ArrayDesc inferSchema(std::vector< ArrayDesc> schemas, boost::shared_ptr< Query> query)
{
assert(schemas.size() == 1);
ArrayDesc const& srcDesc = schemas[0];
ArrayDesc dstDesc = ((boost::shared_ptr<OperatorParamSchema>&)_parameters[0])->getSchema();
//Compile a desc of all possible attributes (aggregate calls first) and source dimensions
ArrayDesc aggregationDesc (srcDesc.getName(), Attributes(), srcDesc.getDimensions());
vector<string> aggregatedNames;
//add aggregate calls first
for (size_t i = 1; i < _parameters.size(); i++)
{
addAggregatedAttribute( (shared_ptr <OperatorParamAggregateCall>&) _parameters[i], srcDesc, aggregationDesc);
aggregatedNames.push_back(aggregationDesc.getAttributes()[aggregationDesc.getAttributes().size()-1].getName());
}
//add other attributes
BOOST_FOREACH(const AttributeDesc &srcAttr, srcDesc.getAttributes())
{
//if there's an attribute with same name as an aggregate call - skip the attribute
bool found = false;
BOOST_FOREACH(const AttributeDesc &aggAttr, aggregationDesc.getAttributes())
{
if( aggAttr.getName() == srcAttr.getName())
{
found = true;
}
}
if (!found)
{
aggregationDesc.addAttribute(AttributeDesc( aggregationDesc.getAttributes().size(),
srcAttr.getName(),
srcAttr.getType(),
srcAttr.getFlags(),
srcAttr.getDefaultCompressionMethod(),
srcAttr.getAliases(),
&srcAttr.getDefaultValue(),
srcAttr.getDefaultValueExpr(),
srcAttr.getComment(),
srcAttr.getVarSize()));
}
}
//Ensure attributes names uniqueness.
if (!dstDesc.getEmptyBitmapAttribute())
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_REDIMENSION_ERROR1);
BOOST_FOREACH(const AttributeDesc &dstAttr, dstDesc.getAttributes())
{
BOOST_FOREACH(const AttributeDesc &srcAttr, aggregationDesc.getAttributes())
{
if (srcAttr.getName() == dstAttr.getName())
{
if (srcAttr.getType() != dstAttr.getType())
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_ATTRIBUTE_TYPE)
<< srcAttr.getName() << srcAttr.getType() << dstAttr.getType();
}
if (!dstAttr.isNullable() && srcAttr.isNullable())
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_ATTRIBUTE_FLAGS)
<< srcAttr.getName();
}
goto NextAttr;
}
}
BOOST_FOREACH(const DimensionDesc &srcDim, aggregationDesc.getDimensions())
{
if (srcDim.hasNameOrAlias(dstAttr.getName()))
{
if (dstAttr.getType() != TID_INT64)
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_DESTINATION_ATTRIBUTE_TYPE)
<< dstAttr.getName() << TID_INT64;
}
if (srcDim.getType() != TID_INT64)
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_SOURCE_DIMENSION_TYPE)
<< dstAttr.getName() << TID_INT64;
}
if (dstAttr.getFlags() != 0)
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_DESTINATION_ATTRIBUTE_FLAGS)
<< dstAttr.getName();
}
goto NextAttr;
}
}
if (dstAttr.isEmptyIndicator() == false)
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_UNEXPECTED_DESTINATION_ATTRIBUTE)
<< dstAttr.getName();
}
NextAttr:;
}
BOOST_FOREACH(const DimensionDesc &dstDim, dstDesc.getDimensions())
{
if (dstDim.getChunkOverlap() != 0)
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_REDIMENSION_STORE_ERROR3);
BOOST_FOREACH(const AttributeDesc &srcAttr, aggregationDesc.getAttributes())
{
if (dstDim.hasNameOrAlias(srcAttr.getName()))
{
for (size_t i = 0; i< aggregatedNames.size(); i++)
{
if (srcAttr.getName() == aggregatedNames[i])
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_REDIMENSION_ERROR2);
}
if (srcAttr.getType() != TID_INT64)
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_SOURCE_ATTRIBUTE_TYPE)
<< srcAttr.getName() << TID_INT64;
}
if (srcAttr.getFlags() != 0)
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_SOURCE_ATTRIBUTE_FLAGS)
<< srcAttr.getName() << TID_INT64;
}
if (dstDim.getType() != TID_INT64)
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_DESTINATION_DIMENSION_TYPE)
<< srcAttr.getName() << TID_INT64;
}
goto NextDim;
}
}
BOOST_FOREACH(const DimensionDesc &srcDim, aggregationDesc.getDimensions())
{
if (srcDim.hasNameOrAlias(dstDim.getBaseName()))
{
if (dstDim.getType() != srcDim.getType() ||
dstDim.getStart() != srcDim.getStart() ||
dstDim.getLength() != srcDim.getLength() ||
dstDim.getChunkInterval() != srcDim.getChunkInterval())
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_DIMENSIONS_DONT_MATCH)
<< srcDim.getBaseName() << dstDim.getBaseName();
}
goto NextDim;
}
}
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_UNEXPECTED_DESTINATION_DIMENSION) << dstDim.getBaseName();
NextDim:;
}
return ArrayDesc(srcDesc.getName()+"_redimension", dstDesc.getAttributes(), dstDesc.getDimensions(), dstDesc.getFlags());
}
ArrayDesc inferSchema(std::vector< ArrayDesc> schemas, boost::shared_ptr< Query> query)
{
assert(schemas.size() == 1);
ArrayDesc const& srcDesc = schemas[0];
ArrayDesc dstDesc = ((boost::shared_ptr<OperatorParamSchema>&)_parameters[0])->getSchema();
//Compile a desc of all possible attributes (aggregate calls first) and source dimensions
ArrayDesc aggregationDesc (srcDesc.getName(), Attributes(), srcDesc.getDimensions());
vector<string> aggregatedNames;
//add aggregate calls first
for (size_t i = 1; i < _parameters.size(); i++)
{
addAggregatedAttribute( (shared_ptr <OperatorParamAggregateCall>&) _parameters[i], srcDesc, aggregationDesc);
string aggName = aggregationDesc.getAttributes()[aggregationDesc.getAttributes().size()-1].getName();
bool aggFound = false;
BOOST_FOREACH(const AttributeDesc &dstAttr, dstDesc.getAttributes()) {
if (dstAttr.getName() == aggName) {
aggFound = true;
break;
}
}
if (!aggFound) {
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_ATTRIBUTE_DOESNT_EXIST) << aggName << dstDesc.getName();
}
aggregatedNames.push_back(aggName);
}
//add other attributes
BOOST_FOREACH(const AttributeDesc &srcAttr, srcDesc.getAttributes())
{
//if there's an attribute with same name as an aggregate call - skip the attribute
bool found = false;
BOOST_FOREACH(const AttributeDesc &aggAttr, aggregationDesc.getAttributes())
{
if( aggAttr.getName() == srcAttr.getName())
{
found = true;
}
}
if (!found)
{
aggregationDesc.addAttribute(AttributeDesc( aggregationDesc.getAttributes().size(),
srcAttr.getName(),
srcAttr.getType(),
srcAttr.getFlags(),
srcAttr.getDefaultCompressionMethod(),
srcAttr.getAliases(),
&srcAttr.getDefaultValue(),
srcAttr.getDefaultValueExpr(),
srcAttr.getVarSize()));
}
}
//Ensure attributes names uniqueness.
if (!dstDesc.getEmptyBitmapAttribute())
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_REDIMENSION_ERROR1);
BOOST_FOREACH(const AttributeDesc &dstAttr, dstDesc.getAttributes())
{
BOOST_FOREACH(const AttributeDesc &srcAttr, aggregationDesc.getAttributes())
{
if (srcAttr.getName() == dstAttr.getName())
{
if (srcAttr.getType() != dstAttr.getType())
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_ATTRIBUTE_TYPE)
<< srcAttr.getName() << srcAttr.getType() << dstAttr.getType();
}
if (!dstAttr.isNullable() && srcAttr.isNullable())
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_ATTRIBUTE_FLAGS)
<< srcAttr.getName();
}
goto NextAttr;
}
}
BOOST_FOREACH(const DimensionDesc &srcDim, aggregationDesc.getDimensions())
{
if (srcDim.hasNameAndAlias(dstAttr.getName()))
{
if (dstAttr.getType() != TID_INT64)
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_DESTINATION_ATTRIBUTE_TYPE)
<< dstAttr.getName() << TID_INT64;
}
if (dstAttr.getFlags() != 0)
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_DESTINATION_ATTRIBUTE_FLAGS)
<< dstAttr.getName();
}
goto NextAttr;
}
}
if (dstAttr.isEmptyIndicator() == false)
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_UNEXPECTED_DESTINATION_ATTRIBUTE)
<< dstAttr.getName();
}
NextAttr:;
}
Dimensions outputDims;
size_t nNewDims = 0;
BOOST_FOREACH(const DimensionDesc &dstDim, dstDesc.getDimensions())
{
if (dstDim.getChunkOverlap() > dstDim.getChunkInterval())
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OVERLAP_CANT_BE_LARGER_CHUNK);
}
BOOST_FOREACH(const AttributeDesc &srcAttr, aggregationDesc.getAttributes())
{
if (dstDim.hasNameAndAlias(srcAttr.getName()))
{
for (size_t i = 0; i< aggregatedNames.size(); i++)
{
if (srcAttr.getName() == aggregatedNames[i])
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_OP_REDIMENSION_ERROR2);
}
if ( !IS_INTEGRAL(srcAttr.getType()) || srcAttr.getType() == TID_UINT64 )
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_SOURCE_ATTRIBUTE_TYPE)
<< srcAttr.getName() << TID_INT64;
}
outputDims.push_back(dstDim);
goto NextDim;
}
}
BOOST_FOREACH(const DimensionDesc &srcDim, aggregationDesc.getDimensions())
{
if (srcDim.hasNameAndAlias(dstDim.getBaseName()))
{
DimensionDesc outputDim = dstDim;
outputDims.push_back(outputDim);
goto NextDim;
}
}
//one synthetic dimension allowed
if (nNewDims++ != 0 || !aggregatedNames.empty() )
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_UNEXPECTED_DESTINATION_DIMENSION) << dstDim.getBaseName();
}
outputDims.push_back(dstDim);
NextDim:;
}
return ArrayDesc(srcDesc.getName(), dstDesc.getAttributes(), outputDims, dstDesc.getFlags());
}
/**
* Perform operator-specific checks of input and return the shape of the output. Currently,
* the output array must exist.
* @param schemas the shapes of the input arrays
* @param query the query context
*/
ArrayDesc inferSchema(std::vector< ArrayDesc> schemas, shared_ptr< Query> query)
{
assert(schemas.size() == 1);
assert(_parameters.size() == 1);
string arrayName = ((shared_ptr<OperatorParamReference>&)_parameters[0])->getObjectName();
ArrayDesc const& srcDesc = schemas[0];
//Ensure attributes names uniqueness.
ArrayDesc dstDesc;
if (!SystemCatalog::getInstance()->getArrayDesc(arrayName, dstDesc, false))
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_ARRAY_DOESNT_EXIST) << arrayName;
}
if(dstDesc.isImmutable())
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_ILLEGAL_OPERATION)
<< "Target of INSERT must be a mutable array";
}
Dimensions const& srcDims = srcDesc.getDimensions();
Dimensions const& dstDims = dstDesc.getDimensions();
if (srcDims.size() != dstDims.size())
{
//TODO: this will get lifted when we allow redimension+insert in the same op
//and when we DO implement redimension+insert - we will need to match attributes/dimensions by name, not position.
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_ILLEGAL_OPERATION)
<< "Temporary restriction: target of INSERT must have same dimensions as the source";
}
for (size_t i = 0, n = srcDims.size(); i < n; i++)
{
if( srcDims[i].getType() != TID_INT64 || dstDims[i].getType() != TID_INT64)
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_ILLEGAL_OPERATION)
<< "Temporary restriction: INSERT only supports integer dimensions";
}
//TODO: we can also allow arrays that are smaller whose length is not evenly divided by chunk interval
//but then we have to detect "edge chunks" and rewrite them cleverly
if( srcDims[i].getStartMin() != dstDims[i].getStartMin() ||
srcDims[i].getChunkInterval() != dstDims[i].getChunkInterval() ||
srcDims[i].getChunkOverlap() != dstDims[i].getChunkOverlap() ||
srcDims[i].getEndMax() > dstDims[i].getEndMax() ||
( srcDims[i].getEndMax() < dstDims[i].getEndMax() &&
srcDims[i].getLength() % srcDims[i].getChunkInterval() != 0))
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_DIMENSIONS_DONT_MATCH)
<< srcDims[i].getBaseName() << dstDims[i].getBaseName();
}
}
Attributes const& srcAttrs = srcDesc.getAttributes(true);
Attributes const& dstAttrs = dstDesc.getAttributes(true);
if (srcAttrs.size() != dstAttrs.size())
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_ILLEGAL_OPERATION)
<< "Temporary restriction: target of INSERT must have same attributes as the source";
}
for (size_t i = 0, n = srcAttrs.size(); i < n; i++)
{
if(srcAttrs[i].getType() != dstAttrs[i].getType())
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_ATTRIBUTE_TYPE)
<< srcAttrs[i].getName() << srcAttrs[i].getType() << dstAttrs[i].getType();
}
//can't store nulls into a non-nullable attribute
if(!dstAttrs[i].isNullable() && srcAttrs[i].isNullable())
{
throw USER_EXCEPTION(SCIDB_SE_INFER_SCHEMA, SCIDB_LE_WRONG_ATTRIBUTE_FLAGS)
<< srcAttrs[i].getName();
}
}
//Note: let us NOT add arrayID numbers to the schema - because we do not have our ArrayID yet.
//We will get our ArrayID when we execute and create the array. Until then - don't bother.
//Old store code adds the arrayID to the schema - but that's the arrayID of the previous version,
//not the new version created by the op. A dangerous fallacy - stupid and unnecessary.
return ArrayDesc(arrayName, dstDesc.getAttributes(), dstDesc.getDimensions(), dstDesc.getFlags());
}
