c++ - 仅在 Boost.Spirit.Qi 语法初始化的优化构建中出现段错误

Question

我在使用 Spirit Qi 编写解析器时遇到了一个奇怪的问题:我在某个地方有一个错误导致 -O 优化崩溃，但并非没有优化。它在语法的构造函数中崩溃:

template <typename Iterator>
struct math_expression_grammar : qi::grammar<Iterator, std::string()>
{
    qi::rule<Iterator, std::string()>
        expression,
        term,
        factorial,
        factor,
        pexpression,
        pfactor,
        nfactor,
        number;

    math_expression_grammar():
        math_expression_grammar::base_type(expression)
    {
        using namespace boost::spirit;
        using namespace boost::spirit::ascii;
        namespace sp = boost::spirit;
        namespace ph = boost::phoenix;

        auto sum = term[_val = sp::_1] >> lit('+') >> term[_val += sp::_1, _val += "+ "];
        auto difference = term[_val = sp::_1] >> lit('-') >> term[_val += sp::_1, _val += "- "];
        auto product = factor[_val = sp::_1] >> lit('*') >> factor[_val += sp::_1, _val += "* "];
        auto dividend = factor[_val = sp::_1] >> lit('/') >> factor[_val += sp::_1, _val += "/ "];

        expression = sum |
                     difference |
                     term;

        term = product |
               dividend |
               factor;

        pfactor = factorial.alias();
        nfactor = (lit('-') >> pfactor)[_val = sp::_1 + "n "];
        factor = nfactor | pfactor;

        pexpression = lit('(') >> expression >> lit(')');

        factorial = (pexpression | number)[_val = sp::_1] >> -lit('!')[_val += "! "];

        number = sp::double_[_val = ph::bind(stringize<double>, sp::_1) + ' '];
    }
};

我必须在 Windows 64 位的 TDM GCC 4.8.2 和 Arch Linux 64 位的 GCC 4.9.0 上测试它；两者都有同样的问题。这是 Valgrind 跟踪的相关部分，已启用优化:

==15671== Use of uninitialised value of size 8
==15671==    at 0x4040DA: void boost::spirit::qi::rule<__gnu_cxx::__normal_iterator<char*, std::string>, std::string (), boost::spirit::unused_type, boost::spirit::unused_type, boost::spiri
t::unused_type>::define<mpl_::bool_<false>, boost::proto::exprns_::expr<boost::proto::tagns_::tag::bitwise_or, boost::proto::argsns_::list2<boost::proto::exprns_::expr<boost::proto::tagns_:
:tag::bitwise_or, boost::proto::argsns_::list2<boost::proto::exprns_::expr<boost::proto::tagns_::tag::shift_right, boost::proto::argsns_::list2<boost::proto::exprns_::expr<boost::proto::tag
ns_::tag::shift_right, boost::proto::argsns_::list2<boost::proto::exprns_::expr<boost::proto::tagns_::tag::subscript, boost::proto::argsns_::list2<boost::spirit::qi::rule<__gnu_cxx::__norma
l_iterator<char*, std::string>, std::string (), boost::spirit::unused_type, boost::spirit::unused_type, boost::spirit::unused_type>&, boost::proto::exprns_::expr<boost::proto::tagns_::tag::
terminal, boost::proto::argsns_::term<boost::phoenix::actor<boost::phoenix::composite<boost::phoenix::assign_eval, boost::fusion::vector<boost::spirit::attribute<0>, boost::spirit::argument
<0>, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_> > > cons
t&>, 0l> >, 2l> const&, boost::proto::exprns_::expr<boost::proto::tagns_::tag::terminal, boost::proto::argsns_::term<boost::spirit::terminal_ex<boost::spirit::tag::lit, boost::fusion::vecto
r1<char> > >, 0l> const&>, 2l> const&, boost::proto::exprns_::expr<boost::proto::tagns_::tag::subscript, boost::proto::argsns_::list2<boost::spirit::qi::rule<__gnu_cxx::__normal_iterator<ch
ar*, std::string>, std::string (), boost::spirit::unused_type, boost::spirit::unused_type, boost::spirit::unused_type>&, boost::proto::exprns_::expr<boost::proto::tagns_::tag::terminal, boo
st::proto::argsns_::term<boost::phoenix::actor<boost::phoenix::composite<boost::phoenix::sequence_eval, boost::fusion::vector<boost::phoenix::composite<boost::phoenix::plus_assign_eval, boo
st::fusion::vector<boost::spirit::attribute<0>, boost::spirit::argument<0>, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boo
st::fusion::void_, boost::fusion::void_, boost::fusion::void_> >, boost::phoenix::composite<boost::phoenix::plus_assign_eval, boost::fusion::vector<boost::spirit::attribute<0>, boost::phoen
ix::value<char const*>, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusi
on::void_> >, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_>
 > > const&>, 0l> >, 2l> const&>, 2l>&, boost::proto::exprns_::expr<boost::proto::tagns_::tag::shift_right, boost::proto::argsns_::list2<boost::proto::exprns_::expr<boost::proto::tagns_::ta
g::shift_right, boost::proto::argsns_::list2<boost::proto::exprns_::expr<boost::proto::tagns_::tag::subscript, boost::proto::argsns_::list2<boost::spirit::qi::rule<__gnu_cxx::__normal_itera
tor<char*, std::string>, std::string (), boost::spirit::unused_type, boost::spirit::unused_type, boost::spirit::unused_type>&, boost::proto::exprns_::expr<boost::proto::tagns_::tag::termina
l, boost::proto::argsns_::term<boost::phoenix::actor<boost::phoenix::composite<boost::phoenix::assign_eval, boost::fusion::vector<boost::spirit::attribute<0>, boost::spirit::argument<0>, bo
ost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_> > > const&>, 0l
> >, 2l> const&, boost::proto::exprns_::expr<boost::proto::tagns_::tag::terminal, boost::proto::argsns_::term<boost::spirit::terminal_ex<boost::spirit::tag::lit, boost::fusion::vector1<char
> > >, 0l> const&>, 2l> const&, boost::proto::exprns_::expr<boost::proto::tagns_::tag::subscript, boost::proto::argsns_::list2<boost::spirit::
==15671==    by 0x404BD9: math_expression_grammar<__gnu_cxx::__normal_iterator<char*, std::string> >::math_expression_grammar() (in /home/collin/programming/parser/parser)
==15671==    by 0x401E6A: main (in /home/collin/programming/parser/parser)
==15671==  Uninitialised value was created by a stack allocation
==15671==    at 0x404672: math_expression_grammar<__gnu_cxx::__normal_iterator<char*, std::string> >::math_expression_grammar() (in /home/collin/programming/parser/parser)

这是关闭优化的整个日志:

==15686== Memcheck, a memory error detector
==15686== Copyright (C) 2002-2013, and GNU GPL'd, by Julian Seward et al.
==15686== Using Valgrind-3.9.0 and LibVEX; rerun with -h for copyright info
==15686== Command: ./parser
==15686==

==15686==
==15686== FILE DESCRIPTORS: 3 open at exit.
==15686== Open file descriptor 2: /dev/pts/5
==15686==    <inherited from parent>
==15686==
==15686== Open file descriptor 1: /dev/pts/5
==15686==    <inherited from parent>
==15686==
==15686== Open file descriptor 0: /dev/pts/5
==15686==    <inherited from parent>
==15686==
==15686==
==15686== HEAP SUMMARY:
==15686==     in use at exit: 0 bytes in 0 blocks
==15686==   total heap usage: 14 allocs, 14 frees, 776 bytes allocated
==15686==
==15686== All heap blocks were freed -- no leaks are possible
==15686==
==15686== For counts of detected and suppressed errors, rerun with: -v
==15686== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 1 from 1)

两次测试之间我的代码没有任何变化。

我在实际调试问题时遇到了麻烦，因为该错误仅在 -O 优化或更高优化时出现，但不幸的是 -Og 不会出现。我确实怀疑这个错误可能在 Boost.Spirit 中，但我也很不确定；我没有发现我的代码有任何问题，但我可能遗漏了一些东西或错误地使用了 Spirit。更有经验的人可以指出我正确的方向吗？

这是完整的可编译代码:

#include <iostream>
#include <string>
#include <vector>
#include <stack>
#include <algorithm>
#include <utility>

#include <boost/mpl/vector.hpp>
#include <boost/spirit/include/qi.hpp>
#include <boost/spirit/include/qi_real.hpp>
#include <boost/spirit/include/phoenix.hpp>
#include <boost/bind.hpp>
#include <boost/ref.hpp>

namespace qi = boost::spirit::qi;

template <typename NumType, typename Iterator>
NumType inline parse_number(Iterator first, Iterator last)
{
    using namespace std;

    istringstream extractor(string(first, last));
    NumType num;
    extractor >> num;

    return num;
}

template <typename NumType = double, typename Iterator>
NumType eval_rpn(Iterator head, Iterator last)
{
    using namespace std;

    const char tokens[] = {'+', '-', '*', '/', '^', 'n', ' '};
    auto tokens_begin = begin(tokens), tokens_end = end(tokens);

    stack<NumType> num_stack;

    while(head != last)
    {
        auto next = find_first_of(head, last, tokens_begin, tokens_end);

        if(head != next) num_stack.push(parse_number<NumType>(head, next));

        if(next != last)
        {
            NumType temp;

            switch(*next)
            {
            case '+':
                temp = num_stack.top();
                num_stack.pop();
                num_stack.top() += temp;
                break;

            case '-':
                temp = num_stack.top();
                num_stack.pop();
                num_stack.top() -= temp;
                break;

            case '*':
                temp = num_stack.top();
                num_stack.pop();
                num_stack.top() *= temp;
                break;

            case '/':
                temp = num_stack.top();
                num_stack.pop();
                num_stack.top() /= temp;
                break;

            case '^':
                temp = num_stack.top();
                num_stack.pop();
                num_stack.top() = pow(num_stack.top(), temp);
                break;

            case 'n':
                num_stack.top() = -num_stack.top();
                break;

            default:
                break; // Do nothing
            }

            head = next+1;
        }
        else head = last;
    }

    return num_stack.top();
}

template <typename T>
std::string stringize(T x)
{
    return std::to_string(x);
}

template <typename Iterator>
struct math_expression_grammar : qi::grammar<Iterator, std::string()>
{
    qi::rule<Iterator, std::string()>
        expression,
        term,
        factorial,
        factor,
        pexpression,
        pfactor,
        nfactor,
        number;

    math_expression_grammar():
        math_expression_grammar::base_type(expression)
    {
        using namespace boost::spirit;
        using namespace boost::spirit::ascii;
        namespace sp = boost::spirit;
        namespace ph = boost::phoenix;

        auto sum = term[_val = sp::_1] >> lit('+') >> term[_val += sp::_1, _val += "+ "];
        auto difference = term[_val = sp::_1] >> lit('-') >> term[_val += sp::_1, _val += "- "];
        auto product = factor[_val = sp::_1] >> lit('*') >> factor[_val += sp::_1, _val += "* "];
        auto dividend = factor[_val = sp::_1] >> lit('/') >> factor[_val += sp::_1, _val += "/ "];

        expression = sum |
                     difference |
                     term;

        term = product |
               dividend |
               factor;

        pfactor = factorial.alias();
        nfactor = (lit('-') >> pfactor)[_val = sp::_1 + "n "];
        factor = nfactor | pfactor;

        pexpression = lit('(') >> expression >> lit(')');

        factorial = (pexpression | number)[_val = sp::_1] >> -lit('!')[_val += "! "];

        number = sp::double_[_val = ph::bind(stringize<double>, sp::_1) + ' '];
    }
};

int main()
{
    using namespace std;

    math_expression_grammar<string::iterator> g;

    string input;
    getline(cin, input);
    while(input.size())
    {
        auto first = input.begin(), last = input.end();
        cout << input << endl;

        string result;
        if(!boost::spirit::qi::parse(first, last, g, result))
        {
            cout << "Error at " << last - first << ":\n\t" << *first << endl;
        }
        else
        {
            cout << result << endl;
            cout << eval_rpn(result.begin(), result.end()) << endl;
        }

        getline(cin, input);
    }
}

Answer 1

您不能(安全地)在 Spirit 表达式模板上使用 auto。这导致了 UB。

将很快发布一个固定版本 ( awaiting information )。同时，请参阅:

• Assigning parsers to auto variables
• boost spirit V2 qi bug associated with optimization level