D7net Mini Sh3LL v1
Current File : //lib/policykit-1/../../lib64/../share/fwupd/../doc/libpam0g/../../intltool-debian/intltool-merge |
#!/usr/bin/perl -w
# -*- Mode: perl; indent-tabs-mode: nil; c-basic-offset: 4 -*-
#
# The Intltool Message Merger
#
# Copyright (C) 2000, 2003 Free Software Foundation.
# Copyright (C) 2000, 2001 Eazel, Inc
#
# Intltool is free software; you can redistribute it and/or
# modify it under the terms of the GNU General Public License
# version 2 published by the Free Software Foundation.
#
# Intltool 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
# General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
#
# As a special exception to the GNU General Public License, if you
# distribute this file as part of a program that contains a
# configuration script generated by Autoconf, you may include it under
# the same distribution terms that you use for the rest of that program.
#
# Authors: Maciej Stachowiak <mjs@noisehavoc.org>
# Kenneth Christiansen <kenneth@gnu.org>
# Darin Adler <darin@bentspoon.com>
#
# Proper XML UTF-8'ification written by Cyrille Chepelov <chepelov@calixo.net>
#
## Release information
my $PROGRAM = "intltool-merge";
my $PACKAGE = "intltool";
my $VERSION = "0.35.0";
## Loaded modules
use strict;
use Getopt::Long;
use File::Basename;
my $must_end_tag = -1;
my $last_depth = -1;
my $translation_depth = -1;
my @tag_stack = ();
my @entered_tag = ();
my @translation_strings = ();
my $leading_space = "";
## Scalars used by the option stuff
my $HELP_ARG = 0;
my $VERSION_ARG = 0;
my $BA_STYLE_ARG = 0;
my $XML_STYLE_ARG = 0;
my $KEYS_STYLE_ARG = 0;
my $DESKTOP_STYLE_ARG = 0;
my $SCHEMAS_STYLE_ARG = 0;
my $RFC822DEB_STYLE_ARG = 0;
my $QUOTED_STYLE_ARG = 0;
my $QUIET_ARG = 0;
my $PASS_THROUGH_ARG = 0;
my $UTF8_ARG = 0;
my $MULTIPLE_OUTPUT = 0;
my $cache_file;
## Handle options
GetOptions
(
"help" => \$HELP_ARG,
"version" => \$VERSION_ARG,
"quiet|q" => \$QUIET_ARG,
"oaf-style|o" => \$BA_STYLE_ARG, ## for compatibility
"ba-style|b" => \$BA_STYLE_ARG,
"xml-style|x" => \$XML_STYLE_ARG,
"keys-style|k" => \$KEYS_STYLE_ARG,
"desktop-style|d" => \$DESKTOP_STYLE_ARG,
"schemas-style|s" => \$SCHEMAS_STYLE_ARG,
"rfc822deb-style|r" => \$RFC822DEB_STYLE_ARG,
"quoted-style" => \$QUOTED_STYLE_ARG,
"pass-through|p" => \$PASS_THROUGH_ARG,
"utf8|u" => \$UTF8_ARG,
"multiple-output|m" => \$MULTIPLE_OUTPUT,
"cache|c=s" => \$cache_file
) or &error;
my $PO_DIR;
my $FILE;
my $OUTFILE;
my %po_files_by_lang = ();
my %translations = ();
my $iconv = $ENV{"ICONV"} || $ENV{"INTLTOOL_ICONV"} || "/usr/bin/iconv";
my $devnull = ($^O eq 'MSWin32' ? 'NUL:' : '/dev/null');
# Use this instead of \w for XML files to handle more possible characters.
my $w = "[-A-Za-z0-9._:]";
# XML quoted string contents
my $q = "[^\\\"]*";
## Check for options.
if ($VERSION_ARG)
{
&print_version;
}
elsif ($HELP_ARG)
{
&print_help;
}
elsif ($BA_STYLE_ARG && @ARGV > 2)
{
&utf8_sanity_check;
&preparation;
&print_message;
&ba_merge_translations;
&finalize;
}
elsif ($XML_STYLE_ARG && @ARGV > 2)
{
&utf8_sanity_check;
&preparation;
&print_message;
&xml_merge_output;
&finalize;
}
elsif ($KEYS_STYLE_ARG && @ARGV > 2)
{
&utf8_sanity_check;
&preparation;
&print_message;
&keys_merge_translations;
&finalize;
}
elsif ($DESKTOP_STYLE_ARG && @ARGV > 2)
{
&utf8_sanity_check;
&preparation;
&print_message;
&desktop_merge_translations;
&finalize;
}
elsif ($SCHEMAS_STYLE_ARG && @ARGV > 2)
{
&utf8_sanity_check;
&preparation;
&print_message;
&schemas_merge_translations;
&finalize;
}
elsif ($RFC822DEB_STYLE_ARG && @ARGV > 2)
{
&preparation;
&rfc822deb_merge_translations;
&finalize;
}
elsif ($QUOTED_STYLE_ARG && @ARGV > 2)
{
&utf8_sanity_check;
&preparation;
&print_message;
"ed_merge_translations;
&finalize;
}
else
{
&print_help;
}
exit;
## Sub for printing release information
sub print_version
{
print <<_EOF_;
${PROGRAM} (${PACKAGE}) ${VERSION}
Written by Maciej Stachowiak, Darin Adler and Kenneth Christiansen.
Copyright (C) 2000-2003 Free Software Foundation, Inc.
Copyright (C) 2000-2001 Eazel, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
_EOF_
exit;
}
## Sub for printing usage information
sub print_help
{
print <<_EOF_;
Usage: ${PROGRAM} [OPTION]... PO_DIRECTORY FILENAME OUTPUT_FILE
Generates an output file that includes some localized attributes from an
untranslated source file.
Mandatory options: (exactly one must be specified)
-b, --ba-style includes translations in the bonobo-activation style
-d, --desktop-style includes translations in the desktop style
-k, --keys-style includes translations in the keys style
-s, --schemas-style includes translations in the schemas style
-r, --rfc822deb-style includes translations in the RFC822 style
--quoted-style includes translations in the quoted string style
-x, --xml-style includes translations in the standard xml style
Other options:
-u, --utf8 convert all strings to UTF-8 before merging
(default for everything except RFC822 style)
-p, --pass-through deprecated, does nothing and issues a warning
-m, --multiple-output output one localized file per locale, instead of
a single file containing all localized elements
-c, --cache=FILE specify cache file name
(usually \$top_builddir/po/.intltool-merge-cache)
-q, --quiet suppress most messages
--help display this help and exit
--version output version information and exit
Report bugs to http://bugzilla.gnome.org/ (product name "$PACKAGE")
or send email to <xml-i18n-tools\@gnome.org>.
_EOF_
exit;
}
## Sub for printing error messages
sub print_error
{
print STDERR "Try `${PROGRAM} --help' for more information.\n";
exit;
}
sub print_message
{
print "Merging translations into $OUTFILE.\n" unless $QUIET_ARG;
}
sub preparation
{
$PO_DIR = $ARGV[0];
$FILE = $ARGV[1];
$OUTFILE = $ARGV[2];
&gather_po_files;
&get_translation_database;
}
# General-purpose code for looking up translations in .po files
sub po_file2lang
{
my ($tmp) = @_;
$tmp =~ s/^.*\/(.*)\.po$/$1/;
return $tmp;
}
sub gather_po_files
{
for my $po_file (glob "$PO_DIR/*.po") {
$po_files_by_lang{po_file2lang($po_file)} = $po_file;
}
}
sub get_local_charset
{
my ($encoding) = @_;
my $alias_file = $ENV{"G_CHARSET_ALIAS"} || "/usr/lib/charset.alias";
# seek character encoding aliases in charset.alias (glib)
if (open CHARSET_ALIAS, $alias_file)
{
while (<CHARSET_ALIAS>)
{
next if /^\#/;
return $1 if (/^\s*([-._a-zA-Z0-9]+)\s+$encoding\b/i)
}
close CHARSET_ALIAS;
}
# if not found, return input string
return $encoding;
}
sub get_po_encoding
{
my ($in_po_file) = @_;
my $encoding = "";
open IN_PO_FILE, $in_po_file or die;
while (<IN_PO_FILE>)
{
## example: "Content-Type: text/plain; charset=ISO-8859-1\n"
if (/Content-Type\:.*charset=([-a-zA-Z0-9]+)\\n/)
{
$encoding = $1;
last;
}
}
close IN_PO_FILE;
if (!$encoding)
{
print STDERR "Warning: no encoding found in $in_po_file. Assuming ISO-8859-1\n" unless $QUIET_ARG;
$encoding = "ISO-8859-1";
}
system ("$iconv -f $encoding -t UTF-8 <$devnull 2>$devnull");
if ($?) {
$encoding = get_local_charset($encoding);
}
return $encoding
}
sub utf8_sanity_check
{
print STDERR "Warning: option --pass-through has been removed.\n" if $PASS_THROUGH_ARG;
$UTF8_ARG = 1;
}
sub get_translation_database
{
if ($cache_file) {
&get_cached_translation_database;
} else {
&create_translation_database;
}
}
sub get_newest_po_age
{
my $newest_age;
foreach my $file (values %po_files_by_lang)
{
my $file_age = -M $file;
$newest_age = $file_age if !$newest_age || $file_age < $newest_age;
}
$newest_age = 0 if !$newest_age;
return $newest_age;
}
sub create_cache
{
print "Generating and caching the translation database\n" unless $QUIET_ARG;
&create_translation_database;
open CACHE, ">$cache_file" || die;
print CACHE join "\x01", %translations;
close CACHE;
}
sub load_cache
{
print "Found cached translation database\n" unless $QUIET_ARG;
my $contents;
open CACHE, "<$cache_file" || die;
{
local $/;
$contents = <CACHE>;
}
close CACHE;
%translations = split "\x01", $contents;
}
sub get_cached_translation_database
{
my $cache_file_age = -M $cache_file;
if (defined $cache_file_age)
{
if ($cache_file_age <= &get_newest_po_age)
{
&load_cache;
return;
}
print "Found too-old cached translation database\n" unless $QUIET_ARG;
}
&create_cache;
}
sub create_translation_database
{
for my $lang (keys %po_files_by_lang)
{
my $po_file = $po_files_by_lang{$lang};
if ($UTF8_ARG)
{
my $encoding = get_po_encoding ($po_file);
if (lc $encoding eq "utf-8")
{
open PO_FILE, "<$po_file";
}
else
{
print "NOTICE: $po_file is not in UTF-8 but $encoding, converting...\n" unless $QUIET_ARG;;
open PO_FILE, "$iconv -f $encoding -t UTF-8 $po_file|";
}
}
else
{
open PO_FILE, "<$po_file";
}
my $nextfuzzy = 0;
my $inmsgid = 0;
my $inmsgstr = 0;
my $msgid = "";
my $msgstr = "";
while (<PO_FILE>)
{
$nextfuzzy = 1 if /^#, fuzzy/;
if (/^msgid "((\\.|[^\\])*)"/ )
{
$translations{$lang, $msgid} = $msgstr if $inmsgstr && $msgid && $msgstr;
$msgid = "";
$msgstr = "";
if ($nextfuzzy) {
$inmsgid = 0;
} else {
$msgid = unescape_po_string($1);
$inmsgid = 1;
}
$inmsgstr = 0;
$nextfuzzy = 0;
}
if (/^msgstr "((\\.|[^\\])*)"/)
{
$msgstr = unescape_po_string($1);
$inmsgstr = 1;
$inmsgid = 0;
}
if (/^"((\\.|[^\\])*)"/)
{
$msgid .= unescape_po_string($1) if $inmsgid;
$msgstr .= unescape_po_string($1) if $inmsgstr;
}
}
$translations{$lang, $msgid} = $msgstr if $inmsgstr && $msgid && $msgstr;
}
}
sub finalize
{
}
sub unescape_one_sequence
{
my ($sequence) = @_;
return "\\" if $sequence eq "\\\\";
return "\"" if $sequence eq "\\\"";
return "\n" if $sequence eq "\\n";
return "\r" if $sequence eq "\\r";
return "\t" if $sequence eq "\\t";
return "\b" if $sequence eq "\\b";
return "\f" if $sequence eq "\\f";
return "\a" if $sequence eq "\\a";
return chr(11) if $sequence eq "\\v"; # vertical tab, see ascii(7)
return chr(hex($1)) if ($sequence =~ /\\x([0-9a-fA-F]{2})/);
return chr(oct($1)) if ($sequence =~ /\\([0-7]{3})/);
# FIXME: Is \0 supported as well? Kenneth and Rodney don't want it, see bug #48489
return $sequence;
}
sub unescape_po_string
{
my ($string) = @_;
$string =~ s/(\\x[0-9a-fA-F]{2}|\\[0-7]{3}|\\.)/unescape_one_sequence($1)/eg;
return $string;
}
## NOTE: deal with < - < but not > - > because it seems its ok to have
## > in the entity. For further info please look at #84738.
sub entity_decode
{
local ($_) = @_;
s/'/'/g; # '
s/"/"/g; # "
s/&/&/g;
s/</</g;
return $_;
}
# entity_encode: (string)
#
# Encode the given string to XML format (encode '<' etc).
sub entity_encode
{
my ($pre_encoded) = @_;
my @list_of_chars = unpack ('C*', $pre_encoded);
# with UTF-8 we only encode minimalistic
return join ('', map (&entity_encode_int_minimalist, @list_of_chars));
}
sub entity_encode_int_minimalist
{
return """ if $_ == 34;
return "&" if $_ == 38;
return "'" if $_ == 39;
return "<" if $_ == 60;
return chr $_;
}
sub entity_encoded_translation
{
my ($lang, $string) = @_;
my $translation = $translations{$lang, $string};
return $string if !$translation;
return entity_encode ($translation);
}
## XML (bonobo-activation specific) merge code
sub ba_merge_translations
{
my $source;
{
local $/; # slurp mode
open INPUT, "<$FILE" or die "can't open $FILE: $!";
$source = <INPUT>;
close INPUT;
}
open OUTPUT, ">$OUTFILE" or die "can't open $OUTFILE: $!";
# Binmode so that selftest works ok if using a native Win32 Perl...
binmode (OUTPUT) if $^O eq 'MSWin32';
while ($source =~ s|^(.*?)([ \t]*<\s*$w+\s+($w+\s*=\s*"$q"\s*)+/?>)([ \t]*\n)?||s)
{
print OUTPUT $1;
my $node = $2 . "\n";
my @strings = ();
$_ = $node;
while (s/(\s)_($w+\s*=\s*"($q)")/$1$2/s) {
push @strings, entity_decode($3);
}
print OUTPUT;
my %langs;
for my $string (@strings)
{
for my $lang (keys %po_files_by_lang)
{
$langs{$lang} = 1 if $translations{$lang, $string};
}
}
for my $lang (sort keys %langs)
{
$_ = $node;
s/(\sname\s*=\s*)"($q)"/$1"$2-$lang"/s;
s/(\s)_($w+\s*=\s*")($q)"/$1 . $2 . entity_encoded_translation($lang, $3) . '"'/seg;
print OUTPUT;
}
}
print OUTPUT $source;
close OUTPUT;
}
## XML (non-bonobo-activation) merge code
# Process tag attributes
# Only parameter is a HASH containing attributes -> values mapping
sub getAttributeString
{
my $sub = shift;
my $do_translate = shift || 0;
my $language = shift || "";
my $result = "";
my $translate = shift;
foreach my $e (reverse(sort(keys %{ $sub }))) {
my $key = $e;
my $string = $sub->{$e};
my $quote = '"';
$string =~ s/^[\s]+//;
$string =~ s/[\s]+$//;
if ($string =~ /^'.*'$/)
{
$quote = "'";
}
$string =~ s/^['"]//g;
$string =~ s/['"]$//g;
if ($do_translate && $key =~ /^_/) {
$key =~ s|^_||g;
if ($language) {
# Handle translation
my $decode_string = entity_decode($string);
my $translation = $translations{$language, $decode_string};
if ($translation) {
$translation = entity_encode($translation);
$string = $translation;
}
$$translate = 2;
} else {
$$translate = 2 if ($translate && (!$$translate)); # watch not to "overwrite" $translate
}
}
$result .= " $key=$quote$string$quote";
}
return $result;
}
# Returns a translatable string from XML node, it works on contents of every node in XML::Parser tree
sub getXMLstring
{
my $ref = shift;
my $spacepreserve = shift || 0;
my @list = @{ $ref };
my $result = "";
my $count = scalar(@list);
my $attrs = $list[0];
my $index = 1;
$spacepreserve = 1 if ((exists $attrs->{"xml:space"}) && ($attrs->{"xml:space"} =~ /^["']?preserve["']?$/));
$spacepreserve = 0 if ((exists $attrs->{"xml:space"}) && ($attrs->{"xml:space"} =~ /^["']?default["']?$/));
while ($index < $count) {
my $type = $list[$index];
my $content = $list[$index+1];
if (! $type ) {
# We've got CDATA
if ($content) {
# lets strip the whitespace here, and *ONLY* here
$content =~ s/\s+/ /gs if (!$spacepreserve);
$result .= $content;
}
} elsif ( "$type" ne "1" ) {
# We've got another element
$result .= "<$type";
$result .= getAttributeString(@{$content}[0], 0); # no nested translatable elements
if ($content) {
my $subresult = getXMLstring($content, $spacepreserve);
if ($subresult) {
$result .= ">".$subresult . "</$type>";
} else {
$result .= "/>";
}
} else {
$result .= "/>";
}
}
$index += 2;
}
return $result;
}
# Translate list of nodes if necessary
sub translate_subnodes
{
my $fh = shift;
my $content = shift;
my $language = shift || "";
my $singlelang = shift || 0;
my $spacepreserve = shift || 0;
my @nodes = @{ $content };
my $count = scalar(@nodes);
my $index = 0;
while ($index < $count) {
my $type = $nodes[$index];
my $rest = $nodes[$index+1];
if ($singlelang) {
my $oldMO = $MULTIPLE_OUTPUT;
$MULTIPLE_OUTPUT = 1;
traverse($fh, $type, $rest, $language, $spacepreserve);
$MULTIPLE_OUTPUT = $oldMO;
} else {
traverse($fh, $type, $rest, $language, $spacepreserve);
}
$index += 2;
}
}
sub isWellFormedXmlFragment
{
my $ret = eval 'require XML::Parser';
if(!$ret) {
die "You must have XML::Parser installed to run $0\n\n";
}
my $fragment = shift;
return 0 if (!$fragment);
$fragment = "<root>$fragment</root>";
my $xp = new XML::Parser(Style => 'Tree');
my $tree = 0;
eval { $tree = $xp->parse($fragment); };
return $tree;
}
sub traverse
{
my $fh = shift;
my $nodename = shift;
my $content = shift;
my $language = shift || "";
my $spacepreserve = shift || 0;
if (!$nodename) {
if ($content =~ /^[\s]*$/) {
$leading_space .= $content;
}
print $fh $content;
} else {
# element
my @all = @{ $content };
my $attrs = shift @all;
my $translate = 0;
my $outattr = getAttributeString($attrs, 1, $language, \$translate);
if ($nodename =~ /^_/) {
$translate = 1;
$nodename =~ s/^_//;
}
my $lookup = '';
$spacepreserve = 0 if ((exists $attrs->{"xml:space"}) && ($attrs->{"xml:space"} =~ /^["']?default["']?$/));
$spacepreserve = 1 if ((exists $attrs->{"xml:space"}) && ($attrs->{"xml:space"} =~ /^["']?preserve["']?$/));
print $fh "<$nodename", $outattr;
if ($translate) {
$lookup = getXMLstring($content, $spacepreserve);
if (!$spacepreserve) {
$lookup =~ s/^\s+//s;
$lookup =~ s/\s+$//s;
}
if ($lookup || $translate == 2) {
my $translation = $translations{$language, $lookup} if isWellFormedXmlFragment($translations{$language, $lookup});
if ($MULTIPLE_OUTPUT && ($translation || $translate == 2)) {
$translation = $lookup if (!$translation);
print $fh " xml:lang=\"", $language, "\"" if $language;
print $fh ">";
if ($translate == 2) {
translate_subnodes($fh, \@all, $language, 1, $spacepreserve);
} else {
print $fh $translation;
}
print $fh "</$nodename>";
return; # this means there will be no same translation with xml:lang="$language"...
# if we want them both, just remove this "return"
} else {
print $fh ">";
if ($translate == 2) {
translate_subnodes($fh, \@all, $language, 1, $spacepreserve);
} else {
print $fh $lookup;
}
print $fh "</$nodename>";
}
} else {
print $fh "/>";
}
for my $lang (sort keys %po_files_by_lang) {
if ($MULTIPLE_OUTPUT && $lang ne "$language") {
next;
}
if ($lang) {
# Handle translation
#
my $translate = 0;
my $localattrs = getAttributeString($attrs, 1, $lang, \$translate);
my $translation = $translations{$lang, $lookup} if isWellFormedXmlFragment($translations{$lang, $lookup});
if ($translate && !$translation) {
$translation = $lookup;
}
if ($translation || $translate) {
print $fh "\n";
$leading_space =~ s/.*\n//g;
print $fh $leading_space;
print $fh "<", $nodename, " xml:lang=\"", $lang, "\"", $localattrs, ">";
if ($translate == 2) {
translate_subnodes($fh, \@all, $lang, 1, $spacepreserve);
} else {
print $fh $translation;
}
print $fh "</$nodename>";
}
}
}
} else {
my $count = scalar(@all);
if ($count > 0) {
print $fh ">";
my $index = 0;
while ($index < $count) {
my $type = $all[$index];
my $rest = $all[$index+1];
traverse($fh, $type, $rest, $language, $spacepreserve);
$index += 2;
}
print $fh "</$nodename>";
} else {
print $fh "/>";
}
}
}
}
sub intltool_tree_comment
{
my $expat = shift;
my $data = shift;
my $clist = $expat->{Curlist};
my $pos = $#$clist;
push @$clist, 1 => $data;
}
sub intltool_tree_cdatastart
{
my $expat = shift;
my $clist = $expat->{Curlist};
my $pos = $#$clist;
push @$clist, 0 => $expat->original_string();
}
sub intltool_tree_cdataend
{
my $expat = shift;
my $clist = $expat->{Curlist};
my $pos = $#$clist;
$clist->[$pos] .= $expat->original_string();
}
sub intltool_tree_char
{
my $expat = shift;
my $text = shift;
my $clist = $expat->{Curlist};
my $pos = $#$clist;
# Use original_string so that we retain escaped entities
# in CDATA sections.
#
if ($pos > 0 and $clist->[$pos - 1] eq '0') {
$clist->[$pos] .= $expat->original_string();
} else {
push @$clist, 0 => $expat->original_string();
}
}
sub intltool_tree_start
{
my $expat = shift;
my $tag = shift;
my @origlist = ();
# Use original_string so that we retain escaped entities
# in attribute values. We must convert the string to an
# @origlist array to conform to the structure of the Tree
# Style.
#
my @original_array = split /\x/, $expat->original_string();
my $source = $expat->original_string();
# Remove leading tag.
#
$source =~ s|^\s*<\s*(\S+)||s;
# Grab attribute key/value pairs and push onto @origlist array.
#
while ($source)
{
if ($source =~ /^\s*([\w:-]+)\s*[=]\s*["]/)
{
$source =~ s|^\s*([\w:-]+)\s*[=]\s*["]([^"]*)["]||s;
push @origlist, $1;
push @origlist, '"' . $2 . '"';
}
elsif ($source =~ /^\s*([\w:-]+)\s*[=]\s*[']/)
{
$source =~ s|^\s*([\w:-]+)\s*[=]\s*[']([^']*)[']||s;
push @origlist, $1;
push @origlist, "'" . $2 . "'";
}
else
{
last;
}
}
my $ol = [ { @origlist } ];
push @{ $expat->{Lists} }, $expat->{Curlist};
push @{ $expat->{Curlist} }, $tag => $ol;
$expat->{Curlist} = $ol;
}
sub readXml
{
my $filename = shift || return;
if(!-f $filename) {
die "ERROR Cannot find filename: $filename\n";
}
my $ret = eval 'require XML::Parser';
if(!$ret) {
die "You must have XML::Parser installed to run $0\n\n";
}
my $xp = new XML::Parser(Style => 'Tree');
$xp->setHandlers(Char => \&intltool_tree_char);
$xp->setHandlers(Start => \&intltool_tree_start);
$xp->setHandlers(CdataStart => \&intltool_tree_cdatastart);
$xp->setHandlers(CdataEnd => \&intltool_tree_cdataend);
my $tree = $xp->parsefile($filename);
# <foo><head id="a">Hello <em>there</em></head><bar>Howdy<ref/></bar>do</foo>
# would be:
# [foo, [{}, head, [{id => "a"}, 0, "Hello ", em, [{}, 0, "there"]], bar, [{},
# 0, "Howdy", ref, [{}]], 0, "do" ] ]
return $tree;
}
sub print_header
{
my $infile = shift;
my $fh = shift;
my $source;
if(!-f $infile) {
die "ERROR Cannot find filename: $infile\n";
}
print $fh qq{<?xml version="1.0" encoding="UTF-8"?>\n};
{
local $/;
open DOCINPUT, "<${FILE}" or die;
$source = <DOCINPUT>;
close DOCINPUT;
}
if ($source =~ /(<!DOCTYPE.*\[.*\]\s*>)/s)
{
print $fh "$1\n";
}
elsif ($source =~ /(<!DOCTYPE[^>]*>)/s)
{
print $fh "$1\n";
}
}
sub parseTree
{
my $fh = shift;
my $ref = shift;
my $language = shift || "";
my $name = shift @{ $ref };
my $cont = shift @{ $ref };
while (!$name || "$name" eq "1") {
$name = shift @{ $ref };
$cont = shift @{ $ref };
}
my $spacepreserve = 0;
my $attrs = @{$cont}[0];
$spacepreserve = 1 if ((exists $attrs->{"xml:space"}) && ($attrs->{"xml:space"} =~ /^["']?preserve["']?$/));
traverse($fh, $name, $cont, $language, $spacepreserve);
}
sub xml_merge_output
{
my $source;
if ($MULTIPLE_OUTPUT) {
for my $lang (sort keys %po_files_by_lang) {
if ( ! -e $lang ) {
mkdir $lang or die "Cannot create subdirectory $lang: $!\n";
}
open OUTPUT, ">$lang/$OUTFILE" or die "Cannot open $lang/$OUTFILE: $!\n";
binmode (OUTPUT) if $^O eq 'MSWin32';
my $tree = readXml($FILE);
print_header($FILE, \*OUTPUT);
parseTree(\*OUTPUT, $tree, $lang);
close OUTPUT;
print "CREATED $lang/$OUTFILE\n" unless $QUIET_ARG;
}
}
open OUTPUT, ">$OUTFILE" or die "Cannot open $OUTFILE: $!\n";
binmode (OUTPUT) if $^O eq 'MSWin32';
my $tree = readXml($FILE);
print_header($FILE, \*OUTPUT);
parseTree(\*OUTPUT, $tree);
close OUTPUT;
print "CREATED $OUTFILE\n" unless $QUIET_ARG;
}
sub keys_merge_translations
{
open INPUT, "<${FILE}" or die;
open OUTPUT, ">${OUTFILE}" or die;
binmode (OUTPUT) if $^O eq 'MSWin32';
while (<INPUT>)
{
if (s/^(\s*)_(\w+=(.*))/$1$2/)
{
my $string = $3;
print OUTPUT;
my $non_translated_line = $_;
for my $lang (sort keys %po_files_by_lang)
{
my $translation = $translations{$lang, $string};
next if !$translation;
$_ = $non_translated_line;
s/(\w+)=.*/[$lang]$1=$translation/;
print OUTPUT;
}
}
else
{
print OUTPUT;
}
}
close OUTPUT;
close INPUT;
}
sub desktop_merge_translations
{
open INPUT, "<${FILE}" or die;
open OUTPUT, ">${OUTFILE}" or die;
binmode (OUTPUT) if $^O eq 'MSWin32';
while (<INPUT>)
{
if (s/^(\s*)_(\w+=(.*))/$1$2/)
{
my $string = $3;
print OUTPUT;
my $non_translated_line = $_;
for my $lang (sort keys %po_files_by_lang)
{
my $translation = $translations{$lang, $string};
next if !$translation;
$_ = $non_translated_line;
s/(\w+)=.*/${1}[$lang]=$translation/;
print OUTPUT;
}
}
else
{
print OUTPUT;
}
}
close OUTPUT;
close INPUT;
}
sub schemas_merge_translations
{
my $source;
{
local $/; # slurp mode
open INPUT, "<$FILE" or die "can't open $FILE: $!";
$source = <INPUT>;
close INPUT;
}
open OUTPUT, ">$OUTFILE" or die;
binmode (OUTPUT) if $^O eq 'MSWin32';
# FIXME: support attribute translations
# Empty nodes never need translation, so unmark all of them.
# For example, <_foo/> is just replaced by <foo/>.
$source =~ s|<\s*_($w+)\s*/>|<$1/>|g;
while ($source =~ s/
(.*?)
(\s+)(<locale\ name="C">(\s*)
(<default>\s*(?:<!--[^>]*?-->\s*)?(.*?)\s*<\/default>)?(\s*)
(<short>\s*(?:<!--[^>]*?-->\s*)?(.*?)\s*<\/short>)?(\s*)
(<long>\s*(?:<!--[^>]*?-->\s*)?(.*?)\s*<\/long>)?(\s*)
<\/locale>)
//sx)
{
print OUTPUT $1;
my $locale_start_spaces = $2 ? $2 : '';
my $default_spaces = $4 ? $4 : '';
my $short_spaces = $7 ? $7 : '';
my $long_spaces = $10 ? $10 : '';
my $locale_end_spaces = $13 ? $13 : '';
my $c_default_block = $3 ? $3 : '';
my $default_string = $6 ? $6 : '';
my $short_string = $9 ? $9 : '';
my $long_string = $12 ? $12 : '';
print OUTPUT "$locale_start_spaces$c_default_block";
$default_string =~ s/\s+/ /g;
$default_string = entity_decode($default_string);
$short_string =~ s/\s+/ /g;
$short_string = entity_decode($short_string);
$long_string =~ s/\s+/ /g;
$long_string = entity_decode($long_string);
for my $lang (sort keys %po_files_by_lang)
{
my $default_translation = $translations{$lang, $default_string};
my $short_translation = $translations{$lang, $short_string};
my $long_translation = $translations{$lang, $long_string};
next if (!$default_translation && !$short_translation &&
!$long_translation);
print OUTPUT "\n$locale_start_spaces<locale name=\"$lang\">";
print OUTPUT "$default_spaces";
if ($default_translation)
{
$default_translation = entity_encode($default_translation);
print OUTPUT "<default>$default_translation</default>";
}
print OUTPUT "$short_spaces";
if ($short_translation)
{
$short_translation = entity_encode($short_translation);
print OUTPUT "<short>$short_translation</short>";
}
print OUTPUT "$long_spaces";
if ($long_translation)
{
$long_translation = entity_encode($long_translation);
print OUTPUT "<long>$long_translation</long>";
}
print OUTPUT "$locale_end_spaces</locale>";
}
}
print OUTPUT $source;
close OUTPUT;
}
sub rfc822deb_merge_translations
{
my %encodings = ();
for my $lang (keys %po_files_by_lang) {
$encodings{$lang} = ($UTF8_ARG ? 'UTF-8' : get_po_encoding($po_files_by_lang{$lang}));
}
my $source;
{
local $/; # slurp mode
open INPUT, "<$FILE" or die "can't open $FILE: $!";
$source = <INPUT>;
close INPUT;
}
open OUTPUT, ">${OUTFILE}" or die;
binmode (OUTPUT) if $^O eq 'MSWin32';
my $last = 0;
while ($source =~ /\G(.*?)(^|\n)(_+)([^:\s]+)(:[ \t]*)(.*?)(?=\n[\S\n]|$)/sg)
{
$last = pos($source);
my ($pre, $newline, $underscore, $tag, $space, $text) = ($1, $2, $3, $4, $5, $6);
my $non_translated_line = $tag.$space;
$underscore = length($underscore);
# Print untranslated fields
my $untranslated_fields = $pre;
$untranslated_fields =~ s/\n#.*//g;
$untranslated_fields =~ s/^#.*(\n|$)//;
print OUTPUT $untranslated_fields;
# Remove [] dummy strings
my $stripped = $text;
$stripped =~ s/\[\s[^\[\]]*\],/,/g
if $underscore == 2;
$stripped =~ s/\[\s[^\[\]]*\]$//;
$non_translated_line .= $stripped;
print OUTPUT $newline.$non_translated_line;
if ($underscore)
{
my @str_list = rfc822deb_split($underscore, $text);
my $partial = 0;
# Process pseudo-comments
my @tfields = ();
if ($pre =~ m/^#flag:/m)
{
my @c = split (/\n#flag:/, $pre);
# The first field is null
shift (@c);
for (@c)
{
if (s/^comment(!?):(\S+)(?=\n|$)//s)
{
# This command is ignored by intltool-merge
}
elsif (s/^translate(!?):(\S+)(?=\n|$)//s)
{
rfc822deb_parse_spec($2, $1, 1+$#str_list, 0, 1, \@tfields);
}
elsif (s/^partial(?=\n|$)//s)
{
$partial = 1;
}
else
{
die "Unknown directive: $_\n\nAborting!\n";
}
}
}
# By default, print all msgids
rfc822deb_parse_spec('*', '', 1+$#str_list, 0, 1, \@tfields)
if $#tfields == -1;
for my $lang (sort keys %po_files_by_lang)
{
my $is_translated = 1;
my $str_translated = '';
my $cnt = 0;
for my $str (@str_list)
{
$cnt++;
my $translation;
if ($tfields[$cnt] && $str ne '')
{
$translation = $translations{$lang, $str};
if (!$translation)
{
if ($partial)
{
$translation = $str;
}
else
{
$is_translated = 0;
last;
}
}
}
else
{
$translation = $str;
}
# $translation may also contain [] dummy
# strings, mostly to indicate an empty string
$translation =~ s/\[\s[^\[\]]*\]$//;
# Escape commas
$translation =~ s/,/\\,/g
if $underscore == 2;
if ($cnt == 1)
{
print STDERR "WARNING: $lang: spurious newline removed\n"
if $translation =~ s/\n/ /g;
$str_translated .= $translation;
}
else
{
if ($underscore == 2)
{
$str_translated .= ', ' . $translation;
}
else
{
$translation =~ s/\n/\n /g;
$str_translated .= "\n ." unless $cnt == 2;
$str_translated .= "\n " . $translation unless $str eq '';
}
}
}
next unless $is_translated;
$str_translated =~ s/\s+$//;
$str_translated = ' '.$str_translated if length ($str_translated) && $str_translated !~ /^\n/s;
$_ = $non_translated_line;
s/^(\w+):\s*.*/$newline${1}-$lang.$encodings{$lang}:$str_translated/s;
print OUTPUT;
}
}
}
my $tail = substr($source, $last);
$tail .= "\n" unless $tail =~ m/\n$/s;
$tail =~ s/^#.*\n//mg;
print OUTPUT $tail;
close OUTPUT;
close INPUT;
}
sub rfc822deb_parse_spec {
my $spec = shift;
my $negate = shift;
my $len = shift;
my $notfound = shift;
my $found = shift;
my $ref = shift;
$spec = ','.$spec.',';
# Replace '*' by all values
my $all = '1-'.$len;
$spec =~ s/\*/$all/g;
# Expand ranges
$spec =~ s/(\d+)-(\d+)/join(",", ($1..$2))/eg;
if ($#{$ref} == -1)
{
for my $cnt (1..$len)
{
$ref->[$cnt] = $notfound;
}
}
for my $cnt (1..$len)
{
if ($spec =~ m/,$cnt,/ && !$negate) {
$ref->[$cnt] .= $found;
} elsif ($spec !~ m/,$cnt,/ && $negate) {
$ref->[$cnt] .= $found;
}
}
}
sub rfc822deb_split
{
# Debian defines a special way to deal with rfc822-style files:
# when a value contain newlines, it consists of
# 1. a short form (first line)
# 2. a long description, all lines begin with a space,
# and paragraphs are separated by a single dot on a line
# This routine returns an array of all paragraphs, and reformat
# them.
# When first argument is 2, the string is a comma separated list of
# values.
my $type = shift;
my $text = shift;
$text =~ s/^[ \t]//mg;
if ($type ne 1)
{
my @values = ();
for my $value (split(/(?<!\\), */, $text, 0))
{
$value =~ s/\\,/,/g;
push @values, $value;
}
return @values;
}
return ($text) if $text !~ /\n/;
$text =~ s/([^\n]*)\n//;
my @list = ($1);
my $str = '';
for my $line (split (/\n/, $text))
{
chomp $line;
if ($line =~ /^\.\s*$/)
{
# New paragraph
$str =~ s/\s*$//;
push(@list, $str);
$str = '';
}
elsif ($line =~ /^\s/)
{
# Line which must not be reformatted
$str .= "\n" if length ($str) && $str !~ /\n$/;
$line =~ s/\s+$//;
$str .= $line."\n";
}
else
{
# Continuation line, remove newline
$str .= " " if length ($str) && $str !~ /\n$/;
$str .= $line;
}
}
$str =~ s/\s*$//;
push(@list, $str) if length ($str);
return @list;
}
sub quoted_translation
{
my ($lang, $string) = @_;
$string =~ s/\\\"/\"/g;
my $translation = $translations{$lang, $string};
$translation = $string if !$translation;
$translation =~ s/\"/\\\"/g;
return $translation
}
sub quoted_merge_translations
{
if (!$MULTIPLE_OUTPUT) {
print "Quoted only supports Multiple Output.\n";
exit(1);
}
for my $lang (sort keys %po_files_by_lang) {
if ( ! -e $lang ) {
mkdir $lang or die "Cannot create subdirectory $lang: $!\n";
}
open INPUT, "<${FILE}" or die;
open OUTPUT, ">$lang/$OUTFILE" or die "Cannot open $lang/$OUTFILE: $!\n";
binmode (OUTPUT) if $^O eq 'MSWin32';
while (<INPUT>)
{
s/\"(([^\"]|\\\")*[^\\\"])\"/"\"" . "ed_translation($lang, $1) . "\""/ge;
print OUTPUT;
}
close OUTPUT;
close INPUT;
}
}
AnonSec - 2021 | Recode By D7net