[Linux] Process TTS callback on the main thread to avoid speech-dispatcher deadlock.

platform/linuxbsd/tts_linux.cpp

@@ -35,6 +35,11 @@
 
 TTS_Linux *TTS_Linux::singleton = nullptr;
 
+void TTS_Linux::_bind_methods() {
+	ClassDB::bind_method(D_METHOD("_speech_event", "msg_id", "client_id", "type"), &TTS_Linux::_speech_event);
+	ClassDB::bind_method(D_METHOD("_speech_index_mark", "msg_id", "client_id", "type", "index_mark"), &TTS_Linux::_speech_index_mark);
+}
+
 void TTS_Linux::speech_init_thread_func(void *p_userdata) {
 	TTS_Linux *tts = (TTS_Linux *)p_userdata;
 	if (tts) {
@@ -76,76 +81,84 @@ void TTS_Linux::speech_init_thread_func(void *p_userdata) {
 
 void TTS_Linux::speech_event_index_mark(size_t p_msg_id, size_t p_client_id, SPDNotificationType p_type, char *p_index_mark) {
 	TTS_Linux *tts = TTS_Linux::get_singleton();
-	if (tts && tts->ids.has(p_msg_id)) {
-		MutexLock thread_safe_method(tts->_thread_safe_);
-		// Get word offset from the index mark injected to the text stream.
-		String mark = String::utf8(p_index_mark);
-		DisplayServer::get_singleton()->tts_post_utterance_event(DisplayServer::TTS_UTTERANCE_BOUNDARY, tts->ids[p_msg_id], mark.to_int());
+	if (tts) {
+		tts->call_deferred(SNAME("_speech_index_mark"), p_msg_id, p_client_id, (int)p_type, String::utf8(p_index_mark));
+	}
+}
+
+void TTS_Linux::_speech_index_mark(size_t p_msg_id, size_t p_client_id, int p_type, const String &p_index_mark) {
+	_THREAD_SAFE_METHOD_
+
+	if (ids.has(p_msg_id)) {
+		DisplayServer::get_singleton()->tts_post_utterance_event(DisplayServer::TTS_UTTERANCE_BOUNDARY, ids[p_msg_id], p_index_mark.to_int());
 	}
 }
 
 void TTS_Linux::speech_event_callback(size_t p_msg_id, size_t p_client_id, SPDNotificationType p_type) {
 	TTS_Linux *tts = TTS_Linux::get_singleton();
 	if (tts) {
-		MutexLock thread_safe_method(tts->_thread_safe_);
-		List<DisplayServer::TTSUtterance> &queue = tts->queue;
-		if (!tts->paused && tts->ids.has(p_msg_id)) {
-			if (p_type == SPD_EVENT_END) {
-				DisplayServer::get_singleton()->tts_post_utterance_event(DisplayServer::TTS_UTTERANCE_ENDED, tts->ids[p_msg_id]);
-				tts->ids.erase(p_msg_id);
-				tts->last_msg_id = -1;
-				tts->speaking = false;
-			} else if (p_type == SPD_EVENT_CANCEL) {
-				DisplayServer::get_singleton()->tts_post_utterance_event(DisplayServer::TTS_UTTERANCE_CANCELED, tts->ids[p_msg_id]);
-				tts->ids.erase(p_msg_id);
-				tts->last_msg_id = -1;
-				tts->speaking = false;
-			}
+		tts->call_deferred(SNAME("_speech_event"), p_msg_id, p_client_id, (int)p_type);
+	}
+}
+
+void TTS_Linux::_speech_event(size_t p_msg_id, size_t p_client_id, int p_type) {
+	_THREAD_SAFE_METHOD_
+
+	if (!paused && ids.has(p_msg_id)) {
+		if ((SPDNotificationType)p_type == SPD_EVENT_END) {
+			DisplayServer::get_singleton()->tts_post_utterance_event(DisplayServer::TTS_UTTERANCE_ENDED, ids[p_msg_id]);
+			ids.erase(p_msg_id);
+			last_msg_id = -1;
+			speaking = false;
+		} else if ((SPDNotificationType)p_type == SPD_EVENT_CANCEL) {
+			DisplayServer::get_singleton()->tts_post_utterance_event(DisplayServer::TTS_UTTERANCE_CANCELED, ids[p_msg_id]);
+			ids.erase(p_msg_id);
+			last_msg_id = -1;
+			speaking = false;
 		}
-		if (!tts->speaking && queue.size() > 0) {
-			DisplayServer::TTSUtterance &message = queue.front()->get();
-
-			// Inject index mark after each word.
-			String text;
-			String language;
-			SPDVoice **voices = spd_list_synthesis_voices(tts->synth);
-			if (voices != nullptr) {
-				SPDVoice **voices_ptr = voices;
-				while (*voices_ptr != nullptr) {
-					if (String::utf8((*voices_ptr)->name) == message.voice) {
-						language = String::utf8((*voices_ptr)->language);
-						break;
-					}
-					voices_ptr++;
-				}
-				free_spd_voices(voices);
-			}
-			PackedInt32Array breaks = TS->string_get_word_breaks(message.text, language);
-			for (int i = 0; i < breaks.size(); i += 2) {
-				const int start = breaks[i];
-				const int end = breaks[i + 1];
-				text += message.text.substr(start, end - start + 1);
-				text += "<mark name=\"" + String::num_int64(end, 10) + "\"/>";
-			}
+	}
+	if (!speaking && queue.size() > 0) {
+		DisplayServer::TTSUtterance &message = queue.front()->get();
 
-			spd_set_synthesis_voice(tts->synth, message.voice.utf8().get_data());
-			spd_set_volume(tts->synth, message.volume * 2 - 100);
-			spd_set_voice_pitch(tts->synth, (message.pitch - 1) * 100);
-			float rate = 0;
-			if (message.rate > 1.f) {
-				rate = log10(MIN(message.rate, 2.5f)) / log10(2.5f) * 100;
-			} else if (message.rate < 1.f) {
-				rate = log10(MAX(message.rate, 0.5f)) / log10(0.5f) * -100;
+		// Inject index mark after each word.
+		String text;
+		String language;
+		SPDVoice **voices = spd_list_synthesis_voices(synth);
+		if (voices != nullptr) {
+			SPDVoice **voices_ptr = voices;
+			while (*voices_ptr != nullptr) {
+				if (String::utf8((*voices_ptr)->name) == message.voice) {
+					language = String::utf8((*voices_ptr)->language);
+					break;
+				}
+				voices_ptr++;
 			}
-			spd_set_voice_rate(tts->synth, rate);
-			spd_set_data_mode(tts->synth, SPD_DATA_SSML);
-			tts->last_msg_id = spd_say(tts->synth, SPD_TEXT, text.utf8().get_data());
-			tts->ids[tts->last_msg_id] = message.id;
-			DisplayServer::get_singleton()->tts_post_utterance_event(DisplayServer::TTS_UTTERANCE_STARTED, message.id);
-
-			queue.pop_front();
-			tts->speaking = true;
+			free_spd_voices(voices);
 		}
+		PackedInt32Array breaks = TS->string_get_word_breaks(message.text, language);
+		for (int i = 0; i < breaks.size(); i += 2) {
+			const int start = breaks[i];
+			const int end = breaks[i + 1];
+			text += message.text.substr(start, end - start + 1);
+			text += "<mark name=\"" + String::num_int64(end, 10) + "\"/>";
+		}
+		spd_set_synthesis_voice(synth, message.voice.utf8().get_data());
+		spd_set_volume(synth, message.volume * 2 - 100);
+		spd_set_voice_pitch(synth, (message.pitch - 1) * 100);
+		float rate = 0;
+		if (message.rate > 1.f) {
+			rate = log10(MIN(message.rate, 2.5f)) / log10(2.5f) * 100;
+		} else if (message.rate < 1.f) {
+			rate = log10(MAX(message.rate, 0.5f)) / log10(0.5f) * -100;
+		}
+		spd_set_voice_rate(synth, rate);
+		spd_set_data_mode(synth, SPD_DATA_SSML);
+		last_msg_id = spd_say(synth, SPD_TEXT, text.utf8().get_data());
+		ids[last_msg_id] = message.id;
+		DisplayServer::get_singleton()->tts_post_utterance_event(DisplayServer::TTS_UTTERANCE_STARTED, message.id);
+
+		queue.pop_front();
+		speaking = true;
 	}
 }
 
@@ -204,7 +217,7 @@ void TTS_Linux::speak(const String &p_text, const String &p_voice, int p_volume,
 	if (is_paused()) {
 		resume();
 	} else {
-		speech_event_callback(0, 0, SPD_EVENT_BEGIN);
+		_speech_event(0, 0, (int)SPD_EVENT_BEGIN);
 	}
 }
 

platform/linuxbsd/tts_linux.h

@@ -45,7 +45,8 @@
 #include <libspeechd.h>
 #endif
 
-class TTS_Linux {
+class TTS_Linux : public Object {
+	GDCLASS(TTS_Linux, Object);
 	_THREAD_SAFE_CLASS_
 
 	List<DisplayServer::TTSUtterance> queue;
@@ -63,6 +64,11 @@ class TTS_Linux {
 
 	static TTS_Linux *singleton;
 
+protected:
+	static void _bind_methods();
+	void _speech_event(size_t p_msg_id, size_t p_client_id, int p_type);
+	void _speech_index_mark(size_t p_msg_id, size_t p_client_id, int p_type, const String &p_index_mark);
+
 public:
 	static TTS_Linux *get_singleton();