#include "ByteArray.class.h" //Define size of a uchar_repr_t #define CHARSZ(x) (x.c[0] == 0 ? 0 : (x.c[1] == 0 ? 1 : (x.c[2] == 0 ? 2 : (x.c[3] == 0 ? 3 : 4)))) using namespace CMem; ByteArray::ByteArray(const char* c) : BasicString() { m_length = strlen(c); memcpy(m_string, (u8int*)c, m_length); } void ByteArray::affect(const String &string, u8int encoding) { m_length = 0; for (u32int i = 0; i < string.size(); i++) { uchar_repr_t a = string[i].encode(encoding); m_length += CHARSZ(a); } if (m_string != 0) delete m_string; if (m_length == 0) { m_string = 0; return; } m_string = new u8int[m_length]; u32int x = 0; for (u32int i = 0; i < string.size(); i++) { uchar_repr_t a = string[i].encode(encoding); memcpy(m_string + x, (u8int*)a.c, CHARSZ(a)); x += CHARSZ(a); } } void ByteArray::resize(u32int size) { if (size == m_length) return; if (size == 0) { delete m_string; m_length = 0; m_string = 0; } u8int *nd = new u8int[size]; if (size < m_length) { memcpy(nd, m_string, size); } else { memcpy(nd, m_string, m_length); memset(nd + m_length, 0, size - m_length); } delete m_string; m_string = nd; m_length = size; } String ByteArray::toString (u8int encoding) { char* c = new char[m_length + 1]; memcpy((u8int*)c, m_string, m_length); c[m_length] = 0; //Add NULL terminator String r(c, encoding); delete c; return r; }