使用 Boost.Spirit 解析混合数据类型的 OBJ 文件？

Question

我确实有一个如下所示的 OBJ 文件：

# This file uses centimeters as units for non-parametric coordinates.

# first element
v -0.017050 -0.017928 0.005579
v -0.014504 -0.017928 0.010577
   .
   .
v -0.000000 -0.017928 0.017967

vt 0.397581 0.004762
vt 0.397544 0.034263
   .
   .
vt 0.397507 0.063764

vn -0.951057 0.000000 0.309016
vn -0.809017 0.000000 0.587785
   .
   .
vn -0.587785 0.000000 0.809017

f 1/1/1 2/2/2 21/4/3
f 21/4/3 2/2/2 22/3/4
   .
   .
f 3/5/5 4/7/7 23/6/6


# second element
v -0.014504 0.017928 -0.010499
v -0.017050 0.017928 -0.005501
   .
   .
v -0.000000 0.017928 0.000039

vt 0.063001 0.262615
vt 0.073837 0.268136
   .
   .
vt 0.089861 0.299584

vn 0.000000 1.000000 -0.000002
vn 0.000000 1.000000 -0.000002
   .
   .
vn 0.000000 1.000000 -0.000000

f 36/80/78 37/81/79 42/66/64
f 37/81/79 38/82/80 42/66/64
   .
   .
f 40/84/82 21/64/62 42/66/64

# third element
   .
   .
etc.

该文件包含四种不同类型的记录，每一种都以v、vt、vn 或f 开头。 我想知道在以下情况下如何使用 boost-spirit 解析这样的文件：

int main(int argc, char **argv) {

    std::vector<double> positions;
    std::vector<double> texcoords;
    std::vector<double> normals;
    std::vector<uint32_t> faces;


    // 1- store each record begins with 'v' into 'positions'
    // 2- store each record begins with 'vt' into 'texcoords'
    // 3- store each record begins with 'vn' into 'normals'
    // 4- store each record begins with 'f' into 'faces'.  The integers are separated either with '/' or a blank-space (Each record with three groups, each group with three integers)
    // 5- ignore each record begins with '#'
    // 6- ignoring all empty lines

    return 0;
}

使用 boost-spirit 最有效的方法是什么？

Answer 1

好的，所以我开始在 Spirit X3 中解释这一点，换个思路。

AST

一如既往，第一件事。我认为我们可以更精确地了解数据结构以匹配数据本身。

这将使它更有用且不易出错。

using Position = std::tuple<double, double, double>;
using Coords   = std::tuple<double, double>;
using Normal   = std::tuple<double, double, double>;
//using Face3    = std::array<uint32_t, 3>;
//using Face9    = std::array<Face3, 3>; // 9
using Face9    = std::vector<uint32_t>; // hmmm

struct Element {
    std::vector<Position> positions;
    std::vector<Coords>   texcoords;
    std::vector<Normal>   normals;
    std::vector<Face9>    faces;
};

struct OBJ {
    std::vector<Element> elements;
};

我很想将std::array 用于固定长度序列（Face3/Face9），但不能轻易使其工作，所以在这里我像您一样选择了更灵活的vector。

解析器

这是一个 1:1 的映射：

// 1- store each record begins with 'v' into 'positions'
auto pos
    = rule<struct _pos, Position> {"pos"}
    = "v" >> double_ >> double_ >> double_;

// 2- store each record begins with 'vt' into 'texcoords'
auto tex
    = rule<struct _tex, Coords> {"tex"}
    = "vt" >> double_ >> double_;

// 3- store each record begins with 'vn' into 'normals'
auto normals
    = rule<struct _norm, Normal> {"normal"}
    = "vn" >> double_ >> double_ >> double_;

// 4- store each record begins with 'f' into 'faces'.
//    The integers are separated either with '/' or a blank-space (Each
//    record with three groups, each group with three integers)

auto face3
    = rule<struct _face3, Face9> {"face3"}
    = uint_ >> '/' >> uint_ >> '/' >> uint_;
auto faces
    = rule<struct _faces, Face9> {"faces"}
    = "f" >> repeat(3) [ face3 ];

我们不关心评论/空格，因为我们将使用一个船长：

// 5- ignore each record begins with '#'
// 6- ignoring all empty lines
auto skipper = blank | '#' >> *(char_ - eol) >> (eol|eoi);

现在，让我们跳到最后阶段：主要的解析器规则：

auto OBJ = *skip(skipper) [
    *eol >> &pos [ new_element ] >>
    lines_of(pos,     &Element::positions) >>
    lines_of(tex,     &Element::texcoords) >>
    lines_of(normals, &Element::normals) >>
    lines_of(faces,   &Element::faces)
];

如您所见，我再次假设了更多限制（因为示例文件中的 cmets 建议出现单独的元素，并且各部分按顺序出现）。

为了解析重复的行，我们有这个工厂：

auto lines_of = [](auto p, auto member) {
    return *(p [ push(member) ] >> (eoi|+eol));
};

push 操作是按功能生成的：

auto push = [](auto member) {
    return [member](auto& ctx) {
        auto& data = get<OBJ>(ctx);
        auto& v = std::invoke(member, data.elements.back());
        v.push_back(std::move(_attr(ctx)));
    };
};

我们将从解析器上下文中获取 OBJ 对象

使用 OBJ 上下文调用解析器

main 代码如下所示：

#include <iostream>
#include <iomanip>
int main() {
    std::ifstream ifs("input.txt");
    boost::spirit::istream_iterator f(ifs >> std::noskipws), l;

    OBJ data;
    if (x3::parse(f, l, x3::with<OBJ>(data) [ Parser::OBJ ])) {
        std::cout << "Yay, " << data.elements.size() << " elements\n";

        for (auto& el : data.elements) {
            dump(el);
        }
    } else {
        std::cout << "Failed\n";
    }

    if (f!=l) {
        std::cout << "Remaining unparsed: " << std::quoted(std::string(f,l)) << "\n";
    }
}

就是这样。对于给定的输入，打印：

Yay, 2 elements

奖励：调试输出

如果你有{fmt} 可用，用-DHAVE_FMT 编译，以获得漂亮的输出：

#ifdef HAVE_FMT
#include <fmt/printf.h>
#include <fmt/ranges.h>
void dump(Element const& el) {
    auto& [pos,tex,nrm,fac] = el;
    fmt::print("positions: {}\n"
            "texcoords: {}\n"
            "normals: {}\n"
            "faces: {}\n\n", pos, tex, nrm, fac);
}
#else
void dump(Element const&) { }
#endif

打印：

Yay, 2 elements
positions: {(-0.01705, -0.017928, 0.005579), (-0.014504, -0.017928, 0.010577), (-0.0, -0.017928, 0.017967)}
texcoords: {(0.397581, 0.004762), (0.397544, 0.034263), (0.397507, 0.063764)}
normals: {(-0.951057, 0.0, 0.309016), (-0.809017, 0.0, 0.587785), (-0.587785, 0.0, 0.809017)}
faces: {{1, 1, 1, 2, 2, 2, 21, 4, 3}, {21, 4, 3, 2, 2, 2, 22, 3, 4}, {3, 5, 5, 4, 7, 7, 23, 6, 6}}

positions: {(-0.014504, 0.017928, -0.010499), (-0.01705, 0.017928, -0.005501), (-0.0, 0.017928, 3.9e-05)}
texcoords: {(0.063001, 0.262615), (0.073837, 0.268136), (0.089861, 0.299584)}
normals: {(0.0, 1.0, -2e-06), (0.0, 1.0, -2e-06), (0.0, 1.0, -0.0)}
faces: {{36, 80, 78, 37, 81, 79, 42, 66, 64}, {37, 81, 79, 38, 82, 80, 42, 66, 64}, {40, 84, 82, 21, 64, 62, 42, 66, 64}}

完整列表

Live On Wandbox

//#define HAVE_FMT
//#define BOOST_SPIRIT_X3_DEBUG
#include <boost/fusion/adapted/std_tuple.hpp>
#include <boost/fusion/adapted/std_array.hpp>
#include <boost/spirit/home/x3.hpp>
#include <boost/spirit/include/support_istream_iterator.hpp>
#include <fstream>

using Position = std::tuple<double, double, double>;
using Coords   = std::tuple<double, double>;
using Normal   = std::tuple<double, double, double>;
//using Face3    = std::array<uint32_t, 3>;
//using Face9    = std::array<Face3, 3>; // 9
using Face9    = std::vector<uint32_t>; // hmmm

struct Element {
    std::vector<Position> positions;
    std::vector<Coords>   texcoords;
    std::vector<Normal>   normals;
    std::vector<Face9>    faces;
};

struct OBJ {
    std::vector<Element> elements;
};

namespace x3 = boost::spirit::x3;

namespace Parser {
    using namespace x3;

    auto new_element = [](auto& ctx) {
        auto& data = get<OBJ>(ctx);
        data.elements.emplace_back();
    };

    // 1- store each record begins with 'v' into 'positions'
    auto pos
        = rule<struct _pos, Position> {"pos"}
        = "v" >> double_ >> double_ >> double_;

    // 2- store each record begins with 'vt' into 'texcoords'
    auto tex
        = rule<struct _tex, Coords> {"tex"}
        = "vt" >> double_ >> double_;

    // 3- store each record begins with 'vn' into 'normals'
    auto normals
        = rule<struct _norm, Normal> {"normal"}
        = "vn" >> double_ >> double_ >> double_;

    // 4- store each record begins with 'f' into 'faces'.
    //    The integers are separated either with '/' or a blank-space (Each
    //    record with three groups, each group with three integers)

    auto face3
        = rule<struct _face3, Face9> {"face3"}
        = uint_ >> '/' >> uint_ >> '/' >> uint_;
    auto faces
        = rule<struct _faces, Face9> {"faces"}
        = "f" >> repeat(3) [ face3 ];

    // 5- ignore each record begins with '#'
    // 6- ignoring all empty lines
    auto skipper = blank | '#' >> *(char_ - eol) >> (eol|eoi);

    auto push = [](auto member) {
        return [member](auto& ctx) {
            auto& data = get<OBJ>(ctx);
            auto& v = std::invoke(member, data.elements.back());
            v.push_back(std::move(_attr(ctx)));
        };
    };

    auto lines_of = [](auto p, auto member) {
        return *(p [ push(member) ] >> (eoi|+eol));
    };

    auto OBJ = *skip(skipper) [
        *eol >> &pos [ new_element ] >>
        lines_of(pos,     &Element::positions) >>
        lines_of(tex,     &Element::texcoords) >>
        lines_of(normals, &Element::normals) >>
        lines_of(faces,   &Element::faces)
    ];
}

#ifdef HAVE_FMT
#include <fmt/printf.h>
#include <fmt/ranges.h>
void dump(Element const& el) {
    auto& [pos,tex,nrm,fac] = el;
    fmt::print("positions: {}\n"
            "texcoords: {}\n"
            "normals: {}\n"
            "faces: {}\n\n", pos, tex, nrm, fac);
}
#else
void dump(Element const&) { }
#endif

#include <iostream>
#include <iomanip>
int main() {
    std::ifstream ifs("input.txt");
    boost::spirit::istream_iterator f(ifs >> std::noskipws), l;

    OBJ data;
    if (x3::parse(f, l, x3::with<OBJ>(data) [ Parser::OBJ ])) {
        std::cout << "Yay, " << data.elements.size() << " elements\n";

        for (auto& el : data.elements) {
            dump(el);
        }
    } else {
        std::cout << "Failed\n";
    }

    if (f!=l) {
        std::cout << "Remaining unparsed: " << std::quoted(std::string(f,l)) << "\n";
    }
}