C++ ShiftJIS to UTF8 conversion

Normally I would recommend using the ICU library, but for this alone, using it is way too much overhead.

First a conversion function which takes an std::string with Shiftjis data, and returns an std::string with UTF8 (note 2019: no idea anymore if it works :))

It uses a uint8_t array of 25088 elements (25088 byte), which is used as convTable in the code. The function does not fill this variable, you have to load it from eg. a file first. The second code part below is a program that can generate the file.

The conversion function doesn't check if the input is valid ShiftJIS data.

std::string sj2utf8(const std::string &input)
{
    std::string output(3 * input.length(), ' '); //ShiftJis won't give 4byte UTF8, so max. 3 byte per input char are needed
    size_t indexInput = 0, indexOutput = 0;

    while(indexInput < input.length())
    {
        char arraySection = ((uint8_t)input[indexInput]) >> 4;

        size_t arrayOffset;
        if(arraySection == 0x8) arrayOffset = 0x100; //these are two-byte shiftjis
        else if(arraySection == 0x9) arrayOffset = 0x1100;
        else if(arraySection == 0xE) arrayOffset = 0x2100;
        else arrayOffset = 0; //this is one byte shiftjis

        //determining real array offset
        if(arrayOffset)
        {
            arrayOffset += (((uint8_t)input[indexInput]) & 0xf) << 8;
            indexInput++;
            if(indexInput >= input.length()) break;
        }
        arrayOffset += (uint8_t)input[indexInput++];
        arrayOffset <<= 1;

        //unicode number is...
        uint16_t unicodeValue = (convTable[arrayOffset] << 8) | convTable[arrayOffset + 1];

        //converting to UTF8
        if(unicodeValue < 0x80)
        {
            output[indexOutput++] = unicodeValue;
        }
        else if(unicodeValue < 0x800)
        {
            output[indexOutput++] = 0xC0 | (unicodeValue >> 6);
            output[indexOutput++] = 0x80 | (unicodeValue & 0x3f);
        }
        else
        {
            output[indexOutput++] = 0xE0 | (unicodeValue >> 12);
            output[indexOutput++] = 0x80 | ((unicodeValue & 0xfff) >> 6);
            output[indexOutput++] = 0x80 | (unicodeValue & 0x3f);
        }
    }

    output.resize(indexOutput); //remove the unnecessary bytes
    return output;
}

About the helper file: I used to have a download here, but nowadays I only know unreliable file hosters. So... either http://s000.tinyupload.com/index.php?file_id=95737652978017682303 works for you, or:

First download the "original" data from ftp://ftp.unicode.org/Public/MAPPINGS/OBSOLETE/EASTASIA/JIS/SHIFTJIS.TXT . I can't paste this here because of the length, so we have to hope at least unicode.org stays online.

Then use this program while piping/redirecting above text file in, and redirecting the binary output to a new file. (Needs a binary-safe shell, no idea if it works on Windows).

#include<iostream>
#include<string>
#include<cstdio>

using namespace std;

// pipe SHIFTJIS.txt in and pipe to (binary) file out
int main()
{
    string s;
    uint8_t *mapping; //same bigendian array as in converting function
    mapping = new uint8_t[2*(256 + 3*256*16)];

    //initializing with space for invalid value, and then ASCII control chars
    for(size_t i = 32; i < 256 + 3*256*16; i++)
    {
        mapping[2 * i] = 0;
        mapping[2 * i + 1] = 0x20;
    }
    for(size_t i = 0; i < 32; i++)
    {
        mapping[2 * i] = 0;
        mapping[2 * i + 1] = i;
    }

    while(getline(cin, s)) //pipe the file SHIFTJIS to stdin
    {
        if(s.substr(0, 2) != "0x") continue; //comment lines

        uint16_t shiftJisValue, unicodeValue;
        if(2 != sscanf(s.c_str(), "%hx %hx", &shiftJisValue, &unicodeValue)) //getting hex values
        {
            puts("Error hex reading");
            continue;
        }

        size_t offset; //array offset
        if((shiftJisValue >> 8) == 0) offset = 0;
        else if((shiftJisValue >> 12) == 0x8) offset = 256;
        else if((shiftJisValue >> 12) == 0x9) offset = 256 + 16*256;
        else if((shiftJisValue >> 12) == 0xE) offset = 256 + 2*16*256;
        else
        {
            puts("Error input values");
            continue;
        }

        offset = 2 * (offset + (shiftJisValue & 0xfff));
        if(mapping[offset] != 0 || mapping[offset + 1] != 0x20)
        {
            puts("Error mapping not 1:1");
            continue;
        }

        mapping[offset] = unicodeValue >> 8;
        mapping[offset + 1] = unicodeValue & 0xff;
    }

    fwrite(mapping, 1, 2*(256 + 3*256*16), stdout);
    delete[] mapping;
    return 0;
}

Notes:
Two-byte big endian raw unicode values (more than two byte not necessary here)
First 256 chars (512 byte) for the single byte ShiftJIS chars, value 0x20 for invalid ones.
Then 3 * 256*16 chars for the groups 0x8???, 0x9??? and 0xE???
= 25088 byte