ChatMessageHistory and message_from_dict question

[XiaoConstantine]

Context:
I m trying to persist ChatMessageHistory to sqlite for session memory control purpose, and noticed following issue (maybe it's expected and I m missing something)

Say I have the following chat history

 cm =ChatMessageHistory(messages=[HumanMessage(content="\n         1. Generate a comprehensive and informative answer and format
    ...:  your answer (no more than 80 words after Final Answer:) in markdown. \n         2. You have to use one of your tools to answer
    ...:  questions. \n         3. You have to provide the sources / links you've used to answer the quesion.\n         4. You may use t
    ...: ools more than once.\n         5. Create your reply in the same language as the search string.\n         6. Use an unbiased and
    ...:  journalistic tone.\n         7. Use this current date and time: Tuesday, April 23, 2024 10:09 UTC\n         Question: Pixel 7
    ...: camera specs\n         "), AIMessage(content='The Pixel 7 camera specs include a 50MP primary lens with optical and digital zoo
    ...: m, an ultra-wide 12MP lens, and a 10.8MP telephoto lens. It also features autofocus technology, dual exposure controls, and Nig
    ...: ht Sight for low-light photography. For more detailed information on the Pixel 7 camera specs, you can refer to this [source](h
    ...: ttps://support.foobar.com/hc/en-us/articles/360058489753-Android-technical-difficulties-What-should-I-do-next).\nPlease no
    ...: te that the sources provided are not directly related to Pixel 7 camera specs, as the tool used (maven_search) primarily return
    ...: ed results related to technical difficulties and general information. However, the information on the Pixel
    ...: 7 camera specs mentioned above was found within the search results.<|im_end|>')], memory_key='chat_history')

I can serialize it to json string with cm.json() without issue

When I try to deserialize it back to messages:

ChatMessageHistory.parse_obj(json.loads(cm.json()))

All the messages are using the base type (BaseMessage) instead of original AIMessage, HumanMessage.

Then I found that there's an utility does handle conversion between BaseMessage and its subclasses, when i try:

d = json.loads(cm.json())
messages_from_dict(d["messages"])

I got:

from_dict(message)
     69 _type = message["type"]
     70 if _type == "human":
---> 71     return HumanMessage(**message["data"])
     72 elif _type == "ai":
     73     return AIMessage(**message["data"])

KeyError: 'data'

What's **message["data"] ? shouldn't it be just HumanMessage(**message) ?
Why does parse_obj doesn't do the additional step which check the message type and initialize it as the proper message? (maybe because it's the pydantic native function?)