Uh oh!
There was an error while loading. Please reload this page.
Revert "GH-3574: Statistics.toParquetStatistics always set null_count (#3575)" - #3688
Conversation
…_count(apache#3575)" This reverts commit 63aebcc.
kevinjqliu
left a comment
There was a problem hiding this comment.
LGTM
The change is valid based on parquet-format, which recommends writing null_count even when it’s zero, but it doesn’t preserve backward compatibility and can cause problems for downstream consumers like Iceberg.
We should revert for now until we find a safer way to introduce this change.
wgtmac
left a comment
There was a problem hiding this comment.
I agree we need to revert that PR to unblock the release. But I do think that fix is valid in essence.
kevinjqliu
commented
Jul 27, 2026
lets reopen #3574 to track this |
Fokko
commented
Jul 27, 2026
I've re-opened the issue, and I agree @wgtmac, the change on its own makes sense, but currently we implicitly have either all the stats, or no stats at all. So we should evolve that contract 👍 |
This reverts commit 63aebcc.
Rationale for this change
What changes are included in this PR?
Are these changes tested?
Are there any user-facing changes?