コード例 #1
0
ファイル: wordlist.cpp プロジェクト: alerque/bibledit
void Wordlist::process_line(ustring & line, set <ustring> &section_entries)
// Processes one line of text:
// - deals with entries.
// - deals with asterisks.
// section_entries - has the entries already made in this section.
{
  // Handle section restart.
  {
    ustring s (line);
    ustring marker = usfm_extract_marker (s);
    if (section_markers.find (marker) != section_markers.end()) {
      section_entries.clear();
    }
  }
  // Remove the asterisk before and the asterisk after the closer, e.g.:
  // \w entry*\w* -> \w entry\w*
  // \w entry\w** -> \w entry\w*
  replace_text(line, "*" + entry_closer, entry_closer);
  replace_text(line, entry_closer + "*", entry_closer);
  // Go through the line looking for the opener.
  size_t startpos = line.find(entry_opener);
  while (startpos != string::npos) {
    // Look for the closer too, after the opener, not before.
    size_t endpos = line.find(entry_closer, startpos);
    if (endpos == string::npos)
      break;
    // Get the word.
    ustring word = line.substr(startpos + entry_opener.length(), endpos - startpos - entry_closer.length());
    if (!word.empty()) {
      // Store the word.      
      words.insert(word);
      wordcount++;
      // Handle asterisks.
      if (use_asterisk) {
        bool insert_asterisk = true;
        if (first_in_chapter)
          if (section_entries.find(word) != section_entries.end())
            insert_asterisk = false;
        if (insert_asterisk) {
          line.insert(endpos + entry_closer.length(), "*");
        }
        section_entries.insert(word);
      }
    }
    startpos = line.find(entry_opener, endpos);
  }
}
コード例 #2
0
ファイル: categorize.cpp プロジェクト: postiffm/bibledit-gtk
void CategorizeLine::CategorizeOneLine(ustring &line)
{
    // Extract the marker for this line.
    ustring marker = usfm_extract_marker(line);
    
    // Marker to work with for this line.
    ustring actual_marker;
    if (marker.empty())
        actual_marker = previous_marker;
    else
        actual_marker = marker;
    
    // Re-insert bible note markers.
    if (marker == footnote_opener(false)) {
        line.insert(0, footnote_opener(true));
    }
    if (marker == endnote_opener(false)) {
        line.insert(0, endnote_opener(true));
    }
    if (marker == xref_opener(false)) {
        line.insert(0, xref_opener(true));
    }
    // Subdivide the line into categories.
    // Each category removes something from this content, until it is empty.
    
    // Footnotes, endnotes.
    if (!line.empty()) {
        size_t beginposition, endposition;
        ustring opening_marker = footnote_opener(true);
        ustring closing_marker = footnote_closer();
        beginposition = line.find(opening_marker);
        endposition = line.find(closing_marker);
        if (endposition == string::npos) {
            // BUG: In main editor, the chapter lines are all combined into a single long ustring.
            // This means that footnotes that are split across lines are "fixed." But that is not
            // the case when just looking at a single verse \\v line that has a footnote started
            // but not finished. Fixed 2/15/2018
            endposition = line.length() - 1;
        }
        while (beginposition != string::npos) {
            assert(beginposition <= endposition); // this cannot be above the while stmt...beginposition == string::npos if no marker is found
            ustring notetext;
            notetext = line.substr(beginposition + opening_marker.length(), endposition - beginposition - closing_marker.length());
            line.erase(beginposition, endposition - beginposition + closing_marker.length());
            clear_out_any_marker(notetext);
            append_text(note, notetext);
            beginposition = line.find(opening_marker);
            endposition = line.find(closing_marker);
        }
        opening_marker = endnote_opener(true);
        closing_marker = endnote_closer();
        beginposition = line.find(opening_marker);
        endposition = line.find(closing_marker);
        while ((beginposition != string::npos) && (endposition != string::npos)) {
            ustring notetext;
            notetext = line.substr(beginposition + opening_marker.length(), endposition - beginposition - closing_marker.length());
            line.erase(beginposition, endposition - beginposition + closing_marker.length());
            clear_out_any_marker(notetext);
            append_text(note, notetext);
            beginposition = line.find(opening_marker);
            endposition = line.find(closing_marker);
        }
    }
    // Crossreferences.
    if (!line.empty()) {
        size_t beginposition, endposition;
        ustring opening_marker = xref_opener(true);
        ustring closing_marker = xref_closer();
        beginposition = line.find(opening_marker);
        endposition = line.find(closing_marker);
        while ((beginposition != string::npos) && (endposition != string::npos)) {
            ustring referencetext;
            referencetext = line.substr(beginposition + opening_marker.length(), endposition - beginposition - closing_marker.length());
            line.erase(beginposition, endposition - beginposition + closing_marker.length());
            clear_out_any_marker(referencetext);
            append_text(ref, referencetext);
            beginposition = line.find(opening_marker);
            endposition = line.find(closing_marker);
        }
    }
    // Identifiers.
    if (!line.empty()) {
        if (is_id_marker(actual_marker)) {
            clear_out_any_marker(line);
            append_text(id, line);
            line.clear();
        }
    }
    // Introduction elements.
    if (!line.empty()) {
        if (is_intro_marker(actual_marker)) {
            clear_out_any_marker(line);
            append_text(intro, line);
            line.clear();
        }
    }
    // Headings, titles, labels.
    if (!line.empty()) {
        if (is_head_marker(actual_marker)) {
            clear_out_any_marker(line);
            append_text(head, line);
            line.clear();
        }
    }
    // Chapter text.
    if (!line.empty()) {
        if (is_chap_marker(actual_marker)) {
            clear_out_any_marker(line);
            append_text(chap, line);
            line.clear();
        }
    }
    // Extended study notes. As these use the existing footnote markers, 
    // deal with the study notes first.
    if (!line.empty()) {
        if (is_study_marker(actual_marker)) {
            clear_out_any_marker(line);
            append_text(study, line);
            line.clear();
        }
    }
    // After everything else has been removed, output the rest as main text.
    // This includes the "Verses" group, the "Paragraph Elements", and the
    // "Poetry Elements", the "Table Elements", and the "Special Text and
    // Character Styles", which have been filtered out already above.
    if (!line.empty()) {
        clear_out_any_marker(line);
        append_text(verse, line);
        line.clear();
    }
    // Store previous marker.
    previous_marker = marker;
    
}