If no one has any objection, I'm going to update HIVE-4022, which I entered a week ago when I thought the behavior was Avro-specific, to indicate it actually affects even native Hive tables.
https://issues.apache.org/jira/browse/HIVE-4022 --- On Fri, 2/15/13, Michael Malak <michaelma...@yahoo.com> wrote: > From: Michael Malak <michaelma...@yahoo.com> > Subject: NULLable STRUCTs > To: user@hive.apache.org > Date: Friday, February 15, 2013, 5:03 PM > It seems that all Hive columns (at > least those of primitive types) are always NULLable? > What about columns of type STRUCT? > > The following: > > echo 1,2 >twovalues.csv > hive > CREATE TABLE tc (x INT, y INT) ROW FORMAT DELIMITED FIELDS > TERMINATED BY ','; > LOAD DATA LOCAL INPATH 'twovalues.csv' INTO TABLE tc; > CREATE TABLE oc (z STRUCT<a: int, b: int>); > INSERT INTO TABLE oc SELECT null FROM tc; > > produces the error > > FAILED: SemanticException [Error 10044]: Line 1:18 Cannot > insert into target table because column number/types are > different 'oc': Cannot convert column 0 from void to > struct<a:int,b:int>. > > I initially discovered such behavior with Avro-backed > tables, and even entered a JIRA > https://issues.apache.org/jira/browse/HIVE-4022 > but now I realized it happens with CSV-backed tables as > well. > > Perhaps related, perhaps not, it seems that all members of a > STRUCT are always non-NULLable. The following: > > INSERT INTO TABLE oc SELECT named_struct('a', null, 'b', > null) FROM tc; > > produces the error: > > FAILED: SemanticException [Error 10044]: Line 1:18 Cannot > insert into target table because column number/types are > different 'oc': Cannot convert column 0 from > struct<a:void,b:void> to struct<a:int,b:int>. > >
