github-actions[bot] commented on code in PR #65253:
URL: https://github.com/apache/doris/pull/65253#discussion_r3586438749
##########
be/src/format/transformer/vorc_transformer.cpp:
##########
@@ -510,16 +573,13 @@ bool VOrcTransformer::_collect_column_bounds(const
orc::ColumnStatistics* col_st
} else if (const auto* decimal_stats =
dynamic_cast<const
orc::DecimalColumnStatistics*>(col_stats)) {
if (decimal_stats->hasMinimum() && decimal_stats->hasMaximum()) {
- has_bounds = true;
(*lower_bounds)[field_id] =
_decimal_to_bytes(decimal_stats->getMinimum());
Review Comment:
ORC decimal bounds still need to be normalized to Iceberg's signed unscaled
decimal bytes before committing. This branch passes
`decimal_stats->getMinimum()` / `getMaximum()` straight into
`_decimal_to_bytes()`, and that helper writes only `orc::Decimal::value`; it
never rescales the ORC statistic to the target `DecimalType::get_scale()`. The
current generated output already shows the scale bug for `decimal_col3
decimal(18,6)`: inserted/read values are `1234.567890` and `8765.432100`,
Parquet encodes field 8 as `0x00000000499602D2` / `0x000000020A75E124`, but ORC
commits `0x075BCD15` / `0x05397FB1`, which Iceberg readable metrics decode as
`123.456789` / `87.654321`. `_decimal_to_bytes()` also re-complements negative
`orc::Int128` values into positive magnitudes before serializing, so negative
decimal bounds would be committed with the wrong sign. Please rescale the ORC
statistic to the Iceberg decimal field scale and then serialize the signed
unscaled integer as two's-complement
minimal bytes; add regression coverage for both trailing-zero scale
normalization and a negative decimal bound.
--
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]
