feat(core): Allow checking tool_calls on any BaseMessage without type casts

[jacoblee93]

We should no longer need to do casts like this:

const prettyPrint = (message: BaseMessage) => {
  let txt = `[${message.getType()}]: ${message.content}`;
  if (
    isAIMessage(message) && (message as AIMessage)?.tool_calls?.length || 0 > 0
  ) {
    ...
  }
  console.log(txt);
};

Instead:

const prettyPrint = (message: BaseMessage) => {
  let txt = `[${message.getType()}]: ${message.content}`;
  if (message.tool_calls?.length) {
    ...
  }
  console.log(txt);
};

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
langchainjs-docs	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Jan 8, 2025 1:36am

1 Skipped Deployment

Name	Status	Preview	Comments	Updated (UTC)
langchainjs-api-refs	⬜️ Ignored (Inspect)			Jan 8, 2025 1:36am

[jacoblee93]

So this will mess up a few automatic type inference situations like this:

// messages is inferred as SystemMessage[]
const messages = [new SystemMessage("foo")];
messages.push(new AIMessage("bar"));

error TS2345: Argument of type 'AIMessage' is not assignable to parameter of type 'HumanMessage'.
  Types of property 'tool_calls' are incompatible.
    Type 'ToolCall[] | undefined' is not assignable to type 'never[] | undefined'.
      Type 'ToolCall[]' is not assignable to type 'never[]'.
        Type 'ToolCall' is not assignable to type 'never'.

I still think it's worth it since you can do const messages: BaseMessage[] = [...] or just use dict (as most people should be doing) syntax and it's just this one type narrowing situation that causes issues. Would love thoughts.

[jacoblee93]

Yeah think I need to find a way to make this non-breaking...