summaryrefslogtreecommitdiff
path: root/sc/addin/util/cl2c.pl
diff options
context:
space:
mode:
Diffstat (limited to 'sc/addin/util/cl2c.pl')
-rw-r--r--sc/addin/util/cl2c.pl242
1 files changed, 242 insertions, 0 deletions
diff --git a/sc/addin/util/cl2c.pl b/sc/addin/util/cl2c.pl
new file mode 100644
index 000000000000..32a3a6cda312
--- /dev/null
+++ b/sc/addin/util/cl2c.pl
@@ -0,0 +1,242 @@
+#!/usr/solar/bin/perl
+
+##########################################################################
+#
+# DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+#
+# Copyright 2008 by Sun Microsystems, Inc.
+#
+# OpenOffice.org - a multi-platform office productivity suite
+#
+# $RCSfile: cl2c.pl,v $
+#
+# $Revision: 1.3 $
+#
+# This file is part of OpenOffice.org.
+#
+# OpenOffice.org is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License version 3
+# only, as published by the Free Software Foundation.
+#
+# OpenOffice.org is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU Lesser General Public License version 3 for more details
+# (a copy is included in the LICENSE file that accompanied this code).
+#
+# You should have received a copy of the GNU Lesser General Public License
+# version 3 along with OpenOffice.org. If not, see
+# <http://www.openoffice.org/license.html>
+# for a copy of the LGPLv3 License.
+#
+##########################################################################
+
+if ( $#ARGV != 3 ) {
+ print STDERR "usage: cl2c.pl <file.cl> <file.c> <file.src> <resname>\n";
+ exit -1;
+}
+
+$CL=$ARGV[0];
+$C=$ARGV[1];
+$SRC=$ARGV[2];
+$RNAME=$ARGV[3];
+
+sub sconv
+{
+ local($s)=@_[0];
+ local($o,$c);
+ $_="";
+ foreach $o ( unpack("C*",$s) ) {
+ $c=chr($o);
+ if ( $o >= 32 && $o < 127 ) {
+ $_ .= $c;
+ } else {
+ $_ .= sprintf("\\%o", $o);
+ }
+ }
+ return $_;
+}
+
+
+sub makeneutral {
+
+ print COUT "\n\n/**\n";
+ print COUT " * Get neutral language for specific language.\n";
+ print COUT " * This simplifies the getText switch cases and allows to handle\n";
+ print COUT " * previously unknown language derivates due to foreign installations.\n";
+ print COUT " * If you want to distinguish between some dialects change this function\n";
+ print COUT " * to return the desired nLang before doing the bit masking stuff.\n";
+ print COUT " * See xlang.h for defined LANGUAGE_*\n";
+ print COUT " */\n";
+
+ # taken from tools/source/intntl/intn.cxx International::GetNeutralLanguage
+ print COUT "static USHORT GetNeutralLanguage( USHORT nLang )\n";
+ print COUT "{\n";
+ print COUT "\tUSHORT nPrimLang;\n";
+ print COUT "\n";
+ print COUT "\t/* ignore LANGUAGE_USER* */\n";
+ print COUT "\tif ( (nLang & 0x03FF) >= 0x0200 )\n";
+ print COUT "\t return nLang;\n";
+ print COUT "\n";
+ print COUT "\tnLang &= 0x03FF;\n";
+ print COUT "\n";
+ print COUT "\tnPrimLang = nLang | 0x0400;\n";
+ print COUT "\n";
+ print COUT "\tswitch ( nPrimLang )\n";
+ print COUT "\t{\n";
+ print COUT "\t\tcase LANGUAGE_CHINESE_TRADITIONAL:\n";
+ print COUT "\t\t\tnLang = LANGUAGE_CHINESE;\n";
+ print COUT "\t\t\tbreak;\n";
+ print COUT "\t\tcase LANGUAGE_ENGLISH_US:\n";
+ print COUT "\t\t\tnLang = LANGUAGE_ENGLISH;\n";
+ print COUT "\t\t\tbreak;\n";
+ print COUT "\t\tcase LANGUAGE_NORWEGIAN_BOKMAL:\n";
+ print COUT "\t\t\tnLang = LANGUAGE_NORWEGIAN;\n";
+ print COUT "\t\t\tbreak;\n";
+ print COUT "\t\tcase LANGUAGE_PORTUGUESE_BRAZILIAN:\n";
+ print COUT "\t\t\tnLang = LANGUAGE_PORTUGUESE;\n";
+ print COUT "\t\t\tbreak;\n";
+ print COUT "\n";
+ print COUT "\t\tdefault:\n";
+ print COUT "\t\t\tnLang = nPrimLang;\n";
+ print COUT "\t\t\tbreak;\n";
+ print COUT "\t}\n";
+ print COUT "\n";
+ print COUT "\treturn nLang;\n";
+ print COUT "}\n";
+ print COUT "\n";
+
+}
+
+
+sub maketext {
+
+ print COUT "\n\n/**\n";
+ print COUT " * Get text resource for current language.\n";
+ print COUT " * Remember that 8-bit characters are shown in\n";
+ print COUT " * system dependend code pages!\n";
+ print COUT " * To get correct results you will have to distuinguish\n";
+ print COUT " * for example between UNIX and WIN and OS2 target systems.\n";
+ print COUT " */\n";
+
+ print COUT "static char* getText( int nResource )\n{\n";
+ print COUT "\tswitch( nResource ) {\n";
+
+ $resflag=0;
+ $strname="";
+ $cnt=0;
+ $text_english="";
+
+ while (<SRCIN>) {
+ $resflag=1 if ( /Resource\s$RNAME/ );
+
+ if ( /\{/ ) {
+ if ( ++$cnt == 2 ) {
+ # start language
+ $text_english="";
+ print COUT "\t\t\tswitch( _nLanguage ) {\n";
+ next;
+ }
+ }
+
+ if ( /\}/ ) {
+ if ( --$cnt == 1 ) {
+ # end language
+
+ if ( $text_english ne "" ) {
+ print COUT "\t\t\t\tcase LANGUAGE_ENGLISH:\n\t\t\t\tdefault:\n";
+ print COUT "\t\t\t\treturn(" . $text_english . ")\;\n";
+ }
+
+ print COUT "\t\t\t}\n\t\t\tbreak;\n";
+ next;
+ } elsif ( $cnt == 0 ) {
+ # end of resource
+ $resflag=0;
+ print COUT "\t\tdefault:\n\t\t\tbreak;\n";
+ print COUT "\t}\n\treturn(\"\");\n}\n";
+ next;
+ }
+
+ }
+
+ if ( $resflag && $cnt == 1) {
+ if ( /\sString\s(([A-Z]|\_|[0-9]|[a-z])*)/ ) {
+ $strname=$1;
+ print COUT "\t\tcase " . $strname . ":\n";
+ }
+ }
+
+ if ( $cnt == 2 && /^\s*Text/ ) {
+ $langname="german";
+ ($textdef,@textx)=split(/=/);
+ $text=join("=",@textx);
+ if ( $textdef =~ /\[\s+(.*)\s+\]/ ) {
+ $langname=$1;
+ }
+ else {
+ $langname="ENGLISH_US"; # no [...] => not to be translated
+ }
+
+ $langname="LANGUAGE_" . uc($langname);
+
+ chop($text) while ( $text=~/(\r|\n|\;)$/ );
+ $text=sconv($text);
+ # english_us, not english because it's developer's pigeon
+ if ( $langname eq "LANGUAGE_ENGLISH_US" ) {
+ $text_english=$text;
+ }
+ # ISO coded, obtain at least the default
+ elsif ( $langname =~ /^LANGUAGE_EN-US$/ ) {
+ $text_english=$text;
+ }
+ # we don't know about USER languages, ENGLISH will be appended later
+ elsif ( ! ( $langname =~ /LANGUAGE_USER/ || $langname =~ /^LANGUAGE_ENGLISH$/ ) ) {
+ # ER 28.04.99: for the moment only German and English are
+ # exported, because we have a problem with non-existing
+ # character code tables for systems other than Windoze.
+ # => Chinese would be definitely mixed up and we don't
+ # want to insult anybody.. others like Spanish would look
+ # very ugly, but we'll have to live with bad German Umlauts.
+ if ( $langname =~ /LANGUAGE_(GERMAN|ENGLISH)/ ) {
+ print COUT "\t\t\t\tcase " . $langname . ":\n";
+ print COUT "\t\t\t\treturn(" . $text . ")\;\n";
+ }
+ }
+
+ }
+ }
+
+ makeneutral();
+
+}
+
+open(CLIN,"<$CL") || die "can not open $CL\n";
+open(SRCIN,"<$SRC") || die "can not open $CL\n";
+open(COUT,">$C") || die "can not open $CL\n";
+
+$ccnt=0;
+$incomment=0;
+while(<CLIN>) {
+ if ( /^\/\*--(-*)/ ) {
+ $incomment=1;
+ $ccnt++;
+ }
+
+ print COUT $_ if ( $incomment==0 || $ccnt==1 );
+
+ &maketext() if ( /^static USHORT _nLanguage=/ );
+
+ if ( /(-*)--\*\/$/ ) {
+ $incomment=0;
+ }
+
+}
+
+close(CLIN);
+close(SRCIN);
+close(COUT);
+
+exit 0;
+
+