Boost Spirit x3 conditional (ternary) operator parser (follow up question)

问题

This question is a follow up question for the one in

Boost Spirit x3 conditional (ternary) operator parser

The original question context did not show (my bad!) the ast attributes and the answer therefore could not take all the moving parts into account. This question now shows how the ast attributes looks like and how the ast is used to evaluate the expression with a symbol table.

The follow up question is therefore that how the correctly spelled ternary conditional should change the ast types and how the conditional and expression interact together (according to my understanding it is now not part of the x3::variant as it is to be removed from primary parser choices)

Here is how the ast attributes and declared symbol definitions look like


namespace x3 = boost::spirit::x3;

namespace ast {

struct nil {};
struct unary_op;
struct binary_op;
struct conditional_op;
struct expression;

struct operand : x3::variant<
                 nil
                 , double
                 , std::string
                 , x3::forward_ast<unary_op>
                 , x3::forward_ast<binary_op>
                 //, x3::forward_ast<conditional_op> // conditional_op not here?
                 , x3::forward_ast<expression>
                 > {
    using base_type::base_type;
    using base_type::operator=;
};

struct unary_op {
    double (*op)(double);
    operand rhs;
};

struct binary_op {
    double (*op)(double, double);
    operand lhs;
    operand rhs;
};

/*
struct conditional_op {
    operand lhs;
    operand rhs_true;
    operand rhs_false;
};
*/

struct conditional_op {
    expression lhs;
    // how the exact type is spelled?
    optional<expression, expression> maybe_rhs;
};

struct operation {
    double (*op)(double, double);
    operand rhs;
};

// what is the type of expression ?
struct expression {
    conditional_op conditional;
};

/*
struct expression {
    operand lhs;
    std::list<operation> rhs;
};
*/

} // namespace ast

struct constant_ : x3::symbols<double> {
    constant_() {
        add
            ("e"      , boost::math::constants::e<double>())
            ("pi"     , boost::math::constants::pi<double>())
            ;
    }
} constant;

struct ufunc_ : x3::symbols<double (*)(double)> {
    ufunc_() {
        add
            ("abs"   , static_cast<double (*)(double)>(&std::abs))
            ;
    }
} ufunc;

struct bfunc_ : x3::symbols<double (*)(double, double)> {
    bfunc_() {
        add
            ("max"  , static_cast<double (*)(double, double)>(&std::fmax))
            ;
    }
} bfunc;

struct unary_op_ : x3::symbols<double (*)(double)> {
    unary_op_() {
        add
            ("+", static_cast<double (*)(double)>(&math::plus))
            ("-", static_cast<double (*)(double)>(&math::minus))
            ("!", static_cast<double (*)(double)>(&math::unary_not))
            ;
    }
} unary_op;

struct additive_op_ : x3::symbols<double (*)(double, double)> {
    additive_op_() {
        add
            ("+", static_cast<double (*)(double, double)>(&math::plus))
            ("-", static_cast<double (*)(double, double)>(&math::minus))
            ;
    }
} additive_op;

struct multiplicative_op_ : x3::symbols<double (*)(double, double)> {
    multiplicative_op_() {
        add
            ("*", static_cast<double (*)(double, double)>(&math::multiplies))
            ("/", static_cast<double (*)(double, double)>(&math::divides))
            ("%", static_cast<double (*)(double, double)>(&std::fmod))
            ;
    }
} multiplicative_op;

struct logical_op_ : x3::symbols<double (*)(double, double)> {
    logical_op_() {
        add
            ("&&", static_cast<double (*)(double, double)>(&math::logical_and))
            ("||", static_cast<double (*)(double, double)>(&math::logical_or))
            ;
    }
} logical_op;

struct relational_op_ : x3::symbols<double (*)(double, double)> {
    relational_op_() {
        add
            ("<" , static_cast<double (*)(double, double)>(&math::less))
            ("<=", static_cast<double (*)(double, double)>(&math::less_equals))
            (">" , static_cast<double (*)(double, double)>(&math::greater))
            (">=", static_cast<double (*)(double, double)>(&math::greater_equals))
            ;
    }
} relational_op;

struct equality_op_ : x3::symbols<double (*)(double, double)> {
    equality_op_() {
        add
            ("==", static_cast<double (*)(double, double)>(&math::equals))
            ("!=", static_cast<double (*)(double, double)>(&math::not_equals))
            ;
    }
} equality_op;

struct power_ : x3::symbols<double (*)(double, double)> {
    power_() {
        add
            ("**", static_cast<double (*)(double, double)>(&std::pow))
            ;
    }
} power;

The more complete grammar and the definition of ast attributes is below (with modifications based on the answer in Boost Spirit x3 conditional (ternary) operator parser)


struct expression_class;
struct logical_class;
struct equality_class;
struct relational_class;
struct additive_class;
struct multiplicative_class;
struct factor_class;
struct primary_class;
struct unary_class;
struct binary_class;
struct conditional_class;
struct variable_class;

// Rule declarations

auto const expression     = x3::rule<expression_class    , ast::expression    >{"expression"};
auto const logical        = x3::rule<logical_class       , ast::expression    >{"logical"};
auto const equality       = x3::rule<equality_class      , ast::expression    >{"equality"};
auto const relational     = x3::rule<relational_class    , ast::expression    >{"relational"};
auto const additive       = x3::rule<additive_class      , ast::expression    >{"additive"};
auto const multiplicative = x3::rule<multiplicative_class, ast::expression    >{"multiplicative"};
auto const factor         = x3::rule<factor_class        , ast::expression    >{"factor"};
auto const primary        = x3::rule<primary_class       , ast::operand       >{"primary"};
auto const unary          = x3::rule<unary_class         , ast::unary_op      >{"unary"};
auto const binary         = x3::rule<binary_class        , ast::binary_op     >{"binary"};
auto const conditional    = x3::rule<conditional_class   , ast::conditional_op>{"conditional"};
auto const variable       = x3::rule<variable_class      , std::string        >{"variable"};

// Rule defintions
/* This is a bit of magic to me. Does this definition now say that expression 
   itself is now initializer list constructible from the conditional (which is spelled below)?
*/
auto const expression_def =
    conditional
    ;

/* now ast::conditional_op type should be constructible from an initialization list consisting 
   of of an expression and optional<tuple<expression,expression>> ? How these types should be 
   spelled in the struct? There is a circular reference between expression and conditional :D ? 
*/  
auto const conditional_def =
    logical >> -('?' > expression > ':'> expression)
    ;

auto const logical_def =
    equality >> *(logical_op > equality)
    ;

auto const equality_def =
    relational >> *(equality_op > relational)
    ;

auto const relational_def =
    additive >> *(relational_op > additive)
    ;

auto const additive_def =
    multiplicative >> *(additive_op > multiplicative)
    ;

auto const multiplicative_def =
    factor >> *(multiplicative_op > factor)
    ;

auto const factor_def =
    primary >> *( power > factor )
    ;

auto const unary_def =
    ufunc > '(' > expression > ')'
    ;

auto const binary_def =
    bfunc > '(' > expression > ',' > expression > ')'
    ;

auto const primary_def =
      x3::double_
    | ('(' > expression > ')')
    | (unary_op > primary)
    | binary
    | unary
 // | conditional // by removing the conditional from primary implies the type of x3::variant changes
    | variable
    ;

BOOST_SPIRIT_DEFINE(
    expression,
    logical,
    equality,
    relational,
    additive,
    multiplicative,
    factor,
    primary,
    unary,
    binary,
    conditional,
    variable
)

Here is how the AST is traversed using boost static visitor to evaluate the expression with a variable symbol table

namespace ast {

// Evaluator

struct Evaluator {
    using result_type = double;

    explicit Evaluator(std::map<std::string, double> sym);

    double operator()(nil) const;

    double operator()(double n) const;

    double operator()(std::string const &c) const;

    double operator()(operation const &x, double lhs) const;

    double operator()(unary_op const &x) const;

    double operator()(binary_op const &x) const;

    double operator()(conditional_op const &x) const;

    double operator()(expression const &x) const;

  private:
    std::map<std::string, double> st;
};

Evaluator::Evaluator(std::map<std::string, double> sym) 
: st(std::move(sym)) {}

double Evaluator::operator()(nil) const {
    BOOST_ASSERT(0);
    return 0;
}

double Evaluator::operator()(double n) const { return n; }

double Evaluator::operator()(std::string const &c) const {
    auto it = st.find(c);
    if (it == st.end()) {
        throw std::invalid_argument("Unknown variable " + c);
    }
    return it->second;
}

double Evaluator::operator()(operation const &x, double lhs) const {
    double rhs = boost::apply_visitor(*this, x.rhs);
    return x.op(lhs, rhs);
}

double Evaluator::operator()(unary_op const &x) const {
    double rhs = boost::apply_visitor(*this, x.rhs);
    return x.op(rhs);
}

double Evaluator::operator()(binary_op const &x) const {
    double lhs = boost::apply_visitor(*this, x.lhs);
    double rhs = boost::apply_visitor(*this, x.rhs);
    return x.op(lhs, rhs);
}

double Evaluator::operator()(conditional_op const &x) const {
    return static_cast<bool>(boost::apply_visitor(*this, x.lhs)) 
        ? boost::apply_visitor(*this, x.rhs_true) 
        : boost::apply_visitor(*this, x.rhs_false);
}

double Evaluator::operator()(expression const &x) const {
    double state = boost::apply_visitor(*this, x.lhs);
    for (operation const &oper : x.rhs) {
        state = (*this)(oper, state);
    }
    return state;
}

} // namespace ast

回答1:

So, the top-level attribute exposed is expression, which is, frankly, not representing an expression at all.

Rather, it is representing an artificial unit of expression input syntax, which could perhaps be dubbed "operation_chain".

This is also going to make it hard to use your AST for semantically correct transformations (like e.g. expression evaluation) because crucial information like precedence of operations is not encoded in it.

In fact, if we're not careful it's very possible that this information - if present in the input - would be lost. I think it's possible in practice to go from your AST and reconstruct the operation tree with dependent operations in order of their precedence. But I usually err on the safe side of explicitly modeling the expression tree to reflect the operation dependencies.

That said, the conditional_op is not a chaining binary operation, so it doesn't fit the mold. I'd suggest making the "top level" rules expose an ast::operand instead (so it can fit the conditional_op or expression both just fine).

However, due the "lazy" way we detect the conditional, this requires some semantic actions to actually build the proper attributes:

auto const conditional_def =
    logical [([](auto& ctx) { _val(ctx) = _attr(ctx); })]
    >> -('?' > expression > ':' > expression) [make_conditional_op]
    ;

The first semantic action is straight-forward, the second one became large enough to define it out-of-line:

auto make_conditional_op = [](auto& ctx) {
    using boost::fusion::at_c;
    x3::_val(ctx) = ast::conditional_op {
        x3::_val(ctx),
        at_c<0>(x3::_attr(ctx)),
        at_c<1>(x3::_attr(ctx)) };
};

Still straight-forward but clumsy. Notice that the reason is that we expose different types depending on the presence of the optional branch.

Here's it all put together working:

Live On Coliru

//#define BOOST_SPIRIT_X3_DEBUG
//#define DEBUG_SYMBOLS
#include <iostream>
#include <functional>
#include <iomanip>
#include <list>
#include <boost/fusion/adapted/struct.hpp>
#include <boost/math/constants/constants.hpp>
#include <boost/spirit/home/x3.hpp>
#include <boost/spirit/home/x3/support/ast/variant.hpp>
namespace x3 = boost::spirit::x3;

namespace ast {

    struct nil {};
    struct unary_op;
    struct binary_op;
    struct conditional_op;
    struct expression;

    using UnFunc = std::function<double(double)>;
    using BinFunc = std::function<double(double, double)>;

    struct operand : x3::variant<
             nil
           , double
           , std::string
           , x3::forward_ast<unary_op>
           , x3::forward_ast<binary_op>
           , x3::forward_ast<conditional_op>
           , x3::forward_ast<expression> >
    {
        using base_type::base_type;
        using base_type::operator=;
    };

    struct unary_op {
        UnFunc op;
        operand rhs;
    };

    struct binary_op {
        BinFunc op;
        operand lhs;
        operand rhs;
    };

    struct conditional_op {
        operand lhs;
        operand rhs_true;
        operand rhs_false;
    };

    struct operation {
        BinFunc op;
        operand rhs;
    };

    struct expression {
        operand lhs;
        std::list<operation> rhs;
    };

} // namespace ast

BOOST_FUSION_ADAPT_STRUCT(ast::expression, lhs, rhs)
BOOST_FUSION_ADAPT_STRUCT(ast::operation, op, rhs)
BOOST_FUSION_ADAPT_STRUCT(ast::conditional_op, lhs, rhs_true, rhs_false)
BOOST_FUSION_ADAPT_STRUCT(ast::binary_op, op, lhs, rhs)
BOOST_FUSION_ADAPT_STRUCT(ast::unary_op, op, rhs)

namespace P {

    struct ehbase {
        template <typename It, typename Ctx>
        x3::error_handler_result on_error(It f, It l, x3::expectation_failure<It> const& e, Ctx const& /*ctx*/) const {
            std::cout << std::string(f,l) << "\n"
                      << std::setw(1+std::distance(f, e.where())) << "^"
                      << "-- expected: " << e.which() << "\n";
            return x3::error_handler_result::fail;
        }
    };

    struct expression_class     : ehbase {};
    struct logical_class        : ehbase {};
    struct equality_class       : ehbase {};
    struct relational_class     : ehbase {};
    struct additive_class       : ehbase {};
    struct multiplicative_class : ehbase {};
    struct factor_class         : ehbase {};
    struct primary_class        : ehbase {};
    struct unary_class          : ehbase {};
    struct binary_class         : ehbase {};
    struct conditional_class    : ehbase {};
    struct variable_class       : ehbase {};

    // Rule declarations
    auto const expression     = x3::rule<expression_class    , ast::operand       >{"expression"};
    auto const conditional    = x3::rule<conditional_class   , ast::operand       >{"conditional"};
    auto const primary        = x3::rule<primary_class       , ast::operand       >{"primary"};
    auto const logical        = x3::rule<logical_class       , ast::expression    >{"logical"};
    auto const equality       = x3::rule<equality_class      , ast::expression    >{"equality"};
    auto const relational     = x3::rule<relational_class    , ast::expression    >{"relational"};
    auto const additive       = x3::rule<additive_class      , ast::expression    >{"additive"};
    auto const multiplicative = x3::rule<multiplicative_class, ast::expression    >{"multiplicative"};
    auto const factor         = x3::rule<factor_class        , ast::expression    >{"factor"};
    auto const unary          = x3::rule<unary_class         , ast::unary_op      >{"unary"};
    auto const binary         = x3::rule<binary_class        , ast::binary_op     >{"binary"};
    auto const variable       = x3::rule<variable_class      , std::string        >{"variable"};

    struct constant_ : x3::symbols<double> {
        constant_() {
            this->add
                ("e"      , boost::math::constants::e<double>())
                ("pi"     , boost::math::constants::pi<double>())
                ;
        }
    } constant;

    struct ufunc_ : x3::symbols<ast::UnFunc> {
        ufunc_() {
            this->add
                ("abs"   , &std::abs<double>)
                ;
        }
    } ufunc;

    struct bfunc_ : x3::symbols<ast::BinFunc> {
        bfunc_() {
            this->add
                ("max"  , [](double a,double b){ return std::fmax(a,b); })
                ("min"  , [](double a,double b){ return std::fmin(a,b); })
                ("pow"  , [](double a,double b){ return std::pow(a,b); })
                ;
        }
    } bfunc;

    struct unary_op_ : x3::symbols<ast::UnFunc> {
        unary_op_() {
            this->add
                ("+", [](double v) { return +v; })
                ("-", std::negate{})
                ("!", [](double v) { return !v; })
                ;
        }
    } unary_op;

    struct additive_op_ : x3::symbols<ast::BinFunc> {
        additive_op_() {
            this->add
                ("+", std::plus{})
                ("-", std::minus{})
                ;
        }
    } additive_op;

    struct multiplicative_op_ : x3::symbols<ast::BinFunc> {
        multiplicative_op_() {
            this->add
                ("*", std::multiplies<>{})
                ("/", std::divides<>{})
                ("%", [](double a, double b) { return std::fmod(a, b); })
                ;
        }
    } multiplicative_op;

    struct logical_op_ : x3::symbols<ast::BinFunc> {
        logical_op_() {
            this->add
                ("&&", std::logical_and{})
                ("||", std::logical_or{})
                ;
        }
    } logical_op;

    struct relational_op_ : x3::symbols<ast::BinFunc> {
        relational_op_() {
            this->add
                ("<" , std::less{})
                ("<=", std::less_equal{})
                (">" , std::greater{})
                (">=", std::greater_equal{})
                ;
        }
    } relational_op;

    struct equality_op_ : x3::symbols<ast::BinFunc> {
        equality_op_() {
            this->add
                ("==", std::equal_to{})
                ("!=", std::not_equal_to{})
                ;
        }
    } equality_op;

    struct power_ : x3::symbols<ast::BinFunc> {
        power_() {
            this->add
                ("**", [](double v, double exp) { return std::pow(v, exp); })
                ;
        }
    } power;

    auto const variable_def = x3::lexeme[x3::alpha >> *x3::alnum];

    // Rule defintions
    auto const expression_def =
        conditional
        ;

    auto make_conditional_op = [](auto& ctx) {
        using boost::fusion::at_c;
        x3::_val(ctx) = ast::conditional_op {
            x3::_val(ctx),
            at_c<0>(x3::_attr(ctx)),
            at_c<1>(x3::_attr(ctx)) };
    };

    auto const conditional_def =
        logical [([](auto& ctx) { _val(ctx) = _attr(ctx); })]
        >> -('?' > expression > ':' > expression) [make_conditional_op]
        ;

    auto const logical_def =
        equality >> *(logical_op > equality)
        ;

    auto const equality_def =
        relational >> *(equality_op > relational)
        ;

    auto const relational_def =
        additive >> *(relational_op > additive)
        ;

    auto const additive_def =
        multiplicative >> *(additive_op > multiplicative)
        ;

    auto const multiplicative_def =
        factor >> *(multiplicative_op > factor)
        ;

    auto const factor_def =
        primary >> *( power > factor )
        ;

    auto const unary_def 
        = (unary_op > primary)
        | (ufunc > '(' > expression > ')')
        ;

    auto const binary_def =
        bfunc > '(' > expression > ',' > expression > ')'
        ;

    auto const primary_def =
        x3::double_
        | ('(' > expression > ')')
        //| (unary_op > primary)
        | binary
        | unary
        | constant
        | variable
        ;

    BOOST_SPIRIT_DEFINE(expression)
    BOOST_SPIRIT_DEFINE(logical)
    BOOST_SPIRIT_DEFINE(equality)
    BOOST_SPIRIT_DEFINE(relational)
    BOOST_SPIRIT_DEFINE(additive)
    BOOST_SPIRIT_DEFINE(multiplicative)
    BOOST_SPIRIT_DEFINE(factor)
    BOOST_SPIRIT_DEFINE(primary)
    BOOST_SPIRIT_DEFINE(unary)
    BOOST_SPIRIT_DEFINE(binary)
    BOOST_SPIRIT_DEFINE(conditional)
    BOOST_SPIRIT_DEFINE(variable)
}

int main() {
    for (std::string const input : {
           "x+(3**pow(2,8))",
           "1 + (2 + abs(x))",
           "min(x,1+y)",
           "(x > y ? 1 : 0) * (y - z)",
           "min(3**4,7))",
           "3***4",
           "(3,4)",
        })
    {
        std::cout << " ===== " << std::quoted(input) << " =====\n";
        auto f = begin(input), l = end(input);
        ast::operand out;
        if (phrase_parse(f, l, P::expression, x3::space, out)) {
            std::cout << "Success\n";
        } else {
            std::cout << "Failed\n";
        }
        if (f!=l) {
            std::cout << "Unparsed: " << std::quoted(std::string(f,l)) << "\n";
        }
    }
}

Printing

 ===== "x+(3**pow(2,8))" =====
Success
 ===== "1 + (2 + abs(x))" =====
Success
 ===== "min(x,1+y)" =====
Success
 ===== "(x > y ? 1 : 0) * (y - z)" =====
Success
 ===== "min(3**4,7))" =====
Success
Unparsed: ")"
 ===== "3***4" =====
3***4
   ^-- expected: factor
Failed
Unparsed: "3***4"
 ===== "(3,4)" =====
(3,4)
  ^-- expected: ')'
Failed
Unparsed: "(3,4)"

I feel it should be possible to be