问题描述
我正在使用 boost 精神来解析一些文本。为此,我有两个语法。第一个将字符串解析为结构,第二个将语法作为模板参数并使用它来解析数据序列。第二个解析器应该足够灵活以处理其他语法返回类型。由于原始解析器太大而无法作为最小示例,因此我尽可能地减少了代码,留下了一些不会解析任何内容的东西,但仍然导致相同的编译错误:(Code on Coliru )
#include <boost/fusion/include/adapt_struct.hpp>
#include <boost/spirit/include/qi.hpp>
#include <vector>
namespace ascii = boost::spirit::ascii;
namespace qi = boost::spirit::qi;
struct Struct1
{
float f;
};
BOOST_FUSION_ADAPT_STRUCT(
Struct1,(float,f))
struct Struct2
{
float f;
int i;
};
BOOST_FUSION_ADAPT_STRUCT(
Struct2,f)
(int,i))
template<typename Iterator,typename Result>
class ElementParser : public qi::grammar<Iterator,Result(),ascii::space_type>
{
public:
using ValueType = Result;
ElementParser() : ElementParser::base_type(element) {}
private:
qi::rule<Iterator,ascii::space_type> element;
};
template<typename Iterator,typename ElementParser,typename Element = typename ElementParser::ValueType>
class SP : public qi::grammar<Iterator,std::vector<Element>(),ascii::space_type>
{
public:
SP()
: SP::base_type(sequence)
{
sequence %= simpleVector % ',';
// The simpleVector hack is really needed because of some other parsing
// stuff,that is going on,but has been left out here.
simpleVector %= qi::repeat(1)[simple];
}
private:
using Rule = qi::rule<Iterator,ascii::space_type>;
Rule sequence;
Rule simpleVector;
ElementParser simple;
};
void sequencetest()
{
using Iterator = std::string::const_iterator;
SP<Iterator,qi::uint_parser<>,std::size_t> uintParser; // OK
SP<Iterator,ElementParser<Iterator,float>> floatParser; // OK
SP<Iterator,std::vector<float>>> vectorParser; // OK
// error: invalid static_cast from type 'const std::vector<Struct1,std::allocator<Struct1> >' to type 'element_type' {aka 'float'}
SP<Iterator,Struct1>> struct1Parser;
// error: no matching function for call to 'Struct2::Struct2(const std::vector<Struct2,std::allocator<Struct2> >&)'
SP<Iterator,Struct2>> struct2Parser;
}
只要我使用简单类型或向量作为 ElementParser
的返回类型,一切正常,但是一旦我解析为一个结构(它本身工作正常),序列解析器 SP
似乎尝试了一些奇怪的分配。为什么struct版本会导致编译错误?
解决方法
我认为您绕过了古老的单元素序列兼容性规则。特别是 Struct1 确实适合作为单元素序列。
但是,在您的代码中,我可以通过删除不必要的 repeat(1)[]
装置轻松使其工作:
#include <boost/fusion/include/adapt_struct.hpp>
#include <boost/spirit/include/qi.hpp>
namespace ascii = boost::spirit::ascii;
namespace qi = boost::spirit::qi;
struct Struct1 { float f; };
struct Struct2 { float f; int i; };
BOOST_FUSION_ADAPT_STRUCT(Struct1,f)
BOOST_FUSION_ADAPT_STRUCT(Struct2,f,i)
template <typename Iterator,typename Result>
class ElementParser
: public qi::grammar<Iterator,Result(),ascii::space_type> {
public:
using ValueType = Result;
ElementParser() : ElementParser::base_type(element) {
}
private:
qi::rule<Iterator,ascii::space_type> element;
};
template <typename Iterator,typename ElementParser,typename Element = typename ElementParser::ValueType>
class SequenceParser
: public qi::grammar<Iterator,std::vector<Element>(),ascii::space_type> {
public:
SequenceParser() : SequenceParser::base_type(sequence) {
sequence = simple % ',';
}
private:
qi::rule<Iterator,ascii::space_type> sequence;
ElementParser simple;
};
void sequenceTest() {
using It = std::string::const_iterator;
SequenceParser<It,qi::uint_parser<>,std::size_t> uintParser; // OK
SequenceParser<It,ElementParser<It,float>> floatParser; // OK
SequenceParser<It,std::vector<float>>>
vectorParser; // OK
SequenceParser<It,Struct1>> struct1Parser;
SequenceParser<It,Struct2>> struct2Parser;
}
int main() {
sequenceTest();
}
奖励:JustParseIt
魔法函数
请注意,在某种程度上,您似乎正在重新启动库设计。看看qi::auto_
。
添加这里的想法:
例如通过专门化 trait 来制作逗号分隔的序列解析器:
template <typename... T>
struct create_parser<std::vector<T...>> : comma_separated_sequence {};
struct comma_separated_sequence {
using type = decltype(qi::copy(qi::auto_ % ','));
static type call() { return qi::copy(qi::auto_ % ','); }
};
现在您可以实现一个适用于 /the world/ 的 JustParseIt
函数:
bool JustParseIt(std::string_view input,auto& val) {
return qi::phrase_parse(input.begin(),input.end(),qi::auto_,qi::space,val);
}
你会惊讶地看到它解析的内容:
#include <boost/fusion/include/adapted.hpp>
#include <boost/fusion/include/io.hpp>
#include <boost/spirit/include/qi.hpp>
#include <fmt/ranges.h>
#include <fmt/ostream.h>
namespace qi = boost::spirit::qi;
namespace MyLib {
struct Struct1 { float f; };
struct Struct2 { float f; int i; };
using boost::fusion::operator<<;
}
BOOST_FUSION_ADAPT_STRUCT(MyLib::Struct1,f)
BOOST_FUSION_ADAPT_STRUCT(MyLib::Struct2,i)
namespace {
struct comma_separated_sequence {
using type = decltype(qi::copy(qi::auto_ % ','));
static type call() { return qi::copy(qi::auto_ % ','); }
};
}
namespace boost::spirit::traits {
template <typename... T>
struct create_parser<std::list<T...>> : comma_separated_sequence {};
template <typename... T>
struct create_parser<std::vector<T...>> : comma_separated_sequence {};
}
bool JustParseIt(std::string_view input,auto& val) {
#ifdef BOOST_SPIRIT_DEBUG
using It = decltype(input.begin());
using Skipper = qi::space_type;
using Attr = std::decay_t<decltype(val)>;
static qi::rule<It,Attr(),Skipper> parser = qi::auto_;
BOOST_SPIRIT_DEBUG_NODE(parser);
return qi::phrase_parse(input.begin(),parser,val);
#else
return qi::phrase_parse(input.begin(),val);
#endif
}
int main() {
using namespace MyLib;
std::cerr << std::boolalpha; // make debug easier to read
float f;
JustParseIt("3.1415",f);
uint64_t u;
JustParseIt("00897823",u);
Struct1 s1;
JustParseIt("3.1415",s1);
Struct2 s2;
JustParseIt("3.1415 00897823",s2);
std::list<float> floats;;
JustParseIt("1.2,3.4",floats);
std::list<Struct1> list1;
JustParseIt("1.2",list1);
JustParseIt("1.2,-inf,9e10,NaN",list1);
std::vector<boost::variant<Struct2,bool> > variants;
JustParseIt("true,9e10 123,NaN 234,false,false",variants);
std::vector<Struct2> vec2;
JustParseIt("9e10 123,NaN 234",vec2);
// this is pushing it - for lack of structurual syntax
std::vector<std::tuple<bool,Struct1,std::vector<Struct2>>> insane;
JustParseIt("true 3.14 1e1 1,2e2 2,3e3 3,false +inf 4e4 4",insane);
fmt::print("float f: {}\n"
"uint64_t u: {}\n"
"std::list<float> floats: {}\n"
"std::list<Struct1> list1: {}\n"
"std::vector<Struct2> vec2: {}\n"
"Struct1 s1: {}\n"
"Struct2 s2: {}\n"
"std::vector<boost::variant<Struct2,bool> > variants: {}\n"
"std::vector<std::tuple<bool,std::vector<Struct2>>> "
"insane: {}\n",u,floats,list1,vec2,s1,s2,variants,insane);
}
印刷品
float f: 3.1415
uint64_t u: 897823
std::list<float> floats: {1.2,3.4}
std::list<Struct1> list1: {(1.2),(1.2),(-inf),(9e+10),(nan)}
std::vector<Struct2> vec2: {(9e+10 123),(nan 234)}
Struct1 s1: (3.1415)
Struct2 s2: (3.1415 897823)
std::vector<boost::variant<Struct2,bool> > variants: {1,(9e+10 123),(nan 234),0}
std::vector<std::tuple<bool,std::vector<Struct2>>> insane: {(true,(3.14),{(10 1),(200 2),(3000 3)}),(false,(inf),{(40000 4)})}
请注意,您可以 define BOOST_SPIRIT_DEBUG
将解析器调试到 stderr,例如
<parser>
<try>true,NaN </try>
<success></success>
<attributes>[[true,[9e+10,123],[nan,234],false]]</attributes>
</parser>
,
这是一个更短的示例,演示了相同的问题 (compiler explorer):
#include <boost/fusion/adapted/std_tuple.hpp>
#include <boost/spirit/include/qi.hpp>
#include <vector>
#include <tuple>
namespace ascii = boost::spirit::ascii;
namespace qi = boost::spirit::qi;
void test()
{
using Iterator = std::string::const_iterator;
// OK
qi::rule<Iterator,std::vector<int>(),ascii::space_type> vecI_src;
qi::rule<Iterator,ascii::space_type> vecI_dst = *vecI_src;
// error: no matching function for call to 'std::tuple<int,float>::tuple(const std::vector<std::tuple<int,float> >&)'
qi::rule<Iterator,std::vector<std::tuple<int,float>>(),ascii::space_type> vecT_src;
qi::rule<Iterator,ascii::space_type> vecT_dst = *vecT_src;
}
我认为,问题在于,向量和元组在底层 boost::fusion
库中的处理方式非常相似,因此在展平向量时,boost::fusion
超出了目标并且分配失败。 (可能通过某种 SFINAE 机制。)现在,扁平化向量不起作用,右侧 tuple
解析器的合成属性是 vector<vector<tuple<int,float>>>
类型,而不是预期的 {{ 1}}。
知道这一点后,我发现的(不是很漂亮)解决方案(对于原始示例)是为两种预期形式手动创建赋值函数重载:
vector<tuple<int,float>>
并通过 static
void flattenAndAppend(std::vector<Element>& into,std::vector<std::vector<Element>> const& vector)
{
for(auto const& subvector: vector)
{
into.insert(into.end(),subvector.begin(),subvector.end());
}
}
static
void flattenAndAppend(std::vector<Element>& into,std::vector<Element> const& vector)
{
into.insert(into.end(),vector.begin(),vector.end());
}
函数在语义操作中调用这些:
boost::phoenix
这是整个工作示例 (compiler explorer):
ph::function append = [](auto& into,auto const& a1)
{
flattenAndAppend(into,a1);
};
sequence = (simpleVector % ',')[append(qi::_val,ql::_1)];