            __                       ____
           /\ \                     /\  _`\           __
           \ \ \      __  __     __ \ \,\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 the following:

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")
local extras = require("luasnip.extras")
local l = extras.lambda
local rep = extras.rep
local p = extras.partial
local m = extras.match
local n = extras.nonempty
local dl = extras.dynamic_lambda
local fmt = require("luasnip.extras.fmt").fmt
local fmta = require("luasnip.extras.fmt").fmta
local conds = require("luasnip.extras.expand_conditions")
local postfix = require("luasnip.extras.postfix").postfix
local types = require("luasnip.util.types")
local parse = require("luasnip.util.parser").parse_snippet
local ms = ls.multi_snippet
local k = require("luasnip.nodes.key_indexer").new_key

As noted in the Loaders-Lua-section:

By default, the names from luasnip.config.snip_env will be used, but it's possible to customize them by setting snip_env in setup.

Furthermore, note that while this document assumes you have defined ls to be require("luasnip"), it is not provided in the default set of variables.

Note: the source code of snippets in GIFs is actually here, and it's slightly different from the code below.

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 ls.expand is called.

Jump-Index

Nodes that can be jumped to (insertNode, choiceNode, dynamicNode, restoreNode, snippetNode) all require a "jump-index" so LuaSnip knows the order in which these nodes are supposed to be visited ("jumped to").

s("trig", {
	i(1), t"text", i(2), t"text again", i(3)
})

These indices don't "run" through the entire snippet, like they do in TextMate-snippets ("$1 ${2: $3 $4}"), they restart at 1 in each nested snippetNode:

s("trig", {
	i(1), t" ", sn(2, {
		t" ", i(1), t" ", i(2)
	})
})

(roughly equivalent to the given TextMate-snippet).

Adding Snippets

The snippets for a given filetype have to be added to LuaSnip via ls.add_snippets(filetype, snippets). Snippets that should be accessible globally (in all filetypes) have to be added to the special filetype all.

ls.add_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 in the section API.

Snippet Insertion

When a new snippet is expanded, it can be connected with the snippets that have already been expanded in the buffer in various ways.
First of all, Luasnip distinguishes between root-snippets and child-snippets. The latter are nested inside other snippets, so when jumping through a snippet, one may also traverse the child-snippets expanded inside it, more or less as if the child just contains more nodes of the parent.
Root-snippets are of course characterized by not being child-snippets.
When expanding a new snippet, it becomes a child of the snippet whose region it is expanded inside, and a root if it is not inside any snippet's region.
If it is inside another snippet, the specific node it is inside is determined, and the snippet then nested inside that node.

If that node is interactive (for example, an insertNode), the new snippet will be traversed when the node is visited, as long as the configuration-option link_children is enabled. If it is not enabled, it is possible to jump from the snippet to the node, but not the other way around.
If that node is not interactive, the snippet will be linked to the currently active node, also such that it will not be jumped to again once it is left. This is to prevent jumping large distances across the buffer as much as possible. There may still be one large jump from the snippet back to the current node it is nested inside, but that seems hard to avoid.
Thus, one should design snippets such that the regions where other snippets may be expanded are inside insertNodes.

If the snippet is not a child, but a root, it can be linked up with the roots immediately adjacent to it by enabling link_roots in setup. Since by default only one root is remembered, one should also set keep_roots if link_roots is enabled. The two are separate options, since roots that are not linked can still be reached by ls.activate_node(). This setup (remember roots, but don't jump to them) is useful for a super-tab like mapping (<Tab> and jump on the same key), where one would like to still enter previous roots. Since there would almost always be more jumps if the roots are linked, regular <Tab> would not work almost all the time, and thus link_roots has to stay disabled.

Node

Every node accepts, as its last parameter, an optional table of arguments. There are some common ones (which are listed here), and some that only apply to some nodes (user_args for function/dynamicNode). These opts are only mentioned if they accept options that are not common to all nodes.

Common opts:

node_ext_opts and merge_node_ext_opts: Control ext_opts (most likely highlighting) of the node. Described in detail in ext_opts
key: The node can be referred to by this key. Useful for either Key Indexer or for finding the node at runtime (See Snippets-API), for example inside a dynamicNode. The keys do not have to be unique across the entire lifetime of the snippet, but at any point in time, the snippet may contain each key only once. This means it is fine to return a keyed node from a dynamicNode, because even if it will be generated multiple times, those will not be valid at the same time.
node_callbacks: Define event-callbacks for this node (see events).
Accepts a table that maps an event, e.g. events.enter to the callback (essentially the same as callbacks passed to s, only that there is no first mapping from jump-index to the table of callbacks).

API

get_jump_index(): this method returns the jump-index of a node. If a node doesn't have a jump-index, this method returns nil instead.
get_buf_position(opts) -> {from_position, to_position}: Determines the range of the buffer occupied by this node. from- and to_position are row,column-tuples, 0,0-indexed (first line is 0, first column is 0) and end-inclusive (see :h api-indexing, this is extmarks indexing).
- opts: table|nil, options, valid keys are:
  - raw: bool, default true. This can be used to switch between byte-columns (raw=true) and visual columns (raw=false). This makes a difference if the line contains characters represented by multiple bytes in UTF, for example ÿ.

Snippets

The most direct way to define snippets is s:

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

(This snippet is useless beyond serving as a minimal example)

s(context, nodes, opts) -> snippet

context: Either table or a string. Passing a string is equivalent to passing
```
{
	trig = context
}
```
The following keys are valid:
- trig: string, the trigger of the snippet. If the text in front of (to the left of) the cursor when ls.expand() is called matches it, the snippet will be expanded.
  By default, "matches" means the text in front of the cursor matches the trigger exactly, this behavior can be modified through trigEngine
- name: string, can be used by e.g. nvim-compe to identify the snippet.
- desc (or 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.
  Consider setting trigEngine to "pattern" instead, it is more expressive, and in line with other settings.
- trigEngine: (function|string), determines how trig is interpreted, and what it means for it to "match" the text in front of the cursor.
  This behavior can be completely customized by passing a function, but the predefined ones, which are accessible by passing their identifier, should suffice in most cases:
  - "plain": the default-behavior, the trigger has to match the text before the cursor exactly.
  - "pattern": the trigger is interpreted as a Lua pattern, and is a match if trig .. "$" matches the line up to the cursor. Capture-groups will be accessible as snippet.captures.
  - "ecma": the trigger is interpreted as an ECMAscript-regex, and is a match if trig .. "$" matches the line up to the cursor. Capture-groups will be accessible as snippet.captures.
    This trigEngine requires jsregexp (see LSP-snippets-transformations) to be installed, if it is not, this engine will behave like "plain".
  - "vim": the trigger is interpreted as a vim-regex, and is a match if trig .. "$" matches the line up to the cursor. As with the other regex/pattern-engines, captures will be available as snippet.captures, but there is one caveat: the matching is done using matchlist, so for now empty-string submatches will be interpreted as unmatched, and the corresponding snippet.capture[i] will be nil (this will most likely change, don't rely on this behavior).
  Besides these predefined engines, it is also possible to create new ones: Instead of a string, pass a function which satisfies trigEngine(trigger, opts) -> (matcher(line_to_cursor, trigger) -> whole_match, captures) (i.e. the function receives trig and trigEngineOpts can, for example, precompile a regex, and then returns a function responsible for determining whether the current cursor-position (represented by the line up to the cursor) matches the trigger (it is passed again here so engines which don't do any trigger-specific work (like compilation) can just return a static matcher), and what the capture-groups are).
  The lua-engine, for example, can be implemented like this:
```
local function matcher(line_to_cursor, trigger)
    -- look for match which ends at the cursor.
    -- put all results into a list, there might be many capture-groups.
    local find_res = { line_to_cursor:find(trigger .. "$") }

    if #find_res > 0 then
        -- if there is a match, determine matching string, and the
        -- capture-groups.
        local captures = {}
        -- find_res[1] is `from`, find_res[2] is `to` (which we already know
        -- anyway).
        local from = find_res[1]
        local match = line_to_cursor:sub(from, #line_to_cursor)
        -- collect capture-groups.
        for i = 3, #find_res do
            captures[i - 2] = find_res[i]
        end
        return match, captures
    else
        return nil
    end
end

local function engine(trigger)
    -- don't do any special work here, can't precompile lua-pattern.
    return matcher
end
```
  The predefined engines are defined in trig_engines.lua, read it for more examples.
- trigEngineOpts: table<string, any>, options for the used trigEngine.
  The valid options are:
  - max_len: number, upper bound on the length of the trigger.
    If this is set, the line_to_cursor will be truncated (from the cursor of course) to max_len characters before performing the match.
    This is implemented because feeding long line_to_cursor into e.g. the pattern-trigEngine will hurt performance quite a bit (see issue Luasnip#1103).
    This option is implemented for all trigEngines.
- docstring: string, textual representation of the snippet, specified like desc. Overrides docstrings loaded from json.
- docTrig: string, used as line_to_cursor during docstring-generation. This might be relevant if the snippet relies on specific values in the capture-groups (for example, numbers, which won't work with the default $CAPTURESN used during docstring-generation)
- hidden: boolean, hint for completion-engines. If set, the snippet should not show up when querying snippets.
- priority: positive number, Priority of the snippet, 1000 by default.
  Snippets with high priority will be matched to a trigger before those with a lower one. The priority for multiple snippets can also be set in add_snippets.
- snippetType: string, should be either snippet or autosnippet (ATTENTION: singular form is used), decides whether this snippet has to be triggered by ls.expand() or whether is triggered automatically (don't forget to set ls.config.setup({ enable_autosnippets = true }) if you want to use this feature). If unset it depends on how the snippet is added of which type the snippet will be.
- resolveExpandParams: fn(snippet, line_to_cursor, matched_trigger, captures) -> table|nil, where
  - snippet: Snippet, the expanding snippet object
  - line_to_cursor: string, the line up to the cursor.
  - matched_trigger: string, the fully matched trigger (can be retrieved from line_to_cursor, but we already have that info here :D)
  - captures: captures as returned by trigEngine.
  This function will be evaluated in Snippet:matches() to decide whether the snippet can be expanded or not.
  Returns a table if the snippet can be expanded, nil if can not. The returned table can contain any of these fields:
  - trigger: string, the fully matched trigger.
  - captures: table, this list could update the capture-groups from parameter in snippet expansion. Both trigger and captures can override the values returned via trigEngine.
  - clear_region: { "from": {<row>, <column>}, "to": {<row>, <column>} }, both (0, 0)-indexed, the region where text has to be cleared before inserting the snippet.
  - env_override: map string->(string[]|string), override or extend the snippet's environment (snip.env)
  If any of these is nil, the default is used (trigger and captures as returned by trigEngine, clear_region such that exactly the trigger is deleted, no overridden environment-variables).
  
  A good example for the usage of resolveExpandParams can be found in the implementation of postfix.
- condition: fn(line_to_cursor, matched_trigger, captures) -> bool, where
  - line_to_cursor: string, the line up to the cursor.
  - matched_trigger: string, the fully matched trigger (can be retrieved from line_to_cursor, but we already have that info here :D)
  - captures: if the trigger is pattern, this list contains the capture-groups. Again, could be computed from line_to_cursor, but we already did so.
- show_condition: f(line_to_cursor) -> bool.
  - line_to_cursor: string, the line up to the cursor.
  This function is (should be) evaluated by completion engines, indicating whether the snippet should be included in current completion candidates.
  Defaults to a function returning true.
  This is different from condition because condition is evaluated by LuaSnip on snippet expansion (and thus has access to the matched trigger and captures), while show_condition is (should be) evaluated by the completion engines when scanning for available snippet candidates.
- filetype: string, the filetype of the snippet. This overrides the filetype the snippet is added (via add_snippet) as.
nodes: A single node or a list of nodes. The nodes that make up the snippet.
opts: A table with the following valid keys:
- callbacks: Contains functions that are called upon entering/leaving a node of this snippet.
  For example: to print text upon entering the second node of a snippet, callbacks should be set as follows:
```
{
	-- position of the node, not the jump-index!!
	-- s("trig", {t"first node", t"second node", i(1, "third node")}).
	[2] = {
		[events.enter] = function(node, _event_args) print("2!") end
	}
}
```
  To register a callback for the snippets' own events, the key [-1] may be used. More info on events in events
- child_ext_opts, merge_child_ext_opts: Control ext_opts applied to the children of this snippet. More info on those in the ext_opts-section.

The opts-table, as described here, can also be passed to e.g. snippetNode and indentSnippetNode.
It is also possible to set condition and show_condition (described in the documentation of the context-table) from opts. They should, however, not be set from both.

Data

Snippets contain some interesting tables during runtime:

snippet.env: Contains variables used in the LSP-protocol, for example TM_CURRENT_LINE or TM_FILENAME. It's possible to add customized variables here too, check Variables-Environment Namespaces
snippet.captures: If the snippet was triggered by a pattern (regTrig), and the pattern contained capture-groups, they can be retrieved here.
snippet.trigger: The string that triggered this snippet. Again, only interesting if the snippet was triggered through regTrig, for getting the full match.

These variables/tables primarily come in handy in dynamic/functionNodes, where the snippet can be accessed through the immediate parent (parent.snippet), which is passed to the function. (in most cases parent == parent.snippet, but the parent of the dynamicNode is not always the surrounding snippet, it could be a snippetNode).

API

invalidate(): call this method to effectively remove the snippet. The snippet will no longer be able to expand via expand or expand_auto. It will also be hidden from lists (at least if the plugin creating the list respects the hidden-key), but it might be necessary to call ls.refresh_notify(ft) after invalidating snippets.
get_keyed_node(key): Returns the currently visible node associated with key.

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."})
})

t(text, node_opts):

text: string or string[]
node_opts: table, see Node

InsertNode

These Nodes contain editable text and can be jumped to- and from (e.g. traditional placeholders and tabstops, 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 Insert Nodes are visited in order 1,2,3,..,n,0.
(The jump-index 0 also has to belong to an insertNode!) So the order of InsertNode-jumps is as follows:

After expansion, the cursor is at InsertNode 1,
after jumping forward once at InsertNode 2,
and after jumping forward again at InsertNode 0.

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

The jump-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-indices restart 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 e.g. the TextMate syntax, where tabstops are snippet-global:

${1:First jump} :: ${2: ${3:Third jump} : ${4:Fourth 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 when parsed).

i(jump_index, text, node_opts)

jump_index: number, this determines when this node will be jumped to (see Basics-Jump-Index).
text: string|string[], a single string for just one line, a list with >1 entries for multiple lines. This text will be SELECTed when the insertNode is jumped into.
node_opts: table, described in Node

If the jump_index is 0, replacing its' text will leave it outside the insertNode (for reasons, check out Luasnip#110).

FunctionNode

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

local function fn(
  args,     -- text from i(2) in this example i.e. { { "456" } }
  parent,   -- parent snippet or parent node
  user_args -- user_args from opts.user_args 
)
   return '[' .. args[1][1] .. user_args .. ']'
end

s("trig", {
  i(1), t '<-i(1) ',
  f(fn,  -- callback (args, parent, user_args) -> string
    {2}, -- node indice(s) whose text is passed to fn, i.e. i(2)
    { user_args = { "user_args_value" }} -- opts
  ),
  t ' i(2)->', i(2), t '<-i(2) i(0)->', i(0)
})

f(fn, argnode_references, node_opts):

fn: function(argnode_text, parent, user_args1,...,user_argsn) -> text
- argnode_text: string[][], the text currently contained in the argnodes (e.g. {{line1}, {line1, line2}}). The snippet indent will be removed from all lines following the first.
- parent: The immediate parent of the functionNode.
  It is included here as it allows easy access to some information that could be useful in functionNodes (see Snippets-Data for some examples).
  Many snippets access the surrounding snippet just as parent, but if the functionNode is nested within a snippetNode, the immediate parent is a snippetNode, not the surrounding snippet (only the surrounding snippet contains data like env or captures).
- user_args: The user_args passed in opts. Note that there may be multiple user_args (e.g. user_args1, ..., user_argsn).
fn shall return a string, which will be inserted as is, or a table of strings for multiline strings, where all lines following the first will be prefixed with the snippets' indentation.
argnode_references: node_reference[]|node_refernce|nil.
Either no, a single, or multiple Node References. Changing any of these will trigger a re-evaluation of fn, and insertion of the updated text.
If no node reference is passed, the functionNode is evaluated once upon expansion.

node_opts: table, see Node. One additional key is supported:

user_args: any[], these will be passed to fn as user_arg1-user_argn. These make it easier to reuse similar functions, for example a functionNode that wraps some text in different delimiters ((), [], ...).

local function reused_func(_,_, user_arg1)
    return user_arg1
end

s("trig", {
    f(reused_func, {}, {
        user_args = {"text"}
    }),
    f(reused_func, {}, {
        user_args = {"different text"}
    }),
})

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, {})
)

argnodes_text during function evaluation:

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

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 (See Extras-Lambda)

Node Reference

Node references are used to refer to other nodes in various parts of LuaSnip's API.
For example, argnodes in functionNode, dynamicNode or lambda are node references.
These references can be either of:

number: the jump-index of the node. This will be resolved relative to the parent of the node this is passed to. (So, only nodes with the same parent can be referenced. This is very easy to grasp, but also limiting)
key_indexer: the key of the node, if it is present. This will come in handy if the node that is being referred to is not in the same snippet/snippetNode as the one the node reference is passed to. Also, it is the proper way to refer to a non-interactive node (a functionNode, for example)
absolute_indexer: the absolute position of the node. Just like key_indexer, it allows addressing non-sibling nodes, but is a bit more awkward to handle since a path from root to node has to be determined, whereas key_indexer just needs the key to match.
Due to this, key_indexer should be generally preferred. (More information in Absolute Indexer).
node: just the node. Usage of this is discouraged since it can lead to subtle errors (for example, if the node passed here is captured in a closure and therefore not copied with the remaining tables in the snippet; there's a big comment about just this in commit 8bfbd61).

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(jump_index, choices, node_opts)

jump_index: number, since choiceNodes can be jumped to, they need a jump-index (Info in Basics-Jump-Index).
choices: node[]|node, the choices. The first will be initially active. A list of nodes will be turned into a snippetNode.
node_opts: table. choiceNode supports the keys common to all nodes described in Node, and one additional key:
- 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 decreased 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 jump-index is the same as the choiceNodes'.

As it is only possible (for now) to change choices from within the choiceNode, make sure that all of the choices have some place for the cursor to stop at!

This means that in sn(nil, {...nodes...}) nodes has to contain e.g. an i(1), otherwise LuaSnip will just "jump through" the nodes, making it impossible to change the choice.

c(1, {
	t"some text", -- textNodes are just stopped at.
	i(nil, "some text"), -- likewise.
	sn(nil, {t"some text"}) -- this will not work!
	sn(nil, {i(1), t"some text"}) -- this will.
})

The active choice for a choiceNode can be changed by either calling one of ls.change_choice(1) (forwards) or ls.change_choice(-1) (backwards), or by calling ls.set_choice(choice_indx).

One way to easily interact with choiceNodes is binding change_choice(1/-1) to keys:

-- set keybinds for both INSERT and VISUAL.
vim.api.nvim_set_keymap("i", "<C-n>", "<Plug>luasnip-next-choice", {})
vim.api.nvim_set_keymap("s", "<C-n>", "<Plug>luasnip-next-choice", {})
vim.api.nvim_set_keymap("i", "<C-p>", "<Plug>luasnip-prev-choice", {})
vim.api.nvim_set_keymap("s", "<C-p>", "<Plug>luasnip-prev-choice", {})

Apart from this, there is also a picker (see select_choice where no cycling is necessary and any choice can be selected right away, via vim.ui.select.

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.

Their syntax is similar to s, however, where snippets require a table specifying when to expand, snippetNodes, similar to insertNodes, expect a jump-index.

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

sn(jump_index, nodes, node_opts)

jump_index: number, the usual Jump-Index.
nodes: node[]|node, just like for s.
Note that snippetNodes don't accept an i(0), so the jump-indices of the nodes inside them have to be in 1,2,...,n.
node_opts: table: again, the keys common to all nodes (documented in Node) are supported, but also
- callbacks,
- child_ext_opts and
- merge_child_ext_opts,
which are further explained in Snippets.

IndentSnippetNode

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

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 line breaks, 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 use case for ISN is inserting text, e.g. // 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 line breaks.

To enable such usage, $PARENT_INDENT in the indentstring is replaced by the parent's indent.

isn(jump_index, nodes, indentstring, node_opts)

All of these parameters except indentstring are exactly the same as in SnippetNode.

indentstring: string, will be used to indent the nodes inside this snippetNode.
All occurrences of "$PARENT_INDENT" are replaced with the actual indent of the parent.

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.

d(jump_index, function, node-references, opts):

jump_index: number, just like all jumpable nodes, its' position in the jump-list (Basics-Jump-Index).
function: fn(args, parent, old_state, user_args) -> snippetNode This function is called when the argnodes' text changes. It should generate and return (wrapped inside a snippetNode) nodes, which will be inserted at the dynamicNode's 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_args: passed through from dynamicNode-opts; may have more than one argument.
node_references: node_reference[]|node_references|nil, Node References to the nodes the dynamicNode depends on: if any of these trigger an update (for example, if the text inside them changes), the dynamicNodes' function will be executed, and the result inserted at the dynamicNodes place.
(dynamicNode behaves exactly the same as functionNode in this regard).
opts: In addition to the common Node-keys, there is, again,
- user_args, which is described in FunctionNode.

Examples:

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

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 snippet makes use of old_state to count the number of updates.

To store/restore values generated by the dynamicNode or entered into insert/choiceNode, consider using the shortly-introduced restoreNode instead of old_state.

local function count(_, _, old_state)
	old_state = old_state or {
		updates = 0
	}

	old_state.updates = old_state.updates + 1

	local snip = sn(nil, {
		t(tostring(old_state.updates))
	})

	snip.old_state = old_state
	return snip
end

ls.add_snippets("all",
	s("trig", {
		i(1, "change to update"),
		d(2, count, {1})
	})
)

As with functionNode, user_args can be used to reuse similar dynamicNode- functions.

RestoreNode

This node can store and restore a snippetNode as is. This includes changed choices and changed text. Its' 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 = {
		-- key passed to restoreNodes.
		["user_text"] = i(1, "default_text")
	}
})

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

r(jump_index, key, nodes, node_opts):

jump_index, when to jump to this node.
key, string: restoreNodes with the same key share their content.
nodes, node[]|node: the content 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 for a given key may be defined multiple times, but if the contents differ, it's undefined which will actually be used.
If a key's content is defined in a dynamicNode, it will not be initially 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 especially useful for storing 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.

Key Indexer

A very flexible way of referencing nodes (Node Reference).
While the straightforward way of addressing nodes via their Jump-Index suffices in most cases, a dynamic/functionNode can only depend on nodes in the same snippet(Node), its siblings (since the index is interpreted as relative to their parent). Accessing a node with a different parent is thus not possible. Secondly, and less relevant, only nodes that actually have a jump-index can be referred to (a functionNode, for example, cannot be depended on).
Both of these restrictions are lifted with key_indexer:
It allows addressing nodes by their key, which can be set when the node is constructed, and is wholly independent of the nodes' position in the snippet, thus enabling descriptive labeling.

The following snippets demonstrate the issue and the solution by using key_indexer:

First, the addressed problem of referring to nodes outside the functionNodes parent:

s("trig", {
	i(1), c(2, {
		sn(nil, {
			t"cannot access the argnode :(",
			f(function(args)
			    return args[1]
            end, {???}) -- can't refer to i(1), since it isn't a sibling of `f`.
		}),
		t"sample_text"
	})
})

And the solution: first give the node we want to refer to a key, and then pass the same to the functionNode.

s("trig", {
	i(1, "", {key = "i1-key"}), c(2, {
		sn(nil, { i(1),
			t"can access the argnode :)",
			f(function(args)
                return args[1]
            end, k("i1-key") )
		}),
		t"sample_text"
	})
})

Absolute Indexer

absolute_indexer allows accessing nodes by their unique jump-index path from the snippet-root. This makes it almost as powerful as Key Indexer, but again removes the possibility of referring to non-jumpable nodes and makes it all a bit more error-prone since the jump-index paths are hard to follow, and (unfortunately) have to be a bit verbose (see the long example of absolute_indexer-positions below). Consider just using Key Indexer instead.

(The solution-snippet from Key Indexer, but using ai instead.)

s("trig", {
	i(1), c(2, {
		sn(nil, { i(1),
			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 jump-index, 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}

are all the same node.

MultiSnippet

There are situations where it might be comfortable to access a snippet in different ways. For example, one might want to enable auto-triggering in regions where the snippets usage is common, while leaving it manual-only in others.
This is where ms should be used: A single snippet can be associated with multiple contexts (the context-table determines the conditions under which a snippet may be triggered).
This has the advantage (compared with just registering copies) that all contexts are backed by a single snippet, and not multiple, and it's (at least should be :D) more comfortable to use.

ms(contexts, nodes, opts) -> addable:

contexts: table containing list of contexts, and some keywords.
context are described in Snippets, here they may also be tables or strings.
So far, there is only one valid keyword:
- common: Accepts yet another context.
  The options in common are applied to (but don't override) the other contexts specified in contexts.
nodes: List of nodes, exactly like in Snippets.
opts: Table, options for this function:
- common_opts: The snippet-options (see also Snippets) applied to the snippet generated from nodes.

The returned object is an addable, something which can be passed to add_snippets, or returned from the lua-loader.

Examples:

ls.add_snippets("all", {
    ms({"a", "b"}, {t"a or b"})
})

ls.add_snippets("all", {
    ms({
        common = {snippetType = "autosnippet"},
        "a",
        "b"
    }, {
        t"a or b (but autotriggered!!)"
    })
})

ls.add_snippets("all", {
    ms({
        common = {snippetType = "autosnippet"},
        {trig = "a", snippetType = "snippet"},
        "b",
        {trig = "c", condition = function(line_to_cursor)
            return line_to_cursor == ""
        end}
    }, {
        t"a or b (but autotriggered!!)"
    })
})

Extras

Lambda

A shortcut for functionNodes that only do very basic string manipulation.

l(lambda, argnodes):

lambda: An object created by applying string-operations to l._n, objects representing the nth argnode.
For example:
- l._1:gsub("a", "e") replaces all occurrences of "a" in the text of the first argnode with "e", or
- l._1 .. l._2 concatenates text of the first and second argnode. If an argnode contains multiple lines of text, they are concatenated with "\n" prior to any operation.
argnodes, a Node Reference, just like in function- and dynamicNode.

There are many examples for lambda in Examples/snippets.lua

Match

match can insert text based on a predicate (again, a shorthand for functionNode).

match(argnodes, condition, then, else):

argnode: A single Node Reference. May not be nil, or a table.
condition may be either of
- 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.
  Should end with match, but any other truthy result will be interpreted as matching.
then is inserted if the condition matches,
else if it does not.

Both then and else can be either text, lambda or function (with the same parameters as specified above).
then's default-value depends on the condition:

pattern: Simply the return value from the match, e.g. 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") .

match(n, lambda._1:match(lambda._1:reverse()), "PALINDROME")

s("trig", {
	i(1), t":",
	i(2), t"::",
	m({1, 2}, l._1:match("^"..l._2.."$"), l._1:gsub("a", "e"))
})

```
  s("extras1", {
    i(1), t { "", "" }, m(1, "^ABC$", "A")
  })
```
Inserts "A" if the node with jump-index n matches "ABC" exactly, nothing otherwise.

s("extras2", {
  i(1, "INPUT"), t { "", "" }, m(1, l._1:match(l._1:reverse()), "PALINDROME")
})

Inserts "PALINDROME" if i(1) contains a palindrome.

s("extras3", {
  i(1), t { "", "" }, i(2), t { "", "" },
  m({ 1, 2 }, l._1:match("^" .. l._2 .. "$"), l._1:gsub("a", "e"))
})

This inserts the text of the node with jump-index 1, with all occurrences of a replaced with e, if the second insertNode matches the first exactly.

Repeat

Inserts the text of the passed node.

rep(node_reference)

node_reference, a single Node Reference.

s("extras4", { i(1), t { "", "" }, extras.rep(1) })

Partial

Evaluates a function on expand and inserts its value.

partial(fn, params...)

fn: any function
params: varargs, any, will be passed to fn.

For example partial(os.date, "%Y") inserts the current year on expansion.

s("extras5", { extras.partial(os.date, "%Y") })

Nonempty

Inserts text if the referenced node doesn't contain any text.

nonempty(node_reference, not_empty, empty):

node_reference, a single Node Reference.
not_empty, string: inserted if the node is not empty.
empty, string: inserted if the node is empty.

s("extras6", { i(1, ""), t { "", "" }, extras.nonempty(1, "not empty!", "empty!") })

Dynamic Lambda

Pretty much the same as lambda, but it inserts the resulting text as an insertNode, and, as such, it can be quickly overridden.

dynamic_lambda(jump_indx, lambda, node_references)

jump_indx, as usual, the jump-index.

The remaining arguments carry over from lambda.

s("extras7", { i(1), t { "", "" }, extras.dynamic_lambda(2, l._1 .. l._1, 1) })

`fmt`

Authoring snippets can be quite clunky, especially since every second node is probably a textNode, inserting a small number of characters between two more complicated nodes.

fmt can be used to define snippets in a much more readable way. This is achieved by borrowing (as the name implies) from format-functionality (our syntax is very similar to python's).

fmt accepts a string and a table of nodes. Each occurrence of a delimiter pair in the string is replaced by one node from the table, while text outside the delimiters is turned into textNodes.

Simple example:

ls.add_snippets("all", {
  -- important! fmt does not return a snippet, it returns a table of nodes.
  s("example1", fmt("just an {iNode1}", {
    iNode1 = i(1, "example")
  })),
  s("example2", fmt([[
  if {} then
    {}
  end
  ]], {
    -- i(1) is at nodes[1], i(2) at nodes[2].
    i(1, "not now"), i(2, "when")
  })),
  s("example3", fmt([[
  if <> then
    <>
  end
  ]], {
    -- i(1) is at nodes[1], i(2) at nodes[2].
    i(1, "not now"), i(2, "when")
  }, {
    delimiters = "<>"
  })),
  s("example4", fmt([[
  repeat {a} with the same key {a}
  ]], {
    a = i(1, "this will be repeat")
  }, {
    repeat_duplicates = true
  }))
})

One important detail here is that the position of the delimiters does not, in any way, correspond to the jump-index of the nodes!

fmt(format:string, nodes:table of nodes, opts:table|nil) -> table of nodes

format: a string. Occurrences of {<somekey>} ( {} are customizable; more on that later) are replaced with content[<somekey>] (which should be a node), while surrounding text becomes textNodes.
To escape a delimiter, repeat it ("{{").
If no key is given ({}) are numbered automatically:
"{} ? {} : {}" becomes "{1} ? {2} : {3}", while "{} ? {3} : {}" becomes "{1} ? {3} : {4}" (the count restarts at each numbered placeholder). If a key appears more than once in format, the node in content[<duplicate_key>] is inserted for the first, and copies of it for subsequent occurrences.
nodes: just a table of nodes.
opts: optional arguments:
- delimiters: string, two characters. Change {} to some other pair, e.g. "<>".
- strict: Warn about unused nodes (default true).
- trim_empty: remove empty ("%s*") first and last line in format. Useful when passing multiline strings via [[]] (default true).
- dedent: remove indent common to all lines in format. Again, makes passing multiline-strings a bit nicer (default true).
- repeat_duplicates: repeat nodes when a key is reused instead of copying the node if it has a jump-index, refer to Basics-Jump-Index to know which nodes have a jump-index (default false).

There is also require("luasnip.extras.fmt").fmta. This only differs from fmt by using angle brackets (<>) as the default delimiter.

Conditions

This module (luasnip.extras.condition) contains functions that can be passed to a snippet's condition or show_condition. These are grouped accordingly into luasnip.extras.conditions.expand and luasnip.extras.conditions.show:

expand:

line_begin: only expand if the cursor is at the beginning of the line.

show:

line_end: only expand at the end of the line.
has_selected_text: only expand if there's selected text stored after pressing store_selection_keys.

Additionally, expand contains all conditions provided by show.

Condition Objects

luasnip.extras.conditions also contains condition objects. These can, just like functions, be passed to condition or show_condition, but can also be combined with each other into logical expressions:

-c1 -> not c1
c1 * c2 -> c1 and c2
c1 + c2 -> c1 or c2
c1 - c2 -> c1 and not c2: This is similar to set differences: A \ B = {a in A | a not in B}. This makes -(a + b) = -a - b an identity representing de Morgan's law: not (a or b) = not a and not b. However, since boolean algebra lacks an additive inverse, a + (-b) = a - b does not hold. Thus, this is NOT the same as c1 + (-c2).
c1 ^ c2 -> c1 xor(!=) c2
c1 % c2 -> c1 xnor(==) c2: This decision may seem weird, considering how there is an overload for the ==-operator. Unfortunately, it's not possible to use this for our purposes (some info here), so we decided to make use of a more obscure symbol (which will hopefully avoid false assumptions about its meaning).

This makes logical combinations of conditions very readable. Compare

condition = conditions.expand.line_end + conditions.expand.line_begin

with the more verbose

condition = function(...) return conditions.expand.line_end(...) or conditions.expand.line_begin(...) end

The conditions provided in show and expand are already condition objects. To create new ones, use require("luasnip.extras.conditions").make_condition(condition_fn)

On The Fly-Snippets

Sometimes it's desirable to create snippets tailored for exactly the current situation. For example inserting repetitive, but just slightly different invocations of some function, or supplying data in some schema.

On-the-fly snippets enable exactly this use case: they can be quickly created and expanded with as little disruption as possible.

Since they should mainly fast to write and don't necessarily need all bells and whistles, they don't make use of lsp/textmate-syntax, but a more simplistic one:

$anytext denotes a placeholder (insertNode) with text "anytext". The text also serves as a unique key: if there are multiple placeholders with the same key, only the first will be editable, the others will just mirror it.
... That's it. $ can be escaped by preceding it with a second $, all other symbols will be interpreted literally.

There is currently only one way to expand on-the-fly snippets:
require('luasnip.extras.otf').on_the_fly("<some-register>") will interpret whatever text is in the register <some-register> as a snippet, and expand it immediately. The idea behind this mechanism is that it enables a very immediate way of supplying and retrieving (expanding) the snippet: write the snippet-body into the buffer, cut/yank it into some register, and call on_the_fly("<register>") to expand the snippet.

Here's one set of example keybindings:

" in the first call: passing the register is optional since `on_the_fly`
" defaults to the unnamed register, which will always contain the previously cut
" text.
vnoremap <c-f>  "ec<cmd>lua require('luasnip.extras.otf').on_the_fly("e")<cr>
inoremap <c-f>  <cmd>lua require('luasnip.extras.otf').on_the_fly("e")<cr>

Obviously, <c-f> is arbitrary and can be changed to any other key combo. Another interesting application is allowing multiple on-the-fly snippets at the same time by retrieving snippets from multiple registers:

" 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>

select_choice

It's possible to leverage vim.ui.select for selecting a choice directly, without cycling through the available choices.
All that is needed for this is calling require("luasnip.extras.select_choice"), most likely via some keybinding, e.g.

inoremap <c-u> <cmd>lua require("luasnip.extras.select_choice")()<cr>

while inside a choiceNode.
The opts.kind hint for vim.ui.select will be set to luasnip.

Filetype-Functions

Contains some utility functions that can be passed to the ft_func or load_ft_func-settings.

from_filetype: the default for ft_func. Simply returns the filetype(s) of the buffer.
from_cursor_pos: uses tree-sitter to determine the filetype at the cursor. With that, it's possible to expand snippets in injected regions, as long as the tree-sitter parser supports them. If this is used in conjunction with lazy_load, extra care must be taken that all the filetypes that can be expanded in a given buffer are also returned by load_ft_func (otherwise their snippets may not be loaded). This can easily be achieved with extend_load_ft.
extend_load_ft: fn(extend_ft:map) -> fn A simple solution to the problem described above is loading more filetypes than just that of the target buffer when lazy_loading. This can be done ergonomically via extend_load_ft: calling it with a table where the keys are filetypes, and the values are the filetypes that should be loaded additionally returns a function that can be passed to load_ft_func and takes care of extending the filetypes properly.
```
ls.setup({
	load_ft_func =
		-- Also load both lua and json when a markdown-file is opened,
		-- javascript for html.
		-- Other filetypes just load themselves.
		require("luasnip.extras.filetype_functions").extend_load_ft({
			markdown = {"lua", "json"},
			html = {"javascript"}
		})
})
```

Postfix-Snippet

Postfix snippets, famously used in rust analyzer and various IDEs, are a type of snippet which alters text before the snippet's trigger. While these can be implemented using regTrig snippets, this helper makes the process easier in most cases.

The simplest example, which surrounds the text preceding the .br with brackets [], looks like:

postfix(".br", {
    f(function(_, parent)
        return "[" .. parent.snippet.env.POSTFIX_MATCH .. "]"
    end, {}),
})

and is triggered with xxx.br and expands to [xxx].

Note the parent.snippet.env.POSTFIX_MATCH in the function node. This is additional field generated by the postfix snippet. This field is generated by extracting the text matched (using a configurable matching string, see below) from before the trigger. In the case above, the field would equal "xxx". This is also usable within dynamic nodes.

This field can also be used within lambdas and dynamic nodes.

postfix(".br", {
	l("[" .. l.POSTFIX_MATCH .. "]"),
})

postfix(".brd", {
	d(1, function (_, parent)
		return sn(nil, {t("[" .. parent.env.POSTFIX_MATCH .. "]")})
	end)
})

The arguments to postfix are identical to the arguments to s but with a few extra options.

The first argument can be either a string or a table. If it is a string, that string will act as the trigger, and if it is a table it has the same valid keys as the table in the same position for s except:

wordTrig: This key will be ignored if passed in, as it must always be false for postfix snippets.
match_pattern: The pattern that the line before the trigger is matched against. The default match pattern is "[%w%.%_%-]+$". Note the $. This matches since only the line up until the beginning of the trigger is matched against the pattern, which makes the character immediately preceding the trigger match as the end of the string.

Some other match strings, including the default, are available from the postfix module. require("luasnip.extras.postfix).matches:

default: [%w%.%_%-%"%']+$
line: ^.+$

The second argument is identical to the second argument for s, that is, a table of nodes.

The optional third argument is the same as the third (opts) argument to the s function, but with one difference:

The postfix snippet works using a callback on the pre_expand event of the snippet. If you pass a callback on the pre_expand event (structure example below) it will get run after the builtin callback.

{
	callbacks = {
	[-1] = {
		[events.pre_expand] = function(snippet, event_args)
		-- function body to match before the dot
		-- goes here
		end
		}
	}
}

Treesitter-Postfix-Snippet

Instead of triggering a postfix-snippet when some pattern matches in front of the trigger, it might be useful to match if some specific tree-sitter nodes surround/are in front of the trigger.
While this functionality can also be implemented by a custom resolveExpandParams, this helper simplifies the common cases.

This matching of tree-sitter nodes can be done either

by providing a query and the name of the capture that should be in front of the trigger (in most cases, the complete match, but requiring specific nodes before/after the matched node may be useful as well), or
by providing a function that manually walks the node-tree, and returns the node in front of the trigger on success (for increased flexibility).

A simple example, which surrounds the previous node's text preceding the .mv with std::move() in C++ files, looks like:

local treesitter_postfix = require("luasnip.extras.treesitter_postfix").treesitter_postfix

treesitter_postfix({
    trig = ".mv",
    matchTSNode = {
        query = [[
            [
              (call_expression)
              (identifier)
              (template_function)
              (subscript_expression)
              (field_expression)
              (user_defined_literal)
            ] @prefix
        ]]
        query_lang = "cpp"
    },
},{
    f(function(_, parent)
        local node_content = table.concat(parent.snippet.env.LS_TSMATCH, '\n')
        local replaced_content = ("std::move(%s)"):format(node_content)
        return vim.split(ret_str, "\n", { trimempty = false })
    end)
})

LS_TSMATCH is the tree-sitter-postfix equivalent to POSTFIX_MATCH, and is populated with the match (in this case the text of a tree-sitter-node) in front of the trigger.

The arguments to treesitter_postfix are identical to the arguments to s but with a few extra options.

The first argument has to be a table, which defines at least trig and matchTSNode. All keys from the regular s may be set here (except for wordTrig, which will be ignored), and additionally the following:

reparseBuffer, string?: Sometimes the trigger may interfere with tree-sitter recognizing queries correctly. With this option, the trigger may either be removed from the live-buffer ("live"), from a copy of the buffer ("copy"), or not at all (nil).
matchTSNode: How to determine whether there is a matching node in front of the cursor. There are two options:
- fun(parser: LuaSnip.extra.TSParser, pos: { [1]: number, [2]: number }): LuaSnip.extra.NamedTSMatch?, TSNode? Manually determine whether there is a matching node that ends just before pos (the beginning of the trigger).
  Return nil,nil if there is no match, otherwise first return a table mapping names to nodes (the text, position and type of these will be provided via snip.env), and second the node that is the matched node.
- LuaSnip.extra.MatchTSNodeOpts, which represents a query and provides all captures of the matched pattern in NamedTSMatch. It contains the following options:
  - query, string: The query, in textual form.
  - query_name, string: The name of the runtime-query to be used (passed to query.get()), defaults to "luasnip" (so one could create a file which only contains queries used by luasnip, like $CONFDIR/queries/<lang>/luasnip.scm, which might make sense to define general concepts independent of a single snippet).
    query and query_name are mutually exclusive, only one of both shall be defined.
  - query_lang, string: The language of the query. This is the only required parameter to this function, since there's no sufficiently straightforward way to determine the language of the query for us. Consider using extend_override to define a ts_postfix-function that automatically fills in the language for the filetype of the snippet-file.
  - match_captures, string|string[]: The capture(s) to use for determining the actual prefix (so the node that should be immediately in front of the trigger). This defaults to just "prefix".
  - select, string?|fun(): LuaSnip.extra.MatchSelector: Since there may be multiple matching captures in front of the cursor, there has to be some way to select the node that will actually be used.
    If this is a string, it has to be one of "any", "shortest", or "longest", which mean that any, the shortest, or the longest match is used.
    If it is a function, it must return a table with two fields, record and retrieve. record is called with a TSMatch and a potential node for the TSMatch, and may return true to abort the selection-procedure. retrieve must return either a TSMatch-TSNode-tuple (which is used as the match) or nil, to signify that there is no match.
    lua/luasnip/extras/_treesitter.lua contains the table builtin_tsnode_selectors, which contains the implementations for any/shortest/longest, which can be used as examples for more complicated custom-selectors.

The text of the matched node can be accessed as snip.env.LS_TSMATCH.
The text of the nodes returned as NamedTSMatch can be accessed as snip.env.LS_TSCAPTURE_<node-name-in-caps>, and their range and type as snip.env.LS_TSDATA.<node-name-NOT-in-caps>.range/type (where range is a tuple of row-col-tuples, both 0-indexed).

For a query like

(function_declaration
  name: (identifier) @fname
  parameters: (parameters) @params
  body: (block) @body
) @prefix

matched against

function add(a, b)
    return a + b
end

snip.env would contain:

LS_TSMATCH: { "function add(a, b)", "\treturn a + b", "end" }

LS_TSDATA:

{
  body = {
    range = { { 1, 1 }, { 1, 13 } },
    type = "block"
  },
  fname = {
    range = { { 0, 9 }, { 0, 12 } },
    type = "identifier"
  },
  params = {
    range = { { 0, 12 }, { 0, 18 } },
    type = "parameters"
  },
  prefix = {
    range = { { 0, 0 }, { 2, 3 } },
    type = "function_declaration"
  }
}

LS_TSCAPTURE_FNAME: { "add" }
LS_TSCAPTURE_PARAMS: { "(a, b)" }
LS_TSCAPTURE_BODY: { "return a + b" }
LS_TSCAPTURE_PREFIX: { "function add(a, b)", "\treturn a + b", "end" }

(note that all variables containing text of nodes are string-arrays, one entry for each line)

There is one important caveat when accessing LS_TSDATA in function/dynamicNodes: It won't contain the values as specified here while generating docstrings (in fact, it won't even be a table).
Since docstrings have to be generated without any runtime-information, we just have to provide dummy-data in env, which will be some kind of string related to the name of the environment variable.
Since the structure of LS_TSDATA obviously does not fit that model, we can't really handle it in a nice way (at least yet). So, for now, best include a check like local static_evaluation = type(env.LS_TSDATA) == "string", and behave accordingly if static_evaluation is true (for example, return some value tailored for displaying it in a docstring).

One more example, which actually uses a few captures:

ts_post({
    matchTSNode = {
        query = [[
            (function_declaration
              name: (identifier) @fname
              parameters: (parameters) @params
              body: (block) @body
            ) @prefix
        ]],
        query_lang = "lua",
    },
    trig = ".var"
}, fmt([[
    local {} = function{}
        {}
    end
]], {
    l(l.LS_TSCAPTURE_FNAME),
    l(l.LS_TSCAPTURE_PARAMS),
    l(l.LS_TSCAPTURE_BODY),
}))

The module luasnip.extras.treesitter_postfix contains a few functions that may be useful for creating more efficient ts-postfix-snippets.
Nested in builtin.tsnode_matcher are:

fun find_topmost_types(types: string[]): MatchTSNodeFunc: Generates a LuaSnip.extra.MatchTSNodeFunc which returns the last parent whose type is in types.
fun find_first_types(types: string[]): MatchTSNodeFunc: Similar to find_topmost_types, only this one matches the first parent whose type is in types.
find_nth_parent(n: number): MatchTSNodeFunc: Simply matches the n-th parent of the innermost node in front of the trigger.

With find_topmost_types, the first example can be implemented more efficiently (without needing a whole query):

local postfix_builtin = require("luasnip.extras.treesitter_postfix").builtin

ls.add_snippets("all", {
	ts_post({
		matchTSNode = postfix_builtin.tsnode_matcher.find_topmost_types({
			"call_expression",
			"identifier",
			"template_function",
			"subscript_expression",
			"field_expression",
			"user_defined_literal"
		}),
		trig = ".mv"
	}, {
		l(l_str.format("std::move(%s)", l.LS_TSMATCH))
	})
}, {key = "asdf"})

Snippet List

local sl = require("luasnip.extras.snippet_list")

Makes an open function available to use to open currently available snippets in a different buffer/window/tab.

sl.open(opts:table|nil)

opts: optional arguments:
- snip_info: snip_info(snippet) -> table representation of snippet
- printer: printer(snippets:table) -> any
- display: display(snippets:any)

Benefits include: syntax highlighting, searching, and customizability.

Simple Example:

sl.open()

Customization Examples:

-- making our own snip_info
local function snip_info(snippet)
	return { name = snippet.name }
end

-- using it
sl.open({snip_info = snip_info})

-- making our own printer
local function printer(snippets)
    local res = ""

    for ft, snips in pairs(snippets) do
        res = res .. ft .. "\n"
        for _, snip in pairs(snips) do
            res = res .. "    " .. "Name: " .. snip.name .. "\n"
            res = res .. "    " .. "Desc: " .. snip.description[1] .. "\n"
            res = res .. "    " .. "Trigger: " .. snip.trigger .. "\n"
            res = res .. "    ----" .. "\n"
        end
    end

    return res
end


-- using it
sl.open({printer = printer})

-- making our own display
local function display(printer_result)
    -- right vertical split
    vim.cmd("botright vnew")

    -- get buf and win handle
    local buf = vim.api.nvim_get_current_buf()
    local win = vim.api.nvim_get_current_win()

    -- setting window and buffer options
    vim.api.nvim_win_set_option(win, "foldmethod", "manual")
    vim.api.nvim_buf_set_option(buf, "filetype", "javascript")

    vim.api.nvim_buf_set_option(buf, "buftype", "nofile")
    vim.api.nvim_buf_set_option(buf, "bufhidden", "wipe")
    vim.api.nvim_buf_set_option(buf, "buflisted", false)

    vim.api.nvim_buf_set_name(buf, "Custom Display buf " .. buf)

    -- dump snippets
    local replacement = vim.split(printer_result)
    vim.api.nvim_buf_set_lines(buf, 0, 0, false, replacement)
end

-- using it
sl.open({display = display})

There is a caveat with implementing your own printer and/or display function. The default behavior for the printer function is to return a string representation of the snippets. The display function uses the results from the printer function, therefore by default the display function is expecting that result to be a string.

However, this doesn't have to be the case. For example, you can implement your own printer function that returns a table representation of the snippets but you would have to then implement your own display function or some other function in order to return the result as a string.

An options table, which has some core functionality that can be used to customize 'common' settings, is provided.

sl.options: options table:
- display: display(opts:table|nil) -> function(printer_result:string)

You can see from the example above that making a custom display is a fairly involved process. What if you just wanted to change a buffer option like the name or just the filetype? This is where sl.options.display comes in. It allows you to customize buffer and window options while keeping the default behavior.

sl.options.display(opts:table|nil) -> function(printer_result:string)

opts: optional arguments:
- win_opts: table which has a {window_option = value} form
- buf_opts: table which has a {buffer_option = value} form
- get_name: get_name(buf) -> string

Let's recreate the custom display example above:

-- keeping the default display behavior but modifying window/buffer
local modified_default_display = sl.options.display({
        buf_opts = {filetype = "javascript"},
        win_opts = {foldmethod = "manual"},
        get_name = function(buf) return "Custom Display buf " .. buf end
    })

-- using it
sl.open({display = modified_default_display})

Snippet Location

This module can consume a snippets source, more specifically, jump to the location referred by it.
This is primarily implemented for snippet which got their source from one of the loaders, but might also work for snippets where the source was set manually.

require("luasnip.extras.snip_location"):

snip_location.jump_to_snippet(snip, opts) Jump to the definition of snip.
- snip: a snippet with attached source-data.
- opts: nil|table, optional arguments, valid keys are:
  - hl_duration_ms: number, duration for which the definition should be highlighted, in milliseconds. 0 disables the highlight.
  - edit_fn: function(file), this function will be called with the file the snippet is located in, and is responsible for jumping to it.
    We assume that after it has returned, the current buffer contains file.
snip_location.jump_to_active_snippet(opts) Jump to definition of active snippet.
- opts: nil|table, accepts the same keys as the opts-parameter of jump_to_snippet.

Extend Decorator

Most of LuaSnip's functions have some arguments to control their behavior.
Examples include s, where wordTrig, regTrig, ... can be set in the first argument to the function, or fmt, where the delimiter can be set in the third argument.
This is all good and well, but if these functions are often used with non-default settings, it can become cumbersome to always explicitly set them.

This is where the extend_decorator comes in: it can be used to create decorated functions which always extend the arguments passed directly with other previously defined ones.

An example:

local fmt = require("luasnip.extras.fmt").fmt

fmt("{}", {i(1)}) -- -> list of nodes, containing just the i(1).

-- when authoring snippets for some filetype where `{` and `}` are common, they
-- would always have to be escaped in the format-string. It might be preferable
-- to use other delimiters, like `<` and `>`.

fmt("<>", {i(1)}, {delimiters = "<>"}) -- -> same as above.

-- but it's quite annoying to always pass the `{delimiters = "<>"}`.

-- with extend_decorator:
local fmt_angle = ls.extend_decorator.apply(fmt, {delimiters = "<>"})
fmt_angle("<>", {i(1)}) -- -> same as above.

-- the same also works with other functions provided by luasnip, for example all
-- node/snippet-constructors and `parse_snippet`.

extend_decorator.apply(fn, ...) requires that fn is previously registered via extend_decorator.register. (This is not limited to LuaSnip's functions; although, for usage outside of LuaSnip, best copy the source file: /lua/luasnip/util/extend_decorator.lua).

register(fn, ...):

fn: the function.
...: any number of tables. Each specifies how to extend an argument of fn. The tables accept:
- arg_indx, number (required): the position of the parameter to override.
- extend, fn(arg, extend_value) -> effective_arg (optional): this function is used to extend the arguments passed to the decorated function. It defaults to a function which just extends the arguments table with the extend table (accepts nil). This extend behavior is adaptable to accommodate s, where the first argument may be string or table.

apply(fn, ...) -> decorated_fn:

fn: the function to decorate.
...: The values to extend with. These should match the descriptions passed in register (the argument first passed to register will be extended with the first value passed here).

One more example for registering a new function:

local function somefn(arg1, arg2, opts1, opts2)
	-- not important
end

-- note the reversed arg_indx!!
extend_decorator.register(somefn, {arg_indx=4}, {arg_indx=3})
local extended = extend_decorator.apply(somefn,
	{key = "opts2 is extended with this"},
	{key = "and opts1 with this"})
extended(...)

LSP-Snippets

LuaSnip is capable of parsing LSP-style snippets using ls.parser.parse_snippet(context, snippet_string, opts):

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

context can be:

string|table: treated like the first argument to ls.s, parse_snippet returns a snippet.
number: parse_snippet returns a snippetNode, with the position context.
nil: parse_snippet returns a flat table of nodes. This can be used like fmt.

Nested placeholders("${1:this is ${2:nested}}") will be turned into choiceNodes with:

the given snippet("this is ${1:nested}") and
an empty insertNode

This behavior can be modified by changing parser_nested_assembler in ls.setup().

LuaSnip will also modify some snippets that it is incapable of representing accurately:

if the $0 is a placeholder with something other than just text inside
if the $0 is a choice
if the $0 is not an immediate child of the snippet (it could be inside a placeholder: "${1: $0 }")

To remedy those incompatibilities, the invalid $0 will be replaced with a tabstop/placeholder/choice which will be visited just before the new $0. This new $0 will be inserted at the (textually) earliest valid position behind the invalid $0.

opts can contain the following keys:

trim_empty: boolean, remove empty lines from the snippet. Default true.
dedent: boolean, remove common indent from the snippet's lines. Default true.

Both trim_empty and dedent will be disabled for snippets parsed via ls.lsp_expand: it might prevent correct expansion of snippets sent by LSP.

SnipMate Parser

It is furthermore possible to parse SnipMate snippets (this includes support for Vim script-evaluation!!)

SnipMate snippets need to be parsed with a different function, ls.parser.parse_snipmate:

ls.parser.parse_snipmate("year", "The year is `strftime('%Y')`")

parse_snipmate accepts the same arguments as parse_snippet, only the snippet body is parsed differently.

Transformations

To apply Variable/Placeholder-transformations, LuaSnip needs to apply ECMAScript regular expressions. This is implemented by relying on jsregexp.

The easiest (but potentially error-prone) way to install it is by calling make install_jsregexp in the repository root.

This process can be automated by packer.nvim:

use { "L3MON4D3/LuaSnip", run = "make install_jsregexp" }

If this fails, first open an issue :P, and then try installing the jsregexp-LuaRock. This is also possible via packer.nvim, although actual usage may require a small workaround, see here or here.

Alternatively, jsregexp can be cloned locally, maked, and the resulting jsregexp.so placed in some place where Neovim can find it (probably ~/.config/nvim/lua/).

If jsregexp is not available, transformations are replaced by a simple copy.

Variables

All TM_something-variables are supported with two additions: LS_SELECT_RAW and LS_SELECT_DEDENT. These were introduced because TM_SELECTED_TEXT is designed to be compatible with VSCode's 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). Besides those we also provide LS_TRIGGER which contains the trigger of the snippet, and LS_CAPTURE_n (where n is a positive integer) that contains the n-th capture when using a regex with capture groups as trig in the snippet definition.

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

s("selected_text", f(function(args, snip)
  local res, env = {}, snip.env
  table.insert(res, "Selected Text (current line is " .. env.TM_LINE_NUMBER .. "):")
  for _, ele in ipairs(env.LS_SELECT_RAW) do table.insert(res, ele) end
  return res
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 visual mode will populate the *SELECT*-vars for the next snippet and then clear them.

Environment Namespaces

You can also add your own variables by using the ls.env_namespace(name, opts) where:

name: string the names the namespace, can't contain the character "_"
opts is a table containing (in every case EnvVal is the same as string|list[string]:
- vars: (fn(name:string)->EnvVal) | map[string, EnvVal] Is a function that receives a string and returns a value for the var with that name or a table from var name to a value (in this case, if the value is a function it will be executed lazily once per snippet expansion).
- init: fn(info: table)->map[string, EnvVal] Returns a table of variables that will set to the environment of the snippet on expansion, use this for vars that have to be calculated in that moment or that depend on each other. The info table argument contains pos (0-based position of the cursor on expansion), the trigger of the snippet and the captures list.
- eager: list[string] names of variables that will be taken from vars and appended eagerly (like those in init)
- multiline_vars: (fn(name:string)->bool)|map[string, bool]|bool|string[] Says if certain vars are a table or just a string, can be a function that get's the name of the var and returns true if the var is a key, a list of vars that are tables or a boolean for the full namespace, it's false by default. Refer to issue#510 for more information.

The four fields of opts are optional but you need to provide either init or vars, and eager can't be without vars. Also, you can't use namespaces that override default vars.

A simple example to make it more clear:

local function random_lang()
    return ({"LUA", "VIML", "VIML9"})[math.floor(math.random()/2 + 1.5)]
end

ls.env_namespace("MY", {vars={ NAME="LuaSnip",  LANG=random_lang }})

-- then you can use  $MY_NAME and $MY_LANG in your snippets

ls.env_namespace("SYS", {vars=os.getenv, eager={"HOME"}})

-- then you can use  $SYS_HOME which was eagerly initialized but also $SYS_USER (or any other system environment var) in your snippets

lsp.env_namespace("POS", {init=function(info) return {VAL=vim.inspect(info.pos)} end})

-- then you can use  $POS_VAL in your snippets

s("custom_env", d(1, function(args, parent)
  local env = parent.snippet.env
  return sn(nil, t {
    "NAME: " .. env.MY_NAME,
    "LANG: " .. env.MY_LANG,
    "HOME: " .. env.SYS_HOME,
    "USER: " .. env.SYS_USER,
    "VAL: " .. env.POS_VAL
  })
end, {}))

LSP-Variables

All variables, even ones added via env_namespace, can be accessed in LSP snippets as $VAR_NAME.

The LSP specification states:

With $name or ${name:default} you can insert the value of a variable.
When a variable isn't set, its default or the empty string is inserted. When a variable is unknown (that is, its name isn't defined) the name of the variable is inserted and it is transformed into a placeholder.

The above necessitates a differentiation between unknown and unset variables:

For LuaSnip, a variable VARNAME is unknown when env.VARNAME returns nil and unset if it returns an empty string.

Consider this when adding environment variables which might be used in LSP snippets.

Loaders

Luasnip is capable of loading snippets from different formats, including both the well-established VSCode and SnipMate format, as well as plain Lua files for snippets written in Lua.

All loaders (except the vscode-standalone-loader) share a similar interface: require("luasnip.loaders.from_{vscode,snipmate,lua}").{lazy_,}load(opts:table|nil)

where opts can contain the following keys:

paths: List of paths to load. Can be a table, or a single comma-separated string. The paths may begin with ~/ or ./ to indicate that the path is relative to your $HOME or to the directory where your $MYVIMRC resides (useful to add your snippets).
If not set, runtimepath is searched for directories that contain snippets. This procedure differs slightly for each loader:
- lua: the snippet-library has to be in a directory named "luasnippets".
- snipmate: similar to Lua, but the directory has to be "snippets".
- vscode: any directory in runtimepath that contains a package.json contributing snippets.
lazy_paths: behaves essentially like paths, with two exceptions: if it is nil, it does not default to runtimepath, and the paths listed here do not need to exist, and will be loaded on creation.
LuaSnip will do its best to determine the path that this should resolve to, but since the resolving we do is not very sophisticated it may produce incorrect paths. Definitely check the log if snippets are not loaded as expected.
exclude: List of languages to exclude, empty by default.
include: List of languages to include, includes everything by default.
{override,default}_priority: These keys are passed straight to the add_snippets-calls (documented in API) and can therefore change the priority of snippets loaded from some collection (or, in combination with {in,ex}clude, only some of its snippets).
fs_event_providers: table<string, boolean>?, specifies which mechanisms should be used to watch files for updates/creation.
If autocmd is set to true, a BufWritePost-hook watches files of this collection, if libuv is set, the file-watcher-api exposed by libuv is used to watch for updates.
Use libuv if you want snippets to update from other Neovim-instances, and autocmd if the collection resides on a file system where the libuv-watchers may not work correctly. Or, of course, just enable both :D
By default, only autocmd is enabled.

While load will immediately load the snippets, lazy_load will defer loading until the snippets are actually needed (whenever a new buffer is created, or the filetype is changed LuaSnip actually loads lazy_loaded snippets for the filetypes associated with this buffer. This association can be changed by customizing load_ft_func in setup: the option takes a function that, passed a bufnr, returns the filetypes that should be loaded (fn(bufnr) -> filetypes (string[]))).

All of the loaders support reloading, so simply editing any file contributing snippets will reload its snippets (according to fs_event_providers in the instance where the file was edited, or in other instances as well).

As an alternative (or addition) to automatic reloading, LuaSnip can also process manual updates to files: Call require("luasnip.loaders").reload_file(path) to reload the file at path.
This may be useful when the collection is controlled by some other plugin, or when enabling the other reload-mechanisms is for some reason undesirable (performance? minimalism?).

For easy editing of these files, LuaSnip provides a vim.ui.select-based dialog (Loaders-edit_snippets) where first the filetype, and then the file can be selected.

Snippet-specific filetypes

Some loaders (vscode,lua) support giving snippets generated in some file their own filetype (vscode via scope, lua via the underlying filetype-option for snippets). These snippet-specific filetypes are not considered when determining which files to lazy_load for some filetype, this is exclusively determined by the language associated with a file in vscodes' package.json, and the file/directory-name in lua.

This can be resolved relatively easily in vscode, where the language advertised in package.json can just be a superset of the scopes in the file.
Another simplistic solution is to set the language to all (in lua, it might make sense to create a directory luasnippets/all/*.lua to group these files together).
Another approach is to modify load_ft_func to load a custom filetype if the snippets should be activated, and store the snippets in a file for that filetype. This can be used to group snippets by e.g. framework, and load them once a file belonging to such a framework is edited.

Example:
react.lua

return {
    s({filetype = "css", trig = ...}, ...),
    s({filetype = "html", trig = ...}, ...),
    s({filetype = "js", trig = ...}, ...),
}

luasnip_config.lua

load_ft_func = function(bufnr)
    if "<bufnr-in-react-framework>" then
        -- will load `react.lua` for this buffer
        return {"react"}
    else
        return require("luasnip.extras.filetype_functions").from_filetype_load
    end
end

See the Troubleshooting-Adding Snippets-Loaders section if one is having issues adding snippets via loaders.

VS-Code

As a reference on the structure of these snippet libraries, see friendly-snippets.

We support a small extension: snippets can contain LuaSnip-specific options in the luasnip-table:

"example1": {
	"prefix": "options",
	"body": [
		"whoa! :O"
	],
	"luasnip": {
		"priority": 2000,
		"autotrigger": true,
		"wordTrig": false
	}
}

Files with the extension jsonc will be parsed as jsonc, json with comments, while *.json are parsed with a regular json parser, where comments are disallowed. (the json parser is a bit faster, so don't default to jsonc if it's not necessary).

Example:

~/.config/nvim/my_snippets/package.json:

{
	"name": "example-snippets",
	"contributes": {
		"snippets": [
			{
				"language": [
					"all"
				],
				"path": "./snippets/all.json"
			},
			{
				"language": [
					"lua"
				],
				"path": "./lua.json"
			}
		]
	}
}

~/.config/nvim/my_snippets/snippets/all.json:

{
	"snip1": {
		"prefix": "all1",
		"body": [
			"expands? jumps? $1 $2 !"
		]
	},
	"snip2": {
		"prefix": "all2",
		"body": [
			"multi $1",
			"line $2",
			"snippet$0"
		]
	}
}

~/.config/nvim/my_snippets/lua.json:

{
	"snip1": {
		"prefix": "lua",
		"body": [
			"lualualua"
		]
	}
}

This collection can be loaded with any of

-- don't pass any arguments, luasnip will find the collection because it is
-- (probably) in rtp.
require("luasnip.loaders.from_vscode").lazy_load()
-- specify the full path...
require("luasnip.loaders.from_vscode").lazy_load({paths = "~/.config/nvim/my_snippets"})
-- or relative to the directory of $MYVIMRC
require("luasnip.loaders.from_vscode").load({paths = "./my_snippets"})

Standalone

Beside snippet-libraries provided by packages, vscode also supports another format which can be used for project-local snippets, or user-defined snippets, .code-snippets.

The layout of these files is almost identical to that of the package-provided snippets, but there is one additional field supported in the snippet-definitions, scope, with which the filetype of the snippet can be set. If scope is not set, the snippet will be added to the global filetype (all).

require("luasnip.loaders.from_vscode").load_standalone(opts)

opts: table, can contain the following keys:
- path: string, Path to the *.code-snippets-file that should be loaded. Just like the paths in load, this one can begin with a "~/" to be relative to $HOME, and a "./" to be relative to the Neovim config directory.
- {override,default}_priority: These keys are passed straight to the add_snippets-calls (documented in API) and can be used to change the priority of the loaded snippets.
- lazy: boolean, if it is set, the file does not have to exist when load_standalone is called, and it will be loaded on creation.
  false by default.

Example: a.code-snippets:

{
    // a comment, since `.code-snippets` may contain jsonc.
    "c/cpp-snippet": {
        "prefix": [
            "trigger1",
            "trigger2"
        ],
        "body": [
            "this is $1",
            "my snippet $2"
        ],
        "description": "A description of the snippet.",
        "scope": "c,cpp"
    },
    "python-snippet": {
        "prefix": "trig",
        "body": [
            "this is $1",
            "a different snippet $2"
        ],
        "description": "Another snippet-description.",
        "scope": "python"
    },
    "global snippet": {
        "prefix": "trigg",
        "body": [
            "this is $1",
            "the last snippet $2"
        ],
        "description": "One last snippet-description.",
    }
}

This file can be loaded by calling

require("luasnip.loaders.from_vscode").load_standalone({path = "a.code-snippets"})

SNIPMATE

Luasnip does not support the full SnipMate format: Only ./{ft}.snippets and ./{ft}/*.snippets will be loaded. See honza/vim-snippets for lots of examples.

Like VSCode, the SnipMate format is also extended to make use of some of LuaSnip's more advanced capabilities:

priority 2000
autosnippet options
	whoa :O

Example:

~/.config/nvim/snippets/c.snippets:

# this is a comment
snippet c c-snippet
	c!

~/.config/nvim/snippets/cpp.snippets:

extends c

snippet cpp cpp-snippet
	cpp!

This can, again, be loaded with any of

require("luasnip.loaders.from_snipmate").load()
-- specify the full path...
require("luasnip.loaders.from_snipmate").lazy_load({paths = "~/.config/nvim/snippets"})
-- or relative to the directory of $MYVIMRC
require("luasnip.loaders.from_snipmate").lazy_load({paths = "./snippets"})

Stuff to watch out for:

Using both extends <ft2> in <ft1>.snippets and ls.filetype_extend("<ft1>", {"<ft2>"}) leads to duplicate snippets.
${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.

Lua

Instead of adding all snippets via add_snippets, it's possible to store them in separate files and load all of those.
The file-structure here is exactly the supported SnipMate-structure, e.g. <ft>.lua or <ft>/*.lua to add snippets for the filetype <ft>.

There are two ways to add snippets:

the files may return two lists of snippets, the snippets in the first are all added as regular snippets, while the snippets in the second will be added as autosnippets (both are the defaults, if a snippet defines a different snippetType, that will have preference)

snippets can also be appended to the global (only for these files - they are not visible anywhere else) tables ls_file_snippets and ls_file_autosnippets. This can be combined with a custom snip_env to define and add snippets with one function call:

ls.setup({
	snip_env = {
		s = function(...)
			local snip = ls.s(...)
			-- we can't just access the global `ls_file_snippets`, since it will be
			-- resolved in the environment of the scope in which it was defined.
			table.insert(getfenv(2).ls_file_snippets, snip)
		end,
		parse = function(...)
			local snip = ls.parser.parse_snippet(...)
			table.insert(getfenv(2).ls_file_snippets, snip)
		end,
		-- remaining definitions.
		...
	},
	...
})

This is more flexible than the previous approach since the snippets don't have to be collected; they just have to be defined using the above s and parse.

As defining all of the snippet constructors (s, c, t, ...) in every file is rather cumbersome, LuaSnip will bring some globals into scope for executing these files. By default, the names from luasnip.config.snip_env will be used, but it's possible to customize them by setting snip_env in setup.

Example:

~/snippets/all.lua:

return {
	s("trig", t("loaded!!"))
}

~/snippets/c.lua:

return {
	s("ctrig", t("also loaded!!"))
}, {
	s("autotrig", t("autotriggered, if enabled"))
}

Load via

require("luasnip.loaders.from_lua").load({paths = "~/snippets"})

Reloading when editing `require`'d files

While the lua-snippet-files will be reloaded on edit, this does not automatically happen if a file the snippet-file depends on (e.g. via require) is changed.
Since this still may still be desirable, there are two functions exposed when a file is loaded by the lua-loader: ls_tracked_dofile and ls_tracked_dopackage. They perform like dofile and (almost like) require, but both register the loaded file internally as a dependency of the snippet-file, so it can be reloaded when the loaded file is edited. As stated, ls_tracked_dofile behaves exactly like dofile, but does the dependency-work as well.
ls_tracked_dopackage mimics require in that it does not take a path, but a module-name like "luasnip.loaders.from_lua", and then searches the runtimepath/lua-directories, and path and cpath for the module.
Unlike require, the file will not be cached, since that would complicate the reload-on-edit-behavior.

edit_snippets

To easily edit snippets for the current session, the files loaded by any loader can be quickly edited via require("luasnip.loaders").edit_snippet_files(opts:table|nil)

When called, it will open a vim.ui.select-dialog to select first a filetype, and then (if there are multiple) the associated file to edit.

opts contains four settings:

ft_filter: fn(filetype:string) -> bool Optionally filter initially listed filetypes. true -> filetype will be listed, false -> not listed. Accepts all filetypes by default.
format: fn(file:string, source_name:string) -> string|nil
file is simply the path to the file, source_name is one of "lua", "snipmate" or "vscode".
If a string is returned, it is used as the title of the item, nil on the other hand will filter out this item.
The default simply replaces some long strings (packer-path and config-path) in file with shorter, symbolic names ("$PLUGINS", "$CONFIG"), but this can be extended to
- filter files from some specific source/path
- more aggressively shorten paths using symbolic names, e.g. "$FRIENDLY_SNIPPETS".
  Example: hide the *.lua snippet files, and shorten the path with $LuaSnip:
```
require "luasnip.loaders" .edit_snippet_files {
  format = function(file, source_name)
    if source_name == "lua" then return nil
    else return file:gsub("/root/.config/nvim/luasnippets", "$LuaSnip")
    end
  end
}
```
edit: fn(file:string) This function is supposed to open the file for editing. The default is a simple vim.cmd("edit " .. file) (replace the current buffer), but one could open the file in a split, a tab, or a floating window, for example.
extend: fn(ft:string, ft_paths:string[]) -> (string,string)[]
This function can be used to create additional choices for the file-selection.
- ft: The filetype snippet-files are queried for.
- ft_paths: list of paths to the known snippet files.
The function should return a list of (string,string)-tuples. The first of each pair is the label that will appear in the selection-prompt, and the second is the path that will be passed to the edit() function if that item was selected.

This can be used to create a new snippet file for the current filetype:

require("luasnip.loaders").edit_snippet_files {
  extend = function(ft, paths)
    if #paths == 0 then
      return {
        { "$CONFIG/" .. ft .. ".snippets",
          string.format("%s/%s.snippets", <PERSONAL_SNIPPETS_FOLDER>, ft) }
      }
    end

    return {}
  end
}

One comfortable way to call this function is registering it as a command:

command! LuaSnipEdit :lua require("luasnip.loaders").edit_snippet_files()

SnippetProxy

SnippetProxy is used internally to alleviate the upfront cost of loading snippets from e.g. a SnipMate library or a VSCode package. This is achieved by only parsing the snippet on expansion, not immediately after reading it from some file. SnippetProxy may also be used from Lua directly to get the same benefits:

This will parse the snippet on startup:

ls.parser.parse_snippet("trig", "a snippet $1!")

while this will parse the snippet upon expansion:

local sp = require("luasnip.nodes.snippetProxy")
sp("trig", "a snippet $1")

sp(context, body, opts) -> snippetProxy

context: exactly the same as the first argument passed to ls.s.
body: the snippet body.
opts: accepts the same opts as ls.s, with some additions:
- parse_fn: the function for parsing the snippet. Defaults to ls.parser.parse_snippet (the parser for LSP snippets), an alternative is the parser for SnipMate snippets (ls.parser.parse_snipmate).

ext_opts

ext_opts can be used to set the opts (see nvim_buf_set_extmark) of the extmarks used for marking node positions, either globally, per snippet or per node. This means that they allow highlighting the text inside of nodes, or adding virtual text to the line the node begins on.

This is an example for the node_ext_opts used to set ext_opts of single nodes:

local ext_opts = {
	-- these ext_opts are applied when the node is active (e.g. it has been
	-- jumped into, and not out yet).
	active = 
	-- this is the table actually passed to `nvim_buf_set_extmark`.
	{
		-- highlight the text inside the node red.
		hl_group = "GruvboxRed"
	},
	-- these ext_opts are applied when the node is not active, but
	-- the snippet still is.
	passive = {
		-- add virtual text on the line of the node, behind all text.
		virt_text = {{"virtual text!!", "GruvboxBlue"}}
	},
	-- visited or unvisited are applied when a node was/was not jumped into.
	visited = {
		hl_group = "GruvboxBlue"
	},
	unvisited = {
		hl_group = "GruvboxGreen"
	},
	-- and these are applied when both the node and the snippet are inactive.
	snippet_passive = {}
}

s("trig", {
	i(1, "text1", {
		node_ext_opts = ext_opts
	}),
	i(2, "text2", {
		node_ext_opts = ext_opts
	})
})

In the above example, the text inside the insertNodes is highlighted in green if they were not yet visited, in blue once they were, and red while they are.
The virtual text "virtual text!!" is visible as long as the snippet is active.

To make defining ext_opts less verbose, more specific states inherit from less specific ones:

passive inherits from snippet_passive
visited and unvisited from passive
active from visited

flowchart TD
	visited --> active
	passive --> visited
	passive --> unvisited
	snippet_passive --> passive

Loading

To disable a key from a less specific state, it has to be explicitly set to its default, e.g. to disable highlighting inherited from passive when the node is active, hl_group should be set to None.

As stated earlier, these ext_opts can also be applied globally or for an entire snippet. For this, it's necessary to specify which kind of node a given set of ext_opts should be applied to:

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

ls.setup({
	ext_opts = {
		[types.insertNode] = {
			active = {...},
			visited = {...},
			passive = {...},
			snippet_passive = {...}
		},
		[types.choiceNode] = {
			active = {...},
			unvisited = {...}
		},
		[types.snippet] = {
			passive = {...}
		}
	}
})

The above applies the given ext_opts to all nodes of these types, in all snippets.

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

s("trig", { i(1, "text1"), i(2, "text2") }, {
	child_ext_opts = {
		[types.insertNode] = {
			passive = {
				hl_group = "GruvboxAqua"
			}
		}
	}
})

However, the ext_opts here are only applied to the insertNodes inside this snippet.

By default, the ext_opts actually used for a node are created by extending the node_ext_opts with the effective_child_ext_opts[node.type] of the parent, which are in turn the parent's child_ext_opts extended with the global ext_opts (those set ls.setup).

It's possible to prevent both of these merges by passing merge_node/child_ext_opts=false to the snippet/node-opts:

ls.setup({
	ext_opts = {
		[types.insertNode] = {
			active = {...}
		}
	}
})

s("trig", {
	i(1, "text1", {
		node_ext_opts = {
			active = {...}
		},
		merge_node_ext_opts = false
	}),
	i(2, "text2")
}, {
	child_ext_opts = {
		[types.insertNode] = {
			passive = {...}
		}
	},
	merge_child_ext_opts = false
})

The hl_group of the global ext_opts can also be set via standard highlight groups:

vim.cmd("hi link LuasnipInsertNodePassive GruvboxRed")
vim.cmd("hi link LuasnipSnippetPassive GruvboxBlue")

-- needs to be called for resolving the effective ext_opts.
ls.setup({})

The names for the used highlight groups are "Luasnip<node>{Passive,Active,SnippetPassive}", where <node> can be any kind of node in PascalCase (or "Snippet").

One problem that might arise when nested nodes are highlighted is that the highlight of inner nodes should be visible, e.g. above that of nodes they are nested inside.

This can be controlled using the priority-key in ext_opts. In nvim_buf_set_extmark, that value is an absolute value, but here it is relative to some base-priority, which is increased for each nesting level of snippet(Nodes)s.

Both the initial base-priority and its' increase and can be controlled using ext_base_prio and ext_prio_increase:

ls.setup({
	ext_opts = {
		[types.insertNode] = {
			active = {
				hl_group = "GruvboxBlue",
				-- the priorities should be \in [0, ext_prio_increase).
				priority = 1
			}
		},
		[types.choiceNode] = {
			active = {
				hl_group = "GruvboxRed"
				-- priority defaults to 0
			}
		}
	}
	ext_base_prio = 200,
	ext_prio_increase = 2
})

Here the highlight of an insertNode nested directly inside a choiceNode is always visible on top of it.

Docstrings

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 e.g. 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 e.g. 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 e.g. 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, {})
})

Refer to #515 for a better example to understand docTrig and docstring.

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(snippets) and its counterpart ls.load_snippet_docstrings(snippets), they may be serialized from or deserialized into the snippets. Both functions accept a table structured like this: {ft1={snippets}, ft2={snippets}}. Such a table containing all snippets can be obtained via ls.get_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

Events can be used to react to some action inside snippets. These callbacks can be defined per snippet (callbacks-key in snippet constructor), per-node by passing them as node_callbacks in node_opts, or globally (autocommand).

callbacks: fn(node[, event_args]) -> event_res
All callbacks receive the node associated with the event and event-specific optional arguments, event_args. event_res is only used in one event, pre_expand, where some properties of the snippet can be changed. If multiple callbacks return event_res, we only guarantee that one of them will be effective, not all of them.

autocommand: Luasnip uses User-events. Autocommands for these can be registered using

au User SomeUserEvent echom "SomeUserEvent was triggered"

or

vim.api.nvim_create_autocommand("User", {
	pattern = "SomeUserEvent",
	command = "echom SomeUserEvent was triggered"
})

The node and event_args can be accessed through require("luasnip").session:

node: session.event_node
event_args: session.event_args

Events:

enter/leave: Called when a node is entered/left (for example when jumping around in a snippet).
User-event: "Luasnip<Node>{Enter,Leave}", with <Node> in PascalCase, e.g. InsertNode or DynamicNode.
event_args: none
change_choice: When the active choice in a choiceNode is changed.
User-event: "LuasnipChangeChoice"
event_args: none
pre_expand: Called before a snippet is expanded. Modifying text is allowed, the expand-position will be adjusted so the snippet expands at the same position relative to existing text.
User-event: "LuasnipPreExpand"
event_args:
- expand_pos: {<row>, <column>}, position at which the snippet will be expanded. <row> and <column> are both 0-indexed.
- expand_pos_mark_id: number, the id of the extmark LuaSnip uses to track expand_pos. This may be moved around freely. event_res:
- env_override: map string->(string[]|string), override or extend the snippet's environment (snip.env).

A pretty useless, beyond serving as an example here, application of these would be printing e.g. the node's text after entering:

vim.api.nvim_create_autocmd("User", {
	pattern = "LuasnipInsertNodeEnter",
	callback = function()
		local node = require("luasnip").session.event_node
		print(table.concat(node:get_text(), "\n"))
	end
})

or some information about expansions:

vim.api.nvim_create_autocmd("User", {
	pattern = "LuasnipPreExpand",
	callback = function()
		-- get event-parameters from `session`.
		local snippet = require("luasnip").session.event_node
		local expand_position =
			require("luasnip").session.event_args.expand_pos

		print(string.format("expanding snippet %s at %s:%s",
			table.concat(snippet:get_docstring(), "\n"),
			expand_position[1],
			expand_position[2]
		))
	end
})

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.

Logging

Luasnip uses logging to report unexpected program states, and information on what's going on in general. If something does not work as expected, taking a look at the log (and potentially increasing the log level) might give some good hints towards what is going wrong.

The log is stored in <vim.fn.stdpath("log")>/luasnip.log (<vim.fn.stdpath("cache")>/luasnip.log for Neovim versions where stdpath("log") does not exist), and can be opened by calling ls.log.open(). You can get the log path through ls.log.log_location(). The log level (granularity of reported events) can be adjusted by calling ls.log.set_loglevel("error"|"warn"|"info"|"debug"). "debug" has the highest granularity, "error" the lowest, the default is "warn". You can also adjust the datetime formatting through the ls.log.time_fmt variable. By default, it uses the '%X' formatting, which results in the full time (hour, minutes and seconds) being shown.

Once this log grows too large (10MiB, currently not adjustable), it will be renamed to luasnip.log.old, and a new, empty log created in its place. If there already exists a luasnip.log.old, it will be deleted.

ls.log.ping() can be used to verify the log is working correctly: it will print a short message to the log.

Source

It is possible to attach, to a snippet, information about its source. This can be done either by the various loaders (if it is enabled in ls.setup (Config-Options, loaders_store_source)), or manually. The attached data can be used by Extras-Snippet-Location to jump to the definition of a snippet.

It is also possible to get/set the source of a snippet via API:

ls.snippet_source:

get(snippet) -> source_data: Retrieve the source-data of snippet. source_data always contains the key file, the file in which the snippet was defined, and may additionally contain line or line_end, the first and last line of the definition.
set(snippet, source): Set the source of a snippet.
- snippet: a snippet which was added via ls.add_snippets.
- source: a source-object, obtained from either from_debuginfo or from_location.
from_location(file, opts) -> source:
- file: string, The path to the file in which the snippet is defined.
- opts: table|nil, optional parameters for the source.
  - line: number, the first line of the definition. 1-indexed.
  - line_end: number, the final line of the definition. 1-indexed.
from_debuginfo(debuginfo) -> source: Generates source from the table returned by debug.getinfo (from now on referred to as debuginfo). debuginfo has to be of a frame of a function which is backed by a file, and has to contain this information, i.e. has to be generated by debug.get_info(*, "Sl") (at least "Sl", it may also contain more info).

Selection

Many snippets use the $TM_SELECTED_TEXT or (for LuaSnip, preferably LS_SELECT_RAW or LS_SELECT_DEDENT) variable, which has to be populated by selecting and then yanking (and usually also cutting) text from the buffer before expanding.

By default, this is disabled (as to not pollute keybindings which may be used for something else), so one has to

either set cut_selection_keys in setup (see Config-Options).
or map ls.cut_keys as the right-hand-side of a mapping
or manually configure the keybinding. For this, create a new keybinding that
1. <Esc>es to NORMAL (to populate the < and >-markers)
2. calls luasnip.pre_yank(<namedreg>)
3. yanks text to some named register <namedreg>
4. calls luasnip.post_yank(<namedreg>) Take care that the yanking actually takes place between the two calls. One way to ensure this is to call the two functions via <cmd>lua ...<Cr>:
```
vim.keymap.set("v", "<Tab>", [[<Esc><cmd>lua require("luasnip.util.select").pre_yank("z")<Cr>gv"zs<cmd>lua require('luasnip.util.select').post_yank("z")<Cr>]])
```
The reason for this specific order is to allow us to take a snapshot of registers (in the pre-callback), and then restore them (in the post-callback) (so that we may get the visual selection directly from the register, which seems to be the most foolproof way of doing this).

Config-Options

These are the settings you can provide to luasnip.setup():

keep_roots: Whether snippet-roots should be linked. See Basics-Snippet-Insertion for more context.
link_roots: Whether snippet-roots should be linked. See Basics-Snippet-Insertion for more context.
exit_roots: Whether snippet-roots should exit at reaching at their last node, $0. This setting is only valid for root snippets, not child snippets. This setting may avoid unexpected behavior by disallowing to jump earlier (finished) snippets. Check Basics-Snippet-Insertion for more information on snippet-roots.
link_children: Whether children should be linked. See Basics-Snippet-Insertion for more context.
history (deprecated): if not nil, keep_roots, link_roots, and link_children will be set to the value of history, and exit_roots will set to inverse value of history. This is just to ensure backwards-compatibility.
update_events: Choose which events trigger an update of the active nodes' dependents. Default is just 'InsertLeave', 'TextChanged,TextChangedI' would update on every change. These, like all other *_events are passed to nvim_create_autocmd as events, so they can be wrapped in a table, like
```
ls.setup({
	update_events = {"TextChanged", "TextChangedI"}
})
```
region_check_events: Events on which to leave the current snippet-root if the cursor is outside its' 'region'. Disabled by default, 'CursorMoved', 'CursorHold' or 'InsertEnter' seem reasonable.
delete_check_events: When to check if the current snippet was deleted, and if so, remove it from the history. Off by default, 'TextChanged' (perhaps 'InsertLeave', to react to changes done in Insert mode) should work just fine (alternatively, this can also be mapped using <Plug>luasnip-delete-check).
cut_selection_keys: Mapping for populating TM_SELECTED_TEXT and related variables (not set by default).
See Selection for more information.
store_selection_keys (deprecated): same as cut_selection_keys
enable_autosnippets: Autosnippets are disabled by default to minimize performance penalty if unused. Set to true to enable.
ext_opts: Additional options passed to extmarks. Can be used to add passive/active highlight on a per-node-basis (more info in DOC.md)
parser_nested_assembler: Override the default behavior of inserting a choiceNode containing the nested snippet and an empty insertNode for nested placeholders ("${1: ${2: this is nested}}"). For an example (behavior more similar to VSCode), check here
ft_func: Source of possible filetypes for snippets. Defaults to a function, which returns vim.split(vim.bo.filetype, ".", true), but check filetype_functions or the Extras-Filetype-Functions-section for more options.
load_ft_func: Function to determine which filetypes belong to a given buffer (used for lazy_loading). fn(bufnr) -> filetypes (string[]). Again, there are some examples in filetype_functions.
snip_env: The best way to author snippets in Lua involves the lua-loader (see Loaders-Lua). Unfortunately, this requires that snippets are defined in separate files, which means that common definitions like s, i, sn, t, fmt, ... have to be repeated in each of them, and that adding more customized functions to ease writing snippets also requires some setup.
snip_env can be used to insert variables into exactly the places where lua-snippets are defined (for now only the file loaded by the lua-loader).
Setting snip_env to { some_global = "a value" } will add (amongst the defaults stated at the beginning of this documentation) the global variable some_global while evaluating these files.
There are special keys which, when set in snip_env change the behavior of this option, and are not passed through to the lua-files:
- __snip_env_behaviour, string: either "set" or "extend" (default "extend")
  If this is "extend", the variables defined in snip_env will complement (and override) the defaults. If this is not desired, "set" will not include the defaults, but only the variables set here.
One side-effect of this is that analysis-tools (most likely lua-language-server) for Lua will generate diagnostics for the usage of undefined symbols. If you mind the (probably) large number of generated warnings, consider adding the undefined globals to the globals recognized by lua-language-server or add ---@diagnostic disable: undefined-global somewhere in the affected files.
loaders_store_source, boolean, whether loaders should store the source of the loaded snippets.
Enabling this means that the definition of any snippet can be jumped to via Extras-Snippet-Location, but also entails slightly increased memory consumption (and load-time, but it's not really noticeable).

Troubleshooting

Adding Snippets

Loaders

Filetypes. LuaSnip uses all as the global filetype. As most snippet collections don't explicitly target LuaSnip, they may not provide global snippets for this filetype, but another, like _ (honza/vim-snippets). In these cases, it's necessary to extend LuaSnip's global filetype with the collection's global filetype:
```
ls.filetype_extend("all", { "_" })
```
In general, if some snippets don't show up when loading a collection, a good first step is checking the filetype LuaSnip is actually looking into (print them for the current buffer via :lua print(vim.inspect(require("luasnip").get_snippet_filetypes()))), against the one the missing snippet is provided for (in the collection).
If there is indeed a mismatch, filetype_extend can be used to also search the collection's filetype:
```
ls.filetype_extend("<luasnip-filetype>", { "<collection-filetype>" })
```
Non-default ft_func loading. As we only load lazy_loaded snippets on some events, lazy_load will probably not play nice when a non-default ft_func is used: if it depends on e.g. the cursor position, only the filetypes for the cursor position when the lazy_load events are triggered will be loaded. Check Extras-Filetype-Function's extend_load_ft for a solution.

General

Snippets sharing triggers. If multiple snippets could be triggered at the current buffer-position, the snippet that was defined first in one's configuration will be expanded first. As a small, real-world LaTeX math example, given the following two snippets with triggers .ov and ov:
```
postfix( -- Insert over-line command to text via post-fix
    { trig = ".ov", snippetType = "autosnippet" },
    {
        f(function(_, parent)
            return "\\overline{" .. parent.snippet.env.POSTFIX_MATCH .. "}"
        end, {}),
    }
),
s( -- Insert over-line command
    { trig = "ov", snippetType="autosnippet" },
    fmt(
        [[\overline{<>}]],
        { i(1) },
        { delimiters = "<>" }
    )
),
```
If one types x followed by .ov, the postfix snippet expands producing \overline{x}. However, if the postfix snippet above is defined after the normal snippet s, then the same key press sequence produces x.\overline{}. This behavior can be overridden by explicitly providing a priority to such snippets. For example, in the above code, if the postfix snippet was defined after the normal snippet s, then adding priority=1001 to the postfix snippet will cause it to expand as if it were defined before the normal snippet s. Snippet priority is discussed in the Snippets section of the documentation.

API

require("luasnip"):

add_snippets(ft:string or nil, snippets:list or table, opts:table or nil): Makes snippets (list of snippets) available in ft.
If ft is nil, snippets should be a table containing lists of snippets, the keys are corresponding filetypes.
opts may contain the following keys:
- type: type of snippets, "snippets" or "autosnippets" (ATTENTION: plural form used here). This serves as default value for the snippetType key of each snippet added by this call see Snippets.
- key: Key that identifies snippets added via this call.
  If add_snippets is called with a key that was already used, the snippets from that previous call will be removed.
  This can be used to reload snippets: pass an unique key to each add_snippets and just redo the add_snippets-call when the snippets have changed.
- override_priority: set priority for all snippets.
- default_priority: set priority only for snippets without snippet priority.
clean_invalidated(opts: table or nil) -> bool: clean invalidated snippets from internal snippet storage.
Invalidated snippets are still stored; it might be useful to actually remove them as they still have to be iterated during expansion.

opts may contain:
- inv_limit: how many invalidated snippets are allowed. If the number of invalid snippets doesn't exceed this threshold, they are not yet cleaned up.
  
  A small number of invalidated snippets (<100) probably doesn't affect runtime at all, whereas recreating the internal snippet storage might.
get_id_snippet(id): returns snippet corresponding to id.
in_snippet(): returns true if the cursor is inside the current snippet.
jumpable(direction): returns true if the current node has a next(direction = 1) or previous(direction = -1), e.g. 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(opts): expands the snippet at(before) the cursor. opts may contain:
- jump_into_func passed through to ls.snip_expand, check its' doc for a description.
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.
locally_jumpable(direction): same as jumpable() except it is ignored if the cursor is not inside the current snippet.
expand_or_jump(): returns true if jump/expand was successful.
expand_auto(): expands the autosnippets before the cursor (not necessary to call manually, will be called via autocmd if enable_autosnippets 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 its 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, override trigger, captures or environment of the snippet.
  This is useful for manually expanding snippets where the trigger passed via trig is not the text triggering the snippet, or those which expect captures (basically, snippets with a non-plaintext trigEngine).
  
  One example:
```
snip_expand(snip, {
	trigger = "override_trigger",
	captures = {"first capture", "second capture"},
	env_override = { this_key = "some value", other_key = {"multiple", "lines"}, TM_FILENAME = "some_other_filename.lua" }
})
```
- 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.
- jump_into_func: fn(snippet) -> node: Callback responsible for jumping into the snippet. The returned node is set as the new active node, i.e. it is the origin of the next jump. The default is basically this:
```
function(snip)
	-- jump_into set the placeholder of the snippet, 1
	-- to jump forwards.
	return snip:jump_into(1)
```
  while this can be used to only insert the snippet:
```
function(snip)
	return snip.insert_nodes[0]
end
```
- indent: bool?, defaults to true. Whether LuaSnip will try to add additional indents to fit current indent level in snippet expanding. This option is useful when some LSP server already take indents into consideration. In such cases, LuaSnip should not try to add additional indents. If you are using nvim-cmp, sample config:
```
require("cmp").setup {
  snippet = {
    expand = function(args)
      local indent_nodes = true
      if vim.api.nvim_get_option_value("filetype", { buf = 0 }) == "dart" then
        indent_nodes = false
      end
      require("luasnip").lsp_expand(args.body, {
        indent = indent_nodes,
      })
    end,
  },
}
```
opts and any of its parameters 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 e.g. deletion of a snippet) and sets the current node behind the snippet, or, if not possible, before it.
lsp_expand(snip_string, opts): expands the LSP 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(snip_info): returns a table of all snippets defined for the current filetypes(s) ({ft1={snip1, snip2}, ft2={snip3, snip4}}). The structure of the snippet is defined by snip_info which is a function (snip_info(snip)) that takes in a snippet (snip), finds the desired information on it, and returns it. snip_info is an optional argument as a default has already been defined. You can use it for more granular control over the table of snippets that is returned.
exit_out_of_region(node): checks whether the cursor is still within the range of the root-snippet node belongs to. If yes, no change occurs; if no, the root-snippet is exited and its $0 will be the new active node.
If a jump causes an error (happens mostly because the text of a snippet was deleted), the snippet is removed from the jumplist and the current node set to the end/beginning of the next/previous snippet.
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 avoid 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, e.g. 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 beginning/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 C++ 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.
set_choice(indx:number): Changes to the indxth choice. If no choiceNode is active, an error is thrown. If the active choiceNode doesn't have an indxth choice, an error is thrown.
get_current_choices() -> string[]: Returns a list of multiline-strings (themselves lists, even if they have only one line), the ith string corresponding to the ith choice of the currently active choiceNode. If no choiceNode is active, an error is thrown.
setup_snip_env(): Adds the variables defined (during setup) in snip_env to the callers environment.
get_snip_env(): Returns snip_env.
jump_destination(direction): Returns the node the next jump in direction (either -1 or 1, for backwards, forwards respectively) leads to, or nil if the destination could not be determined (most likely because there is no node that can be jumped to in the given direction, or there is no active node).
activate_node(opts): Activate a node in any snippet. opts contains the following options:
- pos, {[1]: row, [2]: byte-column}?: The position at which a node should be activated. Defaults to the position of the cursor.
- strict, bool?: If set, throw an error if the node under the cursor can't be jumped into. If not set, fall back to any node of the snippet and enter that instead.
- select, bool?: Whether the text inside the node should be selected. Defaults to true.

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 repository).

Files

DOC.md

Latest commit

History

DOC.md

File metadata and controls

Basics

Jump-Index

Adding Snippets

Snippet Insertion

Node

API

Snippets

Data

API

TextNode

InsertNode

FunctionNode

Node Reference

ChoiceNode

SnippetNode

IndentSnippetNode

DynamicNode

RestoreNode

Key Indexer

Absolute Indexer

MultiSnippet

Extras

Lambda

Match

Repeat

Partial

Nonempty

Dynamic Lambda

fmt

Conditions

Condition Objects

On The Fly-Snippets

select_choice

Filetype-Functions

Postfix-Snippet

Treesitter-Postfix-Snippet

Snippet List

Snippet Location

Extend Decorator

LSP-Snippets

SnipMate Parser

Transformations

Variables

Environment Namespaces

LSP-Variables

Loaders

Snippet-specific filetypes

VS-Code

Standalone

SNIPMATE

Lua

Reloading when editing require'd files

edit_snippets

SnippetProxy

ext_opts

Docstrings

Docstring-Cache

Events

Cleanup

Logging

Source

Selection

Config-Options

Troubleshooting

Adding Snippets

Loaders

General

API

`fmt`

Reloading when editing `require`'d files