/**
* A normal reduce operator's rowObjectInspector looks like a struct containing nested key/value
* structs that contain the column values: { key: { reducesinkkey0:int }, value: { _col0:int,
* _col1:int, .. } }
*
* <p>While the rowObjectInspector looks the same for vectorized queries during compilation time,
* within the tasks at query execution the rowObjectInspector has changed to a flatter structure
* without nested key/value structs: { 'key.reducesinkkey0':int, 'value._col0':int,
* 'value._col1':int, .. }
*
* <p>Trying to fetch 'key.reducesinkkey0' by name from the list of flattened ObjectInspectors
* does not work because the '.' gets interpreted as a field member, even though it is a flattened
* list of column values. This workaround converts the column name referenced in the ExprNodeDesc
* from a nested field name (key.reducesinkkey0) to key_reducesinkkey0, simply by replacing '.'
* with '_'.
*
* @param source
* @return
*/
public static ExprNodeDesc flattenExpr(ExprNodeDesc source) {
if (source instanceof ExprNodeGenericFuncDesc) {
// all children expression should be resolved
ExprNodeGenericFuncDesc function = (ExprNodeGenericFuncDesc) source.clone();
List<ExprNodeDesc> newChildren = flattenExprList(function.getChildren());
for (ExprNodeDesc newChild : newChildren) {
if (newChild == null) {
// Could not resolve all of the function children, fail
return null;
}
}
function.setChildren(newChildren);
return function;
}
if (source instanceof ExprNodeColumnDesc) {
ExprNodeColumnDesc column = (ExprNodeColumnDesc) source;
// Create a new ColumnInfo, replacing STRUCT.COLUMN with STRUCT_COLUMN
String newColumn = column.getColumn().replace('.', '_');
return new ExprNodeColumnDesc(source.getTypeInfo(), newColumn, column.getTabAlias(), false);
}
if (source instanceof ExprNodeFieldDesc) {
// field expression should be resolved
ExprNodeFieldDesc field = (ExprNodeFieldDesc) source.clone();
ExprNodeDesc fieldDesc = flattenExpr(field.getDesc());
if (fieldDesc == null) {
return null;
}
field.setDesc(fieldDesc);
return field;
}
// constant or null expr, just return
return source;
}
public static ArrayList<ExprNodeDesc> clone(List<ExprNodeDesc> sources) {
ArrayList<ExprNodeDesc> result = new ArrayList<ExprNodeDesc>();
for (ExprNodeDesc expr : sources) {
result.add(expr.clone());
}
return result;
}
public static ExprNodeDesc backtrack(
ExprNodeDesc source, Operator<?> current, Operator<?> terminal, boolean foldExpr)
throws SemanticException {
Operator<?> parent = getSingleParent(current, terminal);
if (parent == null) {
return source;
}
if (source instanceof ExprNodeGenericFuncDesc) {
// all children expression should be resolved
ExprNodeGenericFuncDesc function = (ExprNodeGenericFuncDesc) source.clone();
List<ExprNodeDesc> children = backtrack(function.getChildren(), current, terminal, foldExpr);
for (ExprNodeDesc child : children) {
if (child == null) {
// Could not resolve all of the function children, fail
return null;
}
}
function.setChildren(children);
if (foldExpr) {
// fold after replacing, if possible
ExprNodeDesc foldedFunction = ConstantPropagateProcFactory.foldExpr(function);
if (foldedFunction != null) {
return foldedFunction;
}
}
return function;
}
if (source instanceof ExprNodeColumnDesc) {
ExprNodeColumnDesc column = (ExprNodeColumnDesc) source;
return backtrack(column, parent, terminal);
}
if (source instanceof ExprNodeFieldDesc) {
// field expression should be resolved
ExprNodeFieldDesc field = (ExprNodeFieldDesc) source.clone();
ExprNodeDesc fieldDesc = backtrack(field.getDesc(), current, terminal, foldExpr);
if (fieldDesc == null) {
return null;
}
field.setDesc(fieldDesc);
return field;
}
// constant or null expr, just return
return source;
}