VideoContentBlock Not Supported In OpenAI Block Translator
The OpenAI block translator in langchain-core does not handle VideoContentBlock, causing a ValueError when sending video content blocks to OpenAI-compatible models via langchain-openai.
The function convert_to_openai_data_block in langchain_core/messages/block_translators/openai.py lacks a branch for block type 'video', so it falls into the generic else clause and raises ValueError.
1. Initialize a chat model using init_chat_model with an OpenAI-compatible base URL (e.g., DashScope).
2. Create a HumanMessage with content_blocks containing a TextContentBlock and a VideoContentBlock with url and mime_type.
3. Invoke the model with this message. The call fails with ValueError: Block of type video is not supported.
Add a dedicated elif branch for block type 'video' that maps VideoContentBlock to the OpenAI video_url content part format, including url and mime_type. The mime_type defaults to 'video/mp4' if absent to avoid KeyError.
Edge Case Audit
This fix assumes the target OpenAI-compatible endpoint supports the video_url content type. Some models or providers may not support video input, causing a different API error. Additionally, if VideoContentBlock is used without a mime_type, the default may not match the actual video format, leading to rejection. Rollback advice: if you encounter provider-specific issues, revert this change and construct the video_url content part manually as a plain dict in your message content.
