Koha/misc/translator/tmpl_process3.pl
Chris Cormack 873091dcc8 Bug 6958 : Fix for leading spaces being stripped on translation
This will re-add any leading spaces, so formatting is not messed

Signed-off-by: Katrin Fischer <katrin.fischer.83@web.de>
- all examples from the bug report are fixed now
- verified system preferences are still translated
- verified xslt displays are still translated
- verified javascript alerts are still translated
- verified switching languages works

Signed-off-by: Ian Walls <ian.walls@bywatersolutions.com>
2011-10-07 09:42:13 +13:00

557 lines
18 KiB
Perl
Executable file

#!/usr/bin/perl
# This file is part of Koha
# Parts copyright 2003-2004 Paul Poulain
# Parts copyright 2003-2004 Jerome Vizcaino
# Parts copyright 2004 Ambrose Li
=head1 NAME
tmpl_process3.pl - Alternative version of tmpl_process.pl
using gettext-compatible translation files
=cut
use strict;
#use warnings; FIXME - Bug 2505
use Getopt::Long;
use Locale::PO;
use File::Temp qw( :POSIX );
use TmplTokenizer;
use VerboseWarnings qw( :warn :die );
###############################################################################
use vars qw( @in_files $in_dir $str_file $out_dir $quiet );
use vars qw( @excludes $exclude_regex );
use vars qw( $recursive_p );
use vars qw( $pedantic_p );
use vars qw( $href );
use vars qw( $type ); # file extension (DOS form without the dot) to match
use vars qw( $charset_in $charset_out );
###############################################################################
sub find_translation ($) {
my($s) = @_;
my $key = $s;
if ($s =~ /\S/s) {
$key = TmplTokenizer::string_canon($key);
$key = TmplTokenizer::charset_convert($key, $charset_in, $charset_out);
$key = TmplTokenizer::quote_po($key);
}
if (defined $href->{$key} && !$href->{$key}->fuzzy && length Locale::PO->dequote($href->{$key}->msgstr)){
if ($s =~ /^(\s+)/){
return $1 . Locale::PO->dequote($href->{$key}->msgstr);
}
else {
return Locale::PO->dequote($href->{$key}->msgstr);
}
}
else {
return $s;
}
}
sub text_replace_tag ($$) {
my($t, $attr) = @_;
my $it;
# value [tag=input], meta
my $tag = lc($1) if $t =~ /^<(\S+)/s;
my $translated_p = 0;
for my $a ('alt', 'content', 'title', 'value','label') {
if ($attr->{$a}) {
next if $a eq 'label' && $tag ne 'optgroup';
next if $a eq 'content' && $tag ne 'meta';
next if $a eq 'value' && ($tag ne 'input' || (ref $attr->{'type'} && $attr->{'type'}->[1] =~ /^(?:checkbox|hidden|radio|text)$/)); # FIXME
my($key, $val, $val_orig, $order) = @{$attr->{$a}}; #FIXME
if ($val =~ /\S/s) {
my $s = find_translation($val);
if ($attr->{$a}->[1] ne $s) { #FIXME
$attr->{$a}->[1] = $s; # FIXME
$attr->{$a}->[2] = ($s =~ /"/s)? "'$s'": "\"$s\""; #FIXME
$translated_p = 1;
}
}
}
}
if ($translated_p) {
$it = "<$tag"
. join('', map { if ($_ ne '/'){
sprintf(' %s="%s"', $_, $attr->{$_}->[1]);
}
else {
sprintf(' %s',$_);
}
} sort {
$attr->{$a}->[3] <=> $attr->{$b}->[3] #FIXME
} keys %$attr);
$it .= '>';
}
else {
$it = $t;
}
return $it;
}
sub text_replace (**) {
my($h, $output) = @_;
for (;;) {
my $s = TmplTokenizer::next_token $h;
last unless defined $s;
my($kind, $t, $attr) = ($s->type, $s->string, $s->attributes);
if ($kind eq C4::TmplTokenType::TEXT) {
print $output find_translation($t);
} elsif ($kind eq C4::TmplTokenType::TEXT_PARAMETRIZED) {
my $fmt = find_translation($s->form);
print $output TmplTokenizer::parametrize($fmt, 1, $s, sub {
$_ = $_[0];
my($kind, $t, $attr) = ($_->type, $_->string, $_->attributes);
$kind == C4::TmplTokenType::TAG && %$attr?
text_replace_tag($t, $attr): $t });
} elsif ($kind eq C4::TmplTokenType::TAG && %$attr) {
print $output text_replace_tag($t, $attr);
} elsif ($s->has_js_data) {
for my $t (@{$s->js_data}) {
# FIXME for this whole block
if ($t->[0]) {
printf $output "%s%s%s", $t->[2], find_translation $t->[3],
$t->[2];
} else {
print $output $t->[1];
}
}
} elsif (defined $t) {
# Quick fix to bug 4472
$t = "<!DOCTYPE stylesheet [" if $t =~ /DOCTYPE stylesheet/ ;
print $output $t;
}
}
}
sub listfiles ($$$) {
my($dir, $type, $action) = @_;
my @it = ();
if (opendir(DIR, $dir)) {
my @dirent = readdir DIR; # because DIR is shared when recursing
closedir DIR;
for my $dirent (@dirent) {
my $path = "$dir/$dirent";
if ($dirent =~ /^\./ || $dirent eq 'CVS' || $dirent eq 'RCS'
|| (defined $exclude_regex && $dirent =~ /^(?:$exclude_regex)$/)) {
;
} elsif (-f $path) {
push @it, $path if (!defined $type || $dirent =~ /\.(?:$type)$/) || $action eq 'install';
} elsif (-d $path && $recursive_p) {
push @it, listfiles($path, $type, $action);
}
}
} else {
warn_normal "$dir: $!", undef;
}
return @it;
}
###############################################################################
sub mkdir_recursive ($) {
my($dir) = @_;
local($`, $&, $', $1);
$dir = $` if $dir ne /^\/+$/ && $dir =~ /\/+$/;
my ($prefix, $basename) = ($dir =~ /\/([^\/]+)$/s)? ($`, $1): ('.', $dir);
mkdir_recursive($prefix) if $prefix ne '.' && !-d $prefix;
if (!-d $dir) {
print STDERR "Making directory $dir..." unless $quiet;
# creates with rwxrwxr-x permissions
mkdir($dir, 0775) || warn_normal "$dir: $!", undef;
}
}
###############################################################################
sub usage ($) {
my($exitcode) = @_;
my $h = $exitcode? *STDERR: *STDOUT;
print $h <<EOF;
Usage: $0 create [OPTION]
or: $0 update [OPTION]
or: $0 install [OPTION]
or: $0 --help
Create or update PO files from templates, or install translated templates.
-i, --input=SOURCE Get or update strings from SOURCE file.
SOURCE is a directory if -r is also specified.
-o, --outputdir=DIRECTORY Install translation(s) to specified DIRECTORY
--pedantic-warnings Issue warnings even for detected problems
which are likely to be harmless
-r, --recursive SOURCE in the -i option is a directory
-s, --str-file=FILE Specify FILE as the translation (po) file
for input (install) or output (create, update)
-x, --exclude=REGEXP Exclude files matching the given REGEXP
--help Display this help and exit
-q, --quiet no output to screen (except for errors)
The -o option is ignored for the "create" and "update" actions.
Try `perldoc $0 for perhaps more information.
EOF
exit($exitcode);
}#`
###############################################################################
sub usage_error (;$) {
for my $msg (split(/\n/, $_[0])) {
print STDERR "$msg\n";
}
print STDERR "Try `$0 --help for more information.\n";
exit(-1);
}
###############################################################################
GetOptions(
'input|i=s' => \@in_files,
'outputdir|o=s' => \$out_dir,
'recursive|r' => \$recursive_p,
'str-file|s=s' => \$str_file,
'exclude|x=s' => \@excludes,
'quiet|q' => \$quiet,
'pedantic-warnings|pedantic' => sub { $pedantic_p = 1 },
'help' => \&usage,
) || usage_error;
VerboseWarnings::set_application_name $0;
VerboseWarnings::set_pedantic_mode $pedantic_p;
# keep the buggy Locale::PO quiet if it says stupid things
$SIG{__WARN__} = sub {
my($s) = @_;
print STDERR $s unless $s =~ /^Strange line in [^:]+: #~/s
};
my $action = shift or usage_error('You must specify an ACTION.');
usage_error('You must at least specify input and string list filenames.')
if !@in_files || !defined $str_file;
# Type match defaults to *.tt plus *.inc if not specified
$type = "tt|inc|xsl" if !defined($type);
# Check the inputs for being files or directories
for my $input (@in_files) {
usage_error("$input: Input must be a file or directory.\n"
. "(Symbolic links are not supported at the moment)")
unless -d $input || -f $input;;
}
# Generates the global exclude regular expression
$exclude_regex = '(?:'.join('|', @excludes).')' if @excludes;
# Generate the list of input files if a directory is specified
if (-d $in_files[0]) {
die "If you specify a directory as input, you must specify only it.\n"
if @in_files > 1;
# input is a directory, generates list of files to process
$in_dir = $in_files[0];
$in_dir =~ s/\/$//; # strips the trailing / if any
@in_files = listfiles($in_dir, $type, $action);
} else {
for my $input (@in_files) {
die "You cannot specify input files and directories at the same time.\n"
unless -f $input;
}
}
# restores the string list from file
$href = Locale::PO->load_file_ashash($str_file);
# guess the charsets. HTML::Templates defaults to iso-8859-1
if (defined $href) {
die "$str_file: PO file is corrupted, or not a PO file\n" unless defined $href->{'""'};
$charset_out = TmplTokenizer::charset_canon $2 if $href->{'""'}->msgstr =~ /\bcharset=(["']?)([^;\s"'\\]+)\1/;
$charset_in = $charset_out;
warn "Charset in/out: ".$charset_out;
# for my $msgid (keys %$href) {
# if ($msgid =~ /\bcharset=(["']?)([^;\s"'\\]+)\1/) {
# my $candidate = TmplTokenizer::charset_canon $2;
# die "Conflicting charsets in msgid: $charset_in vs $candidate => $msgid\n"
# if defined $charset_in && $charset_in ne $candidate;
# $charset_in = $candidate;
# }
# }
# BUG6464: check consistency of PO messages
# - count number of '%s' in msgid and msgstr
for my $msg ( values %$href ) {
my $id_count = split(/%s/, $msg->{msgid}) - 1;
my $str_count = split(/%s/, $msg->{msgstr}) - 1;
next if $id_count == $str_count ||
$msg->{msgstr} eq '""' ||
grep { /fuzzy/ } @{$msg->{_flags}};
warn_normal
"unconsistent %s count: ($id_count/$str_count):\n" .
" line: " . $msg->{loaded_line_number} . "\n" .
" msgid: " . $msg->{msgid} . "\n" .
" msgstr: " . $msg->{msgstr} . "\n", undef;
}
}
# set our charset in to UTF-8
if (!defined $charset_in) {
$charset_in = TmplTokenizer::charset_canon 'UTF-8';
warn "Warning: Can't determine original templates' charset, defaulting to $charset_in\n";
}
# set our charset out to UTF-8
if (!defined $charset_out) {
$charset_out = TmplTokenizer::charset_canon 'UTF-8';
warn "Warning: Charset Out defaulting to $charset_out\n";
}
my $xgettext = './xgettext.pl'; # actual text extractor script
my $st;
if ($action eq 'create') {
# updates the list. As the list is empty, every entry will be added
if (!-s $str_file) {
warn "Removing empty file $str_file\n";
unlink $str_file || die "$str_file: $!\n";
}
die "$str_file: Output file already exists\n" if -f $str_file;
my($tmph1, $tmpfile1) = tmpnam();
my($tmph2, $tmpfile2) = tmpnam();
close $tmph2; # We just want a name
# Generate the temporary file that acts as <MODULE>/POTFILES.in
for my $input (@in_files) {
print $tmph1 "$input\n";
}
close $tmph1;
warn "I $charset_in O $charset_out";
# Generate the specified po file ($str_file)
$st = system ($xgettext, '-s', '-f', $tmpfile1, '-o', $tmpfile2,
(defined $charset_in? ('-I', $charset_in): ()),
(defined $charset_out? ('-O', $charset_out): ())
);
# Run msgmerge so that the pot file looks like a real pot file
# We need to help msgmerge a bit by pre-creating a dummy po file that has
# the headers and the "" msgid & msgstr. It will fill in the rest.
if ($st == 0) {
# Merge the temporary "pot file" with the specified po file ($str_file)
# FIXME: msgmerge(1) is a Unix dependency
# FIXME: need to check the return value
unless (-f $str_file) {
local(*INPUT, *OUTPUT);
open(INPUT, "<$tmpfile2");
open(OUTPUT, ">$str_file");
while (<INPUT>) {
print OUTPUT;
last if /^\n/s;
}
close INPUT;
close OUTPUT;
}
$st = system('msgmerge', '-U', '-s', $str_file, $tmpfile2);
} else {
error_normal "Text extraction failed: $xgettext: $!\n", undef;
error_additional "Will not run msgmerge\n", undef;
}
# unlink $tmpfile1 || warn_normal "$tmpfile1: unlink failed: $!\n", undef;
# unlink $tmpfile2 || warn_normal "$tmpfile2: unlink failed: $!\n", undef;
} elsif ($action eq 'update') {
my($tmph1, $tmpfile1) = tmpnam();
my($tmph2, $tmpfile2) = tmpnam();
close $tmph2; # We just want a name
# Generate the temporary file that acts as <MODULE>/POTFILES.in
for my $input (@in_files) {
print $tmph1 "$input\n";
}
close $tmph1;
# Generate the temporary file that acts as <MODULE>/<LANG>.pot
$st = system($xgettext, '-s', '-f', $tmpfile1, '-o', $tmpfile2,
'--po-mode',
(defined $charset_in? ('-I', $charset_in): ()),
(defined $charset_out? ('-O', $charset_out): ()));
if ($st == 0) {
# Merge the temporary "pot file" with the specified po file ($str_file)
# FIXME: msgmerge(1) is a Unix dependency
# FIXME: need to check the return value
$st = system('msgmerge', '-U', '-s', $str_file, $tmpfile2);
} else {
error_normal "Text extraction failed: $xgettext: $!\n", undef;
error_additional "Will not run msgmerge\n", undef;
}
# unlink $tmpfile1 || warn_normal "$tmpfile1: unlink failed: $!\n", undef;
# unlink $tmpfile2 || warn_normal "$tmpfile2: unlink failed: $!\n", undef;
} elsif ($action eq 'install') {
if(!defined($out_dir)) {
usage_error("You must specify an output directory when using the install method.");
}
if ($in_dir eq $out_dir) {
warn "You must specify a different input and output directory.\n";
exit -1;
}
# Make sure the output directory exists
# (It will auto-create it, but for compatibility we should not)
-d $out_dir || die "$out_dir: The directory does not exist\n";
# Try to open the file, because Locale::PO doesn't check :-/
open(INPUT, "<$str_file") || die "$str_file: $!\n";
close INPUT;
# creates the new tmpl file using the new translation
for my $input (@in_files) {
die "Assertion failed"
unless substr($input, 0, length($in_dir) + 1) eq "$in_dir/";
# print "$input / $type\n";
if (!defined $type || $input =~ /\.(?:$type)$/) {
my $h = TmplTokenizer->new( $input );
$h->set_allow_cformat( 1 );
VerboseWarnings::set_input_file_name $input;
my $target = $out_dir . substr($input, length($in_dir));
my $targetdir = $` if $target =~ /[^\/]+$/s;
mkdir_recursive($targetdir) unless -d $targetdir;
print STDERR "Creating $target...\n" unless $quiet;
open( OUTPUT, ">$target" ) || die "$target: $!\n";
text_replace( $h, *OUTPUT );
close OUTPUT;
} else {
# just copying the file
my $target = $out_dir . substr($input, length($in_dir));
my $targetdir = $` if $target =~ /[^\/]+$/s;
mkdir_recursive($targetdir) unless -d $targetdir;
system("cp -f $input $target");
print STDERR "Copying $input...\n" unless $quiet;
}
}
} else {
usage_error('Unknown action specified.');
}
if ($st == 0) {
printf "The %s seems to be successful.\n", $action unless $quiet;
} else {
printf "%s FAILED.\n", "\u$action" unless $quiet;
}
exit 0;
###############################################################################
=head1 SYNOPSIS
./tmpl_process3.pl [ I<tmpl_process.pl options> ]
=head1 DESCRIPTION
This is an alternative version of the tmpl_process.pl script,
using standard gettext-style PO files. While there still might
be changes made to the way it extracts strings, at this moment
it should be stable enough for general use; it is already being
used for the Chinese and Polish translations.
Currently, the create, update, and install actions have all been
reimplemented and seem to work.
=head2 Features
=over
=item -
Translation files in standard Uniforum PO format.
All standard tools including all gettext tools,
plus PO file editors like kbabel(1) etc.
can be used.
=item -
Minor changes in whitespace in source templates
do not generally require strings to be re-translated.
=item -
Able to handle <TMPL_VAR> variables in the templates;
<TMPL_VAR> variables are usually extracted in proper context,
represented by a short %s placeholder.
=item -
Able to handle text input and radio button INPUT elements
in the templates; these INPUT elements are also usually
extracted in proper context,
represented by a short %S or %p placeholder.
=item -
Automatic comments in the generated PO files to provide
even more context (line numbers, and the names and types
of the variables).
=item -
The %I<n>$s (or %I<n>$p, etc.) notation can be used
for change the ordering of the variables,
if such a reordering is required for correct translation.
=item -
If a particular <TMPL_VAR> should not appear in the
translation, it can be suppressed with the %0.0s notation.
=item -
Using the PO format also means translators can add their
own comments in the translation files, if necessary.
=item -
Create, update, and install actions are all based on the
same scanner module. This ensures that update and install
have the same idea of what is a translatable string;
attribute names in tags, for example, will not be
accidentally translated.
=back
=head1 NOTES
Anchors are represented by an <AI<n>> notation.
The meaning of this non-standard notation might not be obvious.
The create action calls xgettext.pl to do the actual work;
the update action calls xgettext.pl and msgmerge(1) to do the
actual work.
=head1 BUGS
xgettext.pl must be present in the current directory; the
msgmerge(1) command must also be present in the search path.
The script currently does not check carefully whether these
dependent commands are present.
Locale::PO(3) has a lot of bugs. It can neither parse nor
generate GNU PO files properly; a couple of workarounds have
been written in TmplTokenizer and more is likely to be needed
(e.g., to get rid of the "Strange line" warning for #~).
This script may not work in Windows.
There are probably some other bugs too, since this has not been
tested very much.
=head1 SEE ALSO
xgettext.pl,
TmplTokenizer.pm,
msgmerge(1),
Locale::PO(3),
translator_doc.txt
http://www.saas.nsw.edu.au/koha_wiki/index.php?page=DifficultTerms
=cut