Move inline parser to separate class
Change-Id: I835acf7a234b385dc3d2a4e52c6ada61dbe087db
diff --git a/script/tei2korapxml b/script/tei2korapxml
index 80b0596..5740407 100755
--- a/script/tei2korapxml
+++ b/script/tei2korapxml
@@ -11,9 +11,6 @@
use Encode qw(decode);
-use XML::CompactTree::XS;
-use XML::LibXML::Reader;
-
use FindBin;
BEGIN {
unshift @INC, "$FindBin::Bin/../lib";
@@ -23,10 +20,9 @@
use KorAP::XML::TEI::Tokenizer::External;
use KorAP::XML::TEI::Tokenizer::Conservative;
use KorAP::XML::TEI::Tokenizer::Aggressive;
-use KorAP::XML::TEI::Annotations::Collector;
-use KorAP::XML::TEI::Data;
use KorAP::XML::TEI::Zipper;
use KorAP::XML::TEI::Header;
+use KorAP::XML::TEI::Inline;
eval {
require KorAP::XML::TEI::Tokenizer::KorAP;
@@ -39,16 +35,7 @@
use constant {
# Set to 1 for minimal more debug output (no need to be parametrized)
- DEBUG => $ENV{KORAPXMLTEI_DEBUG} // 0,
-
- # XCT_LINE_NUMBERS is only needed for debugging
- # (see XML::CompactTree::XS)
- XCT_PARAM => (
- XCT_DOCUMENT_ROOT
- | XCT_IGNORE_COMMENTS
- | XCT_ATTRIBUTE_ARRAY
- | ($ENV{KORAPXMLTEI_DEBUG} ? XCT_LINE_NUMBERS : 0)
- )
+ DEBUG => $ENV{KORAPXMLTEI_DEBUG} // 0
};
# Parse options from the command line
@@ -99,9 +86,6 @@
# TODO: IDS-specific (and redundant)
my $_HEADER_TAG = 'idsHeader';
-# name of the tag containing all information stored in $_tokens_file
-my $_TOKENS_TAG = 'w';
-
# Define tokenizers
if ($use_tokenizer_sentence_splits && !$tokenizer_korap) {
@@ -149,45 +133,21 @@
# Handling inline annotations (inside $_TOKENS_TAG)
my $_INLINE_ANNOT = $ENV{KORAPXMLTEI_INLINE} ? 1 : 0;
-# Initialize Token- and Structure-Collector
-my $tokens = KorAP::XML::TEI::Annotations::Collector->new;
-my $structures = KorAP::XML::TEI::Annotations::Collector->new;
-
-# Initialize Data-Collector
-my $data = KorAP::XML::TEI::Data->new;
-
# 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;
-# element from $tree_data
-my $e;
-
# Default encoding of the text
my $input_enc = 'UTF-8';
-# variables for handling ~ whitespace related issue ~
-# (it is sometimes necessary, to correct the from-values for some tags)
-my $add_one;
-
# text line (needed for whitespace handling)
my $text_line = 0;
-# hash for indices of whitespace-nodes
-# (needed to recorrect from-values)
-# IDEA:
-# when closing element, check if it's from-index minus 1 refers to a whitespace-node
-# (means: 'from-index - 1' is a key in %ws).
-# if this is _not_ the case, then the from-value is one
-# to high => correct it by substracting 1
-my %ws;
-
# Input file handle (default: stdin)
my $input_fh = *STDIN;
@@ -202,6 +162,13 @@
binmode $input_fh;
+# Create inline parser object
+my $inline = KorAP::XML::TEI::Inline->new(
+ $skip_inline_tokens,
+ \%skip_inline_tags
+);
+
+
# Reading input document
MAIN: while (<$input_fh>) {
@@ -252,29 +219,20 @@
if ($dir eq '') {
$log->warn(
"Maybe empty textSigle => skipping this text ...\n" .
- 'data=' . substr($data->data, 0, 200)
+ 'data=' . substr($inline->data->data, 0, 200)
);
next MAIN;
};
- my $reader = XML::LibXML::Reader->new(
- string => "<text>$text_buffer</text>",
- huge => 1
- );
-
- my $tree_data = XML::CompactTree::XS::readSubtreeToPerl($reader, XCT_PARAM);
-
- # ~ whitespace related issue ~
- $add_one = 0;
- %ws = ();
-
- # Recursively parse all children
- descend(1, $tree_data->[2]);
+ # 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/${data_file}.xml"),
@@ -291,7 +249,7 @@
);
if ($use_tokenizer_sentence_splits) {
- $ext_tok->sentencize_from_previous_input($structures);
+ $ext_tok->sentencize_from_previous_input($inline->structures);
};
};
@@ -311,29 +269,26 @@
};
# ~ write structures ~
- if (!$structures->empty) {
- $structures->to_zip(
+ if (!$inline->structures->empty) {
+ $inline->structures->to_zip(
$zipper->new_stream("$dir/$_structure_dir/${_structure_file}.xml"),
$text_id_esc,
2 # = structure serialization
- )->reset;
+ );
};
# ~ write tokens ~
- unless ($skip_inline_tokens || $tokens->empty) {
- $tokens->to_zip(
+ unless ($skip_inline_tokens || $inline->tokens->empty) {
+ $inline->tokens->to_zip(
$zipper->new_stream("$dir/$_tokens_dir/${_tokens_file}.xml"),
$text_id_esc,
$_INLINE_ANNOT # Either 0 = tokens without inline or 1 = tokens with inline
- )->reset;
+ );
};
# reinit.
$dir = '';
- # Maybe not necessary
- $data->reset;
-
next MAIN;
};
@@ -427,144 +382,6 @@
close $input_fh;
-exit(0);
-
-
-# Recursively called function to handle XML tree data
-sub descend {
-
- # recursion level
- # (1 = topmost level inside descend() = should always be level of tag $_TEXT_BODY)
- my $depth = shift;
-
- # Iteration through all array elements
- # ($_[0] is a reference to an array reference)
- # See notes on how 'XML::CompactTree::XS' works and
- # see 'NODE TYPES' in manpage of XML::LibXML::Reader
- foreach $e (@{$_[0]}) {
-
- # $e->[1] represents the tag name of an element node
- # or the primary data of a text or ws node
- my $node_info = $e->[1];
-
- # Element node
- if ($e->[0] == XML_READER_TYPE_ELEMENT) {
-
- # Deal with opening tag
-
- # Get the child index depending on the debug state.
- # This is likely to be optimized away by the compiler.
- my $children = $e->[DEBUG ? 5 : 4];
-
- # Skip certain tags
- if ($skip_inline_tags{$node_info}) {
- descend($depth + 1, $children) if defined $children;
- next;
- };
-
- my $anno = $structures->add_new_annotation($node_info);
-
- # Add element also to token list
- if (!$skip_inline_tokens && $node_info eq $_TOKENS_TAG) {
- $tokens->add_annotation($anno);
- };
-
- # Handle attributes (if attributes exist)
- if (defined $e->[3]) {
-
- # with 'XCT_ATTRIBUTE_ARRAY', $node->[3] is an array reference of the form
- # [ name1, value1, name2, value2, ....] of attribute names and corresponding values.
- # NOTE:
- # arrays are faster (see: http://makepp.sourceforge.net/2.0/perl_performance.html)
- for (local $_ = 0; $_ < @{$e->[3]}; $_ += 2) {
- $anno->add_attribute(
- @{$e->[3]}[$_, $_ + 1]
- );
- };
- };
-
- # this is, where a normal tag or tokens-tag ($_TOKENS_TAG) starts
- $anno->set_from($data->position + $add_one);
-
-
- # Call function recursively
- # do no recursion, if $children is not defined
- # (because we have no array of child-nodes, e.g.: <back/>)
- descend($depth+1, $children) if defined $children;
-
-
- # Deal with closing tag
-
- # NOTE:
- # use $pos, because the offsets are _between_ the characters
- # (e.g.: word = 'Hello' => from = 0 (before 'H'), to = 5 (after 'o'))
- my $pos = $data->position;
-
- # Handle structures and tokens
-
- my $from = $anno->from;
-
- # ~ whitespace related issue ~
- if ($from > 0 && not exists $ws{$from - 1}) {
-
- # Previous node was a text-node
- $anno->set_from($from - 1);
- };
-
- # in case this fails, check input
- if (($from - 1) > $pos) {
- die $log->fatal(
- "text_id='$text_id_esc', " .
- 'processing of structures: ' .
- "from-value ($from) is 2 or more greater " .
- "than to-value ($pos) => please check. Aborting"
- );
- };
-
- # TODO:
- # find example for which this case applies
- # maybe this is not necessary anymore, because the
- # above recorrection of the from-value suffices
- #
- # TODO:
- # check, if it's better to remove this line and
- # change above check to 'if ($from - 1) >= $pos;
- # do testing with bigger corpus excerpt (wikipedia?)
- $anno->set_from($pos) if $from == $pos + 1;
- $anno->set_to($pos);
- $anno->set_level($depth);
-
- # Clean up whitespace
- delete $ws{$from - 1} if $from > 0 && exists $ws{$from - 1};
- }
-
- # Text node
- elsif ($e->[0] == XML_READER_TYPE_TEXT) {
-
- $add_one = 1;
- $data->append($node_info);
- }
-
- # Whitespace node
- # (See notes on whitespace handling - regarding XML_READER_TYPE_SIGNIFICANT_WHITESPACE)
- elsif ($e->[0] == XML_READER_TYPE_SIGNIFICANT_WHITESPACE) {
-
- # state, that this from-index belongs to a whitespace-node
- # ('++' doesn't mean a thing here - maybe it could be used for a consistency check)
- $ws{$data->position}++;
-
- $add_one = 0;
- $data->append($node_info);
- }
-
- # not yet handled type
- else {
-
- die $log->fatal('Not yet handled type ($e->[0]=' . $e->[0] . ') ... => Aborting');
- };
- };
-};
-
__END__
@@ -780,186 +597,6 @@
# NOTES
-## Notes on how 'XML::CompactTree::XS' works
-
-Example: <node a="v"><node1>some <n/> text</node1><node2>more-text</node2></node>
-
-Print out name of 'node2' for the above example:
-
-echo '<node a="v"><node1>some <n/> text</node1><node2>more-text</node2></node>' | perl -e 'use XML::CompactTree::XS; use XML::LibXML::Reader; $reader = XML::LibXML::Reader->new(IO => STDIN); $data = XML::CompactTree::XS::readSubtreeToPerl( $reader, XCT_DOCUMENT_ROOT | XCT_IGNORE_COMMENTS | XCT_LINE_NUMBERS ); print "\x27".$data->[2]->[0]->[5]->[1]->[1]."\x27\n"'
-
-Exploring the structure of $data ( = reference to below array ):
-
-[ 0: XML_READER_TYPE_DOCUMENT,
- 1: ?
- 2: [ 0: [ 0: XML_READER_TYPE_ELEMENT <- start recursion with array '$data->[2]' (see descend( \$tree_data->[2] ))
- 1: 'node'
- 2: ?
- 3: HASH (attributes)
- 4: 1 (line number)
- 5: [ 0: [ 0: XML_READER_TYPE_ELEMENT
- 1: 'node1'
- 2: ?
- 3: undefined (no attributes)
- 4: 1 (line number)
- 5: [ 0: [ 0: XML_READER_TYPE_TEXT
- 1: 'some '
- ]
- 1: [ 0: XML_READER_TYPE_ELEMENT
- 1: 'n'
- 2: ?
- 3: undefined (no attributes)
- 4: 1 (line number)
- 5: undefined (no child-nodes)
- ]
- 2: [ 0: XML_READER_TYPE_TEXT
- 1: ' text'
- ]
- ]
- ]
- 1: [ 0: XML_READER_TYPE_ELEMENT
- 1: 'node2'
- 2: ?
- 3: undefined (not attributes)
- 4: 1 (line number)
- 5: [ 0: [ 0: XML_READER_TYPE_TEXT
- 1: 'more-text'
- ]
- ]
- ]
- ]
- ]
- ]
-]
-
-$data->[0] = 9 (=> type == XML_READER_TYPE_DOCUMENT)
-
-ref($data->[2]) == ARRAY (with 1 element for 'node')
-ref($data->[2]->[0]) == ARRAY (with 6 elements)
-
-$data->[2]->[0]->[0] == 1 (=> type == XML_READER_TYPE_ELEMENT)
-$data->[2]->[0]->[1] == 'node'
-ref($data->[2]->[0]->[3]) == HASH (=> ${$data->[2]->[0]->[3]}{a} == 'v')
-$data->[2]->[0]->[4] == 1 (line number)
-ref($data->[2]->[0]->[5]) == ARRAY (with 2 elements for 'node1' and 'node2')
- # child-nodes of actual node (see $children)
-
-ref($data->[2]->[0]->[5]->[0]) == ARRAY (with 6 elements)
-$data->[2]->[0]->[5]->[0]->[0] == 1 (=> type == XML_READER_TYPE_ELEMENT)
-$data->[2]->[0]->[5]->[0]->[1] == 'node1'
-$data->[2]->[0]->[5]->[0]->[3] == undefined (=> no attribute)
-$data->[2]->[0]->[5]->[0]->[4] == 1 (line number)
-ref($data->[2]->[0]->[5]->[0]->[5]) == ARRAY (with 3 elements for 'some ', '<n/>' and ' text')
-
-ref($data->[2]->[0]->[5]->[0]->[5]->[0]) == ARRAY (with 2 elements)
-$data->[2]->[0]->[5]->[0]->[5]->[0]->[0] == 3 (=> type == XML_READER_TYPE_TEXT)
-$data->[2]->[0]->[5]->[0]->[5]->[0]->[1] == 'some '
-
-ref($data->[2]->[0]->[5]->[0]->[5]->[1]) == ARRAY (with 5 elements)
-$data->[2]->[0]->[5]->[0]->[5]->[1]->[0] == 1 (=> type == XML_READER_TYPE_ELEMENT)
-$data->[2]->[0]->[5]->[0]->[5]->[1]->[1] == 'n'
-$data->[2]->[0]->[5]->[0]->[5]->[1]->[3] == undefined (=> no attribute)
-$data->[2]->[0]->[5]->[0]->[5]->[1]->[4] == 1 (line number)
-$data->[2]->[0]->[5]->[0]->[5]->[1]->[5] == undefined (=> no child-nodes)
-
-ref($data->[2]->[0]->[5]->[0]->[5]->[2]) == ARRAY (with 2 elements)
-$data->[2]->[0]->[5]->[0]->[5]->[2]->[0] == 3 (=> type == XML_READER_TYPE_TEXT)
-$data->[2]->[0]->[5]->[0]->[5]->[2]->[1] == ' text'
-
-
-descend() starts with the array reference ${$_[0]} (= \$tree_data->[2]), which corresponds to ${\$data->[2]} in the above example.
-Hence, the expression @{${$_[0]}} corresponds to @{${\$data->[2]}}, $e to ${${\$data->[2]}}[0] (= $data->[2]->[0]) and $e->[0] to
-${${\$data->[2]}}[0]->[0] (= $data->[2]->[0]->[0]).
-
-
-## Notes on whitespace handling
-
-Every whitespace inside the processed text is 'significant' and recognized as a node of type 'XML_READER_TYPE_SIGNIFICANT_WHITESPACE'
-(see function 'descend()').
-
-Definition of significant and insignificant whitespace
-(source: https://www.oracle.com/technical-resources/articles/wang-whitespace.html):
-
-Significant whitespace is part of the document content and should be preserved.
-Insignificant whitespace is used when editing XML documents for readability.
-These whitespaces are typically not intended for inclusion in the delivery of the document.
-
-### Regarding XML_READER_TYPE_SIGNIFICANT_WHITESPACE
-
-The 3rd form of nodes, besides text- (XML_READER_TYPE_TEXT) and tag-nodes (XML_READER_TYPE_ELEMENT) are nodes of the type
- 'XML_READER_TYPE_SIGNIFICANT_WHITESPACE'.
-
-When modifiying the previous example (see: Notes on how 'XML::CompactTree::XS' works) by inserting an additional blank between
- '</node1>' and '<node2>', the output for '$data->[2]->[0]->[5]->[1]->[1]' is a blank (' ') and it's type is '14'
- (XML_READER_TYPE_SIGNIFICANT_WHITESPACE, see 'man XML::LibXML::Reader'):
-
-echo '<node a="v"><node1>some <n/> text</node1> <node2>more-text</node2></node>' | perl -e 'use XML::CompactTree::XS; use XML::LibXML::Reader; $reader = XML::LibXML::Reader->new(IO => STDIN); $data = XML::CompactTree::XS::readSubtreeToPerl( $reader, XCT_DOCUMENT_ROOT | XCT_IGNORE_COMMENTS | XCT_LINE_NUMBERS ); print "node=\x27".$data->[2]->[0]->[5]->[1]->[1]."\x27, type=".$data->[2]->[0]->[5]->[1]->[0]."\n"'
-
-
-Example: '... <head type="main"><s>Campagne in Frankreich</s></head><head type="sub"> <s>1792</s> ...'
-
-Two text-nodes should normally be separated by a blank. In the above example, that would be the 2 text-nodes
- 'Campagne in Frankreich' and '1792', which are separated by the whitespace-node ' ' (see [2]).
-
-The text-node 'Campagne in Frankreich' leads to the setting of '$add_one' to 1, so that when opening the 2nd 'head'-tag,
- it's from-index gets set to the correct start-index of '1792' (and not to the start-index of the whitespace-node ' ').
-
-The assumption here is, that in most cases there _is_ a whitespace node between 2 text-nodes. The below code fragment
- enables a way, to check, if this really _was_ the case for the last 2 'non-tag'-nodes, when closing a tag:
-
-When a whitespace-node is read, its from-index is stored as a hash-key (in %ws), to state that it belongs to a ws-node.
- So when closing a tag, it can be checked, if the previous 'non-tag'-node (text or whitespace), which is the one before
- the last read 'non-tag'-node, was a actually _not_ a ws-node, but instead a text-node. In that case, the from-value of
- the last read 'non-tag'-node has to be corrected (see [1]),
-
-For whitespace-nodes $add_one is set to 0, so when opening the next tag (in the above example the 2nd 's'-tag), no
- additional 1 is added (because this was already done by the whitespace-node itself when incrementing the variable $pos).
-
-[1]
-Now, what happens, when 2 text-nodes are _not_ seperated by a whitespace-node (e.g.: <w>Augen<c>,</c></w>)?
- In this case, the falsely increased from-value has to be decreased again by 1 when closing the enclosing tag
- (see above code fragment '... not exists $ws{ $from - 1 } ...').
-
-[2]
-Comparing the 2 examples '<w>fu</w> <w>bar</w>' and '<w>fu</w><w> </w><w>bar</w>', is ' ' in both cases handled as a
- whitespace-node (XML_READER_TYPE_SIGNIFICANT_WHITESPACE).
-
-The from-index of the 2nd w-tag in the second example refers to 'bar', which may not have been the intention
- (even though '<w> </w>' doesn't make a lot of sense). TODO: could this be a bug?
-
-Empty tags also cling to the next text-token - e.g. in '<w>tok1</w> <w>tok2</w><a><b/></a> <w>tok3</w>' are the from-
- and to-indizes for the tags 'a' and 'b' both 12, which is the start-index of the token 'tok3'.
-
-
-## Notes on whitespace fixing
-
-The idea for the below code fragment was to fix (recreate) missing whitespace in a poorly created corpus, in which linebreaks where inserted
- into the text with the addition that maybe (or not) whitespace before those linebreaks was unintenionally stripped.
-
-It soon turned out, that it was best to suggest considering just avoiding linebreaks and putting all primary text tokens into one line (see
- example further down and notes on 'Input restrictions' in the manpage).
-
-Somehow an old first very poor approach remained, which is not stringent, but also doesn't affect one-line text.
-
-Examples (how primary text with linebreaks would be converted by below code):
-
- '...<w>end</w>\n<w>.</w>...' -> '...<w>end</w> <w>.</w>...'
- '...<w>,</w>\n<w>this</w>\n<w>is</w>\n<w>it</w>\n<w>!</w>...' -> '<w>,<w> <w>this</w> <w>is</w> <w>it</w> <w>!</w>'.
-
-Blanks are inserted before the 1st character:
-
- NOTE: not stringent ('...' stands for text):
-
- beg1............................end1 => no blank before 'beg1'
- beg2....<pb/>...................end2 => no blank before 'beg2'
- beg3....<info attr1="val1"/>....end3 => no blank before 'beg3'
- beg4....<test>ok</test>.........end4 => blank before 'beg4'
-
- => beg1....end1beg2...<pb/>...end2beg3....<info attr1="val1"/>....end3 beg4...<test>ok</test>....end4
- ^
- |_blank between 'end3' and 'beg4'
-
-
## Notes on segfault prevention
binmode on the input handler prevents segfaulting of 'XML::LibXML::Reader' inside the main loop