mirror of
https://github.com/apache/impala.git
synced 2025-12-19 18:12:08 -05:00
52334ba426
The 'case' function resolver in the original Impala planner has a quirk in it which caused issues in the Calcite planner. The function resolver for the original planner resolves all case statements with the "boolean" version. Later on, in the analysis of the CaseExpr, the proper types are assessed and the necessary casting is added. The Calcite planner follows a similar path. The resolver always returns boolean as well and the coerce nodes module determines the proper return type for the case statement. Two other related issues are also fixed here: Literal strings should be treated as type STRING instead of CHAR(X), but a null should literal should not be changed from a CHAR(x) to a STRING. This broke a 'case' test in the test framework where the columns were non-literals with type char(x), and the return value was a "null" which should not have forced a cast to string. A cast from a varchar to a varchar should be ignored. Testing: Added a test to calcite.test. Ensured the existing cast test in test_chars.py passed. Ran through the Jenkins Calcite testing framework. Change-Id: I82d657f4bfce432c458ee8198188dadf9f23f2ef Reviewed-on: http://gerrit.cloudera.org:8080/23560 Reviewed-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com> Tested-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com>