mini-snippets.txt

*mini.snippets* Manage and expand snippets
*MiniSnippets*

MIT License Copyright (c) 2024 Evgeni Chasnovski

==============================================================================

Snippet is a template for a frequently used text. Typical workflow is to type
snippet's (configurable) prefix and expand it into a snippet session.

The template usually contains both pre-defined text and places (called
"tabstops") for user to interactively change/add text during snippet session.

This module supports (only) snippet syntax defined in LSP specification (with
small deviations). See |MiniSnippets-syntax-specification|.

Features:
- Manage snippet collection by adding it explicitly or with a flexible set of
  performant built-in loaders. See |MiniSnippets.gen_loader|.

- Configured snippets are efficiently resolved before every expand based on
  current local context. This, for example, allows using different snippets
  in different local tree-sitter languages (like in markdown code blocks).
  See |MiniSnippets.default_prepare()|.

- Match which snippet to insert based on the currently typed text.
  Supports both exact and fuzzy matching. See |MiniSnippets.default_match()|.

- Select from several matched snippets via `vim.ui.select()`.
  See |MiniSnippets.default_select()|.

- Insert, jump, and edit during snippet session in a configurable manner:
    - Configurable mappings for jumping and stopping.
    - Jumping wraps around the tabstops for easier navigation.
    - Easy to reason rules for when session automatically stops.
    - Text synchronization of linked tabstops preserving relative indent.
    - Dynamic tabstop state visualization (current/visited/unvisited, etc.)
    - Inline visualization of empty tabstops (requires Neovim>=0.10).
    - Works inside comments by preserving comment leader on new lines.
    - Supports nested sessions (expand snippet while there is an active one).
  See |MiniSnippets.default_insert()|.

- Exported function to parse snippet body into easy-to-reason data structure.
  See |MiniSnippets.parse()|.

Notes:
- It does not set up any snippet collection by default. Explicitly populate
  `config.snippets` to have snippets to match from.
- It does not come with a built-in snippet collection. It is expected from
  users to add their own snippets, manually or with dedicated plugin(s).
- It does not support variable/tabstop transformations in default snippet
  session. This requires ECMAScript Regular Expression parser which can not
  be implemented concisely.

Sources with more details:
- |MiniSnippets-glossary|
- |MiniSnippets-overview|
- |MiniSnippets-examples|
- |MiniSnippets-in-other-plugins| (for plugin authors)

# Dependencies ~

This module doesn't come with snippet collection. Either create it manually
or install a dedicated plugin. For example, 'rafamadriz/friendly-snippets'.

# Setup ~

This module needs a setup with `require('mini.snippets').setup({})` (replace `{}`
with your `config` table). It will create global Lua table `MiniSnippets` which
you can use for scripting or manually (with `:lua MiniSnippets.*`).

See |MiniSnippets.config| for `config` structure and default values.

You can override runtime config settings locally to buffer inside
`vim.b.minisnippets_config` which should have same structure as
`Minisnippets.config`. See |mini.nvim-buffer-local-config| for more details.

# Comparisons ~

- 'L3MON4D3/LuaSnip':
    - Both contain functionality to load snippets from file system.
      This module provides several common loader generators while 'LuaSnip'
      contains a more elaborate loading setup.
      Also both require explicit opt-in for which snippets to load.
    - Both support LSP snippet format. 'LuaSnip' also provides own more
      elaborate snippet format which is out of scope for this module.
    - 'LuaSnip' can autoexpand snippets, while this module always requires
      an explicit user action to expand (by design).
    - Both contain snippet expand functionality which differs in some aspects:
        - 'LuaSnip' has an elaborate dynamic tabstop visualization config.
          This module provides a handful of dedicated highlight groups.
        - This module provides configurable visualization of empty tabstops.
        - 'LusSnip' implements nested sessions by essentially merging them
          into one. This module treats each nested session separately (to not
          visually overload) while storing them in stack (first in last out).
        - 'LuaSnip' uses |Select-mode| to power replacing current tabstop,
          while this module always stays in |Insert-mode|. This enables easier
          mapping understanding and more targeted highlighting.
        - This module implements jumping which wraps after final tabstop
          for more flexible navigation (enhanced with by a more flexible
          autostopping rules), while 'LuaSnip' autostops session once
          jumping reached the final tabstop.

- Built-in |vim.snippet| (on Neovim>=0.10):
    - Does not contain functionality to load or match snippets (by design),
      while this module does.
    - Both contain expand functionality based on LSP snippet format.
      Differences in how snippet sessions are handled are similar to
      comparison with 'LuaSnip'.

- 'rafamadriz/friendly-snippets':
    - A snippet collection plugin without features to manage or expand them.
      This module is designed with 'friendly-snippets' compatibility in mind.

- 'abeldekat/cmp-mini-snippets':
    - A source for 'hrsh7th/nvim-cmp' that integrates 'mini.snippets'.

# Highlight groups ~

* `MiniSnippetsCurrent` - current tabstop.
* `MiniSnippetsCurrentReplace` - current tabstop, placeholder is to be replaced.
* `MiniSnippetsFinal` - special `$0` tabstop.
* `MiniSnippetsUnvisited` - not yet visited tabstop(s).
* `MiniSnippetsVisited` - visited tabstop(s).

To change any highlight group, modify it directly with |:highlight|.

# Disabling ~

To disable core functionality, set `vim.g.minisnippets_disable` (globally) or
`vim.b.minisnippets_disable` (for a buffer) to `true`. Considering high number
of different scenarios and customization intentions, writing exact rules
for disabling module's functionality is left to user. See
|mini.nvim-disabling-recipes| for common recipes.

------------------------------------------------------------------------------
                                                         *MiniSnippets-glossary*
`POSITION`        Table representing position in a buffer. Fields:
                - <line> `(number)` - line number (starts at 1).
                - <col> `(number)` - column number (starts at 1).

`REGION`          Table representing region in a buffer.
                Fields: <from> and <to> for inclusive start/end POSITIONs.

`SNIPPET`         Data about template to insert. Should contain fields:
                - <prefix> - string snippet identifier.
                - <body> - string snippet content with appropriate syntax.
                - <desc> - string snippet description in human readable form.
                Can also be used to mean snippet body if distinction is clear.

`SNIPPET SESSION` Interactive state for user to adjust inserted snippet.

`MATCHED SNIPPET` SNIPPET which contains <region> field with REGION that
                matched it. Usually region needs to be removed.

`SNIPPET NODE`    Unit of parsed SNIPPET body. See |MiniSnippets.parse()|.

`TABSTOP`         Dedicated places in SNIPPET body for users to interactively
                adjust. Specified in snippet body with `$` followed by digit(s).

`LINKED TABSTOPS` Different nodes assigned the same tabstop. Updated in sync.

`REFERENCE NODE`  First (from left to right) node of linked tabstops.
                Used to determine synced text and cursor placement after jump.

`EXPAND`          Action to start snippet session based on currently typed text.
                Always done in current buffer at cursor. Executed steps:
                - `PREPARE` - resolve raw config snippets at context.
                - `MATCH` - match resolved snippets at cursor position.
                - `SELECT` - possibly choose among matched snippets.
                - `INSERT` - insert selected snippet and start snippet session.

------------------------------------------------------------------------------
                                                         *MiniSnippets-overview*
Snippet is a template for a frequently used text. Typical workflow is to type
snippet's (configurable) prefix and expand it into a snippet session: add some
pre-defined text and allow user to interactively change/add at certain places.

This overview assumes default config for mappings and expand.
See |MiniSnippets.config| and |MiniSnippets-examples| for more details.

# Snippet structure ~

Snippet consists from three parts:
- `Prefix` - identifier used to match against current text.
- `Body` - actually inserted content with appropriate syntax.
- `Desc` - description in human readable form.

Example: `{ prefix = 'tis', body = 'This is snippet', desc = 'Snip' }`
Typing `tis` and pressing "expand" mapping (<C-j> by default) will remove "tis",
add "This is snippet", and place cursor at the end in Insert mode.

                                             *MiniSnippets-syntax-specification*
# Syntax ~

Inserting just text after typing smaller prefix is already powerful enough.
For more flexibility, snippet body can be formatted in a special way to
provide extra features. This module implements support for syntax defined
in LSP specification (with small deviations). See this link for reference:
https://microsoft.github.io/language-server-protocol/specifications/lsp/3.18/specification/#snippet_syntax

A quick overview of basic syntax features:

- Tabstops are snippet parts meant for interactive editing at their location.
  They are denoted as `$1`, `$2`, etc.
  Navigating between them is called "jumping" and is done in numerical order
  of tabstop identifiers by pressing special keys: <C-l> and <C-h> to jump
  to next and previous tabstop respectively.
  Special tabstop `$0` is called "final tabstop": it is used to decide when
  snippet session is automatically stopped and is visited last during jumping.

  Example: `T1=$1 T2=$2 T0=$0` is expanded as `T1= T2= T0=` with three tabstops.

- Tabstop can have placeholder: a text used if tabstop is not yet edited.
  Text is preserved if no editing is done. It follows this same syntax, which
  means it can itself contain tabstops with placeholders (i.e. be nested).
  Tabstop with placeholder is denoted as `${1:placeholder}` (`$1` is `${1:}`).

  Example: `T1=${1:text} T2=${2:<$1>}` is expanded as `T1=text T2=<text>`;
           typing `x` at first placeholder results in `T1=x T2=<x>`;
           jumping once and typing `y` results in `T1=x T2=y`.

- There can be several tabstops with same identifier. They are linked and
  updated in sync during text editing. Can also have different placeholders;
  they are forced to be the same as in the first (from left to right) tabstop.

  Example: `T1=${1:text} T1=$1` is expanded as `T1=text T1=text`;
           typing `x` at first placeholder results in `T1=x T1=x`.

- Tabstop can also have choices: suggestions about tabstop text. It is denoted
  as `${1|a,b,c|}`. First choice is used as placeholder.

  Example: `T1=${1|left,right|}` is expanded as `T1=left`.

- Variables can be used to automatically insert text without user interaction.
  As tabstops, each one can have a placeholder which is used if variable is
  not defined. There is a special set of variables describing editor state.

  Example: `V1=$TM_FILENAME V2=${NOTDEFINED:placeholder}` is expanded as
           `V1=current-file-basename V2=placeholder`.

What's different from LSP specification:
- Special set of variables is wider and is taken from VSCode specification:
  https://code.visualstudio.com/docs/editor/userdefinedsnippets#_variables
  Exceptions are `BLOCK_COMMENT_START` and `BLOCK_COMMENT_END` as Neovim doesn't
  provide this information.
- Variable `TM_SELECTED_TEXT` is resolved as contents of |quote_quote| register.
  It assumes that text is put there prior to expanding. For example, visually
  select, press |c|, type prefix, and expand.
  See |MiniSnippets-examples| for how to adjust this.
- Environment variables are recognized and supported: `V1=$VIMRUNTIME` will
  use an actual value of |$VIMRUNTIME|.
- Variable transformations are not supported during snippet session. It would
  require interacting with ECMAScript-like regular expressions for which there
  is no easy way in Neovim. It may change in the future.
  Transformations are recognized during parsing, though, with some exceptions:
    - The `}` inside `if` of `${1:?if:else}` needs escaping (for technical reasons).

There is a |MiniSnippets.parse()| function for programmatically parsing
snippet body into a comprehensible data structure.

# Expand ~

Using snippets is done via what is called "expanding". It goes like this:
- Type snippet prefix or its recognizable part.
- Press <C-j> to expand. It will perform the following steps:
    - Prepare available snippets in current context (buffer + local language).
      This allows snippet setup to have general function loaders which return
      different snippets in different contexts.
    - Match text to the left of cursor with available prefixes. It first tries
      to do exact match and falls back to fuzzy matching.
    - If there are several matches, use `vim.ui.select()` to choose one.
    - Insert single matching snippet. If snippet contains tabstops, start
      snippet session.

For more details about each step see:
- |MiniSnippets.default_prepare()|
- |MiniSnippets.default_match()|
- |MiniSnippets.default_select()|
- |MiniSnippets.default_insert()|

Snippet session allows interactive editing at tabstop locations:

- All tabstop locations are visualized depending on tabstop "state" (whether
  it is current/visited/unvisited/final and whether it was already edited).
  Empty tabstops are visualized with inline virtual text ("•"/"∎" for
  regular/final tabstops). It is removed after session is stopped.

- Start session at first tabstop. Type text to replace placeholder.
  When finished with current tabstop, jump to next with <C-l>. Repeat.
  If changed mind about some previous tabstop, jump back with <C-h>.
  Jumping also wraps around the edge (first tabstop is next after final).

- If tabstop has choices, use <C-n> / <C-p> to select next / previous item.

- Starting another snippet session while there is an active one is allowed.
  This creates nested sessions: suspend current, start the new one.
  After newly created is stopped, resume the suspended one.

- Stop session manually by pressing <C-c> or make it stop automatically:
  if final tabstop is current either make a text edit or exit to Normal mode.
  If snippet doesn't explicitly define final tabstop, it is added at the end
  of the snippet.

For more details about snippet session see |MiniSnippets-session|.

# Management ~

Out of the box 'mini.snippets' doesn't load any snippets, it should be done
explicitly inside |MiniSnippets.setup()| following |MiniSnippets.config|.

The suggested approach to snippet management is to create dedicated files with
snippet data and load them through function loaders in `config.snippets`.
See |MiniSnippets-examples| for basic (yet capable) snippet management config.

                                               *MiniSnippets-file-specification*
General idea of supported files is to have at least out of the box experience
with common snippet collections. Namely "rafamadriz/friendly-snippets".
The following files are supported:

- Extensions:
    - Read/decoded as JSON object (|vim.json.decode()|): `*.json`, `*.code-snippets`
    - Executed as Lua file (|dofile()|) and uses returned value: `*.lua`

- Content:
    - Dict-like: object in JSON; returned table in Lua; no order guarantees.
    - Array-like: array in JSON; returned array table in Lua; preserves order.

Example of file content with a single snippet:
- Lua dict-like:   `return { name = { prefix = 't', body = 'Text' } }`
- Lua array-like:  `return { { prefix = 't', body = 'Text', desc = 'name' } }`
- JSON dict-like:  `{ "name": { "prefix": "t", "body": "Text" } }`
- JSON array-like: `[ { "prefix": "t", "body": "Text", "desc": "name" } ]`

General advice:
- Put files in "snippets" subdirectory of any path in 'runtimepath' (like
  "$XDG_CONFIG_HOME/nvim/snippets/global.json").
  This is compatible with |MiniSnippets.gen_loader.from_runtime()| and
  example from |MiniSnippets-examples|.
- Prefer `*.json` files with dict-like content if you want more cross platfrom
  setup. Otherwise use `*.lua` files with array-like content.

Notes:
- There is no built-in support for VSCode-like "package.json" files. Define
  structure manually in |MiniSnippets.setup()| via built-in or custom loaders.
- There is no built-in support for `scope` field of snippet data. Snippets are
  expected to be manually separated into smaller files and loaded on demand.

For supported snippet syntax see |MiniSnippets-syntax-specification|.

# Demo ~

The best way to grasp the design of snippet management and expansion is to
try them out yourself. Here are steps for a basic demo:
- Create 'snippets/global.json' file in the config directory with the content: >

  {
    "Basic":        { "prefix": "ba", "body": "T1=$1 T2=$2 T0=$0"         },
    "Placeholders": { "prefix": "pl", "body": "T1=${1:aa}\nT2=${2:<$1>}"  },
    "Choices":      { "prefix": "ch", "body": "T1=${1|a,b|} T2=${2|c,d|}" },
    "Linked":       { "prefix": "li", "body": "T1=$1\n\tT1=$1"            },
    "Variables":    { "prefix": "va", "body": "Runtime: $VIMRUNTIME\n"    },
    "Complex":      {
      "prefix": "co",
      "body": [ "T1=${1:$RANDOM}", "T3=${3:$1_${2:$1}}", "T2=$2" ]
    }
  }
<
- Set up 'mini.snippets' as recommended in |MiniSnippets-examples|.
- Open Neovim. Type each snippet prefix and press <C-j> (even if there is
  still active session). Explore from there.

------------------------------------------------------------------------------
                                                         *MiniSnippets-examples*
# Basic snippet management config ~

Example of snippet management setup that should cover most cases: >lua

  -- Setup
  local gen_loader = require('mini.snippets').gen_loader
  require('mini.snippets').setup({
    snippets = {
      -- Load custom file with global snippets first
      gen_loader.from_file('~/.config/nvim/snippets/global.json'),

      -- Load snippets based on current language by reading files from
      -- "snippets/" subdirectories from 'runtimepath' directories.
      gen_loader.from_lang(),
    },
  })
<
This setup allows having single file with custom "global" snippets (will be
present in every buffer) and snippets which will be loaded based on the local
language (see |MiniSnippets.gen_loader.from_lang()|).

Create language snippets manually (by creating and populating
'$XDG_CONFIG_HOME/nvim/snippets/lua.json' file) or by installing dedicated
snippet collection plugin (like 'rafamadriz/friendly-snippets').

Note: all built-in loaders and |MiniSnippets.read_file()| cache their output
by default. It means that after a file is first read, changing it won't have
effect during current Neovim session. See |MiniSnippets.gen_loader| about how
to reset cache if necessary.

# Select from all available snippets in current context ~

With |MiniSnippets.default_match()|, expand snippets (<C-j> by default) at line
start or after whitespace. To be able to always select from all current
context snippets, make mapping similar to the following: >lua

  local rhs = function() MiniSnippets.expand({ match = false }) end
  vim.keymap.set('i', '<C-g><C-j>', rhs, { desc = 'Expand all' })
<
# "Supertab"-like <Tab> / <S-Tab> mappings ~

This module intentionally by default uses separate keys to expand and jump as
it enables cleaner use of nested sessions. Here is an example of setting up
custom <Tab> to "expand or jump" and <S-Tab> to "jump to previous": >lua

  local snippets = require('mini.snippets')
  local match_strict = function(snips)
    -- Do not match with whitespace to cursor's left
    return snippets.default_match(snips, { pattern_fuzzy = '%S+' })
  end
  snippets.setup({
    -- ... Set up snippets ...
    mappings = { expand = '', jump_next = '', jump_prev = '' },
    expand   = { match = match_strict },
  })
  local expand_or_jump = function()
    local can_expand = #MiniSnippets.expand({ insert = false }) > 0
    if can_expand then vim.schedule(MiniSnippets.expand); return '' end
    local is_active = MiniSnippets.session.get() ~= nil
    if is_active then MiniSnippets.session.jump('next'); return '' end
    return '\t'
  end
  local jump_prev = function() MiniSnippets.session.jump('prev') end
  vim.keymap.set('i', '<Tab>', expand_or_jump, { expr = true })
  vim.keymap.set('i', '<S-Tab>', jump_prev)
<
# Stop session immediately after jumping to final tabstop ~

Utilize a dedicated |MiniSnippets-events|: >lua

  local fin_stop = function(args)
    if args.data.tabstop_to == '0' then MiniSnippets.session.stop() end
  end
  local au_opts = { pattern = 'MiniSnippetsSessionJump', callback = fin_stop }
  vim.api.nvim_create_autocmd('User', au_opts)
<
# Stop all sessions on Normal mode exit ~

Use |ModeChanged| and |MiniSnippets-events| events: >lua

  local make_stop = function()
    local au_opts = { pattern = '*:n', once = true }
    au_opts.callback = function()
      while MiniSnippets.session.get() do
        MiniSnippets.session.stop()
      end
    end
    vim.api.nvim_create_autocmd('ModeChanged', au_opts)
  end
  local opts = { pattern = 'MiniSnippetsSessionStart', callback = make_stop }
  vim.api.nvim_create_autocmd('User', opts)
<
# Customize variable evaluation ~

Create environment variables and `config.expand.insert` wrapper: >lua

  -- Use evnironment variables with value is same for all snippet sessions
  vim.loop.os_setenv('USERNAME', 'user')

  -- Compute custom lookup for variables with dynamic values
  local insert_with_lookup = function(snippet)
    local lookup = {
      TM_SELECTED_TEXT = table.concat(vim.fn.getreg('a', true, true), '\n'),
    }
    return MiniSnippets.default_insert(snippet, { lookup = lookup })
  end

  require('mini.snippets').setup({
    -- ... Set up snippets ...
    expand = { insert = insert_with_lookup },
  })
<
# Using Neovim's built-ins to insert snippet ~

Define custom `expand.insert` in |MiniSnippets.config| and mappings: >lua

  require('mini.snippets').setup({
    -- ... Set up snippets ...
    expand = {
      insert = function(snippet, _) vim.snippet.expand(snippet.body) end
    }
  })
  -- Make jump mappings or skip to use built-in <Tab>/<S-Tab> in Neovim>=0.11
  local jump_next = function()
    if vim.snippet.active({direction = 1}) then return vim.snippet.jump(1) end
  end
  local jump_prev = function()
    if vim.snippet.active({direction = -1}) then vim.snippet.jump(-1) end
  end
  vim.keymap.set({ 'i', 's' }, '<C-l>', jump_next)
  vim.keymap.set({ 'i', 's' }, '<C-h>', jump_prev)
<
                                                 *MiniSnippets-in-other-plugins*
# Using 'mini.snippets' in other plugins ~

- Perform a `_G.MiniSnippets ~= nil` check before using any feature. This
  ensures that user explicitly set up 'mini.snippets'.

- To insert snippet given its body (like |vim.snippet.expand()|), use: >lua

     -- Use configured `insert` method with falling back to default
     local insert = MiniSnippets.config.expand.insert
       or MiniSnippets.default_insert
     -- Insert at cursor
     insert({ body = snippet })
<
- To get available snippets, use: >lua

  -- Get snippets matched at cursor
  MiniSnippets.expand({ insert = false })

  -- Get all snippets available at cursor context
  MiniSnippets.expand({ match = false, insert = false })
<
------------------------------------------------------------------------------
                                                          *MiniSnippets.setup()*
                         `MiniSnippets.setup`({config})
Module setup

Parameters ~
{config} `(table|nil)` Module config table. See |MiniSnippets.config|.

Usage ~
>lua
  require('mini.snippets').setup({}) -- replace {} with your config table
                                     -- needs `snippets` field present
<
------------------------------------------------------------------------------
                                                           *MiniSnippets.config*
                             `MiniSnippets.config`
Module config

Default values:
>lua
  MiniSnippets.config = {
    -- Array of snippets and loaders (see |MiniSnippets.config| for details).
    -- Nothing is defined by default. Add manually to have snippets to match.
    snippets = {},

    -- Module mappings. Use `''` (empty string) to disable one.
    mappings = {
      -- Expand snippet at cursor position. Created globally in Insert mode.
      expand = '<C-j>',

      -- Interact with default `expand.insert` session.
      -- Created for the duration of active session(s)
      jump_next = '<C-l>',
      jump_prev = '<C-h>',
      stop = '<C-c>',
    },

    -- Functions describing snippet expansion. If `nil`, default values
    -- are `MiniSnippets.default_<field>()`.
    expand = {
      -- Resolve raw config snippets at context
      prepare = nil,
      -- Match resolved snippets at cursor position
      match = nil,
      -- Possibly choose among matched snippets
      select = nil,
      -- Insert selected snippet
      insert = nil,
    },
  }
<
# Loaded snippets ~

`config.snippets` is an array containing snippet data which can be: snippet
table, function loader, or (however deeply nested) array of snippet data.

Snippet is a table with the following fields:

- <prefix> `(string|table|nil)` - string used to match against current text.
   If array, all strings should be used as separate prefixes.
- <body> `(string|table|nil)` - content of a snippet which should follow
   the |MiniSnippets-syntax-specification|. Array is concatenated with "\n".
- <desc> `(string|table|nil)` - description of snippet. Can be used to display
  snippets in a more human readable form. Array is concatenated with "\n".

Function loaders are expected to be called with single `context` table argument
(containing any data about current context) and return same as `config.snippets`
data structure.

`config.snippets` is resolved with `config.prepare` on every expand.
See |MiniSnippets.default_prepare()| for how it is done by default.

For a practical example see |MiniSnippets-examples|.
Here is an illustration of `config.snippets` customization capabilities: >lua

  local gen_loader = require('mini.snippets').gen_loader
  require('mini.snippets').setup({
    snippets = {
      -- Load custom file with global snippets first (order matters)
      gen_loader.from_file('~/.config/nvim/snippets/global.json'),

      -- Or add them here explicitly
      { prefix='cdate', body='$CURRENT_YEAR-$CURRENT_MONTH-$CURRENT_DATE' },

      -- Load snippets based on current language by reading files from
      -- "snippets/" subdirectories from 'runtimepath' directories.
      gen_loader.from_lang(),

      -- Load project-local snippets with `gen_loader.from_file()`
      -- and relative path (file doesn't have to be present)
      gen_loader.from_file('.vscode/project.code-snippets'),

      -- Custom loader for language-specific project-local snippets
      function(context)
        local rel_path = '.vscode/' .. context.lang .. '.code-snippets'
        if vim.fn.filereadable(rel_path) == 0 then return end
        return MiniSnippets.read_file(rel_path)
      end,

      -- Ensure that some prefixes are not used (as there is no `body`)
      { prefix = { 'bad', 'prefix' } },
    }
  })
<
# Mappings ~

`config.mappings` describes which mappings are automatically created.

`mappings.expand` is created globally in Insert mode and is used to expand
snippet at cursor. Use |MiniSnippets.expand()| for custom mappings.

`mappings.jump_next`, `mappings.jump_prev`, and `mappings.stop` are created for
the duration of active snippet session(s) from |MiniSnippets.default_insert()|.
Used to jump to next/previous tabstop and stop active session respectively.
Use |MiniSnippets.session.jump()| and |MiniSnippets.session.stop()| for custom
Insert mode mappings.
Note: do not use `"<C-n>"` or `"<C-p>"` for any action as they conflict with
built-in completion: it forces them to mean "change focus to next/previous
completion item". This matters more frequently than when there is a tabstop
with choices due to how this module handles built-in completion during jumps.

# Expand ~

`config.expand` defines expand steps (see |MiniSnippets-glossary|), either after
pressing `mappings.expand` or starting manually via |MiniSnippets.expand()|.

`expand.prepare` is a function that takes `raw_snippets` in the form of
`config.snippets` and should return a plain array of snippets (as described
in |MiniSnippets-glossary|). Will be called on every |MiniSnippets.expand()| call.
If returns second value, it will be used as context for warning messages.
Default: |MiniSnippets.default_prepare()|.

`expand.match` is a function that takes `expand.prepare` output and returns
an array of matched snippets: one or several snippets user might intend to
eventually insert. Should sort matches in output from best to worst.
Entries can contain `region` field with current buffer region used to do
the match; usually it needs to be removed (similar to how |ins-completion|
and |abbreviations| work).
Default: |MiniSnippets.default_match()|

`expand.select` is a function that takes output of `expand.match` and function
that inserts snippet (and also ensures Insert mode and removes snippet's match
region). Should allow user to perform interactive snippet selection and
insert the chosen one. Designed to be compatible with |vim.ui.select()|.
Called for any non-empty `expand.match` output (even with single entry).
Default: |MiniSnippets.default_select()|

`expand.insert` is a function that takes single snippet table as input and
inserts snippet at cursor position. This is a main entry point for adding
text template to buffer and starting a snippet session.
If called inside |MiniSnippets.expand()| (which is a usual interactive case),
all it has to do is insert snippet at cursor position. Ensuring Insert mode
and removing matched snippet region is done beforehand.
Default: |MiniSnippets.default_insert()|

Illustration of `config.expand` customization: >lua

  -- Supply extra data as context
  local my_p = function(raw_snippets)
    local _, cont = MiniSnippets.default_prepare({})
    cont.cursor = vim.api.nvim_win_get_cursor()
    return MiniSnippets.default_prepare(raw_snippets, { context = cont })
  end
  -- Perform fuzzy match based only on alphanumeric characters
  local my_m = function(snippets)
    return MiniSnippets.default_match(snippets, { pattern_fuzzy = '%w*' })
  end
  -- Always insert the best matched snippet
  local my_s = function(snippets, insert) return insert(snippets[1]) end
  -- Use different string to show empty tabstop as inline virtual text
  local my_i = function(snippet)
    return MiniSnippets.default_insert(snippet, { empty_tabstop = '$' })
  end

  require('mini.snippets').setup({
    -- ... Set up snippets ...
    expand = { prepare = my_p, match = my_m, select = my_s, insert = my_i }
  })
<
------------------------------------------------------------------------------
                                                         *MiniSnippets.expand()*
                         `MiniSnippets.expand`({opts})
Expand snippet at cursor position

Perform expand steps (see |MiniSnippets-glossary|).
Initial raw snippets are taken from `config.snippets` in current buffer.
Snippets from `vim.b.minisnippets_config` are appended to global snippet array.

Parameters ~
{opts} `(table|nil)` Options. Same structure as `expand` in |MiniSnippets.config|
  and uses its values as default. There are differences in allowed values:
  - Use `match = false` to have all buffer snippets as matches.
  - Use `select = false` to always expand the best match (if any).
  - Use `insert = false` to return all matches without inserting.

  Note: `opts.insert` is called after ensuring Insert mode, removing snippet's
  match region, and positioning cursor.

Return ~
`(table|nil)` If `insert` is `false`, an array of matched snippets (`expand.match`
  output). Otherwise `nil`.

Usage ~
>lua
  -- Match, maybe select, and insert
  MiniSnippets.expand()

  -- Match and force expand the best match (if any)
  MiniSnippets.expand({ select = false })

  -- Use all current context snippets as matches
  MiniSnippets.expand({ match = false })

  -- Get all matched snippets
  local matches = MiniSnippets.expand({ insert = false })

  -- Get all current context snippets
  local all = MiniSnippets.expand({ match = false, insert = false })
<
------------------------------------------------------------------------------
                                                       *MiniSnippets.gen_loader*
                           `MiniSnippets.gen_loader`
Generate snippet loader

This is a table with function elements. Call to actually get a loader.

Common features for all produced loaders:
- Designed to work with |MiniSnippets-file-specification|.
- Cache output by default, i.e. second and later calls with same input value
  don't read file system. Different loaders from same generator share cache.
  Disable by setting `opts.cache` to `false`.
  To clear all cache, call |MiniSnippets.setup()|. For example:
  `MiniSnippets.setup(MiniSnippets.config)`
- Use |vim.notify()| to show problems during loading while trying to load as
  much correctly defined snippet data as possible.
  Disable by setting `opts.silent` to `true`.

------------------------------------------------------------------------------
                                           *MiniSnippets.gen_loader.from_lang()*
                  `MiniSnippets.gen_loader.from_lang`({opts})
Generate language loader

Output loads files from "snippets/" subdirectories of 'runtimepath' matching
configured language patterns.
See |MiniSnippets.gen_loader.from_runtime()| for runtime loading details.

Language is taken from <lang> field (if present with string value) of `context`
argument used in loader calls during "prepare" stage.
This is compatible with |MiniSnippets.default_prepare()| and most snippet
collection plugins.

Parameters ~
{opts} `(table|nil)` Options. Possible values:
  - <lang_patterns> `(table)` - map from language to array of runtime patterns
    used to find snippet files, as in |MiniSnippets.gen_loader.from_runtime()|.
    Patterns will be processed in order. With |MiniSnippets.default_prepare()|
    it means if snippets have same prefix, data from later patterns is used.
    To interactively check the current language with default context, execute
    `:=MiniSnippets.default_prepare({})` and see data in the second table.

    Default pattern array (for non-empty language) is constructed as to read
    `*.json` and `*.lua` files that are:
    - Inside "snippets/" subdirectory named as language (files can be however
      deeply nested).
    - Named as language and is in "snippets/" directory (however deep).
    Example for "lua" language: >lua
    { 'lua/**/*.json', 'lua/**/*.lua', '**/lua.json', '**/lua.lua' }
<
    Add entry for `""` (empty string) as language to be sourced when `lang`
    context is empty string (which is usually temporary scratch buffers).

  - <cache> `(boolean)` - whether to use cached output. Default: `true`.
    Note: caching is done per used runtime pattern, not `lang` value to allow
    different `from_lang()` loaders to share cache.
  - <silent> `(boolean)` - whether to hide non-error messages. Default: `false`.

Return ~
`(function)` Snippet loader.

Usage ~
>lua
  -- Adjust language patterns
  local latex_patterns = { 'latex/**/*.json', '**/latex.json' }
  local lang_patterns = { tex = latex_patterns, plaintex = latex_patterns }
  local gen_loader = require('mini.snippets').gen_loader
  require('mini.snippets').setup({
    snippets = {
      gen_loader.from_lang({ lang_patterns = lang_patterns }),
    },
  })
<
------------------------------------------------------------------------------
                                        *MiniSnippets.gen_loader.from_runtime()*
           `MiniSnippets.gen_loader.from_runtime`({pattern}, {opts})
Generate runtime loader

Output loads files which match `pattern` inside "snippets/" directories from
'runtimepath'. This is useful to simultaneously read several similarly
named files from different sources. Order from 'runtimepath' is preserved.

Typical case is loading snippets for a language from files like `xxx.{json,lua}`
but located in different "snippets/" directories inside 'runtimepath'.
- `<config>`/snippets/lua.json - manually curated snippets in user config.
- `<path/to/installed/plugin>`/snippets/lua.json - from installed plugin.
- `<config>`/after/snippets/lua.json - used to adjust snippets from plugins.
  For example, remove some snippets by using prefixes and no body.

Parameters ~
{pattern} `(string)` Pattern of files to read. Can have wildcards as described
  in |nvim_get_runtime_file()|. Example for "lua" language: `'lua.{json,lua}'`.
{opts} `(table|nil)` Options. Possible fields:
  - <all> `(boolean)` - whether to load from all matching runtime files.
    Default: `true`.
  - <cache> `(boolean)` - whether to use cached output. Default: `true`.
    Note: caching is done per `pattern` value, which assumes that both
    'runtimepath' value and snippet files do not change during Neovim session.
    Caching this way gives significant speed improvement by reducing the need
    to traverse file system on every snippet expand.
  - <silent> `(boolean)` - whether to hide non-error messages. Default: `false`.

Return ~
`(function)` Snippet loader.

------------------------------------------------------------------------------
                                           *MiniSnippets.gen_loader.from_file()*
              `MiniSnippets.gen_loader.from_file`({path}, {opts})
Generate single file loader

Output is a thin wrapper around |MiniSnippets.read_file()| which will skip
warning if file is absent (other messages are still shown). Use it to load
file which is not guaranteed to exist (like project-local snippets).

Parameters ~
{path} `(string)` Same as in |MiniSnippets.read_file()|.
{opts} `(table|nil)` Same as in |MiniSnippets.read_file()|.

Return ~
`(function)` Snippet loader.

------------------------------------------------------------------------------
                                                      *MiniSnippets.read_file()*
                    `MiniSnippets.read_file`({path}, {opts})
Read file with snippet data

Parameters ~
{path} `(string)` Path to file with snippets. Can be relative.
  See |MiniSnippets-file-specification| for supported file formats.
{opts} `(table|nil)` Options. Possible fields:
  - <cache> `(boolean)` - whether to use cached output. Default: `true`.
    Note: Caching is done per full path only after successful reading.
  - <silent> `(boolean)` - whether to hide non-error messages. Default: `false`.

Return ~
`(table|nil)` Array of snippets or `nil` if failed (also warn with |vim.notify()|
  about the reason).

------------------------------------------------------------------------------
                                                *MiniSnippets.default_prepare()*
             `MiniSnippets.default_prepare`({raw_snippets}, {opts})
Default prepare

Normalize raw snippets (as in `snippets` from |MiniSnippets.config|) based on
supplied context:
- Traverse and flatten nested arrays. Function loaders are executed with
  `opts.context` as argument and output is processed recursively.
- Ensure unique non-empty prefixes: later ones completely override earlier
  ones (similar to how |ftplugin| and similar runtime design behave).
  Empty string prefixes are all added (to allow inserting without matching).
- Transform and infer fields:
    - Multiply array `prefix` into several snippets with same body/description.
      Infer absent `prefix` as empty string.
    - Concatenate array `body` with "\n". Do not infer absent `body` to have
      it remove previously added snippet with the same prefix.
    - Concatenate array `desc` with "\n". Infer `desc` field from `description`
      (for compatibility) or `body` fields, in that order.
- Sort output by prefix.

Unlike |MiniSnippets.gen_loader| entries, there is no output caching. This
avoids duplicating data from `gen_loader` cache and reduces memory usage.
It also means that every |MiniSnippets.expand()| call prepares snippets, which
is usually fast enough. If not, consider manual caching: >lua

  local cache = {}
  local prepare_cached = function(raw_snippets)
    local _, cont = MiniSnippets.default_prepare({})
    local id = 'buf=' .. cont.buf_id .. ',lang=' .. cont.lang
    if cache[id] then return unpack(vim.deepcopy(cache[id])) end
    local snippets = MiniSnippets.default_prepare(raw_snippets)
    cache[id] = vim.deepcopy({ snippets, cont })
    return snippets, cont
  end
<
Parameters ~
{raw_snippets} `(table)` Array of snippet data as from |MiniSnippets.config|.
{opts} `(table|nil)` Options. Possible fields:
  - <context> `(any)` - Context used as an argument for callable snippet data.
    Default: table with <buf_id> (current buffer identifier) and <lang> (local
    language) fields. Language is computed from tree-sitter parser at cursor
    (allows different snippets in injected languages), 'filetype' otherwise.

Return ~
`(...)` Array of snippets and supplied context (default if none was supplied).

------------------------------------------------------------------------------
                                                  *MiniSnippets.default_match()*
                `MiniSnippets.default_match`({snippets}, {opts})
Default match

Match snippets based on the line before cursor.

Tries two matching approaches consecutively:
- Find exact snippet prefix (if present and non-empty) to the left of cursor.
  It should also be preceded with a byte that matches `pattern_exact_boundary`.
  In case of any match, return the one with the longest prefix.
- Match fuzzily snippet prefixes against the base (text to the left of cursor
  extracted via `opts.pattern_fuzzy`). Matching is done via |matchfuzzy()|.
  Empty base results in all snippets being matched. Return all fuzzy matches.

Parameters ~
{snippets} `(table)` Array of snippets which can be matched.
{opts} `(table|nil)` Options. Possible fields:
  - <pattern_exact_boundary> `(string)` - Lua pattern for the byte to the left
    of exact match to accept it. Line start is matched against empty string;
    use `?` quantifier to allow it as boundary.
    Default: `[%s%p]?` (accept only whitespace and punctuation as boundary,
    allow match at line start).
    Example: prefix "l" matches in lines `l`, `_l`, `x l`; but not `1l`, `ll`.
  - <pattern_fuzzy> `(string)` - Lua pattern to extract base to the left of
    cursor for fuzzy matching. Supply empty string to skip this step.
    Default: `'%S*'` (as many as possible non-whitespace; allow empty string).

Return ~
`(table)` Array of matched snippets ordered from best to worst match.

Usage ~
>lua
  -- Accept any exact match
  MiniSnippets.default_match(snippets, { pattern_exact_boundary = '.?' })

  -- Perform fuzzy match based only on alphanumeric characters
  MiniSnippets.default_match(snippets, { pattern_fuzzy = '%w*' })
<
------------------------------------------------------------------------------
                                                 *MiniSnippets.default_select()*
          `MiniSnippets.default_select`({snippets}, {insert}, {opts})
Default select

Show snippets as |vim.ui.select()| items and insert the chosen one.
For best interactive experience requires `vim.ui.select()` to work from Insert
mode (be properly called and restore Insert mode after choice).
This is the case for at least |MiniPick.ui_select()| and Neovim's default.

Parameters ~
{snippets} `(table)` Array of snippets (as an output of `config.expand.match`).
{insert} `(function|nil)` Function to insert chosen snippet (passed as the only
  argument). Expected to remove snippet's match region (if present as a field)
  and ensure proper cursor position in Insert mode.
  Default: |MiniSnippets.default_insert()|.
{opts} `(table|nil)` Options. Possible fields:
  - <insert_single> `(boolean)` - whether to skip |vim.ui.select()| for `snippets`
    with a single entry and insert it directly. Default: `true`.

------------------------------------------------------------------------------
                                                 *MiniSnippets.default_insert()*
                `MiniSnippets.default_insert`({snippet}, {opts})
Default insert

Prepare for snippet insert and do it:
- Ensure Insert mode.
- Delete snippet's match region (if present as <region> field). Ensure cursor.
- Parse snippet body with |MiniSnippets.parse()| and enabled `normalize`.
  In particular, evaluate variables, ensure final node presence and same
  text for nodes with same tabstops. Stop if not able to.
- Insert snippet at cursor:
    - Add snippet's text. Lines are split at "\n".
      Indent and left comment leaders (inferred from 'commentstring' and
      'comments') of current line are repeated on the next.
      Tabs ("\t") are expanded according to 'expandtab' and 'shiftwidth'.
    - If there is an actionable tabstop (not final), start snippet session.

                                                          *MiniSnippets-session*
# Session life cycle ~

- Start with cursor at first tabstop. If there are linked tabstops, cursor
  is placed at start of reference node (see |MiniSnippets-glossary|).
  All tabstops are visualized with dedicated highlight groups (see "Highlight
  groups" section in |MiniSnippets|).
  Empty tabstops are visualized with inline virtual text ("•"/"∎" for
  regular/final tabstops) meaning that it is not an actual text in the
  buffer and will be removed after session is stopped.

- Decide whether you want to replace the placeholder. If not, jump to next or
  previous tabstop. If yes, edit it: add new and/or delete already added text.
  While doing so, several things happen in all linked tabstops (if any):

    - After first typed character the placeholder is removed and highlighting
      changes from `MiniSnippetsCurrentReplace` to `MiniSnippetsCurrent`.
    - Text in all tabstop nodes is synchronized with the reference one.
      Relative indent of reference tabstop's text is preserved: all but first
      lines in linked tabstops are reindented based on the first line indent.
      Note: text sync is forced only for current tabstop (for performance).

- Jump with <C-l> / <C-h> to next / previous tabstop. Exact keys can be
  adjusted in |MiniSnippets.config| `mappings`.
  See |MiniSnippets.session.jump()| for jumping details.

- If tabstop has choices, all of them are shown after each jump and deleting
  tabstop text. It is done with |complete()|, so use <C-n> / <C-p> to select
  next / previous choice. Type text to narrow down the list.
  Works best when 'completeopt' option contains `menuone` and `noselect` flags.
  Note: deleting character hides the list due to how |complete()| works;
  delete whole tabstop text (for example with one or more |i_CTRL-W|) for
  full list to reappear.

- Nest another session by expanding snippet in the same way as without
  active session (can be even done in another buffer). If snippet has no
  actionable tabstop, text is just inserted. Otherwise start nested session:

    - Suspend current session: hide highlights, keep text change tracking.
    - Start new session and act as if it is the only one (edit/jump/nest).
    - When ready (possibly after even more nested sessions), stop the session.
      This will resume previous one: sync text for its current tabstop and
      show highlighting.
      The experience of text synchronization only after resuming session is
      similar to how editing in |visual-block| mode works.
      Nothing else (like cursor/mode/buffer) is changed for a smoother
      automated session stop.

  Notes about the choice of the "session stack" approach to nesting over more
  common "merge into single session" approach:
  - Does not overload with highlighting.
  - Allows nested sessions in different buffers.
  - Doesn't need a complex logic of injecting one session into another.

- Repeat edit/jump/nest steps any number of times.

- Stop. It can be done in two ways:

    - Manually by pressing <C-c> or calling |MiniSnippets.session.stop()|.
      Exact key can be adjusted in |MiniSnippets.config| `mappings`.
    - Automatically: any text edit or switching to Normal mode stops session
      if final tabstop (`$0`) is current. Its presence is ensured after insert.
      Not stopping session right away after jumping to final mode (as most
      other snippet plugins do) allows going back to other tabstops in case
      of a late missed typo. Wrapping around the edge during jumping also
      helps with that.
      If current tabstop is not final, exiting into Normal mode for quick edit
      outside of snippets range (or carefully inside) is fine. Later get back
      into Insert mode and jump to next tabstop or manually stop session.
  See |MiniSnippets-examples| for how to set up custom stopping rules.

Use |MiniSnippets.session.get()| to get data about active/nested session(s).
Use |MiniSnippets.session.jump()| / |MiniSnippets.session.stop()| in mappings.

What is allowed but not officially supported/recommended:

- Editing text within snippet range but outside of session life cycle. Mostly
  behaves as expected, but may harm tracking metadata (|extmarks|).
  In general anything but deleting tabstop range should be OK.
  Text synchronization of current tabstop would still be active.

                                                         *MiniSnippets-events*
# Events ~

General session activity (autocommand data contains <session> field):
- `MiniSnippetsSessionStart` - after a session is started.
- `MiniSnippetsSessionStop` - before a session is stopped.

Nesting session activity (autocommand data contains <session> field):
- `MiniSnippetsSessionSuspend` - before a session is suspended.
- `MiniSnippetsSessionResume` - after a session is resumed.

Jumping between tabstops (autocommand data contains <tabstop_from> and
<tabstop_new> fields):
- `MiniSnippetsSessionJumpPre` - before jumping to a new tabstop.
- `MiniSnippetsSessionJump` - after jumping to a new tabstop.

Parameters ~
{snippet} `(table)` Snippet table. Field <body> is mandatory.
{opts} `(table|nil)` Options. Possible fields:
  - <empty_tabstop> `(string)` - used to visualize empty regular tabstops.
    Default: "•".
  - <empty_tabstop_final> `(string)` - used to visualize empty final tabstop(s).
    Default: "∎".
  - <lookup> `(table)` - passed to |MiniSnippets.parse()|. Use it to adjust
    how variables are evaluated. Default: `{}`.

------------------------------------------------------------------------------
                                                          *MiniSnippets.session*
                             `MiniSnippets.session`
Work with snippet session from |MiniSnippets.default_insert()|

------------------------------------------------------------------------------
                                                    *MiniSnippets.session.get()*
                       `MiniSnippets.session.get`({all})
Get data about active session

Parameters ~
{all} `(boolean|nil)` Whether to return array with the whole session stack.
  Default: `false`.

Return ~
`(table)` Single table with session data (if `all` is `false`) or array of them.
  Session data contains the following fields:
   - <buf_id> `(number)` - identifier of session's buffer.
   - <cur_tabstop> `(string)` - identifier of session's current tabstop.
   - <extmark_id> `(number)` - |extmark| identifier which track session range.
   - <insert_args> `(table)` - |MiniSnippets.default_insert()| arguments used to
     create the session. A table with <snippet> and <opts> fields.
   - <nodes> `(table)` - parsed array of snippet nodes which is kept up to date
     during session. Has the structure of a normalized |MiniSnippets.parse()|
     output, plus every node contains `extmark_id` field with |extmark| identifier
     which can be used to get data about the current node state.
   - <ns_id> `(number)` - |namespace| identifier for all session's extmarks.
   - <tabstops> `(table)` - data about session's tabstops. Fields are string
     tabstop identifiers and values are tables with the following fields:
       - <is_visited> `(boolean)` - whether tabstop was visited.
       - <next> `(string)` - identifier of the next tabstop.
       - <prev> `(string)` - identifier of the previous tabstop.

------------------------------------------------------------------------------
                                                   *MiniSnippets.session.jump()*
                    `MiniSnippets.session.jump`({direction})
Jump to next/previous tabstop

Make next/previous tabstop be current. Executes the following steps:
- Mark current tabstop as visited.
- Find the next/previous tabstop id assuming they are sorted as numbers.
  Tabstop "0" is always last. Search is wrapped around the edges: first and
  final tabstops are next/previous for one another.
- Focus on target tabstop:
    - Ensure session's buffer is current.
    - Adjust highlighting of affected nodes.
    - Set cursor at tabstop's reference node (first node among linked).
      Cursor is placed on left edge if tabstop has not been edited yet (so
      typing text replaces placeholder), on right edge otherwise (to update
      already edited text).
    - Show all choices for tabstop with choices. Navigating through choices
      will update tabstop's text.

Parameters ~
{direction} `(string)` One of "next" or "prev".

------------------------------------------------------------------------------
                                                   *MiniSnippets.session.stop()*
                         `MiniSnippets.session.stop`()
Stop (only) active session

To stop all nested sessions use the following code: >lua

  while MiniSnippets.session.get() do
    MiniSnippets.session.stop()
  end
<
------------------------------------------------------------------------------
                                                          *MiniSnippets.parse()*
                  `MiniSnippets.parse`({snippet_body}, {opts})
Parse snippet

Parameters ~
{snippet_body} `(string|table)` Snippet body as string or array of strings.
  Should follow |MiniSnippets-syntax-specification|.
{opts} `(table|nil)` Options. Possible fields:
  - <normalize> `(boolean)` - whether to normalize nodes:
    - Evaluate variable nodes and add output as a `text` field.
      If variable is not set, `text` field is `nil`.
      Values from `opts.lookup` are preferred over evaluation output.
      See |MiniSnippets-syntax-specification| for more info about variables.
    - Add `text` field for tabstops present in `opts.lookup`.
    - Ensure every node contains exactly one of `text` or `placeholder` fields.
      If there are none, add default `placeholder` (one text node with first
      choice or empty string). If there are both, remove `placeholder` field.
    - Ensure present final tabstop: append to end if absent.
    - Ensure that nodes for same tabstop have same placeholder. Use the one
      from the first node.
    Default: `false`.
  - <lookup> `(table)` - map from variable/tabstop (string) name to its value.
    Default: `{}`.

Return ~
`(table)` Array of nodes. Node is a table with fields depending on node type:
  - Text node:
    - <text> `(string)` - node's text.
  - Tabstop node:
    - <tabstop> `(string)` - tabstop identifier.
    - <text> `(string|nil)` - tabstop value (if present in <lookup>).
    - <placeholder> `(table|nil)` - array of nodes to be used as placeholder.
    - <choices> `(table|nil)` - array of string choices.
    - <transform> `(table|nil)` - array of transformation string parts.
  - Variable node:
    - <var> `(string)` - variable name.
    - <text> `(string|nil)` - variable value.
    - <placeholder> `(table|nil)` - array of nodes to be used as placeholder.
    - <transform> `(table|nil)` - array of transformation string parts.


 vim:tw=78:ts=8:noet:ft=help:norl: