skyyyy955 commented on PR #10748:
URL: https://github.com/apache/paimon/pull/10748#issuecomment-6891827154

   Thanks for fixing the branch time-travel issue in #10748, @zhoulii. I 
checked a nearby case on current master (`9b6c997198`): selecting a branch 
without a snapshot or tag still keeps the source branch's schema.
   
   In a local filesystem table, I widened `count` from INT to BIGINT only on 
`dev` and wrote `3000000000`. Loading `db.t$branch_dev` reads that value 
correctly, while `main.copy({"branch": "dev"})` returns `-1294967296` without 
an error. Adding `scan.snapshot-id` to the copy reads the correct value, so the 
original time-travel fix works. A column added only on `dev` is also absent 
from the branch-only copy.
   
   I reproduced this with real Parquet writes and reads on Windows, Python 
3.10.11 and PyArrow 19.0.1, using the Python backends. The unchanged-schema 
control and normal main reads pass. I also checked the same path through 
`CatalogSplitProvider`, but have not run distributed Ray workers or the native 
backend.
   
   Would resolving the target branch schema for this latest-read case be a 
suitable follow-up to #10748, or should branch selection use a separate API 
that reloads the schema? I have a standalone reproduction and can help with a 
small fix once the intended behavior is clear.
   


-- 
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]

Reply via email to