async parsing + tex caching

Plugins/ZShell/Llm/session.cpp

@@ -94,8 +94,8 @@ LlmClient* ChatSession::client() const {
 }
 
 void ChatSession::ensureLoaded() {
-	if (m_loaded) return;
-	m_loaded = true;
+	if (m_loadRequested) return;
+	m_loadRequested = true;
 	if (auto* store = qobject_cast<ChatStore*>(parent()))
 		store->loadMessagesInto(this);
 }
@@ -105,6 +105,18 @@ ChatMessageModel* ChatSession::messagesModel() {
 	return m_model;
 }
 
+void ChatSession::markLoaded() {
+	if (m_loaded) return;
+	m_loaded = true;
+	Q_EMIT loaded();
+}
+
+bool ChatSession::takeClearPending() {
+	const bool pending = m_clearPending;
+	m_clearPending = false;
+	return pending;
+}
+
 void ChatSession::adoptMessages(QList<ChatMessage*> messages) {
 	m_model->loadMessages(std::move(messages));
 }
@@ -127,9 +139,20 @@ void ChatSession::clearMessages() {
 }
 
 void ChatSession::startGeneration(ChatMessage* target) {
-	if (auto* generation = target->activeGeneration()) {
-		if (auto* clientObject = client())
+	auto* generation = target->activeGeneration();
+	auto* clientObject = client();
+	if (generation && clientObject) {
+		if (isLoaded()) {
 			clientObject->startGeneration(this, generation);
+		} else {
+			// The store load is still in flight; the request needs the
+			// full history, so start once it lands.
+			connect(
+				this, &ChatSession::loaded, clientObject,
+				[this, generation, clientObject]() {
+					clientObject->startGeneration(this, generation);
+				});
+		}
 	}
 	persist();
 }
@@ -209,6 +232,11 @@ void ChatSession::clear() {
 			return;
 		}
 	}
+	if (!isLoaded()) {
+		// The load lands shortly; drop everything once it does.
+		m_clearPending = true;
+		return;
+	}
 	m_model->clear();
 	persist();
 }