Quanlong Huang created ORC-968:
----------------------------------

             Summary: Column names used to build SearchArgument should be full 
path names
                 Key: ORC-968
                 URL: https://issues.apache.org/jira/browse/ORC-968
             Project: ORC
          Issue Type: Bug
          Components: C++
    Affects Versions: 1.7.0
            Reporter: Quanlong Huang


The C++ reader provides the same interfaces as those in hive-storage-api for 
building SearchArguments, e.g.
{code:cpp}
virtual SearchArgumentBuilder& lessThan(const std::string& column,
                                        PredicateDataType type,
                                        Literal literal) = 0;
{code}
C++: 
[https://github.com/apache/orc/blob/2143841e24abb2e0fef1a3396376682fc3bb6fea/c%2B%2B/include/orc/sargs/SearchArgument.hh#L90-L92]
 Java: 
[https://github.com/apache/hive/blob/620b2b197269041d7f508bd0e4564ed8e5edfcfd/storage-api/src/java/org/apache/hadoop/hive/ql/io/sarg/SearchArgument.java#L219-L220]

However, the column strings in Java codes are treated as full path names 
instead of the leaf field names. E.g. for a nested column 'id' inside a struct 
type column 's', the name string is expected to be 's.id' instead of 'id'.

More details about the string format:
{quote}Find a subtype of this schema by name. If the name is a simple integer, 
it will be used as a column number. Otherwise, this routine will recursively 
search for the name. * Struct fields are selected by name.
 * List children are selected by "_elem".
 * Map children are selected by "_key" or "_value".
 * Union children are selected by number starting at 0.

Names are separated by '.'.
{quote}
[https://javadoc.io/doc/org.apache.orc/orc-core/1.6.10/org/apache/orc/TypeDescription.html#findSubtype-java.lang.String-]

The C++ library currently treats the name strings as struct field names and 
recursively searching it in SargsApplier::findColumn
 
[https://github.com/apache/orc/blob/2143841e24abb2e0fef1a3396376682fc3bb6fea/c%2B%2B/src/sargs/SargsApplier.cc#L25-L39]

To be consistent with Java codes and to avoid ambiguous selection (if there are 
identical sub-field names in different struct columns), we should adjust the 
C++ library to treat name strings as full path names. It should also deal with 
special names like "_elem", "_key", and "_value".



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

Reply via email to