src/Paragraph.cpp

   1 /**
   2  * \file Paragraph.cpp
   3  * This file is part of LyX, the document processor.
   4  * Licence details can be found in the file COPYING.
   5  *
   6  * \author Asger Alstrup
   7  * \author Lars Gullik Bjønnes
   8  * \author Richard Kimberly Heck (XHTML output)
   9  * \author Jean-Marc Lasgouttes
  10  * \author Angus Leeming
  11  * \author John Levon
  12  * \author André Pönitz
  13  * \author Dekel Tsur
  14  * \author Jürgen Vigna
  15  *
  16  * Full author contact details are available in file CREDITS.
  17  */
  18
  19 #include <config.h>
  20
  21 #include "Paragraph.h"
  22
  23 #include "Buffer.h"
  24 #include "BufferParams.h"
  25 #include "BufferEncodings.h"
  26 #include "Changes.h"
  27 #include "Counters.h"
  28 #include "InsetList.h"
  29 #include "Language.h"
  30 #include "LaTeXFeatures.h"
  31 #include "Layout.h"
  32 #include "Font.h"
  33 #include "FontList.h"
  34 #include "LyXRC.h"
  35 #include "OutputParams.h"
  36 #include "output_latex.h"
  37 #include "output_xhtml.h"
  38 #include "output_docbook.h"
  39 #include "ParagraphParameters.h"
  40 #include "Session.h"
  41 #include "SpellChecker.h"
  42 #include "texstream.h"
  43 #include "TexRow.h"
  44 #include "Text.h"
  45 #include "TextClass.h"
  46 #include "WordLangTuple.h"
  47 #include "WordList.h"
  48
  49 #include "frontends/alert.h"
  50
  51 #include "insets/InsetArgument.h"
  52 #include "insets/InsetBibitem.h"
  53 #include "insets/InsetLabel.h"
  54 #include "insets/InsetSpecialChar.h"
  55 #include "insets/InsetText.h"
  56
  57 #include "mathed/InsetMathHull.h"
  58
  59 #include "support/debug.h"
  60 #include "support/docstring_list.h"
  61 #include "support/ExceptionMessage.h"
  62 #include "support/gettext.h"
  63 #include "support/lassert.h"
  64 #include "support/lstrings.h"
  65 #include "support/textutils.h"
  66
  67 #include <atomic>
  68 #include <sstream>
  69 #include <vector>
  70
  71 using namespace std;
  72 using namespace lyx::support;
  73
  74 // OSX clang and msvc < 2015 do not support C++11 thread_local
  75 #if defined(__APPLE__)
  76 #define THREAD_LOCAL_STATIC static __thread
  77 #elif defined(_MSC_VER) && (_MSC_VER < 1900)
  78 #define THREAD_LOCAL_STATIC static __declspec(thread)
  79 #else
  80 #define THREAD_LOCAL_STATIC thread_local static
  81 #endif
  82
  83 namespace lyx {
  84
  85 /////////////////////////////////////////////////////////////////////
  86 //
  87 // SpellResultRange
  88 //
  89 /////////////////////////////////////////////////////////////////////
  90
  91 class SpellResultRange {
  92 public:
  93         SpellResultRange(FontSpan range, SpellChecker::Result result)
  94         : range_(range), result_(result)
  95         {}
  96         ///
  97         FontSpan const & range() const { return range_; }
  98         ///
  99         void range(FontSpan const & r) { range_ = r; }
 100         ///
 101         SpellChecker::Result result() const { return result_; }
 102         ///
 103         void result(SpellChecker::Result r) { result_ = r; }
 104         ///
 105         bool contains(pos_type pos) const { return range_.contains(pos); }
 106         ///
 107         bool covered(FontSpan const & r) const
 108         {
 109                 // 1. first of new range inside current range or
 110                 // 2. last of new range inside current range or
 111                 // 3. first of current range inside new range or
 112                 // 4. last of current range inside new range
 113                 //FIXME: is this the same as !range_.intersect(r).empty() ?
 114                 return range_.contains(r.first) || range_.contains(r.last) ||
 115                         r.contains(range_.first) || r.contains(range_.last);
 116         }
 117         ///
 118         void shift(pos_type pos, int offset)
 119         {
 120                 if (range_.first > pos) {
 121                         range_.first += offset;
 122                         range_.last += offset;
 123                 } else if (range_.last >= pos) {
 124                         range_.last += offset;
 125                 }
 126         }
 127 private:
 128         FontSpan range_ ;
 129         SpellChecker::Result result_ ;
 130 };
 131
 132
 133 /////////////////////////////////////////////////////////////////////
 134 //
 135 // SpellCheckerState
 136 //
 137 /////////////////////////////////////////////////////////////////////
 138
 139 class SpellCheckerState {
 140 public:
 141         SpellCheckerState()
 142         {
 143                 needs_refresh_ = true;
 144                 current_change_number_ = 0;
 145         }
 146
 147         void setRange(FontSpan const & fp, SpellChecker::Result state)
 148         {
 149                 Ranges result;
 150                 RangesIterator et = ranges_.end();
 151                 RangesIterator it = ranges_.begin();
 152                 for (; it != et; ++it) {
 153                         if (!it->covered(fp))
 154                                 result.push_back(SpellResultRange(it->range(), it->result()));
 155                         else if (state == SpellChecker::WORD_OK) {
 156                                 // trim or split the current misspelled range
 157                                 // store misspelled ranges only
 158                                 FontSpan range = it->range();
 159                                 if (fp.first > range.first) {
 160                                         // misspelled area in front of WORD_OK
 161                                         range.last = fp.first - 1;
 162                                         result.push_back(SpellResultRange(range, it->result()));
 163                                         range = it->range();
 164                                 }
 165                                 if (fp.last < range.last) {
 166                                         // misspelled area after WORD_OK range
 167                                         range.first = fp.last + 1;
 168                                         result.push_back(SpellResultRange(range, it->result()));
 169                                 }
 170                         }
 171                 }
 172                 ranges_ = result;
 173                 if (state != SpellChecker::WORD_OK)
 174                         ranges_.push_back(SpellResultRange(fp, state));
 175         }
 176
 177         void increasePosAfterPos(pos_type pos)
 178         {
 179                 correctRangesAfterPos(pos, 1);
 180                 needsRefresh(pos);
 181         }
 182
 183         void decreasePosAfterPos(pos_type pos)
 184         {
 185                 correctRangesAfterPos(pos, -1);
 186                 needsRefresh(pos);
 187         }
 188
 189         void refreshLast(pos_type pos)
 190         {
 191                 if (pos < refresh_.last)
 192                         refresh_.last = pos;
 193         }
 194
 195         SpellChecker::Result getState(pos_type pos) const
 196         {
 197                 SpellChecker::Result result = SpellChecker::WORD_OK;
 198                 RangesIterator et = ranges_.end();
 199                 RangesIterator it = ranges_.begin();
 200                 for (; it != et; ++it) {
 201                         if(it->contains(pos)) {
 202                                 return it->result();
 203                         }
 204                 }
 205                 return result;
 206         }
 207
 208         FontSpan const & getRange(pos_type pos) const
 209         {
 210                 /// empty span to indicate mismatch
 211                 static FontSpan empty_;
 212                 RangesIterator et = ranges_.end();
 213                 RangesIterator it = ranges_.begin();
 214                 for (; it != et; ++it) {
 215                         if(it->contains(pos)) {
 216                                 return it->range();
 217                         }
 218                 }
 219                 return empty_;
 220         }
 221
 222         bool needsRefresh() const
 223         {
 224                 return needs_refresh_;
 225         }
 226
 227         SpellChecker::ChangeNumber currentChangeNumber() const
 228         {
 229                 return current_change_number_;
 230         }
 231
 232         void refreshRange(pos_type & first, pos_type & last) const
 233         {
 234                 first = refresh_.first;
 235                 last = refresh_.last;
 236         }
 237
 238         void needsRefresh(pos_type pos)
 239         {
 240                 if (needs_refresh_ && pos != -1) {
 241                         if (pos < refresh_.first)
 242                                 refresh_.first = pos;
 243                         if (pos > refresh_.last)
 244                                 refresh_.last = pos;
 245                 } else if (pos != -1) {
 246                         // init request check for neighbour positions too
 247                         refresh_.first = pos > 0 ? pos - 1 : 0;
 248                         // no need for special end of paragraph check
 249                         refresh_.last = pos + 1;
 250                 }
 251                 needs_refresh_ = pos != -1;
 252         }
 253
 254         void needsCompleteRefresh(SpellChecker::ChangeNumber change_number)
 255         {
 256                 needs_refresh_ = true;
 257                 refresh_.first = 0;
 258                 refresh_.last = -1;
 259                 current_change_number_ = change_number;
 260         }
 261 private:
 262         typedef vector<SpellResultRange> Ranges;
 263         typedef Ranges::const_iterator RangesIterator;
 264         Ranges ranges_;
 265         /// the area of the paragraph with pending spell check
 266         FontSpan refresh_;
 267         /// spell state cache version number
 268         SpellChecker::ChangeNumber current_change_number_;
 269         bool needs_refresh_;
 270
 271
 272         void correctRangesAfterPos(pos_type pos, int offset)
 273         {
 274                 RangesIterator et = ranges_.end();
 275                 Ranges::iterator it = ranges_.begin();
 276                 for (; it != et; ++it) {
 277                         it->shift(pos, offset);
 278                 }
 279         }
 280
 281 };
 282
 283 /////////////////////////////////////////////////////////////////////
 284 //
 285 // Paragraph::Private
 286 //
 287 /////////////////////////////////////////////////////////////////////
 288
 289 class Paragraph::Private
 290 {
 291         // Enforce our own "copy" constructor
 292         Private(Private const &) = delete;
 293         Private & operator=(Private const &) = delete;
 294         // Unique ID generator
 295         static int make_id();
 296 public:
 297         ///
 298         Private(Paragraph * owner, Layout const & layout);
 299         /// "Copy constructor"
 300         Private(Private const &, Paragraph * owner);
 301         /// Copy constructor from \p beg  to \p end
 302         Private(Private const &, Paragraph * owner, pos_type beg, pos_type end);
 303
 304         ///
 305         void insertChar(pos_type pos, char_type c, Change const & change);
 306
 307         /// Output the surrogate pair formed by \p c and \p next to \p os.
 308         /// \return the number of characters written.
 309         int latexSurrogatePair(BufferParams const &, otexstream & os,
 310                                char_type c, char_type next,
 311                                OutputParams const &) const;
 312
 313         /// Output a space in appropriate formatting (or a surrogate pair
 314         /// if the next character is a combining character).
 315         /// \return whether a surrogate pair was output.
 316         bool simpleTeXBlanks(BufferParams const &,
 317                              OutputParams const &,
 318                              otexstream &,
 319                              pos_type i,
 320                              unsigned int & column,
 321                              Font const & font,
 322                              Layout const & style) const;
 323
 324         /// This could go to ParagraphParameters if we want to.
 325         int startTeXParParams(BufferParams const &, otexstream &,
 326                               OutputParams const &) const;
 327
 328         /// This could go to ParagraphParameters if we want to.
 329         bool endTeXParParams(BufferParams const &, otexstream &,
 330                              OutputParams const &) const;
 331
 332         ///
 333         void latexInset(BufferParams const &,
 334                                    otexstream &,
 335                                    OutputParams &,
 336                                    Font & running_font,
 337                                    Font & basefont,
 338                                    Font const & outerfont,
 339                                    bool & open_font,
 340                                    Change & running_change,
 341                                    Layout const & style,
 342                                    pos_type & i,
 343                                    unsigned int & column,
 344                                    bool const fontswitch_inset,
 345                                    bool const closeLanguage,
 346                                    bool const lang_switched_at_inset) const;
 347
 348         ///
 349         void latexSpecialChar(
 350                                    otexstream & os,
 351                                    BufferParams const & bparams,
 352                                    OutputParams const & runparams,
 353                                    Font const & running_font,
 354                                    string & alien_script,
 355                                    Layout const & style,
 356                                    pos_type & i,
 357                                    pos_type end_pos,
 358                                    unsigned int & column) const;
 359
 360         ///
 361         bool latexSpecialT1(
 362                 char_type const c,
 363                 otexstream & os,
 364                 pos_type i,
 365                 unsigned int & column) const;
 366         ///
 367         bool latexSpecialTU(
 368                 char_type const c,
 369                 otexstream & os,
 370                 pos_type i,
 371                 unsigned int & column) const;
 372         ///
 373         bool latexSpecialT3(
 374                 char_type const c,
 375                 otexstream & os,
 376                 pos_type i,
 377                 unsigned int & column) const;
 378
 379         ///
 380         void validate(LaTeXFeatures & features) const;
 381
 382         /// Checks if the paragraph contains only text and no inset or font change.
 383         bool onlyText(Buffer const & buf, Font const & outerfont,
 384                       pos_type initial) const;
 385
 386         /// a vector of speller skip positions
 387         typedef vector<FontSpan> SkipPositions;
 388         typedef SkipPositions::const_iterator SkipPositionsIterator;
 389
 390         void appendSkipPosition(SkipPositions & skips, pos_type const pos) const;
 391
 392         Language * getSpellLanguage(pos_type const from) const;
 393
 394         Language * locateSpellRange(pos_type & from, pos_type & to,
 395                                     SkipPositions & skips) const;
 396
 397         bool hasSpellerChange() const
 398         {
 399                 SpellChecker::ChangeNumber speller_change_number = 0;
 400                 if (theSpellChecker())
 401                         speller_change_number = theSpellChecker()->changeNumber();
 402                 return speller_change_number > speller_state_.currentChangeNumber();
 403         }
 404
 405         bool ignoreWord(docstring const & word) const;
 406
 407         void setMisspelled(pos_type from, pos_type to, SpellChecker::Result state)
 408         {
 409                 pos_type textsize = owner_->size();
 410                 // check for sane arguments
 411                 if (to <= from || from >= textsize)
 412                         return;
 413                 FontSpan fp = FontSpan(from, to - 1);
 414                 speller_state_.setRange(fp, state);
 415         }
 416
 417         void requestSpellCheck(pos_type pos)
 418         {
 419                 if (pos == -1)
 420                         speller_state_.needsCompleteRefresh(speller_state_.currentChangeNumber());
 421                 else
 422                         speller_state_.needsRefresh(pos);
 423         }
 424
 425         void readySpellCheck()
 426         {
 427                 speller_state_.needsRefresh(-1);
 428         }
 429
 430         bool needsSpellCheck() const
 431         {
 432                 return speller_state_.needsRefresh();
 433         }
 434
 435         void rangeOfSpellCheck(pos_type & first, pos_type & last) const
 436         {
 437                 speller_state_.refreshRange(first, last);
 438                 if (last == -1) {
 439                         last = owner_->size();
 440                         return;
 441                 }
 442                 pos_type endpos = last;
 443                 owner_->locateWord(first, endpos, WHOLE_WORD, true);
 444                 if (endpos < last) {
 445                         endpos = last;
 446                         owner_->locateWord(last, endpos, WHOLE_WORD, true);
 447                 }
 448                 last = endpos;
 449         }
 450
 451         int countSkips(SkipPositionsIterator & it, SkipPositionsIterator const et,
 452                             int & start) const
 453         {
 454                 int numskips = 0;
 455                 while (it != et && it->first < start) {
 456                         long skip = it->last - it->first + 1;
 457                         start += skip;
 458                         numskips += skip;
 459                         ++it;
 460                 }
 461                 return numskips;
 462         }
 463
 464         void markMisspelledWords(Language const * lang,
 465                                  pos_type const & first, pos_type const & last,
 466                                  SpellChecker::Result result,
 467                                  docstring const & word,
 468                                  SkipPositions const & skips);
 469
 470         InsetCode ownerCode() const
 471         {
 472                 return inset_owner_ ? inset_owner_->lyxCode() : NO_CODE;
 473         }
 474
 475         /// Which Paragraph owns us?
 476         Paragraph * owner_;
 477
 478         /// In which Inset?
 479         Inset const * inset_owner_;
 480
 481         ///
 482         FontList fontlist_;
 483
 484         ///
 485         ParagraphParameters params_;
 486
 487         /// for recording and looking up changes
 488         Changes changes_;
 489
 490         ///
 491         InsetList insetlist_;
 492
 493         /// end of label
 494         pos_type begin_of_body_;
 495
 496         typedef docstring TextContainer;
 497         ///
 498         TextContainer text_;
 499
 500         typedef set<docstring> Words;
 501         typedef map<string, Words> LangWordsMap;
 502         ///
 503         LangWordsMap words_;
 504         ///
 505         Layout const * layout_;
 506         ///
 507         SpellCheckerState speller_state_;
 508         ///
 509         int id_;
 510 };
 511
 512
 513 Paragraph::Private::Private(Paragraph * owner, Layout const & layout)
 514         : owner_(owner), inset_owner_(nullptr), begin_of_body_(0), layout_(&layout), id_(-1)
 515 {
 516         text_.reserve(100);
 517 }
 518
 519
 520 //static
 521 int Paragraph::Private::make_id()
 522 {
 523         // The id is unique per session across buffers because it is used in
 524         // LFUN_PARAGRAPH_GOTO to switch to a different buffer, for instance in the
 525         // outliner.
 526         // (thread-safe)
 527         static int next_id(0);
 528         return next_id++;
 529 }
 530
 531
 532 Paragraph::Private::Private(Private const & p, Paragraph * owner)
 533         : owner_(owner), inset_owner_(p.inset_owner_), fontlist_(p.fontlist_),
 534           params_(p.params_), changes_(p.changes_), insetlist_(p.insetlist_),
 535           begin_of_body_(p.begin_of_body_), text_(p.text_), words_(p.words_),
 536           layout_(p.layout_), id_(make_id())
 537 {
 538         requestSpellCheck(p.text_.size());
 539 }
 540
 541
 542 Paragraph::Private::Private(Private const & p, Paragraph * owner,
 543         pos_type beg, pos_type end)
 544         : owner_(owner), inset_owner_(p.inset_owner_),
 545           params_(p.params_), changes_(p.changes_),
 546           insetlist_(p.insetlist_, beg, end),
 547           begin_of_body_(p.begin_of_body_), words_(p.words_),
 548           layout_(p.layout_), id_(make_id())
 549 {
 550         if (beg >= pos_type(p.text_.size()))
 551                 return;
 552         text_ = p.text_.substr(beg, end - beg);
 553
 554         FontList::const_iterator fcit = fontlist_.begin();
 555         FontList::const_iterator fend = fontlist_.end();
 556         for (; fcit != fend; ++fcit) {
 557                 if (fcit->pos() < beg)
 558                         continue;
 559                 if (fcit->pos() >= end) {
 560                         // Add last entry in the fontlist_.
 561                         fontlist_.set(text_.size() - 1, fcit->font());
 562                         break;
 563                 }
 564                 // Add a new entry in the fontlist_.
 565                 fontlist_.set(fcit->pos() - beg, fcit->font());
 566         }
 567         requestSpellCheck(p.text_.size());
 568 }
 569
 570
 571 /////////////////////////////////////////////////////////////////////
 572 //
 573 // Paragraph
 574 //
 575 /////////////////////////////////////////////////////////////////////
 576
 577 namespace {
 578
 579 /** This helper class should be instantiated at the start of methods
 580  * that can create or merge changes. If as a result the value of
 581  * Paragraph::isChanged is modified, it makes sure that updateBuffer()
 582  * will be run.
 583  */
 584 struct ChangesMonitor {
 585         ///
 586         ChangesMonitor(Paragraph & par)
 587                 : par_(par), was_changed_(par.isChanged()) {}
 588         ///
 589         ~ChangesMonitor()
 590         {
 591                 /* We may need to run updateBuffer to check whether the buffer
 592                  * contains changes (and toggle the changes toolbar). We do it
 593                  * when:
 594                  * 1. the `changedness' of the paragraph has changed,
 595                  * 2. and we are not in the situation where the buffer has changes
 596                  * and new changes are added to the paragraph.
 597                  */
 598                 try {
 599                         if (par_.isChanged() != was_changed_
 600                                 && par_.inInset().isBufferValid()
 601                                 && !(par_.inInset().buffer().areChangesPresent() && par_.isChanged()))
 602                                 par_.inInset().buffer().forceUpdate();
 603                 } catch(support::ExceptionMessage const &) {}
 604         }
 605
 606 private:
 607         ///
 608         Paragraph const & par_;
 609         ///
 610         bool was_changed_;
 611 };
 612
 613 }
 614
 615 void Paragraph::addChangesToToc(DocIterator const & cdit, Buffer const & buf,
 616                                 bool output_active, TocBackend & backend) const
 617 {
 618         d->changes_.addToToc(cdit, buf, output_active, backend);
 619 }
 620
 621
 622 bool Paragraph::isDeleted(pos_type start, pos_type end) const
 623 {
 624         LASSERT(start >= 0 && start <= size(), return false);
 625         LASSERT(end > start && end <= size() + 1, return false);
 626
 627         return d->changes_.isDeleted(start, end);
 628 }
 629
 630
 631 bool Paragraph::isChanged(pos_type start, pos_type end) const
 632 {
 633         LASSERT(start >= 0 && start <= size(), return false);
 634         LASSERT(end > start && end <= size() + 1, return false);
 635
 636         return d->changes_.isChanged(start, end);
 637 }
 638
 639 // FIXME: Ideally the diverse isChanged() methods should account for that!
 640 bool Paragraph::hasChangedInsets(pos_type start, pos_type end) const
 641 {
 642         LASSERT(start >= 0 && start <= size(), return false);
 643         LASSERT(end > start && end <= size() + 1, return false);
 644
 645         for (auto const & icit : d->insetlist_) {
 646                 if (icit.pos < start)
 647                         continue;
 648                 if (icit.pos >= end)
 649                         break;
 650                 if (icit.inset && icit.inset->isChanged())
 651                         return true;
 652         }
 653         return false;
 654 }
 655
 656 bool Paragraph::isChanged() const
 657 {
 658         return d->changes_.isChanged();
 659 }
 660
 661
 662 bool Paragraph::isMergedOnEndOfParDeletion(bool trackChanges) const
 663 {
 664         // keep the logic here in sync with the logic of eraseChars()
 665         if (!trackChanges)
 666                 return true;
 667
 668         Change const & change = d->changes_.lookup(size());
 669         return change.inserted() && change.currentAuthor();
 670 }
 671
 672 Change Paragraph::parEndChange() const
 673 {
 674         return d->changes_.lookup(size());
 675 }
 676
 677
 678 void Paragraph::setChange(Change const & change)
 679 {
 680         // Make sure that Buffer::hasChangesPresent is updated
 681         ChangesMonitor cm(*this);
 682
 683         // beware of the imaginary end-of-par character!
 684         d->changes_.set(change, 0, size() + 1);
 685
 686         /*
 687          * Propagate the change recursively - but not in case of DELETED!
 688          *
 689          * Imagine that your co-author makes changes in an existing inset. He
 690          * sends your document to you and you come to the conclusion that the
 691          * inset should go completely. If you erase it, LyX must not delete all
 692          * text within the inset. Otherwise, the change tracked insertions of
 693          * your co-author get lost and there is no way to restore them later.
 694          *
 695          * Conclusion: An inset's content should remain untouched if you delete it
 696          */
 697
 698         if (!change.deleted()) {
 699                 for (pos_type pos = 0; pos < size(); ++pos) {
 700                         if (Inset * inset = getInset(pos))
 701                                 inset->setChange(change);
 702                 }
 703         }
 704 }
 705
 706
 707 void Paragraph::setChange(pos_type pos, Change const & change)
 708 {
 709         // Make sure that Buffer::hasChangesPresent is updated
 710         ChangesMonitor cm(*this);
 711
 712         LASSERT(pos >= 0 && pos <= size(), return);
 713         d->changes_.set(change, pos);
 714
 715         // see comment in setChange(Change const &) above
 716         if (!change.deleted() && pos < size())
 717                 if (Inset * inset = getInset(pos))
 718                         inset->setChange(change);
 719 }
 720
 721
 722 Change const & Paragraph::lookupChange(pos_type pos) const
 723 {
 724         LBUFERR(pos >= 0 && pos <= size());
 725         return d->changes_.lookup(pos);
 726 }
 727
 728
 729 void Paragraph::acceptChanges(pos_type start, pos_type end)
 730 {
 731         // Make sure that Buffer::hasChangesPresent is updated
 732         ChangesMonitor cm(*this);
 733
 734         LASSERT(start >= 0 && start <= size(), return);
 735         LASSERT(end > start && end <= size() + 1, return);
 736
 737         for (pos_type pos = start; pos < end; ++pos) {
 738                 switch (lookupChange(pos).type) {
 739                         case Change::UNCHANGED:
 740                                 // accept changes in nested inset
 741                                 if (Inset * inset = getInset(pos))
 742                                         inset->acceptChanges();
 743                                 break;
 744
 745                         case Change::INSERTED:
 746                                 d->changes_.set(Change(Change::UNCHANGED), pos);
 747                                 // also accept changes in nested inset
 748                                 if (Inset * inset = getInset(pos))
 749                                         inset->acceptChanges();
 750                                 break;
 751
 752                         case Change::DELETED:
 753                                 // Suppress access to non-existent
 754                                 // "end-of-paragraph char"
 755                                 if (pos < size()) {
 756                                         eraseChar(pos, false);
 757                                         --end;
 758                                         --pos;
 759                                 }
 760                                 break;
 761                 }
 762
 763         }
 764 }
 765
 766
 767 void Paragraph::rejectChanges(pos_type start, pos_type end)
 768 {
 769         LASSERT(start >= 0 && start <= size(), return);
 770         LASSERT(end > start && end <= size() + 1, return);
 771
 772         // Make sure that Buffer::hasChangesPresent is updated
 773         ChangesMonitor cm(*this);
 774
 775         for (pos_type pos = start; pos < end; ++pos) {
 776                 switch (lookupChange(pos).type) {
 777                         case Change::UNCHANGED:
 778                                 // reject changes in nested inset
 779                                 if (Inset * inset = getInset(pos))
 780                                                 inset->rejectChanges();
 781                                 break;
 782
 783                         case Change::INSERTED:
 784                                 // Suppress access to non-existent
 785                                 // "end-of-paragraph char"
 786                                 if (pos < size()) {
 787                                         eraseChar(pos, false);
 788                                         --end;
 789                                         --pos;
 790                                 }
 791                                 break;
 792
 793                         case Change::DELETED:
 794                                 d->changes_.set(Change(Change::UNCHANGED), pos);
 795
 796                                 // Do NOT reject changes within a deleted inset!
 797                                 // There may be insertions of a co-author inside of it!
 798
 799                                 break;
 800                 }
 801         }
 802 }
 803
 804
 805 void Paragraph::Private::insertChar(pos_type pos, char_type c,
 806                 Change const & change)
 807 {
 808         LASSERT(pos >= 0 && pos <= int(text_.size()), return);
 809
 810         // Make sure that Buffer::hasChangesPresent is updated
 811         ChangesMonitor cm(*owner_);
 812
 813         // track change
 814         changes_.insert(change, pos);
 815
 816         // This is actually very common when parsing buffers (and
 817         // maybe inserting ascii text)
 818         if (pos == pos_type(text_.size())) {
 819                 // when appending characters, no need to update tables
 820                 text_.push_back(c);
 821                 // but we want spell checking
 822                 requestSpellCheck(pos);
 823                 return;
 824         }
 825
 826         text_.insert(text_.begin() + pos, c);
 827
 828         // Update the font table.
 829         fontlist_.increasePosAfterPos(pos);
 830
 831         // Update the insets
 832         insetlist_.increasePosAfterPos(pos);
 833
 834         // Update list of misspelled positions
 835         speller_state_.increasePosAfterPos(pos);
 836
 837         // Update bookmarks
 838         if (inset_owner_ && inset_owner_->isBufferValid())
 839                 theSession().bookmarks().adjustPosAfterPos(inset_owner_->buffer().fileName(),
 840                                                        id_, pos, 1);
 841 }
 842
 843
 844 bool Paragraph::insertInset(pos_type pos, Inset * inset,
 845                                    Font const & font, Change const & change)
 846 {
 847         LASSERT(inset, return false);
 848         LASSERT(pos >= 0 && pos <= size(), return false);
 849
 850         // Make sure that Buffer::hasChangesPresent is updated
 851         ChangesMonitor cm(*this);
 852
 853         // Paragraph::insertInset() can be used in cut/copy/paste operation where
 854         // d->inset_owner_ is not set yet.
 855         if (d->inset_owner_ && !d->inset_owner_->insetAllowed(inset->lyxCode()))
 856                 return false;
 857
 858         d->insertChar(pos, META_INSET, change);
 859         LASSERT(d->text_[pos] == META_INSET, return false);
 860
 861         // Add a new entry in the insetlist_.
 862         d->insetlist_.insert(inset, pos);
 863
 864         // Some insets require run of spell checker
 865         requestSpellCheck(pos);
 866         setFont(pos, font);
 867         return true;
 868 }
 869
 870
 871 bool Paragraph::eraseChar(pos_type pos, bool trackChanges)
 872 {
 873         LASSERT(pos >= 0 && pos <= size(), return false);
 874
 875         // Make sure that Buffer::hasChangesPresent is updated
 876         ChangesMonitor cm(*this);
 877
 878         // keep the logic here in sync with the logic of isMergedOnEndOfParDeletion()
 879
 880         if (trackChanges) {
 881                 Change change = d->changes_.lookup(pos);
 882
 883                 // set the character to DELETED if
 884                 //  a) it was previously unchanged or
 885                 //  b) it was inserted by a co-author
 886
 887                 if (!change.changed() ||
 888                       (change.inserted() && !change.currentAuthor())) {
 889                         setChange(pos, Change(Change::DELETED));
 890                         // request run of spell checker
 891                         requestSpellCheck(pos);
 892                         return false;
 893                 }
 894
 895                 if (change.deleted())
 896                         return false;
 897         }
 898
 899         // Don't physically access the imaginary end-of-paragraph character.
 900         // eraseChar() can only mark it as DELETED. A physical deletion of
 901         // end-of-par must be handled externally.
 902         if (pos == size()) {
 903                 return false;
 904         }
 905
 906         // track change
 907         d->changes_.erase(pos);
 908
 909         // if it is an inset, delete the inset entry
 910         if (d->text_[pos] == META_INSET)
 911                 d->insetlist_.erase(pos);
 912
 913         d->text_.erase(d->text_.begin() + pos);
 914
 915         // Update the fontlist_
 916         d->fontlist_.erase(pos);
 917
 918         // Update the insetlist_
 919         d->insetlist_.decreasePosAfterPos(pos);
 920
 921         // Update list of misspelled positions
 922         d->speller_state_.decreasePosAfterPos(pos);
 923         d->speller_state_.refreshLast(size());
 924
 925         // Update bookmarks
 926         if (d->inset_owner_ && d->inset_owner_->isBufferValid())
 927                 theSession().bookmarks().adjustPosAfterPos(d->inset_owner_->buffer().fileName(),
 928                                                        d->id_, pos, -1);
 929
 930         return true;
 931 }
 932
 933
 934 int Paragraph::eraseChars(pos_type start, pos_type end, bool trackChanges)
 935 {
 936         LASSERT(start >= 0 && start <= size(), return 0);
 937         LASSERT(end >= start && end <= size() + 1, return 0);
 938
 939         pos_type i = start;
 940         for (pos_type count = end - start; count; --count) {
 941                 if (!eraseChar(i, trackChanges))
 942                         ++i;
 943         }
 944         return end - i;
 945 }
 946
 947
 948 // Handle combining characters
 949 int Paragraph::Private::latexSurrogatePair(BufferParams const & bparams,
 950                 otexstream & os, char_type c, char_type next,
 951                 OutputParams const & runparams) const
 952 {
 953         // Writing next here may circumvent a possible font change between
 954         // c and next. Since next is only output if it forms a surrogate pair
 955         // with c we can ignore this:
 956         // A font change inside a surrogate pair does not make sense and is
 957         // hopefully impossible to input.
 958         // FIXME: change tracking
 959         // Is this correct WRT change tracking?
 960         Encoding const & encoding = *(runparams.encoding);
 961         docstring latex1 = encoding.latexChar(next).first;
 962         if (runparams.inIPA) {
 963                 string const tipashortcut = Encodings::TIPAShortcut(next);
 964                 if (!tipashortcut.empty()) {
 965                         latex1 = from_ascii(tipashortcut);
 966                 }
 967         }
 968         docstring latex2 = encoding.latexChar(c).first;
 969
 970         if (bparams.useNonTeXFonts || docstring(1, next) == latex1) {
 971                 // Encoding supports the combination:
 972                 // output as is (combining char after base char).
 973                 os << latex2 << latex1;
 974                 return latex1.length() + latex2.length();
 975         }
 976
 977         os << latex1 << "{" << latex2 << "}";
 978         return latex1.length() + latex2.length() + 2;
 979 }
 980
 981
 982 bool Paragraph::Private::simpleTeXBlanks(BufferParams const & bparams,
 983                                        OutputParams const & runparams,
 984                                        otexstream & os,
 985                                        pos_type i,
 986                                        unsigned int & column,
 987                                        Font const & font,
 988                                        Layout const & style) const
 989 {
 990         if (style.pass_thru || runparams.pass_thru)
 991                 return false;
 992
 993         if (i + 1 < int(text_.size())) {
 994                 char_type next = text_[i + 1];
 995                 if (Encodings::isCombiningChar(next)) {
 996                         // This space has an accent, so we must always output it.
 997                         column += latexSurrogatePair(bparams, os, ' ', next, runparams) - 1;
 998                         return true;
 999                 }
1000         }
1001
1002         if (runparams.linelen > 0
1003             && column > runparams.linelen
1004             && i
1005             && text_[i - 1] != ' '
1006             && (i + 1 < int(text_.size()))
1007             // same in FreeSpacing mode
1008             && !owner_->isFreeSpacing()
1009             // In typewriter mode, we want to avoid
1010             // ! . ? : at the end of a line
1011             && !(font.fontInfo().family() == TYPEWRITER_FAMILY
1012                  && (text_[i - 1] == '.'
1013                      || text_[i - 1] == '?'
1014                      || text_[i - 1] == ':'
1015                      || text_[i - 1] == '!'))) {
1016                 os << '\n';
1017                 os.texrow().start(owner_->id(), i + 1);
1018                 column = 0;
1019         } else if (style.free_spacing) {
1020                 os << '~';
1021         } else {
1022                 os << ' ';
1023         }
1024         return false;
1025 }
1026
1027
1028 void Paragraph::Private::latexInset(BufferParams const & bparams,
1029                                     otexstream & os,
1030                                     OutputParams & runparams,
1031                                     Font & running_font,
1032                                     Font & basefont,
1033                                     Font const & outerfont,
1034                                     bool & open_font,
1035                                     Change & running_change,
1036                                     Layout const & style,
1037                                     pos_type & i,
1038                                     unsigned int & column,
1039                                     bool const fontswitch_inset,
1040                                     bool const closeLanguage,
1041                                     bool const lang_switched_at_inset) const
1042 {
1043         Inset * inset = owner_->getInset(i);
1044         LBUFERR(inset);
1045
1046         if (style.pass_thru) {
1047                 odocstringstream ods;
1048                 inset->plaintext(ods, runparams);
1049                 os << ods.str();
1050                 return;
1051         }
1052
1053         // FIXME: move this to InsetNewline::latex
1054         if (inset->lyxCode() == NEWLINE_CODE || inset->lyxCode() == SEPARATOR_CODE) {
1055                 // newlines are handled differently here than
1056                 // the default in simpleTeXSpecialChars().
1057                 if (!style.newline_allowed) {
1058                         os << '\n';
1059                 } else {
1060                         if (open_font) {
1061                                 bool needPar = false;
1062                                 column += running_font.latexWriteEndChanges(
1063                                         os, bparams, runparams,
1064                                         basefont, basefont, needPar);
1065                                 open_font = false;
1066                         }
1067
1068                         if (running_font.fontInfo().family() == TYPEWRITER_FAMILY)
1069                                 os << '~';
1070
1071                         basefont = owner_->getLayoutFont(bparams, outerfont);
1072                         running_font = basefont;
1073
1074                         if (runparams.moving_arg)
1075                                 os << "\\protect ";
1076
1077                 }
1078                 os.texrow().start(owner_->id(), i + 1);
1079                 column = 0;
1080         }
1081
1082         if (owner_->isDeleted(i)) {
1083                 if( ++runparams.inDeletedInset == 1)
1084                         runparams.changeOfDeletedInset = owner_->lookupChange(i);
1085         }
1086
1087         if (inset->canTrackChanges()) {
1088                 column += Changes::latexMarkChange(os, bparams, running_change,
1089                         Change(Change::UNCHANGED), runparams);
1090                 running_change = Change(Change::UNCHANGED);
1091         }
1092
1093         bool close = false;
1094         odocstream::pos_type const len = os.os().tellp();
1095
1096         if (inset->forceLTR(runparams)
1097             && running_font.isRightToLeft()
1098             // ERT is an exception, it should be output with no
1099             // decorations at all
1100             && inset->lyxCode() != ERT_CODE) {
1101                 if (runparams.use_polyglossia) {
1102                         os << "\\LRE{";
1103                 } else if (running_font.language()->lang() == "farsi"
1104                            || running_font.language()->lang() == "arabic_arabi")
1105                         os << "\\textLR{" << termcmd;
1106                 else
1107                         os << "\\L{";
1108                 close = true;
1109         }
1110
1111         if (open_font && fontswitch_inset) {
1112                 bool lang_closed = false;
1113                 // Close language if needed
1114                 if (closeLanguage && !lang_switched_at_inset) {
1115                         // We need prev_font here as language changes directly at inset
1116                         // will only be started inside the inset.
1117                         Font const prev_font = (i > 0) ?
1118                                                 owner_->getFont(bparams, i - 1, outerfont)
1119                                               : running_font;
1120                         Font tmpfont(basefont);
1121                         tmpfont.setLanguage(prev_font.language());
1122                         bool needPar = false;
1123                         unsigned int count = tmpfont.latexWriteEndChanges(os, bparams, runparams,
1124                                                                           basefont, basefont,
1125                                                                           needPar, closeLanguage);
1126                         column += count;
1127                         lang_closed = count > 0;
1128                 }
1129                 // Update the running_font, making sure, however,
1130                 // to leave the language as it was.
1131                 // FIXME: probably a better way to keep track of the old
1132                 // language, than copying the entire font?
1133                 Font const copy_font(running_font);
1134                 basefont = owner_->getLayoutFont(bparams, outerfont);
1135                 running_font = basefont;
1136                 if (!closeLanguage)
1137                         running_font.setLanguage(copy_font.language());
1138                 OutputParams rp = runparams;
1139                 rp.encoding = basefont.language()->encoding();
1140                 // For these, we use switches, so they should be taken as
1141                 // base inside the inset.
1142                 basefont.fontInfo().setSize(copy_font.fontInfo().size());
1143                 basefont.fontInfo().setFamily(copy_font.fontInfo().family());
1144                 basefont.fontInfo().setSeries(copy_font.fontInfo().series());
1145                 // Now re-do font changes in a way needed here
1146                 // (using switches with multi-par insets)
1147                 InsetText const * textinset = inset->asInsetText();
1148                 bool const cprotect = textinset
1149                         ? textinset->hasCProtectContent(runparams.moving_arg)
1150                           && !textinset->text().isMainText()
1151                           && inset->lyxCode() != BRANCH_CODE
1152                         : false;
1153                 unsigned int count2 = basefont.latexWriteStartChanges(os, bparams,
1154                                                       rp, running_font,
1155                                                       basefont, true,
1156                                                       cprotect);
1157                 open_font = true;
1158                 column += count2;
1159                 if (count2 == 0 && (lang_closed || lang_switched_at_inset))
1160                         // All fonts closed
1161                         open_font = false;
1162                 if (closeLanguage)
1163                         runparams.local_font = &basefont;
1164         }
1165
1166         if (fontswitch_inset && !closeLanguage) {
1167                 // The directionality has been switched at inset.
1168                 // Force markup inside.
1169                 runparams.local_font = &basefont;
1170         }
1171
1172         size_t const previous_row_count = os.texrow().rows();
1173
1174         try {
1175                 runparams.lastid = id_;
1176                 runparams.lastpos = i;
1177                 inset->latex(os, runparams);
1178         } catch (EncodingException & e) {
1179                 // add location information and throw again.
1180                 e.par_id = id_;
1181                 e.pos = i;
1182                 throw;
1183         }
1184
1185         if (close)
1186                 os << '}';
1187
1188         if (os.texrow().rows() > previous_row_count) {
1189                 os.texrow().start(owner_->id(), i + 1);
1190                 column = 0;
1191         } else {
1192                 column += (unsigned int)(os.os().tellp() - len);
1193         }
1194
1195         if (owner_->isDeleted(i))
1196                 --runparams.inDeletedInset;
1197 }
1198
1199
1200 void Paragraph::Private::latexSpecialChar(otexstream & os,
1201                                           BufferParams const & bparams,
1202                                           OutputParams const & runparams,
1203                                           Font const & running_font,
1204                                           string & alien_script,
1205                                           Layout const & style,
1206                                           pos_type & i,
1207                                           pos_type end_pos,
1208                                           unsigned int & column) const
1209 {
1210         char_type const c = owner_->getUChar(bparams, runparams, i);
1211
1212         if (style.pass_thru || runparams.pass_thru || runparams.find_effective()
1213             || contains(style.pass_thru_chars, c)
1214             || contains(runparams.pass_thru_chars, c)) {
1215                 if (runparams.find_effective()) {
1216                         switch (c) {
1217                         case '\\':
1218                                 os << "\\\\";
1219                                 return;
1220                         case '{':
1221                                 os << "\\braceleft ";
1222                                 return;
1223                         case '}':
1224                                 os << "\\braceright ";
1225                                 return;
1226                         case '$':
1227                                 os << "\\lyxdollar ";
1228                                 return;
1229                         case '~':
1230                                 os << "\\lyxtilde ";
1231                                 return;
1232                         case ' ':
1233                         case '\0':
1234                                 break;
1235                         default:
1236                                 os.put(c);
1237                                 return;
1238                         }
1239                 }
1240                 else if (c != '\0') {
1241                         Encoding const * const enc = runparams.encoding;
1242                         if (enc && !enc->encodable(c))
1243                                 throw EncodingException(c);
1244                         os.put(c);
1245                         return;
1246                 }
1247         }
1248
1249         // TIPA uses its own T3 encoding
1250         if (runparams.inIPA && latexSpecialT3(c, os, i, column))
1251                 return;
1252         // If T1 font encoding is used, use the special
1253         // characters it provides.
1254         // NOTE: Some languages reset the font encoding internally to a
1255         //       non-standard font encoding. If we are using such a language,
1256         //       we do not output special T1 chars.
1257         if (!runparams.inIPA && !running_font.language()->internalFontEncoding()
1258             && !runparams.isFullUnicode() && runparams.main_fontenc == "T1"
1259             && latexSpecialT1(c, os, i, column))
1260                 return;
1261         // NOTE: "fontspec" (non-TeX fonts) sets the font encoding to "TU" (untill 2017 "EU1" or "EU2")
1262         else if (!runparams.inIPA && !running_font.language()->internalFontEncoding()
1263                  && runparams.isFullUnicode() && latexSpecialTU(c, os, i, column))
1264                      return;
1265
1266         // Otherwise, we use what LaTeX provides us.
1267         switch (c) {
1268         case '\\':
1269                 os << "\\textbackslash" << termcmd;
1270                 column += 15;
1271                 break;
1272         case '<':
1273                 os << "\\textless" << termcmd;
1274                 column += 10;
1275                 break;
1276         case '>':
1277                 os << "\\textgreater" << termcmd;
1278                 column += 13;
1279                 break;
1280         case '|':
1281                 os << "\\textbar" << termcmd;
1282                 column += 9;
1283                 break;
1284         case '-':
1285                 os << '-';
1286                 if (i + 1 < static_cast<pos_type>(text_.size()) &&
1287                     (end_pos == -1 || i + 1 < end_pos) &&
1288                     text_[i+1] == '-') {
1289                         // Prevent "--" becoming an en dash and "---" an em dash.
1290                         // (Within \ttfamily, "---" is merged to en dash + hyphen.)
1291                         os << "{}";
1292                         column += 2;
1293                 }
1294                 break;
1295         case '\"':
1296                 os << "\\textquotedbl" << termcmd;
1297                 column += 14;
1298                 break;
1299
1300         case '$': case '&':
1301         case '%': case '#': case '{':
1302         case '}': case '_':
1303                 os << '\\';
1304                 os.put(c);
1305                 column += 1;
1306                 break;
1307
1308         case '~':
1309                 os << "\\textasciitilde" << termcmd;
1310                 column += 16;
1311                 break;
1312
1313         case '^':
1314                 os << "\\textasciicircum" << termcmd;
1315                 column += 17;
1316                 break;
1317
1318         case '*':
1319         case '[':
1320         case ']':
1321                 // avoid being mistaken for optional arguments
1322                 os << '{';
1323                 os.put(c);
1324                 os << '}';
1325                 column += 2;
1326                 break;
1327
1328         case ' ':
1329                 // Blanks are printed before font switching.
1330                 // Sure? I am not! (try nice-latex)
1331                 // I am sure it's correct. LyX might be smarter
1332                 // in the future, but for now, nothing wrong is
1333                 // written. (Asger)
1334                 break;
1335
1336         case 0x2013:
1337         case 0x2014:
1338                 // XeTeX's dash behaviour is determined via a global setting
1339                 if (bparams.use_dash_ligatures
1340                     && owner_->getFontSettings(bparams, i).fontInfo().family() != TYPEWRITER_FAMILY
1341                     && !runparams.inIPA
1342                         // TODO #10961: && not in inset Flex Code
1343                         // TODO #10961: && not in layout LyXCode
1344                     && (!bparams.useNonTeXFonts || runparams.flavor != Flavor::XeTeX)) {
1345                         if (c == 0x2013) {
1346                                 // en-dash
1347                                 os << "--";
1348                                 column +=2;
1349                         } else {
1350                                 // em-dash
1351                                 os << "---";
1352                                 column +=3;
1353                         }
1354                         break;
1355                 }
1356                 // fall through
1357         default:
1358                 if (c == '\0')
1359                         return;
1360
1361                 Encoding const & encoding = *(runparams.encoding);
1362                 char_type next = '\0';
1363                 if (i + 1 < int(text_.size())) {
1364                         next = text_[i + 1];
1365                         if (Encodings::isCombiningChar(next)) {
1366                                 column += latexSurrogatePair(bparams, os, c, next, runparams) - 1;
1367                                 ++i;
1368                                 break;
1369                         }
1370                 }
1371                 pair<docstring, bool> latex = encoding.latexChar(c);
1372                 docstring nextlatex;
1373                 bool nexttipas = false;
1374                 string nexttipashortcut;
1375                 if (next != '\0' && next != META_INSET && !encoding.encodable(next)) {
1376                         nextlatex = encoding.latexChar(next).first;
1377                         if (runparams.inIPA) {
1378                                 nexttipashortcut = Encodings::TIPAShortcut(next);
1379                                 nexttipas = !nexttipashortcut.empty();
1380                         }
1381                 }
1382                 bool tipas = false;
1383                 if (runparams.inIPA) {
1384                         string const tipashortcut = Encodings::TIPAShortcut(c);
1385                         if (!tipashortcut.empty()) {
1386                                 latex.first = from_ascii(tipashortcut);
1387                                 latex.second = false;
1388                                 tipas = true;
1389                         }
1390                 }
1391                 // eventually close "script wrapper" command (see `Paragraph::latex`)
1392                 if (!alien_script.empty()
1393                         && alien_script != Encodings::isKnownScriptChar(next)) {
1394                         column += latex.first.length();
1395                         alien_script.clear();
1396                         os << latex.first << "}";
1397                         break;
1398                 }
1399                 if (latex.second
1400                          && ((!prefixIs(nextlatex, '\\')
1401                                && !prefixIs(nextlatex, '{')
1402                                && !prefixIs(nextlatex, '}'))
1403                              || (nexttipas
1404                                  && !prefixIs(from_ascii(nexttipashortcut), '\\')))
1405                          && !tipas) {
1406                         // Prevent eating of a following space or command corruption by
1407                         // following characters
1408                         if (next == ' ' || next == '\0') {
1409                                 column += latex.first.length() + 1;
1410                                 os << latex.first << "{}";
1411                         } else {
1412                                 column += latex.first.length();
1413                                 os << latex.first << " ";
1414                         }
1415                 } else {
1416                         column += latex.first.length() - 1;
1417                         os << latex.first;
1418                 }
1419                 break;
1420         }
1421 }
1422
1423
1424 bool Paragraph::Private::latexSpecialT1(char_type const c, otexstream & os,
1425         pos_type i, unsigned int & column) const
1426 {
1427         switch (c) {
1428         case '>':
1429         case '<':
1430                 os.put(c);
1431                 // In T1 encoding, these characters exist
1432                 // but we should avoid ligatures
1433                 if (i + 1 >= int(text_.size()) || text_[i + 1] != c)
1434                         return true;
1435                 os << "\\textcompwordmark" << termcmd;
1436                 column += 19;
1437                 return true;
1438         case '|':
1439                 os.put(c);
1440                 return true;
1441         case '\"':
1442                 // soul.sty breaks with \char`\"
1443                 os << "\\textquotedbl" << termcmd;
1444                 column += 14;
1445                 return true;
1446         default:
1447                 return false;
1448         }
1449 }
1450
1451
1452 bool Paragraph::Private::latexSpecialTU(char_type const c, otexstream & os,
1453         pos_type i, unsigned int & column) const
1454 {
1455         // TU encoding is currently on par with T1.
1456         return latexSpecialT1(c, os, i, column);
1457 }
1458
1459
1460 bool Paragraph::Private::latexSpecialT3(char_type const c, otexstream & os,
1461         pos_type /*i*/, unsigned int & column) const
1462 {
1463         switch (c) {
1464         case '*':
1465         case '[':
1466         case ']':
1467         case '\"':
1468                 os.put(c);
1469                 return true;
1470         case '|':
1471                 os << "\\textvertline" << termcmd;
1472                 column += 14;
1473                 return true;
1474         default:
1475                 return false;
1476         }
1477 }
1478
1479
1480 void Paragraph::Private::validate(LaTeXFeatures & features) const
1481 {
1482         if (layout_->inpreamble && inset_owner_) {
1483                 // FIXME: Using a string stream here circumvents the encoding
1484                 // switching machinery of odocstream. Therefore the
1485                 // output is wrong if this paragraph contains content
1486                 // that needs to switch encoding.
1487                 Buffer const & buf = inset_owner_->buffer();
1488                 otexstringstream os;
1489                 os << layout_->preamble();
1490                 size_t const length = os.length();
1491                 TeXOnePar(buf, *inset_owner_->getText(int(buf.getParFromID(owner_->id()).idx())),
1492                           buf.getParFromID(owner_->id()).pit(), os,
1493                           features.runparams(), string(), 0, -1, true);
1494                 if (os.length() > length)
1495                         features.addPreambleSnippet(os.release(), true);
1496         }
1497
1498         if (features.runparams().flavor == Flavor::Html
1499             && layout_->htmltitle()) {
1500                 features.setHTMLTitle(owner_->asString(AS_STR_INSETS | AS_STR_SKIPDELETE));
1501         }
1502
1503         // check the params.
1504         if (!params_.spacing().isDefault())
1505                 features.require("setspace");
1506
1507         // then the layouts
1508         features.useLayout(layout_->name());
1509
1510         // then the fonts
1511         fontlist_.validate(features);
1512
1513         // then the indentation
1514         if (!params_.leftIndent().zero())
1515                 features.require("ParagraphLeftIndent");
1516
1517         // then the insets
1518         InsetList::const_iterator icit = insetlist_.begin();
1519         InsetList::const_iterator iend = insetlist_.end();
1520         for (; icit != iend; ++icit) {
1521                 if (icit->inset) {
1522                         features.inDeletedInset(owner_->isDeleted(icit->pos));
1523                         if (icit->inset->lyxCode() == FOOT_CODE) {
1524                                 // FIXME: an item inset would make things much easier.
1525                                 if ((layout_->latextype == LATEX_LIST_ENVIRONMENT
1526                                      || (layout_->latextype == LATEX_ITEM_ENVIRONMENT
1527                                          && layout_->margintype == MARGIN_FIRST_DYNAMIC))
1528                                     && (icit->pos < begin_of_body_
1529                                         || (icit->pos == begin_of_body_
1530                                             && (icit->pos == 0 || text_[icit->pos - 1] != ' '))))
1531                                         features.saveNoteEnv("description");
1532                         }
1533                         icit->inset->validate(features);
1534                         features.inDeletedInset(false);
1535                         if (layout_->needprotect &&
1536                             icit->inset->lyxCode() == FOOT_CODE)
1537                                 features.require("NeedLyXFootnoteCode");
1538                 }
1539         }
1540
1541         // then the contents
1542         BufferParams const bp = features.runparams().is_child
1543                 ? features.buffer().masterParams() : features.buffer().params();
1544         for (pos_type i = 0; i < int(text_.size()) ; ++i) {
1545                 char_type c = text_[i];
1546                 CharInfo const & ci = Encodings::unicodeCharInfo(c);
1547                 if (c == 0x0022) {
1548                         if (features.runparams().isFullUnicode() && bp.useNonTeXFonts)
1549                                 features.require("textquotedblp");
1550                         else if (features.runparams().main_fontenc != "T1"
1551                                  || ((&owner_->getFontSettings(bp, i))->language()->internalFontEncoding()))
1552                                 features.require("textquotedbl");
1553                 } else if (ci.textFeature() && contains(ci.textPreamble(), '=')) {
1554                         // features that depend on the font or input encoding
1555                         string feats = ci.textPreamble();
1556                         string fontenc = (&owner_->getFontSettings(bp, i))->language()->fontenc(bp);
1557                         if (fontenc.empty())
1558                                 fontenc = features.runparams().main_fontenc;
1559                         while (!feats.empty()) {
1560                                 string feat;
1561                                 feats = split(feats, feat, ',');
1562                                 if (contains(feat, "!=")) {
1563                                         // a feature that is required except for the spcified
1564                                         // font or input encodings
1565                                         string realfeature;
1566                                         string const contexts = ltrim(split(feat, realfeature, '!'), "=");
1567                                         // multiple encodings are separated by semicolon
1568                                         vector<string> context = getVectorFromString(contexts, ";");
1569                                         // require feature if the context matches neither current font
1570                                         // nor input encoding
1571                                         if (std::find(context.begin(), context.end(), fontenc) == context.end()
1572                                             && std::find(context.begin(), context.end(),
1573                                                          features.runparams().encoding->name()) == context.end())
1574                                                 features.require(realfeature);
1575                                 } else if (contains(feat, '=')) {
1576                                         // a feature that is required only for the spcified
1577                                         // font or input encodings
1578                                         string realfeature;
1579                                         string const contexts = split(feat, realfeature, '=');
1580                                         // multiple encodings are separated by semicolon
1581                                         vector<string> context = getVectorFromString(contexts, ";");
1582                                         // require feature if the context matches either current font
1583                                         // or input encoding
1584                                         if (std::find(context.begin(), context.end(), fontenc) != context.end()
1585                                             || std::find(context.begin(), context.end(),
1586                                                          features.runparams().encoding->name()) != context.end())
1587                                                 features.require(realfeature);
1588                                 }
1589                         }
1590                 } else if (!bp.use_dash_ligatures
1591                            && (c == 0x2013 || c == 0x2014)
1592                            && bp.useNonTeXFonts
1593                            && features.runparams().flavor == Flavor::XeTeX)
1594                         // XeTeX's dash behaviour is determined via a global setting
1595                         features.require("xetexdashbreakstate");
1596                 BufferEncodings::validate(c, features);
1597         }
1598 }
1599
1600 /////////////////////////////////////////////////////////////////////
1601 //
1602 // Paragraph
1603 //
1604 /////////////////////////////////////////////////////////////////////
1605
1606 namespace {
1607         Layout const emptyParagraphLayout;
1608 }
1609
1610 Paragraph::Paragraph()
1611         : d(new Paragraph::Private(this, emptyParagraphLayout))
1612 {
1613         itemdepth = 0;
1614         d->params_.clear();
1615 }
1616
1617
1618 Paragraph::Paragraph(Paragraph const & par)
1619         : itemdepth(par.itemdepth),
1620         d(new Paragraph::Private(*par.d, this))
1621 {
1622         registerWords();
1623 }
1624
1625
1626 Paragraph::Paragraph(Paragraph const & par, pos_type beg, pos_type end)
1627         : itemdepth(par.itemdepth),
1628         d(new Paragraph::Private(*par.d, this, beg, end))
1629 {
1630         registerWords();
1631 }
1632
1633
1634 Paragraph & Paragraph::operator=(Paragraph const & par)
1635 {
1636         // needed as we will destroy the private part before copying it
1637         if (&par != this) {
1638                 itemdepth = par.itemdepth;
1639
1640                 deregisterWords();
1641                 delete d;
1642                 d = new Private(*par.d, this);
1643                 registerWords();
1644         }
1645         return *this;
1646 }
1647
1648
1649 Paragraph::~Paragraph()
1650 {
1651         deregisterWords();
1652         delete d;
1653 }
1654
1655
1656 namespace {
1657
1658 // this shall be called just before every "os << ..." action.
1659 void flushString(ostream & os, docstring & s)
1660 {
1661         os << to_utf8(s);
1662         s.erase();
1663 }
1664
1665 } // namespace
1666
1667
1668 void Paragraph::write(ostream & os, BufferParams const & bparams,
1669         depth_type & depth) const
1670 {
1671         // The beginning or end of a deeper (i.e. nested) area?
1672         if (depth != d->params_.depth()) {
1673                 if (d->params_.depth() > depth) {
1674                         while (d->params_.depth() > depth) {
1675                                 os << "\n\\begin_deeper";
1676                                 ++depth;
1677                         }
1678                 } else {
1679                         while (d->params_.depth() < depth) {
1680                                 os << "\n\\end_deeper";
1681                                 --depth;
1682                         }
1683                 }
1684         }
1685
1686         // First write the layout
1687         os << "\n\\begin_layout " << to_utf8(d->layout_->name()) << '\n';
1688
1689         d->params_.write(os);
1690
1691         Font font1(inherit_font, bparams.language);
1692
1693         Change running_change = Change(Change::UNCHANGED);
1694
1695         // this string is used as a buffer to avoid repetitive calls
1696         // to to_utf8(), which turn out to be expensive (JMarc)
1697         docstring write_buffer;
1698
1699         int column = 0;
1700         for (pos_type i = 0; i <= size(); ++i) {
1701
1702                 Change const & change = lookupChange(i);
1703                 if (change != running_change)
1704                         flushString(os, write_buffer);
1705                 Changes::lyxMarkChange(os, bparams, column, running_change, change);
1706                 running_change = change;
1707
1708                 if (i == size())
1709                         break;
1710
1711                 // Write font changes
1712                 Font font2 = getFontSettings(bparams, i);
1713                 if (font2 != font1) {
1714                         flushString(os, write_buffer);
1715                         font2.lyxWriteChanges(font1, os);
1716                         column = 0;
1717                         font1 = font2;
1718                 }
1719
1720                 char_type const c = d->text_[i];
1721                 switch (c) {
1722                 case META_INSET:
1723                         if (Inset const * inset = getInset(i)) {
1724                                 flushString(os, write_buffer);
1725                                 if (inset->directWrite()) {
1726                                         // international char, let it write
1727                                         // code directly so it's shorter in
1728                                         // the file
1729                                         inset->write(os);
1730                                 } else {
1731                                         if (i)
1732                                                 os << '\n';
1733                                         os << "\\begin_inset ";
1734                                         inset->write(os);
1735                                         os << "\n\\end_inset\n\n";
1736                                         column = 0;
1737                                 }
1738                                 // FIXME This can be removed again once the mystery
1739                                 // crash has been resolved.
1740                                 os << flush;
1741                         }
1742                         break;
1743                 case '\\':
1744                         flushString(os, write_buffer);
1745                         os << "\n\\backslash\n";
1746                         column = 0;
1747                         break;
1748                 case '.':
1749                         flushString(os, write_buffer);
1750                         if (i + 1 < size() && d->text_[i + 1] == ' ') {
1751                                 os << ".\n";
1752                                 column = 0;
1753                         } else
1754                                 os << '.';
1755                         break;
1756                 default:
1757                         if ((column > 70 && c == ' ')
1758                             || column > 79) {
1759                                 flushString(os, write_buffer);
1760                                 os << '\n';
1761                                 column = 0;
1762                         }
1763                         // this check is to amend a bug. LyX sometimes
1764                         // inserts '\0' this could cause problems.
1765                         if (c != '\0')
1766                                 write_buffer.push_back(c);
1767                         else
1768                                 LYXERR0("NUL char in structure.");
1769                         ++column;
1770                         break;
1771                 }
1772         }
1773
1774         flushString(os, write_buffer);
1775         os << "\n\\end_layout\n";
1776         // FIXME This can be removed again once the mystery
1777         // crash has been resolved.
1778         os << flush;
1779 }
1780
1781
1782 void Paragraph::validate(LaTeXFeatures & features) const
1783 {
1784         d->validate(features);
1785         bool fragile = features.runparams().moving_arg;
1786         fragile |= layout().needprotect;
1787         if (inInset().getLayout().isNeedProtect())
1788                 fragile = true;
1789         if (needsCProtection(fragile))
1790                 features.require("cprotect");
1791 }
1792
1793
1794 void Paragraph::insert(pos_type pos, docstring const & str,
1795                        Font const & font, Change const & change)
1796 {
1797         for (size_t i = 0, n = str.size(); i != n ; ++i)
1798                 insertChar(pos + i, str[i], font, change);
1799 }
1800
1801
1802 void Paragraph::appendChar(char_type c, Font const & font,
1803                 Change const & change)
1804 {
1805         // Make sure that Buffer::hasChangesPresent is updated
1806         ChangesMonitor cm(*this);
1807
1808         // track change
1809         d->changes_.insert(change, d->text_.size());
1810         // when appending characters, no need to update tables
1811         d->text_.push_back(c);
1812         setFont(d->text_.size() - 1, font);
1813         d->requestSpellCheck(d->text_.size() - 1);
1814 }
1815
1816
1817 void Paragraph::appendString(docstring const & s, Font const & font,
1818                 Change const & change)
1819 {
1820         // Make sure that Buffer::hasChangesPresent is updated
1821         ChangesMonitor cm(*this);
1822
1823         pos_type end = s.size();
1824         size_t oldsize = d->text_.size();
1825         size_t newsize = oldsize + end;
1826         size_t capacity = d->text_.capacity();
1827         if (newsize >= capacity)
1828                 d->text_.reserve(max(capacity + 100, newsize));
1829
1830         // when appending characters, no need to update tables
1831         d->text_.append(s);
1832
1833         // FIXME: Optimize this!
1834         for (size_t i = oldsize; i != newsize; ++i) {
1835                 // track change
1836                 d->changes_.insert(change, i);
1837                 d->requestSpellCheck(i);
1838         }
1839         d->fontlist_.set(oldsize, font);
1840         d->fontlist_.set(newsize - 1, font);
1841 }
1842
1843
1844 void Paragraph::insertChar(pos_type pos, char_type c,
1845                            bool trackChanges)
1846 {
1847         d->insertChar(pos, c, Change(trackChanges ?
1848                            Change::INSERTED : Change::UNCHANGED));
1849 }
1850
1851
1852 void Paragraph::insertChar(pos_type pos, char_type c,
1853                            Font const & font, bool trackChanges)
1854 {
1855         d->insertChar(pos, c, Change(trackChanges ?
1856                            Change::INSERTED : Change::UNCHANGED));
1857         setFont(pos, font);
1858 }
1859
1860
1861 void Paragraph::insertChar(pos_type pos, char_type c,
1862                            Font const & font, Change const & change)
1863 {
1864         d->insertChar(pos, c, change);
1865         setFont(pos, font);
1866 }
1867
1868
1869 void Paragraph::resetFonts(Font const & font)
1870 {
1871         d->fontlist_.clear();
1872         d->fontlist_.set(0, font);
1873         d->fontlist_.set(d->text_.size() - 1, font);
1874 }
1875
1876 // Gets uninstantiated font setting at position.
1877 Font const & Paragraph::getFontSettings(BufferParams const & bparams,
1878                                          pos_type pos) const
1879 {
1880         if (pos > size()) {
1881                 LYXERR0("pos: " << pos << " size: " << size());
1882                 LBUFERR(false);
1883         }
1884
1885         FontList::const_iterator cit = d->fontlist_.fontIterator(pos);
1886         if (cit != d->fontlist_.end())
1887                 return cit->font();
1888
1889         if (pos == size() && !empty())
1890                 return getFontSettings(bparams, pos - 1);
1891
1892         // Optimisation: avoid a full font instantiation if there is no
1893         // language change from previous call.
1894         static Font previous_font;
1895         static Language const * previous_lang = nullptr;
1896         Language const * lang = getParLanguage(bparams);
1897         if (lang != previous_lang) {
1898                 previous_lang = lang;
1899                 previous_font = Font(inherit_font, lang);
1900         }
1901         return previous_font;
1902 }
1903
1904
1905 FontSpan Paragraph::fontSpan(pos_type pos) const
1906 {
1907         LBUFERR(pos <= size());
1908
1909         if (pos == size())
1910                 return FontSpan(pos, pos);
1911
1912         pos_type start = 0;
1913         FontList::const_iterator cit = d->fontlist_.begin();
1914         FontList::const_iterator end = d->fontlist_.end();
1915         for (; cit != end; ++cit) {
1916                 if (cit->pos() >= pos) {
1917                         if (pos >= beginOfBody())
1918                                 return FontSpan(max(start, beginOfBody()),
1919                                                 cit->pos());
1920                         else
1921                                 return FontSpan(start,
1922                                                 min(beginOfBody() - 1,
1923                                                          cit->pos()));
1924                 }
1925                 start = cit->pos() + 1;
1926         }
1927
1928         // This should not happen, but if so, we take no chances.
1929         LYXERR0("Paragraph::fontSpan: position not found in fontinfo table!");
1930         LASSERT(false, return FontSpan(pos, pos));
1931 }
1932
1933
1934 // Gets uninstantiated font setting at position 0
1935 Font const & Paragraph::getFirstFontSettings(BufferParams const & bparams) const
1936 {
1937         if (!empty() && !d->fontlist_.empty())
1938                 return d->fontlist_.begin()->font();
1939
1940         // Optimisation: avoid a full font instantiation if there is no
1941         // language change from previous call.
1942         static Font previous_font;
1943         static Language const * previous_lang = nullptr;
1944         if (bparams.language != previous_lang) {
1945                 previous_lang = bparams.language;
1946                 previous_font = Font(inherit_font, bparams.language);
1947         }
1948
1949         return previous_font;
1950 }
1951
1952
1953 // Gets the fully instantiated font at a given position in a paragraph
1954 // This is basically the same function as Text::GetFont() in text2.cpp.
1955 // The difference is that this one is used for generating the LaTeX file,
1956 // and thus cosmetic "improvements" are disallowed: This has to deliver
1957 // the true picture of the buffer. (Asger)
1958 Font const Paragraph::getFont(BufferParams const & bparams, pos_type pos,
1959                                  Font const & outerfont) const
1960 {
1961         LBUFERR(pos >= 0);
1962
1963         Font font = getFontSettings(bparams, pos);
1964
1965         pos_type const body_pos = beginOfBody();
1966         FontInfo & fi = font.fontInfo();
1967         if (pos < body_pos)
1968                 fi.realize(d->layout_->labelfont);
1969         else
1970                 fi.realize(d->layout_->font);
1971
1972         fi.realize(outerfont.fontInfo());
1973         fi.realize(bparams.getFont().fontInfo());
1974
1975         return font;
1976 }
1977
1978
1979 Font const Paragraph::getLabelFont
1980         (BufferParams const & bparams, Font const & outerfont) const
1981 {
1982         FontInfo tmpfont = d->layout_->labelfont;
1983         tmpfont.realize(outerfont.fontInfo());
1984         tmpfont.realize(bparams.getFont().fontInfo());
1985         return Font(tmpfont, getParLanguage(bparams));
1986 }
1987
1988
1989 Font const Paragraph::getLayoutFont
1990         (BufferParams const & bparams, Font const & outerfont) const
1991 {
1992         FontInfo tmpfont = d->layout_->font;
1993         tmpfont.realize(outerfont.fontInfo());
1994         tmpfont.realize(bparams.getFont().fontInfo());
1995         return Font(tmpfont, getParLanguage(bparams));
1996 }
1997
1998
1999 char_type Paragraph::getUChar(BufferParams const & bparams,
2000                               OutputParams const & rp,
2001                               pos_type pos) const
2002 {
2003         char_type c = d->text_[pos];
2004
2005         // Return unchanged character in LTR languages
2006         // or if we use poylglossia/bidi (XeTeX).
2007         if (rp.useBidiPackage()
2008             || !getFontSettings(bparams, pos).isRightToLeft())
2009                 return c;
2010
2011         // Without polyglossia/bidi, we need to account for some special cases.
2012         // FIXME This needs to be audited!
2013         // Check if:
2014         // * The input is as expected for all delimiters
2015         //   => checked for Hebrew!
2016         // * The output matches the display in the LyX workarea
2017         //   => checked for Hebrew!
2018         // * The special cases below are really necessary
2019         //   => checked for Hebrew!
2020         // * In arabic_arabi, brackets are transformed to Arabic
2021         //   Ornate Parentheses. Is this is really wanted?
2022
2023         string const & lang = getFontSettings(bparams, pos).language()->lang();
2024         char_type uc = c;
2025
2026         // 1. In the following languages, parentheses need to be reversed.
2027         //    Also with polyglodia/luabidi
2028         bool const reverseparens = (lang == "hebrew" || rp.use_polyglossia);
2029
2030         // 2. In the following languages, brackets don't need to be reversed.
2031         bool const reversebrackets = lang != "arabic_arabtex"
2032                         && lang != "arabic_arabi"
2033                         && lang != "farsi";
2034
2035         // Now swap delimiters if needed.
2036         switch (c) {
2037         case '(':
2038                 if (reverseparens)
2039                         uc = ')';
2040                 break;
2041         case ')':
2042                 if (reverseparens)
2043                         uc = '(';
2044                 break;
2045         case '[':
2046                 if (reversebrackets)
2047                         uc = ']';
2048                 break;
2049         case ']':
2050                 if (reversebrackets)
2051                         uc = '[';
2052                 break;
2053         case '{':
2054                 uc = '}';
2055                 break;
2056         case '}':
2057                 uc = '{';
2058                 break;
2059         case '<':
2060                 uc = '>';
2061                 break;
2062         case '>':
2063                 uc = '<';
2064                 break;
2065         }
2066
2067         return uc;
2068 }
2069
2070
2071 void Paragraph::setFont(pos_type pos, Font const & font)
2072 {
2073         LASSERT(pos <= size(), return);
2074
2075         // First, reduce font against layout/label font
2076         // Update: The setCharFont() routine in text2.cpp already
2077         // reduces font, so we don't need to do that here. (Asger)
2078
2079         d->fontlist_.set(pos, font);
2080 }
2081
2082
2083 void Paragraph::makeSameLayout(Paragraph const & par)
2084 {
2085         d->layout_ = par.d->layout_;
2086         d->params_ = par.d->params_;
2087 }
2088
2089
2090 bool Paragraph::stripLeadingSpaces(bool trackChanges)
2091 {
2092         if (isFreeSpacing())
2093                 return false;
2094
2095         int pos = 0;
2096         int count = 0;
2097
2098         while (pos < size() && (isNewline(pos) || isLineSeparator(pos))) {
2099                 if (eraseChar(pos, trackChanges))
2100                         ++count;
2101                 else
2102                         ++pos;
2103         }
2104
2105         return count > 0 || pos > 0;
2106 }
2107
2108
2109 bool Paragraph::hasSameLayout(Paragraph const & par) const
2110 {
2111         return par.d->layout_ == d->layout_
2112                 && d->params_.sameLayout(par.d->params_);
2113 }
2114
2115
2116 depth_type Paragraph::getDepth() const
2117 {
2118         return d->params_.depth();
2119 }
2120
2121
2122 depth_type Paragraph::getMaxDepthAfter() const
2123 {
2124         if (d->layout_->isEnvironment())
2125                 return d->params_.depth() + 1;
2126         else
2127                 return d->params_.depth();
2128 }
2129
2130
2131 LyXAlignment Paragraph::getAlign(BufferParams const & bparams) const
2132 {
2133         if (d->params_.align() == LYX_ALIGN_LAYOUT)
2134                 return getDefaultAlign(bparams);
2135         else
2136                 return d->params_.align();
2137 }
2138
2139
2140 LyXAlignment Paragraph::getDefaultAlign(BufferParams const & bparams) const
2141 {
2142         LyXAlignment res = layout().align;
2143         if (isRTL(bparams)) {
2144                 // Swap sides
2145                 if (res == LYX_ALIGN_LEFT)
2146                         res = LYX_ALIGN_RIGHT;
2147                 else if  (res == LYX_ALIGN_RIGHT)
2148                         res = LYX_ALIGN_LEFT;
2149         }
2150         return res;
2151 }
2152
2153
2154 docstring const & Paragraph::labelString() const
2155 {
2156         return d->params_.labelString();
2157 }
2158
2159
2160 // the next two functions are for the manual labels
2161 docstring const Paragraph::getLabelWidthString() const
2162 {
2163         if (d->layout_->margintype == MARGIN_MANUAL
2164             || d->layout_->latextype == LATEX_BIB_ENVIRONMENT)
2165                 return d->params_.labelWidthString();
2166         else
2167                 return _("Senseless with this layout!");
2168 }
2169
2170
2171 void Paragraph::setLabelWidthString(docstring const & s)
2172 {
2173         d->params_.labelWidthString(s);
2174 }
2175
2176
2177 docstring Paragraph::expandLabel(Layout const & layout,
2178                 BufferParams const & bparams) const
2179 {
2180         return expandParagraphLabel(layout, bparams, true);
2181 }
2182
2183
2184 docstring Paragraph::expandParagraphLabel(Layout const & layout,
2185                 BufferParams const & bparams, bool process_appendix) const
2186 {
2187         DocumentClass const & tclass = bparams.documentClass();
2188         string const & lang = getParLanguage(bparams)->code();
2189         bool const in_appendix = process_appendix && d->params_.appendix();
2190         docstring fmt = translateIfPossible(layout.labelstring(in_appendix), lang);
2191
2192         if (fmt.empty() && !layout.counter.empty())
2193                 return tclass.counters().theCounter(layout.counter, lang);
2194
2195         // handle 'inherited level parts' in 'fmt',
2196         // i.e. the stuff between '@' in   '@Section@.\arabic{subsection}'
2197         size_t const i = fmt.find('@', 0);
2198         if (i != docstring::npos) {
2199                 size_t const j = fmt.find('@', i + 1);
2200                 if (j != docstring::npos) {
2201                         docstring parent(fmt, i + 1, j - i - 1);
2202                         docstring label = from_ascii("??");
2203                         if (tclass.hasLayout(parent))
2204                                 label = expandParagraphLabel(tclass[parent], bparams,
2205                                                       process_appendix);
2206                         fmt = docstring(fmt, 0, i) + label
2207                                 + docstring(fmt, j + 1, docstring::npos);
2208                 }
2209         }
2210
2211         return tclass.counters().counterLabel(fmt, lang);
2212 }
2213
2214
2215 void Paragraph::applyLayout(Layout const & new_layout)
2216 {
2217         d->layout_ = &new_layout;
2218         LyXAlignment const oldAlign = d->params_.align();
2219
2220         if (!(oldAlign & d->layout_->alignpossible)) {
2221                 frontend::Alert::warning(_("Alignment not permitted"),
2222                         _("The new layout does not permit the alignment previously used.\nSetting to default."));
2223                 d->params_.align(LYX_ALIGN_LAYOUT);
2224         }
2225 }
2226
2227
2228 pos_type Paragraph::beginOfBody() const
2229 {
2230         return d->begin_of_body_;
2231 }
2232
2233
2234 void Paragraph::setBeginOfBody()
2235 {
2236         if (d->layout_->labeltype != LABEL_MANUAL) {
2237                 d->begin_of_body_ = 0;
2238                 return;
2239         }
2240
2241         // Unroll the first two cycles of the loop
2242         // and remember the previous character to
2243         // remove unnecessary getChar() calls
2244         pos_type i = 0;
2245         pos_type end = size();
2246         bool prev_char_deleted = false;
2247         if (i < end && (!(isNewline(i) || isEnvSeparator(i)) || isDeleted(i))) {
2248                 ++i;
2249                 if (i < end) {
2250                         char_type previous_char = d->text_[i];
2251                         if (!(isNewline(i) || isEnvSeparator(i))) {
2252                                 ++i;
2253                                 while (i < end && (previous_char != ' ' || prev_char_deleted)) {
2254                                         char_type temp = d->text_[i];
2255                                         prev_char_deleted = isDeleted(i);
2256                                         if (!isDeleted(i) && (isNewline(i) || isEnvSeparator(i)))
2257                                                 break;
2258                                         ++i;
2259                                         previous_char = temp;
2260                                 }
2261                         }
2262                 }
2263         }
2264
2265         d->begin_of_body_ = i;
2266 }
2267
2268
2269 bool Paragraph::allowParagraphCustomization() const
2270 {
2271         return inInset().allowParagraphCustomization();
2272 }
2273
2274
2275 bool Paragraph::usePlainLayout() const
2276 {
2277         return inInset().usePlainLayout();
2278 }
2279
2280
2281 bool Paragraph::isPassThru() const
2282 {
2283         return inInset().isPassThru() || d->layout_->pass_thru;
2284 }
2285
2286
2287 bool Paragraph::parbreakIsNewline() const
2288 {
2289         return inInset().getLayout().parbreakIsNewline() || d->layout_->parbreak_is_newline;
2290 }
2291
2292
2293 bool Paragraph::isPartOfTextSequence() const
2294 {
2295         for (pos_type i = 0; i < size(); ++i) {
2296                 if (!isInset(i) || getInset(i)->isPartOfTextSequence())
2297                         return true;
2298         }
2299         return false;
2300 }
2301
2302 namespace {
2303
2304 // paragraphs inside floats need different alignment tags to avoid
2305 // unwanted space
2306
2307 bool noTrivlistCentering(InsetCode code)
2308 {
2309         return code == FLOAT_CODE
2310                || code == WRAP_CODE
2311                || code == CELL_CODE;
2312 }
2313
2314
2315 string correction(string const & orig)
2316 {
2317         if (orig == "flushleft")
2318                 return "raggedright";
2319         if (orig == "flushright")
2320                 return "raggedleft";
2321         if (orig == "center")
2322                 return "centering";
2323         return orig;
2324 }
2325
2326
2327 bool corrected_env(otexstream & os, string const & suffix, string const & env,
2328         InsetCode code, bool const lastpar, int & col)
2329 {
2330         string macro = suffix + "{";
2331         if (noTrivlistCentering(code)) {
2332                 if (lastpar) {
2333                         // the last paragraph in non-trivlist-aligned
2334                         // context is special (to avoid unwanted whitespace)
2335                         if (suffix == "\\begin") {
2336                                 macro = "\\" + correction(env) + "{}";
2337                                 os << from_ascii(macro);
2338                                 col += macro.size();
2339                                 return true;
2340                         }
2341                         return false;
2342                 }
2343                 macro += correction(env);
2344         } else
2345                 macro += env;
2346         macro += "}";
2347         if (suffix == "\\par\\end") {
2348                 os << breakln;
2349                 col = 0;
2350         }
2351         os << from_ascii(macro);
2352         col += macro.size();
2353         if (suffix == "\\begin") {
2354                 os << breakln;
2355                 col = 0;
2356         }
2357         return true;
2358 }
2359
2360 } // namespace
2361
2362
2363 int Paragraph::Private::startTeXParParams(BufferParams const & bparams,
2364                         otexstream & os, OutputParams const & runparams) const
2365 {
2366         int column = 0;
2367
2368         bool canindent =
2369                 (bparams.paragraph_separation == BufferParams::ParagraphIndentSeparation) ?
2370                         (layout_->toggle_indent != ITOGGLE_NEVER) :
2371                         (layout_->toggle_indent == ITOGGLE_ALWAYS);
2372
2373         LyXAlignment const curAlign = params_.align();
2374
2375         // Do not output \\noindent for paragraphs
2376         // 1. that cannot have indentation or are indented always,
2377         // 2. that are not part of the immediate text sequence (e.g., contain only floats),
2378         // 3. that are PassThru,
2379         // 4. or that are centered.
2380         if (canindent && params_.noindent()
2381             && owner_->isPartOfTextSequence()
2382             && !layout_->pass_thru
2383             && curAlign != LYX_ALIGN_CENTER) {
2384                 if (!owner_->empty()
2385                     && owner_->getInset(0)
2386                     && owner_->getInset(0)->lyxCode() == VSPACE_CODE)
2387                         // If the paragraph starts with a vspace, the \\noindent
2388                         // needs to come after that (as it leaves vmode).
2389                         // If the paragraph consists only of the vspace,
2390                         // \\noindent is not needed at all.
2391                         runparams.need_noindent = owner_->size() > 1;
2392                 else {
2393                         os << "\\noindent" << termcmd;
2394                         column += 10;
2395                 }
2396         }
2397
2398         if (curAlign == layout_->align)
2399                 return column;
2400
2401         switch (curAlign) {
2402         case LYX_ALIGN_NONE:
2403         case LYX_ALIGN_BLOCK:
2404         case LYX_ALIGN_LAYOUT:
2405         case LYX_ALIGN_SPECIAL:
2406         case LYX_ALIGN_DECIMAL:
2407                 break;
2408         case LYX_ALIGN_LEFT:
2409         case LYX_ALIGN_RIGHT:
2410         case LYX_ALIGN_CENTER:
2411                 if (runparams.moving_arg) {
2412                         os << "\\protect";
2413                         column += 8;
2414                 }
2415                 break;
2416         }
2417
2418         string const begin_tag = "\\begin";
2419         InsetCode code = ownerCode();
2420         bool const lastpar = runparams.isLastPar;
2421         // RTL in classic (PDF)LaTeX (without the Bidi package)
2422         // Luabibdi (used by LuaTeX) behaves like classic
2423         bool const rtl_classic = owner_->getParLanguage(bparams)->rightToLeft()
2424                 && !runparams.useBidiPackage();
2425
2426         switch (curAlign) {
2427         case LYX_ALIGN_NONE:
2428         case LYX_ALIGN_BLOCK:
2429         case LYX_ALIGN_LAYOUT:
2430         case LYX_ALIGN_SPECIAL:
2431         case LYX_ALIGN_DECIMAL:
2432                 break;
2433         case LYX_ALIGN_LEFT: {
2434                 if (rtl_classic)
2435                         // Classic (PDF)LaTeX switches the left/right logic in RTL mode
2436                         corrected_env(os, begin_tag, "flushright", code, lastpar, column);
2437                 else
2438                         corrected_env(os, begin_tag, "flushleft", code, lastpar, column);
2439                 break;
2440         } case LYX_ALIGN_RIGHT: {
2441                 if (rtl_classic)
2442                         // Classic (PDF)LaTeX switches the left/right logic in RTL mode
2443                         corrected_env(os, begin_tag, "flushleft", code, lastpar, column);
2444                 else
2445                         corrected_env(os, begin_tag, "flushright", code, lastpar, column);
2446                 break;
2447         } case LYX_ALIGN_CENTER: {
2448                 corrected_env(os, begin_tag, "center", code, lastpar, column);
2449                 break;
2450         }
2451         }
2452
2453         return column;
2454 }
2455
2456
2457 bool Paragraph::Private::endTeXParParams(BufferParams const & bparams,
2458                         otexstream & os, OutputParams const & runparams) const
2459 {
2460         LyXAlignment const curAlign = params_.align();
2461
2462         if (curAlign == layout_->align)
2463                 return false;
2464
2465         switch (curAlign) {
2466         case LYX_ALIGN_NONE:
2467         case LYX_ALIGN_BLOCK:
2468         case LYX_ALIGN_LAYOUT:
2469         case LYX_ALIGN_SPECIAL:
2470         case LYX_ALIGN_DECIMAL:
2471                 break;
2472         case LYX_ALIGN_LEFT:
2473         case LYX_ALIGN_RIGHT:
2474         case LYX_ALIGN_CENTER:
2475                 if (runparams.moving_arg)
2476                         os << "\\protect";
2477                 break;
2478         }
2479
2480         bool output = false;
2481         int col = 0;
2482         string const end_tag = "\\par\\end";
2483         InsetCode code = ownerCode();
2484         bool const lastpar = runparams.isLastPar;
2485         // RTL in classic (PDF)LaTeX (without the Bidi package)
2486         // Luabibdi (used by LuaTeX) behaves like classic
2487         bool const rtl_classic = owner_->getParLanguage(bparams)->rightToLeft()
2488                 && !runparams.useBidiPackage();
2489
2490         switch (curAlign) {
2491         case LYX_ALIGN_NONE:
2492         case LYX_ALIGN_BLOCK:
2493         case LYX_ALIGN_LAYOUT:
2494         case LYX_ALIGN_SPECIAL:
2495         case LYX_ALIGN_DECIMAL:
2496                 break;
2497         case LYX_ALIGN_LEFT: {
2498                 if (rtl_classic)
2499                         // Classic (PDF)LaTeX switches the left/right logic in RTL mode
2500                         output = corrected_env(os, end_tag, "flushright", code, lastpar, col);
2501                 else
2502                         output = corrected_env(os, end_tag, "flushleft", code, lastpar, col);
2503                 break;
2504         } case LYX_ALIGN_RIGHT: {
2505                 if (rtl_classic)
2506                         // Classic (PDF)LaTeX switches the left/right logic in RTL mode
2507                         output = corrected_env(os, end_tag, "flushleft", code, lastpar, col);
2508                 else
2509                         output = corrected_env(os, end_tag, "flushright", code, lastpar, col);
2510                 break;
2511         } case LYX_ALIGN_CENTER: {
2512                 corrected_env(os, end_tag, "center", code, lastpar, col);
2513                 break;
2514         }
2515         }
2516
2517         return output || lastpar;
2518 }
2519
2520
2521 // This one spits out the text of the paragraph
2522 void Paragraph::latex(BufferParams const & bparams,
2523         Font const & outerfont,
2524         otexstream & os,
2525         OutputParams const & runparams,
2526         int start_pos, int end_pos, bool force) const
2527 {
2528         LYXERR(Debug::OUTFILE, "Paragraph::latex...     " << this);
2529
2530         // FIXME This check should not be needed. Perhaps issue an
2531         // error if it triggers.
2532         Layout const & style = inInset().forcePlainLayout() ?
2533                 bparams.documentClass().plainLayout() : *d->layout_;
2534
2535         if (!force && style.inpreamble)
2536                 return;
2537
2538         bool const allowcust = allowParagraphCustomization();
2539
2540         // Current base font for all inherited font changes, without any
2541         // change caused by an individual character, except for the language:
2542         // It is set to the language of the first character.
2543         // As long as we are in the label, this font is the base font of the
2544         // label. Before the first body character it is set to the base font
2545         // of the body.
2546         Font basefont;
2547
2548         // If there is an open font-encoding changing command (script wrapper),
2549         // alien_script is set to its name
2550         string alien_script;
2551         string script;
2552
2553         // Maybe we have to create a optional argument.
2554         pos_type body_pos = beginOfBody();
2555         unsigned int column = 0;
2556
2557         // If we are inside an non inheritFont() inset,
2558         // the outerfont is the buffer's main font
2559         Font const real_outerfont =
2560                 inInset().inheritFont() ? outerfont : Font(bparams.getFont());
2561
2562         if (body_pos > 0) {
2563                 // the optional argument is kept in curly brackets in
2564                 // case it contains a ']'
2565                 // This is not strictly needed, but if this is changed it
2566                 // would be a file format change, and tex2lyx would need
2567                 // to be adjusted, since it unconditionally removes the
2568                 // braces when it parses \item.
2569                 os << "[{";
2570                 column += 2;
2571                 basefont = getLabelFont(bparams, real_outerfont);
2572         } else {
2573                 basefont = getLayoutFont(bparams, real_outerfont);
2574         }
2575
2576         // Which font is currently active?
2577         Font running_font(basefont);
2578         // Do we have an open font change?
2579         bool open_font = false;
2580
2581         Change runningChange =
2582             runparams.inDeletedInset && !inInset().canTrackChanges()
2583             ? runparams.changeOfDeletedInset : Change(Change::UNCHANGED);
2584
2585         Encoding const * const prev_encoding = runparams.encoding;
2586
2587         os.texrow().start(id(), 0);
2588
2589         // if the paragraph is empty, the loop will not be entered at all
2590         if (empty()) {
2591                 // For InTitle commands, we have already opened a group
2592                 // in output_latex::TeXOnePar.
2593                 if (style.isCommand() && !style.intitle) {
2594                         os << '{';
2595                         ++column;
2596                 }
2597                 if (!style.leftdelim().empty()) {
2598                         os << style.leftdelim();
2599                         column += style.leftdelim().size();
2600                 }
2601                 if (allowcust)
2602                         column += d->startTeXParParams(bparams, os, runparams);
2603         }
2604
2605         // Whether a \par can be issued for insets typeset inline with text.
2606         // Yes if greater than 0. This has to be static.
2607         THREAD_LOCAL_STATIC int parInline = 0;
2608
2609         for (pos_type i = 0; i < size(); ++i) {
2610                 // First char in paragraph or after label?
2611                 if (i == body_pos) {
2612                         if (body_pos > 0) {
2613                                 if (open_font) {
2614                                         bool needPar = false;
2615                                         column += running_font.latexWriteEndChanges(
2616                                                 os, bparams, runparams,
2617                                                 basefont, basefont, needPar);
2618                                         open_font = false;
2619                                 }
2620                                 basefont = getLayoutFont(bparams, real_outerfont);
2621                                 running_font = basefont;
2622
2623                                 column += Changes::latexMarkChange(os, bparams,
2624                                                 runningChange, Change(Change::UNCHANGED),
2625                                                 runparams);
2626                                 runningChange = Change(Change::UNCHANGED);
2627
2628                                 os << ((isEnvSeparator(i) && !runparams.find_effective()) ? "}]~" : "}] ");
2629                                 column +=3;
2630                         }
2631                         // For InTitle commands, we have already opened a group
2632                         // in output_latex::TeXOnePar.
2633                         if (style.isCommand() && !style.intitle) {
2634                                 os << '{';
2635                                 ++column;
2636                         }
2637
2638                         if (!style.leftdelim().empty()) {
2639                                 os << style.leftdelim();
2640                                 column += style.leftdelim().size();
2641                         }
2642
2643                         if (allowcust)
2644                                 column += d->startTeXParParams(bparams, os,
2645                                                             runparams);
2646                 }
2647
2648                 runparams.wasDisplayMath = runparams.inDisplayMath;
2649                 runparams.inDisplayMath = false;
2650                 bool deleted_display_math = false;
2651                 Change const & change = runparams.inDeletedInset
2652                         ? runparams.changeOfDeletedInset : lookupChange(i);
2653
2654                 char_type const c = d->text_[i];
2655
2656                 // Check whether a display math inset follows
2657                 bool output_changes;
2658                 if (!runparams.find_effective())
2659                         output_changes = bparams.output_changes;
2660                 else
2661                         output_changes = runparams.find_with_deleted();
2662                 if (c == META_INSET
2663                     && i >= start_pos && (end_pos == -1 || i < end_pos)) {
2664                         if (isDeleted(i))
2665                                 runparams.ctObject = getInset(i)->getCtObject(runparams);
2666
2667                         InsetMath const * im = getInset(i)->asInsetMath();
2668                         if (im && im->asHullInset()
2669                             && im->asHullInset()->outerDisplay()) {
2670                                 runparams.inDisplayMath = true;
2671                                 // runparams.inDeletedInset will be set by
2672                                 // latexInset later, but we need this info
2673                                 // before it is called. On the other hand, we
2674                                 // cannot set it here because it is a counter.
2675                                 deleted_display_math = isDeleted(i);
2676                         }
2677                         if (output_changes && deleted_display_math
2678                             && runningChange == change
2679                             && change.type == Change::DELETED
2680                             && !os.afterParbreak()) {
2681                                 // A display math in the same paragraph follows.
2682                                 // We have to close and then reopen \lyxdeleted,
2683                                 // otherwise the math will be shifted up.
2684                                 OutputParams rp = runparams;
2685                                 if (open_font) {
2686                                         bool needPar = false;
2687                                         column += running_font.latexWriteEndChanges(
2688                                                 os, bparams, rp, basefont,
2689                                                 basefont, needPar);
2690                                         open_font = false;
2691                                 }
2692                                 basefont = (body_pos > i) ? getLabelFont(bparams, real_outerfont)
2693                                                           : getLayoutFont(bparams, real_outerfont);
2694                                 running_font = basefont;
2695                                 column += Changes::latexMarkChange(os, bparams,
2696                                         Change(Change::INSERTED), change, rp);
2697                         }
2698                 }
2699
2700                 if (output_changes && runningChange != change) {
2701                         if (!alien_script.empty()) {
2702                                 column += 1;
2703                                 os << "}";
2704                                 alien_script.clear();
2705                         }
2706                         if (open_font) {
2707                                 bool needPar = false;
2708                                 column += running_font.latexWriteEndChanges(
2709                                                 os, bparams, runparams,
2710                                                 basefont, basefont, needPar);
2711                                 open_font = false;
2712                         }
2713                         basefont = (body_pos > i) ? getLabelFont(bparams, real_outerfont)
2714                                                   : getLayoutFont(bparams, real_outerfont);
2715                         running_font = basefont;
2716                         column += Changes::latexMarkChange(os, bparams, runningChange,
2717                                                            change, runparams);
2718                         runningChange = change;
2719                 }
2720
2721                 // do not output text which is marked deleted
2722                 // if change tracking output is disabled
2723                 if (!output_changes && change.deleted()) {
2724                         continue;
2725                 }
2726
2727                 ++column;
2728
2729                 // Fully instantiated font
2730                 Font current_font = getFont(bparams, i, outerfont);
2731                 // Previous font
2732                 Font const prev_font = (i > 0) ?
2733                                         getFont(bparams, i - 1, outerfont)
2734                                       : current_font;
2735
2736                 Font const last_font = running_font;
2737                 bool const in_ct_deletion = (output_changes
2738                                              && runningChange == change
2739                                              && change.type == Change::DELETED
2740                                              && !os.afterParbreak());
2741                 // Insets where font switches are used (rather than font commands)
2742                 bool const fontswitch_inset =
2743                                 c == META_INSET
2744                                 && getInset(i)
2745                                 && getInset(i)->allowMultiPar()
2746                                 && getInset(i)->lyxCode() != ERT_CODE
2747                                 && (getInset(i)->producesOutput()
2748                                     // FIXME Something more general?
2749                                     // Comments do not "produce output" but are still
2750                                     // part of the TeX source and require font switches
2751                                     // to be closed (otherwise LaTeX fails).
2752                                     || getInset(i)->layoutName() == "Note:Comment");
2753
2754                 bool closeLanguage = false;
2755                 bool lang_switched_at_inset = false;
2756                 if (fontswitch_inset) {
2757                         // Some insets cannot be inside a font change command.
2758                         // However, even such insets *can* be placed in \L or \R
2759                         // or their equivalents (for RTL language switches),
2760                         // so we don't close the language in those cases
2761                         // (= differing isRightToLeft()).
2762                         // ArabTeX, though, doesn't seem to handle this special behavior.
2763                         closeLanguage = basefont.isRightToLeft() == current_font.isRightToLeft()
2764                                         || basefont.language()->lang() == "arabic_arabtex"
2765                                         || current_font.language()->lang() == "arabic_arabtex";
2766                         // We need to check prev_font as language changes directly at inset
2767                         // will only be started inside the inset.
2768                         lang_switched_at_inset = prev_font.language() != current_font.language();
2769                 }
2770
2771                 // Do we need to close the previous font?
2772                 bool langClosed = false;
2773                 if (open_font &&
2774                     ((current_font != running_font
2775                       || current_font.language() != running_font.language())
2776                      || (fontswitch_inset
2777                          && (current_font == prev_font))))
2778                 {
2779                         // ensure there is no open script-wrapper
2780                         if (!alien_script.empty()) {
2781                                 column += 1;
2782                                 os << "}";
2783                                 alien_script.clear();
2784                         }
2785                         if (in_ct_deletion) {
2786                                 // We have to close and then reopen \lyxdeleted,
2787                                 // as strikeout needs to be on lowest level.
2788                                 os << '}';
2789                                 column += 1;
2790                         }
2791                         if (closeLanguage) {
2792                                 // Force language closing
2793                                 current_font.setLanguage(basefont.language());
2794                                 langClosed = true;
2795                         }
2796                         Font const nextfont = (i == body_pos-1) ? basefont : current_font;
2797                         bool needPar = false;
2798                         column += running_font.latexWriteEndChanges(
2799                                     os, bparams, runparams, basefont,
2800                                     nextfont, needPar);
2801                         if (in_ct_deletion) {
2802                                 // We have to close and then reopen \lyxdeleted,
2803                                 // as strikeout needs to be on lowest level.
2804                                 OutputParams rp = runparams;
2805                                 column += Changes::latexMarkChange(os, bparams,
2806                                         Change(Change::UNCHANGED), Change(Change::DELETED), rp);
2807                         }
2808                         // Has the language been closed in the latexWriteEndChanges() call above?
2809                         langClosed |= running_font.language() != basefont.language()
2810                                         && running_font.language() != nextfont.language()
2811                                         && (running_font.language()->encoding()->package() != Encoding::CJK);
2812                         running_font = basefont;
2813                         open_font &= !langClosed;
2814                 }
2815
2816                 // if necessary, close language environment before opening CJK
2817                 string const running_lang = running_font.language()->babel();
2818                 string const lang_end_command = lyxrc.language_command_end;
2819                 if (!lang_end_command.empty() && !bparams.useNonTeXFonts
2820                         && !running_lang.empty()
2821                         && running_lang == openLanguageName()
2822                         && current_font.language()->encoding()->package() == Encoding::CJK) {
2823                         string end_tag = subst(lang_end_command, "$$lang", running_lang);
2824                         os << from_ascii(end_tag);
2825                         column += end_tag.length();
2826                         if (!languageStackEmpty())
2827                                 popLanguageName();
2828                 }
2829
2830                 // Switch file encoding if necessary (and allowed)
2831                 if ((!fontswitch_inset || closeLanguage)
2832                     && !runparams.pass_thru && !style.pass_thru &&
2833                     runparams.encoding->package() != Encoding::none &&
2834                     current_font.language()->encoding()->package() != Encoding::none) {
2835                         pair<bool, int> const enc_switch =
2836                                 switchEncoding(os.os(), bparams, runparams,
2837                                         *(current_font.language()->encoding()));
2838                         if (enc_switch.first) {
2839                                 column += enc_switch.second;
2840                                 runparams.encoding = current_font.language()->encoding();
2841                         }
2842                 }
2843
2844                 // A display math inset inside an ulem command will be output
2845                 // as a box of width \linewidth, so we have to either disable
2846                 // indentation if the inset starts a paragraph, or start a new
2847                 // line to accommodate such box. This has to be done before
2848                 // writing any font changing commands.
2849                 if (runparams.inDisplayMath && !deleted_display_math
2850                     && runparams.inulemcmd) {
2851                         if (os.afterParbreak())
2852                                 os << "\\noindent";
2853                         else
2854                                 os << "\\\\\n";
2855                 }
2856
2857                 // Do we need to change font?
2858                 if ((current_font != running_font ||
2859                      current_font.language() != running_font.language())
2860                     && i != body_pos - 1)
2861                 {
2862                         if (!fontswitch_inset) {
2863                                 if (in_ct_deletion) {
2864                                         // We have to close and then reopen \lyxdeleted,
2865                                         // as strikeout needs to be on lowest level.
2866                                         OutputParams rp = runparams;
2867                                         bool needPar = false;
2868                                         column += running_font.latexWriteEndChanges(
2869                                                 os, bparams, rp, basefont,
2870                                                 basefont, needPar);
2871                                         os << '}';
2872                                         column += 1;
2873                                 }
2874                                 otexstringstream ots;
2875                                 InsetText const * textinset = inInset().asInsetText();
2876                                 bool const cprotect = textinset
2877                                         ? textinset->hasCProtectContent(runparams.moving_arg)
2878                                           && !textinset->text().isMainText()
2879                                           && inInset().lyxCode() != BRANCH_CODE
2880                                         : false;
2881                                 column += current_font.latexWriteStartChanges(ots, bparams,
2882                                                                               runparams, basefont, last_font, false,
2883                                                                               cprotect);
2884                                 // Check again for display math in ulem commands as a
2885                                 // font change may also occur just before a math inset.
2886                                 if (runparams.inDisplayMath && !deleted_display_math
2887                                     && runparams.inulemcmd) {
2888                                         if (os.afterParbreak())
2889                                                 os << "\\noindent";
2890                                         else
2891                                                 os << "\\\\\n";
2892                                 }
2893                                 running_font = current_font;
2894                                 open_font = true;
2895                                 docstring fontchange = ots.str();
2896                                 os << fontchange;
2897                                 // check whether the fontchange ends with a \\textcolor
2898                                 // modifier and the text starts with a space. If so we
2899                                 // need to add } in order to prevent \\textcolor from gobbling
2900                                 // the space (bug 4473).
2901                                 docstring const last_modifier = rsplit(fontchange, '\\');
2902                                 if (prefixIs(last_modifier, from_ascii("textcolor")) && c == ' ')
2903                                         os << from_ascii("{}");
2904                                 else if (ots.terminateCommand())
2905                                         os << termcmd;
2906                                 if (in_ct_deletion) {
2907                                         // We have to close and then reopen \lyxdeleted,
2908                                         // as strikeout needs to be on lowest level.
2909                                         OutputParams rp = runparams;
2910                                         column += Changes::latexMarkChange(os, bparams,
2911                                                 Change(Change::UNCHANGED), change, rp);
2912                                 }
2913                         } else {// if fontswitch_inset
2914                                 if (current_font != running_font || !langClosed)
2915                                         // font is still open in fontswitch_insets if we have
2916                                         // a non-lang font difference or if the language
2917                                         // is the only difference but has not been forcedly
2918                                         // closed meanwhile
2919                                         open_font = true;
2920                                 running_font = current_font;
2921                         }
2922                 }
2923
2924                 // FIXME: think about end_pos implementation...
2925                 if (c == ' ' && i >= start_pos && (end_pos == -1 || i < end_pos)) {
2926                         // FIXME: integrate this case in latexSpecialChar
2927                         // Do not print the separation of the optional argument
2928                         // if style.pass_thru is false. This works because
2929                         // latexSpecialChar ignores spaces if
2930                         // style.pass_thru is false.
2931                         if (i != body_pos - 1) {
2932                                 if (d->simpleTeXBlanks(bparams, runparams, os,
2933                                                 i, column, current_font, style)) {
2934                                         // A surrogate pair was output. We
2935                                         // must not call latexSpecialChar
2936                                         // in this iteration, since it would output
2937                                         // the combining character again.
2938                                         ++i;
2939                                         continue;
2940                                 }
2941                         }
2942                 }
2943
2944                 OutputParams rp = runparams;
2945                 rp.free_spacing = style.free_spacing;
2946                 rp.local_font = &current_font;
2947                 rp.intitle = style.intitle;
2948
2949                 // Two major modes:  LaTeX or plain
2950                 // Handle here those cases common to both modes
2951                 // and then split to handle the two modes separately.
2952                 if (c == META_INSET) {
2953                         if (i >= start_pos && (end_pos == -1 || i < end_pos)) {
2954                                 // Greyedout notes and, in general, all insets
2955                                 // with InsetLayout::isDisplay() == false,
2956                                 // are typeset inline with the text. So, we
2957                                 // can add a \par to the last paragraph of
2958                                 // such insets only if nothing else follows.
2959                                 bool incremented = false;
2960                                 Inset const * inset = getInset(i);
2961                                 InsetText const * textinset = inset
2962                                                         ? inset->asInsetText()
2963                                                         : nullptr;
2964                                 if (i + 1 == size() && textinset
2965                                     && !inset->getLayout().isDisplay()) {
2966                                         ParagraphList const & pars =
2967                                                 textinset->text().paragraphs();
2968                                         pit_type const pit = pars.size() - 1;
2969                                         Font const lastfont =
2970                                                 pit < 0 || pars[pit].empty()
2971                                                 ? pars[pit].getLayoutFont(
2972                                                                 bparams,
2973                                                                 real_outerfont)
2974                                                 : pars[pit].getFont(bparams,
2975                                                         pars[pit].size() - 1,
2976                                                         real_outerfont);
2977                                         if (lastfont.fontInfo().size() !=
2978                                             basefont.fontInfo().size()) {
2979                                                 ++parInline;
2980                                                 incremented = true;
2981                                         }
2982                                 }
2983                                 // We need to restore parts of this after insets with
2984                                 // allowMultiPar() true
2985                                 Font const save_basefont = basefont;
2986                                 d->latexInset(bparams, os, rp, running_font,
2987                                                 basefont, real_outerfont, open_font,
2988                                                 runningChange, style, i, column, fontswitch_inset,
2989                                                 closeLanguage, (lang_switched_at_inset || langClosed));
2990                                 if (fontswitch_inset) {
2991                                         if (open_font) {
2992                                                 bool needPar = false;
2993                                                 column += running_font.latexWriteEndChanges(
2994                                                         os, bparams, runparams,
2995                                                         basefont, basefont, needPar);
2996                                                 open_font = false;
2997                                         }
2998                                         basefont.fontInfo().setSize(save_basefont.fontInfo().size());
2999                                         basefont.fontInfo().setFamily(save_basefont.fontInfo().family());
3000                                         basefont.fontInfo().setSeries(save_basefont.fontInfo().series());
3001                                 }
3002                                 if (incremented)
3003                                         --parInline;
3004
3005                                 if (runparams.ctObject == CtObject::DisplayObject
3006                                     || runparams.ctObject == CtObject::UDisplayObject) {
3007                                         // Close \lyx*deleted and force its
3008                                         // reopening (if needed)
3009                                         os << '}';
3010                                         column++;
3011                                         runningChange = Change(Change::UNCHANGED);
3012                                         runparams.ctObject = CtObject::Normal;
3013                                 }
3014                         }
3015                 } else if (i >= start_pos && (end_pos == -1 || i < end_pos)) {
3016                         if (!bparams.useNonTeXFonts)
3017                           script = Encodings::isKnownScriptChar(c);
3018                         if (script != alien_script) {
3019                                 if (!alien_script.empty()) {
3020                                         os << "}";
3021                                         alien_script.clear();
3022                                 }
3023                                 string fontenc = running_font.language()->fontenc(bparams);
3024                                 if (!script.empty()
3025                                         && !Encodings::fontencSupportsScript(fontenc, script)) {
3026                                         column += script.length() + 2;
3027                                         os << "\\" << script << "{";
3028                                         alien_script = script;
3029                                 }
3030                         }
3031                         try {
3032                                 d->latexSpecialChar(os, bparams, rp, running_font,
3033                                                                         alien_script, style, i, end_pos, column);
3034                         } catch (EncodingException & e) {
3035                                 if (runparams.dryrun) {
3036                                         os << "<" << _("LyX Warning: ")
3037                                            << _("uncodable character") << " '";
3038                                         os.put(c);
3039                                         os << "'>";
3040                                 } else {
3041                                         // add location information and throw again.
3042                                         e.par_id = id();
3043                                         e.pos = i;
3044                                         throw;
3045                                 }
3046                         }
3047                 }
3048
3049                 // Set the encoding to that returned from latexSpecialChar (see
3050                 // comment for encoding member in OutputParams.h)
3051                 runparams.encoding = rp.encoding;
3052
3053                 // Also carry on the info on a closed ulem command for insets
3054                 // such as Note that do not produce any output, so that no
3055                 // command is ever executed but its opening was recorded.
3056                 runparams.inulemcmd = rp.inulemcmd;
3057
3058                 // These need to be passed upstream as well
3059                 runparams.need_maketitle = rp.need_maketitle;
3060                 runparams.have_maketitle = rp.have_maketitle;
3061
3062                 // And finally, pass the post_macros upstream
3063                 runparams.post_macro = rp.post_macro;
3064         }
3065
3066         // Close wrapper for alien script
3067         if (!alien_script.empty()) {
3068                 os << "}";
3069                 alien_script.clear();
3070         }
3071
3072         Font const font = empty()
3073                 ? getLayoutFont(bparams, real_outerfont)
3074                 : getFont(bparams, size() - 1, real_outerfont);
3075
3076         InsetText const * textinset = inInset().asInsetText();
3077
3078         bool const maintext = textinset
3079                 ? textinset->text().isMainText()
3080                 : false;
3081
3082         size_t const numpars = textinset
3083                 ? textinset->text().paragraphs().size()
3084                 : 0;
3085
3086         bool needPar = false;
3087
3088         if (style.resfont.size() != font.fontInfo().size()
3089             && (!runparams.isLastPar || maintext
3090                 || (numpars > 1 && d->ownerCode() != CELL_CODE
3091                     && (inInset().getLayout().isDisplay()
3092                         || parInline)))
3093             && !style.isCommand()) {
3094                 needPar = true;
3095         }
3096
3097         // If we have an open font definition, we have to close it
3098         if (open_font) {
3099                 // Make sure that \\par is done with the font of the last
3100                 // character if this has another size as the default.
3101                 // This is necessary because LaTeX (and LyX on the screen)
3102                 // calculates the space between the baselines according
3103                 // to this font. (Matthias)
3104                 //
3105                 // We must not change the font for the last paragraph
3106                 // of non-multipar insets, tabular cells or commands,
3107                 // since this produces unwanted whitespace.
3108 #ifdef FIXED_LANGUAGE_END_DETECTION
3109                 if (next_) {
3110                         running_font.latexWriteEndChanges(os, bparams,
3111                                         runparams, basefont,
3112                                         next_->getFont(bparams, 0, outerfont),
3113                                                        needPar);
3114                 } else {
3115                         running_font.latexWriteEndChanges(os, bparams,
3116                                         runparams, basefont, basefont, needPar);
3117                 }
3118 #else
3119 //FIXME: For now we ALWAYS have to close the foreign font settings if they are
3120 //FIXME: there as we start another \selectlanguage with the next paragraph if
3121 //FIXME: we are in need of this. This should be fixed sometime (Jug)
3122                 running_font.latexWriteEndChanges(os, bparams, runparams,
3123                                 basefont, basefont, needPar);
3124 #endif
3125         }
3126         if (needPar) {
3127                 // The \par could not be inserted at the same nesting
3128                 // level of the font size change, so do it now.
3129                 os << "{\\" << font.latexSize() << "\\par}";
3130         }
3131
3132         if (!runparams.inDeletedInset || inInset().canTrackChanges())
3133                 column += Changes::latexMarkChange(os, bparams, runningChange,
3134                                         Change(Change::UNCHANGED), runparams);
3135
3136         // Needed if there is an optional argument but no contents.
3137         if (body_pos > 0 && body_pos == size()) {
3138                 os << "}]~";
3139         }
3140
3141         if (!style.rightdelim().empty()) {
3142                 os << style.rightdelim();
3143                 column += style.rightdelim().size();
3144         }
3145
3146         if (allowcust && d->endTeXParParams(bparams, os, runparams)
3147             && runparams.encoding != prev_encoding) {
3148                 runparams.encoding = prev_encoding;
3149                 os << setEncoding(prev_encoding->iconvName());
3150         }
3151
3152         LYXERR(Debug::OUTFILE, "Paragraph::latex... done " << this);
3153 }
3154
3155
3156 bool Paragraph::emptyTag() const
3157 {
3158         for (pos_type i = 0; i < size(); ++i) {
3159                 if (Inset const * inset = getInset(i)) {
3160                         InsetCode lyx_code = inset->lyxCode();
3161                         // FIXME testing like that is wrong. What is
3162                         // the intent?
3163                         if (lyx_code != TOC_CODE &&
3164                             lyx_code != INCLUDE_CODE &&
3165                             lyx_code != GRAPHICS_CODE &&
3166                             lyx_code != ERT_CODE &&
3167                             lyx_code != LISTINGS_CODE &&
3168                             lyx_code != FLOAT_CODE &&
3169                             lyx_code != TABULAR_CODE) {
3170                                 return false;
3171                         }
3172                 } else {
3173                         char_type c = d->text_[i];
3174                         if (c != ' ' && c != '\t')
3175                                 return false;
3176                 }
3177         }
3178         return true;
3179 }
3180
3181
3182 string Paragraph::getID(Buffer const &, OutputParams const &)
3183         const
3184 {
3185         for (pos_type i = 0; i < size(); ++i) {
3186                 if (Inset const * inset = getInset(i)) {
3187                         InsetCode lyx_code = inset->lyxCode();
3188                         if (lyx_code == LABEL_CODE) {
3189                                 InsetLabel const * const il = static_cast<InsetLabel const *>(inset);
3190                                 docstring const & id = il->getParam("name");
3191                                 return "id='" + to_utf8(xml::cleanID(id)) + "'";
3192                         }
3193                 }
3194         }
3195         return string();
3196 }
3197
3198
3199 pos_type Paragraph::firstWordDocBook(XMLStream & xs, OutputParams const & runparams) const
3200 {
3201         pos_type i;
3202         for (i = 0; i < size(); ++i) {
3203                 if (Inset const * inset = getInset(i)) {
3204                         inset->docbook(xs, runparams);
3205                 } else {
3206                         char_type c = d->text_[i];
3207                         if (c == ' ')
3208                                 break;
3209                         xs << c;
3210                 }
3211         }
3212         return i;
3213 }
3214
3215
3216 pos_type Paragraph::firstWordLyXHTML(XMLStream & xs, OutputParams const & runparams)
3217         const
3218 {
3219         pos_type i;
3220         for (i = 0; i < size(); ++i) {
3221                 if (Inset const * inset = getInset(i)) {
3222                         inset->xhtml(xs, runparams);
3223                 } else {
3224                         char_type c = d->text_[i];
3225                         if (c == ' ')
3226                                 break;
3227                         xs << c;
3228                 }
3229         }
3230         return i;
3231 }
3232
3233
3234 bool Paragraph::Private::onlyText(Buffer const & buf, Font const & outerfont, pos_type initial) const
3235 {
3236         Font font_old;
3237         pos_type size = text_.size();
3238         for (pos_type i = initial; i < size; ++i) {
3239                 Font font = owner_->getFont(buf.params(), i, outerfont);
3240                 if (text_[i] == META_INSET)
3241                         return false;
3242                 if (i != initial && font != font_old)
3243                         return false;
3244                 font_old = font;
3245         }
3246
3247         return true;
3248 }
3249
3250
3251 namespace {
3252
3253 void doFontSwitchDocBook(vector<xml::FontTag> & tagsToOpen,
3254                   vector<xml::EndFontTag> & tagsToClose,
3255                   bool & flag, FontState curstate, xml::FontTypes type)
3256 {
3257         if (curstate == FONT_ON) {
3258                 tagsToOpen.push_back(docbookStartFontTag(type));
3259                 flag = true;
3260         } else if (flag) {
3261                 tagsToClose.push_back(docbookEndFontTag(type));
3262                 flag = false;
3263         }
3264 }
3265
3266 class OptionalFontType {
3267 public:
3268         xml::FontTypes ft;
3269         bool has_value;
3270
3271         OptionalFontType(): ft(xml::FT_EMPH), has_value(false) {} // A possible value at random for ft.
3272         OptionalFontType(xml::FontTypes ft): ft(ft), has_value(true) {}
3273 };
3274
3275 OptionalFontType fontShapeToXml(FontShape fs)
3276 {
3277         switch (fs) {
3278         case ITALIC_SHAPE:
3279                 return {xml::FT_ITALIC};
3280         case SLANTED_SHAPE:
3281                 return {xml::FT_SLANTED};
3282         case SMALLCAPS_SHAPE:
3283                 return {xml::FT_SMALLCAPS};
3284         case UP_SHAPE:
3285         case INHERIT_SHAPE:
3286                 return {};
3287         default:
3288                 // the other tags are for internal use
3289                 LATTEST(false);
3290                 return {};
3291         }
3292 }
3293
3294 OptionalFontType fontFamilyToXml(FontFamily fm)
3295 {
3296         switch (fm) {
3297         case ROMAN_FAMILY:
3298                 return {xml::FT_ROMAN};
3299         case SANS_FAMILY:
3300                 return {xml::FT_SANS};
3301         case TYPEWRITER_FAMILY:
3302                 return {xml::FT_TYPE};
3303         case INHERIT_FAMILY:
3304                 return {};
3305         default:
3306                 // the other tags are for internal use
3307                 LATTEST(false);
3308                 return {};
3309         }
3310 }
3311
3312 OptionalFontType fontSizeToXml(FontSize fs)
3313 {
3314         switch (fs) {
3315         case TINY_SIZE:
3316                 return {xml::FT_SIZE_TINY};
3317         case SCRIPT_SIZE:
3318                 return {xml::FT_SIZE_SCRIPT};
3319         case FOOTNOTE_SIZE:
3320                 return {xml::FT_SIZE_FOOTNOTE};
3321         case SMALL_SIZE:
3322                 return {xml::FT_SIZE_SMALL};
3323         case LARGE_SIZE:
3324                 return {xml::FT_SIZE_LARGE};
3325         case LARGER_SIZE:
3326                 return {xml::FT_SIZE_LARGER};
3327         case LARGEST_SIZE:
3328                 return {xml::FT_SIZE_LARGEST};
3329         case HUGE_SIZE:
3330                 return {xml::FT_SIZE_HUGE};
3331         case HUGER_SIZE:
3332                 return {xml::FT_SIZE_HUGER};
3333         case INCREASE_SIZE:
3334                 return {xml::FT_SIZE_INCREASE};
3335         case DECREASE_SIZE:
3336                 return {xml::FT_SIZE_DECREASE};
3337         case INHERIT_SIZE:
3338         case NORMAL_SIZE:
3339                 return {};
3340         default:
3341                 // the other tags are for internal use
3342                 LATTEST(false);
3343                 return {};
3344         }
3345 }
3346
3347 struct DocBookFontState
3348 {
3349         FontShape  curr_fs   = INHERIT_SHAPE;
3350         FontFamily curr_fam  = INHERIT_FAMILY;
3351         FontSize   curr_size = INHERIT_SIZE;
3352
3353         // track whether we have opened these tags
3354         bool emph_flag = false;
3355         bool bold_flag = false;
3356         bool noun_flag = false;
3357         bool ubar_flag = false;
3358         bool dbar_flag = false;
3359         bool sout_flag = false;
3360         bool xout_flag = false;
3361         bool wave_flag = false;
3362         // shape tags
3363         bool shap_flag = false;
3364         // family tags
3365         bool faml_flag = false;
3366         // size tags
3367         bool size_flag = false;
3368 };
3369
3370 std::tuple<vector<xml::FontTag>, vector<xml::EndFontTag>> computeDocBookFontSwitch(FontInfo const & font_old,
3371                                                                                            Font const & font,
3372                                                                                            std::string const & default_family,
3373                                                                                            DocBookFontState & fs)
3374 {
3375         vector<xml::FontTag> tagsToOpen;
3376         vector<xml::EndFontTag> tagsToClose;
3377
3378         // emphasis
3379         FontState curstate = font.fontInfo().emph();
3380         if (font_old.emph() != curstate)
3381                 doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.emph_flag, curstate, xml::FT_EMPH);
3382
3383         // noun
3384         curstate = font.fontInfo().noun();
3385         if (font_old.noun() != curstate)
3386                 doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.noun_flag, curstate, xml::FT_NOUN);
3387
3388         // underbar
3389         curstate = font.fontInfo().underbar();
3390         if (font_old.underbar() != curstate)
3391                 doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.ubar_flag, curstate, xml::FT_UBAR);
3392
3393         // strikeout
3394         curstate = font.fontInfo().strikeout();
3395         if (font_old.strikeout() != curstate)
3396                 doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.sout_flag, curstate, xml::FT_SOUT);
3397
3398         // xout
3399         curstate = font.fontInfo().xout();
3400         if (font_old.xout() != curstate)
3401                 doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.xout_flag, curstate, xml::FT_XOUT);
3402
3403         // double underbar
3404         curstate = font.fontInfo().uuline();
3405         if (font_old.uuline() != curstate)
3406                 doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.dbar_flag, curstate, xml::FT_DBAR);
3407
3408         // wavy line
3409         curstate = font.fontInfo().uwave();
3410         if (font_old.uwave() != curstate)
3411                 doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.wave_flag, curstate, xml::FT_WAVE);
3412
3413         // bold
3414         // a little hackish, but allows us to reuse what we have.
3415         curstate = (font.fontInfo().series() == BOLD_SERIES ? FONT_ON : FONT_OFF);
3416         if (font_old.series() != font.fontInfo().series())
3417                 doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.bold_flag, curstate, xml::FT_BOLD);
3418
3419         // Font shape
3420         fs.curr_fs = font.fontInfo().shape();
3421         FontShape old_fs = font_old.shape();
3422         if (old_fs != fs.curr_fs) {
3423                 if (fs.shap_flag) {
3424                         OptionalFontType tag = fontShapeToXml(old_fs);
3425                         if (tag.has_value)
3426                                 tagsToClose.push_back(docbookEndFontTag(tag.ft));
3427                         fs.shap_flag = false;
3428                 }
3429
3430                 OptionalFontType tag = fontShapeToXml(fs.curr_fs);
3431                 if (tag.has_value)
3432                         tagsToOpen.push_back(docbookStartFontTag(tag.ft));
3433         }
3434
3435         // Font family
3436         fs.curr_fam = font.fontInfo().family();
3437         FontFamily old_fam = font_old.family();
3438         if (old_fam != fs.curr_fam) {
3439                 if (fs.faml_flag) {
3440                         OptionalFontType tag = fontFamilyToXml(old_fam);
3441                         if (tag.has_value)
3442                                 tagsToClose.push_back(docbookEndFontTag(tag.ft));
3443                         fs.faml_flag = false;
3444                 }
3445                 switch (fs.curr_fam) {
3446                         case ROMAN_FAMILY:
3447                                 // we will treat a "default" font family as roman, since we have
3448                                 // no other idea what to do.
3449                                 if (default_family != "rmdefault" && default_family != "default") {
3450                                         tagsToOpen.push_back(docbookStartFontTag(xml::FT_ROMAN));
3451                                         fs.faml_flag = true;
3452                                 }
3453                                 break;
3454                         case SANS_FAMILY:
3455                                 if (default_family != "sfdefault") {
3456                                         tagsToOpen.push_back(docbookStartFontTag(xml::FT_SANS));
3457                                         fs.faml_flag = true;
3458                                 }
3459                                 break;
3460                         case TYPEWRITER_FAMILY:
3461                                 if (default_family != "ttdefault") {
3462                                         tagsToOpen.push_back(docbookStartFontTag(xml::FT_TYPE));
3463                                         fs.faml_flag = true;
3464                                 }
3465                                 break;
3466                         case INHERIT_FAMILY:
3467                                 break;
3468                         default:
3469                                 // the other tags are for internal use
3470                                 LATTEST(false);
3471                                 break;
3472                 }
3473         }
3474
3475         // Font size
3476         fs.curr_size = font.fontInfo().size();
3477         FontSize old_size = font_old.size();
3478         if (old_size != fs.curr_size) {
3479                 if (fs.size_flag) {
3480                         OptionalFontType tag = fontSizeToXml(old_size);
3481                         if (tag.has_value)
3482                                 tagsToClose.push_back(docbookEndFontTag(tag.ft));
3483                         fs.size_flag = false;
3484                 }
3485
3486                 OptionalFontType tag = fontSizeToXml(fs.curr_size);
3487                 if (tag.has_value) {
3488                         tagsToOpen.push_back(docbookStartFontTag(tag.ft));
3489                         fs.size_flag = true;
3490                 }
3491         }
3492
3493         return std::tuple<vector<xml::FontTag>, vector<xml::EndFontTag>>(tagsToOpen, tagsToClose);
3494 }
3495
3496 } // anonymous namespace
3497
3498
3499 std::tuple<std::vector<docstring>, std::vector<docstring>, std::vector<docstring>>
3500     Paragraph::simpleDocBookOnePar(Buffer const & buf,
3501                                                       OutputParams const & runparams,
3502                                                       Font const & outerfont,
3503                                                       pos_type initial,
3504                                                       bool is_last_par,
3505                                                       bool ignore_fonts) const
3506 {
3507         std::vector<docstring> prependedParagraphs;
3508         std::vector<docstring> generatedParagraphs;
3509         std::vector<docstring> appendedParagraphs;
3510         odocstringstream os;
3511
3512         // If there is an argument that must be output before the main tag, do it before handling the rest of the paragraph.
3513         // Also tag all arguments that shouldn't go in the main content right now, so that they are never generated at the
3514         // wrong place.
3515         OutputParams rp = runparams;
3516     for (pos_type i = initial; i < size(); ++i) {
3517         if (getInset(i) && getInset(i)->lyxCode() == ARG_CODE) {
3518             const InsetArgument * arg = getInset(i)->asInsetArgument();
3519             if (arg->docbookargumentbeforemaintag()) {
3520                 auto xs_local = XMLStream(os);
3521                 arg->docbook(xs_local, rp);
3522
3523                 prependedParagraphs.push_back(os.str());
3524                 os.str(from_ascii(""));
3525
3526                 rp.docbook_prepended_arguments.insert(arg);
3527             } else if (arg->docbookargumentaftermaintag()) {
3528                 rp.docbook_appended_arguments.insert(arg);
3529             }
3530         }
3531     }
3532
3533     // State variables for the main loop.
3534     auto xs = new XMLStream(os); // XMLStream has no copy constructor: to create a new object, the only solution
3535     // is to hold a pointer to the XMLStream (xs = XMLStream(os) is not allowed once the first object is built).
3536     std::vector<char_type> delayedChars; // When a font tag ends with a space, output it after the closing font tag.
3537     // This requires to store delayed characters at some point.
3538
3539     DocBookFontState fs; // Track whether we have opened font tags
3540     DocBookFontState old_fs = fs;
3541
3542     Layout const & style = *d->layout_;
3543     FontInfo font_old = style.labeltype == LABEL_MANUAL ? style.labelfont : style.font;
3544     string const default_family = buf.masterBuffer()->params().fonts_default_family;
3545
3546     vector<xml::FontTag> tagsToOpen;
3547     vector<xml::EndFontTag> tagsToClose;
3548
3549         // Parsing main loop.
3550         for (pos_type i = initial; i < size(); ++i) {
3551             bool ignore_fonts_i = ignore_fonts
3552                               || style.docbooknofontinside()
3553                               || (getInset(i) && getInset(i)->getLayout().docbooknofontinside());
3554
3555                 // Don't show deleted material in the output.
3556                 if (isDeleted(i))
3557                         continue;
3558
3559                 // If this is an InsetNewline, generate a new paragraph. Also reset the fonts, so that tags are closed in
3560                 // this paragraph.
3561                 if (getInset(i) && getInset(i)->lyxCode() == NEWLINE_CODE) {
3562                         if (!ignore_fonts_i)
3563                                 xs->closeFontTags();
3564
3565                         // Output one paragraph (i.e. one string entry in generatedParagraphs).
3566                         generatedParagraphs.push_back(os.str());
3567
3568                         // Create a new XMLStream for the new paragraph, completely independent from the previous one. This implies
3569                         // that the string stream must be reset.
3570                         os.str(from_ascii(""));
3571                         delete xs;
3572                         xs = new XMLStream(os);
3573
3574                         // Restore the fonts for the new paragraph, so that the right tags are opened for the new entry.
3575                         if (!ignore_fonts_i) {
3576                                 font_old = outerfont.fontInfo();
3577                                 fs = old_fs;
3578                         }
3579                 }
3580
3581                 // Determine which tags should be opened or closed regarding fonts.
3582                 Font const font = getFont(buf.masterBuffer()->params(), i, outerfont);
3583         tie(tagsToOpen, tagsToClose) = computeDocBookFontSwitch(font_old, font, default_family, fs);
3584
3585                 if (!ignore_fonts_i) {
3586             vector<xml::EndFontTag>::const_iterator cit = tagsToClose.begin();
3587             vector<xml::EndFontTag>::const_iterator cen = tagsToClose.end();
3588             for (; cit != cen; ++cit)
3589                 *xs << *cit;
3590         }
3591
3592         // Deal with the delayed characters *after* closing font tags.
3593         if (!delayedChars.empty()) {
3594             for (char_type c: delayedChars)
3595                 *xs << c;
3596             delayedChars.clear();
3597         }
3598
3599         if (!ignore_fonts_i) {
3600                         vector<xml::FontTag>::const_iterator sit = tagsToOpen.begin();
3601                         vector<xml::FontTag>::const_iterator sen = tagsToOpen.end();
3602                         for (; sit != sen; ++sit)
3603                                 *xs << *sit;
3604
3605                         tagsToClose.clear();
3606                         tagsToOpen.clear();
3607                 }
3608
3609         // Finally, write the next character or inset.
3610                 if (Inset const * inset = getInset(i)) {
3611                     bool inset_is_argument_elsewhere = getInset(i)->asInsetArgument() &&
3612                             rp.docbook_appended_arguments.find(inset->asInsetArgument()) != rp.docbook_appended_arguments.end() &&
3613                             rp.docbook_prepended_arguments.find(inset->asInsetArgument()) != rp.docbook_prepended_arguments.end();
3614
3615                         if ((!rp.for_toc || inset->isInToc()) && !inset_is_argument_elsewhere) {
3616                             // Arguments may need to be output
3617                                 OutputParams np = rp;
3618                                 np.local_font = &font;
3619
3620                                 // TODO: special case will bite here.
3621                                 np.docbook_in_par = true;
3622                                 inset->docbook(*xs, np);
3623                         }
3624                 } else {
3625                         char_type c = getUChar(buf.masterBuffer()->params(), rp, i);
3626                         if (lyx::isSpace(c) && !ignore_fonts)
3627                                 delayedChars.push_back(c);
3628                         else
3629                                 *xs << c;
3630                 }
3631                 font_old = font.fontInfo();
3632         }
3633
3634         // FIXME, this code is just imported from XHTML
3635         // I'm worried about what happens if a branch, say, is itself
3636         // wrapped in some font stuff. I think that will not work.
3637         if (!ignore_fonts)
3638                 xs->closeFontTags();
3639
3640         // Deal with the delayed characters *after* closing font tags.
3641         if (!delayedChars.empty())
3642                 for (char_type c: delayedChars)
3643                         *xs << c;
3644
3645         // In listings, new lines (i.e. \n characters in the output) are very important. Avoid generating one for the
3646         // last line to get a clean output.
3647         if (rp.docbook_in_listing && !is_last_par)
3648                 *xs << xml::CR();
3649
3650         // Finalise the last (and most likely only) paragraph.
3651         generatedParagraphs.push_back(os.str());
3652     os.str(from_ascii(""));
3653     delete xs;
3654
3655     // If there is an argument that must be output after the main tag, do it after handling the rest of the paragraph.
3656     for (pos_type i = initial; i < size(); ++i) {
3657         if (getInset(i) && getInset(i)->lyxCode() == ARG_CODE) {
3658             const InsetArgument * arg = getInset(i)->asInsetArgument();
3659             if (arg->docbookargumentaftermaintag()) {
3660                 // Don't use rp, as this argument would not generate anything.
3661                 auto xs_local = XMLStream(os);
3662                 arg->docbook(xs_local, runparams);
3663
3664                 appendedParagraphs.push_back(os.str());
3665                 os.str(from_ascii(""));
3666             }
3667         }
3668     }
3669
3670         return std::make_tuple(prependedParagraphs, generatedParagraphs, appendedParagraphs);
3671 }
3672
3673
3674 namespace {
3675
3676 void doFontSwitchXHTML(vector<xml::FontTag> & tagsToOpen,
3677                   vector<xml::EndFontTag> & tagsToClose,
3678                   bool & flag, FontState curstate, xml::FontTypes type)
3679 {
3680         if (curstate == FONT_ON) {
3681                 tagsToOpen.push_back(xhtmlStartFontTag(type));
3682                 flag = true;
3683         } else if (flag) {
3684                 tagsToClose.push_back(xhtmlEndFontTag(type));
3685                 flag = false;
3686         }
3687 }
3688
3689 } // anonymous namespace
3690
3691
3692 docstring Paragraph::simpleLyXHTMLOnePar(Buffer const & buf,
3693                                     XMLStream & xs,
3694                                     OutputParams const & runparams,
3695                                     Font const & outerfont,
3696                                     bool start_paragraph, bool close_paragraph,
3697                                     pos_type initial) const
3698 {
3699         docstring retval;
3700
3701         // track whether we have opened these tags
3702         bool emph_flag = false;
3703         bool bold_flag = false;
3704         bool noun_flag = false;
3705         bool ubar_flag = false;
3706         bool dbar_flag = false;
3707         bool sout_flag = false;
3708         bool xout_flag = false;
3709         bool wave_flag = false;
3710         // shape tags
3711         bool shap_flag = false;
3712         // family tags
3713         bool faml_flag = false;
3714         // size tags
3715         bool size_flag = false;
3716
3717         Layout const & style = *d->layout_;
3718
3719         if (start_paragraph)
3720                 xs.startDivision(allowEmpty());
3721
3722         FontInfo font_old =
3723                 style.labeltype == LABEL_MANUAL ? style.labelfont : style.font;
3724
3725         FontShape  curr_fs   = INHERIT_SHAPE;
3726         FontFamily curr_fam  = INHERIT_FAMILY;
3727         FontSize   curr_size = INHERIT_SIZE;
3728
3729         string const default_family =
3730                 buf.masterBuffer()->params().fonts_default_family;
3731
3732         vector<xml::FontTag> tagsToOpen;
3733         vector<xml::EndFontTag> tagsToClose;
3734
3735         // parsing main loop
3736         for (pos_type i = initial; i < size(); ++i) {
3737                 // let's not show deleted material in the output
3738                 if (isDeleted(i))
3739                         continue;
3740
3741                 Font const font = getFont(buf.masterBuffer()->params(), i, outerfont);
3742
3743                 // emphasis
3744                 FontState curstate = font.fontInfo().emph();
3745                 if (font_old.emph() != curstate)
3746                         doFontSwitchXHTML(tagsToOpen, tagsToClose, emph_flag, curstate, xml::FT_EMPH);
3747
3748                 // noun
3749                 curstate = font.fontInfo().noun();
3750                 if (font_old.noun() != curstate)
3751                         doFontSwitchXHTML(tagsToOpen, tagsToClose, noun_flag, curstate, xml::FT_NOUN);
3752
3753                 // underbar
3754                 curstate = font.fontInfo().underbar();
3755                 if (font_old.underbar() != curstate)
3756                         doFontSwitchXHTML(tagsToOpen, tagsToClose, ubar_flag, curstate, xml::FT_UBAR);
3757
3758                 // strikeout
3759                 curstate = font.fontInfo().strikeout();
3760                 if (font_old.strikeout() != curstate)
3761                         doFontSwitchXHTML(tagsToOpen, tagsToClose, sout_flag, curstate, xml::FT_SOUT);
3762
3763                 // xout
3764                 curstate = font.fontInfo().xout();
3765                 if (font_old.xout() != curstate)
3766                         doFontSwitchXHTML(tagsToOpen, tagsToClose, xout_flag, curstate, xml::FT_XOUT);
3767
3768                 // double underbar
3769                 curstate = font.fontInfo().uuline();
3770                 if (font_old.uuline() != curstate)
3771                         doFontSwitchXHTML(tagsToOpen, tagsToClose, dbar_flag, curstate, xml::FT_DBAR);
3772
3773                 // wavy line
3774                 curstate = font.fontInfo().uwave();
3775                 if (font_old.uwave() != curstate)
3776                         doFontSwitchXHTML(tagsToOpen, tagsToClose, wave_flag, curstate, xml::FT_WAVE);
3777
3778                 // bold
3779                 // a little hackish, but allows us to reuse what we have.
3780                 curstate = (font.fontInfo().series() == BOLD_SERIES ? FONT_ON : FONT_OFF);
3781                 if (font_old.series() != font.fontInfo().series())
3782                         doFontSwitchXHTML(tagsToOpen, tagsToClose, bold_flag, curstate, xml::FT_BOLD);
3783
3784                 // Font shape
3785                 curr_fs = font.fontInfo().shape();
3786                 FontShape old_fs = font_old.shape();
3787                 if (old_fs != curr_fs) {
3788                         if (shap_flag) {
3789                                 switch (old_fs) {
3790                                 case ITALIC_SHAPE:
3791                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_ITALIC));
3792                                         break;
3793                                 case SLANTED_SHAPE:
3794                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SLANTED));
3795                                         break;
3796                                 case SMALLCAPS_SHAPE:
3797                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SMALLCAPS));
3798                                         break;
3799                                 case UP_SHAPE:
3800                                 case INHERIT_SHAPE:
3801                                         break;
3802                                 default:
3803                                         // the other tags are for internal use
3804                                         LATTEST(false);
3805                                         break;
3806                                 }
3807                                 shap_flag = false;
3808                         }
3809                         switch (curr_fs) {
3810                         case ITALIC_SHAPE:
3811                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_ITALIC));
3812                                 shap_flag = true;
3813                                 break;
3814                         case SLANTED_SHAPE:
3815                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SLANTED));
3816                                 shap_flag = true;
3817                                 break;
3818                         case SMALLCAPS_SHAPE:
3819                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SMALLCAPS));
3820                                 shap_flag = true;
3821                                 break;
3822                         case UP_SHAPE:
3823                         case INHERIT_SHAPE:
3824                                 break;
3825                         default:
3826                                 // the other tags are for internal use
3827                                 LATTEST(false);
3828                                 break;
3829                         }
3830                 }
3831
3832                 // Font family
3833                 curr_fam = font.fontInfo().family();
3834                 FontFamily old_fam = font_old.family();
3835                 if (old_fam != curr_fam) {
3836                         if (faml_flag) {
3837                                 switch (old_fam) {
3838                                 case ROMAN_FAMILY:
3839                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_ROMAN));
3840                                         break;
3841                                 case SANS_FAMILY:
3842                                     tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SANS));
3843                                     break;
3844                                 case TYPEWRITER_FAMILY:
3845                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_TYPE));
3846                                         break;
3847                                 case INHERIT_FAMILY:
3848                                         break;
3849                                 default:
3850                                         // the other tags are for internal use
3851                                         LATTEST(false);
3852                                         break;
3853                                 }
3854                                 faml_flag = false;
3855                         }
3856                         switch (curr_fam) {
3857                         case ROMAN_FAMILY:
3858                                 // we will treat a "default" font family as roman, since we have
3859                                 // no other idea what to do.
3860                                 if (default_family != "rmdefault" && default_family != "default") {
3861                                         tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_ROMAN));
3862                                         faml_flag = true;
3863                                 }
3864                                 break;
3865                         case SANS_FAMILY:
3866                                 if (default_family != "sfdefault") {
3867                                         tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SANS));
3868                                         faml_flag = true;
3869                                 }
3870                                 break;
3871                         case TYPEWRITER_FAMILY:
3872                                 if (default_family != "ttdefault") {
3873                                         tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_TYPE));
3874                                         faml_flag = true;
3875                                 }
3876                                 break;
3877                         case INHERIT_FAMILY:
3878                                 break;
3879                         default:
3880                                 // the other tags are for internal use
3881                                 LATTEST(false);
3882                                 break;
3883                         }
3884                 }
3885
3886                 // Font size
3887                 curr_size = font.fontInfo().size();
3888                 FontSize old_size = font_old.size();
3889                 if (old_size != curr_size) {
3890                         if (size_flag) {
3891                                 switch (old_size) {
3892                                 case TINY_SIZE:
3893                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_TINY));
3894                                         break;
3895                                 case SCRIPT_SIZE:
3896                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_SCRIPT));
3897                                         break;
3898                                 case FOOTNOTE_SIZE:
3899                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_FOOTNOTE));
3900                                         break;
3901                                 case SMALL_SIZE:
3902                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_SMALL));
3903                                         break;
3904                                 case LARGE_SIZE:
3905                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_LARGE));
3906                                         break;
3907                                 case LARGER_SIZE:
3908                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_LARGER));
3909                                         break;
3910                                 case LARGEST_SIZE:
3911                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_LARGEST));
3912                                         break;
3913                                 case HUGE_SIZE:
3914                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_HUGE));
3915                                         break;
3916                                 case HUGER_SIZE:
3917                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_HUGER));
3918                                         break;
3919                                 case INCREASE_SIZE:
3920                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_INCREASE));
3921                                         break;
3922                                 case DECREASE_SIZE:
3923                                         tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_DECREASE));
3924                                         break;
3925                                 case INHERIT_SIZE:
3926                                 case NORMAL_SIZE:
3927                                         break;
3928                                 default:
3929                                         // the other tags are for internal use
3930                                         LATTEST(false);
3931                                         break;
3932                                 }
3933                                 size_flag = false;
3934                         }
3935                         switch (curr_size) {
3936                         case TINY_SIZE:
3937                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_TINY));
3938                                 size_flag = true;
3939                                 break;
3940                         case SCRIPT_SIZE:
3941                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_SCRIPT));
3942                                 size_flag = true;
3943                                 break;
3944                         case FOOTNOTE_SIZE:
3945                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_FOOTNOTE));
3946                                 size_flag = true;
3947                                 break;
3948                         case SMALL_SIZE:
3949                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_SMALL));
3950                                 size_flag = true;
3951                                 break;
3952                         case LARGE_SIZE:
3953                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_LARGE));
3954                                 size_flag = true;
3955                                 break;
3956                         case LARGER_SIZE:
3957                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_LARGER));
3958                                 size_flag = true;
3959                                 break;
3960                         case LARGEST_SIZE:
3961                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_LARGEST));
3962                                 size_flag = true;
3963                                 break;
3964                         case HUGE_SIZE:
3965                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_HUGE));
3966                                 size_flag = true;
3967                                 break;
3968                         case HUGER_SIZE:
3969                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_HUGER));
3970                                 size_flag = true;
3971                                 break;
3972                         case INCREASE_SIZE:
3973                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_INCREASE));
3974                                 size_flag = true;
3975                                 break;
3976                         case DECREASE_SIZE:
3977                                 tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_DECREASE));
3978                                 size_flag = true;
3979                                 break;
3980                         case INHERIT_SIZE:
3981                         case NORMAL_SIZE:
3982                                 break;
3983                         default:
3984                                 // the other tags are for internal use
3985                                 LATTEST(false);
3986                                 break;
3987                         }
3988                 }
3989
3990                 // FIXME XHTML
3991                 // Other such tags? What about the other text ranges?
3992
3993                 vector<xml::EndFontTag>::const_iterator cit = tagsToClose.begin();
3994                 vector<xml::EndFontTag>::const_iterator cen = tagsToClose.end();
3995                 for (; cit != cen; ++cit)
3996                         xs << *cit;
3997
3998                 vector<xml::FontTag>::const_iterator sit = tagsToOpen.begin();
3999                 vector<xml::FontTag>::const_iterator sen = tagsToOpen.end();
4000                 for (; sit != sen; ++sit)
4001                         xs << *sit;
4002
4003                 tagsToClose.clear();
4004                 tagsToOpen.clear();
4005
4006                 Inset const * inset = getInset(i);
4007                 if (inset) {
4008                         if (!runparams.for_toc || inset->isInToc()) {
4009                                 OutputParams np = runparams;
4010                                 np.local_font = &font;
4011                                 // If the paragraph has size 1, then we are in the "special
4012                                 // case" where we do not output the containing paragraph info
4013                                 if (!inset->getLayout().htmlisblock() && size() != 1)
4014                                         np.html_in_par = true;
4015                                 retval += inset->xhtml(xs, np);
4016                         }
4017                 } else {
4018                         char_type c = getUChar(buf.masterBuffer()->params(),
4019                                                runparams, i);
4020                         if (c == ' ' && (style.free_spacing || runparams.free_spacing))
4021                                 xs << XMLStream::ESCAPE_NONE << "&nbsp;";
4022                         else if (c == '\'')
4023                                 xs << XMLStream::ESCAPE_NONE << "&#8217;";
4024                         else
4025                                 xs << c;
4026                 }
4027                 font_old = font.fontInfo();
4028         }
4029
4030         // FIXME XHTML
4031         // I'm worried about what happens if a branch, say, is itself
4032         // wrapped in some font stuff. I think that will not work.
4033         xs.closeFontTags();
4034         if (close_paragraph)
4035                 xs.endDivision();
4036
4037         return retval;
4038 }
4039
4040
4041 bool Paragraph::isHfill(pos_type pos) const
4042 {
4043         Inset const * inset = getInset(pos);
4044         return inset && inset->isHfill();
4045 }
4046
4047
4048 bool Paragraph::isNewline(pos_type pos) const
4049 {
4050         // U+2028 LINE SEPARATOR
4051         // U+2029 PARAGRAPH SEPARATOR
4052         char_type const c = d->text_[pos];
4053         if (c == 0x2028 || c == 0x2029)
4054                 return true;
4055         Inset const * inset = getInset(pos);
4056         return inset && inset->lyxCode() == NEWLINE_CODE;
4057 }
4058
4059
4060 bool Paragraph::isEnvSeparator(pos_type pos) const
4061 {
4062         Inset const * inset = getInset(pos);
4063         return inset && inset->lyxCode() == SEPARATOR_CODE;
4064 }
4065
4066
4067 bool Paragraph::isLineSeparator(pos_type pos) const
4068 {
4069         char_type const c = d->text_[pos];
4070         if (isLineSeparatorChar(c))
4071                 return true;
4072         Inset const * inset = getInset(pos);
4073         return inset && inset->isLineSeparator();
4074 }
4075
4076
4077 bool Paragraph::isWordSeparator(pos_type pos, bool const ignore_deleted) const
4078 {
4079         if (pos == size())
4080                 return true;
4081         if (ignore_deleted && isDeleted(pos))
4082                 return false;
4083         if (Inset const * inset = getInset(pos))
4084                 return !inset->isLetter();
4085         // if we have a hard hyphen (no en- or emdash) or apostrophe
4086         // we pass this to the spell checker
4087         // FIXME: this method is subject to change, visit
4088         // https://bugzilla.mozilla.org/show_bug.cgi?id=355178
4089         // to get an impression how complex this is.
4090         if (isHardHyphenOrApostrophe(pos))
4091                 return false;
4092         char_type const c = d->text_[pos];
4093         // We want to pass the escape chars to the spellchecker
4094         docstring const escape_chars = from_utf8(lyxrc.spellchecker_esc_chars);
4095         return !isLetterChar(c) && !isDigitASCII(c) && !contains(escape_chars, c);
4096 }
4097
4098
4099 bool Paragraph::isHardHyphenOrApostrophe(pos_type pos) const
4100 {
4101         pos_type const psize = size();
4102         if (pos >= psize)
4103                 return false;
4104         char_type const c = d->text_[pos];
4105         if (c != '-' && c != '\'')
4106                 return false;
4107         pos_type nextpos = pos + 1;
4108         pos_type prevpos = pos > 0 ? pos - 1 : 0;
4109         if ((nextpos == psize || isSpace(nextpos))
4110                 && (pos == 0 || isSpace(prevpos)))
4111                 return false;
4112         return true;
4113 }
4114
4115
4116 bool Paragraph::needsCProtection(bool const fragile) const
4117 {
4118         // first check the layout of the paragraph, but only in insets
4119         InsetText const * textinset = inInset().asInsetText();
4120         bool const maintext = textinset
4121                 ? textinset->text().isMainText()
4122                 : false;
4123
4124         if (!maintext && layout().needcprotect) {
4125                 // Environments need cprotection regardless the content
4126                 if (layout().latextype == LATEX_ENVIRONMENT)
4127                         return true;
4128
4129                 // Commands need cprotection if they contain specific chars
4130                 int const nchars_escape = 9;
4131                 static char_type const chars_escape[nchars_escape] = {
4132                         '&', '_', '$', '%', '#', '^', '{', '}', '\\'};
4133
4134                 docstring const pars = asString();
4135                 for (int k = 0; k < nchars_escape; k++) {
4136                         if (contains(pars, chars_escape[k]))
4137                                 return true;
4138                 }
4139         }
4140
4141         // now check whether we have insets that need cprotection
4142         for (auto const & icit : d->insetlist_) {
4143                 Inset const * ins = icit.inset;
4144                 if (!ins)
4145                         continue;
4146                 if (ins->needsCProtection(maintext, fragile))
4147                         return true;
4148         }
4149
4150         return false;
4151 }
4152
4153
4154 FontSpan const & Paragraph::getSpellRange(pos_type pos) const
4155 {
4156         return d->speller_state_.getRange(pos);
4157 }
4158
4159
4160 bool Paragraph::isChar(pos_type pos) const
4161 {
4162         if (Inset const * inset = getInset(pos))
4163                 return inset->isChar();
4164         char_type const c = d->text_[pos];
4165         return !isLetterChar(c) && !isDigitASCII(c) && !lyx::isSpace(c);
4166 }
4167
4168
4169 bool Paragraph::isSpace(pos_type pos) const
4170 {
4171         if (Inset const * inset = getInset(pos))
4172                 return inset->isSpace();
4173         char_type const c = d->text_[pos];
4174         return lyx::isSpace(c);
4175 }
4176
4177
4178 Language const *
4179 Paragraph::getParLanguage(BufferParams const & bparams) const
4180 {
4181         if (!empty())
4182                 return getFirstFontSettings(bparams).language();
4183         // FIXME: we should check the prev par as well (Lgb)
4184         return bparams.language;
4185 }
4186
4187
4188 bool Paragraph::isRTL(BufferParams const & bparams) const
4189 {
4190         return getParLanguage(bparams)->rightToLeft()
4191                 && !inInset().getLayout().forceLTR();
4192 }
4193
4194
4195 void Paragraph::changeLanguage(BufferParams const & bparams,
4196                                Language const * from, Language const * to)
4197 {
4198         // change language including dummy font change at the end
4199         for (pos_type i = 0; i <= size(); ++i) {
4200                 Font font = getFontSettings(bparams, i);
4201                 if (font.language() == from) {
4202                         font.setLanguage(to);
4203                         setFont(i, font);
4204                         d->requestSpellCheck(i);
4205                 }
4206         }
4207 }
4208
4209
4210 bool Paragraph::isMultiLingual(BufferParams const & bparams) const
4211 {
4212         Language const * doc_language = bparams.language;
4213         for (auto const & f : d->fontlist_)
4214                 if (f.font().language() != ignore_language &&
4215                     f.font().language() != latex_language &&
4216                     f.font().language() != doc_language)
4217                         return true;
4218         return false;
4219 }
4220
4221
4222 void Paragraph::getLanguages(std::set<Language const *> & langs) const
4223 {
4224         for (auto const & f : d->fontlist_) {
4225                 Language const * lang = f.font().language();
4226                 if (lang != ignore_language &&
4227                     lang != latex_language)
4228                         langs.insert(lang);
4229         }
4230 }
4231
4232
4233 docstring Paragraph::asString(int options) const
4234 {
4235         return asString(0, size(), options);
4236 }
4237
4238
4239 docstring Paragraph::asString(pos_type beg, pos_type end, int options, const OutputParams *runparams) const
4240 {
4241         odocstringstream os;
4242
4243         if (beg == 0
4244             && options & AS_STR_LABEL
4245             && !d->params_.labelString().empty())
4246                 os << d->params_.labelString() << ' ';
4247
4248         for (pos_type i = beg; i < end; ++i) {
4249                 if ((options & AS_STR_SKIPDELETE) && isDeleted(i))
4250                         continue;
4251                 char_type const c = d->text_[i];
4252                 if (isPrintable(c) || c == '\t'
4253                     || (c == '\n' && (options & AS_STR_NEWLINES)))
4254                         os.put(c);
4255                 else if (c == META_INSET && (options & AS_STR_INSETS)) {
4256                         if (c == META_INSET && (options & AS_STR_PLAINTEXT)) {
4257                                 LASSERT(runparams != nullptr, return docstring());
4258                                 getInset(i)->plaintext(os, *runparams);
4259                         } else if (c == META_INSET && (options & AS_STR_MATHED)
4260                                    && getInset(i)->lyxCode() == REF_CODE) {
4261                                 Buffer const & buf = getInset(i)->buffer();
4262                                 OutputParams rp(&buf.params().encoding());
4263                                 Font const font(inherit_font, buf.params().language);
4264                                 rp.local_font = &font;
4265                                 otexstream ots(os);
4266                                 getInset(i)->latex(ots, rp);
4267                         } else {
4268                                 getInset(i)->toString(os);
4269                         }
4270                 }
4271         }
4272
4273         return os.str();
4274 }
4275
4276
4277 void Paragraph::forOutliner(docstring & os, size_t const maxlen,
4278                             bool const shorten, bool const label) const
4279 {
4280         size_t tmplen = shorten ? maxlen + 1 : maxlen;
4281         if (label && !labelString().empty())
4282                 os += labelString() + ' ';
4283         if (!layout().isTocCaption())
4284                 return;
4285         for (pos_type i = 0; i < size() && os.length() < tmplen; ++i) {
4286                 if (isDeleted(i))
4287                         continue;
4288                 char_type const c = d->text_[i];
4289                 if (isPrintable(c))
4290                         os += c;
4291                 else if (c == META_INSET)
4292                         getInset(i)->forOutliner(os, tmplen, false);
4293         }
4294         if (shorten)
4295                 Text::shortenForOutliner(os, maxlen);
4296 }
4297
4298
4299 void Paragraph::setInsetOwner(Inset const * inset)
4300 {
4301         d->inset_owner_ = inset;
4302 }
4303
4304
4305 int Paragraph::id() const
4306 {
4307         return d->id_;
4308 }
4309
4310
4311 void Paragraph::setId(int id)
4312 {
4313         d->id_ = id;
4314 }
4315
4316
4317 Layout const & Paragraph::layout() const
4318 {
4319         return *d->layout_;
4320 }
4321
4322
4323 void Paragraph::setLayout(Layout const & layout)
4324 {
4325         d->layout_ = &layout;
4326 }
4327
4328
4329 void Paragraph::setDefaultLayout(DocumentClass const & tc)
4330 {
4331         setLayout(tc.defaultLayout());
4332 }
4333
4334
4335 void Paragraph::setPlainLayout(DocumentClass const & tc)
4336 {
4337         setLayout(tc.plainLayout());
4338 }
4339
4340
4341 void Paragraph::setPlainOrDefaultLayout(DocumentClass const & tc)
4342 {
4343         if (usePlainLayout())
4344                 setPlainLayout(tc);
4345         else
4346                 setDefaultLayout(tc);
4347 }
4348
4349
4350 Inset const & Paragraph::inInset() const
4351 {
4352         LBUFERR(d->inset_owner_);
4353         return *d->inset_owner_;
4354 }
4355
4356
4357 ParagraphParameters & Paragraph::params()
4358 {
4359         return d->params_;
4360 }
4361
4362
4363 ParagraphParameters const & Paragraph::params() const
4364 {
4365         return d->params_;
4366 }
4367
4368
4369 bool Paragraph::isFreeSpacing() const
4370 {
4371         if (d->layout_->free_spacing)
4372                 return true;
4373         return d->inset_owner_ && d->inset_owner_->isFreeSpacing();
4374 }
4375
4376
4377 bool Paragraph::allowEmpty() const
4378 {
4379         if (d->layout_->keepempty)
4380                 return true;
4381         return d->inset_owner_ && d->inset_owner_->allowEmpty();
4382 }
4383
4384
4385 bool Paragraph::brokenBiblio() const
4386 {
4387         // There is a problem if there is no bibitem at position 0 in
4388         // paragraphs that need one, if there is another bibitem in the
4389         // paragraph or if this paragraph is not supposed to have
4390         // a bibitem inset at all.
4391         return ((d->layout_->labeltype == LABEL_BIBLIO
4392                 && (d->insetlist_.find(BIBITEM_CODE) != 0
4393                     || d->insetlist_.find(BIBITEM_CODE, 1) > 0))
4394                 || (d->layout_->labeltype != LABEL_BIBLIO
4395                     && d->insetlist_.find(BIBITEM_CODE) != -1));
4396 }
4397
4398
4399 int Paragraph::fixBiblio(Buffer const & buffer)
4400 {
4401         // FIXME: when there was already an inset at 0, the return value is 1,
4402         // which does not tell whether another inset has been removed; the
4403         // cursor cannot be correctly updated.
4404
4405         bool const track_changes = buffer.params().track_changes;
4406         int bibitem_pos = d->insetlist_.find(BIBITEM_CODE);
4407
4408         // The case where paragraph is not BIBLIO
4409         if (d->layout_->labeltype != LABEL_BIBLIO) {
4410                 if (bibitem_pos == -1)
4411                         // No InsetBibitem => OK
4412                         return 0;
4413                 // There is an InsetBibitem: remove it!
4414                 d->insetlist_.release(bibitem_pos);
4415                 eraseChar(bibitem_pos, track_changes);
4416                 return (bibitem_pos == 0) ? -1 : -bibitem_pos;
4417         }
4418
4419         bool const hasbibitem0 = bibitem_pos == 0;
4420         if (hasbibitem0) {
4421                 bibitem_pos = d->insetlist_.find(BIBITEM_CODE, 1);
4422                 // There was an InsetBibitem at pos 0,
4423                 // and no other one => OK
4424                 if (bibitem_pos == -1)
4425                         return 0;
4426                 // there is a bibitem at the 0 position, but since
4427                 // there is a second one, we copy the second on the
4428                 // first. We're assuming there are at most two of
4429                 // these, which there should be.
4430                 // FIXME: why does it make sense to do that rather
4431                 // than keep the first? (JMarc)
4432                 Inset * inset = releaseInset(bibitem_pos);
4433                 d->insetlist_.begin()->inset = inset;
4434                 // This needs to be done to update the counter (#8499)
4435                 buffer.updateBuffer();
4436                 return -bibitem_pos;
4437         }
4438
4439         // We need to create an inset at the beginning
4440         Inset * inset = nullptr;
4441         if (bibitem_pos > 0) {
4442                 // there was one somewhere in the paragraph, let's move it
4443                 inset = d->insetlist_.release(bibitem_pos);
4444                 eraseChar(bibitem_pos, track_changes);
4445         } else
4446                 // make a fresh one
4447                 inset = new InsetBibitem(const_cast<Buffer *>(&buffer),
4448                                          InsetCommandParams(BIBITEM_CODE));
4449
4450         Font font(inherit_font, buffer.params().language);
4451         insertInset(0, inset, font, Change(track_changes ? Change::INSERTED
4452                                                    : Change::UNCHANGED));
4453
4454         // This is needed to get the counters right
4455         buffer.updateBuffer();
4456         return 1;
4457 }
4458
4459
4460 void Paragraph::checkAuthors(AuthorList const & authorList)
4461 {
4462         d->changes_.checkAuthors(authorList);
4463 }
4464
4465
4466 bool Paragraph::isChanged(pos_type pos) const
4467 {
4468         return lookupChange(pos).changed();
4469 }
4470
4471
4472 bool Paragraph::isInserted(pos_type pos) const
4473 {
4474         return lookupChange(pos).inserted();
4475 }
4476
4477
4478 bool Paragraph::isDeleted(pos_type pos) const
4479 {
4480         return lookupChange(pos).deleted();
4481 }
4482
4483
4484 InsetList const & Paragraph::insetList() const
4485 {
4486         return d->insetlist_;
4487 }
4488
4489
4490 void Paragraph::setInsetBuffers(Buffer & b)
4491 {
4492         d->insetlist_.setBuffer(b);
4493 }
4494
4495
4496 void Paragraph::resetBuffer()
4497 {
4498         d->insetlist_.resetBuffer();
4499 }
4500
4501
4502 Inset * Paragraph::releaseInset(pos_type pos)
4503 {
4504         Inset * inset = d->insetlist_.release(pos);
4505         /// does not honour change tracking!
4506         eraseChar(pos, false);
4507         return inset;
4508 }
4509
4510
4511 Inset * Paragraph::getInset(pos_type pos)
4512 {
4513         return (pos < pos_type(d->text_.size()) && d->text_[pos] == META_INSET)
4514                  ? d->insetlist_.get(pos) : nullptr;
4515 }
4516
4517
4518 Inset const * Paragraph::getInset(pos_type pos) const
4519 {
4520         return (pos < pos_type(d->text_.size()) && d->text_[pos] == META_INSET)
4521                  ? d->insetlist_.get(pos) : nullptr;
4522 }
4523
4524
4525 void Paragraph::changeCase(BufferParams const & bparams, pos_type pos,
4526                 pos_type & right, TextCase action)
4527 {
4528         // process sequences of modified characters; in change
4529         // tracking mode, this approach results in much better
4530         // usability than changing case on a char-by-char basis
4531         // We also need to track the current font, since font
4532         // changes within sequences can occur.
4533         vector<pair<char_type, Font> > changes;
4534
4535         bool const trackChanges = bparams.track_changes;
4536
4537         bool capitalize = true;
4538
4539         for (; pos < right; ++pos) {
4540                 char_type oldChar = d->text_[pos];
4541                 char_type newChar = oldChar;
4542
4543                 // ignore insets and don't play with deleted text!
4544                 if (oldChar != META_INSET && !isDeleted(pos)) {
4545                         switch (action) {
4546                                 case text_lowercase:
4547                                         newChar = lowercase(oldChar);
4548                                         break;
4549                                 case text_capitalization:
4550                                         if (capitalize) {
4551                                                 newChar = uppercase(oldChar);
4552                                                 capitalize = false;
4553                                         }
4554                                         break;
4555                                 case text_uppercase:
4556                                         newChar = uppercase(oldChar);
4557                                         break;
4558                         }
4559                 }
4560
4561                 if (isWordSeparator(pos) || isDeleted(pos)) {
4562                         // permit capitalization again
4563                         capitalize = true;
4564                 }
4565
4566                 if (oldChar != newChar) {
4567                         changes.push_back(make_pair(newChar, getFontSettings(bparams, pos)));
4568                         if (pos != right - 1)
4569                                 continue;
4570                         // step behind the changing area
4571                         pos++;
4572                 }
4573
4574                 int erasePos = pos - changes.size();
4575                 for (auto const & change : changes) {
4576                         insertChar(pos, change.first, change.second, trackChanges);
4577                         if (!eraseChar(erasePos, trackChanges)) {
4578                                 ++erasePos;
4579                                 ++pos; // advance
4580                                 ++right; // expand selection
4581                         }
4582                 }
4583                 changes.clear();
4584         }
4585 }
4586
4587 int Paragraph::find(docstring const & str, bool cs, bool mw,
4588                 pos_type start_pos, bool del) const
4589 {
4590         pos_type pos = start_pos;
4591         int const strsize = str.length();
4592         int i = 0;
4593         pos_type const parsize = d->text_.size();
4594         for (i = 0; i < strsize && pos < parsize; ++i, ++pos) {
4595                 // ignore deleted matter
4596                 if (!del && isDeleted(pos)) {
4597                         if (pos == parsize - 1)
4598                                 break;
4599                         pos++;
4600                         --i;
4601                         continue;
4602                 }
4603                 // Ignore "invisible" letters such as ligature breaks
4604                 // and hyphenation chars while searching
4605                 bool nonmatch = false;
4606                 while (pos < parsize && isInset(pos)) {
4607                         Inset const * inset = getInset(pos);
4608                         if (!inset->isLetter() && !inset->isChar())
4609                                 break;
4610                         odocstringstream os;
4611                         if (inset->lyxCode() == lyx::QUOTE_CODE) {
4612                                 OutputParams op(0);
4613                                 op.find_set_feature(OutputParams::SearchQuick);
4614                                 inset->plaintext(os, op);
4615                         }
4616                         else {
4617                                 inset->toString(os);
4618                         }
4619                         docstring const insetstring = os.str();
4620                         if (!insetstring.empty()) {
4621                                 int const insetstringsize = insetstring.length();
4622                                 for (int j = 0; j < insetstringsize && pos < parsize; ++i, ++j) {
4623                                         if ((cs && str[i] != insetstring[j])
4624                                             || (!cs && uppercase(str[i]) != uppercase(insetstring[j]))) {
4625                                                 nonmatch = true;
4626                                                 break;
4627                                         }
4628                                 }
4629                         }
4630                         pos++;
4631                 }
4632                 if (nonmatch || i == strsize)
4633                         break;
4634                 char_type dp = d->text_[pos];
4635                 if (cs && str[i] != dp)
4636                         break;
4637                 if (!cs && uppercase(str[i]) != uppercase(dp))
4638                         break;
4639         }
4640
4641         if (i != strsize)
4642                 return 0;
4643
4644         // if necessary, check whether string matches word
4645         if (mw) {
4646                 if (start_pos > 0 && !isWordSeparator(start_pos - 1))
4647                         return 0;
4648                 if (pos < parsize
4649                         && !isWordSeparator(pos))
4650                         return 0;
4651         }
4652
4653         return pos - start_pos;
4654 }
4655
4656
4657 char_type Paragraph::getChar(pos_type pos) const
4658 {
4659         return d->text_[pos];
4660 }
4661
4662
4663 pos_type Paragraph::size() const
4664 {
4665         return d->text_.size();
4666 }
4667
4668
4669 bool Paragraph::empty() const
4670 {
4671         return d->text_.empty();
4672 }
4673
4674
4675 bool Paragraph::isInset(pos_type pos) const
4676 {
4677         return d->text_[pos] == META_INSET;
4678 }
4679
4680
4681 bool Paragraph::isSeparator(pos_type pos) const
4682 {
4683         //FIXME: Are we sure this can be the only separator?
4684         return d->text_[pos] == ' ';
4685 }
4686
4687
4688 void Paragraph::deregisterWords()
4689 {
4690         Private::LangWordsMap::const_iterator itl = d->words_.begin();
4691         Private::LangWordsMap::const_iterator ite = d->words_.end();
4692         for (; itl != ite; ++itl) {
4693                 WordList & wl = theWordList(itl->first);
4694                 Private::Words::const_iterator it = (itl->second).begin();
4695                 Private::Words::const_iterator et = (itl->second).end();
4696                 for (; it != et; ++it)
4697                         wl.remove(*it);
4698         }
4699         d->words_.clear();
4700 }
4701
4702
4703 void Paragraph::locateWord(pos_type & from, pos_type & to,
4704         word_location const loc, bool const ignore_deleted) const
4705 {
4706         switch (loc) {
4707         case WHOLE_WORD_STRICT:
4708                 if (from == 0 || from == size()
4709                     || isWordSeparator(from, ignore_deleted)
4710                     || isWordSeparator(from - 1, ignore_deleted)) {
4711                         to = from;
4712                         return;
4713                 }
4714                 // fall through
4715
4716         case WHOLE_WORD:
4717                 // If we are already at the beginning of a word, do nothing
4718                 if (!from || isWordSeparator(from - 1, ignore_deleted))
4719                         break;
4720                 // fall through
4721
4722         case PREVIOUS_WORD:
4723                 // always move the cursor to the beginning of previous word
4724                 while (from && !isWordSeparator(from - 1, ignore_deleted))
4725                         --from;
4726                 break;
4727         case NEXT_WORD:
4728                 LYXERR0("Paragraph::locateWord: NEXT_WORD not implemented yet");
4729                 break;
4730         case PARTIAL_WORD:
4731                 // no need to move the 'from' cursor
4732                 break;
4733         }
4734         to = from;
4735         while (to < size() && !isWordSeparator(to, ignore_deleted))
4736                 ++to;
4737 }
4738
4739
4740 void Paragraph::collectWords()
4741 {
4742         for (pos_type pos = 0; pos < size(); ++pos) {
4743                 if (isWordSeparator(pos))
4744                         continue;
4745                 pos_type from = pos;
4746                 locateWord(from, pos, WHOLE_WORD);
4747                 // Work around MSVC warning: The statement
4748                 // if (pos < from + lyxrc.completion_minlength)
4749                 // triggers a signed vs. unsigned warning.
4750                 // I don't know why this happens, it could be a MSVC bug, or
4751                 // related to LLP64 (windows) vs. LP64 (unix) programming
4752                 // model, or the C++ standard might be ambigous in the section
4753                 // defining the "usual arithmetic conversions". However, using
4754                 // a temporary variable is safe and works on all compilers.
4755                 pos_type const endpos = from + lyxrc.completion_minlength;
4756                 if (pos < endpos)
4757                         continue;
4758                 FontList::const_iterator cit = d->fontlist_.fontIterator(from);
4759                 if (cit == d->fontlist_.end())
4760                         return;
4761                 Language const * lang = cit->font().language();
4762                 docstring const word = asString(from, pos, AS_STR_NONE);
4763                 d->words_[lang->lang()].insert(word);
4764         }
4765 }
4766
4767
4768 void Paragraph::registerWords()
4769 {
4770         Private::LangWordsMap::const_iterator itl = d->words_.begin();
4771         Private::LangWordsMap::const_iterator ite = d->words_.end();
4772         for (; itl != ite; ++itl) {
4773                 WordList & wl = theWordList(itl->first);
4774                 Private::Words::const_iterator it = (itl->second).begin();
4775                 Private::Words::const_iterator et = (itl->second).end();
4776                 for (; it != et; ++it)
4777                         wl.insert(*it);
4778         }
4779 }
4780
4781
4782 void Paragraph::updateWords()
4783 {
4784         deregisterWords();
4785         collectWords();
4786         registerWords();
4787 }
4788
4789
4790 void Paragraph::Private::appendSkipPosition(SkipPositions & skips, pos_type const pos) const
4791 {
4792         SkipPositionsIterator begin = skips.begin();
4793         SkipPositions::iterator end = skips.end();
4794         if (pos > 0 && begin < end) {
4795                 --end;
4796                 if (end->last == pos - 1) {
4797                         end->last = pos;
4798                         return;
4799                 }
4800         }
4801         skips.insert(end, FontSpan(pos, pos));
4802 }
4803
4804
4805 Language * Paragraph::Private::locateSpellRange(
4806         pos_type & from, pos_type & to,
4807         SkipPositions & skips) const
4808 {
4809         // skip leading white space
4810         while (from < to && owner_->isWordSeparator(from))
4811                 ++from;
4812         // don't check empty range
4813         if (from >= to)
4814                 return nullptr;
4815         // get current language
4816         Language * lang = getSpellLanguage(from);
4817         pos_type last = from;
4818         bool samelang = true;
4819         bool sameinset = true;
4820         while (last < to && samelang && sameinset) {
4821                 // hop to end of word
4822                 while (last < to && !owner_->isWordSeparator(last)) {
4823                         Inset const * inset = owner_->getInset(last);
4824                         if (inset && dynamic_cast<const InsetSpecialChar *>(inset)) {
4825                                 // check for "invisible" letters such as ligature breaks
4826                                 odocstringstream os;
4827                                 inset->toString(os);
4828                                 if (os.str().length() != 0) {
4829                                         // avoid spell check of visible special char insets
4830                                         // stop the loop in front of the special char inset
4831                                         sameinset = false;
4832                                         break;
4833                                 }
4834                         } else if (inset) {
4835                                 appendSkipPosition(skips, last);
4836                         } else if (owner_->isDeleted(last)) {
4837                                 appendSkipPosition(skips, last);
4838                         }
4839                         ++last;
4840                 }
4841                 // hop to next word while checking for insets
4842                 while (sameinset && last < to && owner_->isWordSeparator(last)) {
4843                         if (Inset const * inset = owner_->getInset(last))
4844                                 sameinset = inset->isChar() && inset->isLetter();
4845                         if (sameinset && owner_->isDeleted(last)) {
4846                                 appendSkipPosition(skips, last);
4847                         }
4848                         if (sameinset)
4849                                 last++;
4850                 }
4851                 if (sameinset && last < to) {
4852                         // now check for language change
4853                         samelang = lang == getSpellLanguage(last);
4854                 }
4855         }
4856         // if language change detected backstep is needed
4857         if (!samelang)
4858                 --last;
4859         to = last;
4860         return lang;
4861 }
4862
4863
4864 Language * Paragraph::Private::getSpellLanguage(pos_type const from) const
4865 {
4866         Language * lang =
4867                 const_cast<Language *>(owner_->getFontSettings(
4868                         inset_owner_->buffer().params(), from).language());
4869         if (lang == inset_owner_->buffer().params().language
4870                 && !lyxrc.spellchecker_alt_lang.empty()) {
4871                 string lang_code;
4872                 string const lang_variety =
4873                         split(lyxrc.spellchecker_alt_lang, lang_code, '-');
4874                 lang->setCode(lang_code);
4875                 lang->setVariety(lang_variety);
4876         }
4877         return lang;
4878 }
4879
4880
4881 void Paragraph::requestSpellCheck(pos_type pos)
4882 {
4883         d->requestSpellCheck(pos);
4884         if (pos == -1) {
4885                 // Also request spellcheck within (text) insets
4886                 for (auto const & insets : insetList()) {
4887                         if (!insets.inset->asInsetText())
4888                                 continue;
4889                         ParagraphList & inset_pars =
4890                                 insets.inset->asInsetText()->paragraphs();
4891                         ParagraphList::iterator pit = inset_pars.begin();
4892                         ParagraphList::iterator pend = inset_pars.end();
4893                         for (; pit != pend; ++pit)
4894                                 pit->requestSpellCheck();
4895                 }
4896         }
4897 }
4898
4899
4900 bool Paragraph::needsSpellCheck() const
4901 {
4902         SpellChecker::ChangeNumber speller_change_number = 0;
4903         if (theSpellChecker())
4904                 speller_change_number = theSpellChecker()->changeNumber();
4905         if (speller_change_number > d->speller_state_.currentChangeNumber()) {
4906                 d->speller_state_.needsCompleteRefresh(speller_change_number);
4907         }
4908         return d->needsSpellCheck();
4909 }
4910
4911
4912 bool Paragraph::Private::ignoreWord(docstring const & word) const
4913 {
4914         // Ignore words with digits
4915         // FIXME: make this customizable
4916         // (note that some checkers ignore words with digits by default)
4917         docstring::const_iterator cit = word.begin();
4918         docstring::const_iterator const end = word.end();
4919         for (; cit != end; ++cit) {
4920                 if (isNumber((*cit)))
4921                         return true;
4922         }
4923         return false;
4924 }
4925
4926
4927 SpellChecker::Result Paragraph::spellCheck(pos_type & from, pos_type & to,
4928         WordLangTuple & wl, docstring_list & suggestions,
4929         bool do_suggestion, bool check_learned) const
4930 {
4931         SpellChecker::Result result = SpellChecker::WORD_OK;
4932         SpellChecker * speller = theSpellChecker();
4933         if (!speller)
4934                 return result;
4935
4936         if (!d->layout_->spellcheck || !inInset().allowSpellCheck())
4937                 return result;
4938
4939         locateWord(from, to, WHOLE_WORD, true);
4940         if (from == to || from >= size())
4941                 return result;
4942
4943         docstring word = asString(from, to, AS_STR_INSETS | AS_STR_SKIPDELETE);
4944         Language * lang = d->getSpellLanguage(from);
4945
4946         BufferParams const & bparams = d->inset_owner_->buffer().params();
4947
4948         if (getFontSettings(bparams, from).fontInfo().nospellcheck() == FONT_ON)
4949                 return result;
4950
4951         wl = WordLangTuple(word, lang);
4952
4953         if (word.empty())
4954                 return result;
4955
4956         if (needsSpellCheck() || check_learned) {
4957                 pos_type end = to;
4958                 if (!d->ignoreWord(word)) {
4959                         bool const trailing_dot = to < size() && d->text_[to] == '.';
4960                         result = speller->check(wl, bparams.spellignore());
4961                         if (SpellChecker::misspelled(result) && trailing_dot) {
4962                                 wl = WordLangTuple(word.append(from_ascii(".")), lang);
4963                                 result = speller->check(wl, bparams.spellignore());
4964                                 if (!SpellChecker::misspelled(result)) {
4965                                         LYXERR(Debug::GUI, "misspelled word is correct with dot: \"" <<
4966                                            word << "\" [" <<
4967                                            from << ".." << to << "]");
4968                                 } else {
4969                                         // spell check with dot appended failed too
4970                                         // restore original word/lang value
4971                                         word = asString(from, to, AS_STR_INSETS | AS_STR_SKIPDELETE);
4972                                         wl = WordLangTuple(word, lang);
4973                                 }
4974                         }
4975                 }
4976                 if (!SpellChecker::misspelled(result)) {
4977                         // area up to the begin of the next word is not misspelled
4978                         while (end < size() && isWordSeparator(end))
4979                                 ++end;
4980                 }
4981                 d->setMisspelled(from, end, result);
4982         } else {
4983                 result = d->speller_state_.getState(from);
4984         }
4985
4986         if (do_suggestion)
4987                 suggestions.clear();
4988
4989         if (SpellChecker::misspelled(result)) {
4990                 LYXERR(Debug::GUI, "misspelled word: \"" <<
4991                            word << "\" [" <<
4992                            from << ".." << to << "]");
4993                 if (do_suggestion)
4994                         speller->suggest(wl, suggestions);
4995         }
4996         return result;
4997 }
4998
4999
5000 void Paragraph::anonymize()
5001 {
5002         // This is a very crude anonymization for now
5003         for (char_type & c : d->text_)
5004                 if (isLetterChar(c) || isNumber(c))
5005                         c = 'a';
5006 }
5007
5008
5009 void Paragraph::Private::markMisspelledWords(
5010         Language const * lang,
5011         pos_type const & first, pos_type const & last,
5012         SpellChecker::Result result,
5013         docstring const & word,
5014         SkipPositions const & skips)
5015 {
5016         if (!SpellChecker::misspelled(result)) {
5017                 setMisspelled(first, last, SpellChecker::WORD_OK);
5018                 return;
5019         }
5020         int snext = first;
5021         SpellChecker * speller = theSpellChecker();
5022         // locate and enumerate the error positions
5023         int nerrors = speller->numMisspelledWords();
5024         int numskipped = 0;
5025         SkipPositionsIterator it = skips.begin();
5026         SkipPositionsIterator et = skips.end();
5027         for (int index = 0; index < nerrors; ++index) {
5028                 int wstart;
5029                 int wlen = 0;
5030                 speller->misspelledWord(index, wstart, wlen);
5031                 /// should not happen if speller supports range checks
5032                 if (!wlen)
5033                         continue;
5034                 WordLangTuple const candidate(word.substr(wstart, wlen), lang);
5035                 wstart += first + numskipped;
5036                 if (snext < wstart) {
5037                         /// mark the range of correct spelling
5038                         numskipped += countSkips(it, et, wstart);
5039                         setMisspelled(snext,
5040                                 wstart - 1, SpellChecker::WORD_OK);
5041                 }
5042                 snext = wstart + wlen;
5043                 // Check whether the candidate is in the document's local dict
5044                 SpellChecker::Result actresult = result;
5045                 if (inset_owner_->buffer().params().spellignored(candidate))
5046                         actresult = SpellChecker::DOCUMENT_LEARNED_WORD;
5047                 numskipped += countSkips(it, et, snext);
5048                 /// mark the range of misspelling
5049                 setMisspelled(wstart, snext, actresult);
5050                 if (actresult == SpellChecker::DOCUMENT_LEARNED_WORD)
5051                         LYXERR(Debug::GUI, "local dictionary word: \"" <<
5052                                    candidate.word() << "\" [" <<
5053                                    wstart << ".." << (snext-1) << "]");
5054                 else
5055                         LYXERR(Debug::GUI, "misspelled word: \"" <<
5056                                    candidate.word() << "\" [" <<
5057                                    wstart << ".." << (snext-1) << "]");
5058                 ++snext;
5059         }
5060         if (snext <= last) {
5061                 /// mark the range of correct spelling at end
5062                 setMisspelled(snext, last, SpellChecker::WORD_OK);
5063         }
5064 }
5065
5066
5067 void Paragraph::spellCheck() const
5068 {
5069         SpellChecker * speller = theSpellChecker();
5070         if (!speller || empty() ||!needsSpellCheck())
5071                 return;
5072         pos_type start;
5073         pos_type endpos;
5074         d->rangeOfSpellCheck(start, endpos);
5075         if (speller->canCheckParagraph()) {
5076                 // loop until we leave the range
5077                 for (pos_type first = start; first < endpos; ) {
5078                         pos_type last = endpos;
5079                         Private::SkipPositions skips;
5080                         Language * lang = d->locateSpellRange(first, last, skips);
5081                         if (first >= endpos)
5082                                 break;
5083                         // start the spell checker on the unit of meaning
5084                         docstring word = asString(first, last, AS_STR_INSETS + AS_STR_SKIPDELETE);
5085                         WordLangTuple wl = WordLangTuple(word, lang);
5086                         BufferParams const & bparams = d->inset_owner_->buffer().params();
5087                         SpellChecker::Result result = !word.empty() ?
5088                                 speller->check(wl, bparams.spellignore()) : SpellChecker::WORD_OK;
5089                         d->markMisspelledWords(lang, first, last, result, word, skips);
5090                         first = ++last;
5091                 }
5092         } else {
5093                 static docstring_list suggestions;
5094                 pos_type to = endpos;
5095                 while (start < endpos) {
5096                         WordLangTuple wl;
5097                         spellCheck(start, to, wl, suggestions, false);
5098                         start = to + 1;
5099                 }
5100         }
5101         d->readySpellCheck();
5102 }
5103
5104
5105 bool Paragraph::isMisspelled(pos_type pos, bool check_boundary) const
5106 {
5107         bool result = SpellChecker::misspelled(d->speller_state_.getState(pos));
5108         if (result || pos <= 0 || pos > size())
5109                 return result;
5110         if (check_boundary && (pos == size() || isWordSeparator(pos)))
5111                 result = SpellChecker::misspelled(d->speller_state_.getState(pos - 1));
5112         return result;
5113 }
5114
5115
5116 string Paragraph::magicLabel() const
5117 {
5118         stringstream ss;
5119         ss << "magicparlabel-" << id();
5120         return ss.str();
5121 }
5122
5123
5124 } // namespace lyx