summaryrefslogtreecommitdiffstats
path: root/squirrel_3_0_1_stable/_OLD_sqplus/SqPlusOCharBuf.h
diff options
context:
space:
mode:
Diffstat (limited to 'squirrel_3_0_1_stable/_OLD_sqplus/SqPlusOCharBuf.h')
-rw-r--r--squirrel_3_0_1_stable/_OLD_sqplus/SqPlusOCharBuf.h193
1 files changed, 0 insertions, 193 deletions
diff --git a/squirrel_3_0_1_stable/_OLD_sqplus/SqPlusOCharBuf.h b/squirrel_3_0_1_stable/_OLD_sqplus/SqPlusOCharBuf.h
deleted file mode 100644
index 24fe16b16..000000000
--- a/squirrel_3_0_1_stable/_OLD_sqplus/SqPlusOCharBuf.h
+++ /dev/null
@@ -1,193 +0,0 @@
-
-#ifndef SQPLUS_OCHARBUF_H
-#define SQPLUS_OCHARBUF_H
-
-// Conversion routines between different character encodings, to be used in
-// Push/Get/Match handlers of SqPlus. It enables using both of char* and
-// wchar_t* as function arguments.
-//
-// When converting from wchar_t to char, the choice is between using a Latin1
-// or a UTF8 representation in 8-bit mode. This is probably most useful if
-// Squirrel is compiled in ANSI mode and the app uses some wchar_t strings
-// (UNICODE mode on Win32).
-//
-
-// A char is either
-// - A Latin1 character (first code page, coincides with Unicode points 0..255)
-// - Part of an UTF8 character
-//
-// Note: SQUTF8 requires Squirrel sources that have been patched for UTF8 strings
-// internally (when converting function arguments).
-
-#ifndef SQUTF8
- // Default to Latin1 conversion
- //#undef SQPLUS_USE_LATIN1
- #ifndef SQPLUS_USE_LATIN1
- #define SQPLUS_USE_LATIN1 1
- #endif
-#else
- // Using Squirrel with internal UTF8 string encoding
- //#undef SQPLUS_USE_LATIN1
- #ifndef SQPLUS_USE_LATIN1
- #define SQPLUS_USE_LATIN1 0
- #endif
-#endif
-
-
-// The various ol_... functions sets up for a conversion scheme
-// that can be used in both wchar_t => char and char => wchar_t.
-int ol_strlen(const wchar_t* pwc);
-int ol_buflen(const wchar_t* pwc);
-int ol_charlen(const wchar_t* pwc);
-int ol_char(const wchar_t* pwc);
-bool ol_writechar(wchar_t*& pwc, int ch);
-
-int ol_strlen(const char* pc);
-int ol_buflen(const char* pc);
-int ol_charlen(const char* pc);
-int ol_char(const char* pc);
-bool ol_writechar(char*& pc, int ch);
-
-#ifdef SQUNICODE
- // SQChar is wchar_t, convert to/from either Latin1 or UTF8
- typedef char SQOtherChar;
-#else
- // SQChar is char, convert to/from wchar_t
- typedef wchar_t SQOtherChar;
-#endif
-
-
-// Buffer to hold a string and release it in destructor.
-// SQT is input character type
-// SQOT is the opposite type of SQChar
-// If 'try_borrow' is set, it will not allocate a new buffer when no
-// conversion is done, it will just keep the input pointer (beware).
-template<class SQT, class SQOT, bool use_latin1>
-struct SQCharBufImpl {
-
- // Allocate a buffer from string of same type
- SQCharBufImpl(const SQT *poc=NULL, bool try_borrow=true ) : m_poc(0), m_own(false) {
- Init( poc, try_borrow );
- }
-
- // This does conversion from other char type
- SQCharBufImpl(const SQOT *ps, bool try_borrow=false ) : m_poc(0), m_own(false) {
- Init( ps ); // Cannot borrow pointer when doing conversion
- }
-
- void Init( const SQT *poc, bool try_borrow ){
- m_own = !try_borrow;
- if( try_borrow ){
- m_poc = (SQT*)poc;
- }
- else {
- int sl = poc ? ol_buflen(poc) : 0;
- if( poc ){
- m_poc = poc ? new SQT[sl+1] : NULL;
- if( m_poc ) memcpy( m_poc, poc, (sl+1)*sizeof(SQT) );
- }
- else
- m_poc = NULL;
- }
- }
-
- void Init( const SQOT *ps ){
- m_own = true;
- if( ps ){
- int sl = ps ? ol_strlen(ps) : 0; // Length of string in characters (not bytes)
- int scl = 0; // Length of converted string in char/wchar_t count
- // How much storage needed?
- if( !use_latin1 && sizeof(SQT)<sizeof(SQOT) ){
- // Converting wchar_t => UTF8
- const SQOT *ps1 = ps; // It is wchar_t* here
- SQT tmp_buf[8];
- SQT *ptmp;
- while( *ps1 ){
- ptmp = tmp_buf;
- if( ol_writechar(ptmp,*ps1++) )
- scl += ol_charlen(tmp_buf);
- else
- scl++;
- }
- }
- else scl = sl; // Converting to wchar_t or Latin1, keep length
-
- m_poc = new SQT[scl+1];
- if( !m_poc ) return;
-
- // Convert
- SQT *poc = m_poc;
- while( *ps ){
- ol_writechar( poc, ol_char(ps) );
- ps += ol_charlen(ps);
- }
- *poc = 0; // Terminating zero
- }
- else
- m_poc = NULL;
- }
-
- ~SQCharBufImpl(){ Release(); }
-
- void Release( ){
- if(m_poc && m_own )
- delete [] m_poc;
- m_poc = NULL;
- }
-
- SQT* Set( const SQOT *ps ){
- Release( );
- Init( ps );
- return m_poc;
- }
-
- operator SQT*() const { return m_poc; }
-
- SQCharBufImpl<SQT,SQOT,use_latin1>& operator = (const SQT *ps){
- Release();
- Init( ps, false );
- return *this;
- }
-
- SQCharBufImpl<SQT,SQOT,use_latin1>& operator = (const SQOT *ps){
- Release();
- Init( ps );
- return *this;
- }
-
- // Move string from other here - Note: Empties the input argument (other)
- SQCharBufImpl<SQT,SQOT,use_latin1>& operator = (const SQCharBufImpl<SQT,SQOT,use_latin1>& other){
- Release();
- m_poc = other.m_poc;
- m_own = other.m_own;
- SQT **psqt = (SQT**)&other.m_poc;
- *psqt = NULL;
- return *this;
- }
-
-protected:
- SQT *m_poc;
- bool m_own;
-};
-
-typedef SQCharBufImpl<SQOtherChar, SQChar, bool(SQPLUS_USE_LATIN1)> SQOthCharBuf;
-typedef SQCharBufImpl<SQChar, SQOtherChar, bool(SQPLUS_USE_LATIN1)> SQDefCharBuf;
-typedef SQCharBufImpl<wchar_t, char, bool(SQPLUS_USE_LATIN1)> SQWCharBuf;
-typedef SQCharBufImpl<char, wchar_t, bool(SQPLUS_USE_LATIN1)> SQACharBuf;
-
-#ifdef SQUNICODE
- // SQChar is wchar_t, convert to/from to either Latin1 or UTF8
- #define WC2SQ(s) s
- #define SQ2WC(s) s
- SQDefCharBuf CH2SQ( const char* ps );
- SQOthCharBuf SQ2CH( const wchar_t* pws );
-#else
- // SQChar is char, convert to/from wchar_t
- SQDefCharBuf WC2SQ( const wchar_t* pws );
- SQOthCharBuf SQ2WC( const char* pws );
- #define CH2SQ(s) s
- #define SQ2CH(s) s
-#endif
-
-#endif // SQPLUS_CHARBUF_H
-