[cdt-debug-dev] Displaying values of custom data types

[Matthias Spycher <matthias@xxxxxxxxxx>]

We have extended to the gdb/mi layer in our product to treat certain 
struct types as well-known fundamental types whose values can be 
displayed directly as numbers in the variables/expressions views. For 
example, we have a C++ library of fixed point types and big integer 
types. There appears to be a limitation in the current cdt.debug.core 
API when it comes to value formatting. Some of our struct types also 
implement ICDICharType so they can be converted to string values, but 
this doesn't allow for the formatting of numbers in hex, decimal, etc. 
The following code in CDebugModelPresentation checks for character and 
floating point types, but doesn't care to check integer types before 
eliminating arrays and structs (which is why our struct types implement 
ICDICharType).

           try {
               String valueString = value.getValueString();
               if ( valueString != null ) {
                   valueString = valueString.trim();
                   if ( type != null && type.isCharacter() ) {
                       if ( valueString.length() == 0 )
                           valueString = "."; //$NON-NLS-1$
                       label.append( valueString );
                   }
                   else if ( type != null && type.isFloatingPointType() ) {
                       Number floatingPointValue = 
CDebugUtils.getFloatingPointValue( (ICValue)value );
                       if ( CDebugUtils.isNaN( floatingPointValue ) )
                           valueString = "NAN"; //$NON-NLS-1$
                       if ( CDebugUtils.isPositiveInfinity( 
floatingPointValue ) )
                           valueString = CDebugUIMessages.getString( 
"CDTDebugModelPresentation.23" ); //$NON-NLS-1$
                       if ( CDebugUtils.isNegativeInfinity( 
floatingPointValue ) )
                           valueString = CDebugUIMessages.getString( 
"CDTDebugModelPresentation.24" ); //$NON-NLS-1$
                       label.append( valueString );
                   }
                   else if ( type == null || (!type.isArray() && 
!type.isStructure()) ) {
                       if ( valueString.length() > 0 ) {
                           label.append( valueString );
                       }
                   }
               }
           }
           catch( DebugException e1 ) {
           }


Furthermore, the number types are bounded in range when the values are 
retrieved in CValue. For example:

   private String getIntValueString( ICDIIntValue value ) throws 
CDIException {
       CVariableFormat format = getParentVariable().getFormat();
       if ( CVariableFormat.NATURAL.equals( format ) || 
CVariableFormat.DECIMAL.equals( format ) ) {
           return (isUnsigned()) ? Long.toString( value.longValue() ) : 
Integer.toString( value.intValue() );
       }
       else if ( CVariableFormat.HEXADECIMAL.equals( format ) ) {
           StringBuffer sb = new StringBuffer( "0x" ); //$NON-NLS-1$
           String stringValue = (isUnsigned()) ? Long.toHexString( 
value.longValue() ) : Integer.toHexString( value.intValue() );
           sb.append( (stringValue.length() > 8) ? 
stringValue.substring( stringValue.length() - 8 ) : stringValue );
           return sb.toString();
       }
       return null;
   }

We could add new interfaces that would allow for the conversion of 
stucts into strings or numbers in the debug presentation model, or we 
could work with the existing interfaces and eliminate the assumption 
that signed integers are bounded 32-bit quantities, etc. I would prefer 
the former solution, but I am interested in hearing what the cdt debug 
experts think.

Matthias