autocomplete.md

Autocomplete

The text editor provides an optional autocomplete framework. Once configured, the editor takes care of activation events (triggering), state tracking, visualization and insertion of suggestions with full undo/redo. The application is responsible for providing the list of suggestions through a callback or the API. This allows simple implementations to provide suggestions in realtime and allows other implementations to do things asynchronously like reaching out to a language server. Autocomplete can't be triggered when multiple cursors are active as this causes a mess. Try it in Visual Studio Code if you want to see what I mean.

To activate the feature, the app must provide a configuration like:

TextEditor::AutoCompleteConfig config;

config.callback = [this](TextEditor::AutoCompleteState& state) {
	....
};

editor.SetAutoCompleteConfig(&config);

Deactivation can be achieved by passing nullptr to SetAutoCompleteConfig. The TextEditor::AutoCompleteConfig class contains all the configuration options and is defined as (please note defaults):

class AutoCompleteConfig {
public:
	// specifies whether typing by the user triggers autocomplete
	bool triggerOnTyping = true;

	// specifies whether the specified shortcut triggers autocomplete
	bool triggerOnShortcut = true;

	// specifies whether typing (or shortcut) in comments or strings triggers autocomplete
	bool triggerInComments = false;
	bool triggerInStrings = false;

	// manual trigger key sequence (default is Ctrl+space on all platforms, even MacOS)
	// remember Dear ImGui reverses Ctrl and Command on MacOS
#if __APPLE__
	ImGuiKeyChord triggerShortcut = ImGuiMod_Super | ImGuiKey_Space;
#else
	ImGuiKeyChord triggerShortcut = ImGuiMod_Ctrl | ImGuiKey_Space;
#endif

	// see if single suggestions are automatically inserted
	// this only works when triggered manually
	bool autoInsertSingleSuggestions = false;

	// delay in milliseconds between autocomplete trigger and suggestions popup
	std::chrono::milliseconds triggerDelay{200};

	// text label used when no suggestions are available (this allows for internationalization)
	std::string noSuggestionsLabel = "No suggestions";

	// called when autocomplete is configured, active and the editor needs an updated suggestions list
	// callback must populate and order suggestions in state object
	// suggestion list is not cleared by editor between callbacks
	// callback is called during the rendering process (so don't take too long)

	// if it takes too long, applications should do search in separate thread and
	// use API to report results (see SetAutoCompleteSuggestions)
	std::function<void(AutoCompleteState&)> callback;

	// optional opaque void* that must be managed externally but passed to callback
	void* userData = nullptr;
};

When the callback is activated, a TextEditor::AutoCompleteState object is passed informing the app about the context and providing space to return suggestions. It is defined as:

class AutoCompleteState {
public:
	// current context (strings = UTF-8, columns = Nth visible column and indices = Nth codepoint)
	// to understand the difference between column and index, think like a tab :-)
	std::string searchTerm;
	size_t line;
	size_t searchTermStartColumn;
	size_t searchTermStartIndex;
	size_t searchTermEndColumn;
	size_t searchTermEndIndex;

	bool inIdentifier;
	bool inNumber;
	bool inComment;
	bool inString;

	// currently selected language (could be nullptr if no language is selected)
	const Language* language;

	// opaque void* provided by app when autocomplete was setup
	void* userData;

	// auto complete suggestions te be provided by app callback (the app is responsible for sorting)

	// the editor does not automatically include language specific keywords or identifiers in the suggestion list
	// this is left to the application so it can be context specific in case a language server is used
	// a pointer to the current language definition is provided so callbacks have easy access
	std::vector<std::string> suggestions;
};

The editor also comes with a Trie class that implements fuzzy searching and the example app shows how it can be used. Given that this is a primitive, poor-man's solution, more sophisticated solutions probably required external language engines/services that are beyond the scope of this editor. With the provided API however, connections to external capabilities can be established and context-sensitive suggestions can be provided based on the most advanced algorithms (even good-old AI slop :-).

Below is a quick snippet that shows how to use the Trie class to implement a poor-man's autocomplete without using a language server. This snippet was taken from the example application so can see it in context.

void Editor::setAutocompleteMode(bool flag) {
	// see we are turning autocomplete on or off
	if (flag) {
		// rebuild word list
		buildAutocompleteTrie();

		// setup autocomplete by submitting a new configuration
		TextEditor::AutoCompleteConfig config;

		config.callback = [this](TextEditor::AutoCompleteState& state) {
			trie.findSuggestions(state.suggestions, state.searchTerm);
		};

		editor.SetAutoCompleteConfig(&config);

		// enable change tracking
		// we don't track every keystroke, callbacks can be delayed up to 3000 milliseconds
		// if you want live tracking, change 3000 to 0 (performance hit will be minimal for small documents)
		editor.SetChangeCallback([this]() {
			buildAutocompleteTrie();
		}, 3000);

	} else {
		// disable autocomplete and change tracking
		editor.SetAutoCompleteConfig(nullptr);
		editor.SetChangeCallback(nullptr);
	}
}

void Editor::buildAutocompleteTrie() {
	// empty list first
	trie.clear();

	// add language words (if required)
	auto language = editor.GetLanguage();

	if (language) {
		for (auto& word : language->keywords) { trie.insert(word); }
		for (auto& word : language->declarations) { trie.insert(word); }
		for (auto& word : language->identifiers) { trie.insert(word); }
	}

	// add all identifiers in current document
	editor.IterateIdentifiers([this](const std::string& identifier) {
		trie.insert(identifier);
	});
}