Editor scripts

You can create custom menu items and editor lifecycle hooks using Lua files with special extension: .editor_script. Using this system, you can tweak editor to enhance your development workflow.

Editor script runtime

Editor scripts run inside an editor, in a Lua VM emulated by Java VM. All scripts share the same single environment, which means they can interact with each other. You can require Lua modules, just as with .script files, but Lua version that is running inside the editor is different, so make sure your shared code is compatible. Editor uses Lua version 5.2.x, more specifically luaj runtime, which is currently the only viable solution to run Lua on JVM. Besides that, there are some restrictions:

• there is no debug package;
• there is no os.execute, though we provide a similar editor.execute();
• there is no os.tmpname and io.tmpfile — currently editor scripts can access files only inside the project directory;
• there is currently no os.rename, although we want to add it;
• there is no os.exit and os.setlocale.
• it’s not allowed to use some long-running functions in contexts where the editor needs an immediate response from the script, see Execution Modes for more details.

All editor extensions defined in editor scripts are loaded when you open a project. When you fetch libraries, extensions are reloaded, since there might be new editor scripts in a libraries you depend on. During this reload, no changes in your own editor scripts are picked up, since you might be in the middle of changing them. To reload them as well, you should run Project → Reload Editor Scripts command.

Anatomy of .editor_script

Every editor script should return a module, like that:

local M = {}

function M.get_commands()
  -- TODO - define editor commands
end

function M.get_language_servers()
  -- TODO - define language servers
end

function M.get_prefs_schema()
  -- TODO - define preferences
end

return M

Editor then collects all editor scripts defined in project and libraries, loads them into single Lua VM and calls into them when needed (more on that in commands and lifecycle hooks sections).

Editor API

You can interact with the editor using editor package that defines this API:

• editor.platform — a string, either "x86_64-win32" for Windows, "x86_64-macos" for macOS or "x86_64-linux" for Linux.
• editor.version — a string, version name of Defold, e.g. "1.4.8"
• editor.engine_sha1 — a string, SHA1 of Defold engine
• editor.editor_sha1 — a string, SHA1 of Defold editor
• editor.get(node_id, property) — get a value of some node inside the editor. Nodes in the editor are various entities, such as script or collection files, game objects inside collections, json files loaded as resources, etc. node_id is a userdata that is passed to the editor script by the editor. Alternatively, you can pass resource path instead of node id, for example "/main/game.script". property is a string. Currently these properties are supported:
  • "path" — file path from the project folder for resources — entities that exist as files. Example of returned value: "/main/game.script"
  • "text" — text content of a resource editable as text (such as script files or json). Example of returned value: "function init(self)\nend". Please note that this is not the same as reading file with io.open(), because you can edit a file without saving it, and these edits are available only when accessing "text" property.
  • some properties that are shown in the Properties view when you have selected something in the Outline view. These types of outline properties supported:
    • strings
    • booleans
    • numbers
    • vec2/vec3/vec4
    • resources

    Please note that some of these properties might be read-only, and some might be unavailable in different contexts, so you should use editor.can_get before reading them and editor.can_set before making editor set them. Hover over property name in Properties view to see a tooltip with information about how this property is named in editor scripts. You can set resource properties to nil by supplying "" value.

• editor.can_get(node_id, property) — check if you can get this property so editor.get() won’t throw an error.
• editor.can_set(node_id, property) — check if editor.tx.set() transaction step with this property won’t throw an error.
• editor.create_directory(resource_path) — create a directory if it does not exist, and all non-existent parent directories.
• editor.delete_directory(resource_path) — delete a directory if it exists, and all existent child directories and files.
• editor.execute(cmd, [...args], [options]) — run a shell command, optionally capturing its output.
• editor.save() — persist all unsaved changed to disk.
• editor.transact(txs) — modify the editor in-memory state using 1 or more transaction steps created with editor.tx.* functions.
• editor.ui.* — various UI-related functions, see UI manual.
• editor.prefs.* — functions for interacting with editor preferences, see prefs.

You can find the full editor API reference here.

Commands

If editor script module defines function get_commands, it will be called on extension reload, and returned commands will be available for use inside the editor in menu bar or in context menus in Assets and Outline panes. Example:

local M = {}

function M.get_commands()
  return {
    {
      label = "Remove Comments",
      locations = {"Edit", "Assets"},
      query = {
        selection = {type = "resource", cardinality = "one"}
      },
      active = function(opts)
        local path = editor.get(opts.selection, "path")
        return ends_with(path, ".lua") or ends_with(path, ".script")
      end,
      run = function(opts)
        local text = editor.get(opts.selection, "text")
        editor.transact({
          editor.tx.set(opts.selection, "text", strip_comments(text))
        })
      end
    },
    {
      label = "Minify JSON"
      locations = {"Assets"},
      query = {
        selection = {type = "resource", cardinality = "one"}
      },
      active = function(opts)
        return ends_with(editor.get(opts.selection, "path"), ".json")
      end,
      run = function(opts)
        local path = editor.get(opts.selection, "path")
        editor.execute("./scripts/minify-json.sh", path:sub(2))
      end
    }
  }
end

return M

Editor expects get_commands() to return an array of tables, each describing a separate command. Command description consists of:

• label (required) — text on a menu item that will be displayed to the user
• locations (required) — an array of either "Edit", "View", "Assets" or "Outline", describes a place where this command should be available. "Edit" and "View" mean menu bar at the top, "Assets" means context menu in Assets pane, and "Outline" means context menu in Outline pane.
• query — a way for command to ask editor for relevant information and define what data it operates on. For every key in query table there will be corresponding key in opts table that active and run callbacks receive as argument. Supported keys:
  • selection means this command is valid when there is something selected, and it operates on this selection.
    • type is a type of selected nodes command is interested in, currently these types are allowed:
      • "resource" — in Assets and Outline, resource is selected item that has a corresponding file. In menu bar (Edit or View), resource is a currently open file;
      • "outline" — something that can be shown in the Outline. In Outline it’s a selected item, in menu bar it’s a currently open file;
    • cardinality defines how many selected items there should be. If "one", selection passed to command callback will be a single node id. If "many", selection passed to command callback will be an array of one or more node ids.
• active - a callback that is executed to check that command is active, expected to return boolean. If locations include "Assets" or "Outline", active will be called when showing context menu. If locations include "Edit" or "View", active will be called on every user interaction, such as typing on keyboard or clicking with mouse, so be sure that active is relatively fast.
• run - a callback that is executed when user selects the menu item.

Use commands to change the in-memory editor state

Inside the run handler, you can query and change the in-memory editor state. Querying is done using editor.get() function, where you can ask the editor about the current state of files and selection (if using query = {selection = ...}). You can get the "text" property of script files, and also some properties shown in the Properties view — hover over property name to see a tooltip with information about how this property is named in editor scripts. Changing the editor state is done using editor.transact(), where you bundle 1 or more modifications in a single undoable step. For example, if you want to be able to reset transform of a game object, you could write a command like that:

{
  label = "Reset transform",
  locations = {"Outline"},
  query = {selection = {type = "outline", cardinality = "one"}},
  active = function(opts)
    local node = opts.selection
    return editor.can_set(node, "position") 
       and editor.can_set(node, "rotation") 
       and editor.can_set(node, "scale")
  end,
  run = function(opts)
    local node = opts.selection
    editor.transact({
      editor.tx.set(node, "position", {0, 0, 0}),
      editor.tx.set(node, "rotation", {0, 0, 0}),
      editor.tx.set(node, "scale", {1, 1, 1})
    })
  end
}

Use shell commands

Inside the run handler, you can write to files (using io module) and execute shell commands (using editor.execute() command). When executing shell commands, it’s possible to capture the output of a shell command as a string and then use it in code. For example, if you want to make a command for formatting JSON that shells out to globally installed jq, you can write the following command:

{
  label = "Format JSON",
  locations = {"Assets"},
  query = {selection = {type = "resource", cardinality = "one"}},
  action = function(opts)
    local path = editor.get(opts.selection, "path")
    return path:match(".json$") ~= nil
  end,
  run = function(opts)
    local text = editor.get(opts.selection, "text")
    local new_text = editor.execute("jq", "-n", "--argjson", "data", text, "$data", {
      reload_resources = false, -- don't reload resources since jq does not touch disk
      out = "capture" -- return text output instead of nothing
    })
    editor.transact({ editor.tx.set(opts.selection, "text", new_text) })
  end
}

Since this command invokes shell program in a read-only way (and notifies the editor about it using reload_resources = false), you get the benefit of making this action undoable.

Lifecycle hooks

There is a specially treated editor script file: hooks.editor_script, located in a root of your project, in the same directory as game.project. This and only this editor script will receive lifecycle events from the editor. Example of such file:

local M = {}

function M.on_build_started(opts)
  local file = io.open("assets/build.json", "w")
  file:write('{"build_time": "' .. os.date() .. '"}')
  file:close()
end

return M

We decided to limit lifecycle hooks to single editor script file because order in which build hooks happen is more important than how easy it is to add another build step. Commands are independent from each other, so it does not really matter in what order they are shown in the menu, in the end user executes a particular command they selected. If it was possible to specify build hooks in different editor scripts, it would create a problem: in which order do hooks execute? You probably want to create a checksums of content after you compress it… And having a single file that establishes order of build steps by calling each step function explicitly is a way to solve this problem.

Existing lifecycle hooks that /hooks.editor_script may specify:

• on_build_started(opts) — executed when game is Built to run locally or on some remote target using either the Project Build or Debug Start options. Your changes will appear in the built game. Raising an error from this hook will abort a build. opts is a table that contains following keys:
  • platform — a string in %arch%-%os% format describing what platform it’s built for, currently always the same value as in editor.platform.
• on_build_finished(opts) — executed when build is finished, be it successful or failed. opts is a table with following keys:
  • platform — same as in on_build_started
  • success — whether build is successful, either true or false
• on_bundle_started(opts) — executed when you create a bundle or Build HTML5 version of a game. As with on_build_started, changes triggered by this hook will appear in a bundle, and errors will abort a bundle. opts will have these keys:
  • output_directory — a file path pointing to a directory with bundle output, for example "/path/to/project/build/default/__htmlLaunchDir"
  • platform — platform the game is bundled for. See a list of possible platform values in Bob manual.
  • variant — bundle variant, either "debug", "release" or "headless"
• on_bundle_finished(opts) — executed when bundle is finished, be it successful or not. opts is a table with the same data as opts in on_bundle_started, plus success key indicating whether build is successful.
• on_target_launched(opts) — executed when user launched a game and it successfully started. opts contains an url key pointing to a launched engine service, for example, "http://127.0.0.1:35405"
• on_target_terminated(opts) — executed when launched game is closed, has same opts as on_target_launched

Please note that lifecycle hooks currently are an editor-only feature, and they are not executed by Bob when bundling from command line.

Language servers

The editor supports a subset Language Server Protocol. While we aim to expand the editor’s support for LSP features in the future, currently it can only show diagnostics (i.e. lints) in the edited files and provide completions.

To define the language server, you need to edit your editor script’s get_language_servers function like so:

function M.get_language_servers()
  local command = 'build/plugins/my-ext/plugins/bin/' .. editor.platform .. '/lua-lsp'
  if editor.platform == 'x86_64-win32' then
    command = command .. '.exe'
  end
  return {
    {
      languages = {'lua'},
      watched_files = {
        { pattern = '**/.luacheckrc' }
      },
      command = {command, '--stdio'}
    }
  }
end

The editor will start the language server using the specified command, using the server process’s standard input and output for communication.

Language server definition table may specify:

• languages (required) — a list of languages the server is interested in, as defined here (file extensions also work);
• command (required) - an array of command and its arguments
• watched_files - an array of tables with pattern keys (a glob) that will trigger the server’s watched files changed notification.

Editor scripts in libraries

You can publish libraries for other people to use that contain commands, and they will be automatically picked up by the editor. Hooks, on the other hand, can’t be picked up automatically, since they have to be defined in a file that is in a root folder of a project, but libraries expose only subfolders. This is intended to give more control over build process: you still can create lifecycle hooks as simple functions in .lua files, so users of your library can require and use them in their /hooks.editor_script.

Also note that although dependencies are shown in Assets view, they do not exist as files (they are entries in a zip archive). It’s possible to make the editor extract some files from the dependencies into build/plugins/ folder. To do it, you need to create ext.manifest file in your library folder, and then create plugins/bin/${platform} folder in the same folder where the ext.manifest file is located. Files in that folder will be automatically extracted to /build/plugins/${extension-path}/plugins/bin/${platform} folder, so your editor scripts can reference them.

Prefs

Editor scripts can define and use preferences — persistent, uncommitted pieces of data stored on the user’s computer. These preferences have three key characteristics:

• typed: every preference has a schema definition that includes the data type and other metadata like default value
• scoped: prefs are scoped either per project or per user
• nested: every preference key is a dot-separated string, where the first path segment identifies an editor script, and the rest

All preferences must be registered by defining their schema:

function M.get_prefs_schema()
  return {
    ["my_json_formatter.jq_path"] = editor.prefs.schema.string(),
    ["my_json_formatter.indent.size"] = editor.prefs.schema.integer({default = 2, scope = editor.prefs.SCOPE.PROJECT}),
    ["my_json_formatter.indent.type"] = editor.prefs.schema.enum({values = {"spaces", "tabs"}, scope = editor.prefs.SCOPE.PROJECT}),
  }
end

After such editor script is reloaded, the editor registers this schema. Then the editor script can get and set the prefs, e.g.:

-- Get a specific preference
editor.prefs.get("my_json_formatter.indent.type")
-- Returns: "spaces"

-- Get an entire preference group
editor.prefs.get("my_json_formatter")
-- Returns:
-- {
--   jq_path = "",
--   indent = {
--     size = 2,
--     type = "spaces"
--   }
-- }

-- Set multiple nested preferences at once
editor.prefs.set("my_json_formatter.indent", {
    type = "tabs",
    size = 1
})

Execution modes

The editor script runtime uses 2 execution modes that are mostly transparent to editor scripts: immediate and long-running.

Immediate mode is used when the editor needs to receive a response from the script as fast as possible. For instance, menu commands’ active callbacks are executed in immediate mode, because these checks are performed on the editors UI thread in response to user interacting with the editor, and should update the UI within the same frame.

Long-running mode is used when the editor doesn’t need an instantaneous response from the script. For example, menu commands’ run callbacks are executed in a long-running mode, allowing the script to take more time to complete its work.

Some of the functions that the editor scripts can use may take a lot of time to run. For example, editor.execute("git", "status", {reload_resources=false, out="capture"}) can take up to a second on sufficiently large projects. To maintain editor responsiveness and performance, functions that may be time-consuming are not allowed in contexts where the editor needs an immediate response. Attempting to use such a function in an immediate context will result in an error: Cannot use long-running editor function in immediate context. To resolve this error, avoid using such functions in immediate contexts.

The following functions are considered long-running and cannot be used in immediate mode:

• editor.create_directory(), editor.delete_directory(), editor.save(), os.remove() and file:write(): these functions modify the files on disc, causing the editor to synchronize its in-memory resource tree with the disc state, which can take seconds in large projects.
• editor.execute(): execution of shell commands can take an unpredictable amount of time.
• editor.transact(): large transactions on widely-referenced nodes may take hundreds of milliseconds, which is too slow for UI responsiveness.

The following code execution contexts use immediate mode:

• Menu command’s active callbacks: the editor needs a response from the script within the same UI frame.
• Top-level of editor scripts: we don’t expect the act of reloading editor scripts to have any side effects.

Actions

Editor scripts may return an array of actions from a command’s run function or from /hooks.editor_script’s hook functions. These actions will then be performed by the editor.

Action is a table describing what editor should do. Every action has an action key. Actions come in 2 flavors: undoable and non-undoable.

Undoable actions

Undoable action can be undone after it is executed. If a command returns multiple undoable actions, they are performed together, and get undone together. You should use undoable actions if you can. Their downside is that they are more limited.

Existing undoable actions:

• "set" — set a property of a node in the editor to some value. Example:

  {
    action = "set",
    node_id = opts.selection,
    property = "text",
    value = "current time is " .. os.date()
  }
  

  "set" action requires these keys:

  • node_id — node id userdata. Alternatively, you can use resource path here instead of node id you received from the editor, for example "/main/game.script";
  • property — a property of a node to set, e.g. "text";
  • value — new value for a property. For "text" property it should be a string.

Non-undoable actions

Non-undoable action clears undo history, so if you want to undo such action, you will have to use other means, such as version control.

Existing non-undoable actions:

• "shell" — execute a shell script. Example:

  {
    action = "shell",
    command = {
      "./scripts/minify-json.sh",
      editor.get(opts.selection, "path"):sub(2) -- trim leading "/"
    }
  }
  

  "shell" action requires command key, which is an array of command and it’s arguments.

Mixing actions and side effects

You can mix undoable and non-undoable actions. Actions are executed sequentially, hence depending on an order of actions you will end up losing ability to undo parts of that command.

Instead of returning actions from functions that expect them, you can just read and write to files directly using io.open(). This will trigger a resource reload that will clear undo history.