zhoulii opened a new issue, #9696: URL: https://github.com/apache/paimon/issues/9696
### Search before asking - [x] I searched in the [issues](https://github.com/apache/paimon/issues) and found nothing similar. ### Paimon version master ### Compute Engine java & python api ### Minimal reproduce step **Java** 1. Create a table with `id` and `embedding ARRAY<FLOAT> COMMENT '__VECTOR_FIELD;3'`, with row tracking and data evolution enabled. 2. Insert a row and create a tag named `before_add`. 3. Add `embedding_v2` using the same vector directive. 4. Reload the table and read the `before_add` tag, projecting only `id`. **Python** 1. Create a data-evolution table with a `__BLOB_DESCRIPTOR_FIELD` or `__BLOB_VIEW_FIELD` column. Write a non-null reference and retain the snapshot. 2. Drop that column, reload the table, and read the retained snapshot. ### What doesn't meet your expectations? Expected: historical reads should use the field declarations from the selected snapshot. Adding or dropping columns should not break earlier snapshots or change their returned values. Actual: - Java fails before reading data with `Some of the columns specified as vector-field are unknown.` Similar validation failures occur after adding BLOB columns. - Python returns serialized descriptor/view bytes instead of resolving the original payload. Both paths combine historical fields with current schema options, leaving the field declarations inconsistent. ### Anything else? _No response_ ### Are you willing to submit a PR? - [x] I'm willing to submit a PR! -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
