use of org.apache.drill.exec.expr.ValueVectorReadExpression in project drill by apache.
the class ProjectRecordBatch method setupNewSchema.
@Override
protected boolean setupNewSchema() throws SchemaChangeException {
if (allocationVectors != null) {
for (final ValueVector v : allocationVectors) {
v.clear();
}
}
this.allocationVectors = Lists.newArrayList();
if (complexWriters != null) {
container.clear();
} else {
container.zeroVectors();
}
final List<NamedExpression> exprs = getExpressionList();
final ErrorCollector collector = new ErrorCollectorImpl();
final List<TransferPair> transfers = Lists.newArrayList();
final ClassGenerator<Projector> cg = CodeGenerator.getRoot(Projector.TEMPLATE_DEFINITION, context.getFunctionRegistry(), context.getOptions());
cg.getCodeGenerator().plainJavaCapable(true);
// Uncomment out this line to debug the generated code.
// cg.getCodeGenerator().saveCodeForDebugging(true);
final IntHashSet transferFieldIds = new IntHashSet();
final boolean isAnyWildcard = isAnyWildcard(exprs);
final ClassifierResult result = new ClassifierResult();
final boolean classify = isClassificationNeeded(exprs);
for (int i = 0; i < exprs.size(); i++) {
final NamedExpression namedExpression = exprs.get(i);
result.clear();
if (classify && namedExpression.getExpr() instanceof SchemaPath) {
classifyExpr(namedExpression, incoming, result);
if (result.isStar) {
// The value indicates which wildcard we are processing now
final Integer value = result.prefixMap.get(result.prefix);
if (value != null && value.intValue() == 1) {
int k = 0;
for (final VectorWrapper<?> wrapper : incoming) {
final ValueVector vvIn = wrapper.getValueVector();
if (k > result.outputNames.size() - 1) {
assert false;
}
// get the renamed column names
final String name = result.outputNames.get(k++);
if (name == EMPTY_STRING) {
continue;
}
if (isImplicitFileColumn(vvIn)) {
continue;
}
final FieldReference ref = new FieldReference(name);
final ValueVector vvOut = container.addOrGet(MaterializedField.create(ref.getAsNamePart().getName(), vvIn.getField().getType()), callBack);
final TransferPair tp = vvIn.makeTransferPair(vvOut);
transfers.add(tp);
}
} else if (value != null && value.intValue() > 1) {
// subsequent wildcards should do a copy of incoming valuevectors
int k = 0;
for (final VectorWrapper<?> wrapper : incoming) {
final ValueVector vvIn = wrapper.getValueVector();
final SchemaPath originalPath = SchemaPath.getSimplePath(vvIn.getField().getPath());
if (k > result.outputNames.size() - 1) {
assert false;
}
// get the renamed column names
final String name = result.outputNames.get(k++);
if (name == EMPTY_STRING) {
continue;
}
if (isImplicitFileColumn(vvIn)) {
continue;
}
final LogicalExpression expr = ExpressionTreeMaterializer.materialize(originalPath, incoming, collector, context.getFunctionRegistry());
if (collector.hasErrors()) {
throw new SchemaChangeException(String.format("Failure while trying to materialize incoming schema. Errors:\n %s.", collector.toErrorString()));
}
final MaterializedField outputField = MaterializedField.create(name, expr.getMajorType());
final ValueVector vv = container.addOrGet(outputField, callBack);
allocationVectors.add(vv);
final TypedFieldId fid = container.getValueVectorId(SchemaPath.getSimplePath(outputField.getPath()));
final ValueVectorWriteExpression write = new ValueVectorWriteExpression(fid, expr, true);
final HoldingContainer hc = cg.addExpr(write, ClassGenerator.BlkCreateMode.TRUE_IF_BOUND);
}
}
continue;
}
} else {
// For the columns which do not needed to be classified,
// it is still necessary to ensure the output column name is unique
result.outputNames = Lists.newArrayList();
final String outputName = getRef(namedExpression).getRootSegment().getPath();
addToResultMaps(outputName, result, true);
}
String outputName = getRef(namedExpression).getRootSegment().getPath();
if (result != null && result.outputNames != null && result.outputNames.size() > 0) {
boolean isMatched = false;
for (int j = 0; j < result.outputNames.size(); j++) {
if (!result.outputNames.get(j).equals(EMPTY_STRING)) {
outputName = result.outputNames.get(j);
isMatched = true;
break;
}
}
if (!isMatched) {
continue;
}
}
final LogicalExpression expr = ExpressionTreeMaterializer.materialize(namedExpression.getExpr(), incoming, collector, context.getFunctionRegistry(), true, unionTypeEnabled);
final MaterializedField outputField = MaterializedField.create(outputName, expr.getMajorType());
if (collector.hasErrors()) {
throw new SchemaChangeException(String.format("Failure while trying to materialize incoming schema. Errors:\n %s.", collector.toErrorString()));
}
// add value vector to transfer if direct reference and this is allowed, otherwise, add to evaluation stack.
if (expr instanceof ValueVectorReadExpression && incoming.getSchema().getSelectionVectorMode() == SelectionVectorMode.NONE && !((ValueVectorReadExpression) expr).hasReadPath() && !isAnyWildcard && !transferFieldIds.contains(((ValueVectorReadExpression) expr).getFieldId().getFieldIds()[0])) {
final ValueVectorReadExpression vectorRead = (ValueVectorReadExpression) expr;
final TypedFieldId id = vectorRead.getFieldId();
final ValueVector vvIn = incoming.getValueAccessorById(id.getIntermediateClass(), id.getFieldIds()).getValueVector();
Preconditions.checkNotNull(incoming);
final FieldReference ref = getRef(namedExpression);
final ValueVector vvOut = container.addOrGet(MaterializedField.create(ref.getAsUnescapedPath(), vectorRead.getMajorType()), callBack);
final TransferPair tp = vvIn.makeTransferPair(vvOut);
transfers.add(tp);
transferFieldIds.add(vectorRead.getFieldId().getFieldIds()[0]);
} else if (expr instanceof DrillFuncHolderExpr && ((DrillFuncHolderExpr) expr).getHolder().isComplexWriterFuncHolder()) {
// Lazy initialization of the list of complex writers, if not done yet.
if (complexWriters == null) {
complexWriters = Lists.newArrayList();
} else {
complexWriters.clear();
}
// The reference name will be passed to ComplexWriter, used as the name of the output vector from the writer.
((DrillFuncHolderExpr) expr).getFieldReference(namedExpression.getRef());
cg.addExpr(expr, ClassGenerator.BlkCreateMode.TRUE_IF_BOUND);
if (complexFieldReferencesList == null) {
complexFieldReferencesList = Lists.newArrayList();
}
// save the field reference for later for getting schema when input is empty
complexFieldReferencesList.add(namedExpression.getRef());
} else {
// need to do evaluation.
final ValueVector vector = container.addOrGet(outputField, callBack);
allocationVectors.add(vector);
final TypedFieldId fid = container.getValueVectorId(SchemaPath.getSimplePath(outputField.getPath()));
final boolean useSetSafe = !(vector instanceof FixedWidthVector);
final ValueVectorWriteExpression write = new ValueVectorWriteExpression(fid, expr, useSetSafe);
final HoldingContainer hc = cg.addExpr(write, ClassGenerator.BlkCreateMode.TRUE_IF_BOUND);
// We cannot do multiple transfers from the same vector. However we still need to instantiate the output vector.
if (expr instanceof ValueVectorReadExpression) {
final ValueVectorReadExpression vectorRead = (ValueVectorReadExpression) expr;
if (!vectorRead.hasReadPath()) {
final TypedFieldId id = vectorRead.getFieldId();
final ValueVector vvIn = incoming.getValueAccessorById(id.getIntermediateClass(), id.getFieldIds()).getValueVector();
vvIn.makeTransferPair(vector);
}
}
logger.debug("Added eval for project expression.");
}
}
try {
CodeGenerator<Projector> codeGen = cg.getCodeGenerator();
codeGen.plainJavaCapable(true);
// Uncomment out this line to debug the generated code.
// codeGen.saveCodeForDebugging(true);
this.projector = context.getImplementationClass(codeGen);
projector.setup(context, incoming, this, transfers);
} catch (ClassTransformationException | IOException e) {
throw new SchemaChangeException("Failure while attempting to load generated class", e);
}
if (container.isSchemaChanged()) {
container.buildSchema(SelectionVectorMode.NONE);
return true;
} else {
return false;
}
}
Aggregations