1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
4 <meta http-equiv="Content-Type" content="text/html; charset=ISO-8859-1">
5 <link href="style.css" rel="stylesheet" type="text/css" title="normal" media=screen>
6 <title>Function Parser for C++ v4.3 : Documentation</title>
10 <h1>Function Parser for C++ v4.3 </h1>
12 <p>Authors: Juha Nieminen
13 (<a href="http://iki.fi/warp/">http://iki.fi/warp/</a>),
15 (<a href="http://iki.fi/bisqwit/">http://iki.fi/bisqwit/</a>).
17 <p>The usage license of this library is located at the end of this file.
19 <h2>Table of contents:</h2>
22 <li><a href="#whatsnew">What's new</a>
23 <li><a href="#preface">Preface</a>
24 <li><a href="#usage">Usage</a>
26 <li><a href="#parsertypes">Parser types</a>
27 <li><a href="#configuring">Configuring the compilation</a>
28 <li><a href="#copyassignment">Copying and assignment</a>
29 <li><a href="#shortdesc">Short descriptions of FunctionParser methods</a>
30 <li><a href="#longdesc">Long descriptions of FunctionParser methods</a>
32 <li><a href="#longdesc_Parse"><code>Parse()</code></a>
33 <li><a href="#longdesc_setDelimiterChar"><code>setDelimiterChar()</code></a>
34 <li><a href="#longdesc_ErrorMsg"><code>ErrorMsg()</code></a>
35 <li><a href="#longdesc_GetParseErrorType"><code>GetParseErrorType()</code></a>
36 <li><a href="#longdesc_Eval"><code>Eval()</code></a>
37 <li><a href="#longdesc_EvalError"><code>EvalError()</code></a>
38 <li><a href="#longdesc_Optimize"><code>Optimize()</code></a>
39 <li><a href="#longdesc_AddConstant"><code>AddConstant()</code></a>
40 <li><a href="#longdesc_AddUnit"><code>AddUnit()</code></a>
41 <li><a href="#longdesc_AddFunction1"><code>AddFunction()</code></a> (C++ function)
42 <li><a href="#longdesc_AddFunction2"><code>AddFunction()</code></a> (FunctionParser)
43 <li><a href="#longdesc_RemoveIdentifier"><code>RemoveIdentifier()</code></a>
44 <li><a href="#longdesc_ParseAndDeduceVariables"><code>ParseAndDeduceVariables()</code></a>
49 <li><a href="#literals">Numeric literals</a>
50 <li><a href="#identifiers">Identifier names</a>
51 <li><a href="#functionsyntax">The function string syntax</a>
52 <li><a href="#inlinevars">Inline variables</a>
53 <li><a href="#whitespace">Whitespace</a>
57 <li><a href="#evaluationchecks">About evaluation-time checks</a>
58 <li><a href="#threadsafety">About thread safety</a>
59 <li><a href="#tipsandtricks">Tips and tricks</a>
60 <li><a href="#contact">Contacting the author</a>
62 <!-- <li><a href="#algorithm">The algorithm used in the library</a> -->
63 <li><a href="#license">Usage license</a>
66 <a name="whatsnew"></a>
71 <li>Function syntax enhancement: Added possibility of defining new
72 variables in the function string itself. (See documentation for details.)
73 <li>Fixed some bugs in the optimizer (among others,
74 <code>"atan2(-x,-y)"</code> was being wrongly optimized into
75 <code>"atan2(x,y)"</code>
80 <li>The <code>Optimize()</code> method now works also with the
81 <code>float</code> and <code>long double</code> versions of the library.
83 <li>Some new optimizations added.</li>
84 <li>There was a call to the C99/C++0x function <code>isinf()</code> in the
85 optimizer which made it not compile with some compilers. This call has
91 <li>Official support for hexadecimal literals (for all the parser types).
92 Previously there was support only if the <code>strto...()</code> C
93 library functions happened to support them (which is a non-standard
94 extension of those functions). See documentation for details.</li>
95 <li>Significant amount of new optimizations performed by
96 <code>Parse()</code>.</li>
97 <li>Minor bugfixes.</li>
101 <!-- -------------------------------------------------------------------- -->
102 <a name="preface"></a>
105 <p>This C++ library offers a class which can be used to parse and evaluate a
106 mathematical function from a string (which might be eg. requested from the
107 user). The syntax of the function string is similar to mathematical expressions
108 written in C/C++ (the exact syntax is specified later in this document).
109 The function can then be evaluated with different values of variables.
111 <p>For example, a function like "<code>sin(sqrt(x*x+y*y))</code>" can be
112 parsed from a string (either <code>std::string</code> or a C-style string)
113 and then evaluated with different values of <code>x</code> and <code>y</code>.
114 This library can be useful for evaluating user-inputted functions, or in
115 some cases interpreting mathematical expressions in a scripting language.
117 <p>This library aims for maximum speed in both parsing and evaluation, while
118 keeping maximum portability. The library should compile and work with any
119 standard-conforming C++ compiler.
121 <p>Different numerical types are supported: <code>double</code>,
122 <code>float</code>, <code>long double</code>, <code>long int</code>,
123 multiple-precision floating point numbers using the MPFR library, and
124 arbitrary precision integers using the GMP library. (Note that it's
125 not necessary for these two libraries to exist in the system in order
126 to use the Function Parser library with the other numerical types. Support
127 for these libraries is optionally compiled in using preprocessor settings.)
130 <!-- -------------------------------------------------------------------- -->
134 <p>To use the <code>FunctionParser</code> class, you have to include
135 <code>"fparser.hh"</code> in your source code files which use the
136 <code>FunctionParser</code> class.
138 <p>If you are going to use the MPFR version of the library, you need to
139 include <code>"fparser_mpfr.hh"</code>. If you are going to use the GMP
140 version of the library, you need to include <code>"fparser_gmpint.hh"</code>.
141 (Note that support for these special parser versions needs to be specified
142 with preprocessor macros. See the <a href="#parsertypes">documentation
143 below</a> for details.)
145 <p>When compiling, you have to compile <code>fparser.cc</code> and
146 <code>fpoptimizer.cc</code> and link them to the main program. In many
147 developement environments it's enough to add those two files to your
148 current project (usually header files don't have to be added to the
149 project for the compilation to work).
151 <p>If you are going to use the MPFR or the GMP versions of the library,
152 you also need to add <code>mpfr/MpfrFloat.cc</code> or
153 <code>mpfr/GmpInt.cc</code> files to your project, respectively. Otherwise
154 they should not be added to the project.
156 <p>Note that part of the library source code is inside several
157 <code>.inc</code> files (these files contain auto-generated C++ code),
158 provided in the library package. These files are used by
159 <code>fparser.cc</code> and don't need to be added explicitly to the
160 project in most IDEs (such as Visual Studio). Basically, you don't need
161 to do anything with these files, other than keep them in the same directory
162 as <code>fparser.cc</code>.
164 <p>Simple usage example of the library:
168 fp.Parse("sqrt(x*x + y*y)", "x,y");
169 double variables[2] = { 1.5, 2.9 };
170 double result = fp.Eval(variables);
173 <!-- -------------------------------------------------------------------- -->
174 <a name="parsertypes"></a>
175 <h3>Parser types</h3>
177 <p>Different versions of the function parser class are supported, using
178 different floating point or integral types for function evaluation.
180 <p>All the classes other than the default one, <code>FunctionParser</code>,
181 need to be enabled at compile time by defining a preprocessor macro
182 (specified below) either in the <code>fpconfig.hh</code> file or your
183 compiler settings. (The reason for this is that all the other parser types
184 use either C99 standard libraries not yet available in the official C++
185 standard or the third-party libraries GMP and MPFR. It also makes
186 compilation faster and the resulting binary smaller when unused classes
187 are not compiled in.).
189 <p>Note that if you try to use the other class types without enabling them
190 with the correspondent preprocessor macro, you will get a linker error
191 (rather than a compiler error) because those classes will not have been
192 instantiated when the library was compiled.
194 <p>Currently the <code>Optimize()</code> method works only for the
195 <code>FunctionParser</code>, <code>FunctionParser_f</code> and
196 <code>FunctionParser_ld</code> classes. For the other types it can be
197 called but it does nothing.
201 <dt><p><code>FunctionParser</code></dt>
203 <p>This is the default class, which uses <code>double</code> as its
204 numerical type. This is the only class enabled by default.
205 <p>If you use some other type than this one, and you don't want this
206 version of the class compiled into the library, you can define the
207 preprocessor macro <code>FP_DISABLE_DOUBLE_TYPE</code>.
210 <dt><p><code>FunctionParser_f</code></dt>
212 <p>This parser uses <code>float</code> as its numerical type.
213 <p>The <code>FP_SUPPORT_FLOAT_TYPE</code> preprocessor macro needs to be
214 defined for this class to be enabled.
215 <p>Note that this version of the class uses C99 standard math functions
216 not yet available in all C++ compilers. Also the <code>ld</code> version
217 of the class described below use such functions.
220 <dt><p><code>FunctionParser_ld</code></dt>
222 <p>This parser uses <code>long double</code> as its numerical type.
223 <p>The <code>FP_SUPPORT_LONG_DOUBLE_TYPE</code> preprocessor macro needs
224 to be defined for this class to be enabled.
227 <dt><p><code>FunctionParser_li</code></dt>
229 <p>This parser uses <code>long int</code> as its numerical type.
230 <p>The <code>FP_SUPPORT_LONG_INT_TYPE</code> preprocessor macro needs
231 to be defined for this class to be enabled.
232 <p>Note that this version of the class uses a reduced function syntax
233 with support only for functions which are feasible to be used with
234 integral types (namely <code>abs()</code>, <code>eval()</code>,
235 <code>if()</code>, <code>min()</code> and <code>max()</code>, besides
236 basic arithmetic operators, except for the power operator).
239 <dt><p><code>FunctionParser_mpfr</code></dt>
241 <p>This parser uses <code>MpfrFloat</code> as its numerical type.
242 <p>The <code>FP_SUPPORT_MPFR_FLOAT_TYPE</code> preprocessor macro needs
243 to be defined for this class to be enabled.
244 <p>Note that to use this version of the parser,
245 <code>"fparser_mpfr.hh"</code> needs to be included.
246 <p><code>MpfrFloat</code> is an auxiliary class which uses the MPFR
247 library for multiple-precision floating point numbers. The class
248 behaves largely like a floating point type, and is declared in the
249 <code>mpfr/MpfrFloat.hh</code> file (see that file for info about
250 the public interface of the class).
251 <p>If this class is enabled, <code>mpfr/MpfrFloat.cc</code>
252 needs to be compiled into the project, as well as the GMP and MPFR
253 libraries. (With the gcc compiler this means using the linker options
254 "<code>-lgmp -lmpfr</code>".)
257 <dt><p><code>FunctionParser_gmpint</code></dt>
259 <p>This parser uses <code>GmpInt</code> as its numerical type.
260 <p>The <code>FP_SUPPORT_GMP_INT_TYPE</code> preprocessor macro needs
261 to be defined for this class to be enabled.
262 <p>Note that to use this version of the parser,
263 <code>"fparser_gmpint.hh"</code> needs to be included.
264 <p><code>GmpInt</code> is an auxiliary class which uses the GMP
265 library for arbitrary-precision integer numbers. The class
266 behaves largely like an integer type, and is declared in the
267 <code>mpfr/GmpInt.hh</code> file (see that file for info about
268 the public interface of the class).
269 <p>If this class is enabled, <code>mpfr/GmpInt.cc</code>
270 needs to be compiled into the project, as well as the GMP library.
271 <p>This version of the class also uses a reduced version of the syntax,
272 like the <code>long int</code> version.
273 <p><b>Note:</b> Since there's no upper limit to the size of GMP
274 integers, this version of the class should be used with care in
275 situations where malicious users might be able to exploit it to
276 make the program run out of memory. An example of this would be
277 a server-side application usable through the WWW.
281 <p>Note that these different classes are completely independent and
282 instances of different classes cannot be given to each other using the
283 <code>AddFunction()</code> method. Only objects of the same type can
284 be given to that method.
286 <p>The rest of the documentation assumes that <code>FunctionParser</code>
287 (which uses the <code>double</code> type) is used. The usage of the other
288 classes is identical except that <code>double</code> is replaced with the
289 correspondent type used by that class. (In other words, whenever the
290 rest of this documentation uses the type keyword '<code>double</code>',
291 the correspondent type should be used instead, when using another version
294 <!-- -------------------------------------------------------------------- -->
295 <a name="configuring"></a>
296 <h3>Configuring the compilation</h3>
298 <p>There is a set of precompiler options in the <code>fpconfig.hh</code> file
299 which can be used for setting certain features on or off. All of these options
300 can also be specified from the outside, using precompiler settings (eg. the
301 <code>-D</code> option in gcc), and thus it's not necessary to modify this
305 <dt><p><code>FP_SUPPORT_TR1_MATH_FUNCS</code> : (Default off)</dt>
306 <dd><p>Define this precompiler constant to make the library use additional
307 math functions defined in the C99 standard and the C++ TR1 standard
308 proposal (but not yet in the official C++ standard). This can make
309 evaluation faster when these functions are involved.
310 <p>The C++ TR1 math functions in question are: <code>asinh()</code>,
311 <code>acosh()</code>, <code>atanh()</code>, <code>exp2()</code> and
315 <dt><p><code>FP_ENABLE_EVAL</code> : (Default off)</dt>
316 <dd><p>Even though the maximum recursion level of the <code>eval()</code>
317 function is limited, it is still possible to write functions which never
318 reach this maximum recursion level but take enormous amounts of
319 time to evaluate (this can be undesirable eg. in web server-side
320 applications). For this reason this function is disabled by default.
321 You can add support for the <code>eval()</code> function by
322 defining this precompiler constant.
325 <dt><p><code>FP_EVAL_MAX_REC_LEVEL</code> : (Default 1000)</dt>
326 <dd><p>Sets the maximum recursion level allowed for <code>eval()</code>.
329 <dt><p><code>FP_SUPPORT_OPTIMIZER</code> : (Default on)</dt>
330 <dd><p>If you are not going to use the <code>Optimize()</code> method, you
331 can comment this line out to speed-up the compilation a bit, as
332 well as making the binary a bit smaller. (<code>Optimize()</code> can
333 still be called, but it will not do anything.)
335 <p>You can also disable the optimizer by specifying the
336 <code>FP_NO_SUPPORT_OPTIMIZER</code> precompiler constant in your
340 <dt><p><code>FP_EPSILON</code> : (Default <code>1e-14</code>)</dt>
341 <dd><p>Epsilon value used in comparison operators.
342 If this line is commented out, then no epsilon will be used.
345 <dt><p><code>FP_USE_THREAD_SAFE_EVAL</code> : (Default off)</dt>
346 <dd><p>Define this precompiler constant to make <code>Eval()</code>
347 thread-safe. Refer to the <a href="#threadsafety">thread safety
348 section</a> later in this document for more information.
349 Note that defining this may make <code>Eval()</code> slightly slower.
350 <p>Also note that the MPFR and GMP versions of the library cannot be
351 made thread-safe, and thus this setting has no effect on them.
354 <dt><p><code>FP_USE_THREAD_SAFE_EVAL_WITH_ALLOCA</code> : (Default off)</dt>
355 <dd><p>This is like the previous, but makes <code>Eval()</code> use the
356 <code>alloca()</code> function (instead of <code>std::vector</code>).
357 This will make it faster, but the <code>alloca()</code>
358 function is not standard and thus not supported by all compilers.
361 <dt><p><code>FP_NO_EVALUATION_CHECKS</code> : (Default off)</dt>
362 <dd><p>If this precompiler constant is defined, no evaluation-time checks
363 will be performed. This may give a slight boost in speed in certain
364 situations. Consult the <a href="#evaluationchecks">evaluation
365 checks section</a> below for more information on this subject.
370 <!-- -------------------------------------------------------------------- -->
371 <a name="copyassignment"></a>
372 <h3>Copying and assignment</h3>
374 <p>The class implements a safe copy constructor and assignment operator.
376 <p>It uses the copy-on-write technique for efficiency. This means that
377 when copying or assigning a FunctionParser instance, the internal data
378 (which in some cases can be quite lengthy) is not immediately copied
379 but only when the contents of the copy (or the original) are changed.
381 <p>This means that copying/assigning is a very fast operation, and if
382 the copies are never modified then actual data copying never happens
385 <p>The <code>Eval()</code> and <code>EvalError()</code> methods of the
386 copy can be called without the internal data being copied.
388 <p>Calling <code>Parse()</code>, <code>Optimize()</code> or the user-defined
389 constant/function adding methods will cause a deep-copy.
392 <!-- -------------------------------------------------------------------- -->
393 <a name="shortdesc"></a>
394 <h3>Short descriptions of FunctionParser methods</h3>
397 int Parse(const std::string& Function, const std::string& Vars,
398 bool useDegrees = false);
400 int Parse(const char* Function, const std::string& Vars,
401 bool useDegrees = false);
404 <p>Parses the given function and compiles it to internal format.
405 Return value is -1 if successful, else the index value to the location
410 void setDelimiterChar(char);
413 <p>Sets an ending delimiter character for the function string. (See the
414 long description for more details.)
418 const char* ErrorMsg(void) const;
421 <p>Returns an error message corresponding to the error in
422 <code>Parse()</code>, or an empty string if no such error occurred.
426 ParseErrorType GetParseErrorType() const;
429 <p>Returns the type of parsing error which occurred. Possible return types
430 are described in the long description.
434 double Eval(const double* Vars);
437 <p>Evaluates the function given to <code>Parse()</code>.
441 int EvalError(void) const;
444 <p>Returns <code>0</code> if no error happened in the previous call to
445 <code>Eval()</code>, else an error code <code>>0</code>.
452 <p>Tries to optimize the bytecode for faster evaluation.
456 bool AddConstant(const std::string& name, double value);
459 <p>Add a constant to the parser. Returns <code>false</code> if the name of
460 the constant is invalid, else <code>true</code>.
464 bool AddUnit(const std::string& name, double value);
467 <p>Add a new unit to the parser. Returns <code>false</code> if the name of
468 the unit is invalid, else <code>true</code>.
472 bool AddFunction(const std::string& name,
473 double (*functionPtr)(const double*),
474 unsigned paramsAmount);
477 <p>Add a user-defined function to the parser (as a function pointer).
478 Returns <code>false</code> if the name of the function is invalid, else
483 bool AddFunction(const std::string& name, FunctionParser&);
486 <p>Add a user-defined function to the parser (as a <code>FunctionParser</code>
487 instance). Returns <code>false</code> if the name of the function is invalid,
488 else <code>true</code>.
492 bool RemoveIdentifier(const std::string& name);
495 <p>Removes the constant, unit or user-defined function with the specified
496 name from the parser.
500 int ParseAndDeduceVariables(const std::string& function,
501 int* amountOfVariablesFound = 0,
502 bool useDegrees = false);
503 int ParseAndDeduceVariables(const std::string& function,
504 std::string& resultVarString,
505 int* amountOfVariablesFound = 0,
506 bool useDegrees = false);
507 int ParseAndDeduceVariables(const std::string& function,
508 std::vector<std::string>& resultVars,
509 bool useDegrees = false);
512 <p>Like <code>Parse()</code>, but the variables in the function are deduced
513 automatically. The amount of found variables and the variable names themselves
514 are returned by the different versions of the function.
516 <!-- -------------------------------------------------------------------- -->
517 <a name="longdesc"></a>
518 <h3>Long descriptions of FunctionParser methods</h3>
521 <a name="longdesc_Parse"></a>
523 int Parse(const std::string& Function, const std::string& Vars,
524 bool useDegrees = false);
526 int Parse(const char* Function, const std::string& Vars,
527 bool useDegrees = false);
530 <p>Parses the given function (and compiles it to internal format).
531 Destroys previous function. Following calls to <code>Eval()</code> will evaluate
534 <p>The strings given as parameters are not needed anymore after parsing.
540 <td><code>Function</code></td>
541 <td>String containing the function to parse.</td>
543 <td><code>Vars</code></td>
544 <td>String containing the variable names, separated by commas.<br>
545 Eg. <code>"x,y"</code>, <code>"VarX,VarY,VarZ,n"</code> or
546 <code>"x1,x2,x3,x4,__VAR__"</code>.
548 <td><code>useDegrees</code></td>
549 <td>(Optional.) Whether to use degrees or radians in
550 trigonometric functions. (Default: radians)</td>
554 <p>If a <code>char*</code> is given as the <code>Function</code> parameter,
555 it must be a null-terminated string.
557 <p>Variables can have any size and they are case sensitive (ie.
558 <code>"var"</code>, <code>"VAR"</code> and <code>"Var"</code> are
559 <em>different</em> variable names). Letters, digits, underscores and
560 UTF8-encoded characters can be used in variable names, but the name of
561 a variable can't begin with a digit. Each variable name can appear only
562 once in the '<code>Vars</code>' string. Function names are not legal
565 <p>Using longer variable names causes no overhead whatsoever to the
566 <code>Eval()</code> method, so it's completely safe to use variable names
569 <p>The third, optional parameter specifies whether angles should be
570 interpreted as radians or degrees in trigonometrical functions.
571 If not specified, the default value is radians.
576 <li>On success the function returns <code>-1</code>.
577 <li>On error the function returns an index to where the error was found
578 (<code>0</code> is the first character, <code>1</code> the second, etc).
579 If the error was not a parsing error returns an index to the end of the
583 <p>Example: <code>parser.Parse("3*x+y", "x,y");</code>
587 <a name="longdesc_setDelimiterChar"></a>
589 void setDelimiterChar(char);
592 <p>By default the parser expects the entire function string to be valid
593 (ie. the entire contents of the given <code>std::string</code>, or a C string
594 ending in the null character <code>'\0'</code>).
596 <p>If a delimiter character is specified with this function, then if it's
597 encountered at the outermost parsing level by the <code>Parse()</code>
598 function, and the input function has been valid so far, <code>Parse()</code>
599 will return an index to this character inside the input string, but rather
600 than set an error code, <code>FP_NO_ERROR</code> will be set.
602 <p>The idea is that this can be used to more easily parse functions which
603 are embedded inside larger strings, containing surrounding data, without
604 having to explicitly extract the function to a separate string.
606 <p>For example, suppose you are writing an interpreter for a scripting
607 language, which can have commands like this:
609 <p><code>let MyFunction(x,y) = { sin(x*x+y*y) } // A 2-dimensional function</code>
611 <p>Normally when parsing such a line you would have to extract the part
612 inside the curly brackets into a separate string and parse it that way.
613 With this feature what you can do instead is to set <code>'}'</code> as
614 the delimiter character and then simply give a pointer to the character
615 which comes after the <code>'{'</code>. If all goes well, the
616 <code>Parse()</code> function will return an index to the <code>'}'</code>
617 character (from the given starting point) and <code>GetParseErrorType()</code>
618 will return <code>FP_NO_ERROR</code>. You can use the return
619 value (if it's not <code>-1</code>) to jump forward in the string to the
622 <p>Note that a null character (<code>'\0'</code>) or the end of the
623 <code>std::string</code> (if one was given) will still be a valid end of
624 the function string even if a delimiter character was specified. (In this
625 case <code>Parse()</code> will return <code>-1</code> if there was no error,
628 <p>Also note that the delimiter character cannot be any valid operator
629 or alphanumeric (including the underscore) character, nor the other
630 characters defined in the function syntax. It must be a character not
631 supported by the function parser (such as <code>'}'</code>,
632 <code>'"'</code>, <code>']'</code>, etc).
636 <a name="longdesc_ErrorMsg"></a>
638 const char* ErrorMsg(void) const;
641 <p>Returns a pointer to an error message string corresponding to the error
642 caused by <code>Parse()</code> (you can use this to print the proper error
643 message to the user). If no such error has occurred, returns an empty string.
647 <a name="longdesc_GetParseErrorType"></a>
649 ParseErrorType GetParseErrorType() const;
652 <p>Returns the type of parse error which occurred.
654 <p>This method can be used to get the error type if <code>ErrorMsg()</code>
655 is not enough for printing the error message. In other words, this can be
656 used for printing customized error messages (eg. in another language).
657 If the default error messages suffice, then this method doesn't need
660 <code>FunctionParser::ParseErrorType</code> is an enumerated type inside
661 the class (ie. its values are accessed like
662 "<code>FunctionParser::SYNTAX_ERROR</code>").
664 <p>The possible values for FunctionParser::ParseErrorType are listed below,
665 along with their equivalent error message returned by the
666 <code>ErrorMsg()</code> method:
670 <td><code>FP_NO_ERROR</code></td>
671 <td>If no error occurred in the previous call to <code>Parse()</code>.</td>
673 <td><code>SYNTAX_ERROR</code></td>
674 <td>"Syntax error"</td>
676 <td><code>MISM_PARENTH</code></td>
677 <td>"Mismatched parenthesis"</td>
679 <td><code>MISSING_PARENTH</code></td>
680 <td>"Missing ')'"</td>
682 <td><code>EMPTY_PARENTH</code></td>
683 <td>"Empty parentheses"</td>
685 <td><code>EXPECT_OPERATOR</code></td>
686 <td>"Syntax error: Operator expected"</td>
688 <td><code>OUT_OF_MEMORY</code></td>
689 <td>"Not enough memory"</td>
691 <td><code>UNEXPECTED_ERROR</code></td>
692 <td>"An unexpected error occurred. Please make a full bug report to the
695 <td><code>INVALID_VARS</code></td>
696 <td>"Syntax error in parameter 'Vars' given to FunctionParser::Parse()"</td>
698 <td><code>ILL_PARAMS_AMOUNT</code></td>
699 <td>"Illegal number of parameters to function"</td>
701 <td><code>PREMATURE_EOS</code></td>
702 <td>"Syntax error: Premature end of string"</td>
704 <td><code>EXPECT_PARENTH_FUNC</code></td>
705 <td>"Syntax error: Expecting ( after function"</td>
707 <td><code>UNKNOWN_IDENTIFIER</code></td>
708 <td>"Syntax error: Unknown identifier"</td>
710 <td><code>NO_FUNCTION_PARSED_YET</code></td>
711 <td>"(No function has been parsed yet)"</td>
717 <a name="longdesc_Eval"></a>
719 double Eval(const double* Vars);
722 <p>Evaluates the function given to <code>Parse()</code>.
723 The array given as parameter must contain the same amount of values as
724 the amount of variables given to <code>Parse()</code>. Each value corresponds
725 to each variable, in the same order.
729 <li>On success returns the evaluated value of the function given to
730 <code>Parse()</code>.
731 <li>On error (such as division by 0) the return value is unspecified,
737 <p><code>double Vars[] = {1, -2.5};</code><br>
738 <code>double result = parser.Eval(Vars);</code>
742 <a name="longdesc_EvalError"></a>
744 int EvalError(void) const;
747 <p>Used to test if the call to <code>Eval()</code> succeeded.
751 <p>If there was no error in the previous call to <code>Eval()</code>,
752 returns <code>0</code>, else returns a positive value as follows:
754 <li>1: division by zero
755 <li>2: sqrt error (sqrt of a negative value)
756 <li>3: log error (logarithm of a negative value)
757 <li>4: trigonometric error (asin or acos of illegal value)
758 <li>5: maximum recursion level in <code>eval()</code> reached
763 <a name="longdesc_Optimize"></a>
768 <p>This method can be called after calling the <code>Parse()</code> method.
769 It will try to simplify the internal bytecode so that it will evaluate faster
770 (it tries to reduce the amount of opcodes in the bytecode).
772 <p>For example, the bytecode for the function <code>"5+x*y-25*4/8"</code> will
773 be reduced to a bytecode equivalent to the function <code>"x*y-7.5"</code> (the
774 original 11 opcodes will be reduced to 5). Besides calculating constant
775 expressions (like in the example), it also performs other types of
776 simplifications with variable and function expressions.
778 <p>This method is quite slow and the decision of whether to use it or
779 not should depend on the type of application. If a function is parsed
780 once and evaluated millions of times, then calling <code>Optimize()</code>
781 may speed-up noticeably. However, if there are tons of functions to parse
782 and each one is evaluated once or just a few times, then calling
783 <code>Optimize()</code> will only slow down the program.
785 <p>Also, if the original function is expected to be optimal, then calling
786 <code>Optimize()</code> would be useless.
788 <p>Note: Currently this method does not make any checks (like
789 <code>Eval()</code> does) and thus things like <code>"1/0"</code> will cause
790 undefined behaviour. (On the other hand, if such expression is given to the
791 parser, <code>Eval()</code> will always give an error code, no matter what
792 the parameters.) If caching this type of errors is important, a work-around
793 is to call <code>Eval()</code> once before calling <code>Optimize()</code>
794 and checking <code>EvalError()</code>.
796 <p>If the destination application is not going to use this method,
797 the compiler constant <code>FP_SUPPORT_OPTIMIZER</code> can be undefined in
798 <code>fpconfig.hh</code> to make the library smaller (<code>Optimize()</code>
799 can still be called, but it will not do anything).
801 <p>(If you are interested in seeing how this method optimizes the opcode,
802 you can call the <code>PrintByteCode()</code> method before and after the
803 call to <code>Optimize()</code> to see the difference.)
807 <a name="longdesc_AddConstant"></a>
809 bool AddConstant(const std::string& name, double value);
812 <p>This method can be used to add constants to the parser. Syntactically
813 constants are identical to variables (ie. they follow the same naming
814 rules and they can be used in the function string in the same way as
815 variables), but internally constants are directly replaced with their
818 <p>Constants used by a function must be added before calling
819 <code>Parse()</code> for that function. Constants are preserved between
820 <code>Parse()</code> calls in the current FunctionParser instance, so
821 they don't need to be added but once. (If you use the same constant in
822 several instances of FunctionParser, you will need to add it to all the
823 instances separately.)
825 <p>Constants can be added at any time and the value of old constants can
826 be changed, but new additions and changes will only have effect the next
827 time <code>Parse()</code> is called. (That is, changing the value of a constant
828 after calling <code>Parse()</code> and before calling <code>Eval()</code>
829 will have no effect.)
831 <p>The return value will be <code>false</code> if the '<code>name</code>' of
832 the constant was illegal, else <code>true</code>. If the name was illegal,
833 the method does nothing.
835 <p>Example: <code>parser.AddConstant("pi", 3.1415926535897932);</code>
837 <p>Now for example <code>parser.Parse("x*pi", "x");</code> will be identical
838 to the call <code>parser.Parse("x*3.1415926535897932", "x");</code>
842 <a name="longdesc_AddUnit"></a>
844 bool AddUnit(const std::string& name, double value);
847 <p>In some applications it is desirable to have units of measurement.
848 A typical example is an application which creates a page layout to be
849 printed. When printing, distances are usually measured in points
850 (defined by the resolution of the printer). However, it is often more
851 useful for the user to be able to specify measurements in other units
852 such as centimeters or inches.
854 <p>A unit is simply a value by which the preceding element is multiplied.
855 For example, if the printing has been set up to 300 DPI, one inch is
856 then 300 points (dots). Thus saying eg. <code>"5in"</code> is the same as saying
857 <code>"5*300"</code> or <code>"1500"</code> (assuming <code>"in"</code> has
858 been added as a unit with the value 300).
860 <p>Note that units are slightly different from a multiplication in
861 that they have a higher precedence than any other operator (except
862 parentheses). Thus for example <code>"5/2in"</code> is parsed as
863 <code>"5/(2*300)"</code>.
864 (If 5/2 inches is what one wants, it has to be written <code>"(5/2)in"</code>.)
866 <p>You can use the <code>AddUnit()</code> method to add a new unit. The
867 unit can then be used after any element in the function (and will work as
868 a multiplier for that element). An element is a float literal, a constant,
869 a variable, a function or any expression in parentheses. When the element
870 is not a float literal nor an expression in parentheses, there has to naturally
871 be at least one whitespace between the element and the unit (eg.
872 <code>"x in"</code>). To change the value of a unit, call
873 <code>AddUnit()</code> again with the same unit name and the new value.
875 <p>Unit names share the same namespace as constants, functions and
876 variables, and thus should be distinct from those.
878 <p>Example: <code>parser.AddUnit("in", 300);</code>
880 <p>Now for example the function <code>"5in"</code> will be identical to
881 <code>"(5*300)"</code>. Other usage examples include <code>"x in"</code>,
882 <code>"3in+2"</code>, <code>"pow(x,2)in"</code>, <code>"(x+2)in"</code>.
886 <a name="longdesc_AddFunction1"></a>
888 bool AddFunction(const std::string& name,
889 double (*functionPtr)(const double*),
890 unsigned paramsAmount);
893 This method can be used to add new functions to the parser. For example,
894 if you would like to add a function "<code>sqr(A)</code>" which squares the
895 value of <code>A</code>, you can do it with this method (so that you don't
896 need to touch the source code of the parser).
898 <p>The method takes three parameters:
901 <li>The name of the function. The name follows the same naming conventions
904 <li>A C++ function, which will be called when evaluating the function
905 string (if the user-given function is called there). The C++ function
907 <p><code>double functionName(const double* params);</code>
909 <li>The number of parameters the function takes. 0 is a valid value
910 in which case the function takes no parameters (such function
911 should simply ignore the <code>double*</code> it gets as a parameter).
914 <p>The return value will be <code>false</code> if the given name was invalid
915 (either it did not follow the variable naming conventions, or the name was
916 already reserved), else <code>true</code>. If the return value is
917 <code>false</code>, nothing is added.
919 <p>Example: Suppose we have a C++ function like this:
921 <p><code>double Square(const double* p)</code><br>
923 <code> return p[0]*p[0];</code><br>
926 <p>Now we can add this function to the parser like this:
928 <p><code>parser.AddFunction("sqr", Square, 1);</code><br>
929 <code>parser.Parse("2*sqr(x)", "x");</code>
931 <p>An example of a useful function taking no parameters is a function
932 returning a random value. For example:
934 <p><code>double Rand(const double*)</code><br>
936 <code> return drand48();</code><br
939 <p><code>parser.AddFunction("rand", Rand, 0);</code>
941 <p><em>Important note</em>: If you use the <code>Optimize()</code> method,
942 it will assume that the user-given function has no side-effects, that is,
943 it always returns the same value for the same parameters. The optimizer will
944 optimize the function call away in some cases, making this assumption.
945 (The <code>Rand()</code> function given as example above is one such
950 <a name="longdesc_AddFunction2"></a>
952 bool AddFunction(const std::string& name, FunctionParser&);
955 <p>This method is almost identical to the previous <code>AddFunction()</code>,
956 but instead of taking a C++ function, it takes another FunctionParser
959 <p>There are some important restrictions on making a FunctionParser instance
963 <li>The FunctionParser instance given as parameter must be initialized
964 with a <code>Parse()</code> call before giving it as parameter. That
965 is, if you want to use the parser <code>A</code> in the parser
966 <code>B</code>, you must call <code>A.Parse()</code> before you can
967 call <code>B.AddFunction("name", A)</code>.
969 <li>The amount of variables in the FunctionParser instance given as
970 parameter must not change after it has been given to the
971 <code>AddFunction()</code>
972 of another instance. Changing the number of variables will result in
975 <li><code>AddFunction()</code> will fail (ie. return <code>false</code>)
976 if a recursive loop is
977 formed. The method specifically checks that no such loop is built.
979 <li>The FunctionParser instance given as parameter will <em>not</em> be
980 copied internally, only referenced. Thus the FunctionParser instance
981 given as parameter must exist for as long as the other FunctionParser
987 <p><code>FunctionParser f1, f2;</code><br>
988 <p><code>f1.Parse("x*x", "x");</code><br>
989 <p><code>f2.AddFunction("sqr", f1);</code>
991 <p>This version of the <code>AddFunction()</code> method can be useful to
992 eg. chain user-given functions. For example, ask the user for a function F1,
993 and then ask the user another function F2, but now the user can
994 call F1 in this second function if he wants (and so on with a third
995 function F3, where he can call F1 and F2, etc).
998 <a name="longdesc_RemoveIdentifier"></a>
1000 bool RemoveIdentifier(const std::string& name);
1003 <p>If a constant, unit or user-defined function with the specified name
1004 exists in the parser, it will be removed and the return value will be
1005 <code>true</code>, else nothing will be done and the return value will be
1008 <p>(Note: If you want to remove <em>everything</em> from an existing
1009 FunctionParser instance, simply assign a fresh instance to it, ie. like
1010 "<code>parser = FunctionParser();</code>")
1013 <a name="longdesc_ParseAndDeduceVariables"></a>
1015 int ParseAndDeduceVariables(const std::string& function,
1016 int* amountOfVariablesFound = 0,
1017 bool useDegrees = false);
1018 int ParseAndDeduceVariables(const std::string& function,
1019 std::string& resultVarString,
1020 int* amountOfVariablesFound = 0,
1021 bool useDegrees = false);
1022 int ParseAndDeduceVariables(const std::string& function,
1023 std::vector<std::string>& resultVars,
1024 bool useDegrees = false);
1027 <p>These functions work in the same way as the <code>Parse()</code> function,
1028 but the variables in the input function string are deduced automatically. The
1032 <li><code>function</code>: The input function string, as with
1033 <code>Parse()</code>.
1034 <li><code>amountOfVariablesFound</code>: If non-null, the amount of found
1035 variables will be assigned here.
1036 <li><code>resultVarString</code>: The found variables will be written to
1037 this string, in the same format as accepted by the <code>Parse()</code>
1038 function. The variable names will be sorted using the <code><</code>
1039 operator of <code>std::string</code>.
1040 <li><code>resultVars</code>: The found variables will be written to this
1041 vector, each element being one variable name. They will be sorted using
1042 the <code><</code> operator of <code>std::string</code>. (The amount
1043 of found variables can be retrieved, rather obviously, with the
1044 <code>size()</code> method of the vector.)
1045 <li><code>useDegrees</code>: As with <code>Parse()</code>.
1048 <p>As with <code>Parse()</code>, the return value will be <code>-1</code> if
1049 the parsing succeeded, else an index to the location of the error. None of
1050 the specified return values will be modified in case of error.
1053 <!-- -------------------------------------------------------------------- -->
1056 <a name="literals"></a>
1057 <h3>Numeric literals</h3>
1059 <p>A numeric literal is a fixed numerical value in the input function string
1060 (either a floating point value or an integer value, depending on the parser
1063 <p>An integer literal can consist solely of numerical digits (possibly with
1064 a preceding unary minus). For example, "<code>12345</code>".
1066 <p>If the literal is preceded by the characters "<code>0x</code>", it
1067 will be interpreted as a hexadecimal literal, where digits can also include
1068 the letters from '<code>A</code>' to '<code>F</code>' (in either uppercase
1069 or lowercase). For example, "<code>0x89ABC</code>" (which corresponds to the
1072 <p>A floating point literal (only supported by the floating point type parsers)
1073 may additionally include a decimal point followed by the decimal part of the
1074 value, such as for example "<code>12.34</code>", optionally followed by a
1077 <p>A decimal exponent consists of an '<code>E</code>' or '<code>e</code>',
1078 followed by an optional plus or minus sign, followed by decimal digits, and
1079 indicates multiplication by a power of 10. For example, "<code>1.2e5</code>"
1080 (which is equivalent to the value 120000).
1082 <p>If a floating point literal is preceded by the characters "<code>0x</code>"
1083 it will be interpreted in hexadecimal. A hexadecimal floating point
1084 literal consists of a hexadecimal value, with an optional decimal point,
1085 followed optionally by a binary exponent in base 10 (in other words, the
1086 exponent is not in hexadecimal).
1088 <p>A binary exponent has the same format as a decimal exponent, except that
1089 '<code>P</code>' or '<code>p</code>' is used. A binary exponent indicates
1090 multiplication by a power of 2. For example, "<code>0xA.Bp10</code>"
1091 (which is equivalent to the value 10944).
1093 <a name="identifiers"></a>
1094 <h3>Identifier names</h3>
1096 <p>An identifier is the name of a function (internal or user-defined),
1097 variable, constant or unit. New identifiers can be specified with the
1098 functions described in the earlier subsections in this document.
1100 <p>The name of an identifier can use any alphanumeric characters, the
1101 underscore character and any UTF8-encoded unicode character, excluding
1102 those denoting whitespace.
1103 The first character of the name cannot be a numeric digit, though.
1105 <p>All functions, variables, constants and units must use unique names.
1106 It's not possible to add two different identifiers with the same name.
1109 <!-- -------------------------------------------------------------------- -->
1110 <a name="functionsyntax"></a>
1111 <h3>The function string syntax</h3>
1113 <p>The function string understood by the class is very similar (but not
1114 completely identical in all aspects) to mathematical expressions in the
1116 Arithmetic float expressions can be created from float literals, variables
1117 or functions using the following operators in this order of precedence:
1121 <td><code>()</code></td>
1122 <td>expressions in parentheses first</td>
1124 <td><code>A unit</code></td>
1125 <td>a unit multiplier (if one has been added)</td>
1127 <td><code>A^B</code></td>
1128 <td>exponentiation (A raised to the power B)</td>
1130 <td><code>-A</code></td>
1131 <td>unary minus</td>
1133 <td><code>!A</code></td>
1134 <td>unary logical not (result is 1 if <code>int(A)</code> is 0, else 0)</td>
1136 <td><code>A*B A/B A%B</code></td>
1137 <td>multiplication, division and modulo</td>
1139 <td><code>A+B A-B</code></td>
1140 <td>addition and subtraction</td>
1142 <td><code>A=B A<B A<=B<br>A!=B A>B A>=B</code></td>
1143 <td>comparison between A and B (result is either 0 or 1)</td>
1145 <td><code>A&B</code></td>
1146 <td>result is 1 if <code>int(A)</code> and <code>int(B)</code> differ from
1148 Note: Regardless of the values, both operands are always
1149 evaluated. However, if the expression is optimized, it may
1150 be changed such that only one of the operands is evaluated,
1151 according to standard shortcut logical operation semantics.</td>
1153 <td><code>A|B</code></td>
1154 <td>result is 1 if <code>int(A)</code> or <code>int(B)</code> differ from 0,
1156 Note: Regardless of the values, both operands are always
1157 evaluated. However, if the expression is optimized, it may
1158 be changed such that only one of the operands is evaluated,
1159 according to standard shortcut logical operation semantics.</td>
1163 <p>(Note that currently the exponentiation operator is not supported for
1164 <code>FunctionParser_li</code> nor <code>FunctionParser_gmpint</code>.
1165 With the former the result would very easily overflow, making its
1166 usefulness questionable. With the latter it could be easily abused to
1167 make the program run out of memory; think of a function like
1170 <p>Since the unary minus has higher precedence than any other operator, for
1171 example the following expression is valid: <code>x*-y</code>
1173 <p>The comparison operators use an epsilon value, so expressions which may
1174 differ in very least-significant digits should work correctly. For example,
1175 <code>"0.1+0.1+0.1+0.1+0.1+0.1+0.1+0.1+0.1+0.1 = 1"</code> should always
1176 return 1, and the same comparison done with "<code>></code>" or
1177 "<code><</code>" should always return 0. (The epsilon value can be
1178 configured in the <code>fpconfig.hh</code> file.)
1179 Without epsilon this comparison probably returns the wrong value.
1181 <p>The class supports these functions:
1185 <td><code>abs(A)</code></td>
1186 <td>Absolute value of A. If A is negative, returns -A otherwise returns A.</td>
1188 <td><code>acos(A)</code></td>
1189 <td>Arc-cosine of A. Returns the angle, measured in radians, whose cosine
1192 <td><code>acosh(A)</code></td>
1193 <td>Same as acos() but for hyperbolic cosine.</td>
1195 <td><code>asin(A)</code></td>
1196 <td>Arc-sine of A. Returns the angle, measured in radians, whose sine
1199 <td><code>asinh(A)</code></td>
1200 <td>Same as asin() but for hyperbolic sine.</td>
1202 <td><code>atan(A)</code></td>
1203 <td>Arc-tangent of (A). Returns the angle, measured in radians,
1204 whose tangent is (A).</td>
1206 <td><code>atan2(A,B)</code></td>
1207 <td>Arc-tangent of A/B. The two main differences to atan() is
1208 that it will return the right angle depending on the signs of
1209 A and B (atan() can only return values betwen -pi/2 and pi/2),
1210 and that the return value of pi/2 and -pi/2 are possible.</td>
1212 <td><code>atanh(A)</code></td>
1213 <td>Same as atan() but for hyperbolic tangent.</td>
1215 <td><code>cbrt(A)</code></td>
1216 <td>Cube root of A. Returns the value whose cube is A.</td>
1218 <td><code>ceil(A)</code></td>
1219 <td>Ceiling of A. Returns the smallest integer not smaller than A.
1220 Rounds up to the next higher integer. E.g. -2.9, -2.5 and -2.1 are
1221 rounded to -2.0, and 2.9, 2.5 and 2.1 are rounded to 3.0.</td>
1223 <td><code>cos(A)</code></td>
1224 <td>Cosine of A. Returns the cosine of the angle A, where A is
1225 measured in radians.</td>
1227 <td><code>cosh(A)</code></td>
1228 <td>Same as cos() but for hyperbolic cosine.</td>
1230 <td><code>cot(A)</code></td>
1231 <td>Cotangent of A (equivalent to 1/tan(A)).</td>
1233 <td><code>csc(A)</code></td>
1234 <td>Cosecant of A (equivalent to 1/sin(A)).</td>
1236 <td><code>eval(...)</code></td>
1237 <td>This a recursive call to the function to be evaluated. The
1238 number of parameters must be the same as the number of parameters
1239 taken by the function. Must be called inside <code>if()</code> to avoid
1240 infinite recursion.</td>
1242 <td><code>exp(A)</code></td>
1243 <td>Exponential of A. Returns the value of e raised to the power
1244 A where e is the base of the natural logarithm, i.e. the
1245 non-repeating value approximately equal to 2.71828182846.</td>
1247 <td><code>floor(A)</code></td>
1248 <td>Floor of A. Returns the largest integer not greater than A. Rounds
1249 down to the next lower integer.
1250 E.g. -2.9, -2.5 and -2.1 are rounded to -3.0,
1251 and 2.9, 2.5 and 2.1 are rounded to 2.0.</td>
1253 <td><code>if(A,B,C)</code></td>
1254 <td>If int(A) differs from 0, the return value of this function is B,
1255 else C. Only the parameter which needs to be evaluated is
1256 evaluated, the other parameter is skipped; this makes it safe to
1257 use <code>eval()</code> in them.</td>
1259 <td><code>int(A)</code></td>
1260 <td>Rounds A to the closest integer. Equidistant values are rounded up.
1261 E.g. -2.9 is rounded to -3.0; -2.5 and -2.1 are rounded to -2.0,
1262 and 2.9 and 2.5 are rounded to 3.0; 2.1 is rounded to 2.0.</td>
1264 <td><code>log(A)</code></td>
1265 <td>Natural (base e) logarithm of A.</td>
1267 <td><code>log10(A)</code></td>
1268 <td>Base 10 logarithm of A.</td>
1270 <td><code>max(A,B)</code></td>
1271 <td>If A>B, the result is A, else B.</td>
1273 <td><code>min(A,B)</code></td>
1274 <td>If A<B, the result is A, else B.</td>
1276 <td><code>pow(A,B)</code></td>
1277 <td>Exponentiation (A raised to the power B).</td>
1279 <td><code>sec(A)</code></td>
1280 <td>Secant of A (equivalent to 1/cos(A)).</td>
1282 <td><code>sin(A)</code></td>
1283 <td>Sine of A. Returns the sine of the angle A, where A is
1284 measured in radians.</td>
1286 <td><code>sinh(A)</code></td>
1287 <td>Same as sin() but for hyperbolic sine.</td>
1289 <td><code>sqrt(A)</code></td>
1290 <td>Square root of A. Returns the value whose square is A.</td>
1292 <td><code>tan(A)</code></td>
1293 <td>Tangent of A. Returns the tangent of the angle A, where A
1294 is measured in radians.</td>
1296 <td><code>tanh(A)</code></td>
1297 <td>Same as tan() but for hyperbolic tangent.</td>
1299 <td><code>trunc(A)</code></td>
1300 <td>Truncated value of A. Returns an integer corresponding to the value
1301 of A without its fractional part.
1302 E.g. -2.9, -2.5 and -2.1 are rounded to -2.0,
1303 and 2.9, 2.5 and 2.1 are rounded to 2.0.</td>
1307 <p>(Note that for <code>FunctionParser_li</code> and
1308 <code>FunctionParser_gmpint</code> only the functions
1309 <code>abs()</code>, <code>eval()</code>, <code>if()</code>,
1310 <code>min()</code> and <code>max()</code> are supported.)
1312 <p>Examples of function string understood by the class:
1314 <p><code>"1+2"</code><br>
1315 <code>"x-1"</code><br>
1316 <code>"-sin(sqrt(x^2+y^2))"</code><br>
1317 <code>"sqrt(XCoord*XCoord + YCoord*YCoord)"</code><br>
1319 <p>An example of a recursive function is the factorial function:
1321 <code>"if(n>1, n*eval(n-1), 1)"</code>
1323 <p>Note that a recursive call has some overhead, which makes it a bit slower
1324 than any other operation. It may be a good idea to avoid recursive functions
1325 in very time-critical applications. Recursion also takes some memory, so
1326 extremely deep recursions should be avoided (eg. millions of nested recursive
1329 <p>Also note that even though the maximum recursion level of
1330 <code>eval()</code> is limited, it is possible to write functions which
1331 never reach that level but still take enormous amounts of time to evaluate.
1332 This can sometimes be undesirable because it is prone to exploitation,
1333 which is why <code>eval()</code> is disabled by default. It can be enabled
1334 in the <code>fpconfig.hh</code> file.
1337 <!-- -------------------------------------------------------------------- -->
1338 <a name="inlinevars"></a>
1339 <h3>Inline variables</h3>
1341 <p>The function syntax supports defining new variables inside the function
1342 string itself. This can be done with the following syntax:
1344 <p><code>"<variable name> := <expression>; <function>"</code>
1348 <p><code>"length := sqrt(x*x+y*y); 2*length*sin(length)"</code>
1350 <p>(Spaces around the '<code>:=</code>' operator are optional.)
1352 <p>The obvious benefit of this is that if a long expression needs to be
1353 used in the function several times, this allows writing it only once and
1354 using a named variable from that point forward.
1356 <p>The variable name must be an unused identifier (in other words, not an
1357 existing function, variable or unit name).
1359 <p>The <code><function></code> part can have further inline variable
1360 definitions, and thus it's possible to have any amount of them, for example:
1362 <p><code>"A := x^2; B := y^2; C := z^2; sqrt(A+B+C)"</code>
1364 <p>The expressions in subsequent inline variable definitions can use any
1365 of the previous inline variables. It is also possible to redefine an inline
1366 variable. For example:
1368 <p><code>"A := x^2; A := 2*A; sqrt(A)"</code>
1371 <!-- -------------------------------------------------------------------- -->
1372 <a name="whitespace"></a>
1375 <p>Arbitrary amounts of whitespace can optionally be included between
1376 elements in the function string.
1377 The following unicode characters are interpreted as whitespace:
1380 <th>Character number</th>
1381 <th>Character name</th>
1382 <th>UTF-8 byte sequence</th>
1384 <tr><td>U+0009</td><td>HORIZONTAL TABULATION </td><td>09</td></tr>
1385 <tr><td>U+000A</td><td>LINE FEED </td><td>0A</td></tr>
1386 <tr><td>U+000B</td><td>VERTICAL TABULATION </td><td>0B</td></tr>
1387 <tr><td>U+000D</td><td>CARRIAGE RETURN </td><td>0D</td></tr>
1388 <tr><td>U+0020</td><td>SPACE </td><td>20</td></tr>
1389 <tr><td>U+00A0</td><td>NO-BREAK SPACE </td><td>C2 A0</td></tr>
1390 <tr><td>U+2000</td><td>EN QUAD </td><td>E2 80 80</td></tr>
1391 <tr><td>U+2001</td><td>EM QUAD </td><td>E2 80 81</td></tr>
1392 <tr><td>U+2002</td><td>EN SPACE </td><td>E2 80 82</td></tr>
1393 <tr><td>U+2003</td><td>EM SPACE </td><td>E2 80 83</td></tr>
1394 <tr><td>U+2004</td><td>THREE-PER-EM SPACE </td><td>E2 80 84</td></tr>
1395 <tr><td>U+2005</td><td>FOUR-PER-EM SPACE </td><td>E2 80 85</td></tr>
1396 <tr><td>U+2006</td><td>SIX-PER-EM SPACE </td><td>E2 80 86</td></tr>
1397 <tr><td>U+2007</td><td>FIGURE SPACE </td><td>E2 80 87</td></tr>
1398 <tr><td>U+2008</td><td>PUNCTUATION SPACE </td><td>E2 80 88</td></tr>
1399 <tr><td>U+2009</td><td>THIN SPACE </td><td>E2 80 89</td></tr>
1400 <tr><td>U+200A</td><td>HAIR SPACE </td><td>E2 80 8A</td></tr>
1401 <tr><td>U+200B</td><td>ZERO WIDTH SPACE </td><td>E2 80 8B</td></tr>
1402 <tr><td>U+202F</td><td>NARROW NO-BREAK SPACE </td><td>E2 80 AF</td></tr>
1403 <tr><td>U+205F</td><td>MEDIUM MATHEMATICAL SPACE</td><td>E2 81 9F</td></tr>
1404 <tr><td>U+3000</td><td>IDEOGRAPHIC SPACE </td><td>E3 80 80</td></tr>
1408 <!-- -------------------------------------------------------------------- -->
1409 <h2>Miscellaneous</h2>
1411 <a name="evaluationchecks"></a>
1412 <h3>About evaluation-time checks</h3>
1414 <p>By default <code>FunctionParser::Eval()</code> will perform certain sanity
1415 checks before performing certain operations. For example, before calling the
1416 <code>sqrt</code> function, it will check if the parameter is negative, and
1417 if so, it will set the proper error code instead of calling the function.
1418 These checks include:
1421 <li>Division by (the exact value of) zero.
1422 <li>Square root of a negative value.
1423 <li>Logarithm of a non-positive value.
1424 <li>Arcsine or arccosine of a value not in the range [-1, 1]. (This includes
1425 hyperbolic versions of the functions.)
1428 <p>However, the library <em>can not</em> guarantee that it will catch all
1429 possible floating point errors before performing them, because this is
1430 impossible to do with standard C++. For example, dividing a very large
1431 value by a value which is very close to zero, or calculating the logarithm
1432 of a very small value may overflow the result, as well as multiplying two
1433 very large values. Raising a negative number to a non-integral power may
1434 cause a <em>NaN</em> result, etc.
1436 <p>As a rule of thumb, the library will (by default) detect invalid operations
1437 if they are invalid for a range of values. For example, square root is undefined
1438 for all negative values, and arc sine is undefined only values outside the range
1439 [-1, 1]. In general, operations which are invalid for only one single value
1440 (rather than a contiguous range of values) will not be detected (division by
1441 the exact value of zero is an exception to this rule) nor will
1442 overflow/underflow situations.
1444 <p>The library cannot guarantee that floating point
1445 errors will never happen during evaluation. This can make the library to
1446 return the floating point values <em>inf</em> and <em>NaN</em>. Moreover,
1447 if floating point errors cause an interrupt in the target computer
1448 architecture and/or when using certain compiler settings, this library
1449 cannot guarantee that it will never happen.
1451 <p>Since not all error situations can be caught, and since the sanity checks
1452 only slow down the evaluation (although only very slightly), the precompiler
1453 constant <code>FP_NO_EVALUATION_CHECKS</code> can be used to turn all the
1454 checks off. This might make the evaluation slightly faster in certain
1457 <p>Note that the optimizer never performs any sanity checks.
1460 <!-- -------------------------------------------------------------------- -->
1461 <a name="threadsafety"></a>
1462 <h3>About thread safety</h3>
1464 <p>None of the member functions of the FunctionParser class are thread-safe.
1465 Most prominently, the <code>Eval()</code> function is not thread-safe.
1466 (In other words, the <code>Eval()</code> function of a single FunctionParser
1467 instance cannot be safely called simultaneously by two threads.)
1469 <p>There are ways to use this library in a thread-safe way, though. If each
1470 thread uses its own FunctionParser instance, no problems will obviously
1471 happen. Note, however, that if these instances need to be a copy of a given
1472 FunctionParser instance (eg. one where the user has entered a function),
1473 a deep copy of this instance has to be performed for each thread. By
1474 default FunctionParser uses shallow-copying (copy-on-write), which means
1475 that a simple assignment of copy construction will not copy the data itself.
1476 To force a deep copy you can all the <code>ForceDeepCopy()</code> function on
1477 each of the instances of each thread after the assignment or copying has been
1480 <p>Another possibility is to compile the FunctionParser library so that
1481 its <code>Eval()</code> function will be thread-safe. (This can be done by
1482 defining the <code>FP_USE_THREAD_SAFE_EVAL</code> or the
1483 <code>FP_USE_THREAD_SAFE_EVAL_WITH_ALLOCA</code>
1484 precompiler constant.) As long as only one thread calls the other functions
1485 of FunctionParser, the other threads can safely call the <code>Eval()</code>
1486 of this one instance.
1488 <p>Note, however, that compiling the library like this can make
1489 <code>Eval()</code> slightly slower. (The <code>alloca</code> version, if
1490 supported by the compiler, will not be as slow.)
1492 <p>Also note that the MPFR and GMP versions of the library cannot be
1493 made thread-safe, and thus this setting has no effect on them.
1496 <!-- -------------------------------------------------------------------- -->
1497 <a name="tipsandtricks"></a>
1498 <h3>Tips and tricks</h3>
1500 <h4>Add constants automatically to all parser objects</h4>
1502 <p>Often the same constants (such as <em>pi</em> and <em>e</em>) and other
1503 user-defined identifiers (such as units) are always used in all the
1504 <code>FunctionParser</code> objects throughout the program. It would be
1505 troublesome to always have to manually add these constants every time a
1506 new parser object is created.
1508 <p>There is, however, a simple way to always add these user-defined identifiers
1509 to all instances. Write a class like this:
1512 class ParserWithConsts: public FunctionParser
1517 AddConstant("pi", 3.14159265358979323846);
1518 AddConstant("e", 2.71828182845904523536);
1523 <p>Now instead of using <code>FunctionParser</code>, always use
1524 <code>ParserWithConsts</code>. It will behave identically except that the
1525 constants (and possibly other user-defined identifiers) will always be
1526 automatically defined. (Objects of this type even survive
1527 <a href="http://en.wikipedia.org/wiki/Object_slicing">slicing</a>, so
1528 they are completely safe to use anywhere.)
1531 <!-- -------------------------------------------------------------------- -->
1532 <a name="contact"></a>
1533 <h3>Contacting the author</h3>
1535 <p>Any comments, bug reports, etc. should be sent to warp@iki.fi
1538 <!-- -------------------------------------------------------------------- -->
1540 <a name="algorithm"></a>
1541 <h2>The algorithm used in the library</h2>
1543 <p>The whole idea behind the algorithm is to convert the regular infix
1544 format (the regular syntax for mathematical operations in most languages,
1545 like C and the input of the library) to postfix format. The postfix format
1546 is also called stack arithmetic since an expression in postfix format
1547 can be evaluated using a stack and operating with the top of the stack.
1552 <tr><th>infix</th> <th>postfix</th></tr>
1553 <tr><td><code>2+3</code></td><td><code>2 3 +</code></td></tr>
1554 <tr><td><code>1+2+3</code></td><td><code>1 2 + 3 +</code></td></tr>
1555 <tr><td><code>5*2+8/2</code></td><td><code>5 2 * 8 2 / +</code></td></tr>
1556 <tr><td><code>(5+9)*3</code></td><td><code>5 9 + 3 *</code></td></tr>
1559 <p>The postfix notation should be read in this way:
1561 <p>Let's take for example the expression: <code>5 2 * 8 2 / +</code>
1563 <li>Put 5 on the stack
1564 <li>Put 2 on the stack
1565 <li>Multiply the two values on the top of the stack and put the result on
1566 the stack (removing the two old values)
1567 <li>Put 8 on the stack
1568 <li>Put 2 on the stack
1569 <li>Divide the two values on the top of the stack
1570 <li>Add the two values on the top of the stack (which are in this case
1571 the result of 5*2 and 8/2, that is, 10 and 4).
1574 <p>At the end there's only one value in the stack, and that value is the
1575 result of the expression.
1577 <p>Why stack arithmetic?
1579 <p>The last example above can give you a hint.
1580 In infix format operators have precedence and we have to use parentheses to
1581 group operations with lower precedence to be calculated before operations
1582 with higher precedence.
1583 This causes a problem when evaluating an infix expression, specially
1584 when converting it to byte code. For example in this kind of expression:
1585 <code>(x+1)/(y+2)</code>
1586 we have to calculate first the two additions before we can calculate the
1587 division. We have to also keep counting parentheses, since there can be
1588 a countless amount of nested parentheses. This usually means that you
1589 have to do some type of recursion.
1591 <p>The simplest and mostefficient way of calculating this is to convert it
1592 to postfix notation.
1593 The postfix notation has the advantage that you can make all operations
1594 in a straightforward way. You just evaluate the expression from left to
1595 right, applying each operation directly and that's it. There are no
1596 parentheses to worry about. You don't need recursion anywhere.
1597 You have to keep a stack, of course, but that's extremely easily done.
1598 Also you just operate with the top of the stack, which makes it very easy.
1599 You never have to go deeper than 2 items in the stack.
1600 And even better: Evaluating an expression in postfix format is never
1601 slower than in infix format. All the contrary, in many cases it's a lot
1602 faster (eg. because all parentheses are optimized away).
1603 The above example could be expressed in postfix format:
1604 <code>x 1 + y 2 + /</code>
1606 <p>The good thing about the postfix notation is also the fact that it can
1607 be extremely easily expressed in bytecode form.
1608 You only need a byte value for each operation, for each variable and
1609 to push a constant to the stack.
1610 Then you can interpret this bytecode straightforwardly. You just interpret
1611 it byte by byte, from the beginning to the end. You never have to go back,
1612 make loops or anything.
1614 <p>This is what makes byte-coded stack arithmetic so fast.
1618 <!-- -------------------------------------------------------------------- -->
1619 <a name="license"></a>
1620 <h2>Usage license</h2>
1622 <p>Copyright © 2003-2010 Juha Nieminen, Joel Yliluoma
1624 <p>This Library is distributed under the
1625 <a href="http://www.gnu.org/copyleft/lesser.html">Lesser General Public
1626 License</a> (LGPL) version 3.