251 lines
5.6 KiB
C++
251 lines
5.6 KiB
C++
#include <STDInclude.hpp>
|
|
#ifdef ENABLE_BASE128
|
|
#include "base128.h"
|
|
#endif
|
|
|
|
namespace Utils::String
|
|
{
|
|
const char* VA(const char* fmt, ...)
|
|
{
|
|
static VAProvider<4, 256> globalProvider;
|
|
static thread_local VAProvider<8, 256> provider;
|
|
|
|
va_list ap;
|
|
va_start(ap, fmt);
|
|
|
|
const char* result;
|
|
if (Components::Loader::IsUninitializing()) result = globalProvider.get(fmt, ap);
|
|
else result = provider.get(fmt, ap);
|
|
|
|
va_end(ap);
|
|
return result;
|
|
}
|
|
|
|
std::string ToLower(const std::string& text)
|
|
{
|
|
std::string result;
|
|
std::ranges::transform(text, std::back_inserter(result), [](const unsigned char input) -> char
|
|
{
|
|
return static_cast<char>(std::tolower(input));
|
|
});
|
|
|
|
return result;
|
|
}
|
|
|
|
std::string ToUpper(const std::string& text)
|
|
{
|
|
std::string result;
|
|
std::ranges::transform(text, std::back_inserter(result), [](const unsigned char input) -> char
|
|
{
|
|
return static_cast<char>(std::toupper(input));
|
|
});
|
|
|
|
return result;
|
|
}
|
|
|
|
bool Compare(const std::string& lhs, const std::string& rhs)
|
|
{
|
|
return std::ranges::equal(lhs, rhs, [](const unsigned char a, const unsigned char b) -> bool
|
|
{
|
|
return std::tolower(a) == std::tolower(b);
|
|
});
|
|
}
|
|
|
|
std::string DumpHex(const std::string& data, const std::string& separator)
|
|
{
|
|
std::string result;
|
|
|
|
for (std::size_t i = 0; i < data.size(); ++i)
|
|
{
|
|
if (i > 0)
|
|
{
|
|
result.append(separator);
|
|
}
|
|
|
|
result.append(VA("%02X", data[i] & 0xFF));
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
std::string XOR(std::string str, char value)
|
|
{
|
|
for (std::size_t i = 0; i < str.size(); ++i)
|
|
{
|
|
str[i] ^= static_cast<unsigned char>(value);
|
|
}
|
|
|
|
return str;
|
|
}
|
|
|
|
std::vector<std::string> Split(const std::string& str, const char delim)
|
|
{
|
|
std::stringstream ss(str);
|
|
std::string item;
|
|
std::vector<std::string> elems;
|
|
|
|
while (std::getline(ss, item, delim))
|
|
{
|
|
elems.push_back(item); // elems.push_back(std::move(item)); // if C++11 (based on comment from S1x)
|
|
}
|
|
|
|
return elems;
|
|
}
|
|
|
|
void Replace(std::string& str, const std::string& from, const std::string& to)
|
|
{
|
|
std::size_t nPos = 0;
|
|
|
|
while ((nPos = str.find(from, nPos)) != std::string::npos)
|
|
{
|
|
str = str.replace(nPos, from.length(), to);
|
|
nPos += to.length();
|
|
}
|
|
}
|
|
|
|
bool StartsWith(const std::string& haystack, const std::string& needle)
|
|
{
|
|
return haystack.find(needle) == 0; // If the pos of the first found char is 0, string starts with 'needle'
|
|
}
|
|
|
|
bool EndsWith(const std::string& haystack, const std::string& needle)
|
|
{
|
|
if (needle.size() > haystack.size()) return false;
|
|
return std::equal(needle.rbegin(), needle.rend(), haystack.rbegin());
|
|
}
|
|
|
|
bool IsNumber(const std::string& str)
|
|
{
|
|
return !str.empty() && std::find_if(str.begin(), str.end(), [](unsigned char input)
|
|
{
|
|
return !std::isdigit(input);
|
|
}) == str.end();
|
|
}
|
|
|
|
// Trim from start
|
|
std::string& LTrim(std::string& str)
|
|
{
|
|
str.erase(str.begin(), std::find_if(str.begin(), str.end(), [](const unsigned char input)
|
|
{
|
|
return !std::isspace(input);
|
|
}));
|
|
|
|
return str;
|
|
}
|
|
|
|
// Trim from end
|
|
std::string& RTrim(std::string& str)
|
|
{
|
|
str.erase(std::find_if(str.rbegin(), str.rend(), [](const unsigned char input)
|
|
{
|
|
return !std::isspace(input);
|
|
}).base(), str.end());
|
|
|
|
return str;
|
|
}
|
|
|
|
// Trim from both ends
|
|
void Trim(std::string& str)
|
|
{
|
|
LTrim(RTrim(str));
|
|
}
|
|
|
|
std::string Convert(const std::wstring& wstr)
|
|
{
|
|
std::string result;
|
|
result.reserve(wstr.size());
|
|
|
|
for (const auto& chr : wstr)
|
|
{
|
|
result.push_back(static_cast<char>(chr));
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
std::wstring Convert(const std::string& str)
|
|
{
|
|
std::wstring result;
|
|
result.reserve(str.size());
|
|
|
|
for (const auto& chr : str)
|
|
{
|
|
result.push_back(static_cast<wchar_t>(chr));
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
std::string FormatTimeSpan(int milliseconds)
|
|
{
|
|
int secondsTotal = milliseconds / 1000;
|
|
int seconds = secondsTotal % 60;
|
|
int minutesTotal = secondsTotal / 60;
|
|
int minutes = minutesTotal % 60;
|
|
int hoursTotal = minutesTotal / 60;
|
|
|
|
return VA("%02d:%02d:%02d", hoursTotal, minutes, seconds);
|
|
}
|
|
|
|
std::string FormatBandwidth(std::size_t bytes, int milliseconds)
|
|
{
|
|
static const char* sizes[] =
|
|
{
|
|
"B",
|
|
"KB",
|
|
"MB",
|
|
"GB",
|
|
"TB",
|
|
};
|
|
|
|
if (!milliseconds) return "0.00 B/s";
|
|
|
|
double bytesPerSecond = (1000.0 / milliseconds) * bytes;
|
|
|
|
std::size_t i;
|
|
for (i = 0; bytesPerSecond > 1000 && i < ARRAYSIZE(sizes); ++i) // 1024 or 1000?
|
|
{
|
|
bytesPerSecond /= 1000;
|
|
}
|
|
|
|
return VA("%.2f %s/s", static_cast<float>(bytesPerSecond), sizes[i]);
|
|
}
|
|
|
|
#ifdef ENABLE_BASE64
|
|
// Encodes a given string in Base64
|
|
std::string EncodeBase64(const char* input, const unsigned long inputSize)
|
|
{
|
|
unsigned long outlen = long(inputSize + (inputSize / 3.0) + 16);
|
|
unsigned char* outbuf = new unsigned char[outlen]; //Reserve output memory
|
|
base64_encode(reinterpret_cast<unsigned char*>(const_cast<char*>(input)), inputSize, outbuf, &outlen);
|
|
std::string ret(reinterpret_cast<char*>(outbuf), outlen);
|
|
delete[] outbuf;
|
|
return ret;
|
|
}
|
|
|
|
// Encodes a given string in Base64
|
|
std::string EncodeBase64(const std::string& input)
|
|
{
|
|
return EncodeBase64(input.data(), input.size());
|
|
}
|
|
#endif
|
|
|
|
#ifdef ENABLE_BASE128
|
|
// Encodes a given string in Base128
|
|
std::string EncodeBase128(const std::string& input)
|
|
{
|
|
base128 encoder;
|
|
|
|
void* inbuffer = const_cast<char*>(input.data());
|
|
char* buffer = encoder.encode(inbuffer, input.size());
|
|
/*
|
|
Interesting to see that the buffer returned by the encoder is not a standalone string copy
|
|
but instead is a pointer to the internal "encoded" field of the encoder. So if you deinitialize
|
|
the encoder that string will probably become garbage.
|
|
*/
|
|
std::string retval(buffer);
|
|
return retval;
|
|
}
|
|
#endif
|
|
}
|