Aleksandr Efimov has posted comments on this change. ( http://gerrit.cloudera.org:8080/24260 )
Change subject: IMPALA-14946: Catch "trunc" error within Calcite planner ...................................................................... Patch Set 2: (1 comment) http://gerrit.cloudera.org:8080/#/c/24260/2/java/calcite-planner/src/main/java/org/apache/impala/calcite/operators/ImpalaAdjustScaleFunction.java File java/calcite-planner/src/main/java/org/apache/impala/calcite/operators/ImpalaAdjustScaleFunction.java: http://gerrit.cloudera.org:8080/#/c/24260/2/java/calcite-planner/src/main/java/org/apache/impala/calcite/operators/ImpalaAdjustScaleFunction.java@63 PS2, Line 63: throw new RuntimeException("Invalid Truncate Unit for scale"); > Made the cleaner exception handling, added tests On master every one of these forms ends in NullPointerException: Cannot invoke "java.lang.Integer.intValue()" because "scale" is null; with PS3 the two cases the original planner also rejects come out word for word the same. Measured the rest on PS3 through both planners, frontend only: trunc(2.1, int_col) orig: must be called with a constant calcite: same trunc(2.1, null) orig: cannot be called with a NULL calcite: same trunc(2.1, cast(1 as int)) orig: plans fine calcite: must be called with a constant trunc(2.1, 1 + 1) orig: plans fine calcite: must be called with a constant trunc(2.1, cast(null as int)) orig: cannot be called with a NULL calcite: must be called with a constant round(2.1, cast(1 as int)) orig: plans fine calcite: must be called with a constant isOperandLiteral(1, false) and isOperandNull(1, false) count a cast of a literal as non-constant, while the original planner checks children_.get(1).isConstant(). Passing true for allowCast lines up both cast rows, and nothing else has to change for it - getOperandLiteralValue already unwraps a CAST in SqlCallBinding.valueAs. The "1 + 1" row needs the operand reduced first, which is probably beyond this patch. None of these are regressions; they were all NPEs before. Also worth knowing: for the query in the commit message, dayofyear(trunc('2014-11-11', string_col)), the frontend now says "trunc() must be called with a constant second argument" - the original planner gets through analysis there, and the second argument is a time unit rather than a scale. One question: this reaches the caller as java.lang.IllegalArgumentException rather than an AnalysisException - CalciteAnalysisDriver catches ImpalaException and CalciteContextException only. I went through parseAndAnalyze, so it may be converted higher up. Is it? -- To view, visit http://gerrit.cloudera.org:8080/24260 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-Project: Impala-ASF Gerrit-Branch: master Gerrit-MessageType: comment Gerrit-Change-Id: I766c138fae027ba0d919daa5cce38dc93d7144d1 Gerrit-Change-Number: 24260 Gerrit-PatchSet: 2 Gerrit-Owner: Steve Carlin <[email protected]> Gerrit-Reviewer: Aleksandr Efimov <[email protected]> Gerrit-Reviewer: Aman Sinha <[email protected]> Gerrit-Reviewer: Impala Public Jenkins <[email protected]> Gerrit-Reviewer: Joe McDonnell <[email protected]> Gerrit-Reviewer: Michael Smith <[email protected]> Gerrit-Reviewer: Quanlong Huang <[email protected]> Gerrit-Reviewer: Steve Carlin <[email protected]> Gerrit-Comment-Date: Sun, 23 Aug 2026 21:16:44 +0000 Gerrit-HasComments: Yes
