ulysses-you commented on PR #58339:
URL: https://github.com/apache/spark/pull/58339#issuecomment-5474284642

   Round 2 is addressed in 9a450ca; per-finding replies are on the four inline 
threads.
   
   **3 (description regressed):** rewritten. The coarsening-sites bullet now 
names all four -- key-dropping projection, union merge, `GroupPartitionsExec` 
reducers/join-key projection, and the join-key projection in 
`KeyedPartitioning.createShuffleSpec` -- plus a separate bullet for the 
spurious-marker scoping (`ShuffledJoin`'s `InnerLike` arm is the only merge 
site; broadcast joins pass through or expand one side only). The test list 
separates the wrong-results and regression repros (each measured failing 
pre-fix: 2/4, 3/4, 0/4 rows, planning throw, 2-vs-1 shuffles) from the 
invariant pins. No shapes are knowingly left broken.
   
   **Coordination with #58351:** acknowledged. This PR's narrowing refusals 
test the projection arity directly (`projectablePositions.length < 
numPositions`, `joinKeyPositions.length < expressions.length`) and deliberately 
never route through `isNarrowed`/`isCollapsed`, per your finding-10 note; the 
marker scoping is per-member semantics, not an OR-normalized layout property, 
so it should not gain the require-agreement either. Whatever lands second 
rebases; I have both the `partitioning.scala` and `GroupPartitionsExec` 
conflict shapes memorized by now.
   


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to