mrhhsg opened a new pull request, #67962:
URL: https://github.com/apache/doris/pull/67962
### What problem does this PR solve?
Issue Number: None
Problem Summary:
`ColumnStruct::update_crc32c_batch` / `update_crc32c_single` (and the
xxHash / zlib-crc range variants) ignored the `null_map` passed in by an
outer `ColumnNullable` and kept hashing the nested field columns. A NULL
struct row can still carry an arbitrary hidden payload in its fields,
for example `IF(cond, NULL, named_struct(...))` keeps the else-branch
value under the NULL flag. Two logically equal NULL structs therefore got
different hashes and the local hash shuffle sent them to different
aggregation instances, so `GROUP BY` on a nullable struct key silently
produced one extra group per distinct hidden payload.
`ColumnArray` and `ColumnMap` already skip the NULL rows of the outer
mask; this change applies the same pattern to `ColumnStruct` for every
hash entry point that receives a null map. Rows without a null map keep
the existing column-wise batch path, so hashes of non-NULL rows are
unchanged.
### Release note
None
### Check List (For Author)
- Test:
- Unit Test: `ColumnStructTest.*RespectsOuterNullMap` cover crc32c
batch/single, xxHash batch/range and zlib-crc range hashing of a
nullable struct whose NULL rows carry different payloads.
- Regression test:
`query_p0/aggregate/aggregate_groupby_nullable_struct_local_shuffle`
groups `IF(cond, NULL, named_struct(...))` keys through a local hash
shuffle and checks the result matches the non-local-shuffle plan.
- Behavior changed: No
- Does this need documentation: No
https://claude.ai/code/session_01WTHEM5JUWYV8RqrDr9B62p
--
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]
