Function Parser for C++ v3.3.2

Authors: Juha Nieminen (http://iki.fi/warp/), Joel Yliluoma (http://iki.fi/bisqwit/).

The usage license of this library is located at the end of this file.

Table of contents:

What's new

What's new in v3.3.2

What's new in v3.3.1

What's new in v3.3

Preface

Often people need to ask some mathematical expression from the user and then evaluate values for that expression. The simplest example is a program which draws the graphic of a user-defined function on screen.

This library adds C-style function string parsing to the program. This means that you can evaluate the string "sqrt(1-x^2+y^2)" with given values of 'x' and 'y'.

The library is intended to be very fast. It byte-compiles the function string at parse time and interpretes this byte-code at evaluation time. The evaluation is straightforward and no recursions are done (uses stack arithmetic).

Empirical tests show that it indeed is very fast (specially compared to libraries which evaluate functions by just interpreting the raw function string).

The library is made in ISO C++ and requires a standard-conforming C++ compiler.

Usage

To use the FunctionParser class, you have to include "fparser.hh" in your source code files which use the FunctionParser class.

When compiling, you have to compile fparser.cc and fpoptimizer.cc and link them to the main program. In some developement environments it's enough to add those two files to your current project (usually header files don't have to be added to the project for the compilation to work).

If you are not going to use the optimizer (ie. you have commented out FP_SUPPORT_OPTIMIZER in fpconfig.hh), you can leave the latter file out.

Configuring the compilation

There is a set of precompiler options in the fpconfig.hh file which can be used for setting certain features on or off. All of these options can also be specified from the outside, using precompiler settings (eg. the -D option in gcc), and thus it's not necessary to modify this file.

FP_SUPPORT_TR1_MATH_FUNCS : (Default off)

Define this precompiler constant to make the library use additional math functions defined in the C99 standard and the C++ TR1 standard proposal (but not yet in the official C++ standard). This can make evaluation faster when these functions are involved.

The C++ TR1 math functions in question are: asinh(), acosh(), atanh(), exp2() and log2().

FP_ENABLE_EVAL : (Default off)

Even though the maximum recursion level of the eval() function is limited, it is still possible to write functions which never reach this maximum recursion level but take enormous amounts of time to evaluate (this can be undesirable eg. in web server-side applications). For this reason this function is disabled by default. You can add support for the eval() function by defining this precompiler constant.

FP_EVAL_MAX_REC_LEVEL : (Default 1000)

Sets the maximum recursion level allowed for eval().

FP_SUPPORT_OPTIMIZER : (Default on)

If you are not going to use the Optimize() method, you can comment this line out to speed-up the compilation a bit, as well as making the binary a bit smaller. (Optimize() can still be called, but it will not do anything.)

You can also disable the optimizer by specifying the FP_NO_SUPPORT_OPTIMIZER precompiler constant in your compiler settings.

FP_EPSILON : (Default 1e-14)

Epsilon value used in comparison operators. If this line is commented out, then no epsilon will be used.

FP_USE_THREAD_SAFE_EVAL : (Default off)

Define this precompiler constant to make Eval() thread-safe. Refer to the thread safety section later in this document for more information. Note that defining this may make Eval() slightly slower.

FP_USE_THREAD_SAFE_EVAL_WITH_ALLOCA : (Default off)

This is like the previous, but makes Eval() use the alloca() function (instead of std::vector). This will make it faster, but the alloca() function is not standard and thus not supported by all compilers.

FP_NO_EVALUATION_CHECKS : (Default off)

If this precompiler constant is defined, no evaluation-time checks will be performed. This may give a slight boost in speed in certain situations. Consult the evaluation checks section below for more information on this subject.

Copying and assignment

The class implements a safe copy constructor and assignment operator.

It uses the copy-on-write technique for efficiency. This means that when copying or assigning a FunctionParser instance, the internal data (which in some cases can be quite lengthy) is not immediately copied but only when the contents of the copy (or the original) are changed.

This means that copying/assigning is a very fast operation, and if the copies are never modified then actual data copying never happens either.

The Eval() and EvalError() methods of the copy can be called without the internal data being copied.

Calling Parse(), Optimize() or the user-defined constant/function adding methods will cause a deep-copy.

Identifier names

An identifier is a function (internal or user-defined), variable, constant or unit. New identifiers can be specified with the functions described in the next subsections.

The name of an identifier can use any alphanumeric characters, the underscore character and any UTF8-encoded unicode character. The first character of the name cannot be a numeric digit, though.

All functions, variables, constants and units must use unique names. It's not possible to add two different identifiers with the same name.

Short descriptions of FunctionParser methods

int Parse(const std::string& Function, const std::string& Vars,
          bool useDegrees = false);

int Parse(const char* Function, const std::string& Vars,
          bool useDegrees = false);

Parses the given function and compiles it to internal format. Return value is -1 if successful, else the index value to the location of the error.


void setDelimiterChar(char);

Sets an ending delimiter character for the function string. (See the long description for more details.)


const char* ErrorMsg(void) const;

Returns an error message corresponding to the error in Parse(), or an empty string if no such error occurred.


ParseErrorType GetParseErrorType() const;

Returns the type of parsing error which occurred. Possible return types are described in the long description.


double Eval(const double* Vars);

Evaluates the function given to Parse().


int EvalError(void) const;

Returns 0 if no error happened in the previous call to Eval(), else an error code >0.


void Optimize();

Tries to optimize the bytecode for faster evaluation.


bool AddConstant(const std::string& name, double value);

Add a constant to the parser. Returns false if the name of the constant is invalid, else true.


bool AddUnit(const std::string& name, double value);

Add a new unit to the parser. Returns false if the name of the unit is invalid, else true.


bool AddFunction(const std::string& name,
                 double (*functionPtr)(const double*),
                 unsigned paramsAmount);

Add a user-defined function to the parser (as a function pointer). Returns false if the name of the function is invalid, else true.


bool AddFunction(const std::string& name, FunctionParser&);

Add a user-defined function to the parser (as a FunctionParser instance). Returns false if the name of the function is invalid, else true.


bool RemoveIdentifier(const std::string& name);

Removes the constant, unit or user-defined function with the specified name from the parser.


int ParseAndDeduceVariables(const std::string& function,
                            int* amountOfVariablesFound = 0,
                            bool useDegrees = false);
int ParseAndDeduceVariables(const std::string& function,
                            std::string& resultVarString,
                            int* amountOfVariablesFound = 0,
                            bool useDegrees = false);
int ParseAndDeduceVariables(const std::string& function,
                            std::vector<std::string>& resultVars,
                            bool useDegrees = false);

Like Parse(), but the variables in the function are deduced automatically. The amount of found variables and the variable names themselves are returned by the different versions of the function.

Long descriptions of FunctionParser methods


int Parse(const std::string& Function, const std::string& Vars,
          bool useDegrees = false);

int Parse(const char* Function, const std::string& Vars,
          bool useDegrees = false);

Parses the given function (and compiles it to internal format). Destroys previous function. Following calls to Eval() will evaluate the given function.

The strings given as parameters are not needed anymore after parsing.

Parameters:
Function String containing the function to parse.
Vars String containing the variable names, separated by commas.
Eg. "x,y", "VarX,VarY,VarZ,n" or "x1,x2,x3,x4,__VAR__".
useDegrees (Optional.) Whether to use degrees or radians in trigonometric functions. (Default: radians)

If a char* is given as the Function parameter, it must be a null-terminated string.

Variables can have any size and they are case sensitive (ie. "var", "VAR" and "Var" are different variable names). Letters, digits, underscores and UTF8-encoded characters can be used in variable names, but the name of a variable can't begin with a digit. Each variable name can appear only once in the 'Vars' string. Function names are not legal variable names.

Using longer variable names causes no overhead whatsoever to the Eval() method, so it's completely safe to use variable names of any size.

The third, optional parameter specifies whether angles should be interpreted as radians or degrees in trigonometrical functions. If not specified, the default value is radians.

Return values:

Example: parser.Parse("3*x+y", "x,y");


void setDelimiterChar(char);

By default the parser expects the entire function string to be valid (ie. the entire contents of the given std::string, or a C string ending in the null character '\0').

If a delimiter character is specified with this function, then if it's encountered at the outermost parsing level by the Parse() function, and the input function has been valid so far, Parse() will return an index to this character inside the input string, but rather than set an error code, FP_NO_ERROR will be set.

The idea is that this can be used to more easily parse functions which are embedded inside larger strings, containing surrounding data, without having to explicitly extract the function to a separate string.

For example, suppose you are writing an interpreter for a scripting language, which can have commands like this:

let MyFunction(x,y) = { sin(x*x+y*y) } // A 2-dimensional function

Normally when parsing such a line you would have to extract the part inside the curly brackets into a separate string and parse it that way. With this feature what you can do instead is to set '}' as the delimiter character and then simply give a pointer to the character which comes after the '{'. If all goes well, the Parse() function will return an index to the '}' character (from the given starting point) and GetParseErrorType() will return FP_NO_ERROR. You can use the return value (if it's not -1) to jump forward in the string to the delimiter character.

Note that a null character ('\0') or the end of the std::string (if one was given) will still be a valid end of the function string even if a delimiter character was specified. (In this case Parse() will return -1 if there was no error, as usual.)

Also note that the delimiter character cannot be any valid operator or alphanumeric (including the underscore) character, nor the other characters defined in the function syntax. It must be a character not supported by the function parser (such as '}', '"', ']', etc).


const char* ErrorMsg(void) const;

Returns a pointer to an error message string corresponding to the error caused by Parse() (you can use this to print the proper error message to the user). If no such error has occurred, returns an empty string.


ParseErrorType GetParseErrorType() const;

Returns the type of parse error which occurred.

This method can be used to get the error type if ErrorMsg() is not enough for printing the error message. In other words, this can be used for printing customized error messages (eg. in another language). If the default error messages suffice, then this method doesn't need to be called. FunctionParser::ParseErrorType is an enumerated type inside the class (ie. its values are accessed like "FunctionParser::SYNTAX_ERROR").

The possible values for FunctionParser::ParseErrorType are listed below, along with their equivalent error message returned by the ErrorMsg() method:

FP_NO_ERROR If no error occurred in the previous call to Parse().
SYNTAX_ERROR "Syntax error"
MISM_PARENTH "Mismatched parenthesis"
MISSING_PARENTH "Missing ')'"
EMPTY_PARENTH "Empty parentheses"
EXPECT_OPERATOR "Syntax error: Operator expected"
OUT_OF_MEMORY "Not enough memory"
UNEXPECTED_ERROR "An unexpected error occurred. Please make a full bug report to the author"
INVALID_VARS "Syntax error in parameter 'Vars' given to FunctionParser::Parse()"
ILL_PARAMS_AMOUNT "Illegal number of parameters to function"
PREMATURE_EOS "Syntax error: Premature end of string"
EXPECT_PARENTH_FUNC "Syntax error: Expecting ( after function"
NO_FUNCTION_PARSED_YET "(No function has been parsed yet)"


double Eval(const double* Vars);

Evaluates the function given to Parse(). The array given as parameter must contain the same amount of values as the amount of variables given to Parse(). Each value corresponds to each variable, in the same order.

Return values:

Example:

double Vars[] = {1, -2.5};
double result = parser.Eval(Vars);


int EvalError(void) const;

Used to test if the call to Eval() succeeded.

Return values:

If there was no error in the previous call to Eval(), returns 0, else returns a positive value as follows:


void Optimize();

This method can be called after calling the Parse() method. It will try to simplify the internal bytecode so that it will evaluate faster (it tries to reduce the amount of opcodes in the bytecode).

For example, the bytecode for the function "5+x*y-25*4/8" will be reduced to a bytecode equivalent to the function "x*y-7.5" (the original 11 opcodes will be reduced to 5). Besides calculating constant expressions (like in the example), it also performs other types of simplifications with variable and function expressions.

This method is quite slow and the decision of whether to use it or not should depend on the type of application. If a function is parsed once and evaluated millions of times, then calling Optimize() may speed-up noticeably. However, if there are tons of functions to parse and each one is evaluated once or just a few times, then calling Optimize() will only slow down the program.

Also, if the original function is expected to be optimal, then calling Optimize() would be useless.

Note: Currently this method does not make any checks (like Eval() does) and thus things like "1/0" will cause undefined behaviour. (On the other hand, if such expression is given to the parser, Eval() will always give an error code, no matter what the parameters.) If caching this type of errors is important, a work-around is to call Eval() once before calling Optimize() and checking EvalError().

If the destination application is not going to use this method, the compiler constant FP_SUPPORT_OPTIMIZER can be undefined in fpconfig.hh to make the library smaller (Optimize() can still be called, but it will not do anything).

(If you are interested in seeing how this method optimizes the opcode, you can call the PrintByteCode() method before and after the call to Optimize() to see the difference.)


bool AddConstant(const std::string& name, double value);

This method can be used to add constants to the parser. Syntactically constants are identical to variables (ie. they follow the same naming rules and they can be used in the function string in the same way as variables), but internally constants are directly replaced with their value at parse time.

Constants used by a function must be added before calling Parse() for that function. Constants are preserved between Parse() calls in the current FunctionParser instance, so they don't need to be added but once. (If you use the same constant in several instances of FunctionParser, you will need to add it to all the instances separately.)

Constants can be added at any time and the value of old constants can be changed, but new additions and changes will only have effect the next time Parse() is called. (That is, changing the value of a constant after calling Parse() and before calling Eval() will have no effect.)

The return value will be false if the 'name' of the constant was illegal, else true. If the name was illegal, the method does nothing.

Example: parser.AddConstant("pi", 3.1415926535897932);

Now for example parser.Parse("x*pi", "x"); will be identical to the call parser.Parse("x*3.1415926535897932", "x");


bool AddUnit(const std::string& name, double value);

In some applications it is desirable to have units of measurement. A typical example is an application which creates a page layout to be printed. When printing, distances are usually measured in points (defined by the resolution of the printer). However, it is often more useful for the user to be able to specify measurements in other units such as centimeters or inches.

A unit is simply a value by which the preceding element is multiplied. For example, if the printing has been set up to 300 DPI, one inch is then 300 points (dots). Thus saying eg. "5in" is the same as saying "5*300" or "1500" (assuming "in" has been added as a unit with the value 300).

Note that units are slightly different from a multiplication in that they have a higher precedence than any other operator (except parentheses). Thus for example "5/2in" is parsed as "5/(2*300)". (If 5/2 inches is what one wants, it has to be written "(5/2)in".)

You can use the AddUnit() method to add a new unit. The unit can then be used after any element in the function (and will work as a multiplier for that element). An element is a float literal, a constant, a variable, a function or any expression in parentheses. When the element is not a float literal nor an expression in parentheses, there has to naturally be at least one whitespace between the element and the unit (eg. "x in"). To change the value of a unit, call AddUnit() again with the same unit name and the new value.

Unit names share the same namespace as constants, functions and variables, and thus should be distinct from those.

Example: parser.AddUnit("in", 300);

Now for example the function "5in" will be identical to "(5*300)". Other usage examples include "x in", "3in+2", "pow(x,2)in", "(x+2)in".


bool AddFunction(const std::string& name,
                 double (*functionPtr)(const double*),
                 unsigned paramsAmount);
This method can be used to add new functions to the parser. For example, if you would like to add a function "sqr(A)" which squares the value of A, you can do it with this method (so that you don't need to touch the source code of the parser).

The method takes three parameters:

The return value will be false if the given name was invalid (either it did not follow the variable naming conventions, or the name was already reserved), else true. If the return value is false, nothing is added.

Example: Suppose we have a C++ function like this:

double Square(const double* p)
{
    return p[0]*p[0];
}

Now we can add this function to the parser like this:

parser.AddFunction("sqr", Square, 1);
parser.Parse("2*sqr(x)", "x");

An example of a useful function taking no parameters is a function returning a random value. For example:

double Rand(const double*)
{
    return drand48();
}

parser.AddFunction("rand", Rand, 0);

Important note: If you use the Optimize() method, it will assume that the user-given function has no side-effects, that is, it always returns the same value for the same parameters. The optimizer will optimize the function call away in some cases, making this assumption. (The Rand() function given as example above is one such problematic case.)


bool AddFunction(const std::string& name, FunctionParser&);

This method is almost identical to the previous AddFunction(), but instead of taking a C++ function, it takes another FunctionParser instance.

There are some important restrictions on making a FunctionParser instance call another:

Example:

FunctionParser f1, f2;

f1.Parse("x*x", "x");

f2.AddFunction("sqr", f1);

This version of the AddFunction() method can be useful to eg. chain user-given functions. For example, ask the user for a function F1, and then ask the user another function F2, but now the user can call F1 in this second function if he wants (and so on with a third function F3, where he can call F1 and F2, etc).


bool RemoveIdentifier(const std::string& name);

If a constant, unit or user-defined function with the specified name exists in the parser, it will be removed and the return value will be true, else nothing will be done and the return value will be false.

(Note: If you want to remove everything from an existing FunctionParser instance, simply assign a fresh instance to it, ie. like "parser = FunctionParser();")


int ParseAndDeduceVariables(const std::string& function,
                            int* amountOfVariablesFound = 0,
                            bool useDegrees = false);
int ParseAndDeduceVariables(const std::string& function,
                            std::string& resultVarString,
                            int* amountOfVariablesFound = 0,
                            bool useDegrees = false);
int ParseAndDeduceVariables(const std::string& function,
                            std::vector<std::string>& resultVars,
                            bool useDegrees = false);

These functions work in the same way as the Parse() function, but the variables in the input function string are deduced automatically. The parameters are:

As with Parse(), the return value will be -1 if the parsing succeeded, else an index to the location of the error. None of the specified return values will be modified in case of error.

About evaluation-time checks

By default FunctionParser::Eval() will perform certain sanity checks before performing certain operations. For example, before calling the sqrt function, it will check if the parameter is negative, and if so, it will set the proper error code instead of calling the function. These checks include:

However, the library can not guarantee that it will catch all possible floating point errors before performing them, because this is impossible to do with standard C++. For example, dividing a very large value by a value which is very close to zero, or calculating the logarithm of a very small value may overflow the result, as well as multiplying two very large values. Raising a negative number to a non-integral power may cause a NaN result, etc.

In general, as stated, the library cannot guarantee that floating point errors will never happen during evaluation. This can make the library to return the floating point values inf and NaN. Moreover, if floating point errors cause an interrupt in the target computer architecture and/or when using certain compiler settings, this library cannot guarantee that it will never happen.

Since not all error situations can be caught, and since the sanity checks only slow down the evaluation (although only very slightly), the precompiler constant FP_NO_EVALUATION_CHECKS can be used to turn all the checks off. This might make the evaluation slightly faster in certain situations.

Note that the optimizer never performs any sanity checks.

About thread safety

None of the member functions of the FunctionParser class are thread-safe. Most prominently, the Eval() function is not thread-safe. (In other words, the Eval() function of a single FunctionParser instance cannot be safely called simultaneously by two threads.)

There are ways to use this library in a thread-safe way, though. If each thread uses its own FunctionParser instance, no problems will obviously happen. Note, however, that if these instances need to be a copy of a given FunctionParser instance (eg. one where the user has entered a function), a deep copy of this instance has to be performed for each thread. By default FunctionParser uses shallow-copying (copy-on-write), which means that a simple assignment of copy construction will not copy the data itself. To force a deep copy you can all the ForceDeepCopy() function on each of the instances of each thread after the assignment or copying has been done.

Another possibility is to compile the FunctionParser library so that its Eval() function will be thread-safe. (This can be done by defining the FP_USE_THREAD_SAFE_EVAL or the FP_USE_THREAD_SAFE_EVAL_WITH_ALLOCA precompiler constant.) As long as only one thread calls the other functions of FunctionParser, the other threads can safely call the Eval() of this one instance.

Note, however, that compiling the library like this can make Eval() slightly slower. (The alloca version, if supported by the compiler, will not be as slow.)

The function string syntax

The function string understood by the class is very similar to the C-syntax. Arithmetic float expressions can be created from float literals, variables or functions using the following operators in this order of precedence:

() expressions in parentheses first
A unit a unit multiplier (if one has been added)
A^B exponentiation (A raised to the power B)
-A unary minus
!A unary logical not (result is 1 if int(A) is 0, else 0)
A*B A/B A%B multiplication, division and modulo
A+B A-B addition and subtraction
A=B A<B A<=B
A!=B A>B A>=B
comparison between A and B (result is either 0 or 1)
A&B result is 1 if int(A) and int(B) differ from 0, else 0
A|B result is 1 if int(A) or int(B) differ from 0, else 0

Since the unary minus has higher precedence than any other operator, for example the following expression is valid: x*-y

The comparison operators use an epsilon value, so expressions which may differ in very least-significant digits should work correctly. For example, "0.1+0.1+0.1+0.1+0.1+0.1+0.1+0.1+0.1+0.1 = 1" should always return 1, and the same comparison done with ">" or "<" should always return 0. (The epsilon value can be configured in the fpconfig.hh file.) Without epsilon this comparison probably returns the wrong value.

The class supports these functions:

abs(A) Absolute value of A. If A is negative, returns -A otherwise returns A.
acos(A) Arc-cosine of A. Returns the angle, measured in radians, whose cosine is A.
acosh(A) Same as acos() but for hyperbolic cosine.
asin(A) Arc-sine of A. Returns the angle, measured in radians, whose sine is A.
asinh(A) Same as asin() but for hyperbolic sine.
atan(A) Arc-tangent of (A). Returns the angle, measured in radians, whose tangent is (A).
atan2(A,B) Arc-tangent of A/B. The two main differences to atan() is that it will return the right angle depending on the signs of A and B (atan() can only return values betwen -pi/2 and pi/2), and that the return value of pi/2 and -pi/2 are possible.
atanh(A) Same as atan() but for hyperbolic tangent.
ceil(A) Ceiling of A. Returns the smallest integer greater than A. Rounds up to the next higher integer.
cos(A) Cosine of A. Returns the cosine of the angle A, where A is measured in radians.
cosh(A) Same as cos() but for hyperbolic cosine.
cot(A) Cotangent of A (equivalent to 1/tan(A)).
csc(A) Cosecant of A (equivalent to 1/sin(A)).
eval(...) This a recursive call to the function to be evaluated. The number of parameters must be the same as the number of parameters taken by the function. Must be called inside if() to avoid infinite recursion.
exp(A) Exponential of A. Returns the value of e raised to the power A where e is the base of the natural logarithm, i.e. the non-repeating value approximately equal to 2.71828182846.
floor(A) Floor of A. Returns the largest integer less than A. Rounds down to the next lower integer.
if(A,B,C) If int(A) differs from 0, the return value of this function is B, else C. Only the parameter which needs to be evaluated is evaluated, the other parameter is skipped; this makes it safe to use eval() in them.
int(A) Rounds A to the closest integer. 0.5 is rounded to 1.
log(A) Natural (base e) logarithm of A.
log10(A) Base 10 logarithm of A.
max(A,B) If A>B, the result is A, else B.
min(A,B) If A<B, the result is A, else B.
pow(A,B) Exponentiation (A raised to the power B).
sec(A) Secant of A (equivalent to 1/cos(A)).
sin(A) Sine of A. Returns the sine of the angle A, where A is measured in radians.
sinh(A) Same as sin() but for hyperbolic sine.
sqrt(A) Square root of A. Returns the value whose square is A.
tan(A) Tangent of A. Returns the tangent of the angle A, where A is measured in radians.
tanh(A) Same as tan() but for hyperbolic tangent.

Examples of function string understood by the class:

"1+2"
"x-1"
"-sin(sqrt(x^2+y^2))"
"sqrt(XCoord*XCoord + YCoord*YCoord)"

An example of a recursive function is the factorial function: "if(n>1, n*eval(n-1), 1)"

Note that a recursive call has some overhead, which makes it a bit slower than any other operation. It may be a good idea to avoid recursive functions in very time-critical applications. Recursion also takes some memory, so extremely deep recursions should be avoided (eg. millions of nested recursive calls).

Also note that even though the maximum recursion level of eval() is limited, it is possible to write functions which never reach that level but still take enormous amounts of time to evaluate. This can sometimes be undesirable because it is prone to exploitation, which is why eval() is disabled by default. It can be enabled in the fpconfig.hh file.

Tips and tricks

Add constants automatically to all parser objects

Often the same constants (such as pi and e) and other user-defined identifiers (such as units) are always used in all the FunctionParser objects throughout the program. It would be troublesome to always have to manually add these constants every time a new parser object is created.

There is, however, a simple way to always add these user-defined identifiers to all instances. Write a class like this:

    class ParserWithConsts: public FunctionParser
    {
     public:
        ParserWithConsts()
        {
            AddConstant("pi", 3.14159265358979323846);
            AddConstant("e", 2.71828182845904523536);
        }
    };

Now instead of using FunctionParser, always use ParserWithConsts. It will behave identically except that the constants (and possibly other user-defined identifiers) will always be automatically defined. (Objects of this type even survive slicing, so they are completely safe to use anywhere.)

Contacting the author

Any comments, bug reports, etc. should be sent to warp@iki.fi

Usage license

Copyright © 2003-2008 Juha Nieminen, Joel Yliluoma

This library is distributed under two distinct usage licenses depending on the software ("Software" below) which uses the Function Parser library ("Library" below).

The reason for having two distinct usage licenses is to make the library compatible with the GPL license while still being usable in other non-GPL (even commercial) software.


License A

If the Software using the Library is distributed under the GPL license, then the Library can be used under the GPL license as well.

The Library will be under the GPL license only when used with the Software. If the Library is separated from the Software and used in another different software under a different license, then the Library will have the B) license below.

Exception to the above: If the Library is modified for the GPL Software, then the Library cannot be used with the B) license without the express permission of the author of the modifications. A modified library will be under the GPL license by default. That is, only the original, unmodified version of the Library can be taken to another software with the B) license below.

The author of the Software should provide an URL to the original version of the Library if the one used in the Software has been modified. (http://iki.fi/warp/FunctionParser/)

This text file must be distributed in its original intact form along with the sources of the Library. (Documentation about possible modifications to the library should be put in a different text file.)

As per the GPL license, the copyright notification at the beginning of this section should accompany any significant portions of the Library. Proper credits in any Software using the Library is appreciated.


License B

If the Software using the Library is not distributed under the GPL license but under any other license, then the following usage license applies to the Library:

  1. This library is free for non-commercial usage. You can do whatever you like with it as long as you don't claim you made it yourself.
  2. It is possible to use this library in a commercial program, but in this case you MUST contact me first (warp@iki.fi) and ask express permission for this. (Read explanation at the end of the file.)

    If you are making a free program or a shareware program with just a nominal price (5 US dollars or less), you don't have to ask for permission.

    In any case, I DON'T WANT MONEY for the usage of this library. It is free, period.

  3. You can make any modifications you want to it so that it conforms your needs. If you make modifications to it, you have, of course, credits for the modified parts.
  4. If you use this library in your own program, you don't have to provide the source code if you don't want to (ie. the source code of your program or this library).

    If you DO include the source code for this library, this text file must be included in its original intact form.

  5. If you distribute a program which uses this library, and specially if you provide the source code, proper credits MUST be included. Trying to obfuscate the fact that this library is not made by you or that it is free is expressly prohibited. When crediting the usage of this library, it's enough to include a copyright notice and the name of the authors, that is: Juha Nieminen, Joel Yliluoma.

    Also a URL to the library download page would be nice, although not required. The official URL is:

    http://iki.fi/warp/FunctionParser/

  6. And the necessary "lawyer stuff":

    The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

    The software is provided "as is", without warranty of any kind, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose and noninfringement. In no event shall the authors or copyright holders be liable for any claim, damages or other liability, whether in an action of contract, tort or otherwise, arising from, out of or in connection with the software or the use or other dealings in the software.

Explanation of the section 2 of the B) license above:

The section 2 tries to define "fair use" of the library in commercial programs.

"Fair use" of the library means that the program is not heavily dependent on the library, but the library only provides a minor secondary feature to the program.

"Heavily dependent" means that the program depends so much on the library that without it the functionality of the program would be seriously degraded or the program would even become completely non-functional.

In other words: If the program does not depend heavily on the library, that is, the library only provides a minor secondary feature which could be removed without the program being degraded in any considerable way, then it's OK to use the library in the commercial program.

If, however, the program depends so heavily on the library that removing it would make the program non-functional or degrade its functionality considerably, then it's NOT OK to use the library.

The ideology behind this is that it's not fair to use a free library as a base for a commercial program, but it's fair if the library is just a minor, unimportant extra.

If you are going to ask me for permission to use the library in a commercial program, please describe the feature which the library will be providing and how important it is to the program.