[ 
https://issues.apache.org/jira/browse/PARQUET-1655?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16932644#comment-16932644
 ] 

Deepak Majeti commented on PARQUET-1655:
----------------------------------------

The Decimal values comparator indeed needs to be fixed.  The java code (below) 
has the decimal comparator and additionally handles decimals of different 
lengths via padding.  (I don't see the msb flip but I know it is one way to 
compare twos-complement binary )

[https://github.com/apache/parquet-mr/blob/master/parquet-column/src/main/java/org/apache/parquet/schema/PrimitiveComparator.java#L230]

[https://github.com/apache/parquet-mr/blob/master/parquet-column/src/main/java/org/apache/parquet/schema/PrimitiveType.java#L379]

 

 

 

> [C++] Decimal comparisons used for min/max statistics are not correct
> ---------------------------------------------------------------------
>
>                 Key: PARQUET-1655
>                 URL: https://issues.apache.org/jira/browse/PARQUET-1655
>             Project: Parquet
>          Issue Type: Bug
>          Components: parquet-cpp
>            Reporter: Philip Felton
>            Priority: Major
>
> The [Parquet Format 
> specifications|https://github.com/apache/parquet-format/blob/master/LogicalTypes.md]
>  says
> bq. If the column uses int32 or int64 physical types, then signed comparison 
> of the integer values produces the correct ordering. If the physical type is 
> fixed, then the correct ordering can be produced by flipping the 
> most-significant bit in the first byte and then using unsigned byte-wise 
> comparison.
> However this isn't followed in the C++ Parquet code. 16-byte decimal 
> comparison is implemented using a lexicographical comparison of signed chars.
> This appears to be because the function 
> [https://github.com/apache/arrow/blob/master/cpp/src/parquet/statistics.cc#L183]
>  just goes off the sort_order (signed) and physical_type 
> (FIXED_LENGTH_BYTE_ARRAY), there is no override for decimal.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to