1
0
mirror of https://github.com/OpenRCT2/OpenRCT2 synced 2026-01-15 11:03:00 +01:00
Files
OpenRCT2/src/openrct2/core/String.cpp
Ted John 7da60f4950 Make log functions atomic
Use only one print call to stdout in log functions so that messages are not malformed.
2018-04-26 21:26:30 +02:00

565 lines
15 KiB
C++

#pragma region Copyright (c) 2014-2017 OpenRCT2 Developers
/*****************************************************************************
* OpenRCT2, an open source clone of Roller Coaster Tycoon 2.
*
* OpenRCT2 is the work of many authors, a full list can be found in contributors.md
* For more information, visit https://github.com/OpenRCT2/OpenRCT2
*
* OpenRCT2 is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* A full copy of the GNU General Public License can be found in licence.txt
*****************************************************************************/
#pragma endregion
#include <cwctype>
#include <stdexcept>
#include <vector>
#ifdef _WIN32
#ifndef NOMINMAX
#define NOMINMAX
#endif
#define WIN32_LEAN_AND_MEAN
#include <windows.h>
#endif
#include "../localisation/Language.h"
#include "../util/Util.h"
#include "Math.hpp"
#include "Memory.hpp"
#include "String.hpp"
namespace String
{
std::string ToStd(const utf8 * str)
{
if (str == nullptr) return std::string();
else return std::string(str);
}
std::string StdFormat_VA(const utf8 * format, va_list args)
{
auto buffer = Format_VA(format, args);
auto returnValue = ToStd(buffer);
Memory::Free(buffer);
return returnValue;
}
std::string StdFormat(const utf8 * format, ...)
{
va_list args;
va_start(args, format);
const utf8 * buffer = Format_VA(format, args);
va_end(args);
std::string returnValue = ToStd(buffer);
Memory::Free(buffer);
return returnValue;
}
std::string ToUtf8(const std::wstring &s)
{
std::string result;
utf8 * cstr = widechar_to_utf8(s.c_str());
if (cstr != nullptr)
{
result = std::string(cstr);
}
free(cstr);
return result;
}
std::wstring ToUtf16(const std::string &s)
{
std::wstring result;
wchar_t * wcstr = utf8_to_widechar(s.c_str());
if (wcstr != nullptr)
{
result = std::wstring(wcstr);
}
free(wcstr);
return result;
}
bool IsNullOrEmpty(const utf8 * str)
{
return str == nullptr || str[0] == '\0';
}
sint32 Compare(const std::string &a, const std::string &b, bool ignoreCase)
{
return Compare(a.c_str(), b.c_str(), ignoreCase);
}
sint32 Compare(const utf8 * a, const utf8 * b, bool ignoreCase)
{
if (a == b) return 0;
if (a == nullptr) a = "";
if (b == nullptr) b = "";
if (ignoreCase)
{
return _stricmp(a, b);
}
else
{
return strcmp(a, b);
}
}
bool Equals(const std::string &a, const std::string &b, bool ignoreCase)
{
return Equals(a.c_str(), b.c_str(), ignoreCase);
}
bool Equals(const utf8 * a, const utf8 * b, bool ignoreCase)
{
if (a == b) return true;
if (a == nullptr || b == nullptr) return false;
if (ignoreCase)
{
return _stricmp(a, b) == 0;
}
else
{
return strcmp(a, b) == 0;
}
}
bool StartsWith(const utf8 * str, const utf8 * match, bool ignoreCase)
{
if (ignoreCase)
{
while (*str != '\0' && *match != '\0')
{
if (tolower(*str++) != tolower(*match++))
{
return false;
}
}
return true;
}
else
{
while (*str != '\0' && *match != '\0')
{
if (*str++ != *match++)
{
return false;
}
}
return true;
}
}
bool StartsWith(const std::string &str, const std::string &match, bool ignoreCase)
{
return StartsWith(str.c_str(), match.c_str(), ignoreCase);
}
size_t IndexOf(const utf8 * str, utf8 match, size_t startIndex)
{
const utf8 * ch = str + startIndex;
for (; *ch != '\0'; ch++)
{
if (*ch == match)
{
return (size_t)(ch - str);
}
}
return SIZE_MAX;
}
ptrdiff_t LastIndexOf(const utf8 * str, utf8 match)
{
const utf8 * lastOccurance = nullptr;
const utf8 * ch = str;
for (; *ch != '\0'; ch++)
{
if (*ch == match)
{
lastOccurance = ch;
}
}
if (lastOccurance == nullptr)
{
return -1;
}
else
{
return lastOccurance - str;
}
}
size_t LengthOf(const utf8 * str)
{
return utf8_length(str);
}
size_t SizeOf(const utf8 * str)
{
return strlen(str);
}
utf8 * Set(utf8 * buffer, size_t bufferSize, const utf8 * src)
{
return safe_strcpy(buffer, src, bufferSize);
}
utf8 * Set(utf8 * buffer, size_t bufferSize, const utf8 * src, size_t srcSize)
{
utf8 * dst = buffer;
size_t minSize = Math::Min(bufferSize - 1, srcSize);
for (size_t i = 0; i < minSize; i++)
{
*dst++ = *src;
if (*src == '\0') break;
src++;
}
*dst = '\0';
return buffer;
}
utf8 * Append(utf8 * buffer, size_t bufferSize, const utf8 * src)
{
return safe_strcat(buffer, src, bufferSize);
}
utf8 * Format(utf8 * buffer, size_t bufferSize, const utf8 * format, ...)
{
va_list args;
va_start(args, format);
vsnprintf(buffer, bufferSize, format, args);
va_end(args);
// Terminate buffer in case formatted string overflowed
buffer[bufferSize - 1] = '\0';
return buffer;
}
utf8 * Format(const utf8 * format, ...)
{
va_list args;
va_start(args, format);
utf8 * result = Format_VA(format, args);
va_end(args);
return result;
}
utf8 * Format_VA(const utf8 * format, va_list args)
{
va_list args1, args2;
va_copy(args1, args);
va_copy(args2, args);
// Try to format to a initial buffer, enlarge if not big enough
size_t bufferSize = 4096;
utf8 * buffer = Memory::Allocate<utf8>(bufferSize);
// Start with initial buffer
sint32 len = vsnprintf(buffer, bufferSize, format, args);
if (len < 0)
{
Memory::Free(buffer);
va_end(args1);
va_end(args2);
// An error occurred...
return nullptr;
}
size_t requiredSize = (size_t)len + 1;
if (requiredSize > bufferSize)
{
// Try again with bigger buffer
buffer = Memory::Reallocate<utf8>(buffer, bufferSize);
len = vsnprintf(buffer, bufferSize, format, args);
if (len < 0)
{
Memory::Free(buffer);
va_end(args1);
va_end(args2);
// An error occurred...
return nullptr;
}
}
else
{
// Reduce buffer size to only what was required
bufferSize = requiredSize;
buffer = Memory::Reallocate<utf8>(buffer, bufferSize);
}
// Ensure buffer is terminated
buffer[bufferSize - 1] = '\0';
va_end(args1);
va_end(args2);
return buffer;
}
utf8 * AppendFormat(utf8 * buffer, size_t bufferSize, const utf8 * format, ...)
{
utf8 * dst = buffer;
size_t i;
for (i = 0; i < bufferSize; i++)
{
if (*dst == '\0') break;
dst++;
}
if (i < bufferSize - 1)
{
va_list args;
va_start(args, format);
vsnprintf(dst, bufferSize - i - 1, format, args);
va_end(args);
// Terminate buffer in case formatted string overflowed
buffer[bufferSize - 1] = '\0';
}
return buffer;
}
utf8 * Duplicate(const std::string &src)
{
return String::Duplicate(src.c_str());
}
utf8 * Duplicate(const utf8 * src)
{
utf8 * result = nullptr;
if (src != nullptr)
{
size_t srcSize = SizeOf(src) + 1;
result = Memory::Allocate<utf8>(srcSize);
memcpy(result, src, srcSize);
}
return result;
}
utf8 * DiscardUse(utf8 * * ptr, utf8 * replacement)
{
Memory::Free(*ptr);
*ptr = replacement;
return replacement;
}
utf8 * DiscardDuplicate(utf8 * * ptr, const utf8 * replacement)
{
return DiscardUse(ptr, String::Duplicate(replacement));
}
std::vector<std::string> Split(const std::string &s, const std::string &delimiter)
{
if (delimiter.empty())
{
throw std::invalid_argument(nameof(delimiter) " can not be empty.");
}
std::vector<std::string> results;
if (!s.empty())
{
size_t index = 0;
size_t nextIndex;
do
{
nextIndex = s.find(delimiter, index);
std::string value;
if (nextIndex == std::string::npos)
{
value = s.substr(index);
}
else
{
value = s.substr(index, nextIndex - index);
}
results.push_back(value);
index = nextIndex + delimiter.size();
}
while (nextIndex != SIZE_MAX);
}
return results;
}
utf8 * SkipBOM(utf8 * buffer)
{
return (utf8*)SkipBOM((const utf8 *)buffer);
}
const utf8 * SkipBOM(const utf8 * buffer)
{
if ((uint8)buffer[0] == 0xEF && (uint8)buffer[1] == 0xBB && (uint8)buffer[2] == 0xBF)
{
return buffer + 3;
}
return buffer;
}
size_t GetCodepointLength(codepoint_t codepoint)
{
return utf8_get_codepoint_length(codepoint);
}
codepoint_t GetNextCodepoint(utf8 * ptr, utf8 * * nextPtr)
{
return GetNextCodepoint((const utf8 *)ptr, (const utf8 * *)nextPtr);
}
codepoint_t GetNextCodepoint(const utf8 * ptr, const utf8 * * nextPtr)
{
return utf8_get_next(ptr, nextPtr);
}
utf8 * WriteCodepoint(utf8 * dst, codepoint_t codepoint)
{
return utf8_write_codepoint(dst, codepoint);
}
bool IsWhiteSpace(codepoint_t codepoint)
{
// 0x3000 is the 'ideographic space', a 'fullwidth' character used in CJK languages.
return iswspace((wchar_t)codepoint) || codepoint == 0x3000;
}
utf8 * Trim(utf8 * str)
{
utf8 * firstNonWhitespace = nullptr;
codepoint_t codepoint;
utf8 * ch = str;
utf8 * nextCh;
while ((codepoint = GetNextCodepoint(ch, &nextCh)) != '\0')
{
if (codepoint <= WCHAR_MAX && !IsWhiteSpace(codepoint))
{
if (firstNonWhitespace == nullptr)
{
firstNonWhitespace = ch;
}
}
ch = nextCh;
}
if (firstNonWhitespace != nullptr &&
firstNonWhitespace != str)
{
// Take multibyte characters into account: use the last byte of the
// current character.
size_t newStringSize = (nextCh - 1) - firstNonWhitespace;
#ifdef DEBUG
size_t currentStringSize = String::SizeOf(str);
Guard::Assert(newStringSize < currentStringSize, GUARD_LINE);
#endif
std::memmove(str, firstNonWhitespace, newStringSize);
str[newStringSize] = '\0';
}
else
{
*ch = '\0';
}
return str;
}
const utf8 * TrimStart(const utf8 * str)
{
codepoint_t codepoint;
const utf8 * ch = str;
const utf8 * nextCh;
while ((codepoint = GetNextCodepoint(ch, &nextCh)) != '\0')
{
if (codepoint <= WCHAR_MAX && !IsWhiteSpace(codepoint))
{
return ch;
}
ch = nextCh;
}
// String is all whitespace
return ch;
}
utf8 * TrimStart(utf8 * buffer, size_t bufferSize, const utf8 * src)
{
return String::Set(buffer, bufferSize, TrimStart(src));
}
std::string TrimStart(const std::string &s)
{
const utf8 * trimmed = TrimStart(s.c_str());
return std::string(trimmed);
}
std::string Trim(const std::string &s)
{
codepoint_t codepoint;
const utf8 * ch = s.c_str();
const utf8 * nextCh;
const utf8 * startSubstr = nullptr;
const utf8 * endSubstr = nullptr;
while ((codepoint = GetNextCodepoint(ch, &nextCh)) != '\0')
{
bool isWhiteSpace = codepoint <= WCHAR_MAX && IsWhiteSpace(codepoint);
if (!isWhiteSpace)
{
if (startSubstr == nullptr)
{
startSubstr = ch;
}
// Take multibyte characters into account: move pointer towards
// the last byte of the current character.
endSubstr = nextCh - 1;
}
ch = nextCh;
}
if (startSubstr == nullptr)
{
// String is all whitespace
return std::string();
}
size_t stringLength = endSubstr - startSubstr + 1;
return std::string(startSubstr, stringLength);
}
std::string Convert(const std::string_view& src, sint32 srcCodePage, sint32 dstCodePage)
{
#ifdef _WIN32
// Convert from source code page to UTF-16
std::wstring u16;
{
int srcLen = (int)src.size();
int sizeReq = MultiByteToWideChar(srcCodePage, 0, src.data(), srcLen, nullptr, 0);
u16 = std::wstring(sizeReq, 0);
MultiByteToWideChar(srcCodePage, 0, src.data(), srcLen, u16.data(), sizeReq);
}
// Convert from UTF-16 to destination code page
std::string dst;
{
int srcLen = (int)u16.size();
int sizeReq = WideCharToMultiByte(dstCodePage, 0, u16.data(), srcLen, nullptr, 0, nullptr, nullptr);
dst = std::string(sizeReq, 0);
WideCharToMultiByte(dstCodePage, 0, u16.data(), srcLen, dst.data(), sizeReq, nullptr, nullptr);
}
return dst;
#else
STUB();
return std::string(src);
#endif
}
}