Examples with ErrorCollectorImpl org.apache.drill.common.expression.ErrorCollectorImpl used on opensource projects

Example 41 with ErrorCollectorImpl

use of org.apache.drill.common.expression.ErrorCollectorImpl in project drill by apache.

the class TopNBatch method createNewPriorityQueue.

public static PriorityQueue createNewPriorityQueue(MappingSet mainMapping, MappingSet leftMapping, MappingSet rightMapping, List<Ordering> orderings, VectorAccessible batch, boolean unionTypeEnabled, boolean codegenDump, int limit, BufferAllocator allocator, SelectionVectorMode mode, FragmentContext context) {
    OptionSet optionSet = context.getOptions();
    FunctionLookupContext functionLookupContext = context.getFunctionRegistry();
    CodeGenerator<PriorityQueue> cg = CodeGenerator.get(PriorityQueue.TEMPLATE_DEFINITION, optionSet);
    cg.plainJavaCapable(true);
    cg.saveCodeForDebugging(codegenDump);
    // Uncomment out this line to debug the generated code.
    // cg.saveCodeForDebugging(true);
    ClassGenerator<PriorityQueue> g = cg.getRoot();
    g.setMappingSet(mainMapping);
    for (Ordering od : orderings) {
        // first, we rewrite the evaluation stack for each side of the comparison.
        ErrorCollector collector = new ErrorCollectorImpl();
        final LogicalExpression expr = ExpressionTreeMaterializer.materialize(od.getExpr(), batch, collector, functionLookupContext, unionTypeEnabled);
        collector.reportErrors(logger);
        g.setMappingSet(leftMapping);
        HoldingContainer left = g.addExpr(expr, ClassGenerator.BlkCreateMode.FALSE);
        g.setMappingSet(rightMapping);
        HoldingContainer right = g.addExpr(expr, ClassGenerator.BlkCreateMode.FALSE);
        g.setMappingSet(mainMapping);
        // next we wrap the two comparison sides and add the expression block for the comparison.
        LogicalExpression fh = FunctionGenerationHelper.getOrderingComparator(od.nullsSortHigh(), left, right, functionLookupContext);
        HoldingContainer out = g.addExpr(fh, ClassGenerator.BlkCreateMode.FALSE);
        JConditional jc = g.getEvalBlock()._if(out.getValue().ne(JExpr.lit(0)));
        if (od.getDirection() == Direction.ASCENDING) {
            jc._then()._return(out.getValue());
        } else {
            jc._then()._return(out.getValue().minus());
        }
        g.rotateBlock();
    }
    g.rotateBlock();
    g.getEvalBlock()._return(JExpr.lit(0));
    PriorityQueue q = context.getImplementationClass(cg);
    try {
        q.init(limit, allocator, mode == BatchSchema.SelectionVectorMode.TWO_BYTE);
    } catch (SchemaChangeException e) {
        throw TopNBatch.schemaChangeException(e, "Top N", logger);
    }
    return q;
}

Example 42 with ErrorCollectorImpl

use of org.apache.drill.common.expression.ErrorCollectorImpl in project drill by apache.

the class ChainedHashTable method createAndSetupHashTable.

public HashTable createAndSetupHashTable(TypedFieldId[] outKeyFieldIds) throws ClassTransformationException, IOException, SchemaChangeException {
    CodeGenerator<HashTable> top = CodeGenerator.get(HashTable.TEMPLATE_DEFINITION, context.getOptions());
    top.plainJavaCapable(true);
    // Uncomment out this line to debug the generated code.
    // This code is called from generated code, so to step into this code,
    // persist the code generated in HashAggBatch also.
    // top.saveCodeForDebugging(true);
    // use a subclass
    top.preferPlainJava(true);
    ClassGenerator<HashTable> cg = top.getRoot();
    ClassGenerator<HashTable> cgInner = cg.getInnerGenerator("BatchHolder");
    LogicalExpression[] keyExprsBuild = new LogicalExpression[htConfig.getKeyExprsBuild().size()];
    LogicalExpression[] keyExprsProbe = null;
    boolean isProbe = (htConfig.getKeyExprsProbe() != null);
    if (isProbe) {
        keyExprsProbe = new LogicalExpression[htConfig.getKeyExprsProbe().size()];
    }
    ErrorCollector collector = new ErrorCollectorImpl();
    // original ht container from which others may be cloned
    VectorContainer htContainerOrig = new VectorContainer();
    TypedFieldId[] htKeyFieldIds = new TypedFieldId[htConfig.getKeyExprsBuild().size()];
    int i = 0;
    for (NamedExpression ne : htConfig.getKeyExprsBuild()) {
        final LogicalExpression expr = ExpressionTreeMaterializer.materialize(ne.getExpr(), incomingBuild, collector, context.getFunctionRegistry());
        if (collector.hasErrors()) {
            throw new SchemaChangeException("Failure while materializing expression. " + collector.toErrorString());
        }
        if (expr == null) {
            continue;
        }
        keyExprsBuild[i] = expr;
        i++;
    }
    if (isProbe) {
        i = 0;
        for (NamedExpression ne : htConfig.getKeyExprsProbe()) {
            final LogicalExpression expr = ExpressionTreeMaterializer.materialize(ne.getExpr(), incomingProbe, collector, context.getFunctionRegistry());
            if (collector.hasErrors()) {
                throw new SchemaChangeException("Failure while materializing expression. " + collector.toErrorString());
            }
            if (expr == null) {
                continue;
            }
            keyExprsProbe[i] = expr;
            i++;
        }
        JoinUtils.addLeastRestrictiveCasts(keyExprsProbe, incomingProbe, keyExprsBuild, incomingBuild, context);
    }
    i = 0;
    /*
     * Once the implicit casts have been added, create the value vectors for the corresponding
     * type and add it to the hash table's container.
     * Note: Adding implicit casts may have a minor impact on the memory foot print. For example
     * if we have a join condition with bigint on the probe side and int on the build side then
     * after this change we will be allocating a bigint vector in the hashtable instead of an int
     * vector.
     */
    for (NamedExpression ne : htConfig.getKeyExprsBuild()) {
        LogicalExpression expr = keyExprsBuild[i];
        final MaterializedField outputField = MaterializedField.create(ne.getRef().getLastSegment().getNameSegment().getPath(), expr.getMajorType());
        ValueVector vv = TypeHelper.getNewVector(outputField, allocator);
        htKeyFieldIds[i] = htContainerOrig.add(vv);
        i++;
    }
    // Only in case of a join: Generate a special method to check if both the new key and the existing key (in this HT bucket) are nulls
    // (used by Hash-Join to avoid creating a long hash-table chain of null keys, which can lead to useless O(n^2) work on that chain.)
    // The logic is: Nulls match on build, and don't match on probe. Note that this logic covers outer joins as well.
    setupIsKeyMatchInternal(cgInner, bothKeysNullIncomingBuildMapping, bothKeysNullHtableMapping, keyExprsBuild, htConfig.getComparators(), htKeyFieldIds, SetupWork.CHECK_BOTH_NULLS);
    // generate code for isKeyMatch(), setValue(), getHash() and outputRecordKeys()
    setupIsKeyMatchInternal(cgInner, KeyMatchIncomingBuildMapping, KeyMatchHtableMapping, keyExprsBuild, htConfig.getComparators(), htKeyFieldIds, SetupWork.DO_BUILD);
    setupIsKeyMatchInternal(cgInner, KeyMatchIncomingProbeMapping, KeyMatchHtableProbeMapping, keyExprsProbe, htConfig.getComparators(), htKeyFieldIds, SetupWork.DO_PROBE);
    setupSetValue(cgInner, keyExprsBuild, htKeyFieldIds);
    if (outgoing != null) {
        if (outKeyFieldIds.length > htConfig.getKeyExprsBuild().size()) {
            throw new IllegalArgumentException("Mismatched number of output key fields.");
        }
    }
    setupOutputRecordKeys(cgInner, htKeyFieldIds, outKeyFieldIds);
    setupGetHash(cg, /* use top level code generator for getHash */
    GetHashIncomingBuildMapping, incomingBuild, keyExprsBuild);
    setupGetHash(cg, /* use top level code generator for getHash */
    GetHashIncomingProbeMapping, incomingProbe, keyExprsProbe);
    HashTable ht = context.getImplementationClass(top);
    ht.setup(htConfig, allocator, incomingBuild.getContainer(), incomingProbe, outgoing, htContainerOrig, context, cgInner);
    return ht;
}

Example 43 with ErrorCollectorImpl

use of org.apache.drill.common.expression.ErrorCollectorImpl in project drill by apache.

the class PhysicalOperatorUtil method materializeExpression.

/**
 * Helper method to materialize the given logical expression using the
 * {@code ExpressionTreeMaterializer}.
 * @param expr Logical expression to materialize
 * @param incoming Incoming record batch
 * @param context Fragment context
 */
public static LogicalExpression materializeExpression(LogicalExpression expr, VectorAccessible incoming, FragmentContext context) {
    ErrorCollector collector = new ErrorCollectorImpl();
    LogicalExpression mle = ExpressionTreeMaterializer.materialize(expr, incoming, collector, context.getFunctionRegistry());
    collector.reportErrors(logger);
    return mle;
}

Example 44 with ErrorCollectorImpl

use of org.apache.drill.common.expression.ErrorCollectorImpl 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;
    }
}

Example 45 with ErrorCollectorImpl

use of org.apache.drill.common.expression.ErrorCollectorImpl in project drill by apache.

the class TopNBatch method createNewPriorityQueue.

public PriorityQueue createNewPriorityQueue(FragmentContext context, List<Ordering> orderings, VectorAccessible batch, MappingSet mainMapping, MappingSet leftMapping, MappingSet rightMapping) throws ClassTransformationException, IOException, SchemaChangeException {
    CodeGenerator<PriorityQueue> cg = CodeGenerator.get(PriorityQueue.TEMPLATE_DEFINITION, context.getFunctionRegistry(), context.getOptions());
    cg.plainJavaCapable(true);
    // Uncomment out this line to debug the generated code.
    //    cg.saveCodeForDebugging(true);
    ClassGenerator<PriorityQueue> g = cg.getRoot();
    g.setMappingSet(mainMapping);
    for (Ordering od : orderings) {
        // first, we rewrite the evaluation stack for each side of the comparison.
        ErrorCollector collector = new ErrorCollectorImpl();
        final LogicalExpression expr = ExpressionTreeMaterializer.materialize(od.getExpr(), batch, collector, context.getFunctionRegistry(), unionTypeEnabled);
        if (collector.hasErrors()) {
            throw new SchemaChangeException("Failure while materializing expression. " + collector.toErrorString());
        }
        g.setMappingSet(leftMapping);
        HoldingContainer left = g.addExpr(expr, ClassGenerator.BlkCreateMode.FALSE);
        g.setMappingSet(rightMapping);
        HoldingContainer right = g.addExpr(expr, ClassGenerator.BlkCreateMode.FALSE);
        g.setMappingSet(mainMapping);
        // next we wrap the two comparison sides and add the expression block for the comparison.
        LogicalExpression fh = FunctionGenerationHelper.getOrderingComparator(od.nullsSortHigh(), left, right, context.getFunctionRegistry());
        HoldingContainer out = g.addExpr(fh, ClassGenerator.BlkCreateMode.FALSE);
        JConditional jc = g.getEvalBlock()._if(out.getValue().ne(JExpr.lit(0)));
        if (od.getDirection() == Direction.ASCENDING) {
            jc._then()._return(out.getValue());
        } else {
            jc._then()._return(out.getValue().minus());
        }
        g.rotateBlock();
    }
    g.rotateBlock();
    g.getEvalBlock()._return(JExpr.lit(0));
    PriorityQueue q = context.getImplementationClass(cg);
    q.init(config.getLimit(), context, oContext.getAllocator(), schema.getSelectionVectorMode() == BatchSchema.SelectionVectorMode.TWO_BYTE);
    return q;
}