assimp/code/irrXML/irrString.h

665 lines
13 KiB
C
Raw Normal View History

// Copyright (C) 2002-2005 Nikolaus Gebhardt
// This file is part of the "Irrlicht Engine" and the "irrXML" project.
// For conditions of distribution and use, see copyright notice in irrlicht.h and irrXML.h
#ifndef __IRR_STRING_H_INCLUDED__
#define __IRR_STRING_H_INCLUDED__
#include "irrTypes.h"
namespace irr
{
namespace core
{
//! Very simple string class with some useful features.
/** string<c8> and string<wchar_t> work both with unicode AND ascii,
so you can assign unicode to string<c8> and ascii to string<wchar_t>
(and the other way round) if your ever would want to.
Note that the conversation between both is not done using an encoding.
Known bugs:
Special characters like '<EFBFBD>', '<EFBFBD>' and '<EFBFBD>' are ignored in the
methods make_upper, make_lower and equals_ignore_case.
*/
template <class T>
class string
{
public:
//! Default constructor
string()
: allocated(1), used(1), array(0)
{
array = new T[1];
array[0] = 0x0;
}
//! Constructor
string(const string<T>& other)
: allocated(0), used(0), array(0)
{
*this = other;
}
//! Constructs a string from an int
string(int number)
: allocated(0), used(0), array(0)
{
// store if negative and make positive
bool negative = false;
if (number < 0)
{
number *= -1;
negative = true;
}
// temporary buffer for 16 numbers
c8 tmpbuf[16];
tmpbuf[15] = 0;
s32 idx = 15;
// special case '0'
if (!number)
{
tmpbuf[14] = '0';
*this = &tmpbuf[14];
return;
}
// add numbers
while(number && idx)
{
idx--;
tmpbuf[idx] = (c8)('0' + (number % 10));
number = number / 10;
}
// add sign
if (negative)
{
idx--;
tmpbuf[idx] = '-';
}
*this = &tmpbuf[idx];
}
//! Constructor for copying a string from a pointer with a given lenght
template <class B>
string(const B* c, s32 lenght)
: allocated(0), used(0), array(0)
{
if (!c)
return;
allocated = used = lenght+1;
array = new T[used];
for (s32 l = 0; l<lenght; ++l)
array[l] = (T)c[l];
array[lenght] = 0;
}
//! Constructor for unicode and ascii strings
template <class B>
string(const B* c)
: allocated(0), used(0), array(0)
{
*this = c;
}
//! destructor
~string()
{
delete [] array;
}
//! Assignment operator
string<T>& operator=(const string<T>& other)
{
if (this == &other)
return *this;
delete [] array;
allocated = used = other.size()+1;
array = new T[used];
const T* p = other.c_str();
for (s32 i=0; i<used; ++i, ++p)
array[i] = *p;
return *this;
}
//! Assignment operator for strings, ascii and unicode
template <class B>
string<T>& operator=(const B* c)
{
if (!c)
{
if (!array)
{
array = new T[1];
allocated = 1;
used = 1;
}
array[0] = 0x0;
return *this;
}
if ((void*)c == (void*)array)
return *this;
s32 len = 0;
const B* p = c;
while(*p)
{
++len;
++p;
}
// we'll take the old string for a while, because the new string could be
// a part of the current string.
T* oldArray = array;
allocated = used = len+1;
array = new T[used];
for (s32 l = 0; l<len+1; ++l)
array[l] = (T)c[l];
delete [] oldArray;
return *this;
}
//! Add operator for other strings
string<T> operator+(const string<T>& other)
{
string<T> str(*this);
str.append(other);
return str;
}
//! Add operator for strings, ascii and unicode
template <class B>
string<T> operator+(const B* c)
{
string<T> str(*this);
str.append(c);
return str;
}
//! Direct access operator
T& operator [](const s32 index) const
{
_IRR_DEBUG_BREAK_IF(index>=used) // bad index
return array[index];
}
//! Comparison operator
bool operator ==(const T* str) const
{
int i;
for(i=0; array[i] && str[i]; ++i)
if (array[i] != str[i])
return false;
return !array[i] && !str[i];
}
//! Comparison operator
bool operator ==(const string<T>& other) const
{
for(s32 i=0; array[i] && other.array[i]; ++i)
if (array[i] != other.array[i])
return false;
return used == other.used;
}
//! Is smaller operator
bool operator <(const string<T>& other) const
{
for(s32 i=0; array[i] && other.array[i]; ++i)
if (array[i] != other.array[i])
return (array[i] < other.array[i]);
return used < other.used;
}
//! Equals not operator
bool operator !=(const string<T>& other) const
{
return !(*this == other);
}
//! Returns length of string
/** \return Returns length of the string in characters. */
s32 size() const
{
return used-1;
}
//! Returns character string
/** \return Returns pointer to C-style zero terminated string. */
const T* c_str() const
{
return array;
}
//! Makes the string lower case.
void make_lower()
{
const T A = (T)'A';
const T Z = (T)'Z';
const T diff = (T)'a' - A;
for (s32 i=0; i<used; ++i)
{
if (array[i]>=A && array[i]<=Z)
array[i] += diff;
}
}
//! Makes the string upper case.
void make_upper()
{
const T a = (T)'a';
const T z = (T)'z';
const T diff = (T)'A' - a;
for (s32 i=0; i<used; ++i)
{
if (array[i]>=a && array[i]<=z)
array[i] += diff;
}
}
//! Compares the string ignoring case.
/** \param other: Other string to compare.
\return Returns true if the string are equal ignoring case. */
bool equals_ignore_case(const string<T>& other) const
{
for(s32 i=0; array[i] && other[i]; ++i)
if (toLower(array[i]) != toLower(other[i]))
return false;
return used == other.used;
}
//! compares the first n characters of the strings
bool equalsn(const string<T>& other, int len)
{
int i;
for(i=0; array[i] && other[i] && i < len; ++i)
if (array[i] != other[i])
return false;
// if one (or both) of the strings was smaller then they
// are only equal if they have the same lenght
return (i == len) || (used == other.used);
}
//! compares the first n characters of the strings
bool equalsn(const T* str, int len)
{
int i;
for(i=0; array[i] && str[i] && i < len; ++i)
if (array[i] != str[i])
return false;
// if one (or both) of the strings was smaller then they
// are only equal if they have the same lenght
return (i == len) || (array[i] == 0 && str[i] == 0);
}
//! Appends a character to this string
/** \param character: Character to append. */
void append(T character)
{
if (used + 1 > allocated)
reallocate((s32)used + 1);
used += 1;
array[used-2] = character;
array[used-1] = 0;
}
//! Appends a string to this string
/** \param other: String to append. */
void append(const string<T>& other)
{
--used;
s32 len = other.size();
if (used + len + 1 > allocated)
reallocate((s32)used + (s32)len + 1);
for (s32 l=0; l<len+1; ++l)
array[l+used] = other[l];
used = used + len + 1;
}
//! Appends a string of the length l to this string.
/** \param other: other String to append to this string.
\param length: How much characters of the other string to add to this one. */
void append(const string<T>& other, s32 length)
{
s32 len = other.size();
if (len < length)
{
append(other);
return;
}
len = length;
--used;
if (used + len > allocated)
reallocate((s32)used + (s32)len);
for (s32 l=0; l<len; ++l)
array[l+used] = other[l];
used = used + len;
}
//! Reserves some memory.
/** \param count: Amount of characters to reserve. */
void reserve(s32 count)
{
if (count < allocated)
return;
reallocate(count);
}
//! finds first occurrence of character in string
/** \param c: Character to search for.
\return Returns position where the character has been found,
or -1 if not found. */
s32 findFirst(T c) const
{
for (s32 i=0; i<used; ++i)
if (array[i] == c)
return i;
return -1;
}
//! finds first occurrence of a character of a list in string
/** \param c: List of strings to find. For example if the method
should find the first occurance of 'a' or 'b', this parameter should be "ab".
\param count: Amount of characters in the list. Ususally,
this should be strlen(ofParameter1)
\return Returns position where one of the character has been found,
or -1 if not found. */
s32 findFirstChar(T* c, int count) const
{
for (s32 i=0; i<used; ++i)
for (int j=0; j<count; ++j)
if (array[i] == c[j])
return i;
return -1;
}
//! Finds first position of a character not in a given list.
/** \param c: List of characters not to find. For example if the method
should find the first occurance of a character not 'a' or 'b', this parameter should be "ab".
\param count: Amount of characters in the list. Ususally,
this should be strlen(ofParameter1)
\return Returns position where the character has been found,
or -1 if not found. */
template <class B>
s32 findFirstCharNotInList(B* c, int count) const
{
for (int i=0; i<used; ++i)
{
int j;
for (j=0; j<count; ++j)
if (array[i] == c[j])
break;
if (j==count)
return i;
}
return -1;
}
//! Finds last position of a character not in a given list.
/** \param c: List of characters not to find. For example if the method
should find the first occurance of a character not 'a' or 'b', this parameter should be "ab".
\param count: Amount of characters in the list. Ususally,
this should be strlen(ofParameter1)
\return Returns position where the character has been found,
or -1 if not found. */
template <class B>
s32 findLastCharNotInList(B* c, int count) const
{
for (int i=used-2; i>=0; --i)
{
int j;
for (j=0; j<count; ++j)
if (array[i] == c[j])
break;
if (j==count)
return i;
}
return -1;
}
//! finds next occurrence of character in string
/** \param c: Character to search for.
\param startPos: Position in string to start searching.
\return Returns position where the character has been found,
or -1 if not found. */
s32 findNext(T c, s32 startPos) const
{
for (s32 i=startPos; i<used; ++i)
if (array[i] == c)
return i;
return -1;
}
//! finds last occurrence of character in string
//! \param c: Character to search for.
//! \return Returns position where the character has been found,
//! or -1 if not found.
s32 findLast(T c) const
{
for (s32 i=used-1; i>=0; --i)
if (array[i] == c)
return i;
return -1;
}
//! Returns a substring
//! \param begin: Start of substring.
//! \param length: Length of substring.
string<T> subString(s32 begin, s32 length)
{
if (length <= 0)
return string<T>("");
string<T> o;
o.reserve(length+1);
for (s32 i=0; i<length; ++i)
o.array[i] = array[i+begin];
o.array[length] = 0;
o.used = o.allocated;
return o;
}
void operator += (T c)
{
append(c);
}
void operator += (const string<T>& other)
{
append(other);
}
void operator += (int i)
{
append(string<T>(i));
}
//! replaces all characters of a special type with another one
void replace(T toReplace, T replaceWith)
{
for (s32 i=0; i<used; ++i)
if (array[i] == toReplace)
array[i] = replaceWith;
}
//! trims the string.
/** Removes whitespace from begin and end of the string. */
void trim()
{
const char whitespace[] = " \t\n";
const int whitespacecount = 3;
// find start and end of real string without whitespace
int begin = findFirstCharNotInList(whitespace, whitespacecount);
if (begin == -1)
return;
int end = findLastCharNotInList(whitespace, whitespacecount);
if (end == -1)
return;
*this = subString(begin, (end +1) - begin);
}
//! Erases a character from the string. May be slow, because all elements
//! following after the erased element have to be copied.
//! \param index: Index of element to be erased.
void erase(int index)
{
_IRR_DEBUG_BREAK_IF(index>=used || index<0) // access violation
for (int i=index+1; i<used; ++i)
array[i-1] = array[i];
--used;
}
private:
//! Returns a character converted to lower case
T toLower(const T& t) const
{
if (t>=(T)'A' && t<=(T)'Z')
return t + ((T)'a' - (T)'A');
else
return t;
}
//! Reallocate the array, make it bigger or smaler
void reallocate(s32 new_size)
{
T* old_array = array;
array = new T[new_size];
allocated = new_size;
s32 amount = used < new_size ? used : new_size;
for (s32 i=0; i<amount; ++i)
array[i] = old_array[i];
if (allocated < used)
used = allocated;
delete [] old_array;
}
//--- member variables
T* array;
s32 allocated;
s32 used;
};
//! Typedef for character strings
typedef string<irr::c8> stringc;
//! Typedef for wide character strings
typedef string<wchar_t> stringw;
} // end namespace core
} // end namespace irr
#endif