【问题标题】:Spirit-Qi: How can I write a nonterminal parser?Spirit-Qi:如何编写非终端解析器?
【发布时间】:2016-10-06 19:11:21
【问题描述】:

我想写一个可以使用的解析器(作为qi扩展) via my_parser(p1, p2, ...) 其中p1, p2, ... 是qi 解析器表达式。

实际上,我想实现一个 best_matchparser ,它的工作原理类似于 qi 替代方案,但不是选择第一个匹配规则,而是选择“解释”大部分输入的规则。

给定两个规则simple_id = +(qi::alpha)complex_id = simple_id >> *(qi::string("::") > simple_id),它将在输入willy::anton 上选择complex_id。并且这样做不会产生中间属性。这些好处可以通过运行时得到回报,因为需要先行解析。

在我看来,这种解析器构造有几个用例。 benchmark(p1, ...) 作为优化的支持解析器,仅举一个例子。一旦我知道如何提供,我会提供。

这个解析器将是一个非终端。我尝试了(努力),但我无法在 qi 中找到我的问题的答案。我的猜测是,C++ 机制与 c qi 紧密集成,以至于没有可以理解的入口点,至少对我来说是这样。

很容易实现我想要引入运算符的内容。我在下面附加了当前的解决方案。它按预期编译,但不完整。

qi 操作符得到一个 fusion::vector/sequence (whatever) 现成的并对其进行操作。似乎没有图书馆产品可以解决我的问题。甚至 make_nary_composite 已经期望将 args 编译为 Elements

我尝试了很多,但都失败了,所以我不想让你厌烦。

我可以想象的一种解决方法是提供一个有状态运算符,,这将使p1, ... 成为合法的qi 表达式。然后实现一个 unary_parser best_match 或指令来处理该表达式。 , 将获得两种模式:获取当前(成功)解析器消耗的输入长度,并实际解析阶段 1 中选定的输入。包装器首先在模式 1 中运行逗号解析器,然后在模式 2 中运行。它可能很丑,但可以工作。

运算符实现 p1 |= p2 |= ... 对于 n > 2 的运行时而言是最昂贵的。我很乐意解决这个问题。

最小(但仍然是合理的)开销将有best_match(p1, ...)。 这有可能吗?

因为我对 boost::fusion 不太熟悉,所以我也在代码中添加了一些问题(关于 fusion 的方法)。

感觉不对,将实际上是一个三元非终端解析器的东西压入一元解析器方案。但由于我的理解很差,这似乎是完成它的唯一方法。我非常感谢任何帮助我摆脱困境的帮助。

我的环境:Boost 1.61,MSVC 2015 Upd 2,目标 win32console。

进展:

我认为,我正在缓慢但肯定地得到它。我很高兴看到error_invalid_expression。由于以下 proto 表达式,编译失败:

boost::proto::exprns_::expr<
    boost::proto::tagns_::tag::function,boost::proto::argsns_::list5<
        const boost::proto::exprns_::expr<
            boost::proto::tagns_::tag::terminal,boost::proto::argsns_::term<
                mxc::qitoo::tag::best_match
            >
        ,0> &
        ,const boost::spirit::qi::rule<
            iterator_type,std::string (void),
            boost::spirit::standard::space_type,
            boost::spirit::unused_type,boost::spirit::unused_type> &
        ,const boost::spirit::qi::rule<
            iterator_type,std::string (void),
            boost::spirit::standard::space_type,
            boost::spirit::unused_type,
            boost::spirit::unused_type> &
        , const boost::spirit::qi::rule<
            iterator_type,std::string (void),
            boost::spirit::standard::space_type,
            boost::spirit::unused_type,
            boost::spirit::unused_type> &
        ,const boost::spirit::qi::rule<
            iterator_type,std::string (void),
            boost::spirit::standard::space_type,
            boost::spirit::unused_type,
            boost::spirit::unused_type> &
        >
    ,5>

这实际上是描述我的 best_match 解析器的函数样式用法的表达式。我的测试规则看起来像

qitoo::best_match(id, qualified_id, id, qualified_id)

其中idqualified_id 是上面提到的规则。我想要的一切。 发生错误是因为此表达式没有成员 parse。好的。深入挖掘,我发现 qi 的元编译器确实支持一元、二元和一元(即可变参数)解析器。但它不支持函数样式语法。 Qi 似乎仅将一元、二元和一元用于运算符。虽然支持 nary 类型,但它或多或少已经过时了,因为 qi 运算符是二元最大值。

结束进度编辑

#include <boost/spirit/home/qi.hpp>

namespace boost {
    namespace spirit
    {
        ///////////////////////////////////////////////////////////////////////////
        // Enablers
        ///////////////////////////////////////////////////////////////////////////
        template <>
        struct use_operator<qi::domain, proto::tag::bitwise_or_assign> // enables |=
            : mpl::true_ {};

        template <>
        struct flatten_tree<qi::domain, proto::tag::bitwise_or_assign> // flattens |=
            : mpl::true_ {};
    }
}

namespace mxc {
    namespace qitoo {

        namespace spirit = boost::spirit;
        namespace qi = spirit::qi;
        namespace fusion = boost::fusion;

        template <typename Elements>
        struct best_match_parser : qi::nary_parser<mxc::qitoo::best_match_parser<Elements>>
        {
            // This one does a lookahead parse of each qi expression to find out which rule matches 
            // most of the input 
            template <typename Iterator, typename Context, typename Skipper>
            struct best_function
            {
                best_function(Iterator& first, Iterator const& last, Context& context,
                    Skipper const& skipper, int& best, int& idx, int& size)
                    : first(first), last(last), context(context), skipper(skipper), best(best),
                    idx(idx), size(size) {};

                template <typename Component>
                void operator()(Component& component) const
                {
                    Iterator f = first;
                    if (component.parse(f, last, context, skipper, qi::unused)) {
                        // please have a look on how I transport best, idx and size
                        // into the parser context
                        //
                        // I guess, this is a nasty hack and could be done better
                        // Any advice would be highliy appreciated

                        int l = f - first;
                        if (l > size) {
                            size = l;
                            best = idx++;
                        }
                        idx++;
                    }
                }

            private:
                int& best;
                int& idx;
                int& size;

                Iterator&       first;
                Iterator const& last;
                Context&        context;
                Skipper const&  skipper;
            };

            template <typename Context, typename Iterator>
            struct attribute
            {
                // Put all the element attributes in a tuple
                typedef typename spirit::traits::build_attribute_sequence <
                    Elements, Context, spirit::traits::alternative_attribute_transform
                    , Iterator, qi::domain
                >::type all_attributes;

                // Ok, now make a variant over the attribute sequence. Note that
                // build_variant makes sure that 1) all attributes in the variant
                // are unique 2) puts the unused attribute, if there is any, to
                // the front and 3) collapses single element variants, variant<T>
                // to T.
                typedef typename
                    spirit::traits::build_variant<all_attributes>::type
                    type;
            };

            best_match_parser(Elements const& elements_) : elements(elements_), size(0), idx(0), best(-1) {}


            template <typename Iterator, typename Context, typename Skipper, typename Attribute>
            bool parse(Iterator& first, Iterator const& last
                , Context& context, Skipper const& skipper
                , Attribute& attr_) const
            {
                best_function<Iterator, Context, Skipper>  f(first, last, context, skipper, best, idx, size);

                // find out which parser matches most of the input
                fusion::for_each(elements, f);

                // best >= 0 if at least one parser was successful
                if (best >= 0) {
                    // now that I have the best parser identified, how do I access it?
                    // I understand that the next line can not work, but I'm looking for something like that

                    // --> auto v = fusion::at<boost::mpl::int_<best>>(elements);
                };

                return false;
            }

            template <typename Context>
            qi::info what(Context& context) const
            {
                qi::info result("best_match");
                fusion::for_each(elements,
                    spirit::detail::what_function<Context>(result, context));

                return result;
            }

            Elements elements;
            mutable int  best;
            mutable int  idx;
            mutable int  size;

        };
    }
}

namespace boost {
    namespace spirit {
        namespace qi {

            ///////////////////////////////////////////////////////////////////////////
            // Parser generators: make_xxx function (objects)
            ///////////////////////////////////////////////////////////////////////////
            template <typename Elements, typename Modifiers>
            struct make_composite<proto::tag::bitwise_or_assign, Elements, Modifiers>
                : make_nary_composite < Elements, mxc::qitoo::best_match_parser >
            {};
        }

        namespace traits {
            ///////////////////////////////////////////////////////////////////////////
            template <typename Elements>
            struct has_semantic_action<mxc::qitoo::best_match_parser<Elements> >
                : nary_has_semantic_action<Elements> {};

            ///////////////////////////////////////////////////////////////////////////
            template <typename Elements, typename Attribute, typename Context
                , typename Iterator>
                struct handles_container<mxc::qitoo::best_match_parser<Elements>, Attribute, Context
                , Iterator>
                : nary_handles_container<Elements, Attribute, Context, Iterator> {};
        }
    }
}
namespace qi = boost::spirit::qi;
namespace qitoo = mxc::qitoo;

using iterator_type = std::string::const_iterator;
using result_type = std::string;

template<typename Parser>
void parse(const std::string message, const std::string& input, const std::string& rule, const Parser& parser) {
    iterator_type iter = input.begin(), end = input.end();

    std::vector<result_type> parsed_result;

    std::cout << "-------------------------\n";
    std::cout << message << "\n";
    std::cout << "Rule: " << rule << std::endl;
    std::cout << "Parsing: \"" << input << "\"\n";

    bool result = qi::phrase_parse(iter, end, parser, qi::space, parsed_result);
    if (result)
    {
        std::cout << "Parser succeeded.\n";
        std::cout << "Parsed " << parsed_result.size() << " elements:";
        for (const auto& str : parsed_result)
            std::cout << "[" << str << "]";
        std::cout << std::endl;
    }
    else
    {
        std::cout << "Parser failed" << std::endl;
    }
    if (iter == end) {
        std::cout << "EOI reached." << std::endl;
    }
    else {
        std::cout << "EOI not reached. Unparsed: \"" << std::string(iter, end) << "\"" << std::endl;
    }
    std::cout << "-------------------------\n";

}


int main()
{
    namespace qi = boost::spirit::qi;

    qi::rule < iterator_type, std::string(), qi::space_type>
        id = (qi::alpha | qi::char_('_')) >> *(qi::alnum | qi::char_('_'));

    qi::rule < iterator_type, std::string(), qi::space_type>
        qualified_id = id >> *(qi::string("::") > id);


    namespace qitoo = mxc::qitoo;
    namespace qi = boost::spirit::qi;

    parse("best match operator, select second rule"
        , "willy::anton::lutz"
        , "id |= qualified_id"
        , id |= qualified_id);
}

【问题讨论】:

  • 找到了一个解决方案,但涉及修改 qi 的元编译器和 make_component 部分。如果有其他解决方案我没有设法找出。我对自己的非标准 qi 不太满意。

标签: c++ parsing boost boost-spirit boost-spirit-qi


【解决方案1】:

你的例子

我看不出您的样本对这些有何要求。只需重新排序您的分支,然后意识到短分支只是 n=1 的合格案例的一个特例:Live On Coliru¹(或 using X3 version,如果您愿意)。

一般情况

现在,提到 x3,它可以让您的生活更轻松!

这就是我认为你想要的,在一般情况下:

namespace parser {

    template <typename... Parsers>
    struct longest_parser : x3::parser_base {
        longest_parser(Parsers... sub) : _alternatives {sub...} { }

        template <typename It, typename Ctx, typename Other, typename Attr>
        bool parse(It& f, It l, Ctx& ctx, Other const& other, Attr& attr) const {
            auto const saved = f;

            //// To exclude pre-skip from length comparisons, do pre-skip here:
            // x3::skip_over(f, l, ctx);
            auto seq = std::make_index_sequence<sizeof...(Parsers)>();

            auto best = select_best(f, l, ctx, seq);
            //std::cout << "Longest match at index #" << best << "\n";

            bool ok = dispatch(f, l, ctx, other, attr, best, seq);

            if (!ok)
                f = saved;

            return ok;
        }

      private:
        template <typename It, typename Ctx, typename P>
        size_t length_of(It f, It l, Ctx ctx, P const& p) const {
            boost::iterator_range<It> matched;
            return x3::raw[p].parse(f, l, ctx, x3::unused, matched)? boost::size(matched) : 0;
        }

        template <typename It, typename Ctx, size_t... I>
            size_t select_best(It f, It l, Ctx& ctx, std::index_sequence<I...>) const {
                std::array<size_t, sizeof...(I)> lengths { length_of(f, l, ctx, std::get<I>(_alternatives))... };
                return std::distance(lengths.begin(), std::max_element(lengths.begin(), lengths.end()));
            }

        template <typename It, typename Ctx, typename Other, typename Attr, size_t... I>
        bool dispatch(It& f, It l, Ctx& ctx, Other const& other, Attr& attr, size_t targetIdx, std::index_sequence<I...>) const {
            //return (real_parse<I>(f, l, ctx, other, attr, targetIdx) || ...);
            std::array<bool, sizeof...(I)> b = { real_parse<I>(f, l, ctx, other, attr, targetIdx)... };

            return std::accumulate(b.begin(), b.end(), false, std::logical_or<bool>());
        }

        template <size_t Idx, typename It, typename Ctx, typename Other, typename Attr>
        bool real_parse(It& f, It l, Ctx& ctx, Other const& other, Attr& attr, size_t targetIdx) const {
            if (targetIdx != Idx)
                return false;

            return std::get<Idx>(_alternatives).parse(f, l, ctx, other, attr);
        }

        std::tuple<Parsers...> _alternatives;
    };

    template <typename... Ps>
        longest_parser<Ps...> longest(Ps... p) { return {x3::as_parser(p)...}; }
}

如果您的编译器支持,请注意您可以在 dispatch 中使用的折叠表达式(Coliru 支持,edit it to see!)。

还要注意关于可跳过(可能是空格)的微妙选择;如果它对长度比较不重要,请取消注释前跳过。

现场演示

Live On Coliru

#include <boost/spirit/home/x3.hpp>
#include <type_traits>
#include <iostream>
#include <numeric>

namespace x3 = boost::spirit::x3;

namespace std {
    template <typename T> // just for easy debug printing; hack
    static std::ostream& operator<<(std::ostream& os, std::vector<T> const& v) {
        for (auto& el : v) std::cout << '[' << el << ']';
        return os;
    }
}

using string_vec  = std::vector<std::string>;
using result_type = boost::variant<std::string, double, string_vec>;

template <typename Parser>
void parse(const std::string message, const std::string &input, const std::string &rule, const Parser &parser) {
    auto iter = input.begin(), end = input.end();

    std::cout << "-------------------------\n";
    std::cout << message << "\n";
    std::cout << "Rule:     " << rule  << "\n";
    std::cout << "Parsing: '" << input << "'\n";

    result_type parsed_result;
    bool result = phrase_parse(iter, end, parser, x3::space, parsed_result);

    if (result) {
        std::cout << "Parsed " << parsed_result << "\n";
    } else {
        std::cout << "Parser failed\n";
    }
    if (iter != end)
        std::cout << "EOI not reached. Unparsed: '" << std::string(iter, end) << "'\n";
}

namespace parser {

    template <typename... Parsers>
    struct longest_parser : x3::parser_base {
        longest_parser(Parsers... sub) : _alternatives {sub...} { }

        template <typename It, typename Ctx, typename Other, typename Attr>
        bool parse(It& f, It l, Ctx& ctx, Other const& other, Attr& attr) const {
            auto const saved = f;

            //// To exclude pre-skip from length comparisons, do pre-skip here:
            // x3::skip_over(f, l, ctx);
            auto seq = std::make_index_sequence<sizeof...(Parsers)>();

            auto best = select_best(f, l, ctx, seq);
            //std::cout << "Longest match at index #" << best << "\n";

            bool ok = dispatch(f, l, ctx, other, attr, best, seq);

            if (!ok)
                f = saved;

            return ok;
        }

      private:
        template <typename It, typename Ctx, typename P>
        size_t length_of(It f, It l, Ctx ctx, P const& p) const {
            boost::iterator_range<It> matched;
            return x3::raw[p].parse(f, l, ctx, x3::unused, matched)? boost::size(matched) : 0;
        }

        template <typename It, typename Ctx, size_t... I>
            size_t select_best(It f, It l, Ctx& ctx, std::index_sequence<I...>) const {
                std::array<size_t, sizeof...(I)> lengths { length_of(f, l, ctx, std::get<I>(_alternatives))... };
                return std::distance(lengths.begin(), std::max_element(lengths.begin(), lengths.end()));
            }

        template <typename It, typename Ctx, typename Other, typename Attr, size_t... I>
        bool dispatch(It& f, It l, Ctx& ctx, Other const& other, Attr& attr, size_t targetIdx, std::index_sequence<I...>) const {
            //return (real_parse<I>(f, l, ctx, other, attr, targetIdx) || ...);
            std::array<bool, sizeof...(I)> b = { real_parse<I>(f, l, ctx, other, attr, targetIdx)... };

            return std::accumulate(b.begin(), b.end(), false, std::logical_or<bool>());
        }

        template <size_t Idx, typename It, typename Ctx, typename Other, typename Attr>
        bool real_parse(It& f, It l, Ctx& ctx, Other const& other, Attr& attr, size_t targetIdx) const {
            if (targetIdx != Idx)
                return false;

            return std::get<Idx>(_alternatives).parse(f, l, ctx, other, attr);
        }

        std::tuple<Parsers...> _alternatives;
    };

    template <typename... Ps>
        longest_parser<Ps...> longest(Ps... p) { return {x3::as_parser(p)...}; }
}

int main() {
    auto id        = x3::rule<void, std::string> {} = x3::lexeme [ x3::char_("a-zA-Z_") >> *x3::char_("a-zA-Z0-9_") ];
    auto qualified = x3::rule<void, string_vec>  {} = id % "::";

#define TEST_CASE(label, input, rule) parse(label, input, #rule, rule)
    TEST_CASE("unqualified"                , "willy"                , parser::longest(id, x3::int_, x3::double_));
    TEST_CASE("unqualified with whitespace", " willy \t"            , parser::longest(id, x3::int_, x3::double_));
    TEST_CASE("integral or number"         , "123.78::anton::lutz"  , parser::longest(id, x3::int_, x3::double_));
    TEST_CASE("qualified"                  , "willy anton::lutz"    , parser::longest(id, x3::int_, x3::double_));
    TEST_CASE("qualified with whitespace"  , "willy \tanton::lutz"  , parser::longest(id, x3::int_, x3::double_));

    TEST_CASE("unqualified"                , "willy"                , parser::longest(id, x3::int_, x3::double_, qualified));
    TEST_CASE("unqualified with whitespace", " willy \t"            , parser::longest(id, x3::int_, x3::double_, qualified));
    TEST_CASE("integral or number"         , "123.78::anton::lutz"  , parser::longest(id, x3::int_, x3::double_, qualified));
    TEST_CASE("qualified"                  , "willy::anton::lutz"   , parser::longest(id, x3::int_, x3::double_, qualified));
    TEST_CASE("qualified with whitespace"  , "willy ::\tanton::lutz", parser::longest(id, x3::int_, x3::double_, qualified));

    TEST_CASE("unqualified"                , "willy"                , parser::longest(x3::int_, x3::double_, qualified));
    TEST_CASE("unqualified with whitespace", " willy \t"            , parser::longest(x3::int_, x3::double_, qualified));
    TEST_CASE("integral or number"         , "123.78::anton::lutz"  , parser::longest(x3::int_, x3::double_, qualified));
    TEST_CASE("qualified"                  , "willy::anton::lutz"   , parser::longest(x3::int_, x3::double_, qualified));
    TEST_CASE("qualified with whitespace"  , "willy ::\tanton::lutz", parser::longest(x3::int_, x3::double_, qualified));
}

打印

-------------------------
unqualified
Rule:     parser::longest(id, x3::int_, x3::double_)
Parsing: 'willy'
Parsed willy
-------------------------
unqualified with whitespace
Rule:     parser::longest(id, x3::int_, x3::double_)
Parsing: ' willy    '
Parsed willy
-------------------------
integral or number
Rule:     parser::longest(id, x3::int_, x3::double_)
Parsing: '123.78::anton::lutz'
Parsed 123.78
EOI not reached. Unparsed: '::anton::lutz'
-------------------------
qualified
Rule:     parser::longest(id, x3::int_, x3::double_)
Parsing: 'willy anton::lutz'
Parsed willy
EOI not reached. Unparsed: 'anton::lutz'
-------------------------
qualified with whitespace
Rule:     parser::longest(id, x3::int_, x3::double_)
Parsing: 'willy     anton::lutz'
Parsed willy
EOI not reached. Unparsed: 'anton::lutz'
-------------------------
unqualified
Rule:     parser::longest(id, x3::int_, x3::double_, qualified)
Parsing: 'willy'
Parsed willy
-------------------------
unqualified with whitespace
Rule:     parser::longest(id, x3::int_, x3::double_, qualified)
Parsing: ' willy    '
Parsed willy
-------------------------
integral or number
Rule:     parser::longest(id, x3::int_, x3::double_, qualified)
Parsing: '123.78::anton::lutz'
Parsed 123.78
EOI not reached. Unparsed: '::anton::lutz'
-------------------------
qualified
Rule:     parser::longest(id, x3::int_, x3::double_, qualified)
Parsing: 'willy::anton::lutz'
Parsed [willy][anton][lutz]
-------------------------
qualified with whitespace
Rule:     parser::longest(id, x3::int_, x3::double_, qualified)
Parsing: 'willy ::  anton::lutz'
Parsed [willy][anton][lutz]
-------------------------
unqualified
Rule:     parser::longest(x3::int_, x3::double_, qualified)
Parsing: 'willy'
Parsed [willy]
-------------------------
unqualified with whitespace
Rule:     parser::longest(x3::int_, x3::double_, qualified)
Parsing: ' willy    '
Parsed [willy]
-------------------------
integral or number
Rule:     parser::longest(x3::int_, x3::double_, qualified)
Parsing: '123.78::anton::lutz'
Parsed 123.78
EOI not reached. Unparsed: '::anton::lutz'
-------------------------
qualified
Rule:     parser::longest(x3::int_, x3::double_, qualified)
Parsing: 'willy::anton::lutz'
Parsed [willy][anton][lutz]
-------------------------
qualified with whitespace
Rule:     parser::longest(x3::int_, x3::double_, qualified)
Parsing: 'willy ::  anton::lutz'
Parsed [willy][anton][lutz]

注意不同的结果取决于替代方案中的解析器表达式。

【讨论】:

  • 哇,谢谢。是的,我知道,X3 让生活更轻松。已经调查了一下。但对我来说,qi 2 是针对特定问题领域的复杂元编程的杰作。我需要一个问题域来激发动力。我永远不会仅仅因为可以学习元编程。 ;) 所以,在我深入探讨 X3 的“汽车易用性”之前,我先介绍一下 qi 2 的细节。而且,X3 还不稳定,所以对于不想处于“前沿”的客户来说,这不是一个机会。 ;) 无论如何,我试图创建一个例子来激发这个想法。显然不是最好的。
  • 听起来很棒的计划。祝你好运!
  • 我在这里草草写下我的一些发现,省略了“修改气核”部分。我不想只是问一些愚蠢的问题来偷走人们的时间。见MxcQitoo on GitHub。在这里mailing list
  • 更好的问题是:如何实现一个解析器组件,它充当任意数量的运算符?
  • 我为 qi 创建了一个拉取请求,以支持使用函数样式语法的 n 元运算符。 Joel 喜欢该实施,但要求我提出社区意见。您是少数对 qi 和 X3 有深入了解的人之一。我可以请你评估github.com/boostorg/spirit/pull/200吗?值钱吗?
猜你喜欢
  • 1970-01-01
  • 1970-01-01
  • 1970-01-01
  • 1970-01-01
  • 1970-01-01
  • 1970-01-01
  • 1970-01-01
  • 2011-01-26
  • 1970-01-01
相关资源
最近更新 更多