peter-toth commented on PR #58659:
URL: https://github.com/apache/spark/pull/58659#issuecomment-5718191565

   Thanks @cloud-fan. All five are addressed in e0cc1ebf53a.
   
   **The columnar guard.** You are right, and the test was weaker than it 
looked: removing the `doExecuteColumnar` call left it green. The invariant test 
now drives a columnar child through `executeColumnar()` as well, and it fails 
with either call removed.
   
   **The factory doc.** Also right, and the claim was false. 
`plannedPartitioning` holds `AttributeReference`s and is normalized. What 
actually makes `doCanonicalize`'s `copy` safe is that `super.doCanonicalize()` 
has already normalized both partitioning fields by then, and the reducer 
rewrite touches neither. Reworded to say that.
   
   **`keysSatisfy` and the link.** Both taken. The predicate says nothing about 
duplicate keys, which a grouping node also coalesces, so an admitted layout can 
still need one and `satisfies` is what adds `isGrouped`. And 
`[[GroupPartitionsExec]]` does not resolve from `sql/catalyst`; both 
occurrences are plain backticks now, like the other twenty-five in that file.
   
   **The `pickCoPartitionTarget` doc.** Taken. It still said the largest 
partition count picks the target, where a child with no `ShuffleExchangeLike` 
comes first. Worth noting that the sentence and the preference are both older 
than this PR, so the contradiction was already there; this change moved the 
sentence into a new scaladoc, so it is fixed here.
   
   **The eight malformed sentences**, with one correction. Six of them are not 
new: `in the child's on top`, `groups partition every input index`, `it's 
physical size less than`, `the number represent`, `either one of below` and 
`according to other shuffle spec` all appear on master and show as added lines 
only because the restructure moves those blocks into new methods. The two 
written here are the dangling `raw partition keys'` and `The claim goes on to 
declare lives`.
   
   I fixed all eight rather than only mine, since leaving the older six would 
have left the new methods reading badly.
   
   For the rename: thank you, I will leave it in then.
   


-- 
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