Calc/Implementation/Spreadsheet Functions

From Apache OpenOffice Wiki
< Calc‎ | Implementation
Revision as of 20:06, 24 August 2008 by ErAck (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Procedure to add a new Calc spreadsheet function.


Let's assume you want to implement a new spreadsheet function and the function was defined by the OASIS OpenDocument Format Formula subcommittee, see latest revision of the specification (draft). Let's further assume the function's name is MYFUNC and will take 2 parameters, of which the second parameter is optional and defaulted to 0, and returns a number, as following:

Syntax: MYFUNC( Number Param1 [ ; Number Param2 = 0 ] )
Returns: Number


These are the defines used by the resources for function names and the Function Wizard, and the numerical values of OpCode for the formula compiler and interpreter. Note that once defined the names must not be changed because they are used by the localization tools as identifiers.

Add a new define, in this case for 2 parameters append it to the section for functions with more than 1 parameter near the end of the file. Name the define SC_OPCODE_MYFUNC and insert it right before the define of SC_OPCODE_STOP_2_PAR, give it the value SC_OPCODE_STOP_2_PAR had, and increment the values of SC_OPCODE_STOP_2_PAR and SC_OPCODE_LAST_OPCODE_ID. If before the section looked like

#define SC_OPCODE_STOP_2_PAR        393
#define SC_OPCODE_LAST_OPCODE_ID    392      /* last OpCode */

it should then be

#define SC_OPCODE_MYFUNC            393
#define SC_OPCODE_STOP_2_PAR        394
#define SC_OPCODE_LAST_OPCODE_ID    393      /* last OpCode */


Here the OpCodeEnum values are defined. Note that in a non-product build (--enable-dbgutil during configure) there is a typedef OpCodeEnum OpCode; to show enum names in the debugger, while in a product build it is typedef OpCodeEnum USHORT; to save some memory, since compilers tend to produce an int for an enum.

Find a "right" place for the new enum. Although the way the definitions are setup the placement doesn't matter, there are sections with different topics, such as String functions and Statistical functions. Maybe the correct place for MYFUNC would be under miscellaneous. Best practice is to add a new OpCode to the end of such section. Name the OpCode ocMyFunc and add the line

        ocMyFunc            = SC_OPCODE_MYFUNC,


These are the resources for function names. There are 3 resource bundles:

English UI display names. These get localized for the UI of other languages.
These English names are used internally to store/load ODF v1.0/v1.1 and for API XFunctionAccess. Usually the name is identical to that in RID_SC_FUNCTION_NAMES. Once defined and "in the wild", the name must not be changed.
These English names are used internally to store/load ODFF aka OpenFormula as of ODF v1.2. Once defined, the name must not be changed.

The new function name must be defined for all 3 resource bundles.

To the end of

        Text [ en-US ] = "MYFUNC" ;

The [ en-US ] field tells the localization tools that the name may be localized.

To the end of

    String SC_OPCODE_MYFUNC { Text = "MYFUNC" ; };

The absence of the [ en-US ] field tells the localization tools that the name must not be localized.

To the end of

    String SC_OPCODE_MYFUNC { Text = "MYFUNC" ; };

Again, the absence of the [ en-US ] field tells the localization tools that the name must not be localized.

The compiler knows the function

After having added the necessary entries to sc/inc/compiler.hrc, sc/inc/opcode.hxx and sc/source/core/src/compiler.src, the formula compiler now knows the new function name and is able to compile an expression where it is used, and it can be stored in and loaded from a document. Of course nothing else works, the interpreter doesn't know how to handle it and will generate an error if encountered. The function and its parameters will not appear in the Function Wizard.


Function groups (categories) and HelpIDs for functions are defined here. Lookup the group where functions of ID_FUNCTION_GRP_... matching the new function's category are defined and append an entry, incrementing the offset of the last entry by one. For our function that could be



This large resource file contains all elements necessary to display functions in the Function Wizard. It defines the function's short description, the number of parameters, whether they are optional, and the description of each parameter. For a detailed description of fields see the comment on top of the file.

Add the new function to the end of one of the two resource blocks RID_SC_FUNCTION_DESCRIPTIONS1 or RID_SC_FUNCTION_DESCRIPTIONS2. Which one doesn't really matter, but functions should be more or less equally distributed over the two blocks. There are two blocks because one resource block couldn't have more than 64k data, just another legacy from Win16 times.. looked up again right now (2008-08-23) in the meantime this restriction seems to have been obsoleted, which would have to be verified by a test build using one resource block only though.

        String 1 // Description
            Text [ en-US ] = "Calculates foo, optionally using bar." ;
        ExtraData =
            U2S( HID_FUNC_MYFUNC );
            2;  0;  1;
        String 2 // Name of Parameter 1
            Text [ en-US ] = "Number" ;
        String 3 // Description of Parameter 1
            Text [ en-US ] = "A value for which foo is to be calculated." ;
        String 4 // Name of Parameter 2
            Text [ en-US ] = "Number" ;
        String 5 // Description of Parameter 2
            Text [ en-US ] = "The bar value." ;

Some comments on the ExtraData block, for more details see scfuncs.src:

ExtraData =
    0;                      // The function is not suppressed and available in UI.
    ID_FUNCTION_GRP_MATH;   // The category in which the function is displayed.
    U2S( HID_FUNC_MYFUNC ); // The HelpID of this function.
    2;  0;  1;              // 2 parameters, of which the 2nd is optional.
    0;                      // None of the parameters are suppressed in the UI.


Here the HelpIDs to be used within the Function Wizard are propagated to the help system. Go to the end of the section containing HID_FUNC_... and append an entry,

hidspecial HID_FUNC_MYFUNC          { HelpID = HID_FUNC_MYFUNC; };

The Function Wizard knows the function

Now the Function Wizard can display the function and its parameters, and online-help may be authored.


Add a member method to class ScInterpreter that will handle the new function. Here this would be

void ScMyFunc();

Take care that the new member function doesn't resemble the name of some already existing class, for example the method for the ADDRESS() function is named ScAddressFunc() because ScAddress(), if called without this-> prefix, would be the ctor of class ScAddress instead. If in doubt, add ...Func() to the name.


In method ScInterpreter::Interpret() add to switch ( eOp ) the call to the member function for the OpCode:

                case ocMyFunc           : ScMyFunc();                   break;


Pick one of the interpr?.cxx source files where the new method may fit. There is no general advice which file exactly that might be, be sensible. As a guide line

Basic math functions and informational functions.
Date functions and financial functions.
Statistical functions.
Interpreter managing and stack related functions, you shouldn't need to add anything there.
Matrix functions.
Some number crunching that in the past had to be compiled without optimizations, though in the mean time code changed and this probably is not necessary anymore.

For our function, since that expects two, one optional, numerical scalar arguments, this would be:

void ScInterpreter::ScMyFunc()
    BYTE nParamCount = GetByte();
    // The MustHaveParamCount...() functions check the number of parameters and
    // if they do not fit push an error on the stack, if the method fails
    // (returns false) we return immediately.
    if ( !MustHaveParamCount( nParamCount, 1, 2 ) )
    // Arguments are popped from the stack from right to left.
    double fParam2;
    if (nParamCount == 2)
        fParam2 = GetDouble();
        fParam2 = 0.0;
    double fVal = GetDouble();
    if ( /* does fVal meet all constraints */ )
        double fResult = /* calculate foo */ ;
        PushDouble( fResult );

The not so easy case of non-scalar arguments

TODO: section needs elaboration.

If the new function would handle parameters that are not scalar values, for example a NumberSequence or matrix/array, they would have to be treated explicitly, checking and reacting on the type of each argument. Lookout for functions that use the GetType() call and handle StackVar svDoubleRef or similar. Ask on the dev@sc mailing list if in doubt.

In this case the parameter types have to be added to file


that describes how parameters are to be treated in detail. Please read the comments there and in sc/source/core/inc/parclass.hxx. Ask on the dev@sc mailing list if in doubt.

Personal tools