The TableExprNodeRecordField
and TableExprId classes form
the means by which TaQL can deal with any set of data.
First the TaQL expression has to be setup. This is done by
constructing a TableExprNodeRecordField object for each
'field' to be used in the expression. TableExprNodeRecordField
uses a RecordInterface object
to make the data type of a field in the data set known and to
map a field name to a field index (the index is the sequence number
of the field in the record description).
When evaluating the expression for each member in the data set,
a TableExprData> needs to be passed (which is automatically
converted to TableExprId).
So a class needs to be written to access the data in the data set.
It needs to be derived from the abstract base class TableExprData
defined in this file. An example is given below.
It is also possible that the data set contains records and that the selection is based on fields in those records. In such a case the record passed to TableExprNodeRecordField should contain subrecords representing those records. The field index in the various functions as passed as a Block<Int> to denote the fields in the subrecords (and possibly subsubrecords, etc.. However, normally records won't be used and fieldNrs[0] gives the field index.
// Write a class derived from TableExprData to handle the vectors.
class MyTestClass : public TableExprData
{
public:
// Constructor checks if both vectors have equal length.
MyTestClass (const Vector<Int>& fld1, const Vector<String>& fld2)
: itsFld1(fld1), itsFld2(fld2), itsEntry(0)
{ AlwaysAssert (fld1.nelements() == fld2.nelements(), AipsError); }
virtual ~MyTestClass()
{}
void next()
{ itsEntry++; }
// Note that only the get functions for the possible types are needed.
// Also note that all numeric types are handled by TaQL as Double.
// The exception should never be thrown unless things are screwed up.
virtual Double getDouble (const Block<Int>& fieldNrs) const
{ switch (fieldNrs[0]) {
case 0:
return itsFld1(itsEntry);
default:
throw AipsError();
}
}
virtual String getString (const Block<Int>& fieldNrs) const
{ switch (fieldNrs[0]) {
case 1:
return itsFld2(itsEntry);
default:
throw AipsError();
}
}
virtual DataType dataType (const Block<Int>& fieldNrs) const
{ switch (fieldNrs[0]) {
case 0:
return TpInt;
case 1:
return TpString;
default:
throw AipsError();
}
}
// Make a Record to give to vectors a name.
// The order in which the fields are defined determines the fieldnrs
// passed to the get functions.
static Record makeRecord()
{ RecordDesc desc;
desc.addField ("fld1", TpInt);
desc.addField ("fld2", TpString);
return Record(desc);
}
private:
Vector<Int> itsFld1;
Vector<String> itsFld2;
uInt itsEntry;
};
Vector<uInt> findMatches (const Vector<Int>& fld1,
const Vector<String>& fld2)
{
// Make some expression.
// First create a Record to make the names and types known.
Record rec(MyTestClass::makeRecord());
TableExprNode expr (makeRecordExpr(rec,"fld1") > 10 &&
makeRecordExpr(rec,"fld2") != pattern("*xxx*"));
// Now evaluate the expression for each entry in the vector.
// Make a MyTestClass object to handle the vectors and put it in
// a TableExprId object for the TaQL evaluator.
// Note that TableExprId holds a pointer to the original MyTestClass
// object, so the TaQL evaluator 'sees' the changes we make by
// using the its next() function.
MyTestClass subj(fld1, fld2);
TableExprId eid(subj);
// The matching entry numbers are stored in a vector.
Vector<uInt> result(fld1.nelements());
uInt nr=0;
Bool valb;
for (uInt i=0; i<fld1.nelements(); i++) {
expr.get (eid, valb);
if (valb) {
result(nr++) = i;
}
subj.next(); // Next time the next entry must be used
}
result.resize (nr, True);
return result;
}
Get the shape of the given field. Need only be implemented if there are arrays in the data. The default implementation returns an empty IPosition.
Get the data type of the given field. Note that TpArray types have to be returned for arrays. If the field is unknown, TpOther should be returned. It is used for the isdefined function to check if the field is really defined.
Get a scalar in the given type. This might involve converting for Double and DComplex. Most default implementations throws an "not possible" exception. The default getDComplex invokes getDouble.
Get an array in the given type. This might involve converting for Double and DComplex. Most default implementations throws an "not possible" exception. The default getArrayDComplex invokes getArrayDouble.