Let's consider following code:
#include <boost/spirit/include/lex_lexertl.hpp>
#include <boost/spirit/include/qi.hpp>
#include <algorithm>
#include <iostream>
#include <string>
#include <utility>
#include <vector>
namespace lex = boost::spirit::lex;
namespace qi = boost::spirit::qi;
template<typename Lexer>
class expression_lexer
: public lex::lexer<Lexer>
{
public:
typedef lex::token_def<> operator_token_type;
typedef lex::token_def<> value_token_type;
typedef lex::token_def<> variable_token_type;
typedef lex::token_def<lex::omit> parenthesis_token_type;
typedef std::pair<parenthesis_token_type, parenthesis_token_type> parenthesis_token_pair_type;
typedef lex::token_def<lex::omit> whitespace_token_type;
expression_lexer()
: operator_add('+'),
operator_sub('-'),
operator_mul("[x*]"),
operator_div("[:/]"),
value("\\d+(\\.\\d+)?"),
variable("%(\\w+)"),
parenthesis({
std::make_pair(parenthesis_token_type('('), parenthesis_token_type(')')),
std::make_pair(parenthesis_token_type('['), parenthesis_token_type(']'))
}),
whitespace("[ \\t]+")
{
this->self
= operator_add
| operator_sub
| operator_mul
| operator_div
| value
| variable
;
std::for_each(parenthesis.cbegin(), parenthesis.cend(),
[&](parenthesis_token_pair_type const& token_pair)
{
this->self += token_pair.first | token_pair.second;
}
);
this->self("WS") = whitespace;
}
operator_token_type operator_add;
operator_token_type operator_sub;
operator_token_type operator_mul;
operator_token_type operator_div;
value_token_type value;
variable_token_type variable;
std::vector<parenthesis_token_pair_type> parenthesis;
whitespace_token_type whitespace;
};
template<typename Iterator, typename Skipper>
class expression_grammar
: public qi::grammar<Iterator, Skipper>
{
public:
template<typename Tokens>
explicit expression_grammar(Tokens const& tokens)
: expression_grammar::base_type(start)
{
start %= expression >> qi::eoi;
expression %= sum_operand >> -(sum_operator >> expression);
sum_operator %= tokens.operator_add | tokens.operator_sub;
sum_operand %= fac_operand >> -(fac_operator >> sum_operand);
fac_operator %= tokens.operator_mul | tokens.operator_div;
if(!tokens.parenthesis.empty())
fac_operand %= parenthesised | terminal;
else
fac_operand %= terminal;
terminal %= tokens.value | tokens.variable;
if(!tokens.parenthesis.empty())
{
parenthesised %= tokens.parenthesis.front().first >> expression >> tokens.parenthesis.front().second;
std::for_each(tokens.parenthesis.cbegin() + 1, tokens.parenthesis.cend(),
[&](typename Tokens::parenthesis_token_pair_type const& token_pair)
{
parenthesised %= parenthesised.copy() | (token_pair.first >> expression >> token_pair.second);
}
);
}
}
private:
qi::rule<Iterator, Skipper> start;
qi::rule<Iterator, Skipper> expression;
qi::rule<Iterator, Skipper> sum_operand;
qi::rule<Iterator, Skipper> sum_operator;
qi::rule<Iterator, Skipper> fac_operand;
qi::rule<Iterator, Skipper> fac_operator;
qi::rule<Iterator, Skipper> terminal;
qi::rule<Iterator, Skipper> parenthesised;
};
int main()
{
typedef lex::lexertl::token<std::string::const_iterator> token_type;
typedef expression_lexer<lex::lexertl::lexer<token_type>> expression_lexer_type;
typedef expression_lexer_type::iterator_type expression_lexer_iterator_type;
typedef qi::in_state_skipper<expression_lexer_type::lexer_def> skipper_type;
typedef expression_grammar<expression_lexer_iterator_type, skipper_type> expression_grammar_type;
expression_lexer_type lexer;
expression_grammar_type grammar(lexer);
while(std::cin)
{
std::string line;
std::getline(std::cin, line);
std::string::const_iterator first = line.begin();
std::string::const_iterator const last = line.end();
bool const result = lex::tokenize_and_phrase_parse(first, last, lexer, grammar, qi::in_state("WS")[lexer.self]);
if(!result)
std::cout << "Parsing failed! Reminder: >" << std::string(first, last) << "<" << std::endl;
else
{
if(first != last)
std::cout << "Parsing succeeded! Reminder: >" << std::string(first, last) << "<" << std::endl;
else
std::cout << "Parsing succeeded!" << std::endl;
}
}
}
It is a simple parser for arithmetic expressions with values and variables. It is build using expression_lexer for extracting tokens, and then with expression_grammar to parse the tokens.
Use of lexer for such a small case might seem an overkill and probably is one. But that is the cost of simplified example. Also note that use of lexer allows to easily define tokens with regular expression while that allows to easily define them by external code (and user provided configuration in particular). With the example provided it would be no issue at all to read definition of tokens from an external config file and for example allow user to change variables from %name to $name.
The code seems to be working fine (checked on Visual Studio 2013 with Boost 1.61). Except that I have noticed that if I provide string like 5++5 it properly fails but reports as reminder just 5 rather than +5 which means the offending + was "unrecoverably" consumed. Apparently a token that was produced but did not match grammar is in no way returned to the original input. But that is not what I'm asking about. Just a side note I realized when checking the code.
Now the problem is with whitespace skipping. I very much don't like how it is done. While I have done it this way as it seems to be the one provided by many examples including answers to questions here on StackOverflow.
The worst thing seems to be that (nowhere documented?) qi::in_state_skipper. Also it seems that I have to add the whitespace token like that (with a name) rather than like all the other ones as using lexer.whitespace instead of "WS" doesn't seem to work.
And finally having to "clutter" the grammar with the Skipper argument doesn't seem nice. Shouldn't I be free of it? After all I want to make the grammar based on tokens rather than direct input and I want the whitespace to be excluded from tokens stream - it is not needed there anymore!
What other options do I have to skip whitespaces? What are advantages of doing it like it is now?
For some strange reason only now I found a different question, Boost.Spirit SQL grammar/lexer failure, where some other solution to whitespace skipping is provided. A better one!
So below is the example code reworked along the suggestions there:
#include <boost/spirit/include/lex_lexertl.hpp>
#include <boost/spirit/include/qi.hpp>
#include <algorithm>
#include <iostream>
#include <string>
#include <utility>
#include <vector>
namespace lex = boost::spirit::lex;
namespace qi = boost::spirit::qi;
template<typename Lexer>
class expression_lexer
: public lex::lexer<Lexer>
{
public:
typedef lex::token_def<> operator_token_type;
typedef lex::token_def<> value_token_type;
typedef lex::token_def<> variable_token_type;
typedef lex::token_def<lex::omit> parenthesis_token_type;
typedef std::pair<parenthesis_token_type, parenthesis_token_type> parenthesis_token_pair_type;
typedef lex::token_def<lex::omit> whitespace_token_type;
expression_lexer()
: operator_add('+'),
operator_sub('-'),
operator_mul("[x*]"),
operator_div("[:/]"),
value("\\d+(\\.\\d+)?"),
variable("%(\\w+)"),
parenthesis({
std::make_pair(parenthesis_token_type('('), parenthesis_token_type(')')),
std::make_pair(parenthesis_token_type('['), parenthesis_token_type(']'))
}),
whitespace("[ \\t]+")
{
this->self
+= operator_add
| operator_sub
| operator_mul
| operator_div
| value
| variable
| whitespace [lex::_pass = lex::pass_flags::pass_ignore]
;
std::for_each(parenthesis.cbegin(), parenthesis.cend(),
[&](parenthesis_token_pair_type const& token_pair)
{
this->self += token_pair.first | token_pair.second;
}
);
}
operator_token_type operator_add;
operator_token_type operator_sub;
operator_token_type operator_mul;
operator_token_type operator_div;
value_token_type value;
variable_token_type variable;
std::vector<parenthesis_token_pair_type> parenthesis;
whitespace_token_type whitespace;
};
template<typename Iterator>
class expression_grammar
: public qi::grammar<Iterator>
{
public:
template<typename Tokens>
explicit expression_grammar(Tokens const& tokens)
: expression_grammar::base_type(start)
{
start %= expression >> qi::eoi;
expression %= sum_operand >> -(sum_operator >> expression);
sum_operator %= tokens.operator_add | tokens.operator_sub;
sum_operand %= fac_operand >> -(fac_operator >> sum_operand);
fac_operator %= tokens.operator_mul | tokens.operator_div;
if(!tokens.parenthesis.empty())
fac_operand %= parenthesised | terminal;
else
fac_operand %= terminal;
terminal %= tokens.value | tokens.variable;
if(!tokens.parenthesis.empty())
{
parenthesised %= tokens.parenthesis.front().first >> expression >> tokens.parenthesis.front().second;
std::for_each(tokens.parenthesis.cbegin() + 1, tokens.parenthesis.cend(),
[&](typename Tokens::parenthesis_token_pair_type const& token_pair)
{
parenthesised %= parenthesised.copy() | (token_pair.first >> expression >> token_pair.second);
}
);
}
}
private:
qi::rule<Iterator> start;
qi::rule<Iterator> expression;
qi::rule<Iterator> sum_operand;
qi::rule<Iterator> sum_operator;
qi::rule<Iterator> fac_operand;
qi::rule<Iterator> fac_operator;
qi::rule<Iterator> terminal;
qi::rule<Iterator> parenthesised;
};
int main()
{
typedef lex::lexertl::token<std::string::const_iterator> token_type;
typedef expression_lexer<lex::lexertl::actor_lexer<token_type>> expression_lexer_type;
typedef expression_lexer_type::iterator_type expression_lexer_iterator_type;
typedef expression_grammar<expression_lexer_iterator_type> expression_grammar_type;
expression_lexer_type lexer;
expression_grammar_type grammar(lexer);
while(std::cin)
{
std::string line;
std::getline(std::cin, line);
std::string::const_iterator first = line.begin();
std::string::const_iterator const last = line.end();
bool const result = lex::tokenize_and_parse(first, last, lexer, grammar);
if(!result)
std::cout << "Parsing failed! Reminder: >" << std::string(first, last) << "<" << std::endl;
else
{
if(first != last)
std::cout << "Parsing succeeded! Reminder: >" << std::string(first, last) << "<" << std::endl;
else
std::cout << "Parsing succeeded!" << std::endl;
}
}
}
The differences are following:
whitespace token is added to lexer's self as all other tokens.expression_grammar no longer takes Skipper template argument. And so it is also removed from rules.lex::lexertl::actor_lexer is used instead of lex::lexertl::lexer since now there is an action associated with a token.tokenize_and_parse instead of tokenize_and_phrase_parse as I don't need to pass skipper anymore.this->self in lexer from = to += as it seems more flexible (resistant to order changes). But it doesn't affect the solution here.I'm good with this. It suites my needs (or better to say my taste) perfectly. However I wonder whether there are any other consequences of such change? Is any approach preferred in some situations? That I don't know.