mirror of
https://github.com/apache/impala.git
synced 2026-01-02 12:00:33 -05:00
e39de94316
I looked around some and I think having create/drop/show [aggregate] function seems reasonable and extends nicely for UDTs. The create aggregate function can accept a lot of arguments. The non-essential one, I went with resolving them by name rather than position (i.e. argName="value"). I think this is better for the user than specifying it by position. The grammar is: CREATE AGGREGATE <name>(<arg_types>) RETURNS <type> [INTERMEDIATE <type>] LOCATION '/path' UpdateFn='Fn' [comment='comment'] [SerializeFn='symbol'] [MergeFn='symbol'] [InitFn='symbol'] [FinalizeFn='symbol'] The optional args at the end can be in any order. If the other symbols are not specified, we derive them from the UpdateFn symbol that's required. The analyzer would try to figure it out and fail if we can't find the derived symbol in the binary. The simplest example would be: CREATE AGGREGATE FUNCTION count(float) RETURNS BIGINT LOCATION '/path' UpdateFn='CountUpdateFn'; In which case we assume the intermediate type is the return type and the other functions are called 'CountInitFn', 'CountSerializeFn', 'CountMergeFn' 'CountFinalizeFn'. Change-Id: Iefc5741293050f5b295df28e9d1a7d039ead8675 Reviewed-on: http://gerrit.ent.cloudera.com:8080/513 Reviewed-by: Nong Li <nong@cloudera.com> Tested-by: Nong Li <nong@cloudera.com>