script/tei2korapxml

#!/usr/bin/env perl
use strict;
use warnings;

use Log::Any '$log';
use Log::Any::Adapter;
use Pod::Usage;
use Getopt::Long qw(GetOptions :config no_auto_abbrev);

use File::Basename qw(dirname);

use Encode qw(decode);

use FindBin;
BEGIN {
  unshift @INC, "$FindBin::Bin/../lib";
};

use KorAP::XML::TEI qw!remove_xml_comments replace_entities!;
use KorAP::XML::TEI::Tokenizer::External;
use KorAP::XML::TEI::Tokenizer::Conservative;
use KorAP::XML::TEI::Tokenizer::Aggressive;
use KorAP::XML::TEI::Zipper;
use KorAP::XML::TEI::Header;
use KorAP::XML::TEI::Inline;

our $VERSION = '2.5.0';

our $VERSION_MSG = "\ntei2korapxml - v$VERSION\n";

use constant {
  # Set to 1 for minimal more debug output (no need to be parametrized)
  DEBUG => $ENV{KORAPXMLTEI_DEBUG} // 0
};

if ($ENV{KORAPXMLTEI_INLINE}) {
  warn 'KORAPXMLTEI_INLINE is deprecated in favor of --skip-inline-token-annotations';
};

# Inline tokens won't be stored in the structure file
my $inline_tokens_exclusive = 0;

# Parse options from the command line
GetOptions(
  'root|r=s'              => \(my $root_dir    = '.'),
  'input|i=s'             => \(my $input_fname = ''),
  'tokenizer-call|tc=s'   => \(my $tokenizer_call),
  'tokenizer-korap|tk'    => \(my $tokenizer_korap),
  'tokenizer-internal|ti' => \(my $tokenizer_intern),
  'no-tokenizer'          => \(my $no_tokenizer),
  'use-tokenizer-sentence-splits|s' => \(my $use_tokenizer_sentence_splits),
  'inline-tokens=s'       => \(my $inline_tokens = 'tokens#morpho'),
  'inline-structures=s'   => \(my $inline_structures = 'struct#structure'),
  'skip-inline-tokens'    => \(my $skip_inline_tokens = 0),
  'skip-inline-token-annotations' => \(
    my $skip_inline_token_annotations = ($ENV{KORAPXMLTEI_INLINE} ? 0 : 1)),
  'skip-inline-tags=s'    => \(my $skip_inline_tags_str = ''),
  'base-foundry=s'        => \(my $base_dir    = 'base'),
  'data-file=s'           => \(my $data_file   = 'data'),
  'header-file=s'         => \(my $header_file = 'header'),
  'tokens-file=s'         => \(my $tokens_file = 'tokens'),
  'xmlid-to-textsigle|x=s'=> \(my $xmlid_to_textsigle = ''),
  'log|l=s'               => \(my $log_level   = 'notice'),
  'required-version|rv=s' => \(my $required_version),
  ''                      => \(my $stdio),
  'help|h' => sub {
    pod2usage(
      -verbose => 99,
      -sections => 'NAME|DESCRIPTION|SYNOPSIS|ARGUMENTS|OPTIONS',
      -msg => $VERSION_MSG,
      -output => '-'
    )
  },
  'version|v' => sub {
    pod2usage(
      -verbose => 0,
      -msg => $VERSION_MSG,
      -output => '-'
    );
  }
);


# Establish logger
binmode(STDERR, ':encoding(UTF-8)');
Log::Any::Adapter->set('Stderr', log_level => $log_level);
$log->notice('Debugging is activated') if DEBUG;


if ($required_version) {
  $required_version =~ /^\s*(\d+\.\d+\.\d+(-TRIAL)?)\s*$/;
  if (!$1 || $1 ne $VERSION) {
    $log->error("Required version $required_version mismatches version $VERSION");
    exit(1);
  };
};


my ($what, $with);
if ($xmlid_to_textsigle ne '') {
  ($what, $with) = split('@', $xmlid_to_textsigle);
  $what = qr!$what!;
};

# tag (without attributes), which contains the primary text
my $_TEXT_BODY = 'text';
# optional

# Remember to skip certain inline tags
my %skip_inline_tags = ();
if ($skip_inline_tags_str) {
  foreach (split /\s*,\s*/, $skip_inline_tags_str) {
    $skip_inline_tags{$_} = 1;
  };
};

# External tokenization
my $ext_tok;
if ($tokenizer_call) {
  $ext_tok = KorAP::XML::TEI::Tokenizer::External->new($tokenizer_call);
  $ext_tok->sentence_splits(1) if $use_tokenizer_sentence_splits;
}

# KorAP tokenization
elsif ($tokenizer_korap) {
  eval {
    require KorAP::XML::TEI::Tokenizer::KorAP;
    1;
  };

  my $korap_tok_ver = $KorAP::XML::TEI::Tokenizer::KorAP::VERSION;
  if ($korap_tok_ver ne $VERSION) {
    $log->error("KorAP-Tokenizer version ($korap_tok_ver) differs from the expected version ($VERSION)");
    exit(1);
  };

  $ext_tok = KorAP::XML::TEI::Tokenizer::KorAP->new($use_tokenizer_sentence_splits);
}

# No internal tokenizer chosen
elsif (!$tokenizer_intern && !$no_tokenizer) {
  $log->error("No tokenizer chosen. If only internal tokens should be used, pass the --no-tokenizer flag");
  exit(1);
};

if ($use_tokenizer_sentence_splits) {
  $skip_inline_tags{s} = 1;
};

# Internal tokenization
my $aggr_tok = KorAP::XML::TEI::Tokenizer::Aggressive->new;
my $cons_tok = KorAP::XML::TEI::Tokenizer::Conservative->new;


# Name of the directory and the file containing all inline structure informations
# except for $_TOKENS_TAG information
my ($_structure_dir, $_structure_file) = split '#', $inline_structures . '#structure';

# Name of the directory and the file containing all inline token informations
# i.e. tokens of the $_TOKENS_TAG, if $_TOKENS_PROC is set
my ($_tokens_dir, $_tokens_file) = split '#', $inline_tokens . '#morpho';

if (index($_tokens_dir, '!') == 0) {
  $_tokens_dir = substr($_tokens_dir, 1);
  $inline_tokens_exclusive = 1;
};

# Initialize zipper
my $zipper = KorAP::XML::TEI::Zipper->new($root_dir);

# text directory (below $root_dir)
my $dir = '';

# Escaped version of text id
my $text_id_esc;

# Default encoding of the text
my $input_enc = 'UTF-8';

# text line (needed for whitespace handling)
my $text_line = 0;


# Input file handle (default: stdin)
my $input_fh;

# Single dash was set
if ($stdio) {
  $input_fh = *STDIN;
}

# Input flag was passed
elsif ($input_fname ne '') {
  unless (open($input_fh, '<', $input_fname)) {
    die $log->fatal("File '$input_fname' could not be opened.");
  };
}

# No input to process
else {
  pod2usage(
    -verbose => 99,
    -sections => 'NAME|SYNOPSIS',
    -msg => $VERSION_MSG,
    -output => '-'
  );
  exit;
};

# Prevents segfaulting (see notes on segfault prevention)
binmode $input_fh;


# Create inline parser object
my $inline = KorAP::XML::TEI::Inline->new(
  $skip_inline_tokens,
  \%skip_inline_tags,
  $inline_tokens_exclusive
);


# Reading input document
MAIN: while (<$input_fh>) {

  # remove HTML (multi-line) comments (<!--...-->)
  $_ = remove_xml_comments($input_fh, $_);

  # Set input encoding
  if (index($_, '<?xml') == 0 && $_ =~ /\sencoding=(['"])([^\1]+?)\1/) {
    $input_enc = $2;
    next;
  };

  $_ = decode($input_enc, $_);
  $_ = replace_entities($_);

  # Start of text body
  if (index($_, $_TEXT_BODY) >= 0 && m#^(.*)<${_TEXT_BODY}(?: [^>]*)?>(.*)$#) {
    my $suffix = $2;

    if ($1 !~ /^\s*$/ || $suffix !~ /^\s*$/) {
      die $log->fatal("input line number $.: " .
                        "line with opening text-body tag '${_TEXT_BODY}' " .
                        "contains additional information ... => Aborting (line=$_)");
    };

    # Text body data extracted from input document ($input_fh),
    # further processed by XML::LibXML::Reader
    my $text_buffer = '';

    # Iterate over all lines in the text body
    while (<$input_fh>) {

      $_ = remove_xml_comments($input_fh, $_);
      $_ = decode($input_enc, $_);
      $_ = replace_entities($_);

      # End of text body
      if ((my $pos = index($_, "</$_TEXT_BODY>")) >= 0) {

        # write data.xml, structure.xml and evtl. morpho.xml and/or tokenization files

        if ((substr($_, 0, $pos) . substr($_, length("</$_TEXT_BODY>") + $pos)) !~ /^\s*$/) {
          die $log->fatal("input line number $.: " .
                            "line with closing text-body tag '${_TEXT_BODY}'".
                            " contains additional information ... => Aborting (line=$_)");
        };

        if ($dir eq '') {
          $log->warn(
            "Maybe empty textSigle => skipping this text ...\n" .
              'data=' . substr($inline->data->data, 0, 200)
            );
          next MAIN;
        };

        # Parse inline structure
        $inline->parse($text_id_esc, \$text_buffer);

        if (DEBUG) {
          $log->debug("Writing (utf8-formatted) xml file $dir/${data_file}.xml");
        };

        my $data = $inline->data;

        # Write data.xml
        $data->to_zip(
          $zipper->new_stream("$dir/${data_file}.xml"),
          $text_id_esc
        );

        # Tokenize with external tokenizer
        if ($ext_tok) {

          # Tokenize and output
          $ext_tok->tokenize($data->data)->to_zip(
            $zipper->new_stream("$dir/$base_dir/${tokens_file}.xml"),
            $text_id_esc
          );

          if ($use_tokenizer_sentence_splits) {
            $ext_tok->sentencize_from_previous_input($inline->structures);
          };
        };

        # Tokenize with internal tokenizer
        if ($tokenizer_intern) {

          # Tokenize and output
          $cons_tok->tokenize($data->data)->to_zip(
            $zipper->new_stream("$dir/$base_dir/" . $cons_tok->name . '.xml'),
            $text_id_esc
          )->reset;

          $aggr_tok->tokenize($data->data)->to_zip(
            $zipper->new_stream("$dir/$base_dir/" . $aggr_tok->name . '.xml'),
            $text_id_esc
          )->reset;
        };

        # ~ write structures ~
        if (!$inline->structures->empty) {
          $inline->structures->to_zip(
            $zipper->new_stream("$dir/$_structure_dir/${_structure_file}.xml"),
            $text_id_esc,
            2 # = structure serialization
          );
        };

        # ~ write tokens ~
        unless ($skip_inline_tokens || $inline->tokens->empty) {
          $inline->tokens->to_zip(
            $zipper->new_stream("$dir/$_tokens_dir/${_tokens_file}.xml"),
            $text_id_esc,
            # Either 0 = tokens without inline or 1 = tokens with inline
            !$skip_inline_token_annotations
          );
        };

        # reinit.
        $dir = '';

        next MAIN;
      };


      # ~ whitespace handling ~

      # Fix whitespaces (see notes on whitespace fixing)

      # TODO:
      #   Maybe it's best, to keep the stripping of whitespace and
      #   to just remove the if-clause and to insert a blank by default
      #   (with possibly an option on how newlines in primary text should
      #   be handled (stripped or replaced by a whitespace)).

      # Remove consecutive whitespace at beginning and end (mostly one newline)
      s/^\s+//; s/\s+$//;

      # NOTE:
      #   this is only relevant, if a text consists of more than one line

      # TODO:
      #   find a better solution, or create a warning, if a text has more
      #   than one line ($text_line > 1)

      # TODO:
      #   do testing with 2 different corpora
      #   (one with only one-line texts, the other with several lines per text)

      # line contains at least one non-tag character
      if (m/^[^<]*$/ || m/(?:<[^>]+>[^<])|(?:[^<]<[^>]+>)/) {

        # Increment counter for text lines
        $text_line++;

        # insert blank before 1st character
        # (for 2nd line and consecutive lines)
        $_ = ' ' . $_ if $text_line > 1;
      }

      # add line to buffer
      $text_buffer .= $_;
    };
  }

  elsif (m#^(.*)\<TEI\s+[^>]*?xml:id=(["'])(.+?)\2#) {
    my $leadin = $1;
    my $id = $3;
    my $sigle = $3;

    if ($what) {
      $_ = $id;
      eval "s|$what|$with|";  # s@ICC.German\.([^.]+\.[^.]+)\.(.+)@ICCGER/$1/$2@;
      $sigle = $_;
      $log->debug("Converted text id `$id' to sigle `$sigle'");
    };
    $sigle =~ s/\./-/g;

    my @parts = split(/[\/_]/, $sigle);
    if (@parts != 3) {
      die $log->fatal(
          "input line number $.: " .
              "ids must have exactly three parts split by '/', but `$id` only has " . scalar(@parts) . " ".
              "=> Aborting (line=$_)");
    };

    $dir = join("/", @parts);
    $text_id_esc = "$parts[0]/$parts[1].$parts[2]";
    $log->notice("$0: text_id=$text_id_esc");

    if ($leadin !~ /^\s*$/) {
      die $log->fatal(
          "input line number $.: " .
              'line with opening header tag is not in expected format ... ' .
              "=> Aborting (line=$_)");
    };
  }

  # Start of header section
  elsif (m#^(.*)(\<(?:ids|tei)Header.*)$#) {
    my $content = "$2\n";

    if ($1 !~ /^\s*$/) {
      die $log->fatal(
        "input line number $.: " .
          'line with opening header tag is not in expected format ... ' .
          "=> Aborting (line=$_)");
    };

    # Parse header
    my $header = KorAP::XML::TEI::Header->new($content, $input_enc, $text_id_esc)->parse($input_fh);

    # Header was parseable
    if ($header) {

      # Write header to zip
      my $file = $header->dir . '/' . $header_file . '.xml';

      $log->debug("Writing file $file") if DEBUG;

      $header->to_zip($zipper->new_stream($file));

      # Header is for text level
      if ($header->type eq 'text') {

        # Remember dir and sigles
        $dir         = $header->dir;
        $text_id_esc = $header->id_esc;

        # log output for seeing progression
        $log->notice("$0: text_id=$text_id_esc");

        # Reset counter for text lines
        # (needed for whitespace handling)
        $text_line = 0;
      };
    };
  };
};

$zipper->close;

$ext_tok->close if $ext_tok;

close $input_fh;


__END__

=pod

=encoding utf8

=head1 NAME

tei2korapxml - Conversion of TEI P5 based formats to KorAP-XML

=head1 SYNOPSIS

  cat corpus.i5.xml | tei2korapxml - > corpus.korapxml.zip

=head1 DESCRIPTION

C<tei2korapxml> is a script to convert TEI P5 and
L<I5|https://www.ids-mannheim.de/digspra/kl/projekte/korpora/textmodell>
based documents to the
L<KorAP-XML format|https://github.com/KorAP/KorAP-XML-Krill#about-korap-xml>.

This program is usually called from inside another script.

=head1 FORMATS

=head2 Input restrictions

=over 2

=item

TEI P5 formatted input with certain restrictions:

=over 4

=item

B<mandatory>: text-header with integrated textsigle
(or convertable identifier), text-body

=item

B<optional>: corp-header with integrated corpsigle,
doc-header with integrated docsigle

=back

=item

All tokens inside the primary text may not be
newline seperated, because newlines are removed
(see L<KorAP::XML::TEI::Data>) and a conversion of newlines
into blanks between 2 tokens could lead to additional blanks,
where there should be none (e.g.: punctuation characters like C<,> or
C<.> should not be seperated from their predecessor token).
(see also code section C<~ whitespace handling ~> in C<script/tei2korapxml>).

=item

Header types, like C<E<lt>idsHeader [...] type="document" [...] E<gt>>
need to be defined in the same line as the header tag.

=back

=head2 Notes on the output

=over 2

=item

zip file output (default on C<stdout>) with utf8 encoded entries
(which together form the KorAP-XML format)

=back

=head1 INSTALLATION

C<tei2korapxml> requires C<libxml2-dev> bindings and L<File::ShareDir::Install> to be installed.
When these requirements are met, the preferred way to install the script is
to use L<cpanm|App::cpanminus>.

  $ cpanm https://github.com/KorAP/KorAP-XML-TEI.git

In case everything went well, the C<tei2korapxml> tool will
be available on your command line immediately.

Minimum requirement for L<KorAP::XML::TEI> is Perl 5.16.

=head1 OPTIONS

=over 2

=item B<--input|-i>

The input file to process. If no specific input is defined and a single
dash C<-> is passed as an argument, data is read from C<STDIN>.


=item B<--root|-r>

The root directory for output. Defaults to C<.>.

=item B<--help|-h>

Print help information.

=item B<--version|-v>

Print version information.

=item B<--tokenizer-korap|-tk>

Use the standard KorAP/DeReKo tokenizer.

=item B<--tokenizer-internal|-ti>

Tokenize the data using two embedded tokenizers,
that will take an I<aggressive> and a I<conservative>
approach.

=item B<--tokenizer-call|-tc>

Call an external tokenizer process, that will tokenize
from STDIN and outputs the offsets of all tokens.

Texts are separated using C<\x04\n>. The external process
should add a new line per text.

If the L</--use-tokenizer-sentence-splits> option is activated,
sentences are marked by offset as well in new lines.

To use L<Datok|https://github.com/KorAP/Datok> including sentence
splitting, call C<tei2korap> as follows:

  $ cat corpus.i5.xml | tei2korapxml -s \
  $   -tc 'datok tokenize \
  $        -t ./tokenizer.matok \
  $        -p --newline-after-eot --no-sentences \
  $        --no-tokens --sentence-positions -' - \
  $        > corpus.korapxml.zip

=item B<--no-tokenizer>

Boolean flag indicating that no tokenizer should be used.
This is meant to ensure that by default a final token layer always
exists.
If a separate tokenizer is chosen, this flag is ignored.

=item B<--skip-inline-tokens>

Boolean flag indicating that inline tokens should not
be processed. Defaults to false (meaning inline tokens will be processed).

=item B<--skip-inline-token-annotations>

Boolean flag indicating that inline token annotations should not
be processed. Defaults to true (meaning inline token annotations
won't be processed).

=item B<--skip-inline-tags> <tags>

Expects a comma-separated list of tags to be ignored when the structure
is parsed. Content of these tags however will be processed.

=item B<--xmlid-to-textsigle> <from-regex>@<to-c/to-d/to-t>

Expects a regular replacement expression (separated by B<@> between the
search and the replacement) to convert text id attributes to text sigles
with three parts (separated by B</>).

Example:

  tei2korapxml  \
    --xmlid-to-textsigle 'ICC.German\.([^.]+\.[^.]+)\.(.+)@ICCGER/$1/$2' \
    -tk - < t/data/icc_german_sample.p5.xml

Converts text id C<ICC.German.DeReKo.WPD17.G11.00238> to
sigle C<ICCGER/DeReKo.WPD17/G11.00238>.

=item B<--inline-tokens> <foundry>#[<file>]

Define the foundry and file (without extension)
to store inline token information in.
Unless C<--skip-inline-token-annotations> is set,
this will contain annotations as well.
Defaults to C<tokens> and C<morpho>.

The inline token data will also be stored in the
inline structures file (see I<--inline-structures>),
unless the inline token foundry is prepended
by an B<!> exclamation mark, indicating that inline
tokens are stored exclusively in the inline tokens
file.

Example:

  tei2korapxml --inline-tokens '!gingko#morpho' < data.i5.xml > korapxml.zip

=item B<--inline-structures> <foundry>#[<file>]

Define the foundry and file (without extension)
to store inline structure information in.
Defaults to C<struct> and C<structures>.

=item B<--base-foundry> <foundry>

Define the base foundry to store newly generated
token information in.
Defaults to C<base>.

=item B<--data-file> <file>

Define the file (without extension)
to store primary data information in.
Defaults to C<data>.

=item B<--header-file> <file>

Define the file name (without extension)
to store header information on
the corpus, document, and text level in.
Defaults to C<header>.

=item B<--use-tokenizer-sentence-splits|-s>

Replace existing with, or add new, sentence boundary information
provided by the tokenizer.
Currently KorAP-tokenizer and certain external tokenizers support
these boundaries.

=item B<--tokens-file> <file>

Define the file (without extension)
to store generated token information in
(either from the KorAP tokenizer or an externally called tokenizer).
Defaults to C<tokens>.

=item B<--log|-l>

Loglevel for I<Log::Any>. Defaults to C<notice>.

=back

=head1 ENVIRONMENT VARIABLES

=over 2

=item B<KORAPXMLTEI_DEBUG>

Activate minimal debugging.
Defaults to C<false>.

=back

=head1 COPYRIGHT AND LICENSE

Copyright (C) 2021-2023, L<IDS Mannheim|https://www.ids-mannheim.de/>

Author: Peter Harders

Contributors: Nils Diewald, Marc Kupietz, Carsten Schnober

L<KorAP::XML::TEI> is developed as part of the L<KorAP|https://korap.ids-mannheim.de/>
Corpus Analysis Platform at the
L<Leibniz Institute for the German Language (IDS)|https://www.ids-mannheim.de/>,
member of the
L<Leibniz-Gemeinschaft|http://www.leibniz-gemeinschaft.de/>.

This program is free software published under the
L<BSD-2 License|https://opensource.org/licenses/BSD-2-Clause>.

=cut

# NOTES

## Notes on segfault prevention

binmode on the input handler prevents segfaulting of 'XML::LibXML::Reader' inside the main loop
(see notes on 'PerlIO layers' in  'man XML::LibXML'),
removing 'use open qw(:std :utf8)' would fix this problem too, but using binmode on input is more granular
see in perluniintro: You can switch encodings on an already opened stream by using "binmode()
see in perlfunc: If LAYER is omitted or specified as ":raw" the filehandle is made suitable for passing binary data.