (snippetNode.old_state).
      The second example below illustrates the usage of old_state.
    • user_args1, ..., user_argsn: passed through from dynamicNode-constructor.
  3. argnodes: Indices of nodes the dynamicNode depends on: if any of these trigger an update, the dynamicNodes' function will be executed and the result inserted at the dynamicNodes place.
    Can be a single index or a table of indices.
  4. user_args1, ..., user_argsn: Anything passed here will also be passed to the function (easy to reuse similar functions with small changes) (This may be removed, alternatively use partial functions to reuse functions).

Examples:

s("trig", {
	t"text: ", i(1), t{"", "copy: "},
	d(2, function(args)
			-- the returned snippetNode doesn't need a position; it's inserted
			-- "inside" the dynamicNode.
			return sn(nil, {
				-- jump-indices are local to each snippetNode, so restart at 1.
				i(1, args[1])
			})
		end,
	{1})
})

This dynamicNode inserts an insertNode which copies the text inside the first insertNode.

local function lines(args, parent, old_state, initial_text)
	local nodes = {}
	if not old_state then old_state = {} end

	-- count is nil for invalid input.
	local count = tonumber(args[1][1])
	-- Make sure there's a number in args[1].
	if count then
		for j=1, count do
			local iNode
			if old_state and old_state[j] then
				-- old_text is used internally to determine whether
				-- dependents should be updated. It is updated whenever the
				-- node is left, but remains valid when the node is no
				-- longer 'rendered', whereas node:get_text() grabs the text
				-- directly from the node.
				iNode = i(j, old_state[j].old_text)
			else
			  iNode = i(j, initial_text)
			end
			nodes[2*j-1] = iNode

			-- linebreak
			nodes[2*j] = t({"",""})
			-- Store insertNode in old_state, potentially overwriting older
			-- nodes.
			old_state[j] = iNode
		end
	else
		nodes[1] = t("Enter a number!")
	end

	local snip = sn(nil, nodes)
	snip.old_state = old_state
	return snip
end

...

s("trig", {
	i(1, "1"),
	-- pos, function, argnodes, user_arg1
	d(2, lines, {1}, "Sample Text")
})

This snippet would start out as "1\nSample Text" and, upon changing the 1 to eg. 3, it would change to "3\nSample Text\nSample Text\nSample Text". Text that was inserted into any of the dynamicNodes insertNodes is kept when changing to a bigger number.
(old_state is no longer the best way to preserve user-input across multiple recreations: the shortly-explained restoreNode is much more user-friendly)

RESTORENODE

This node can store and restore a snippetNode that was modified (changed choices, inserted text) by the user. It's usage is best demonstrated by an example:

s("paren_change", {
	c(1, {
		sn(nil, { t("("), r(1, "user_text"), t(")") }),
		sn(nil, { t("["), r(1, "user_text"), t("]") }),
		sn(nil, { t("{"), r(1, "user_text"), t("}") }),
	}),
}, {
	stored = {
		user_text = i(1, "default_text")
	}
})

Here the text entered into user_text is preserved upon changing choice.

The constructor for the restoreNode, r, takes (at most) three parameters:

  • pos, when to jump to this node.
  • key, the key that identifies which restoreNodes should share their content.
  • nodes, the contents of the restoreNode. Can either be a single node or a table of nodes (both of which will be wrapped inside a snippetNode, except if the single node already is a snippetNode). The content of a given key may be defined multiple times, but if the contents differ, it's undefined which will actually be used. If a keys content is defined in a dynamicNode, it will not be used for restoreNodes outside that dynamicNode. A way around this limitation is defining the content in the restoreNode outside the dynamicNode.

The content for a key may also be defined in the opts-parameter of the snippet-constructor, as seen in the example above. The stored-table accepts the same values as the nodes-parameter passed to r. If no content is defined for a key, it defaults to the empty insertNode.

An important-to-know limitation of restoreNode is that, for a given key, only one may be visible at a time. See this issue for details.

The restoreNode is also useful for storing user-input across updates of a dynamicNode. Consider this:

local function simple_restore(args, _)
	return sn(nil, {i(1, args[1]), i(2, "user_text")})
end

s("rest", {
	i(1, "preset"), t{"",""},
	d(2, simple_restore, 1)
}),

Every time the i(1) in the outer snippet is changed, the text inside the dynamicNode is reset to "user_text". This can be prevented by using a restoreNode:

local function simple_restore(args, _)
	return sn(nil, {i(1, args[1]), r(2, "dyn", i(nil, "user_text"))})
end

s("rest", {
	i(1, "preset"), t{"",""},
	d(2, simple_restore, 1)
}),

Now the entered text is stored.

RestoreNodes indent is not influenced by indentSnippetNodes right now. If that really bothers you feel free to open an issue.

ABSOLUTE_INDEXER

The absolute_indexer can be used to pass text of nodes to a function/dynamicNode that it doesn't share a parent with.
Normally, accessing the outer i(1) isn't possible from inside eg. a snippetNode (nested inside a choiceNode to make this example more practical):

s("trig", {
	i(1), c(2, {
		sn(nil, {
			t"cannot access the argnode :(", f(function(args) return args[1] end, {???})
		}),
		t"sample_text"
	})
})

Using absolute_indexer, it's possible to do so:

s("trig", {
	i(1), c(2, {
		sn(nil, {
			t"can access the argnode :)", f(function(args) return args[1] end, ai[1])
		}),
		t"sample_text"
	})
})

There are some quirks in addressing nodes:

s("trig", {
	i(2), -- ai[2]: indices based on insert-order, not position.
	sn(1, { -- ai[1]
		i(1), -- ai[1][1]
		t"lel", -- not addressable.
		i(2) -- ai[1][2]
	}),
	c(3, { -- ai[3]
		i(nil), -- ai[3][1]
		t"lel", -- ai[3][2]: choices are always addressable.
	}),
	d(4, function() -- ai[4]
		return sn(nil, { -- ai[4][0]
			i(1), -- ai[4][0][1]
		})
	end, {})
	}))
	r(5, "restore_key", -- ai[5]
		i(1) -- ai[5][0][1]: restoreNodes always store snippetNodes.
	)
	r(6, "restore_key_2", -- ai[6]
		sn(nil, { -- ai[6][0]
			i(1) -- ai[6][0][1]
		})
	)
	}))
})

Note specifically that the index of a dynamicNode differs from that of the generated snippetNode, and that restoreNodes (internally) always store a snippetNode, so even if the restoreNode only contains one node, that node has to be accessed as ai[restoreNodeIndx][0][1].

absolute_indexers' can be constructed in different ways:

ai[1][2][3] == ai(1, 2, 3) == ai{1, 2, 3}

EXTRAS

The module "luasnip.extras" contains nodes that ease writing snippets (This is only a short outline, their usage is shown more expansively in Examples/snippets.lua):

  • lambda: A shortcut for functionNodes that only do very basic string- manipulation. For example, to replace all occurences of "a" in the nth insert with "e", one could use lambda(lambda._1:gsub("a", "e"), n) (signature is similar to that of functionNode). If a node has multiple lines, they will be concatenated using "\n".

  • match: Can insert text based on a predicate (shorthand for functionNodes).
    The complete signature for the node is match(argnodes, condition, then, else), where

    • argnodes can be specified as in functionNode,

    • condition may be a

      • string: interpreted as a lua-pattern. Matched on the \n-joined (in case it's multiline) text of the first argnode (args[1]:match(condition)).
      • function: fn(args, snip) -> bool: takes the same parameters as the functionNode-function, any value other than nil or false is interpreted as a match.
      • lambda: l._n is the \n-joined text of the nth argnode. Useful if string-manipulations have to be performed before the string is matched.

    • then is inserted if the condition matches, else if it doesn't. They can both be either text, lambda or function (with the same parameters as specified above). If then is not given, the then-value depends on what was specified as the condition:

      • pattern: Simply the return value from the match, eg. the entire match, or, if there were capture groups, the first capture group.
      • function: the return value of the function if it is either a string or a table (if there is no then, the function cannot return a table containing something other than strings).
      • lambda: Simply the first value returned by the lambda.

    Examples:

    • match(n, "^ABC$", "A") inserts "A" if the nth jumpable node matches "ABC" exactly, nothing otherwise.
    • match(n, lambda._1:match(lambda._1:reverse()), "PALINDROME") inserts "PALINDROME" if the nth jumpable node is a palindrome.
    • s("trig", {
	i(1), t":",
	i(2), t"::",
	m({1, 2}, lambda._1:match("^"..lambda._2.."$"), lambda._1:gsub("a", "e"))
})
      This inserts the text of the first insertNode, with all occurences of a replaced with e if the second insertNode matches the first exactly.

  • rep: repeats the node with the passed index. rep(1) to repeat the content of the first insert.

  • partial: directly inserts the output of a function. Useful for eg. partial(os.date, "%Y") (arguments passed after the function are passed to it).

  • nonempty: inserts text if the insert at the given index doesn't contain any text. nonempty(n, "empty!", "not empty!") inserts "empty!" if insert n is empty, "not empty!" it it isn't.

  • dynamic_lambda: Operates almost exactly like lambda, only that it can be jumped to and it's contents therfore be easily overridden. dynamic_lambda(2, lambda._1..lambda._1, 1) will first contain the content of insert 1 appended to itself, but the second jump will lead to it, making it easy to override the generated text. The text will only be changed when a argnode updates it.

On The Fly snippets

You can create snippets that are not for being used all the time but only in a single session.

This behaves as an "operator" takes what is in a register and transforms it into a snippet using words prefixed as $ as inputs or copies (depending if the same word appears more than once). You can escape $ by repeating it.

In order to use add something like this to your config:

vnoremap <c-f>  "ec<cmd>lua require('luasnip.extras.otf').on_the_fly()<cr>
inoremap <c-f>  <cmd>lua require('luasnip.extras.otf').on_the_fly("e")<cr>

Notice that you can use your own mapping instead of and you can pick another register instead of "p. You can even use it several times, as if it where a macro if you add several mapppings like:

; For register a
vnoremap <c-f>a  "ac<cmd>lua require('luasnip.extras.otf').on_the_fly()<cr
inoremap <c-f>a  <cmd>lua require('luasnip.extras.otf').on_the_fly("a")<cr>


; For register b
vnoremap <c-f>a  "bc<cmd>:lua require('luasnip.extras.otf').on_the_fly()<cr
inoremap <c-f>b  <cmd>lua require('luasnip.extras.otf').on_the_fly("b")<cr>

LSP-SNIPPETS

Luasnip is capable of parsing lsp-style snippets using ls.parser.parse_snippet(context, snippet_string):

ls.parser.parse_snippet({trig = "lsp"}, "$1 is ${2|hard,easy,challenging|}")

Nested placeholders("${1:this is ${2:nested}}") will be turned into choiceNode's with: - the given snippet("this is ${1:nested}") and - an empty insertNode

VARIABLES

All TM_something-variables are supported with two additions: SELECT_RAW and SELECT_DEDENT. These were introduced because TM_SELECTED_TEXT is designed to be compatible with vscodes' behavior, which can be counterintuitive when the snippet can be expanded at places other than the point where selection started (or when doing transformations on selected text).

All variables can be used outside of lsp-parsed snippets as their values are stored in a snippets' snip.env-table:

s("selected_text", {
	-- the surrounding snippet is passed in args after all argnodes (none,
	-- in this case).
	f(function(args, snip) return snip.env.SELECT_RAW end, {})
})

To use any *SELECT* variable, the store_selection_keys must be set via require("luasnip").config.setup({store_selection_keys="<Tab>"}). In this case, hitting <Tab> while in Visualmode will populate the *SELECT*-vars for the next snippet and then clear them.

VSCODE SNIPPETS LOADER

As luasnip is capable of loading the same format of plugins as vscode, it also includes an easy way for loading those automatically. You just have to call:

 	require("luasnip.loaders.from_vscode").load(opts) -- opts can be ommited

Where opts is a table containing the keys: - paths: List of paths to load. Can be a table or a single, comma-separated string. If not set, runtimepath is used. The paths may begin with ~/ or ./ to indicate that the path is relative to your home or to the folder where your $MYVIMRC resides (useful to add your snippets). The directories passed this way must be structured like friendly-snippets eg. include a package.json. - exclude: List of languages to exclude, by default is empty. - include: List of languages to include, by default is not set.

The last two are useful mainly to avoid loading snippets from 3erd parties you don't wanna include.

Keep in mind that it will extend your snippets table, so do it after setting your snippets or you will have to extend the table as well.

Another way of using the loader is making it lazily

 	require("luasnip.loaders.from_vscode").lazy_load(opts) -- opts can be ommited

In this case opts only accepts paths (runtimepath if any). That will load the general snippets (the ones of filetype 'all') and those of the filetype of the buffers, you open every time you open a new one (but it won't reload them).

Apart from what is stipulated by the start each snippet in the json file can contain a "luasnip" field which is a table for extra parameters for the snippet, till now the only valid one is autotrigger.

After snippets were lazy-loaded, the User LuasnipSnippetsAdded-event will be triggered.

Note load vscode-style packages using require("luasnip.loaders.from_vscode").load(), if you've configured luasnip to detect the filetype based on the cursor position. Else the snippets won't be available to the from_cursor_pos function.

SNIPMATE SNIPPETS LOADER

As the snipmate snippet format is fundamentally the same as vscode, it can also be loaded.

 	require("luasnip.loaders.from_snipmate").load(opts) -- opts can be ommited

See from_vscode for an explanation of opts. If opts.paths is ommited, snippets are loaded from any directory named snippets located in the runtimepath.

Luasnip is compatible with honza/vim-snippets. Please use it as a reference for your directory structure.

When using honza/vim-snippets, the file with the global snippets is _.snippets, So we need to tell luasnip that _ also contains global snippets:

ls.filetype_extend("all", { "_" })

Something similar may have to be done for other snippet-repos as well.

Lazy loading is also available with the snipmate-loader.

 	require("luasnip.loaders.from_snipmate").lazy_load(opts) -- opts can be ommited

Here is a summary of the differences from the original snipmate format.

  • Only ./{ft}.snippets and ./{ft}/*.snippets will be loaded.
  • The file name or folder name will be used as file type.
  • You can use the comment and extends syntax.
  • ${VISUAL} will be replaced by $TM_SELECTED_TEXT to make the snippets compatible with luasnip
  • We do not implement eval using ` (backtick). This may be implemented in the future.

EXT_OPTS

ext_opts are probably best explained with a short example:

local types = require("luasnip.util.types")

vim.api.nvim_command("hi LuasnipChoiceNodePassive cterm=italic")
ls.config.setup({
	ext_opts = {
		[types.insertNode] = {
			passive = {
				hl_group = "GruvboxRed"
			}
		},
		[types.choiceNode] = {
			active = {
				virt_text = {{"choiceNode", "GruvboxOrange"}}
			}
		},
		[types.textNode] = {
			snippet_passive = {
				hl_group = "GruvboxGreen"
			}
		},
	},
	ext_base_prio = 200,
	ext_prio_increase = 3,
})

This highlights insertNodes red (both when active and passive) and adds virtualText and italics to choiceNode while it is active (both only if the snippet is active). textNodes are highlighted green, even if the snippet is no longer active. (unspecified values in passive are populated with values from snippet_passive, those in active with those from the new passive).

The active/ passive-tables are passed to nvim_buf_set_extmark as opts which means only entries valid there can be used here. priority, while still affecting the priority of highlighting, is interpreted as a relative value here, not absolute (0 <= priority < ext_prio_increase). The absolute range of priorities can still be somewhat controlled using ext_base_prio and ext_prio_increase (all highlights start out with ext_base_prio+their own priority, for highlights belonging to a nested snippet(Node), ext_base_prio is increased by ext_prio_increase)).

As a shortcut for setting hl_group, the highlight-groups Luasnip*Node{Active,Passive,SnippetPassive} may be defined (to be actually used by LuaSnip, ls.config.setup has to be called after defining). They are overridden by the values defined in ext_opts directly, but otherwise behave the same (active is extended by passive).

DOCSTRING

Snippet-docstrings can be queried using snippet:get_docstring(). The function evaluates the snippet as if it was expanded regularly, which can be problematic if eg. a dynamicNode in the snippet relies on inputs other than the argument-nodes. snip.env and snip.captures are populated with the names of the queried variable and the index of the capture respectively (snip.env.TM_SELECTED_TEXT -> '$TM_SELECTED_TEXT', snip.captures[1] -> '$CAPTURES1'). Although this leads to more expressive docstrings, it can cause errors in functions that eg. rely on a capture being a number:

s({trig = "(%d)", regTrig = true}, {
	f(function(args, snip)
		return string.rep("repeatme ", tonumber(snip.captures[1]))
	end, {})
}),

This snippet works fine because snippet.captures[1] is always a number. During docstring-generation, however, snippet.captures[1] is '$CAPTURES1', which will cause an error in the functionNode. Issues with snippet.captures can be prevented by specifying docTrig during snippet-definition:

s({trig = "(%d)", regTrig = true, docTrig = "3"}, {
	f(function(args, snip)
		return string.rep("repeatme ", tonumber(snip.captures[1]))
	end, {})
}),

snippet.captures and snippet.trigger will be populated as if actually triggered with 3.

Other issues will have to be handled manually by checking the contents of eg. snip.env or predefining the docstring for the snippet:

s({trig = "(%d)", regTrig = true, docstring = "repeatmerepeatmerepeatme"}, {
	f(function(args, snip)
		return string.rep("repeatme ", tonumber(snip.captures[1]))
	end, {})
}),

DOCSTRING-CACHE

Although generation of docstrings is pretty fast, it's preferable to not redo it as long as the snippets haven't changed. Using ls.store_snippet_docstrings(ls.snippets) and its counterpart ls.load_snippet_docstrings(ls.snippets), they may be serialized from or deserialized into the snippets. Both functions accept a table structured like ls.snippets, ie. {ft1={snippets}, ft2={snippets}}. load should be called before any of the loader-functions as snippets loaded from vscode-style packages already have their docstring set (docstrings wouldn't be overwritten, but there'd be unnecessary calls).

The cache is located at stdpath("cache")/luasnip/docstrings.json (probably ~/.cache/nvim/luasnip/docstrings.json).

EVENTS

Upon leaving/entering nodes or changing a choice an event is triggered: User Luasnip<Node>{Enter,Leave}, where <Node> is the name of a node in PascalCase, eg. InsertNode or DynamicNode or Snippet. The event triggered when changing the choice in a choiceNode is User LuasnipChangeChoice.

A pretty useless, beyond serving as an example here, application of these would be printing eg. the nodes' text after entering:

au User LuasnipInsertNodeEnter
	\lua print(require("luasnip").session.event_node:get_text()[1])

CLEANUP

The function ls.cleanup() triggers the LuasnipCleanup user-event, that you can listen to do some kind of cleaning in your own snippets, by default it will empty the snippets table and the caches of the lazy_load.

API-REFERENCE

require("luasnip"):

  • jumpable(direction): returns true if the current node has a next(direction = 1) or previous(direction = -1), eg. whether it's possible to jump forward or backward to another node.

  • jump(direction): returns true if the jump was successful.

  • expandable(): true if a snippet can be expanded at the current cursor position.

  • expand(): expands the snippet at(before) the cursor.

  • expand_or_jumpable(): returns expandable() or jumpable(1) (exists only because commonly, one key is used to both jump forward and expand).

  • expand_or_locally_jumpable(): same as expand_or_jumpable() except jumpable is ignored if the cursor is not inside the current snippet.

  • expand_or_jump(): returns true if jump/expand was succesful.

  • expand_auto(): expands the autosnippets before the cursor (not necessary to call manually, will be called via autocmd if enable_autosnippet is set in the config).

  • snip_expand(snip, opts): expand snip at the current cursor position. opts may contain the following keys:

    • clear_region: A region of text to clear after expanding (but before jumping into) snip. It has to be at this point (and therefore passed to this function) as clearing before expansion will populate TM_CURRENT_LINE and TM_CURRENT_WORD with wrong values (they would miss the snippet trigger) and clearing after expansion may move the text currently under the cursor and have it end up not at the i(1), but a #trigger chars to it's right. The actual values used for clearing are from and to, both (0,0)-indexed byte-positions. If the variables don't have to be populated with the correct values, it's safe to remove the text manually.
    • expand_params: table, for overriding the trigger used in the snippet and setting the captures (useful for pattern-triggered nodes where the trigger has to be changed from the pattern to the actual text triggering the node). Pass as trigger and captures.
    • pos: position ({line, col}), (0,0)-indexed (in bytes, as returned by nvim_win_get_cursor()), where the snippet should be expanded. The snippet will be put between (line,col-1) and (line,col). The snippet will be expanded at the current cursor if pos is nil. opts and either of its parameter may be nil.

  • get_active_snip(): returns the currently active snippet (not node!).

  • choice_active(): true if inside a choiceNode.

  • change_choice(direction): changes the choice in the innermost currently active choiceNode forward (direction = 1) or backward (direction = -1).

  • unlink_current(): removes the current snippet from the jumplist (useful if luasnip fails to automatically detect eg. deletion of a snippet) and sets the current node behind the snippet, or, if not possible, before it.

  • lsp_expand(snip_string, opts): expand the lsp-syntax-snippet defined via snip_string at the cursor. opts can have the same options as opts in snip_expand.

  • active_update_dependents(): update all function/dynamicNodes that have the current node as an argnode (will actually only update them if the text in any of the argnodes changed).

  • available(): return a table of all snippets defined for the current filetypes(s) ({ft1={snip1, snip2}, ft2={snip3, snip4}}).

  • exit_out_of_region(node): checks whether the cursor is still within the range of the snippet node belongs to. If yes, no change occurs, if No, the snippet is exited and following snippets' regions are checked and potentially exited (the next active node will be the 0-node of the snippet before the one the cursor is inside.
    If the cursor isn't inside any snippet, the active node will be the last node in the jumplist).
    If a jump causes an error (happens mostly because a snippet was deleted), the snippet is removed from the jumplist.

  • store_snippet_docstrings(snippet_table): Stores the docstrings of all snippets in snippet_table to a file (stdpath("cache")/luasnip/docstrings.json).
    Calling store_snippet_docstrings(snippet_table) after adding/modifying snippets and load_snippet_docstrings(snippet_table) on startup after all snippets have been added to snippet_table is a way to avoide regenerating the (unchanged) docstrings on each startup.
    (Depending on when the docstrings are required and how luasnip is loaded, it may be more sensible to let them load lazily, eg. just before they are required).
    snippet_table should be laid out just like luasnip.snippets (it will most likely always be luasnip.snippets).

  • load_snippet_docstrings(snippet_table): Load docstrings for all snippets in snippet_table from stdpath("cache")/luasnip/docstrings.json. The docstrings are stored and restored via trigger, meaning if two snippets for one filetype have the same(very unlikely to happen in actual usage), bugs could occur.
    snippet_table should be laid out as described in store_snippet_docstrings.

  • unlink_current_if_deleted(): Checks if the current snippet was deleted, if so, it is removed from the jumplist. This is not 100% reliable as luasnip only sees the extmarks and their begin/end may not be on the same position, even if all the text between them was deleted.

  • filetype_extend(filetype:string, extend_filetypes:table of string): Tells luasnip that for a buffer with ft=filetype, snippets from extend_filetypes should be searched as well. extend_filetypes is a lua-array ({ft1, ft2, ft3}).
    luasnip.filetype_extend("lua", {"c", "cpp"}) would search and expand c-and cpp-snippets for lua-files.

  • filetype_set(filetype:string, replace_filetypes:table of string): Similar to filetype_extend, but where append appended filetypes, set sets them: filetype_set("lua", {"c"}) causes only c-snippets to be expanded in lua-files, lua-snippets aren't even searched.

  • cleanup(): clears all snippets. Not useful for regular usage, only when authoring and testing snippets.

  • refresh_notify(ft:string): Triggers an autocmd that other plugins can hook into to perform various cleanup for the refreshed filetype.
    Useful for signaling that new snippets were added for the filetype ft.

Not covered in this section are the various node-constructors exposed by the module, their usage is shown either previously in this file or in Examples/snippets.lua (in the repo).