axmol/cocos2dx/label_nodes/CCLabelBMFont.cpp

1462 lines
42 KiB
C++
Raw Normal View History

/****************************************************************************
Copyright (c) 2010-2012 cocos2d-x.org
Copyright (c) 2008-2010 Ricardo Quesada
Copyright (c) 2011 Zynga Inc.
http://www.cocos2d-x.org
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
Use any of these editors to generate BMFonts:
http://glyphdesigner.71squared.com/ (Commercial, Mac OS X)
http://www.n4te.com/hiero/hiero.jnlp (Free, Java)
http://slick.cokeandcode.com/demos/hiero.jnlp (Free, Java)
http://www.angelcode.com/products/bmfont/ (Free, Windows only)
****************************************************************************/
#include "CCLabelBMFont.h"
#include "cocoa/CCString.h"
#include "platform/platform.h"
#include "cocoa/CCDictionary.h"
#include "CCConfiguration.h"
2012-11-16 14:23:14 +08:00
#include "draw_nodes/CCDrawingPrimitives.h"
#include "sprite_nodes/CCSprite.h"
#include "support/CCPointExtension.h"
#include "platform/CCFileUtils.h"
2012-03-22 09:46:07 +08:00
#include "CCDirector.h"
#include "textures/CCTextureCache.h"
using namespace std;
NS_CC_BEGIN
2012-03-22 09:46:07 +08:00
static int cc_wcslen(const unsigned short* str)
{
int i=0;
while(*str++) i++;
return i;
}
/* Code from GLIB gutf8.c starts here. */
#define UTF8_COMPUTE(Char, Mask, Len) \
if (Char < 128) \
{ \
Len = 1; \
Mask = 0x7f; \
} \
else if ((Char & 0xe0) == 0xc0) \
{ \
Len = 2; \
Mask = 0x1f; \
} \
else if ((Char & 0xf0) == 0xe0) \
{ \
Len = 3; \
Mask = 0x0f; \
} \
else if ((Char & 0xf8) == 0xf0) \
{ \
Len = 4; \
Mask = 0x07; \
} \
else if ((Char & 0xfc) == 0xf8) \
{ \
Len = 5; \
Mask = 0x03; \
} \
else if ((Char & 0xfe) == 0xfc) \
{ \
Len = 6; \
Mask = 0x01; \
} \
else \
2012-03-22 09:46:07 +08:00
Len = -1;
#define UTF8_LENGTH(Char) \
((Char) < 0x80 ? 1 : \
((Char) < 0x800 ? 2 : \
((Char) < 0x10000 ? 3 : \
((Char) < 0x200000 ? 4 : \
2012-03-22 09:46:07 +08:00
((Char) < 0x4000000 ? 5 : 6)))))
#define UTF8_GET(Result, Chars, Count, Mask, Len) \
(Result) = (Chars)[0] & (Mask); \
for ((Count) = 1; (Count) < (Len); ++(Count)) \
{ \
if (((Chars)[(Count)] & 0xc0) != 0x80) \
{ \
(Result) = -1; \
break; \
} \
(Result) <<= 6; \
(Result) |= ((Chars)[(Count)] & 0x3f); \
2012-03-22 09:46:07 +08:00
}
#define UNICODE_VALID(Char) \
((Char) < 0x110000 && \
(((Char) & 0xFFFFF800) != 0xD800) && \
((Char) < 0xFDD0 || (Char) > 0xFDEF) && \
2012-03-22 09:46:07 +08:00
((Char) & 0xFFFE) != 0xFFFE)
static const char utf8_skip_data[256] = {
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2,
3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 5,
5, 5, 5, 6, 6, 1, 1
};
static const char *const g_utf8_skip = utf8_skip_data;
#define cc_utf8_next_char(p) (char *)((p) + g_utf8_skip[*(unsigned char *)(p)])
/*
* @str: the string to search through.
* @c: the character to find.
2012-03-22 09:46:07 +08:00
*
* Returns the index of the first occurrence of the character, if found. Otherwise -1 is returned.
*
* Return value: the index of the first occurrence of the character if found or -1 otherwise.
* */
static unsigned int cc_utf8_find_char(std::vector<unsigned short> str, unsigned short c)
{
unsigned int len = str.size();
2012-03-22 09:46:07 +08:00
for (unsigned int i = 0; i < len; ++i)
if (str[i] == c) return i;
2012-03-22 09:46:07 +08:00
return -1;
2012-03-22 09:46:07 +08:00
}
/*
* @str: the string to search through.
* @c: the character to not look for.
2012-03-22 09:46:07 +08:00
*
* Return value: the index of the last character that is not c.
* */
static unsigned int cc_utf8_find_last_not_char(std::vector<unsigned short> str, unsigned short c)
{
int len = str.size();
2012-03-22 09:46:07 +08:00
int i = len - 1;
for (; i >= 0; --i)
if (str[i] != c) return i;
2012-03-22 09:46:07 +08:00
return i;
2012-03-22 09:46:07 +08:00
}
/*
* @str: the string to trim
* @index: the index to start trimming from.
2012-03-22 09:46:07 +08:00
*
* Trims str st str=[0, index) after the operation.
*
* Return value: the trimmed string.
* */
static void cc_utf8_trim_from(std::vector<unsigned short>* str, int index)
{
int size = str->size();
if (index >= size || index < 0)
return;
2012-03-22 09:46:07 +08:00
str->erase(str->begin() + index, str->begin() + size);
2012-03-22 09:46:07 +08:00
}
/*
* @ch is the unicode character whitespace?
*
* Reference: http://en.wikipedia.org/wiki/Whitespace_character#Unicode
*
* Return value: weather the character is a whitespace character.
* */
static bool isspace_unicode(unsigned short ch)
{
return (ch >= 0x0009 && ch <= 0x000D) || ch == 0x0020 || ch == 0x0085 || ch == 0x00A0 || ch == 0x1680
|| (ch >= 0x2000 && ch <= 0x200A) || ch == 0x2028 || ch == 0x2029 || ch == 0x202F
|| ch == 0x205F || ch == 0x3000;
2012-03-22 09:46:07 +08:00
}
static void cc_utf8_trim_ws(std::vector<unsigned short>* str)
{
int len = str->size();
if ( len <= 0 )
return;
int last_index = len - 1;
// Only start trimming if the last character is whitespace..
if (isspace_unicode((*str)[last_index]))
{
for (int i = last_index - 1; i >= 0; --i)
{
if (isspace_unicode((*str)[i]))
last_index = i;
else
break;
}
cc_utf8_trim_from(str, last_index);
}
2012-03-22 09:46:07 +08:00
}
/*
* g_utf8_strlen:
* @p: pointer to the start of a UTF-8 encoded string.
* @max: the maximum number of bytes to examine. If @max
* is less than 0, then the string is assumed to be
* null-terminated. If @max is 0, @p will not be examined and
* may be %NULL.
*
* Returns the length of the string in characters.
*
* Return value: the length of the string in characters
**/
long
2012-03-22 09:46:07 +08:00
cc_utf8_strlen (const char * p, int max)
{
2012-11-15 17:16:51 +08:00
long len = 0;
const char *start = p;
2012-03-22 09:46:07 +08:00
2012-11-15 17:16:51 +08:00
if (!(p != NULL || max == 0))
2012-03-22 09:46:07 +08:00
{
2012-11-15 17:16:51 +08:00
return 0;
2012-03-22 09:46:07 +08:00
}
2012-11-15 17:16:51 +08:00
if (max < 0)
{
while (*p)
{
p = cc_utf8_next_char (p);
++len;
}
2012-03-22 09:46:07 +08:00
}
2012-11-15 17:16:51 +08:00
else
2012-03-22 09:46:07 +08:00
{
2012-11-15 17:16:51 +08:00
if (max == 0 || !*p)
return 0;
2012-03-22 09:46:07 +08:00
2012-11-15 17:16:51 +08:00
p = cc_utf8_next_char (p);
2012-03-22 09:46:07 +08:00
2012-11-15 17:16:51 +08:00
while (p - start < max && *p)
{
++len;
p = cc_utf8_next_char (p);
}
2012-03-22 09:46:07 +08:00
2012-11-15 17:16:51 +08:00
/* only do the last len increment if we got a complete
* char (don't count partial chars)
*/
if (p - start == max)
++len;
2012-03-22 09:46:07 +08:00
}
2012-11-15 17:16:51 +08:00
return len;
2012-03-22 09:46:07 +08:00
}
/*
* g_utf8_get_char:
* @p: a pointer to Unicode character encoded as UTF-8
*
* Converts a sequence of bytes encoded as UTF-8 to a Unicode character.
* If @p does not point to a valid UTF-8 encoded character, results are
* undefined. If you are not sure that the bytes are complete
* valid Unicode characters, you should use g_utf8_get_char_validated()
* instead.
*
* Return value: the resulting character
**/
static unsigned int
cc_utf8_get_char (const char * p)
{
int i, mask = 0, len;
unsigned int result;
unsigned char c = (unsigned char) *p;
UTF8_COMPUTE (c, mask, len);
if (len == -1)
return (unsigned int) - 1;
UTF8_GET (result, p, i, mask, len);
return result;
}
/*
2012-05-31 15:32:18 +08:00
* cc_utf16_from_utf8:
2012-03-22 09:46:07 +08:00
* @str_old: pointer to the start of a C string.
*
* Creates a utf8 string from a cstring.
2012-11-15 17:16:51 +08:00
*
2012-03-22 09:46:07 +08:00
* Return value: the newly created utf8 string.
* */
2012-05-31 15:32:18 +08:00
static unsigned short* cc_utf16_from_utf8(const char* str_old)
2012-03-22 09:46:07 +08:00
{
int len = cc_utf8_strlen(str_old, -1);
2012-03-22 09:46:07 +08:00
unsigned short* str_new = new unsigned short[len + 1];
str_new[len] = 0;
2012-03-22 09:46:07 +08:00
for (int i = 0; i < len; ++i)
{
str_new[i] = cc_utf8_get_char(str_old);
str_old = cc_utf8_next_char(str_old);
}
2012-03-22 09:46:07 +08:00
return str_new;
2012-03-22 09:46:07 +08:00
}
2012-05-31 15:32:18 +08:00
static std::vector<unsigned short> cc_utf16_vec_from_utf16_str(const unsigned short* str)
2012-03-22 09:46:07 +08:00
{
int len = cc_wcslen(str);
std::vector<unsigned short> str_new;
2012-03-22 09:46:07 +08:00
for (int i = 0; i < len; ++i)
2012-05-31 15:32:18 +08:00
{
str_new.push_back(str[i]);
2012-05-31 15:32:18 +08:00
}
return str_new;
2012-03-22 09:46:07 +08:00
}
//
//FNTConfig Cache - free functions
//
static CCDictionary* s_pConfigurations = NULL;
CCBMFontConfiguration* FNTConfigLoadFile( const char *fntFile)
{
CCBMFontConfiguration* pRet = NULL;
if( s_pConfigurations == NULL )
{
s_pConfigurations = new CCDictionary();
}
2010-08-12 17:39:25 +08:00
pRet = (CCBMFontConfiguration*)s_pConfigurations->objectForKey(fntFile);
if( pRet == NULL )
{
pRet = CCBMFontConfiguration::create(fntFile);
if (pRet)
{
s_pConfigurations->setObject(pRet, fntFile);
}
}
return pRet;
}
void FNTConfigRemoveCache( void )
{
if (s_pConfigurations)
{
s_pConfigurations->removeAllObjects();
CC_SAFE_RELEASE_NULL(s_pConfigurations);
}
}
2012-03-22 09:46:07 +08:00
//
//BitmapFontConfiguration
//
CCBMFontConfiguration * CCBMFontConfiguration::create(const char *FNTfile)
{
CCBMFontConfiguration * pRet = new CCBMFontConfiguration();
if (pRet->initWithFNTfile(FNTfile))
{
pRet->autorelease();
return pRet;
}
CC_SAFE_DELETE(pRet);
return NULL;
}
bool CCBMFontConfiguration::initWithFNTfile(const char *FNTfile)
{
m_pKerningDictionary = NULL;
m_pFontDefDictionary = NULL;
2012-11-15 17:16:51 +08:00
m_pCharacterSet = this->parseConfigFile(FNTfile);
if (! m_pCharacterSet)
{
return false;
}
return true;
}
2012-11-15 17:16:51 +08:00
std::set<unsigned int>* CCBMFontConfiguration::getCharacterSet() const
{
return m_pCharacterSet;
}
CCBMFontConfiguration::CCBMFontConfiguration()
2012-11-15 17:16:51 +08:00
: m_pFontDefDictionary(NULL)
, m_nCommonHeight(0)
, m_pKerningDictionary(NULL)
, m_pCharacterSet(NULL)
{
}
CCBMFontConfiguration::~CCBMFontConfiguration()
{
CCLOGINFO( "cocos2d: deallocing CCBMFontConfiguration" );
this->purgeFontDefDictionary();
this->purgeKerningDictionary();
m_sAtlasName.clear();
2012-11-15 17:16:51 +08:00
CC_SAFE_DELETE(m_pCharacterSet);
}
2012-03-22 09:46:07 +08:00
const char* CCBMFontConfiguration::description(void)
{
return CCString::createWithFormat(
"<CCBMFontConfiguration = %08X | Glphys:%d Kernings:%d | Image = %s>",
this,
HASH_COUNT(m_pFontDefDictionary),
HASH_COUNT(m_pKerningDictionary),
m_sAtlasName.c_str()
)->getCString();
}
2012-03-22 09:46:07 +08:00
void CCBMFontConfiguration::purgeKerningDictionary()
{
2012-11-15 17:16:51 +08:00
tCCKerningHashElement *current;
while(m_pKerningDictionary)
{
current = m_pKerningDictionary;
HASH_DEL(m_pKerningDictionary,current);
free(current);
}
}
2012-03-22 09:46:07 +08:00
void CCBMFontConfiguration::purgeFontDefDictionary()
{
2012-11-15 17:16:51 +08:00
tCCFontDefHashElement *current, *tmp;
HASH_ITER(hh, m_pFontDefDictionary, current, tmp) {
HASH_DEL(m_pFontDefDictionary, current);
free(current);
}
}
2012-03-22 09:46:07 +08:00
2012-11-15 17:16:51 +08:00
std::set<unsigned int>* CCBMFontConfiguration::parseConfigFile(const char *controlFile)
{
std::string fullpath = CCFileUtils::sharedFileUtils()->fullPathFromRelativePath(controlFile);
CCString *contents = CCString::createWithContentsOfFile(fullpath.c_str());
CCAssert(contents, "CCBMFontConfiguration::parseConfigFile | Open file error.");
2012-11-15 17:16:51 +08:00
set<unsigned int> *validCharsString = new set<unsigned int>();
if (!contents)
{
CCLOG("cocos2d: Error parsing FNTfile %s", controlFile);
2012-11-15 17:16:51 +08:00
return NULL;
}
// parse spacing / padding
std::string line;
std::string strLeft = contents->getCString();
while (strLeft.length() > 0)
{
int pos = strLeft.find('\n');
if (pos != (int)std::string::npos)
{
// the data is more than a line.get one line
line = strLeft.substr(0, pos);
strLeft = strLeft.substr(pos + 1);
}
else
{
// get the left data
line = strLeft;
strLeft.erase();
}
if(line.substr(0,strlen("info face")) == "info face")
{
// XXX: info parsing is incomplete
// Not needed for the Hiero editors, but needed for the AngelCode editor
// [self parseInfoArguments:line];
this->parseInfoArguments(line);
}
// Check to see if the start of the line is something we are interested in
else if(line.substr(0,strlen("common lineHeight")) == "common lineHeight")
{
this->parseCommonArguments(line);
}
else if(line.substr(0,strlen("page id")) == "page id")
{
this->parseImageFileName(line, controlFile);
}
else if(line.substr(0,strlen("chars c")) == "chars c")
{
// Ignore this line
}
else if(line.substr(0,strlen("char")) == "char")
{
// Parse the current line and create a new CharDef
2012-11-15 17:16:51 +08:00
tCCFontDefHashElement* element = (tCCFontDefHashElement*)malloc( sizeof(*element) );
this->parseCharacterDefinition(line, &element->fontDef);
element->key = element->fontDef.charID;
HASH_ADD_INT(m_pFontDefDictionary, key, element);
2012-11-15 17:16:51 +08:00
validCharsString->insert(element->fontDef.charID);
}
// else if(line.substr(0,strlen("kernings count")) == "kernings count")
// {
// this->parseKerningCapacity(line);
// }
else if(line.substr(0,strlen("kerning first")) == "kerning first")
{
this->parseKerningEntry(line);
}
}
2012-11-15 17:16:51 +08:00
return validCharsString;
}
2012-03-22 09:46:07 +08:00
void CCBMFontConfiguration::parseImageFileName(std::string line, const char *fntFile)
{
//////////////////////////////////////////////////////////////////////////
// line to parse:
// page id=0 file="bitmapFontTest.png"
//////////////////////////////////////////////////////////////////////////
// page ID. Sanity check
int index = line.find('=')+1;
int index2 = line.find(' ', index);
std::string value = line.substr(index, index2-index);
CCAssert(atoi(value.c_str()) == 0, "LabelBMFont file could not be found");
// file
index = line.find('"')+1;
index2 = line.find('"', index);
value = line.substr(index, index2-index);
m_sAtlasName = CCFileUtils::sharedFileUtils()->fullPathFromRelativeFile(value.c_str(), fntFile);
}
2012-03-22 09:46:07 +08:00
void CCBMFontConfiguration::parseInfoArguments(std::string line)
{
//////////////////////////////////////////////////////////////////////////
// possible lines to parse:
// info face="Script" size=32 bold=0 italic=0 charset="" unicode=1 stretchH=100 smooth=1 aa=1 padding=1,4,3,2 spacing=0,0 outline=0
// info face="Cracked" size=36 bold=0 italic=0 charset="" unicode=0 stretchH=100 smooth=1 aa=1 padding=0,0,0,0 spacing=1,1
//////////////////////////////////////////////////////////////////////////
// padding
int index = line.find("padding=");
int index2 = line.find(' ', index);
std::string value = line.substr(index, index2-index);
sscanf(value.c_str(), "padding=%d,%d,%d,%d", &m_tPadding.top, &m_tPadding.right, &m_tPadding.bottom, &m_tPadding.left);
CCLOG("cocos2d: padding: %d,%d,%d,%d", m_tPadding.left, m_tPadding.top, m_tPadding.right, m_tPadding.bottom);
}
2012-03-22 09:46:07 +08:00
void CCBMFontConfiguration::parseCommonArguments(std::string line)
{
//////////////////////////////////////////////////////////////////////////
// line to parse:
// common lineHeight=104 base=26 scaleW=1024 scaleH=512 pages=1 packed=0
//////////////////////////////////////////////////////////////////////////
// Height
int index = line.find("lineHeight=");
int index2 = line.find(' ', index);
std::string value = line.substr(index, index2-index);
sscanf(value.c_str(), "lineHeight=%d", &m_nCommonHeight);
// scaleW. sanity check
index = line.find("scaleW=") + strlen("scaleW=");
index2 = line.find(' ', index);
value = line.substr(index, index2-index);
CCAssert(atoi(value.c_str()) <= CCConfiguration::sharedConfiguration()->getMaxTextureSize(), "CCLabelBMFont: page can't be larger than supported");
// scaleH. sanity check
index = line.find("scaleH=") + strlen("scaleH=");
index2 = line.find(' ', index);
value = line.substr(index, index2-index);
CCAssert(atoi(value.c_str()) <= CCConfiguration::sharedConfiguration()->getMaxTextureSize(), "CCLabelBMFont: page can't be larger than supported");
// pages. sanity check
index = line.find("pages=") + strlen("pages=");
index2 = line.find(' ', index);
value = line.substr(index, index2-index);
CCAssert(atoi(value.c_str()) == 1, "CCBitfontAtlas: only supports 1 page");
// packed (ignore) What does this mean ??
}
2012-03-22 09:46:07 +08:00
void CCBMFontConfiguration::parseCharacterDefinition(std::string line, ccBMFontDef *characterDefinition)
{
//////////////////////////////////////////////////////////////////////////
// line to parse:
// char id=32 x=0 y=0 width=0 height=0 xoffset=0 yoffset=44 xadvance=14 page=0 chnl=0
//////////////////////////////////////////////////////////////////////////
// Character ID
int index = line.find("id=");
int index2 = line.find(' ', index);
std::string value = line.substr(index, index2-index);
sscanf(value.c_str(), "id=%u", &characterDefinition->charID);
// Character x
index = line.find("x=");
index2 = line.find(' ', index);
value = line.substr(index, index2-index);
sscanf(value.c_str(), "x=%f", &characterDefinition->rect.origin.x);
// Character y
index = line.find("y=");
index2 = line.find(' ', index);
value = line.substr(index, index2-index);
sscanf(value.c_str(), "y=%f", &characterDefinition->rect.origin.y);
// Character width
index = line.find("width=");
index2 = line.find(' ', index);
value = line.substr(index, index2-index);
sscanf(value.c_str(), "width=%f", &characterDefinition->rect.size.width);
// Character height
index = line.find("height=");
index2 = line.find(' ', index);
value = line.substr(index, index2-index);
sscanf(value.c_str(), "height=%f", &characterDefinition->rect.size.height);
// Character xoffset
index = line.find("xoffset=");
index2 = line.find(' ', index);
value = line.substr(index, index2-index);
sscanf(value.c_str(), "xoffset=%hd", &characterDefinition->xOffset);
// Character yoffset
index = line.find("yoffset=");
index2 = line.find(' ', index);
value = line.substr(index, index2-index);
sscanf(value.c_str(), "yoffset=%hd", &characterDefinition->yOffset);
// Character xadvance
index = line.find("xadvance=");
index2 = line.find(' ', index);
value = line.substr(index, index2-index);
sscanf(value.c_str(), "xadvance=%hd", &characterDefinition->xAdvance);
}
2012-03-22 09:46:07 +08:00
void CCBMFontConfiguration::parseKerningEntry(std::string line)
{
//////////////////////////////////////////////////////////////////////////
// line to parse:
// kerning first=121 second=44 amount=-7
//////////////////////////////////////////////////////////////////////////
// first
int first;
int index = line.find("first=");
int index2 = line.find(' ', index);
std::string value = line.substr(index, index2-index);
sscanf(value.c_str(), "first=%d", &first);
// second
int second;
index = line.find("second=");
index2 = line.find(' ', index);
value = line.substr(index, index2-index);
sscanf(value.c_str(), "second=%d", &second);
// amount
int amount;
index = line.find("amount=");
index2 = line.find(' ', index);
value = line.substr(index, index2-index);
sscanf(value.c_str(), "amount=%d", &amount);
2012-11-15 17:16:51 +08:00
tCCKerningHashElement *element = (tCCKerningHashElement *)calloc( sizeof( *element ), 1 );
element->amount = amount;
element->key = (first<<16) | (second&0xffff);
HASH_ADD_INT(m_pKerningDictionary,key, element);
}
//
//CCLabelBMFont
//
//LabelBMFont - Purge Cache
void CCLabelBMFont::purgeCachedData()
{
FNTConfigRemoveCache();
}
2010-08-12 17:39:25 +08:00
CCLabelBMFont * CCLabelBMFont::create()
{
CCLabelBMFont * pRet = new CCLabelBMFont();
if (pRet && pRet->init())
{
pRet->autorelease();
return pRet;
}
CC_SAFE_DELETE(pRet);
return NULL;
}
CCLabelBMFont * CCLabelBMFont::create(const char *str, const char *fntFile, float width, CCTextAlignment alignment)
{
return CCLabelBMFont::create(str, fntFile, width, alignment, CCPointZero);
}
CCLabelBMFont * CCLabelBMFont::create(const char *str, const char *fntFile, float width)
{
return CCLabelBMFont::create(str, fntFile, width, kCCTextAlignmentLeft, CCPointZero);
}
CCLabelBMFont * CCLabelBMFont::create(const char *str, const char *fntFile)
{
return CCLabelBMFont::create(str, fntFile, kCCLabelAutomaticWidth, kCCTextAlignmentLeft, CCPointZero);
}
2012-03-22 09:46:07 +08:00
//LabelBMFont - Creation & Init
CCLabelBMFont *CCLabelBMFont::create(const char *str, const char *fntFile, float width/* = kCCLabelAutomaticWidth*/, CCTextAlignment alignment/* = kCCTextAlignmentLeft*/, CCPoint imageOffset/* = CCPointZero*/)
2012-03-22 09:46:07 +08:00
{
CCLabelBMFont *pRet = new CCLabelBMFont();
if(pRet && pRet->initWithString(str, fntFile, width, alignment, imageOffset))
{
pRet->autorelease();
return pRet;
}
CC_SAFE_DELETE(pRet);
return NULL;
2012-03-22 09:46:07 +08:00
}
bool CCLabelBMFont::init()
{
return initWithString(NULL, NULL, kCCLabelAutomaticWidth, kCCTextAlignmentLeft, CCPointZero);
}
bool CCLabelBMFont::initWithString(const char *theString, const char *fntFile, float width/* = kCCLabelAutomaticWidth*/, CCTextAlignment alignment/* = kCCTextAlignmentLeft*/, CCPoint imageOffset/* = CCPointZero*/)
2012-03-22 09:46:07 +08:00
{
CCAssert(!m_pConfiguration, "re-init is no longer supported");
CCAssert( (theString && fntFile) || (theString==NULL && fntFile==NULL), "Invalid params for CCLabelBMFont");
CCTexture2D *texture = NULL;
if (fntFile)
{
CCBMFontConfiguration *newConf = FNTConfigLoadFile(fntFile);
2012-11-15 17:16:51 +08:00
if (!newConf)
{
CCLOG("cocos2d: WARNING. CCLabelBMFont: Impossible to create font. Please check file: '%s'", fntFile);
release();
return false;
}
newConf->retain();
CC_SAFE_RELEASE(m_pConfiguration);
m_pConfiguration = newConf;
m_sFntFile = fntFile;
texture = CCTextureCache::sharedTextureCache()->addImage(m_pConfiguration->getAtlasName());
}
else
{
texture = new CCTexture2D();
texture->autorelease();
}
if (theString == NULL)
{
theString = "";
}
if (CCSpriteBatchNode::initWithTexture(texture, strlen(theString)))
{
m_pAlignment = alignment;
m_tImageOffset = imageOffset;
m_fWidth = width;
m_cOpacity = 255;
m_tColor = ccWHITE;
2012-11-14 18:05:15 +08:00
m_obContentSize = CCSizeZero;
m_bIsOpacityModifyRGB = m_pobTextureAtlas->getTexture()->hasPremultipliedAlpha();
2012-11-15 17:16:51 +08:00
m_obAnchorPoint = ccp(0.5f, 0.5f);
m_pReusedChar = new CCSprite();
2012-11-15 17:16:51 +08:00
m_pReusedChar->initWithTexture(m_pobTextureAtlas->getTexture(), CCRectMake(0, 0, 0, 0), false);
m_pReusedChar->setBatchNode(this);
2012-11-15 17:16:51 +08:00
this->setString(theString);
2012-11-15 17:16:51 +08:00
return true;
}
return false;
}
2012-03-22 09:46:07 +08:00
CCLabelBMFont::CCLabelBMFont()
: m_cOpacity(0)
, m_bIsOpacityModifyRGB(false)
, m_pConfiguration(NULL)
2012-04-10 15:24:06 +08:00
, m_sString(NULL)
2012-03-22 09:46:07 +08:00
, m_bLineBreakWithoutSpaces(false)
, m_tImageOffset(CCPointZero)
{
}
CCLabelBMFont::~CCLabelBMFont()
{
CC_SAFE_RELEASE(m_pReusedChar);
CC_SAFE_DELETE(m_sString);
CC_SAFE_RELEASE(m_pConfiguration);
}
2010-08-12 17:39:25 +08:00
// LabelBMFont - Atlas generation
int CCLabelBMFont::kerningAmountForFirst(unsigned short first, unsigned short second)
{
int ret = 0;
unsigned int key = (first<<16) | (second & 0xffff);
if( m_pConfiguration->m_pKerningDictionary ) {
2012-11-15 17:16:51 +08:00
tCCKerningHashElement *element = NULL;
HASH_FIND_INT(m_pConfiguration->m_pKerningDictionary, &key, element);
if(element)
ret = element->amount;
}
return ret;
}
void CCLabelBMFont::createFontChars()
{
int nextFontPositionX = 0;
int nextFontPositionY = 0;
//unsigned short prev = -1;
int kerningAmount = 0;
2010-08-12 17:39:25 +08:00
CCSize tmpSize = CCSizeZero;
2010-12-24 17:07:31 +08:00
int longestLine = 0;
unsigned int totalHeight = 0;
unsigned int quantityOfLines = 1;
unsigned int stringLen = cc_wcslen(m_sString);
if (stringLen == 0)
{
return;
}
set<unsigned int> *charSet = m_pConfiguration->getCharacterSet();
for (unsigned int i = 0; i < stringLen - 1; ++i)
{
2012-03-22 09:46:07 +08:00
unsigned short c = m_sString[i];
if (c == '\n')
{
quantityOfLines++;
}
}
totalHeight = m_pConfiguration->m_nCommonHeight * quantityOfLines;
nextFontPositionY = 0-(m_pConfiguration->m_nCommonHeight - m_pConfiguration->m_nCommonHeight * quantityOfLines);
2012-11-15 17:16:51 +08:00
CCRect rect;
ccBMFontDef fontDef;
for (unsigned int i= 0; i < stringLen; i++)
{
unsigned short c = m_sString[i];
if (c == '\n')
{
nextFontPositionX = 0;
nextFontPositionY -= m_pConfiguration->m_nCommonHeight;
continue;
}
2012-11-15 17:16:51 +08:00
if (charSet->find(c) == charSet->end())
{
CCLOG("CCLabelBMFont: Attempted to use character not defined in this bitmap: %d", c);
continue;
}
2012-11-15 17:16:51 +08:00
tCCFontDefHashElement *element = NULL;
// unichar is a short, and an int is needed on HASH_FIND_INT
unsigned int key = c;
HASH_FIND_INT(m_pConfiguration->m_pFontDefDictionary, &key, element);
2012-11-15 17:16:51 +08:00
if (! element)
{
CCLOG("cocos2d: LabelBMFont: characer not found %d", c);
continue;
}
2012-11-15 17:16:51 +08:00
fontDef = element->fontDef;
2012-11-15 17:16:51 +08:00
rect = fontDef.rect;
rect = CC_RECT_PIXELS_TO_POINTS(rect);
rect.origin.x += m_tImageOffset.x;
rect.origin.y += m_tImageOffset.y;
CCSprite *fontChar;
2012-11-15 17:16:51 +08:00
bool hasSprite = true;
fontChar = (CCSprite*)(this->getChildByTag(i));
if( ! fontChar )
{
2012-11-15 17:16:51 +08:00
if( 0 )
{
/* WIP: Doesn't support many features yet.
But this code is super fast. It doesn't create any sprite.
Ideal for big labels.
*/
fontChar = m_pReusedChar;
fontChar->setBatchNode(NULL);
hasSprite = false;
}
else
{
fontChar = new CCSprite();
2012-11-15 17:16:51 +08:00
fontChar->initWithTexture(m_pobTextureAtlas->getTexture(), rect);
addChild(fontChar, i, i);
fontChar->release();
}
}
else
{
2012-11-15 17:16:51 +08:00
// updating previous sprite
fontChar->setTextureRect(rect, false, rect.size);
2010-12-24 17:07:31 +08:00
// restore to default in case they were modified
fontChar->setVisible(true);
fontChar->setOpacity(255);
}
2010-12-24 17:07:31 +08:00
// See issue 1343. cast( signed short + unsigned integer ) == unsigned integer (sign is lost!)
int yOffset = m_pConfiguration->m_nCommonHeight - fontDef.yOffset;
CCPoint fontPos = ccp( (float)nextFontPositionX + fontDef.xOffset + fontDef.rect.size.width*0.5f + kerningAmount,
(float)nextFontPositionY + yOffset - rect.size.height*0.5f * CC_CONTENT_SCALE_FACTOR() );
fontChar->setPosition(CC_POINT_PIXELS_TO_POINTS(fontPos));
// update kerning
nextFontPositionX += fontDef.xAdvance + kerningAmount;
//prev = c;
// Apply label properties
fontChar->setOpacityModifyRGB(m_bIsOpacityModifyRGB);
// Color MUST be set before opacity, since opacity might change color if OpacityModifyRGB is on
fontChar->setColor(m_tColor);
// only apply opacity if it is different than 255 )
// to prevent modifying the color too (issue #610)
if( m_cOpacity != 255 )
{
fontChar->setOpacity(m_cOpacity);
}
2010-08-12 17:39:25 +08:00
if (longestLine < nextFontPositionX)
{
longestLine = nextFontPositionX;
}
2012-11-15 17:16:51 +08:00
if (! hasSprite)
{
updateQuadFromSprite(fontChar, i);
}
}
2010-12-28 15:05:55 +08:00
2012-11-15 17:16:51 +08:00
// If the last character processed has an xAdvance which is less that the width of the characters image, then we need
// to adjust the width of the string to take this into account, or the character will overlap the end of the bounding
// box
if (fontDef.xAdvance < fontDef.rect.size.width)
{
tmpSize.width = longestLine + fontDef.rect.size.width - fontDef.xAdvance;
}
else
{
tmpSize.width = longestLine;
}
tmpSize.height = totalHeight;
this->setContentSize(CC_SIZE_PIXELS_TO_POINTS(tmpSize));
}
//LabelBMFont - CCLabelProtocol protocol
void CCLabelBMFont::setString(const char *newString)
2012-03-22 09:46:07 +08:00
{
this->setString(newString, false);
2012-03-22 09:46:07 +08:00
}
void CCLabelBMFont::setString(const char *newString, bool fromUpdate)
{
CC_SAFE_DELETE_ARRAY(m_sString);
2012-05-31 15:32:18 +08:00
m_sString = cc_utf16_from_utf8(newString);
// MARMALADE CHANGE
// THE ASSIGMENT OF STRINGS BELOW PERFORMS AN OVERLAPPING MEMCPY, WHEN fromUpdate IS TRUE
// ADDED THE "IF" LINE TO AVOID THIS
if (strcmp(m_sInitialString.c_str(), newString))
{
m_sInitialString = newString;
}
updateString(fromUpdate);
2012-03-22 09:46:07 +08:00
}
void CCLabelBMFont::updateString(bool fromUpdate)
{
if (m_pChildren && m_pChildren->count() != 0)
{
CCObject* child;
CCARRAY_FOREACH(m_pChildren, child)
{
CCNode* pNode = (CCNode*) child;
if (pNode)
{
pNode->setVisible(false);
}
}
}
this->createFontChars();
if (!fromUpdate)
updateLabel();
}
const char* CCLabelBMFont::getString(void)
{
return m_sInitialString.c_str();
}
void CCLabelBMFont::setCString(const char *label)
{
setString(label);
}
//LabelBMFont - CCRGBAProtocol protocol
void CCLabelBMFont::setColor(const ccColor3B& var)
{
m_tColor = var;
if (m_pChildren && m_pChildren->count() != 0)
{
CCObject* child;
CCARRAY_FOREACH(m_pChildren, child)
{
CCSprite* pNode = (CCSprite*) child;
if (pNode)
{
pNode->setColor(m_tColor);
}
}
}
}
ccColor3B CCLabelBMFont::getColor()
{
return m_tColor;
}
void CCLabelBMFont::setOpacity(GLubyte var)
{
m_cOpacity = var;
2010-08-12 17:39:25 +08:00
if (m_pChildren && m_pChildren->count() != 0)
{
CCObject* child;
CCARRAY_FOREACH(m_pChildren, child)
{
CCNode* pNode = (CCNode*) child;
if (pNode)
{
CCRGBAProtocol *pRGBAProtocol = dynamic_cast<CCRGBAProtocol*>(pNode);
if (pRGBAProtocol)
{
pRGBAProtocol->setOpacity(m_cOpacity);
}
}
}
}
}
GLubyte CCLabelBMFont::getOpacity()
{
return m_cOpacity;
}
void CCLabelBMFont::setOpacityModifyRGB(bool var)
{
m_bIsOpacityModifyRGB = var;
if (m_pChildren && m_pChildren->count() != 0)
{
CCObject* child;
CCARRAY_FOREACH(m_pChildren, child)
{
CCNode* pNode = (CCNode*) child;
if (pNode)
{
CCRGBAProtocol *pRGBAProtocol = dynamic_cast<CCRGBAProtocol*>(pNode);
if (pRGBAProtocol)
{
pRGBAProtocol->setOpacityModifyRGB(m_bIsOpacityModifyRGB);
}
}
}
}
}
bool CCLabelBMFont::isOpacityModifyRGB()
{
return m_bIsOpacityModifyRGB;
}
2010-08-12 17:39:25 +08:00
// LabelBMFont - AnchorPoint
void CCLabelBMFont::setAnchorPoint(const CCPoint& point)
{
2012-11-14 18:05:15 +08:00
if( ! point.equals(m_obAnchorPoint))
{
CCSpriteBatchNode::setAnchorPoint(point);
updateLabel();
}
2012-03-22 09:46:07 +08:00
}
// LabelBMFont - Alignment
void CCLabelBMFont::updateLabel()
{
this->setString(m_sInitialString.c_str(), true);
2012-05-31 15:32:18 +08:00
if (m_fWidth > 0)
{
// Step 1: Make multiline
2012-05-31 15:32:18 +08:00
vector<unsigned short> str_whole = cc_utf16_vec_from_utf16_str(m_sString);
unsigned int stringLength = str_whole.size();
vector<unsigned short> multiline_string;
multiline_string.reserve( stringLength );
vector<unsigned short> last_word;
last_word.reserve( stringLength );
unsigned int line = 1, i = 0;
bool start_line = false, start_word = false;
float startOfLine = -1, startOfWord = -1;
int skip = 0;
CCArray* children = getChildren();
for (unsigned int j = 0; j < children->count(); j++)
{
CCSprite* characterSprite;
while (!(characterSprite = (CCSprite*)this->getChildByTag(j + skip)))
skip++;
2012-11-15 17:16:51 +08:00
if (!characterSprite->isVisible())
continue;
if (i >= stringLength)
break;
unsigned short character = str_whole[i];
if (!start_word)
{
startOfWord = getLetterPosXLeft( characterSprite );
start_word = true;
}
if (!start_line)
{
startOfLine = startOfWord;
start_line = true;
}
// Newline.
if (character == '\n')
{
cc_utf8_trim_ws(&last_word);
last_word.push_back('\n');
multiline_string.insert(multiline_string.end(), last_word.begin(), last_word.end());
last_word.clear();
start_word = false;
start_line = false;
startOfWord = -1;
startOfLine = -1;
i++;
line++;
if (i >= stringLength)
break;
character = str_whole[i];
if (!startOfWord)
{
startOfWord = getLetterPosXLeft( characterSprite );
start_word = true;
}
if (!startOfLine)
{
startOfLine = startOfWord;
start_line = true;
}
}
// Whitespace.
if (isspace_unicode(character))
{
last_word.push_back(character);
multiline_string.insert(multiline_string.end(), last_word.begin(), last_word.end());
last_word.clear();
start_word = false;
startOfWord = -1;
i++;
continue;
}
// Out of bounds.
if ( getLetterPosXRight( characterSprite ) - startOfLine > m_fWidth )
{
if (!m_bLineBreakWithoutSpaces)
{
last_word.push_back(character);
int found = cc_utf8_find_last_not_char(multiline_string, ' ');
if (found != -1)
cc_utf8_trim_ws(&multiline_string);
else
multiline_string.clear();
if (multiline_string.size() > 0)
multiline_string.push_back('\n');
line++;
start_line = false;
startOfLine = -1;
i++;
}
else
{
cc_utf8_trim_ws(&last_word);
last_word.push_back('\n');
multiline_string.insert(multiline_string.end(), last_word.begin(), last_word.end());
last_word.clear();
start_word = false;
start_line = false;
startOfWord = -1;
startOfLine = -1;
line++;
if (i >= stringLength)
break;
if (!startOfWord)
{
startOfWord = getLetterPosXLeft( characterSprite );
start_word = true;
}
if (!startOfLine)
{
startOfLine = startOfWord;
start_line = true;
}
j--;
}
continue;
}
else
{
// Character is normal.
last_word.push_back(character);
i++;
continue;
}
}
multiline_string.insert(multiline_string.end(), last_word.begin(), last_word.end());
int size = multiline_string.size();
unsigned short* str_new = new unsigned short[size + 1];
for (int i = 0; i < size; ++i)
{
str_new[i] = multiline_string[i];
}
str_new[size] = 0;
CC_SAFE_DELETE_ARRAY(m_sString);
m_sString = str_new;
updateString(true);
}
// Step 2: Make alignment
if (m_pAlignment != kCCTextAlignmentLeft)
{
int i = 0;
int lineNumber = 0;
int str_len = cc_wcslen(m_sString);
vector<unsigned short> last_line;
for (int ctr = 0; ctr <= str_len; ++ctr)
{
if (m_sString[ctr] == '\n' || m_sString[ctr] == 0)
{
float lineWidth = 0.0f;
unsigned int line_length = last_line.size();
2012-09-06 14:09:29 +08:00
// if last line is empty we must just increase lineNumber and work with next line
if (line_length == 0)
{
lineNumber++;
continue;
}
int index = i + line_length - 1 + lineNumber;
if (index < 0) continue;
CCSprite* lastChar = (CCSprite*)getChildByTag(index);
if ( lastChar == NULL )
continue;
lineWidth = lastChar->getPosition().x + lastChar->getContentSize().width/2.0f;
float shift = 0;
switch (m_pAlignment)
{
case kCCTextAlignmentCenter:
shift = getContentSize().width/2.0f - lineWidth/2.0f;
break;
case kCCTextAlignmentRight:
shift = getContentSize().width - lineWidth;
break;
default:
break;
}
if (shift != 0)
{
for (unsigned j = 0; j < line_length; j++)
{
index = i + j + lineNumber;
if (index < 0) continue;
CCSprite* characterSprite = (CCSprite*)getChildByTag(index);
characterSprite->setPosition(ccpAdd(characterSprite->getPosition(), ccp(shift, 0.0f)));
}
}
i += line_length;
lineNumber++;
last_line.clear();
continue;
}
last_line.push_back(m_sString[ctr]);
}
}
2012-03-22 09:46:07 +08:00
}
// LabelBMFont - Alignment
void CCLabelBMFont::setAlignment(CCTextAlignment alignment)
{
this->m_pAlignment = alignment;
updateLabel();
2012-03-22 09:46:07 +08:00
}
void CCLabelBMFont::setWidth(float width)
{
this->m_fWidth = width;
updateLabel();
2012-03-22 09:46:07 +08:00
}
void CCLabelBMFont::setLineBreakWithoutSpace( bool breakWithoutSpace )
{
m_bLineBreakWithoutSpaces = breakWithoutSpace;
updateLabel();
}
2010-08-12 17:39:25 +08:00
void CCLabelBMFont::setScale(float scale)
{
CCSpriteBatchNode::setScale(scale);
updateLabel();
}
void CCLabelBMFont::setScaleX(float scaleX)
{
CCSpriteBatchNode::setScaleX(scaleX);
updateLabel();
}
void CCLabelBMFont::setScaleY(float scaleY)
{
CCSpriteBatchNode::setScaleY(scaleY);
updateLabel();
}
float CCLabelBMFont::getLetterPosXLeft( CCSprite* sp )
{
return sp->getPosition().x * m_fScaleX - (sp->getContentSize().width * m_fScaleX * sp->getAnchorPoint().x);
}
float CCLabelBMFont::getLetterPosXRight( CCSprite* sp )
{
return sp->getPosition().x * m_fScaleX + (sp->getContentSize().width * m_fScaleX * sp->getAnchorPoint().x);
}
// LabelBMFont - FntFile
void CCLabelBMFont::setFntFile(const char* fntFile)
{
if (fntFile != NULL && strcmp(fntFile, m_sFntFile.c_str()) != 0 )
{
CCBMFontConfiguration *newConf = FNTConfigLoadFile(fntFile);
CCAssert( newConf, "CCLabelBMFont: Impossible to create font. Please check file");
m_sFntFile = fntFile;
CC_SAFE_RETAIN(newConf);
CC_SAFE_RELEASE(m_pConfiguration);
m_pConfiguration = newConf;
this->setTexture(CCTextureCache::sharedTextureCache()->addImage(m_pConfiguration->getAtlasName()));
this->createFontChars();
}
}
const char* CCLabelBMFont::getFntFile()
{
return m_sFntFile.c_str();
}
//LabelBMFont - Debug draw
#if CC_LABELBMFONT_DEBUG_DRAW
void CCLabelBMFont::draw()
{
CCSpriteBatchNode::draw();
CCSize s = this->getContentSize();
CCPoint vertices[4]={
ccp(0,0),ccp(s.width,0),
ccp(s.width,s.height),ccp(0,s.height),
};
ccDrawPoly(vertices, 4, true);
}
2012-03-22 09:46:07 +08:00
#endif // CC_LABELBMFONT_DEBUG_DRAW
NS_CC_END