private void checkJoinColumn(String context, String side, Schema schema, String colName) {
int count = schema.numColumnsWithName(colName);
if (count == 0) {
throw new SchemaNameException(
context + " error: column name \"" + colName + "\" doesn't appear on " + side);
} else if (count > 1) {
throw new SchemaNameException(
context + " error: column name \"" + colName + "\" is ambiguous on " + side);
}
}
/**
* This helper function handles a <tt>NATURAL</tt> join or a join with a <tt>USING</tt> clause,
* determining what columns to use in the join, computing the join predicate to use, and computing
* the schema of the join result.
*
* <p>The schema of the result follows the SQL standard:
*
* <ul>
* <li>Common columns first, in the order specified, if any.
* <li>Columns that appear only in the left table, if any.
* <li>Columns that appear only in the right table, if any.
* </ul>
*
* <p>Of course, any of these sets of columns could be empty.
*
* @param context a string used in logging messages and exceptions to indicate the context in
* which this method was called.
* @param leftSchema the schema of the table on the left side of the join
* @param rightSchema the schema of the table on the right side of the join
* @param commonCols the set of common columns to use
* @param result this output-parameter is updated to hold the schema of the result, as produced by
* the join operation.
*/
private void buildJoinSchema(
String context,
Schema leftSchema,
Schema rightSchema,
Set<String> commonCols,
Schema result) {
preparedJoinExpr = null;
preparedSelectValues = null;
if (!commonCols.isEmpty()) {
preparedSelectValues = new ArrayList<SelectValue>();
// We will need to generate a join expression using the common
// columns. We will also need a project-spec that will project down
// to only one copy of the common columns.
BooleanOperator andOp = new BooleanOperator(BooleanOperator.Type.AND_EXPR);
// Handle the shared columns. We need to check that the
// names aren't ambiguous on one or the other side.
for (String name : commonCols) {
checkJoinColumn(context, "left", leftSchema, name);
checkJoinColumn(context, "right", leftSchema, name);
// Seems OK, so add it to the result.
ColumnInfo lhsColInfo = leftSchema.getColumnInfo(name);
ColumnInfo rhsColInfo = rightSchema.getColumnInfo(name);
result.addColumnInfo(new ColumnInfo(lhsColInfo.getName(), lhsColInfo.getType()));
// Add an equality test between the common columns to the join
// condition.
CompareOperator eq =
new CompareOperator(
CompareOperator.Type.EQUALS,
new ColumnValue(lhsColInfo.getColumnName()),
new ColumnValue(rhsColInfo.getColumnName()));
andOp.addTerm(eq);
// Add a select-value that projects the appropriate source
// column down to the common column.
SelectValue selVal;
switch (joinType) {
case INNER:
case LEFT_OUTER:
// We can use the left column in the result, as it will
// always be non-NULL.
selVal = new SelectValue(new ColumnValue(lhsColInfo.getColumnName()), name);
preparedSelectValues.add(selVal);
break;
case RIGHT_OUTER:
// We can use the right column in the result, as it will
// always be non-NULL.
selVal = new SelectValue(new ColumnValue(rhsColInfo.getColumnName()), name);
preparedSelectValues.add(selVal);
break;
case FULL_OUTER:
// In this case, the LHS column-value could be null, or the
// RHS column-value could be null. Thus, we need to produce
// a result of COALESCE(lhs.col, rhs.col) AS col.
Expression coalesce =
new FunctionCall(
"COALESCE",
new ColumnValue(lhsColInfo.getColumnName()),
new ColumnValue(rhsColInfo.getColumnName()));
selVal = new SelectValue(coalesce, name);
preparedSelectValues.add(selVal);
}
}
preparedJoinExpr = andOp;
}
// Handle the non-shared columns
for (ColumnInfo colInfo : leftSchema) {
if (!commonCols.contains(colInfo.getName())) {
result.addColumnInfo(colInfo);
SelectValue selVal = new SelectValue(new ColumnValue(colInfo.getColumnName()), null);
preparedSelectValues.add(selVal);
}
}
for (ColumnInfo colInfo : rightSchema) {
if (!commonCols.contains(colInfo.getName())) {
result.addColumnInfo(colInfo);
SelectValue selVal = new SelectValue(new ColumnValue(colInfo.getColumnName()), null);
preparedSelectValues.add(selVal);
}
}
}
/**
* This function prepares the from-clause for use by the planner to generate an actual execution
* plan from the clause. This includes several important tasks:
*
* <ul>
* <li>The schema of the from-clause's output is determined.
* <li>If the from-clause is a join, and the join specifies <tt>NATURAL</tt>, <tt>USING (col,
* ...)</tt>, or an <tt>ON</tt> clause, the join condition is determined
* </ul>
*
* <p>Since a from-clause can be a SQL subquery or a join expression, it should be evident that
* the method will recursively prepare its children as well.
*
* <p><b>This method really shouldn't be called directly.</b> Instead, the enclosing {@link
* SelectClause} will calls this method when the clause's {@link SelectClause#computeSchema}
* method is invoked.
*
* @return the schema of the from-clause's output
* @throws IOException if a table's schema cannot be loaded, or some other IO issue occurs.
*/
public Schema prepare() throws IOException {
Schema result = null;
switch (clauseType) {
case BASE_TABLE:
logger.debug("Preparing BASE_TABLE from-clause.");
TableFileInfo tblFileInfo = StorageManager.getInstance().openTable(tableName);
result = tblFileInfo.getSchema();
if (aliasName != null) {
// Make a copy of the result schema and change the table names.
result = new Schema(result);
result.setTableName(aliasName);
}
break;
case SELECT_SUBQUERY:
logger.debug("Preparing SELECT_SUBQUERY from-clause.");
result = derivedTable.computeSchema();
assert aliasName != null;
// Make a copy of the result schema and change the table names.
result = new Schema(result);
result.setTableName(aliasName);
break;
case JOIN_EXPR:
logger.debug("Preparing JOIN_EXPR from-clause. Condition type = " + condType);
result = new Schema();
Schema leftSchema = leftChild.prepare();
Schema rightSchema = rightChild.prepare();
// Depending on the join type, we might eliminate duplicate column
// names.
if (condType == JoinConditionType.NATURAL_JOIN) {
// Make sure that each side of the join doesn't have any
// duplicate column names. (A schema can have columns with the
// same name, but from different table names. This is not
// allowed for natural joins.)
if (leftSchema.hasMultipleColumnsWithSameName()) {
throw new SchemaNameException(
"Natural join error: "
+ "left child table has multiple columns with same "
+ "column name");
}
if (rightSchema.hasMultipleColumnsWithSameName()) {
throw new SchemaNameException(
"Natural join error: "
+ "right child table has multiple columns with same "
+ "column name");
}
Set<String> commonCols = leftSchema.getCommonColumnNames(rightSchema);
if (commonCols.isEmpty()) {
// TODO: According to the SQL99 standard, this shouldn't
// generate an error.
throw new SchemaNameException(
"Natural join error: " + "child tables share no common column names!");
}
buildJoinSchema("Natural join", leftSchema, rightSchema, commonCols, result);
} else if (condType == JoinConditionType.JOIN_USING) {
LinkedHashSet<String> commonCols = new LinkedHashSet<String>();
for (String name : joinUsingNames) {
if (!commonCols.add(name)) {
throw new SchemaNameException(
"Column name " + name + " was specified multiple times in USING clause");
}
}
buildJoinSchema("Join USING", leftSchema, rightSchema, commonCols, result);
} else {
// This join condition-type doesn't alter the result schema.
// Just lump together the result schemas.
result.append(leftSchema);
result.append(rightSchema);
preparedJoinExpr = joinOnExpr;
}
}
// Don't need to do any schema validation in this function, since all the
// validation takes place in the SELECT clause schema-computation code.
preparedSchema = result;
return result;
}