/// <summary>
/// Remove any columns from a report that are not required to achieve a rollup result.
/// </summary>
/// <remarks>
/// If columns get removed here then the query optimiser will later remove various joins.
/// This can affect whether some rows are repeated.
/// Some aggregate types (e.g. max/min) are unaffected by this.
/// Some types (e.g. Count) are very affected by this, so we replace columns with simpler ones, rather than removing them completely.
/// Some types (e.g. Sum) in principle could be affected by this, but in practice are OK because they will reference the relationship branch that is relevant to them anyway.
/// </remarks>
/// <param name="query">The original query.</param>
/// <param name="clientAggregate">Aggregate settings that are used to determine what columns are used.</param>
/// <returns>A clone of the query, with unused columns removed.</returns>
public static StructuredQuery RemoveUnusedColumns(StructuredQuery query, ClientAggregate clientAggregate, bool supportQuickSearch = false)
{
StructuredQuery queryCopy = query.DeepCopy( );
// Determine columns that are used by the rollup
HashSet <Guid> referencedColumns = new HashSet <Guid>(clientAggregate.AggregatedColumns.Select(a => a.ReportColumnId)
.Concat(clientAggregate.GroupedColumns.Select(g => g.ReportColumnId)));
// Also include columns that are referenced by analyzer conditions
foreach (QueryCondition condition in query.Conditions)
{
ColumnReference colRefExpr = condition.Expression as ColumnReference;
if (colRefExpr == null)
{
continue;
}
referencedColumns.Add(colRefExpr.ColumnId);
}
// Ensure that the inner report returns at least something. (This will typically be the ID column).
if (referencedColumns.Count == 0 && query.SelectColumns.Count > 0)
{
referencedColumns.Add(query.SelectColumns [0].ColumnId);
}
// There are two types of optimisations. Either we can just pull out all unused columns, and let the structured query optimiser
// remove the subsequent relationship joins - which is OK for things like max & min in particular.
// But in some cases (e.g. for Count) we need to ensure we capture all relationships to get the true fanout.
// This is safer, but less efficient.
bool strictlyMaintainRowPresence = clientAggregate.AggregatedColumns.Any(ag => ag.AggregateMethod == AggregateMethod.Count);
if (!strictlyMaintainRowPresence)
{
// Remove all unused columns
queryCopy.SelectColumns.RemoveAll(column => !referencedColumns.Contains(column.ColumnId));
}
else
{
// Visit each column and determine if it can be removed, or converted to a simpler type
List <SelectColumn> columns = queryCopy.SelectColumns;
List <Guid> toRemove = new List <Guid>( );
for (int i = 0; i < columns.Count; i++)
{
SelectColumn column = columns [i];
if (referencedColumns.Contains(column.ColumnId))
{
continue;
}
// Replace field lookups with an equivalent ID column .. to maintain the relationship join, but remove the field join.
ResourceDataColumn fieldColumnExpr = column.Expression as ResourceDataColumn;
if (fieldColumnExpr != null)
{
// TODO: we could delete this column entire IF the entire relationship path to it is 'to one' relationships, without any advance properties to enforce rows.
// UPDATE: for quick serach purpose, the column expression should be ResourceDataColumn but skip in select clause.
// however for performance reasons, if without quick search, still change to IdExpression
if (supportQuickSearch)
{
column.IsHidden = true;
}
else
{
column.Expression = new IdExpression {
NodeId = fieldColumnExpr.NodeId
}
};
continue;
}
// Remove aggregate expressions if they don't have group-bys. (A group-by would cause the aggregate to return more than one row).
AggregateExpression aggExpr = column.Expression as AggregateExpression;
if (aggExpr != null)
{
AggregateEntity aggNode = StructuredQueryHelper.FindNode(queryCopy.RootEntity, aggExpr.NodeId) as AggregateEntity;
if (aggNode != null)
{
if (aggNode.GroupBy == null || aggNode.GroupBy.Count == 0)
{
toRemove.Add(column.ColumnId);
}
}
}
// Would be nice to remove calculated columns .. but probably too risky
}
queryCopy.SelectColumns.RemoveAll(column => toRemove.Contains(column.ColumnId));
}
// Remove any obsolete order-by instructions
queryCopy.OrderBy.RemoveAll(orderBy =>
{
ColumnReference colRefExpr = orderBy.Expression as ColumnReference;
if (colRefExpr == null)
{
return(false);
}
bool colStillPresent = queryCopy.SelectColumns.Any(column => column.ColumnId == colRefExpr.ColumnId);
return(!colStillPresent);
});
return(queryCopy);
}
}
