cppjson/cppjson/parse.cpp at main · StephenChips/cppjson

409 lines (360 loc) · 13.5 KB
#include <array>
#include <cctype>
#include <stdexcept>
#include <string>
#include <utility>
#include <cstdlib>
#include <cmath>
#include <sstream>
#include <cstring>
#include <limits>
#include <cerrno>

#include "cppjson.hpp"

using std::isdigit;
using std::logic_error;
using std::pair;
using std::string;

pair<JSON, string::size_type> parseJSON(const string &input, string::size_type index);
pair<JSON, string::size_type> parseJSONObject(const string &input, string::size_type index);
pair<JSON, string::size_type> parseJSONArray(const string &input, string::size_type index);
pair<std::string, string::size_type> parseStringLiteral(const string &input, string::size_type index);
pair<JSON, string::size_type> parseJSONNumber(const string &input, string::size_type index);
string::size_type expectString(const string &input, string::size_type index, const string &keyword);
string::size_type skipWhitespaces(const string &input, string::size_type index);
inline bool isHexLiteral(const string &input, string::size_type index);
inline bool isOctalLiteral(const string &input, string::size_type index);
std::pair<char16_t, string::size_type> parseUTF16EscapedValue(const string &input, string::size_type index);
inline bool isLeadSurrogate(char16_t v);
inline bool isTrailSurrogate(char16_t v);
char32_t calculateCodepoint(char16_t leadSurroatge, char16_t trailSurrogate);
void writeAsUTF8CodeUnits(std::ostringstream &oss, char32_t codepoint);
void writeHexValueToBuffer(char buf[], const string &input, string::size_type index);

JSON parse(const string &str)
{
    auto result = parseJSON(str, 0);
    auto end = skipWhitespaces(str, result.second);
    if (end != str.size())
        throw SyntaxError();
    return result.first;
}

pair<JSON, string::size_type> parseJSON(const string &input, string::size_type index)
{
    index = skipWhitespaces(input, index);
    auto ch = input[index];
    if (ch == '{')
        return parseJSONObject(input, index);
    else if (ch == '[')
        return parseJSONArray(input, index);
    else if (ch == '"')
    {
        auto result = parseStringLiteral(input, index);
        return std::make_pair(JSON(result.first), result.second);
    }
    else if (isdigit(ch) || ch == '-')
        return parseJSONNumber(input, index);
    else if (ch == 't')
    {
        index = expectString(input, index, "true");
        return std::make_pair(JSON(true), index);
    }
    else if (ch == 'f')
    {
        index = expectString(input, index, "false");
        return std::make_pair(JSON(false), index);
    }
    else if (ch == 'n')
    {
        index = expectString(input, index, "null");
        return std::make_pair(JSON(nullptr), index);
    }
    else
        throw SyntaxError();
}

// The program looked ahead the next character first, and confirmed the what
// JSON value would be next, before it actually calls the function to parse a
// value, so the initial `index` value must points to the first character of
// a JSON value, hence we can simply skip it without checking.

pair<JSON, string::size_type> parseJSONObject(const string &input, string::size_type index)
{
    std::map<std::string, JSON> result;
    index = skipWhitespaces(input, index + 1);

    if (index == input.size())
        throw SyntaxError();
    if (input[index] == '}')
        return std::make_pair(JSON(result), index + 1);

    while (true)
    {
        index = skipWhitespaces(input, index);
        auto objectKey = parseStringLiteral(input, index);
        index = skipWhitespaces(input, objectKey.second);
        index = expectString(input, index, ":");
        index = skipWhitespaces(input, index);
        auto objectValue = parseJSON(input, index);
        index = skipWhitespaces(input, objectValue.second);

        if (input[index] == '}')
        {
            result[objectKey.first] = objectValue.first;
            return std::make_pair(JSON(result), index + 1);
        }
        else if (input[index] == ',')
        {
            result[objectKey.first] = objectValue.first;
            index++;
        }
        else
            throw SyntaxError();
    }

    throw SyntaxError();
}

pair<JSON, string::size_type> parseJSONArray(const string &input, string::size_type index)
{
    std::vector<JSON> result;
    index = skipWhitespaces(input, index + 1);

    if (index == input.size())
        throw SyntaxError();
    if (input[index] == ']')
        return std::make_pair(JSON(result), index + 1);

    while (true)
    {
        auto jsonVal = parseJSON(input, index);
        index = skipWhitespaces(input, jsonVal.second);

        if (index == input.size())
            throw SyntaxError();
        else if (input[index] == ']')
        {
            result.push_back(jsonVal.first);
            return std::make_pair(JSON(result), index + 1);
        }
        else if (input[index] == ',')
        {
            result.push_back(jsonVal.first);
            index = skipWhitespaces(input, index + 1);
        }
        else
            throw SyntaxError();
    }
}

pair<std::string, string::size_type> parseStringLiteral(const string &input, string::size_type index)
{
    // Skip the double-quote at the begining
    index++;

    string::size_type start = index;
    std::ostringstream oss;

    while (index != input.size())
    {
        if (input[index] == '\\')
        {
            if (index + 1 >= input.size())
                throw SyntaxError();

            oss << input.substr(start, index - start);

            auto ch = input[index + 1];
            if (ch == 'u')
            {
                auto v1 = parseUTF16EscapedValue(input, index);
                if (isLeadSurrogate(v1.first))
                {
                    // If the escaped value we just read is a lead surrogate, there must be a trail surrogate right after it.
                    // Together they indicate a single Unicode character.
                    //
                    // Since the value we have just parsed is a lead surrogate, we need to get its correspondent trail surrogate.

                    if (v1.second + 1 < input.size() && input[v1.second] == '\\' && input[v1.second + 1] == 'u')
                    {
                        auto v2 = parseUTF16EscapedValue(input, v1.second);
                        if (isTrailSurrogate(v2.first))
                        {
                            auto leadSurrogate = v1.first;
                            auto trailSurrogate = v2.first;
                            auto codePoint = calculateCodepoint(leadSurrogate, trailSurrogate);
                            writeAsUTF8CodeUnits(oss, codePoint);
                        }
                        else
                        {
                            // The trail surrogate isn't valid because it's value is out of range (which should be between 0xDC00 to 0xDFFF).
                            // In such case, two values are treated as two codepoints individually and will be written to the output together.
                            writeAsUTF8CodeUnits(oss, v1.first);
                            writeAsUTF8CodeUnits(oss, v2.first);
                        }
                        index = v2.second;
                    }
                    else
                    {
                        // There are no another UTF-16 escaped value after the lead surrogate, which is invalid. The program will write the
                        // lead surrogate to the output then continue.
                        writeAsUTF8CodeUnits(oss, v1.first);
                        index = v1.second;
                    }
                }
                else
                {
                    // If the first escaped value isn't a lead surrogate, we will regard it as an Unicode codepoint and will write it to the
                    // output (It might be a trail surrogate but we need not to distinguish it).
                    writeAsUTF8CodeUnits(oss, v1.first);
                    index = v1.second;
                }
            }
            else
            {
                if (ch == '\\')
                    oss << '\\';
                else if (ch == '/')
                    oss << '/';
                else if (ch == 'b')
                    oss << '\b';
                else if (ch == 'f')
                    oss << '\f';
                else if (ch == 'n')
                    oss << '\n';
                else if (ch == 'r')
                    oss << '\r';
                else if (ch == 't')
                    oss << '\t';
                else
                    throw SyntaxError();

                index += 2;
            }

            start = index;
        }
        else if (input[index] == '"')
        {
            if (start < index)
            {
                oss << input.substr(start, index - start);
            }
            return std::make_pair(oss.str(), index + 1);
        }
        else
        {
            index++;
        }
    }

    throw SyntaxError();
}

pair<JSON, string::size_type> parseJSONNumber(const string &input, string::size_type index)
{
    // correspondent regex:
    // /^-?(0|([1-9][0-9]*))(\.[0-9]+)?((e|E)(-|\+)?[0-9]+)?$/

    // We could parse a number with our own codes, but for now we will just use `std::strtod`
    // instead.

    // JSON doesn't supports hex literals and octal literals, but `strtod` can parse them,
    // so we should exclude them before parsing.
    // Hex literal is something like 0x3, 0XE2, a literal started with 0x or 0X, and Octal
    // literals is somthing like 045, 022, a literal started with a leading zero.
    //
    // In JSON, if a number's interal part's first digit is zero, it must be followed by the
    // decimal point (.), no other digits can be appeared behind it.
    if (isHexLiteral(input, index) || isOctalLiteral(input, index))
        throw SyntaxError();

    const char *cStr = input.c_str();
    char *endPtr = const_cast<char *>(cStr);

    double result = std::strtod(cStr + index, &endPtr);

    if (endPtr == cStr)
        throw SyntaxError();
    else if (result == HUGE_VAL)
        result = std::numeric_limits<double>::infinity();
    else if (errno == ERANGE)
        // Underflow, like 1.0E-1000
        result = result > 0 ? 0 : -0;

    return std::make_pair(JSON(result), endPtr - cStr);
}

string::size_type expectString(const string &input, string::size_type index, const string &str)
{
    auto i = index;
    auto j = 0u;

    while (i != input.size() && j != str.size())
    {
        if (input[i] != str[j])
            break;

        i++, j++;
    }

    if (j != str.size())
        throw SyntaxError();

    return i;
}

string::size_type skipWhitespaces(const string &input, string::size_type index)
{
    while (index != input.size())
    {
        auto ch = input[index];
        if (ch == ' ' || ch == '\n' || ch == '\r' || ch == '\t')
            index++;
        else
            break;
    }
    return index;
}

inline bool isHexLiteral(const string &input, string::size_type index)
{
    return index + 1 < input.size() && (input[index + 1] == 'x' || input[index + 1] == 'X');
}

inline bool isOctalLiteral(const string &input, string::size_type index)
{
    return index + 1 < input.size() && input[index] == '0' && isdigit(input[index + 1]);
}

std::pair<char16_t, string::size_type> parseUTF16EscapedValue(const string &input, string::size_type index)
{
    char digitBuffer[5];
    writeHexValueToBuffer(digitBuffer, input, index + 2);
    char16_t v = strtoul(digitBuffer, nullptr, 16);
    return std::make_pair(v, index + 6);
}

inline bool isLeadSurrogate(char16_t v)
{
    return v >= 0xD800u && v <= 0xDBFFu;
}

inline bool isTrailSurrogate(char16_t v)
{
    return v >= 0xDC00u && v <= 0xDFFFu;
}

char32_t calculateCodepoint(char16_t leadSurroatge, char16_t trailSurrogate)
{
    return (static_cast<char32_t>(leadSurroatge - 0xD800u) << 10) + (trailSurrogate - 0xDC00u) + 0x10000;
}

void writeAsUTF8CodeUnits(std::ostringstream &oss, char32_t codepoint)
{
    if (codepoint <= 0x00007F)
    {
        oss << static_cast<char>(codepoint);
    }
    else if (codepoint <= 0x0007FF)
    {
        oss << static_cast<char>(0b11000000 | (codepoint >> 6));
        oss << static_cast<char>(0b10000000 | (codepoint & 0b111111));
    }
    else if (codepoint <= 0x00FFFF)
    {
        // Strictly speaking, codepoints 0x00D800..0x00DFFF should only be used in UTF-16 text as
        // surrogate pair. This is how JSON escapes unicode that takes more than two bytes.
        // But in case we met an unpaired surrogate, instead throwing a syntax error, we will still
        // translate it to correspondent UTF-8 codeunits as if it was a normal unicode character
        // for this is how JSON.parse(...) behaves in browser.

        oss << static_cast<char>(0b11100000 | (codepoint >> 12));
        oss << static_cast<char>(0b10000000 | ((codepoint >> 6) & 0b111111));
        oss << static_cast<char>(0b10000000 | (codepoint & 0b111111));
    }
    else if (codepoint <= 0x10FFFF)
    {
        oss << static_cast<char>(0b11110000 | (codepoint >> 18));
        oss << static_cast<char>(0b10000000 | (codepoint >> 12) & 0b111111);
        oss << static_cast<char>(0b10000000 | (codepoint >> 6) & 0b111111);
        oss << static_cast<char>(0b10000000 | (codepoint & 0b111111));
    }
    else
    {
        throw std::range_error("The codepoint is out of range.");
    }
}

void writeHexValueToBuffer(char xdigits[], const string &input, string::size_type index)
{
    if (index + 4 >= input.size())
        throw SyntaxError();

    for (auto i = 0; i < 4; i++)
    {
        if (!std::isxdigit(input[index + i]))
            throw SyntaxError();
    }

    for (auto i = 0; i < 4; i++)
    {
        xdigits[i] = input[index + i];
    }

    xdigits[4] = '\0';
}
Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Expand file tree

Search code, repositories, users, issues, pull requests...

FilesExpand file tree

parse.cpp

Latest commit

History

parse.cpp

File metadata and controls

Expand file tree
