Package io.trino.sql.planner.rowpattern

Class AggregateArgumentsRewriter

java.lang.Object

io.trino.sql.planner.rowpattern.AggregateArgumentsRewriter

public class AggregateArgumentsRewriter
extends Object

Method Summary

Modifier and Type	Method	Description
static List<Expression>	rewrite(List<Expression> arguments, Symbol classifierSymbol, Symbol matchNumberSymbol)	This rewriter is dedicated for aggregation arguments in pattern recognition context.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Method Details

rewrite

public static List<Expression> rewrite(List<Expression> arguments,
 Symbol classifierSymbol,
 Symbol matchNumberSymbol)

This rewriter is dedicated for aggregation arguments in pattern recognition context.

Rewrite expressions so that they do not contain any elements specific to row pattern recognition: - remove labels from `LabelDereference`s and `CLASSIFIER()` calls, - replace `CLASSIFIER()` and `MATCH_NUMBER()` calls with new symbols, - the expressions do not contain navigations by analysis.

NOTE: Unlike `LogicalIndexExtractor`, this rewriter does not re-allocate all symbols. The rewritten expressions contain all the original symbols, and additionally they contain new symbols replacing `CLASSIFIER()` and `MATCH_NUMBER()` calls. It is correct, because each of the expressions (by analysis) is effectively evaluated within a single row, so any expression optimizations based on symbols are applicable. Additionally, that makes the expressions eligible for pre-projection on the condition that there were no `CLASSIFIER()` or `MATCH_NUMBER()` calls. The `PushDownProjectionsFromPatternRecognition` rule pushes down argument computations and replaces them with single symbols.

Because the expressions are effectively evaluated within a single row, it is correct to replace all `CLASSIFIER()` calls with the same symbol and all `MATCH_NUMBER()` calls with the same symbol.