20550e2018-03-06Pontus Östlund #include "sass.hpp" #include <string> #include <vector> #include <cstdlib> #include <cmath> #include "utf8.h" namespace Sass { namespace UTF_8 { using std::string; // naming conventions: // offset: raw byte offset (0 based) // position: code point offset (0 based) // index: code point offset (1 based or negative) // function that will count the number of code points (utf-8 characters) from the given beginning to the given end size_t code_point_count(const string& str, size_t start, size_t end) { return utf8::distance(str.begin() + start, str.begin() + end); } size_t code_point_count(const string& str) { return utf8::distance(str.begin(), str.end()); } // function that will return the byte offset at a code point position size_t offset_at_position(const string& str, size_t position) { string::const_iterator it = str.begin(); utf8::advance(it, position, str.end()); return std::distance(str.begin(), it); } // function that returns number of bytes in a character at offset size_t code_point_size_at_offset(const string& str, size_t offset) { // get iterator from string and forward by offset string::const_iterator stop = str.begin() + offset; // check if beyond boundary if (stop == str.end()) return 0; // advance by one code point utf8::advance(stop, 1, str.end()); // calculate offset for code point return stop - str.begin() - offset; } // function that will return a normalized index, given a crazy one size_t normalize_index(int index, size_t len) { long signed_len = static_cast<long>(len); // assuming the index is 1-based // we are returning a 0-based index if (index > 0 && index <= signed_len) { // positive and within string length return index-1; } else if (index > signed_len) { // positive and past string length return len; } else if (index == 0) { return 0; } else if (std::abs((double)index) <= signed_len) { // negative and within string length return index + signed_len; } else { // negative and past string length return 0; } } #ifdef _WIN32 // utf16 functions using std::wstring; // convert from utf16/wide string to utf8 string string convert_from_utf16(const wstring& utf16) { string utf8; // pre-allocate expected memory utf8.reserve(sizeof(utf16)/2); utf8::utf16to8(utf16.begin(), utf16.end(), back_inserter(utf8)); return utf8; } // convert from utf8 string to utf16/wide string wstring convert_to_utf16(const string& utf8) { wstring utf16; // pre-allocate expected memory utf16.reserve(code_point_count(utf8)*2); utf8::utf8to16(utf8.begin(), utf8.end(), back_inserter(utf16)); return utf16; } #endif } }