|
|
|
/**********************************************************************
|
|
|
|
** Copyright (C) 2005-2008 Trolltech ASA. All rights reserved.
|
|
|
|
**
|
|
|
|
** This file is part of TQt Linguist.
|
|
|
|
**
|
|
|
|
** This file may be used under the terms of the GNU General
|
|
|
|
** Public License versions 2.0 or 3.0 as published by the Free
|
|
|
|
** Software Foundation and appearing in the files LICENSE.GPL2
|
|
|
|
** and LICENSE.GPL3 included in the packaging of this file.
|
|
|
|
** Alternatively you may (at your option) use any later version
|
|
|
|
** of the GNU General Public License if such license has been
|
|
|
|
** publicly approved by Trolltech ASA (or its successors, if any)
|
|
|
|
** and the KDE Free TQt Foundation.
|
|
|
|
**
|
|
|
|
** Please review the following information to ensure GNU General
|
|
|
|
** Public Licensing requirements will be met:
|
|
|
|
** http://trolltech.com/products/qt/licenses/licensing/opensource/.
|
|
|
|
** If you are unsure which license is appropriate for your use, please
|
|
|
|
** review the following information:
|
|
|
|
** http://trolltech.com/products/qt/licenses/licensing/licensingoverview
|
|
|
|
** or contact the sales department at sales@trolltech.com.
|
|
|
|
**
|
|
|
|
** Licensees holding valid TQt Commercial licenses may use this file in
|
|
|
|
** accordance with the TQt Commercial License Agreement provided with
|
|
|
|
** the Software.
|
|
|
|
**
|
|
|
|
** This file is provided "AS IS" with NO WARRANTY OF ANY KIND,
|
|
|
|
** INCLUDING THE WARRANTIES OF DESIGN, MERCHANTABILITY AND FITNESS FOR
|
|
|
|
** A PARTICULAR PURPOSE. Trolltech reserves all rights not granted
|
|
|
|
** herein.
|
|
|
|
**
|
|
|
|
**********************************************************************/
|
|
|
|
|
|
|
|
#include <metatranslator.h>
|
|
|
|
|
|
|
|
#include <ntqfile.h>
|
|
|
|
#include <ntqregexp.h>
|
|
|
|
#include <ntqstring.h>
|
|
|
|
#include <ntqtextstream.h>
|
|
|
|
#include <ntqvaluestack.h>
|
|
|
|
#include <ntqxml.h>
|
|
|
|
|
|
|
|
#include <ctype.h>
|
|
|
|
#include <errno.h>
|
|
|
|
#include <stdio.h>
|
|
|
|
#include <string.h>
|
|
|
|
|
|
|
|
/* qmake ignore TQ_OBJECT */
|
|
|
|
|
|
|
|
static const char MagicComment[] = "TRANSLATOR ";
|
|
|
|
|
|
|
|
static TQMap<TQCString, int> needs_TQ_OBJECT;
|
|
|
|
static TQMap<TQCString, int> lacks_TQ_OBJECT;
|
|
|
|
|
|
|
|
/*
|
|
|
|
The first part of this source file is the C++ tokenizer. We skip
|
|
|
|
most of C++; the only tokens that interest us are defined here.
|
|
|
|
Thus, the code fragment
|
|
|
|
|
|
|
|
int main()
|
|
|
|
{
|
|
|
|
printf( "Hello, world!\n" );
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
is broken down into the following tokens (Tok_ omitted):
|
|
|
|
|
|
|
|
Ident Ident LeftParen RightParen
|
|
|
|
LeftBrace
|
|
|
|
Ident LeftParen String RightParen Semicolon
|
|
|
|
return Semicolon
|
|
|
|
RightBrace.
|
|
|
|
|
|
|
|
The 0 doesn't produce any token.
|
|
|
|
*/
|
|
|
|
|
|
|
|
enum { Tok_Eof, Tok_class, Tok_namespace, Tok_return, Tok_tr,
|
|
|
|
Tok_trUtf8, Tok_translate, Tok_TQ_OBJECT, Tok_Ident,
|
|
|
|
Tok_Comment, Tok_String, Tok_Arrow, Tok_Colon,
|
|
|
|
Tok_Gulbrandsen, Tok_LeftBrace, Tok_RightBrace, Tok_LeftParen,
|
|
|
|
Tok_RightParen, Tok_Comma, Tok_Semicolon };
|
|
|
|
|
|
|
|
/*
|
|
|
|
The tokenizer maintains the following global variables. The names
|
|
|
|
should be self-explanatory.
|
|
|
|
*/
|
|
|
|
static TQCString yyFileName;
|
|
|
|
static int yyCh;
|
|
|
|
static char yyIdent[128];
|
|
|
|
static size_t yyIdentLen;
|
|
|
|
static char yyComment[65536];
|
|
|
|
static size_t yyCommentLen;
|
|
|
|
static char yyString[65536];
|
|
|
|
static size_t yyStringLen;
|
|
|
|
static TQValueStack<int> yySavedBraceDepth;
|
|
|
|
static TQValueStack<int> yySavedParenDepth;
|
|
|
|
static int yyBraceDepth;
|
|
|
|
static int yyParenDepth;
|
|
|
|
static int yyLineNo;
|
|
|
|
static int yyCurLineNo;
|
|
|
|
static int yyBraceLineNo;
|
|
|
|
static int yyParenLineNo;
|
|
|
|
|
|
|
|
// the file to read from (if reading from a file)
|
|
|
|
static FILE *yyInFile;
|
|
|
|
|
|
|
|
// the string to read from and current position in the string (otherwise)
|
|
|
|
static TQString yyInStr;
|
|
|
|
static int yyInPos;
|
|
|
|
|
|
|
|
static int (*getChar)();
|
|
|
|
|
|
|
|
static int getCharFromFile()
|
|
|
|
{
|
|
|
|
int c = getc( yyInFile );
|
|
|
|
if ( c == '\n' )
|
|
|
|
yyCurLineNo++;
|
|
|
|
return c;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int getCharFromString()
|
|
|
|
{
|
|
|
|
if ( yyInPos == (int) yyInStr.length() ) {
|
|
|
|
return EOF;
|
|
|
|
} else {
|
|
|
|
return yyInStr[yyInPos++].latin1();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void startTokenizer( const char *fileName, int (*getCharFunc)() )
|
|
|
|
{
|
|
|
|
yyInPos = 0;
|
|
|
|
getChar = getCharFunc;
|
|
|
|
|
|
|
|
yyFileName = fileName;
|
|
|
|
yyCh = getChar();
|
|
|
|
yySavedBraceDepth.clear();
|
|
|
|
yySavedParenDepth.clear();
|
|
|
|
yyBraceDepth = 0;
|
|
|
|
yyParenDepth = 0;
|
|
|
|
yyCurLineNo = 1;
|
|
|
|
yyBraceLineNo = 1;
|
|
|
|
yyParenLineNo = 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int getToken()
|
|
|
|
{
|
|
|
|
const char tab[] = "abfnrtv";
|
|
|
|
const char backTab[] = "\a\b\f\n\r\t\v";
|
|
|
|
uint n;
|
|
|
|
|
|
|
|
yyIdentLen = 0;
|
|
|
|
yyCommentLen = 0;
|
|
|
|
yyStringLen = 0;
|
|
|
|
|
|
|
|
while ( yyCh != EOF ) {
|
|
|
|
yyLineNo = yyCurLineNo;
|
|
|
|
|
|
|
|
if ( isalpha(yyCh) || yyCh == '_' ) {
|
|
|
|
do {
|
|
|
|
if ( yyIdentLen < sizeof(yyIdent) - 1 )
|
|
|
|
yyIdent[yyIdentLen++] = (char) yyCh;
|
|
|
|
yyCh = getChar();
|
|
|
|
} while ( isalnum(yyCh) || yyCh == '_' );
|
|
|
|
yyIdent[yyIdentLen] = '\0';
|
|
|
|
|
|
|
|
switch ( yyIdent[0] ) {
|
|
|
|
case 'Q':
|
|
|
|
if ( strcmp(yyIdent + 1, "_OBJECT") == 0 ) {
|
|
|
|
return Tok_TQ_OBJECT;
|
|
|
|
} else if ( strcmp(yyIdent + 1, "T_TR_NOOP") == 0 ) {
|
|
|
|
return Tok_tr;
|
|
|
|
} else if ( strcmp(yyIdent + 1, "T_TRANSLATE_NOOP") == 0 ) {
|
|
|
|
return Tok_translate;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case 'T':
|
|
|
|
// TR() for when all else fails
|
|
|
|
if ( tqstricmp(yyIdent + 1, "R") == 0 )
|
|
|
|
return Tok_tr;
|
|
|
|
break;
|
|
|
|
case 'c':
|
|
|
|
if ( strcmp(yyIdent + 1, "lass") == 0 )
|
|
|
|
return Tok_class;
|
|
|
|
break;
|
|
|
|
case 'f':
|
|
|
|
/*
|
|
|
|
TQTranslator::findMessage() has the same parameters as
|
|
|
|
TQApplication::translate().
|
|
|
|
*/
|
|
|
|
if ( strcmp(yyIdent + 1, "indMessage") == 0 )
|
|
|
|
return Tok_translate;
|
|
|
|
break;
|
|
|
|
case 'n':
|
|
|
|
if ( strcmp(yyIdent + 1, "amespace") == 0 )
|
|
|
|
return Tok_namespace;
|
|
|
|
break;
|
|
|
|
case 'r':
|
|
|
|
if ( strcmp(yyIdent + 1, "eturn") == 0 )
|
|
|
|
return Tok_return;
|
|
|
|
break;
|
|
|
|
case 's':
|
|
|
|
if ( strcmp(yyIdent + 1, "truct") == 0 )
|
|
|
|
return Tok_class;
|
|
|
|
break;
|
|
|
|
case 't':
|
|
|
|
if ( strcmp(yyIdent + 1, "r") == 0 ) {
|
|
|
|
return Tok_tr;
|
|
|
|
} else if ( qstrcmp(yyIdent + 1, "rUtf8") == 0 ) {
|
|
|
|
return Tok_trUtf8;
|
|
|
|
} else if ( qstrcmp(yyIdent + 1, "ranslate") == 0 ) {
|
|
|
|
return Tok_translate;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return Tok_Ident;
|
|
|
|
} else {
|
|
|
|
switch ( yyCh ) {
|
|
|
|
case '#':
|
|
|
|
/*
|
|
|
|
Early versions of lupdate complained about
|
|
|
|
unbalanced braces in the following code:
|
|
|
|
|
|
|
|
#ifdef ALPHA
|
|
|
|
while ( beta ) {
|
|
|
|
#else
|
|
|
|
while ( gamma ) {
|
|
|
|
#endif
|
|
|
|
delta;
|
|
|
|
}
|
|
|
|
|
|
|
|
The code contains, indeed, two opening braces for
|
|
|
|
one closing brace; yet there's no reason to panic.
|
|
|
|
|
|
|
|
The solution is to remember yyBraceDepth as it was
|
|
|
|
when #if, #ifdef or #ifndef was met, and to set
|
|
|
|
yyBraceDepth to that value when meeting #elif or
|
|
|
|
#else.
|
|
|
|
*/
|
|
|
|
do {
|
|
|
|
yyCh = getChar();
|
|
|
|
} while ( isspace(yyCh) && yyCh != '\n' );
|
|
|
|
|
|
|
|
switch ( yyCh ) {
|
|
|
|
case 'i':
|
|
|
|
yyCh = getChar();
|
|
|
|
if ( yyCh == 'f' ) {
|
|
|
|
// if, ifdef, ifndef
|
|
|
|
yySavedBraceDepth.push( yyBraceDepth );
|
|
|
|
yySavedParenDepth.push( yyParenDepth );
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case 'e':
|
|
|
|
yyCh = getChar();
|
|
|
|
if ( yyCh == 'l' ) {
|
|
|
|
// elif, else
|
|
|
|
if ( !yySavedBraceDepth.isEmpty() ) {
|
|
|
|
yyBraceDepth = yySavedBraceDepth.top();
|
|
|
|
yyParenDepth = yySavedParenDepth.top();
|
|
|
|
}
|
|
|
|
} else if ( yyCh == 'n' ) {
|
|
|
|
// endif
|
|
|
|
if ( !yySavedBraceDepth.isEmpty() ) {
|
|
|
|
yySavedBraceDepth.pop();
|
|
|
|
yySavedParenDepth.pop();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
while ( isalnum(yyCh) || yyCh == '_' )
|
|
|
|
yyCh = getChar();
|
|
|
|
break;
|
|
|
|
case '/':
|
|
|
|
yyCh = getChar();
|
|
|
|
if ( yyCh == '/' ) {
|
|
|
|
do {
|
|
|
|
yyCh = getChar();
|
|
|
|
} while ( yyCh != EOF && yyCh != '\n' );
|
|
|
|
} else if ( yyCh == '*' ) {
|
|
|
|
bool metAster = FALSE;
|
|
|
|
bool metAsterSlash = FALSE;
|
|
|
|
|
|
|
|
while ( !metAsterSlash ) {
|
|
|
|
yyCh = getChar();
|
|
|
|
if ( yyCh == EOF ) {
|
|
|
|
fprintf( stderr,
|
|
|
|
"%s: Unterminated C++ comment starting at"
|
|
|
|
" line %d\n",
|
|
|
|
(const char *) yyFileName, yyLineNo );
|
|
|
|
yyComment[yyCommentLen] = '\0';
|
|
|
|
return Tok_Comment;
|
|
|
|
}
|
|
|
|
if ( yyCommentLen < sizeof(yyComment) - 1 )
|
|
|
|
yyComment[yyCommentLen++] = (char) yyCh;
|
|
|
|
|
|
|
|
if ( yyCh == '*' )
|
|
|
|
metAster = TRUE;
|
|
|
|
else if ( metAster && yyCh == '/' )
|
|
|
|
metAsterSlash = TRUE;
|
|
|
|
else
|
|
|
|
metAster = FALSE;
|
|
|
|
}
|
|
|
|
yyCh = getChar();
|
|
|
|
yyCommentLen -= 2;
|
|
|
|
yyComment[yyCommentLen] = '\0';
|
|
|
|
return Tok_Comment;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case '"':
|
|
|
|
yyCh = getChar();
|
|
|
|
|
|
|
|
while ( yyCh != EOF && yyCh != '\n' && yyCh != '"' ) {
|
|
|
|
if ( yyCh == '\\' ) {
|
|
|
|
yyCh = getChar();
|
|
|
|
|
|
|
|
if ( yyCh == '\n' ) {
|
|
|
|
yyCh = getChar();
|
|
|
|
} else if ( yyCh == 'x' ) {
|
|
|
|
TQCString hex = "0";
|
|
|
|
|
|
|
|
yyCh = getChar();
|
|
|
|
while ( isxdigit(yyCh) ) {
|
|
|
|
hex += (char) yyCh;
|
|
|
|
yyCh = getChar();
|
|
|
|
}
|
|
|
|
sscanf( hex, "%x", &n );
|
|
|
|
if ( yyStringLen < sizeof(yyString) - 1 )
|
|
|
|
yyString[yyStringLen++] = (char) n;
|
|
|
|
} else if ( yyCh >= '0' && yyCh < '8' ) {
|
|
|
|
TQCString oct = "";
|
|
|
|
|
|
|
|
do {
|
|
|
|
oct += (char) yyCh;
|
|
|
|
yyCh = getChar();
|
|
|
|
} while ( yyCh >= '0' && yyCh < '8' );
|
|
|
|
sscanf( oct, "%o", &n );
|
|
|
|
if ( yyStringLen < sizeof(yyString) - 1 )
|
|
|
|
yyString[yyStringLen++] = (char) n;
|
|
|
|
} else {
|
|
|
|
const char *p = strchr( tab, yyCh );
|
|
|
|
if ( yyStringLen < sizeof(yyString) - 1 )
|
|
|
|
yyString[yyStringLen++] = ( p == 0 ) ?
|
|
|
|
(char) yyCh : backTab[p - tab];
|
|
|
|
yyCh = getChar();
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if ( yyStringLen < sizeof(yyString) - 1 )
|
|
|
|
yyString[yyStringLen++] = (char) yyCh;
|
|
|
|
yyCh = getChar();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
yyString[yyStringLen] = '\0';
|
|
|
|
|
|
|
|
if ( yyCh != '"' )
|
|
|
|
tqWarning( "%s:%d: Unterminated C++ string",
|
|
|
|
(const char *) yyFileName, yyLineNo );
|
|
|
|
|
|
|
|
if ( yyCh == EOF ) {
|
|
|
|
return Tok_Eof;
|
|
|
|
} else {
|
|
|
|
yyCh = getChar();
|
|
|
|
return Tok_String;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case '-':
|
|
|
|
yyCh = getChar();
|
|
|
|
if ( yyCh == '>' ) {
|
|
|
|
yyCh = getChar();
|
|
|
|
return Tok_Arrow;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case ':':
|
|
|
|
yyCh = getChar();
|
|
|
|
if ( yyCh == ':' ) {
|
|
|
|
yyCh = getChar();
|
|
|
|
return Tok_Gulbrandsen;
|
|
|
|
}
|
|
|
|
return Tok_Colon;
|
|
|
|
case '\'':
|
|
|
|
yyCh = getChar();
|
|
|
|
if ( yyCh == '\\' )
|
|
|
|
yyCh = getChar();
|
|
|
|
|
|
|
|
do {
|
|
|
|
yyCh = getChar();
|
|
|
|
} while ( yyCh != EOF && yyCh != '\'' );
|
|
|
|
yyCh = getChar();
|
|
|
|
break;
|
|
|
|
case '{':
|
|
|
|
if (yyBraceDepth == 0)
|
|
|
|
yyBraceLineNo = yyCurLineNo;
|
|
|
|
yyBraceDepth++;
|
|
|
|
yyCh = getChar();
|
|
|
|
return Tok_LeftBrace;
|
|
|
|
case '}':
|
|
|
|
if (yyBraceDepth == 0)
|
|
|
|
yyBraceLineNo = yyCurLineNo;
|
|
|
|
yyBraceDepth--;
|
|
|
|
yyCh = getChar();
|
|
|
|
return Tok_RightBrace;
|
|
|
|
case '(':
|
|
|
|
if (yyParenDepth == 0)
|
|
|
|
yyParenLineNo = yyCurLineNo;
|
|
|
|
yyParenDepth++;
|
|
|
|
yyCh = getChar();
|
|
|
|
return Tok_LeftParen;
|
|
|
|
case ')':
|
|
|
|
if (yyParenDepth == 0)
|
|
|
|
yyParenLineNo = yyCurLineNo;
|
|
|
|
yyParenDepth--;
|
|
|
|
yyCh = getChar();
|
|
|
|
return Tok_RightParen;
|
|
|
|
case ',':
|
|
|
|
yyCh = getChar();
|
|
|
|
return Tok_Comma;
|
|
|
|
case ';':
|
|
|
|
yyCh = getChar();
|
|
|
|
return Tok_Semicolon;
|
|
|
|
default:
|
|
|
|
yyCh = getChar();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return Tok_Eof;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
The second part of this source file is the parser. It accomplishes
|
|
|
|
a very easy task: It finds all strings inside a tr() or translate()
|
|
|
|
call, and possibly finds out the context of the call. It supports
|
|
|
|
three cases: (1) the context is specified, as in
|
|
|
|
FunnyDialog::tr("Hello") or translate("FunnyDialog", "Hello");
|
|
|
|
(2) the call appears within an inlined function; (3) the call
|
|
|
|
appears within a function defined outside the class definition.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int yyTok;
|
|
|
|
|
|
|
|
static bool match( int t )
|
|
|
|
{
|
|
|
|
bool matches = ( yyTok == t );
|
|
|
|
if ( matches )
|
|
|
|
yyTok = getToken();
|
|
|
|
return matches;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool matchString( TQCString *s )
|
|
|
|
{
|
|
|
|
bool matches = ( yyTok == Tok_String );
|
|
|
|
*s = "";
|
|
|
|
while ( yyTok == Tok_String ) {
|
|
|
|
*s += yyString;
|
|
|
|
yyTok = getToken();
|
|
|
|
}
|
|
|
|
return matches;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool matchEncoding( bool *utf8 )
|
|
|
|
{
|
|
|
|
if ( yyTok == Tok_Ident ) {
|
|
|
|
if ( strcmp(yyIdent, "TQApplication") == 0 ) {
|
|
|
|
yyTok = getToken();
|
|
|
|
if ( yyTok == Tok_Gulbrandsen )
|
|
|
|
yyTok = getToken();
|
|
|
|
}
|
|
|
|
*utf8 = TQString( yyIdent ).endsWith( TQString("UTF8") );
|
|
|
|
yyTok = getToken();
|
|
|
|
return TRUE;
|
|
|
|
} else {
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void parse( MetaTranslator *tor, const char *initialContext,
|
|
|
|
const char *defaultContext )
|
|
|
|
{
|
|
|
|
TQMap<TQCString, TQCString> qualifiedContexts;
|
|
|
|
TQStringList namespaces;
|
|
|
|
TQCString context;
|
|
|
|
TQCString text;
|
|
|
|
TQCString com;
|
|
|
|
TQCString functionContext = initialContext;
|
|
|
|
TQCString prefix;
|
|
|
|
bool utf8 = FALSE;
|
|
|
|
bool missing_TQ_OBJECT = FALSE;
|
|
|
|
|
|
|
|
yyTok = getToken();
|
|
|
|
while ( yyTok != Tok_Eof ) {
|
|
|
|
switch ( yyTok ) {
|
|
|
|
case Tok_class:
|
|
|
|
/*
|
|
|
|
Partial support for inlined functions.
|
|
|
|
*/
|
|
|
|
yyTok = getToken();
|
|
|
|
if ( yyBraceDepth == (int) namespaces.count() &&
|
|
|
|
yyParenDepth == 0 ) {
|
|
|
|
do {
|
|
|
|
/*
|
|
|
|
This code should execute only once, but we play
|
|
|
|
safe with impure definitions such as
|
|
|
|
'class Q_EXPORT TQMessageBox', in which case
|
|
|
|
'TQMessageBox' is the class name, not 'Q_EXPORT'.
|
|
|
|
*/
|
|
|
|
functionContext = yyIdent;
|
|
|
|
yyTok = getToken();
|
|
|
|
} while ( yyTok == Tok_Ident );
|
|
|
|
|
|
|
|
while ( yyTok == Tok_Gulbrandsen ) {
|
|
|
|
yyTok = getToken();
|
|
|
|
functionContext += "::";
|
|
|
|
functionContext += yyIdent;
|
|
|
|
yyTok = getToken();
|
|
|
|
}
|
|
|
|
|
|
|
|
if ( yyTok == Tok_Colon ) {
|
|
|
|
missing_TQ_OBJECT = TRUE;
|
|
|
|
} else {
|
|
|
|
functionContext = defaultContext;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case Tok_namespace:
|
|
|
|
yyTok = getToken();
|
|
|
|
if ( yyTok == Tok_Ident ) {
|
|
|
|
TQCString ns = yyIdent;
|
|
|
|
yyTok = getToken();
|
|
|
|
if ( yyTok == Tok_LeftBrace &&
|
|
|
|
yyBraceDepth == (int) namespaces.count() + 1 )
|
|
|
|
namespaces.append( TQString(ns) );
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case Tok_tr:
|
|
|
|
case Tok_trUtf8:
|
|
|
|
utf8 = ( yyTok == Tok_trUtf8 );
|
|
|
|
yyTok = getToken();
|
|
|
|
if ( match(Tok_LeftParen) && matchString(&text) ) {
|
|
|
|
com = "";
|
|
|
|
if ( match(Tok_RightParen) || (match(Tok_Comma) &&
|
|
|
|
matchString(&com) && match(Tok_RightParen)) ) {
|
|
|
|
if ( prefix.isNull() ) {
|
|
|
|
context = functionContext;
|
|
|
|
if ( !namespaces.isEmpty() )
|
|
|
|
context.prepend( (namespaces.join(TQString("::")) +
|
|
|
|
TQString("::")).latin1() );
|
|
|
|
} else {
|
|
|
|
context = prefix;
|
|
|
|
}
|
|
|
|
prefix = (const char *) 0;
|
|
|
|
|
|
|
|
if ( qualifiedContexts.contains(context) )
|
|
|
|
context = qualifiedContexts[context];
|
|
|
|
tor->insert( MetaTranslatorMessage(context, text, com,
|
|
|
|
TQString::null, utf8) );
|
|
|
|
|
|
|
|
if ( lacks_TQ_OBJECT.contains(context) ) {
|
|
|
|
tqWarning( "%s:%d: Class '%s' lacks TQ_OBJECT macro",
|
|
|
|
(const char *) yyFileName, yyLineNo,
|
|
|
|
(const char *) context );
|
|
|
|
lacks_TQ_OBJECT.remove( context );
|
|
|
|
} else {
|
|
|
|
needs_TQ_OBJECT.insert( context, 0 );
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case Tok_translate:
|
|
|
|
utf8 = FALSE;
|
|
|
|
yyTok = getToken();
|
|
|
|
if ( match(Tok_LeftParen) &&
|
|
|
|
matchString(&context) &&
|
|
|
|
match(Tok_Comma) &&
|
|
|
|
matchString(&text) ) {
|
|
|
|
com = "";
|
|
|
|
if ( match(Tok_RightParen) ||
|
|
|
|
(match(Tok_Comma) &&
|
|
|
|
matchString(&com) &&
|
|
|
|
(match(Tok_RightParen) ||
|
|
|
|
match(Tok_Comma) &&
|
|
|
|
matchEncoding(&utf8) &&
|
|
|
|
match(Tok_RightParen))) )
|
|
|
|
tor->insert( MetaTranslatorMessage(context, text, com,
|
|
|
|
TQString::null, utf8) );
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case Tok_TQ_OBJECT:
|
|
|
|
missing_TQ_OBJECT = FALSE;
|
|
|
|
yyTok = getToken();
|
|
|
|
break;
|
|
|
|
case Tok_Ident:
|
|
|
|
if ( !prefix.isNull() )
|
|
|
|
prefix += "::";
|
|
|
|
prefix += yyIdent;
|
|
|
|
yyTok = getToken();
|
|
|
|
if ( yyTok != Tok_Gulbrandsen )
|
|
|
|
prefix = (const char *) 0;
|
|
|
|
break;
|
|
|
|
case Tok_Comment:
|
|
|
|
com = yyComment;
|
|
|
|
com = com.simplifyWhiteSpace();
|
|
|
|
if ( com.left(sizeof(MagicComment) - 1) == MagicComment ) {
|
|
|
|
com.remove( 0, sizeof(MagicComment) - 1 );
|
|
|
|
int k = com.find( ' ' );
|
|
|
|
if ( k == -1 ) {
|
|
|
|
context = com;
|
|
|
|
} else {
|
|
|
|
context = com.left( k );
|
|
|
|
com.remove( 0, k + 1 );
|
|
|
|
tor->insert( MetaTranslatorMessage(context, "", com,
|
|
|
|
TQString::null, FALSE) );
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
Provide a backdoor for people using "using
|
|
|
|
namespace". See the manual for details.
|
|
|
|
*/
|
|
|
|
k = 0;
|
|
|
|
while ( (k = context.find("::", k)) != -1 ) {
|
|
|
|
qualifiedContexts.insert( context.mid(k + 2), context );
|
|
|
|
k++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
yyTok = getToken();
|
|
|
|
break;
|
|
|
|
case Tok_Arrow:
|
|
|
|
yyTok = getToken();
|
|
|
|
if ( yyTok == Tok_tr || yyTok == Tok_trUtf8 )
|
|
|
|
tqWarning( "%s:%d: Cannot invoke tr() like this",
|
|
|
|
(const char *) yyFileName, yyLineNo );
|
|
|
|
break;
|
|
|
|
case Tok_Gulbrandsen:
|
|
|
|
// at top level?
|
|
|
|
if ( yyBraceDepth == (int) namespaces.count() && yyParenDepth == 0 )
|
|
|
|
functionContext = prefix;
|
|
|
|
yyTok = getToken();
|
|
|
|
break;
|
|
|
|
case Tok_RightBrace:
|
|
|
|
case Tok_Semicolon:
|
|
|
|
if ( yyBraceDepth >= 0 &&
|
|
|
|
yyBraceDepth + 1 == (int) namespaces.count() )
|
|
|
|
namespaces.remove( namespaces.fromLast() );
|
|
|
|
if ( yyBraceDepth == (int) namespaces.count() ) {
|
|
|
|
if ( missing_TQ_OBJECT ) {
|
|
|
|
if ( needs_TQ_OBJECT.contains(functionContext) ) {
|
|
|
|
tqWarning( "%s:%d: Class '%s' lacks TQ_OBJECT macro",
|
|
|
|
(const char *) yyFileName, yyLineNo,
|
|
|
|
(const char *) functionContext );
|
|
|
|
} else {
|
|
|
|
lacks_TQ_OBJECT.insert( functionContext, 0 );
|
|
|
|
}
|
|
|
|
}
|
|
|
|
functionContext = defaultContext;
|
|
|
|
missing_TQ_OBJECT = FALSE;
|
|
|
|
}
|
|
|
|
yyTok = getToken();
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
yyTok = getToken();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if ( yyBraceDepth != 0 )
|
|
|
|
fprintf( stderr,
|
|
|
|
"%s:%d: Unbalanced braces in C++ code (or abuse of the C++"
|
|
|
|
" preprocessor)\n",
|
|
|
|
(const char *)yyFileName, yyBraceLineNo );
|
|
|
|
else if ( yyParenDepth != 0 )
|
|
|
|
fprintf( stderr,
|
|
|
|
"%s:%d: Unbalanced parentheses in C++ code (or abuse of the C++"
|
|
|
|
" preprocessor)\n",
|
|
|
|
(const char *)yyFileName, yyParenLineNo );
|
|
|
|
}
|
|
|
|
|
|
|
|
void fetchtr_cpp( const char *fileName, MetaTranslator *tor,
|
|
|
|
const char *defaultContext, bool mustExist )
|
|
|
|
{
|
|
|
|
yyInFile = fopen( fileName, "r" );
|
|
|
|
if ( yyInFile == 0 ) {
|
|
|
|
if ( mustExist )
|
|
|
|
fprintf( stderr,
|
|
|
|
"lupdate error: Cannot open C++ source file '%s': %s\n",
|
|
|
|
fileName, strerror(errno) );
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
startTokenizer( fileName, getCharFromFile );
|
|
|
|
parse( tor, 0, defaultContext );
|
|
|
|
fclose( yyInFile );
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
In addition to C++, we support TQt Designer UI files.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
Fetches tr() calls in C++ code in UI files (inside "<function>"
|
|
|
|
tag). This mechanism is obsolete.
|
|
|
|
*/
|
|
|
|
void fetchtr_inlined_cpp( const char *fileName, const TQString& in,
|
|
|
|
MetaTranslator *tor, const char *context )
|
|
|
|
{
|
|
|
|
yyInStr = in;
|
|
|
|
startTokenizer( fileName, getCharFromString );
|
|
|
|
parse( tor, context, 0 );
|
|
|
|
yyInStr = TQString::null;
|
|
|
|
}
|
|
|
|
|
|
|
|
class UiHandler : public TQXmlDefaultHandler
|
|
|
|
{
|
|
|
|
public:
|
|
|
|
UiHandler( MetaTranslator *translator, const char *fileName )
|
|
|
|
: tor( translator ), fname( fileName ), comment( "" ) { }
|
|
|
|
|
|
|
|
virtual bool startElement( const TQString& namespaceURI,
|
|
|
|
const TQString& localName, const TQString& qName,
|
|
|
|
const TQXmlAttributes& atts );
|
|
|
|
virtual bool endElement( const TQString& namespaceURI,
|
|
|
|
const TQString& localName, const TQString& qName );
|
|
|
|
virtual bool characters( const TQString& ch );
|
|
|
|
virtual bool fatalError( const TQXmlParseException& exception );
|
|
|
|
|
|
|
|
private:
|
|
|
|
void flush();
|
|
|
|
|
|
|
|
MetaTranslator *tor;
|
|
|
|
TQCString fname;
|
|
|
|
TQString context;
|
|
|
|
TQString source;
|
|
|
|
TQString comment;
|
|
|
|
|
|
|
|
TQString accum;
|
|
|
|
};
|
|
|
|
|
|
|
|
bool UiHandler::startElement( const TQString& /* namespaceURI */,
|
|
|
|
const TQString& /* localName */,
|
|
|
|
const TQString& qName,
|
|
|
|
const TQXmlAttributes& atts )
|
|
|
|
{
|
|
|
|
if ( qName == TQString("item") ) {
|
|
|
|
flush();
|
|
|
|
if ( !atts.value(TQString("text")).isEmpty() )
|
|
|
|
source = atts.value( TQString("text") );
|
|
|
|
} else if ( qName == TQString("string") ) {
|
|
|
|
flush();
|
|
|
|
}
|
|
|
|
accum.truncate( 0 );
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool UiHandler::endElement( const TQString& /* namespaceURI */,
|
|
|
|
const TQString& /* localName */,
|
|
|
|
const TQString& qName )
|
|
|
|
{
|
|
|
|
accum.replace( TQRegExp(TQString("\r\n")), "\n" );
|
|
|
|
|
|
|
|
if ( qName == TQString("class") ) {
|
|
|
|
if ( context.isEmpty() )
|
|
|
|
context = accum;
|
|
|
|
} else if ( qName == TQString("string") ) {
|
|
|
|
source = accum;
|
|
|
|
} else if ( qName == TQString("comment") ) {
|
|
|
|
comment = accum;
|
|
|
|
flush();
|
|
|
|
} else if ( qName == TQString("function") ) {
|
|
|
|
fetchtr_inlined_cpp( (const char *) fname, accum, tor,
|
|
|
|
context.latin1() );
|
|
|
|
} else {
|
|
|
|
flush();
|
|
|
|
}
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool UiHandler::characters( const TQString& ch )
|
|
|
|
{
|
|
|
|
accum += ch;
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool UiHandler::fatalError( const TQXmlParseException& exception )
|
|
|
|
{
|
|
|
|
TQString msg;
|
|
|
|
msg.sprintf( "Parse error at line %d, column %d (%s).",
|
|
|
|
exception.lineNumber(), exception.columnNumber(),
|
|
|
|
exception.message().latin1() );
|
|
|
|
fprintf( stderr, "XML error: %s\n", msg.latin1() );
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
void UiHandler::flush()
|
|
|
|
{
|
|
|
|
if ( !context.isEmpty() && !source.isEmpty() )
|
|
|
|
tor->insert( MetaTranslatorMessage(context.utf8(), source.utf8(),
|
|
|
|
comment.utf8(), TQString::null,
|
|
|
|
TRUE) );
|
|
|
|
source.truncate( 0 );
|
|
|
|
comment.truncate( 0 );
|
|
|
|
}
|
|
|
|
|
|
|
|
void fetchtr_ui( const char *fileName, MetaTranslator *tor,
|
|
|
|
const char * /* defaultContext */, bool mustExist )
|
|
|
|
{
|
|
|
|
TQFile f( fileName );
|
|
|
|
if ( !f.open(IO_ReadOnly) ) {
|
|
|
|
if ( mustExist )
|
|
|
|
fprintf( stderr, "lupdate error: cannot open UI file '%s': %s\n",
|
|
|
|
fileName, strerror(errno) );
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
TQTextStream t( &f );
|
|
|
|
TQXmlInputSource in( t );
|
|
|
|
TQXmlSimpleReader reader;
|
|
|
|
reader.setFeature( "http://xml.org/sax/features/namespaces", FALSE );
|
|
|
|
reader.setFeature( "http://xml.org/sax/features/namespace-prefixes", TRUE );
|
|
|
|
reader.setFeature( "http://trolltech.com/xml/features/report-whitespace"
|
|
|
|
"-only-CharData", FALSE );
|
|
|
|
TQXmlDefaultHandler *hand = new UiHandler( tor, fileName );
|
|
|
|
reader.setContentHandler( hand );
|
|
|
|
reader.setErrorHandler( hand );
|
|
|
|
|
|
|
|
if ( !reader.parse(in) )
|
|
|
|
fprintf( stderr, "%s: Parse error in UI file\n", fileName );
|
|
|
|
reader.setContentHandler( 0 );
|
|
|
|
reader.setErrorHandler( 0 );
|
|
|
|
delete hand;
|
|
|
|
f.close();
|
|
|
|
}
|