JingsongLi commented on code in PR #11178:
URL: https://github.com/apache/paimon/pull/11178#discussion_r4583489915


##########
paimon-core/src/main/java/org/apache/paimon/mergetree/compact/PartialUpdateMergeFunction.java:
##########
@@ -206,8 +206,9 @@ public void add(KeyValue kv) {
             }
 
             if (removeRecordOnDelete) {
+                // -U also removes the row, a later +I or +U restores it
+                currentDeleteRow = true;

Review Comment:
   [P1] Gate Native readers until their retraction semantics match
   
   This changes the interpretation of files that the currently supported Native 
backend still reads with the old semantics. I wrote a real Parquet 
partial-update table with remove-record-on-delete=true and write-only=true, 
committing +I(1,10,20) and then -U(1,10,NULL) separately. Reading those exact 
same files with this Java head returns no row; pypaimon-rust 0.5.0 returns the 
stale (1,10,20) row, with Python fallback explicitly forbidden. The clean Java 
merge-base also returns that row, confirming the divergence is introduced here. 
pyproject.toml still accepts Rust >=0.5.0, Native reads default to enabled, and 
the bridge has no capability gate for this option. Consequently Java/Flink and 
supported PyPaimon Native readers disagree on whether a committed key exists. 
Align Rust and require its compatible version/capability, or explicitly reject 
affected Native configurations before enabling this contract change. Pure 
Python already rejects this option, so silently falling back is insuf
 ficient.



-- 
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