ArrayDesc inferSchema(std::vector< ArrayDesc> schemas, boost::shared_ptr< Query> query)
{
assert(schemas.size() == 1);
ArrayDesc const& desc = schemas[0];
Dimensions const& dims = desc.getDimensions();
Attributes const& attrs = desc.getAttributes();
AttributeID aid = 0;
if (_parameters.size() >= 1) {
aid = ((boost::shared_ptr<OperatorParamReference>&)_parameters[0])->getObjectNo();
}
AggregatePtr maxAggregate = AggregateLibrary::getInstance()->createAggregate("max", TypeLibrary::getType(attrs[aid].getType()));
Attributes aggAttrs(1);
aggAttrs[0] = AttributeDesc((AttributeID)0,
attrs[aid].getName() + "_max",
maxAggregate->getResultType().typeId(),
AttributeDesc::IS_NULLABLE,
0);
if (_parameters.size() <= 1) {
Dimensions aggDims(1);
aggDims[0] = DimensionDesc("i", 0, 0, 0, 0, 1, 0);
return ArrayDesc(desc.getName(), aggAttrs, aggDims);
} else {
vector<int> groupBy(_parameters.size()-1);
for (size_t i = 0; i < groupBy.size(); i++) {
groupBy[i] = ((boost::shared_ptr<OperatorParamReference>&)_parameters[i + 1])->getObjectNo();
}
Dimensions aggDims(groupBy.size());
for (size_t i = 0, n = aggDims.size(); i < n; i++) {
DimensionDesc const& srcDim = dims[groupBy[i]];
aggDims[i] = DimensionDesc( srcDim.getBaseName(),
srcDim.getStartMin(),
srcDim.getCurrStart(),
srcDim.getCurrEnd(),
srcDim.getEndMax(),
i == 0 && groupBy[i] == 0 ? srcDim.getChunkInterval() : srcDim.getCurrLength(),
0,
srcDim.getType(),
srcDim.getFlags(),
srcDim.getMappingArrayName(),
srcDim.getComment(),
srcDim.getFuncMapOffset(),
srcDim.getFuncMapScale());
}
return ArrayDesc(desc.getName(), aggAttrs, aggDims);
}
}
ArrayDesc inferSchema(std::vector<ArrayDesc> schemas, boost::shared_ptr<Query> query)
{
assert(schemas.size() == 1);
ArrayDesc const& schema = schemas[0];
Dimensions const& dims = schema.getDimensions();
size_t nDims = dims.size();
size_t nParams = _parameters.size();
assert((nParams & 1) == 0 || nParams >= nDims*2);
Dimensions newDims(nDims - nParams/2);
if (newDims.size() <= 0)
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_SLICE_ERROR1);
std::vector<std::string> sliceDimName(nParams/2);
for (size_t i = 0; i < nParams; i+=2) {
sliceDimName[i >> 1] = ((boost::shared_ptr<OperatorParamReference>&)_parameters[i])->getObjectName();
}
size_t j = 0;
for (size_t i = 0; i < nDims; i++) {
const std::string dimName = dims[i].getBaseName();
int k = sliceDimName.size();
while (--k >= 0
&& sliceDimName[k] != dimName
&& !(sliceDimName[k][0] == '_' && (size_t)atoi(sliceDimName[k].c_str()+1) == i+1));
if (k < 0)
{
if (j >= newDims.size())
throw USER_QUERY_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_DUPLICATE_DIMENSION_NAME,
_parameters[i]->getParsingContext()) << dimName;
newDims[j++] = dims[i];
}
}
return ArrayDesc(schema.getName(), schema.getAttributes(), newDims);
}
/**
* Determine the schema of the output. inferSchema is called on the coordinator instance during query planning and
* may be called several times as the planner gets its act together. It will always be called with the same inputs
* for the same query. This function must behave deterministically, but the shape of the output may vary based on
* inputs and parameters.
* @param schemas all of the schemas of the input arrays (if the operator accepts any)
* @param query the query context
* @return the schema of the outpt, as described above.
*/
ArrayDesc inferSchema(vector< ArrayDesc> schemas, shared_ptr< Query> query)
{
/*
* Make one string attribute: id=0, name="instance_status" of type string, no flags, no default compression.
* The ID of the attribute is simply a number from 0 to num_attributes-1 and must equal to its position
* in the attributes vector.
*/
AttributeDesc outputAttribute (0, "instance_status", TID_STRING, 0, 0);
Attributes outputAttributes(1, outputAttribute);
/* Add the empty tag attribute. Arrays with the empty tag are "emptyable" meaning that some cells may be empty.
* It is a good practice to add this to every constructed array. In fact, in the future it may become the
* default for all arrays.
*/
outputAttributes = addEmptyTagAttribute(outputAttributes);
/* The output dimension: from 0 to "*" with a chunk size of 1. The amount of data returned is so small that the
* chunk size is not relevant.
*/
DimensionDesc outputDimension("instance_no", 0, MAX_COORDINATE, 1, 0);
Dimensions outputDimensions(1, outputDimension);
/* The first argument is the name of the returned array. */
return ArrayDesc("hello_instances", outputAttributes, outputDimensions);
}
ArrayDesc inferSchema(std::vector< ArrayDesc> schemas, std::shared_ptr< Query> query)
{
assert(schemas.size() == 1);
for (Parameters::const_iterator it = _parameters.begin(); it != _parameters.end(); ++it)
assert(((std::shared_ptr<OperatorParamReference>&)*it)->getParamType() == PARAM_ATTRIBUTE_REF);
Attributes newAttributes;
const Attributes &oldAttributes = schemas[0].getAttributes();
bool includesIndicator = false;
size_t n = _parameters.size();
for (size_t i = 0; i < n; i++)
{
const AttributeDesc &attr =
oldAttributes[((std::shared_ptr<OperatorParamReference>&)_parameters[i])->getObjectNo()];
newAttributes.push_back(AttributeDesc(i, attr.getName(), attr.getType(),
attr.getFlags(), attr.getDefaultCompressionMethod(),
attr.getAliases(), &attr.getDefaultValue(),
attr.getDefaultValueExpr()));
includesIndicator |= attr.isEmptyIndicator();
}
if (!includesIndicator) {
AttributeDesc const* indicator = schemas[0].getEmptyBitmapAttribute();
if (indicator != NULL) {
newAttributes.push_back(AttributeDesc(n, indicator->getName(), indicator->getType(),
indicator->getFlags(), indicator->getDefaultCompressionMethod(),
indicator->getAliases()));
}
}
return ArrayDesc(schemas[0].getName(), newAttributes, schemas[0].getDimensions(), defaultPartitioning());
}
ArrayDesc inferSchema(std::vector< ArrayDesc> schemas, boost::shared_ptr< Query> query)
{
Attributes outputAttrs;
outputAttrs.push_back(AttributeDesc(0, "dummy", TID_DOUBLE, AttributeDesc::IS_NULLABLE, 0));
Dimensions outputDims;
outputDims.push_back(DimensionDesc("i",0,0,1,0));
return ArrayDesc("test_cache", outputAttrs, outputDims);
}
ArrayDesc inferSchema(std::vector< ArrayDesc> schemas, boost::shared_ptr< Query> query)
{
assert(schemas.size() == 2);
ArrayDesc const& patternDesc = schemas[0];
ArrayDesc const& catalogDesc = schemas[1];
Attributes const& catalogAttributes = catalogDesc.getAttributes(true);
Dimensions const& catalogDimensions = catalogDesc.getDimensions();
Attributes const& patternAttributes = patternDesc.getAttributes(true);
Dimensions resultDimensions = patternDesc.getDimensions();
size_t totalAttributes = catalogAttributes.size() + patternAttributes.size() + 1 + catalogDimensions.size();
Attributes matchAttributes(totalAttributes);
if (catalogDimensions.size() != resultDimensions.size())
{
stringstream left, right;
printDimNames(left, resultDimensions);
printDimNames(right, catalogDimensions);
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_DIMENSION_COUNT_MISMATCH)
<< "match" << left.str() << right.str();
}
for (size_t i = 0, n = catalogDimensions.size(); i < n; i++) {
if (!(catalogDimensions[i].getStartMin() == resultDimensions[i].getStartMin()
&& catalogDimensions[i].getChunkInterval() == resultDimensions[i].getChunkInterval()
&& catalogDimensions[i].getChunkOverlap() == resultDimensions[i].getChunkOverlap()))
{
// XXX To do: implement requiresRepart() method, remove interval/overlap checks
// above, use SCIDB_LE_START_INDEX_MISMATCH here.
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_ARRAYS_NOT_CONFORMANT);
}
}
size_t j = 0;
for (size_t i = 0, n = patternAttributes.size(); i < n; i++, j++) {
AttributeDesc const& attr = patternAttributes[i];
matchAttributes[j] = AttributeDesc(j, attr.getName(), attr.getType(), attr.getFlags(),
attr.getDefaultCompressionMethod(), attr.getAliases(), &attr.getDefaultValue(),
attr.getDefaultValueExpr());
}
for (size_t i = 0, n = catalogAttributes.size(); i < n; i++, j++) {
AttributeDesc const& attr = catalogAttributes[i];
matchAttributes[j] = AttributeDesc(j, "match_" + attr.getName(), attr.getType(), attr.getFlags(),
attr.getDefaultCompressionMethod(), attr.getAliases(), &attr.getDefaultValue(),
attr.getDefaultValueExpr());
}
for (size_t i = 0, n = catalogDimensions.size(); i < n; i++, j++) {
matchAttributes[j] = AttributeDesc(j, "match_" + catalogDimensions[i].getBaseName(), TID_INT64, 0, 0);
}
matchAttributes[j] = AttributeDesc(j, DEFAULT_EMPTY_TAG_ATTRIBUTE_NAME, TID_INDICATOR, AttributeDesc::IS_EMPTY_INDICATOR, 0);
int64_t maxCollisions = evaluate(((boost::shared_ptr<OperatorParamLogicalExpression>&)_parameters[1])->getExpression(),
query, TID_INT64).getInt64();
if (maxCollisions <= 0 || (int32_t)maxCollisions != maxCollisions) {
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_OPERATOR_ARGUMENT2) << "positive";
}
resultDimensions.push_back(DimensionDesc("collision", 0, 0, maxCollisions-1, maxCollisions-1, (uint32_t)maxCollisions, 0));
return ArrayDesc("match", matchAttributes, resultDimensions);
}
/* inferSchema helps the query planner decide on the shape of
* the output array. All operators must define this function.
*/
ArrayDesc inferSchema(vector< ArrayDesc> schemas, shared_ptr< Query> query)
{
ArrayDesc const& matrix = schemas[0];
if(matrix.getAttributes(true)[0].getType() != TID_STRING)
throw SYSTEM_EXCEPTION(SCIDB_SE_INTERNAL, SCIDB_LE_ILLEGAL_OPERATION) << "cu requires a single string-valued attribute";
Attributes outputAttributes(matrix.getAttributes());
Dimensions outputDimensions(matrix.getDimensions());
return ArrayDesc(matrix.getName(), outputAttributes, outputDimensions);
}
ArrayDesc inferSchema(std::vector< ArrayDesc> schemas, boost::shared_ptr< Query> query)
{
SplitSettings settings (_parameters, true, query); //construct and check to ensure settings are legit
vector<AttributeDesc> attributes(1);
attributes[0] = AttributeDesc((AttributeID)0, "value", TID_STRING, 0, 0);
vector<DimensionDesc> dimensions(2);
dimensions[0] = DimensionDesc("source_instance_id", 0, 0, MAX_COORDINATE, MAX_COORDINATE, 1, 0);
dimensions[1] = DimensionDesc("chunk_no", 0, 0, MAX_COORDINATE, MAX_COORDINATE, 1, 0);
return ArrayDesc("split", attributes, dimensions);
}
ArrayDesc inferSchema(std::vector< ArrayDesc> schemas, shared_ptr< Query> query)
{
ArrayDesc const& inputSchema = schemas[0];
//FastCountSettings settings (_parameters, true, query);
vector<DimensionDesc> dimensions(1);
size_t const nInstances = query->getInstancesCount();
dimensions[0] = DimensionDesc("i", 0, 0, CoordinateBounds::getMax(), CoordinateBounds::getMax(), 1, 0);
// dimensions[0] = DimensionDesc("i", 0, 0, nInstances-1, nInstances-1, 1, 0);
vector<AttributeDesc> attributes;
attributes.push_back(AttributeDesc((AttributeID)0, "count", TID_UINT64, AttributeDesc::IS_NULLABLE, 0));
return ArrayDesc("fast_count", attributes, dimensions, defaultPartitioning(), inputSchema.getResidency(),false);
}
ArrayDesc inferSchema(vector<ArrayDesc> inputSchemas, shared_ptr<Query> query)
{
Attributes atts(1);
atts[0] = AttributeDesc((AttributeID)0, "success", TID_BOOL, 0, CompressorType::NONE );
Dimensions dims(1);
dims[0] = DimensionDesc("i", 0, 0, 0, 0, 1, 0);
//#ifdef CPP11
return ArrayDesc("", atts, dims, defaultPartitioning(), query->getDefaultArrayResidency());
//#else
//return ArrayDesc("", atts, dims);
//#endif
}
ArrayDesc inferSchema(std::vector< ArrayDesc> schemas, boost::shared_ptr< Query> query)
{
Attributes atts(1);
TypeId type = schemas[0].getAttributes()[0].getType();
AttributeDesc multAttr((AttributeID)0, "matricize", type, 0, 0);
atts[0] = multAttr;
Dimensions const& inputDims = schemas[0].getDimensions();
Coordinate ndims = inputDims.size();
Coordinate rowmode = 0;
Coordinate colmode = 1;
if (_parameters.size() == 1) {
rowmode = evaluate(((boost::shared_ptr<OperatorParamLogicalExpression>&)_parameters[0])->getExpression(), query, TID_INT64).getInt64();
if(rowmode < 1 || rowmode > ndims) {
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_MATRICIZE_ERROR1);
}
if(rowmode < ndims)
colmode = ndims;
else
colmode = ndims-1;
rowmode -= 1;
colmode -= 1;
} else {
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_MATRICIZE_ERROR2);
}
Coordinate len = 1;
Coordinate chunklen = 1;
for(Coordinate i=ndims-1;i>=0;i--) {
if(i == rowmode) continue;
len *= inputDims[i].getLength();
chunklen *= inputDims[i].getChunkInterval();
}
Dimensions dims(2);
dims[0] = inputDims[rowmode];
dims[1] = DimensionDesc(inputDims[rowmode].getBaseName(),
1,
len,
chunklen,
0,
inputDims[rowmode].getType(),
inputDims[rowmode].getFlags(),
inputDims[rowmode].getMappingArrayName(),
inputDims[rowmode].getComment());
return ArrayDesc("Matricize",atts,dims);
}
ArrayDesc inferSchema(std::vector<ArrayDesc> schemas, boost::shared_ptr<Query> query)
{
assert(schemas.size() == 0);
assert(_parameters.size() == 2);
assert(((boost::shared_ptr<OperatorParam>&)_parameters[1])->getParamType() == PARAM_ARRAY_REF);
const string &newArrayName = ((boost::shared_ptr<OperatorParamReference>&)_parameters[1])->getObjectName();
if (SystemCatalog::getInstance()->containsArray(newArrayName))
{
throw USER_QUERY_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_ARRAY_ALREADY_EXIST,
_parameters[1]->getParsingContext()) << newArrayName;
}
return ArrayDesc();
}
ArrayDesc inferSchema(std::vector< ArrayDesc> schemas, boost::shared_ptr< Query> query)
{
assert(schemas.size() == 1);
const ArrayDesc& desc = schemas[0];
const Attributes &oldAttributes = desc.getAttributes();
Attributes newAttributes = desc.getAttributes();
for (size_t paramNo = 0, paramCount = _parameters.size(); paramNo < paramCount; paramNo+=2)
{
int32_t attNo = ((boost::shared_ptr<OperatorParamReference>&)_parameters[paramNo])->getObjectNo();
AttributeDesc attr = oldAttributes[attNo];
newAttributes[attNo] = AttributeDesc(attNo,
((boost::shared_ptr<OperatorParamReference>&)_parameters[paramNo + 1])->getObjectName(),
attr.getType(), attr.getFlags(), attr.getDefaultCompressionMethod(),
attr.getAliases(), &attr.getDefaultValue(), attr.getDefaultValueExpr());
}
return ArrayDesc(desc.getId(), desc.getUAId(), desc.getVersionId(), desc.getName(), newAttributes, desc.getDimensions());
}
ArrayDesc inferSchema(std::vector< ArrayDesc> schemas, boost::shared_ptr< Query> query)
{
assert(schemas.size() == 1);
assert(_parameters[0]->getParamType() == PARAM_ATTRIBUTE_REF);
assert(_parameters[1]->getParamType() == PARAM_LOGICAL_EXPRESSION);
if ( _parameters.size() % 2 != 0 )
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_OPERATOR_ARGUMENTS_COUNT2) << "tile_apply";
}
Attributes outAttrs;
AttributeID nextAttrId =0;
for (size_t i=0; i<schemas[0].getAttributes().size(); i++) {
AttributeDesc const& attr = schemas[0].getAttributes()[i];
if(attr.getType()!=TID_INDICATOR) {
outAttrs.push_back( AttributeDesc(nextAttrId++,
attr.getName(),
attr.getType(),
attr.getFlags(),
attr.getDefaultCompressionMethod(),
attr.getAliases(),
attr.getReserve(),
&attr.getDefaultValue(),
attr.getDefaultValueExpr(),
attr.getVarSize()));
}
}
size_t k;
for (k=0; k<_parameters.size(); k+=2) {
const string &attributeName = ((boost::shared_ptr<OperatorParamReference>&)_parameters[k])->getObjectName();
Expression expr;
expr.compile(((boost::shared_ptr<OperatorParamLogicalExpression>&)_parameters[k+1])->getExpression(),
query, _properties.tile, TID_VOID, schemas);
assert(!_properties.tile);
int flags = 0;
if (expr.isNullable()) {
flags = (int)AttributeDesc::IS_NULLABLE;
}
for (size_t j = 0; j < nextAttrId; j++) {
AttributeDesc const& attr = outAttrs[j];
if (attr.getName() == attributeName)
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_DUPLICATE_ATTRIBUTE_NAME) << attributeName;
}
}
outAttrs.push_back(AttributeDesc(nextAttrId++,
attributeName,
expr.getType(),
flags,
0));
}
if(schemas[0].getEmptyBitmapAttribute()) {
AttributeDesc const* emptyTag = schemas[0].getEmptyBitmapAttribute();
for (size_t j = 0; j < nextAttrId; j++) {
AttributeDesc const& attr = outAttrs[j];
if (attr.getName() == emptyTag->getName()) {
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_DUPLICATE_ATTRIBUTE_NAME) << attr.getName();
}
}
outAttrs.push_back( AttributeDesc(nextAttrId,
emptyTag->getName(),
emptyTag->getType(),
emptyTag->getFlags(),
emptyTag->getDefaultCompressionMethod(),
emptyTag->getAliases(),
emptyTag->getReserve(),
&emptyTag->getDefaultValue(),
emptyTag->getDefaultValueExpr(),
emptyTag->getVarSize()));
}
return ArrayDesc(schemas[0].getName(), outAttrs, schemas[0].getDimensions());
}
ArrayDesc inferSchema(std::vector< ArrayDesc> schemas, boost::shared_ptr< Query> query)
{
assert(schemas.size() == 2);
if (!hasSingleAttribute(schemas[0]) || !hasSingleAttribute(schemas[1]))
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_MULTIPLY_ERROR2);
if (schemas[0].getDimensions().size() != 2 || schemas[1].getDimensions().size() != 2)
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_MULTIPLY_ERROR3);
if (schemas[0].getDimensions()[0].getLength() == INFINITE_LENGTH
|| schemas[0].getDimensions()[1].getLength() == INFINITE_LENGTH
|| schemas[1].getDimensions()[0].getLength() == INFINITE_LENGTH
|| schemas[1].getDimensions()[1].getLength() == INFINITE_LENGTH)
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_MULTIPLY_ERROR4);
if (schemas[0].getDimensions()[1].getLength() != schemas[1].getDimensions()[1].getLength()
|| schemas[0].getDimensions()[1].getStart() != schemas[1].getDimensions()[1].getStart())
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_MULTIPLY_ERROR5);
// FIXME: This condition needs to go away later
if (schemas[0].getDimensions()[1].getChunkInterval() != schemas[1].getDimensions()[1].getChunkInterval())
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_MULTIPLY_ERROR6);
if (schemas[0].getAttributes()[0].getType() != schemas[1].getAttributes()[0].getType())
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_MULTIPLY_ERROR7);
if (schemas[0].getAttributes()[0].isNullable() || schemas[1].getAttributes()[0].isNullable())
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_MULTIPLY_ERROR8);
Attributes atts(1);
TypeId type = schemas[0].getAttributes()[0].getType();
AttributeDesc multAttr((AttributeID)0, "multiply", type, 0, 0);
atts[0] = multAttr;
Dimensions dims(2);
DimensionDesc const& d1 = schemas[0].getDimensions()[0];
dims[0] = DimensionDesc(d1.getBaseName(),
d1.getNamesAndAliases(),
d1.getStartMin(),
d1.getCurrStart(),
d1.getCurrEnd(),
d1.getEndMax(),
d1.getChunkInterval(),
0,
d1.getType(),
d1.getFlags(),
d1.getMappingArrayName(),
d1.getComment(),
d1.getFuncMapOffset(),
d1.getFuncMapScale());
DimensionDesc const& d2 = schemas[1].getDimensions()[0];
dims[1] = DimensionDesc(d1.getBaseName() == d2.getBaseName() ? d1.getBaseName() + "2" : d2.getBaseName(),
d2.getNamesAndAliases(),
d2.getStartMin(),
d2.getCurrStart(),
d2.getCurrEnd(),
d2.getEndMax(),
d2.getChunkInterval(),
0,
d2.getType(),
d2.getFlags(),
d2.getMappingArrayName(),
d2.getComment(),
d2.getFuncMapOffset(),
d2.getFuncMapScale());
return ArrayDesc("MultiplyRow",atts,dims);
}
