Spring AI OllamaChatModel doesn't respect custom AbstractMessage's implementation

Bug description I'm using spring-ai along with local ollama. The OllamaChatModel's implementation doesn't consider external AbstractMessage implementation in method OllamaApi.ChatRequest ollamaChatRequest(Prompt prompt, boolean stream). There method body makes few class cast checks and then throws exception IllegalArgumentException with relevant message about unsupported message type. However, org.springframework.ai.chat.messages.AbstractMessage is open to extend and I can come up with my own independent implementation.

Environment spring-ai:1.0.0

Steps to reproduce

Implement MyChatMemory with method like this

   @Override
    public List<Message> get(String conversationId) {

        return List.of(new AbstractMessage(MessageType.ASSISTANT, "any text", Map.of()) {

        });
    }

Expose bean

    @Bean
    public ChatClient chatClient(ChatClient.Builder builder, MyChatMemory chatMemory) {

        return builder.defaultAdvisors(MessageChatMemoryAdvisor.builder(chatMemory).build()).build();
    }

Evoke using ChatClient

 chatClient.prompt(prompt) //
                .advisors(advisorSpec -> advisorSpec.param(ChatMemory.CONVERSATION_ID, id))
                .stream() //
                .chatResponse() //
                .subscribe(response -> {
                   //process response
                });

Expected behavior

The org.springframework.ai.ollama.OllamaChatModel#ollamaChatRequest doesn't throw exception java.lang.IllegalArgumentException: Unsupported message type: ASSISTANT

Comment From: sunyuhan1998

I agree with your point. Relying solely on instanceof to determine the message type in this approach is unreasonable, as it imposes limitations on custom implementations of AbstractMessage and is also inconsistent with the implementation in other models. I've submitted a PR to address this issue, which you can follow here: #3999