clintropolis commented on a change in pull request #10613:
URL: https://github.com/apache/druid/pull/10613#discussion_r553818153
##########
File path:
processing/src/main/java/org/apache/druid/segment/filter/ExpressionFilter.java
##########
@@ -55,6 +69,105 @@ public ExpressionFilter(final Supplier<Expr> expr, final
FilterTuning filterTuni
this.filterTuning = filterTuning;
}
+ @Override
+ public boolean canVectorizeMatcher(ColumnInspector inspector)
+ {
+ return expr.get().canVectorize(inspector);
+ }
+
+ @Override
+ public VectorValueMatcher makeVectorMatcher(VectorColumnSelectorFactory
factory)
+ {
+ final Expr theExpr = expr.get();
+
+ DruidPredicateFactory predicateFactory = new DruidPredicateFactory()
+ {
+ @Override
+ public Predicate<String> makeStringPredicate()
+ {
+ return Evals::asBoolean;
+ }
+
+ @Override
+ public DruidLongPredicate makeLongPredicate()
+ {
+ return Evals::asBoolean;
+ }
+
+ @Override
+ public DruidFloatPredicate makeFloatPredicate()
+ {
+ return Evals::asBoolean;
+ }
+
+ @Override
+ public DruidDoublePredicate makeDoublePredicate()
+ {
+ return Evals::asBoolean;
+ }
+
+ // The hashcode and equals are to make
SubclassesMustOverrideEqualsAndHashCodeTest stop complaining..
+ // DruidPredicateFactory doesn't really need equals or hashcode, in fact
only the 'toString' method is called
+ // when testing equality of DimensionPredicateFilter, so it's the truly
required method, but even that seems
+ // strange at best.
+ // Rather than tackle removing the annotation and reworking equality
tests for now, will leave this to refactor
+ // as part of https://github.com/apache/druid/issues/8256 which suggests
combining Filter and DimFilter
+ // implementations, which should clean up some of this mess.
+ @Override
+ public int hashCode()
+ {
+ return super.hashCode();
+ }
+
+ @Override
+ public boolean equals(Object obj)
+ {
+ return super.equals(obj);
+ }
+ };
+
+
+ final ExprType outputType = theExpr.getOutputType(factory);
+
+ if (outputType == null) {
+ // if an expression is vectorizable, but the output type is null, the
result will be null (or the default
+ // value in default mode) because expression is either all null
constants or missing columns
+
+ // in sql compatible mode, this means no matches ever, so just use the
false matcher:
+ if (NullHandling.sqlCompatible()) {
+ return new FalseVectorMatcher(factory.getVectorSizeInspector());
+ }
+ // in default mode, just fallback to using a long matcher since nearly
all boolean-ish expressions
+ // output a long value so it is probably a safe bet? idk, ending up here
by using all null-ish things
+ // in default mode is dancing on the edge of insanity anyway...
+ return
VectorValueMatcherColumnProcessorFactory.instance().makeLongProcessor(
+
ColumnCapabilitiesImpl.createSimpleNumericColumnCapabilities(ValueType.LONG),
+ ExpressionVectorSelectors.makeVectorValueSelector(factory, theExpr)
+ ).makeMatcher(predicateFactory);
+ }
+
+ switch (outputType) {
+ case LONG:
+ return
VectorValueMatcherColumnProcessorFactory.instance().makeLongProcessor(
+
ColumnCapabilitiesImpl.createSimpleNumericColumnCapabilities(ValueType.LONG),
+ ExpressionVectorSelectors.makeVectorValueSelector(factory, theExpr)
+ ).makeMatcher(predicateFactory);
+ case DOUBLE:
+ return
VectorValueMatcherColumnProcessorFactory.instance().makeDoubleProcessor(
+
ColumnCapabilitiesImpl.createSimpleNumericColumnCapabilities(ValueType.DOUBLE),
+ ExpressionVectorSelectors.makeVectorValueSelector(factory, theExpr)
+ ).makeMatcher(predicateFactory);
+ case STRING:
+ return
VectorValueMatcherColumnProcessorFactory.instance().makeObjectProcessor(
+ // using 'numeric' capabilities creator so we are configured to
NOT be dictionary encoded, etc
+
ColumnCapabilitiesImpl.createSimpleNumericColumnCapabilities(ValueType.STRING),
Review comment:
it is strange, but on purpose for the reason the comment says since the
numeric capabilities defaults are correct in this case. I will make sure all
current users are numeric types, and if so then just make these capabilites
from scratch because we should be counting on something geared towards numeric
behavior in case those change in some unexpected way. Or, if it is used by
other non-numbers then rename it to something more appropriate, like
`createSimpleSingleValueColumnCapabilites` or similar
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
