/**
 * \file Paragraph.cpp
 * This file is part of LyX, the document processor.
 * Licence details can be found in the file COPYING.
 *
 * \author Asger Alstrup
 * \author Lars Gullik Bjønnes
 * \author Richard Kimberly Heck (XHTML output)
 * \author Jean-Marc Lasgouttes
 * \author Angus Leeming
 * \author John Levon
 * \author André Pönitz
 * \author Dekel Tsur
 * \author Jürgen Vigna
 *
 * Full author contact details are available in file CREDITS.
 */

#include <config.h>

#include "Paragraph.h"

#include "Buffer.h"
#include "BufferParams.h"
#include "BufferEncodings.h"
#include "Changes.h"
#include "Counters.h"
#include "Cursor.h"
#include "InsetList.h"
#include "Language.h"
#include "LaTeXFeatures.h"
#include "Layout.h"
#include "Font.h"
#include "FontList.h"
#include "LyXRC.h"
#include "OutputParams.h"
#include "output_latex.h"
#include "output_xhtml.h"
#include "output_docbook.h"
#include "ParagraphParameters.h"
#include "Session.h"
#include "SpellChecker.h"
#include "texstream.h"
#include "TexRow.h"
#include "Text.h"
#include "TextClass.h"
#include "WordLangTuple.h"
#include "WordList.h"

#include "frontends/alert.h"

#include "insets/InsetArgument.h"
#include "insets/InsetBibitem.h"
#include "insets/InsetLabel.h"
#include "insets/InsetSpecialChar.h"
#include "insets/InsetText.h"

#include "mathed/InsetMathHull.h"

#include "support/debug.h"
#include "support/docstring_list.h"
#include "support/ExceptionMessage.h"
#include "support/gettext.h"
#include "support/lassert.h"
#include "support/lstrings.h"
#include "support/textutils.h"

#include <atomic>
#include <sstream>
#include <vector>

using namespace std;
using namespace lyx::support;

// (old?) macOS clang and msvc < 2015 do not support C++11 thread_local
#if defined(__APPLE__)
#define THREAD_LOCAL_STATIC static __thread
#elif defined(_MSC_VER) && (_MSC_VER < 1900)
#define THREAD_LOCAL_STATIC static __declspec(thread)
#else
#define THREAD_LOCAL_STATIC thread_local static
#endif

namespace lyx {

/////////////////////////////////////////////////////////////////////
//
// SpellResultRange
//
/////////////////////////////////////////////////////////////////////

class SpellResultRange {
public:
	SpellResultRange(FontSpan range, SpellChecker::Result result)
	: range_(range), result_(result)
	{}
	///
	FontSpan const & range() const { return range_; }
	///
	void range(FontSpan const & r) { range_ = r; }
	///
	SpellChecker::Result result() const { return result_; }
	///
	void result(SpellChecker::Result r) { result_ = r; }
	///
	bool contains(pos_type pos) const { return range_.contains(pos); }
	///
	bool covered(FontSpan const & r) const
	{
		// 1. first of new range inside current range or
		// 2. last of new range inside current range or
		// 3. first of current range inside new range or
		// 4. last of current range inside new range
		//FIXME: is this the same as !range_.intersect(r).empty() ?
		return range_.contains(r.first) || range_.contains(r.last) ||
			r.contains(range_.first) || r.contains(range_.last);
	}
	///
	void shift(pos_type pos, int offset)
	{
		if (range_.first > pos) {
			range_.first += offset;
			range_.last += offset;
		} else if (range_.last >= pos) {
			range_.last += offset;
		}
	}
private:
	FontSpan range_ ;
	SpellChecker::Result result_ ;
};


/////////////////////////////////////////////////////////////////////
//
// SpellCheckerState
//
/////////////////////////////////////////////////////////////////////

class SpellCheckerState {
public:
	SpellCheckerState()
	{
		needs_refresh_ = true;
		current_change_number_ = 0;
	}

	void setRange(FontSpan const & fp, SpellChecker::Result state)
	{
		Ranges result;
		RangesIterator et = ranges_.end();
		RangesIterator it = ranges_.begin();
		for (; it != et; ++it) {
			if (!it->covered(fp))
				result.push_back(SpellResultRange(it->range(), it->result()));
			else if (state == SpellChecker::WORD_OK) {
				// trim or split the current misspelled range
				// store misspelled ranges only
				FontSpan range = it->range();
				if (fp.first > range.first) {
					// misspelled area in front of WORD_OK
					range.last = fp.first - 1;
					result.push_back(SpellResultRange(range, it->result()));
					range = it->range();
				}
				if (fp.last < range.last) {
					// misspelled area after WORD_OK range
					range.first = fp.last + 1;
					result.push_back(SpellResultRange(range, it->result()));
				}
			}
		}
		ranges_ = result;
		if (state != SpellChecker::WORD_OK)
			ranges_.push_back(SpellResultRange(fp, state));
	}

	void increasePosAfterPos(pos_type pos)
	{
		correctRangesAfterPos(pos, 1);
		needsRefresh(pos);
	}

	void decreasePosAfterPos(pos_type pos)
	{
		correctRangesAfterPos(pos, -1);
		needsRefresh(pos);
	}

	void refreshLast(pos_type pos)
	{
		if (pos < refresh_.last)
			refresh_.last = pos;
	}

	SpellChecker::Result getState(pos_type pos) const
	{
		SpellChecker::Result result = SpellChecker::WORD_OK;
		RangesIterator et = ranges_.end();
		RangesIterator it = ranges_.begin();
		for (; it != et; ++it) {
			if(it->contains(pos)) {
				return it->result();
			}
		}
		return result;
	}

	FontSpan const & getRange(pos_type pos) const
	{
		/// empty span to indicate mismatch
		static FontSpan empty_;
		RangesIterator et = ranges_.end();
		RangesIterator it = ranges_.begin();
		for (; it != et; ++it) {
			if(it->contains(pos)) {
				return it->range();
			}
		}
		return empty_;
	}

	bool needsRefresh() const
	{
		return needs_refresh_;
	}

	SpellChecker::ChangeNumber currentChangeNumber() const
	{
		return current_change_number_;
	}

	void refreshRange(pos_type & first, pos_type & last) const
	{
		first = refresh_.first;
		last = refresh_.last;
	}

	void needsRefresh(pos_type pos)
	{
		if (needs_refresh_ && pos != -1) {
			if (pos < refresh_.first)
				refresh_.first = pos;
			if (pos > refresh_.last)
				refresh_.last = pos;
		} else if (pos != -1) {
			// init request check for neighbour positions too
			refresh_.first = pos > 0 ? pos - 1 : 0;
			// no need for special end of paragraph check
			refresh_.last = pos + 1;
		}
		needs_refresh_ = pos != -1;
	}

	void needsCompleteRefresh(SpellChecker::ChangeNumber change_number)
	{
		needs_refresh_ = true;
		refresh_.first = 0;
		refresh_.last = -1;
		current_change_number_ = change_number;
	}
private:
	typedef vector<SpellResultRange> Ranges;
	typedef Ranges::const_iterator RangesIterator;
	Ranges ranges_;
	/// the area of the paragraph with pending spell check
	FontSpan refresh_;
	/// spell state cache version number
	SpellChecker::ChangeNumber current_change_number_;
	bool needs_refresh_;


	void correctRangesAfterPos(pos_type pos, int offset)
	{
		RangesIterator et = ranges_.end();
		Ranges::iterator it = ranges_.begin();
		for (; it != et; ++it) {
			it->shift(pos, offset);
		}
	}

};

/////////////////////////////////////////////////////////////////////
//
// Paragraph::Private
//
/////////////////////////////////////////////////////////////////////

class Paragraph::Private
{
	// Enforce our own "copy" constructor
	Private(Private const &) = delete;
	Private & operator=(Private const &) = delete;
	// Unique ID generator
	static int make_id();
public:
	///
	Private(Paragraph * owner, Layout const & layout);
	/// "Copy constructor"
	Private(Private const &, Paragraph * owner);
	/// Copy constructor from \p beg  to \p end
	Private(Private const &, Paragraph * owner, pos_type beg, pos_type end);

	///
	void insertChar(pos_type pos, char_type c, Change const & change);

	/// Output the surrogate pair formed by \p c and \p next to \p os.
	/// \return the number of characters written.
	int latexSurrogatePair(BufferParams const &, otexstream & os,
			       char_type c, char_type next,
			       OutputParams const &) const;

	/// Output a space in appropriate formatting (or a surrogate pair
	/// if the next character is a combining character).
	/// \return whether a surrogate pair was output.
	bool simpleTeXBlanks(BufferParams const &,
			     OutputParams const &,
			     otexstream &,
			     pos_type i,
			     unsigned int & column,
			     Font const & font,
			     Layout const & style) const;

	/// This could go to ParagraphParameters if we want to.
	int startTeXParParams(BufferParams const &, otexstream &,
			      OutputParams const &) const;

	/// This could go to ParagraphParameters if we want to.
	bool endTeXParParams(BufferParams const &, otexstream &,
			     OutputParams const &) const;

	///
	void latexInset(BufferParams const &,
				   otexstream &,
				   OutputParams &,
				   Font & running_font,
				   Font & basefont,
				   Font const & outerfont,
				   bool & open_font,
				   Change & running_change,
				   Layout const & style,
				   pos_type & i,
				   unsigned int & column,
				   bool const fontswitch_inset,
				   bool const closeLanguage,
				   bool const lang_switched_at_inset) const;

	///
	void latexSpecialChar(
				   otexstream & os,
				   BufferParams const & bparams,
				   OutputParams const & runparams,
				   Font const & running_font,
				   string & alien_script,
				   Layout const & style,
				   InsetLayout const & il,
				   pos_type & i,
				   pos_type end_pos,
				   unsigned int & column) const;

	///
	bool latexSpecialT1(
		char_type const c,
		otexstream & os,
		pos_type i,
		unsigned int & column) const;
	///
	bool latexSpecialTU(
		char_type const c,
		otexstream & os,
		pos_type i,
		unsigned int & column) const;
	///
	bool latexSpecialT3(
		char_type const c,
		otexstream & os,
		pos_type i,
		unsigned int & column) const;

	///
	void validate(LaTeXFeatures & features) const;

	/// Checks if the paragraph contains only text and no inset or font change.
	bool onlyText(Buffer const & buf, Font const & outerfont,
		      pos_type initial) const;

	/// a vector of speller skip positions
	typedef vector<FontSpan> SkipPositions;
	typedef SkipPositions::const_iterator SkipPositionsIterator;

	void appendSkipPosition(SkipPositions & skips, pos_type const pos) const;

	Language * getSpellLanguage(pos_type const from) const;

	Language * locateSpellRange(pos_type & from, pos_type & to,
				    SkipPositions & skips) const;

	bool hasSpellerChange() const
	{
		SpellChecker::ChangeNumber speller_change_number = 0;
		if (theSpellChecker())
			speller_change_number = theSpellChecker()->changeNumber();
		return speller_change_number > speller_state_.currentChangeNumber();
	}

	bool ignoreWord(docstring const & word) const;

	void setMisspelled(pos_type from, pos_type to, SpellChecker::Result state)
	{
		pos_type textsize = owner_->size();
		// check for sane arguments
		if (to <= from || from >= textsize)
			return;
		FontSpan fp = FontSpan(from, to - 1);
		speller_state_.setRange(fp, state);
	}

	void requestSpellCheck(pos_type pos)
	{
		if (pos == -1)
			speller_state_.needsCompleteRefresh(speller_state_.currentChangeNumber());
		else
			speller_state_.needsRefresh(pos);
	}

	void readySpellCheck()
	{
		speller_state_.needsRefresh(-1);
	}

	bool needsSpellCheck() const
	{
		return speller_state_.needsRefresh();
	}

	void rangeOfSpellCheck(pos_type & first, pos_type & last) const
	{
		speller_state_.refreshRange(first, last);
		if (last == -1) {
			last = owner_->size();
			return;
		}
		pos_type endpos = last;
		owner_->locateWord(first, endpos, WHOLE_WORD, true);
		if (endpos < last) {
			endpos = last;
			owner_->locateWord(last, endpos, WHOLE_WORD, true);
		}
		last = endpos;
	}

	int countSkips(SkipPositionsIterator & it, SkipPositionsIterator const et,
			    int & start) const
	{
		int numskips = 0;
		while (it != et && it->first < start) {
			long skip = it->last - it->first + 1;
			start += skip;
			numskips += skip;
			++it;
		}
		return numskips;
	}

	void markMisspelledWords(Language const * lang,
				 pos_type const & first, pos_type const & last,
				 SpellChecker::Result result,
				 docstring const & word,
				 SkipPositions const & skips);

	InsetCode ownerCode() const
	{
		return inset_owner_ ? inset_owner_->lyxCode() : NO_CODE;
	}

	/// Which Paragraph owns us?
	Paragraph * owner_;

	/// In which Inset?
	Inset const * inset_owner_;

	///
	FontList fontlist_;

	///
	ParagraphParameters params_;

	/// for recording and looking up changes
	Changes changes_;

	///
	InsetList insetlist_;

	/// end of label
	pos_type begin_of_body_;

	typedef docstring TextContainer;
	///
	TextContainer text_;

	typedef set<docstring> Words;
	typedef map<string, Words> LangWordsMap;
	///
	LangWordsMap words_;
	///
	Layout const * layout_;
	///
	SpellCheckerState speller_state_;
	///
	int id_;
};


Paragraph::Private::Private(Paragraph * owner, Layout const & layout)
	: owner_(owner), inset_owner_(nullptr), begin_of_body_(0), layout_(&layout), id_(-1)
{
	text_.reserve(100);
}


//static
int Paragraph::Private::make_id()
{
	// The id is unique per session across buffers because it is used in
	// LFUN_PARAGRAPH_GOTO to switch to a different buffer, for instance in the
	// outliner.
	// (thread-safe)
	static int next_id(0);
	return next_id++;
}


Paragraph::Private::Private(Private const & p, Paragraph * owner)
	: owner_(owner), inset_owner_(p.inset_owner_), fontlist_(p.fontlist_),
	  params_(p.params_), changes_(p.changes_), insetlist_(p.insetlist_),
	  begin_of_body_(p.begin_of_body_), text_(p.text_), words_(p.words_),
	  layout_(p.layout_), id_(make_id())
{
	requestSpellCheck(p.text_.size());
}


Paragraph::Private::Private(Private const & p, Paragraph * owner,
	pos_type beg, pos_type end)
	: owner_(owner), inset_owner_(p.inset_owner_),
	  params_(p.params_), changes_(p.changes_),
	  insetlist_(p.insetlist_, beg, end),
	  begin_of_body_(p.begin_of_body_), words_(p.words_),
	  layout_(p.layout_), id_(make_id())
{
	if (beg >= pos_type(p.text_.size()))
		return;
	text_ = p.text_.substr(beg, end - beg);

	FontList::const_iterator fcit = fontlist_.begin();
	FontList::const_iterator fend = fontlist_.end();
	for (; fcit != fend; ++fcit) {
		if (fcit->pos() < beg)
			continue;
		if (fcit->pos() >= end) {
			// Add last entry in the fontlist_.
			fontlist_.set(text_.size() - 1, fcit->font());
			break;
		}
		// Add a new entry in the fontlist_.
		fontlist_.set(fcit->pos() - beg, fcit->font());
	}
	requestSpellCheck(p.text_.size());
}


/////////////////////////////////////////////////////////////////////
//
// Paragraph
//
/////////////////////////////////////////////////////////////////////

namespace {

/** This helper class should be instantiated at the start of methods
 * that can create or merge changes. If as a result the value of
 * Paragraph::isChanged is modified, it makes sure that updateBuffer()
 * will be run.
 */
struct ChangesMonitor {
	///
	ChangesMonitor(Paragraph & par)
		: par_(par), was_changed_(par.isChanged()) {}
	///
	~ChangesMonitor()
	{
		/* We may need to run updateBuffer to check whether the buffer
		 * contains changes (and toggle the changes toolbar). We do it
		 * when:
		 * 1. the `changedness' of the paragraph has changed,
		 * 2. and we are not in the situation where the buffer has changes
		 * and new changes are added to the paragraph.
		 */
		try {
			if (par_.isChanged() != was_changed_
				&& par_.inInset().isBufferValid()
				&& !(par_.inInset().buffer().areChangesPresent() && par_.isChanged()))
				par_.inInset().buffer().forceUpdate();
		} catch(support::ExceptionMessage const &) {}
	}

private:
	///
	Paragraph const & par_;
	///
	bool was_changed_;
};

}

void Paragraph::addChangesToToc(DocIterator const & cdit, Buffer const & buf,
                                bool output_active, TocBackend & backend) const
{
	d->changes_.addToToc(cdit, buf, output_active, backend);
}


bool Paragraph::isDeleted(pos_type start, pos_type end) const
{
	LASSERT(start >= 0 && start <= size(), return false);
	LASSERT(end > start && end <= size() + 1, return false);

	return d->changes_.isDeleted(start, end);
}


bool Paragraph::isChanged(pos_type start, pos_type end) const
{
	LASSERT(start >= 0 && start <= size(), return false);
	LASSERT(end > start && end <= size() + 1, return false);

	return d->changes_.isChanged(start, end);
}

// FIXME: Ideally the diverse isChanged() methods should account for that!
bool Paragraph::hasChangedInsets(pos_type start, pos_type end) const
{
	LASSERT(start >= 0 && start <= size(), return false);
	LASSERT(end > start && end <= size() + 1, return false);

	for (auto const & icit : d->insetlist_) {
		if (icit.pos < start)
			continue;
		if (icit.pos >= end)
			break;
		if (icit.inset && icit.inset->isChanged())
			return true;
	}
	return false;
}

bool Paragraph::isChanged() const
{
	return d->changes_.isChanged();
}


bool Paragraph::isMergedOnEndOfParDeletion(bool trackChanges) const
{
	// keep the logic here in sync with the logic of eraseChars()
	if (!trackChanges)
		return true;

	Change const & change = d->changes_.lookup(size());
	return change.inserted() && change.currentAuthor();
}

Change Paragraph::parEndChange() const
{
	return d->changes_.lookup(size());
}


void Paragraph::setChange(Change const & change)
{
	// Make sure that Buffer::hasChangesPresent is updated
	ChangesMonitor cm(*this);

	// beware of the imaginary end-of-par character!
	d->changes_.set(change, 0, size() + 1);

	/*
	 * Propagate the change recursively - but not in case of DELETED!
	 *
	 * Imagine that your co-author makes changes in an existing inset. He
	 * sends your document to you and you come to the conclusion that the
	 * inset should go completely. If you erase it, LyX must not delete all
	 * text within the inset. Otherwise, the change tracked insertions of
	 * your co-author get lost and there is no way to restore them later.
	 *
	 * Conclusion: An inset's content should remain untouched if you delete it
	 */

	if (!change.deleted()) {
		for (pos_type pos = 0; pos < size(); ++pos) {
			if (Inset * inset = getInset(pos))
				inset->setChange(change);
		}
	}
}


void Paragraph::setChange(pos_type pos, Change const & change)
{
	// Make sure that Buffer::hasChangesPresent is updated
	ChangesMonitor cm(*this);

	LASSERT(pos >= 0 && pos <= size(), return);
	d->changes_.set(change, pos);

	// see comment in setChange(Change const &) above
	if (!change.deleted() && pos < size())
		if (Inset * inset = getInset(pos))
			inset->setChange(change);
}


Change const & Paragraph::lookupChange(pos_type pos) const
{
	LBUFERR(pos >= 0 && pos <= size());
	return d->changes_.lookup(pos);
}


void Paragraph::acceptChanges(pos_type start, pos_type end)
{
	// Make sure that Buffer::hasChangesPresent is updated
	ChangesMonitor cm(*this);

	LASSERT(start >= 0 && start <= size(), return);
	LASSERT(end > start && end <= size() + 1, return);

	for (pos_type pos = start; pos < end; ++pos) {
		switch (lookupChange(pos).type) {
			case Change::UNCHANGED:
				// accept changes in nested inset
				if (Inset * inset = getInset(pos))
					inset->acceptChanges();
				break;

			case Change::INSERTED:
				d->changes_.set(Change(Change::UNCHANGED), pos);
				// also accept changes in nested inset
				if (Inset * inset = getInset(pos))
					inset->acceptChanges();
				break;

			case Change::DELETED:
				// Suppress access to non-existent
				// "end-of-paragraph char"
				if (pos < size()) {
					eraseChar(pos, false);
					--end;
					--pos;
				}
				break;
		}
	}
}


void Paragraph::rejectChanges(pos_type start, pos_type end)
{
	LASSERT(start >= 0 && start <= size(), return);
	LASSERT(end > start && end <= size() + 1, return);

	// Make sure that Buffer::hasChangesPresent is updated
	ChangesMonitor cm(*this);

	for (pos_type pos = start; pos < end; ++pos) {
		switch (lookupChange(pos).type) {
			case Change::UNCHANGED:
				// reject changes in nested inset
				if (Inset * inset = getInset(pos))
						inset->rejectChanges();
				break;

			case Change::INSERTED:
				// Suppress access to non-existent
				// "end-of-paragraph char"
				if (pos < size()) {
					eraseChar(pos, false);
					--end;
					--pos;
				}
				break;

			case Change::DELETED:
				d->changes_.set(Change(Change::UNCHANGED), pos);

				// Do NOT reject changes within a deleted inset!
				// There may be insertions of a co-author inside of it!

				break;
		}
	}
}


void Paragraph::Private::insertChar(pos_type pos, char_type c,
		Change const & change)
{
	LASSERT(pos >= 0 && pos <= int(text_.size()), return);

	// Make sure that Buffer::hasChangesPresent is updated
	ChangesMonitor cm(*owner_);

	// track change
	changes_.insert(change, pos);

	// This is actually very common when parsing buffers (and
	// maybe inserting ascii text)
	if (pos == pos_type(text_.size())) {
		// when appending characters, no need to update tables
		text_.push_back(c);
		// but we want spell checking
		requestSpellCheck(pos);
		return;
	}

	text_.insert(text_.begin() + pos, c);

	// Update the font table.
	fontlist_.increasePosAfterPos(pos);

	// Update the insets
	insetlist_.increasePosAfterPos(pos);

	// Update list of misspelled positions
	speller_state_.increasePosAfterPos(pos);

	// Update bookmarks
	if (inset_owner_ && inset_owner_->isBufferValid())
		theSession().bookmarks().adjustPosAfterPos(inset_owner_->buffer().fileName(),
	                                               id_, pos, 1);
}


bool Paragraph::insertInset(pos_type pos, Inset * inset,
				   Font const & font, Change const & change)
{
	LASSERT(inset, return false);
	LASSERT(pos >= 0 && pos <= size(), return false);

	// Make sure that Buffer::hasChangesPresent is updated
	ChangesMonitor cm(*this);

	// Paragraph::insertInset() can be used in cut/copy/paste operation where
	// d->inset_owner_ is not set yet.
	if (d->inset_owner_ && !d->inset_owner_->insetAllowed(inset->lyxCode()))
		return false;

	d->insertChar(pos, META_INSET, change);
	LASSERT(d->text_[pos] == META_INSET, return false);

	// Add a new entry in the insetlist_.
	d->insetlist_.insert(inset, pos);

	// Some insets require run of spell checker
	requestSpellCheck(pos);
	setFont(pos, font);
	return true;
}


bool Paragraph::eraseChar(pos_type pos, bool trackChanges)
{
	LASSERT(pos >= 0 && pos <= size(), return false);

	// Make sure that Buffer::hasChangesPresent is updated
	ChangesMonitor cm(*this);

	// keep the logic here in sync with the logic of isMergedOnEndOfParDeletion()

	if (trackChanges) {
		Change change = d->changes_.lookup(pos);

		// set the character to DELETED if
		//  a) it was previously unchanged or
		//  b) it was inserted by a co-author

		if (!change.changed() ||
		      (change.inserted() && !change.currentAuthor())) {
			setChange(pos, Change(Change::DELETED));
			// request run of spell checker
			requestSpellCheck(pos);
			return false;
		}

		if (change.deleted())
			return false;
	}

	// Don't physically access the imaginary end-of-paragraph character.
	// eraseChar() can only mark it as DELETED. A physical deletion of
	// end-of-par must be handled externally.
	if (pos == size()) {
		return false;
	}

	// track change
	d->changes_.erase(pos);

	// if it is an inset, delete the inset entry
	if (d->text_[pos] == META_INSET)
		d->insetlist_.erase(pos);

	d->text_.erase(d->text_.begin() + pos);

	// Update the fontlist_
	d->fontlist_.erase(pos);

	// Update the insetlist_
	d->insetlist_.decreasePosAfterPos(pos);

	// Update list of misspelled positions
	d->speller_state_.decreasePosAfterPos(pos);
	d->speller_state_.refreshLast(size());

	// Update bookmarks
	if (d->inset_owner_ && d->inset_owner_->isBufferValid())
		theSession().bookmarks().adjustPosAfterPos(d->inset_owner_->buffer().fileName(),
	                                               d->id_, pos, -1);

	return true;
}


int Paragraph::eraseChars(pos_type start, pos_type end, bool trackChanges)
{
	LASSERT(start >= 0 && start <= size(), return 0);
	LASSERT(end >= start && end <= size() + 1, return 0);

	pos_type i = start;
	for (pos_type count = end - start; count; --count) {
		if (!eraseChar(i, trackChanges))
			++i;
	}
	return end - i;
}


// Handle combining characters
int Paragraph::Private::latexSurrogatePair(BufferParams const & bparams,
		otexstream & os, char_type c, char_type next,
		OutputParams const & runparams) const
{
	// Writing next here may circumvent a possible font change between
	// c and next. Since next is only output if it forms a surrogate pair
	// with c we can ignore this:
	// A font change inside a surrogate pair does not make sense and is
	// hopefully impossible to input.
	// FIXME: change tracking
	// Is this correct WRT change tracking?
	Encoding const & encoding = *(runparams.encoding);
	docstring latex1 = encoding.latexChar(next).first;
	if (runparams.inIPA) {
		string const tipashortcut = Encodings::TIPAShortcut(next);
		if (!tipashortcut.empty()) {
			latex1 = from_ascii(tipashortcut);
		}
	}
	docstring latex2 = encoding.latexChar(c).first;

	if (bparams.useNonTeXFonts || docstring(1, next) == latex1) {
		// Encoding supports the combination:
		// output as is (combining char after base char).
		os << latex2 << latex1;
		return latex1.length() + latex2.length();
	}

	os << latex1 << "{" << latex2 << "}";
	return latex1.length() + latex2.length() + 2;
}


bool Paragraph::Private::simpleTeXBlanks(BufferParams const & bparams,
				       OutputParams const & runparams,
				       otexstream & os,
				       pos_type i,
				       unsigned int & column,
				       Font const & font,
				       Layout const & style) const
{
	if (style.pass_thru || runparams.pass_thru)
		return false;

	if (i + 1 < int(text_.size())) {
		char_type next = text_[i + 1];
		if (Encodings::isCombiningChar(next)) {
			// This space has an accent, so we must always output it.
			column += latexSurrogatePair(bparams, os, ' ', next, runparams) - 1;
			return true;
		}
	}

	if (runparams.linelen > 0
	    && column > runparams.linelen
	    && i
	    && text_[i - 1] != ' '
	    && (i + 1 < int(text_.size()))
	    // same in FreeSpacing mode
	    && !owner_->isFreeSpacing()
	    // In typewriter mode, we want to avoid
	    // ! . ? : at the end of a line
	    && !(font.fontInfo().family() == TYPEWRITER_FAMILY
		 && (text_[i - 1] == '.'
		     || text_[i - 1] == '?'
		     || text_[i - 1] == ':'
		     || text_[i - 1] == '!'))) {
		os << '\n';
		os.texrow().start(owner_->id(), i + 1);
		column = 0;
	} else if (style.free_spacing) {
		os << '~';
	} else {
		os << ' ';
	}
	return false;
}


void Paragraph::Private::latexInset(BufferParams const & bparams,
				    otexstream & os,
				    OutputParams & runparams,
				    Font & running_font,
				    Font & basefont,
				    Font const & outerfont,
				    bool & open_font,
				    Change & running_change,
				    Layout const & style,
				    pos_type & i,
				    unsigned int & column,
				    bool const fontswitch_inset,
				    bool const closeLanguage,
				    bool const lang_switched_at_inset) const
{
	Inset * inset = owner_->getInset(i);
	LBUFERR(inset);

	if (style.pass_thru) {
		odocstringstream ods;
		inset->plaintext(ods, runparams);
		os << ods.str();
		return;
	}

	// FIXME: move this to InsetNewline::latex
	if (inset->lyxCode() == NEWLINE_CODE || inset->lyxCode() == SEPARATOR_CODE) {
		// newlines are handled differently here than
		// the default in simpleTeXSpecialChars().
		if (!style.newline_allowed) {
			os << '\n';
		} else {
			if (open_font) {
				bool needPar = false;
				column += running_font.latexWriteEndChanges(
					os, bparams, runparams,
					basefont, basefont, needPar);
				open_font = false;
			}

			if (running_font.fontInfo().family() == TYPEWRITER_FAMILY)
				os << '~';

			basefont = owner_->getLayoutFont(bparams, outerfont);
			running_font = basefont;

			if (runparams.moving_arg)
				os << "\\protect ";

		}
		os.texrow().start(owner_->id(), i + 1);
		column = 0;
	}

	if (owner_->isDeleted(i)) {
		if( ++runparams.inDeletedInset == 1)
			runparams.changeOfDeletedInset = owner_->lookupChange(i);
	}

	if (inset->canTrackChanges()) {
		column += Changes::latexMarkChange(os, bparams, running_change,
			Change(Change::UNCHANGED), runparams);
		running_change = Change(Change::UNCHANGED);
	}

	unsigned int close_brace = 0;
	bool const disp_env = (inset->isEnvironment() && inset->getLayout().isDisplay())
			|| runparams.inDisplayMath;
	string close_env;
	odocstream::pos_type const len = os.os().tellp();

	if (inset->forceLTR(runparams)
	    // babel with LuaTeX does not need a switch
	    // babel with XeTeX needs a switch only if bidi is used
	    // and \L is not defined there.
	    && (!runparams.isFullUnicode() || bparams.useBidiPackage(runparams) || runparams.use_polyglossia)
	    && running_font.isRightToLeft()) {
		if (bparams.useBidiPackage(runparams) || runparams.use_polyglossia) {
			// (lua)bidi
			// Displayed environments go in an LTR environment
			if (disp_env) {
				os << "\\begin{LTR}";
				close_env = "LTR";
			} else {
				if (runparams.flavor == Flavor::LuaTeX) {
					// luabidi's \LRE needs extra grouping
					// (possibly a LuaTeX bug)
					os << '{';
					close_brace = 1;
				}
				os << "\\LRE{";
				close_brace += 1;
			}
		} else if (running_font.language()->lang() == "farsi"
			 || running_font.language()->lang() == "arabic_arabi") {
			os << "\\textLR{" << termcmd;
			close_brace = 1;
		} else {
			// babel classic
			os << "\\L{";
			if (disp_env)
				os << safebreakln;
			close_brace = 1;
		}
	}

	if (open_font && fontswitch_inset) {
		bool lang_closed = false;
		// Close language if needed
		if (closeLanguage && !lang_switched_at_inset) {
			// We need prev_font here as language changes directly at inset
			// will only be started inside the inset.
			Font const prev_font = (i > 0) ?
						owner_->getFont(bparams, i - 1, outerfont)
					      : running_font;
			Font tmpfont(basefont);
			tmpfont.setLanguage(prev_font.language());
			bool needPar = false;
			unsigned int count = tmpfont.latexWriteEndChanges(os, bparams, runparams,
									  basefont, basefont,
									  needPar, closeLanguage);
			column += count;
			lang_closed = count > 0;
		}
		// Update the running_font, making sure, however,
		// to leave the language as it was.
		// FIXME: probably a better way to keep track of the old
		// language, than copying the entire font?
		Font const copy_font(running_font);
		basefont = owner_->getLayoutFont(bparams, outerfont);
		running_font = basefont;
		if (!closeLanguage)
			running_font.setLanguage(copy_font.language());
		OutputParams rp = runparams;
		rp.encoding = basefont.language()->encoding();
		// For these, we use switches, so they should be taken as
		// base inside the inset.
		basefont.fontInfo().setSize(copy_font.fontInfo().size());
		basefont.fontInfo().setFamily(copy_font.fontInfo().family());
		basefont.fontInfo().setSeries(copy_font.fontInfo().series());
		// Now re-do font changes in a way needed here
		// (using switches with multi-par insets)
		InsetText const * textinset = inset->asInsetText();
		bool const cprotect = textinset
			? textinset->hasCProtectContent(runparams.moving_arg)
			  && !textinset->text().isMainText()
			  && inset->lyxCode() != BRANCH_CODE
			  && !runparams.no_cprotect
			  && !inset->getLayout().noCProtect()
			: false;
		unsigned int count2 = basefont.latexWriteStartChanges(os, bparams,
						      rp, running_font,
						      basefont, true,
						      cprotect);
		open_font = true;
		column += count2;
		if (count2 == 0 && (lang_closed || lang_switched_at_inset))
			// All fonts closed
			open_font = false;
		if (closeLanguage)
			runparams.local_font = &basefont;
	}

	if (fontswitch_inset && !closeLanguage) {
		// The directionality has been switched at inset.
		// Force markup inside.
		runparams.local_font = &basefont;
	}

	size_t const previous_row_count = os.texrow().rows();

	try {
		runparams.lastid = id_;
		runparams.lastpos = i;
		inset->latex(os, runparams);
	} catch (EncodingException & e) {
		// add location information and throw again.
		e.par_id = id_;
		e.pos = i;
		throw;
	}

	if (!close_env.empty())
		os << "\\end{" << close_env << "}";

	if (close_brace > 0) {
		for (unsigned i = 0; i < close_brace; ++i)
			os << '}';
		if (disp_env)
			os << safebreakln;
	}

	if (os.texrow().rows() > previous_row_count) {
		os.texrow().start(owner_->id(), i + 1);
		column = 0;
	} else {
		column += (unsigned int)(os.os().tellp() - len);
	}

	if (owner_->isDeleted(i))
		--runparams.inDeletedInset;
}


void Paragraph::Private::latexSpecialChar(otexstream & os,
					  BufferParams const & bparams,
					  OutputParams const & runparams,
					  Font const & running_font,
					  string & alien_script,
					  Layout const & style,
					  InsetLayout const & il,
					  pos_type & i,
					  pos_type end_pos,
					  unsigned int & column) const
{
	char_type const c = owner_->getUChar(bparams, runparams, i);

	// Special case: URLs with hyperref need to escape backslash, # and % (#13012).
	// Both a layout tag and a dedicated inset seem too much effort for this.
	string const hr_url_escape_chars = "\\#%";
	if (contains(hr_url_escape_chars, c) && runparams.use_hyperref && il.latexname() == "url"
	    && il.requiresFeature("url")) {
		os << "\\";
		os.put(c);
		return;
	}

	if (!style.pass_thru && !runparams.pass_thru && !style.escape_chars.empty()
	    && contains(style.escape_chars.substr(1), c)){
		os << style.escape_chars.substr(0,1);
		os.put(c);
		return;
	}
	if (!style.pass_thru && !runparams.pass_thru && !runparams.escape_chars.empty()
	    && contains(runparams.escape_chars.substr(1), c)){
		os << runparams.escape_chars.substr(0,1);
		os.put(c);
		return;
	}

	if ((style.pass_thru || runparams.pass_thru || runparams.find_effective()
	    || contains(style.pass_thru_chars, c)
	    || contains(runparams.pass_thru_chars, c))
	    && !contains(runparams.no_pass_thru_chars, c)) {
		if (runparams.find_effective()) {
			switch (c) {
			case '\\':
				os << "\\\\";
				return;
			case '{':
				os << "\\braceleft ";
				return;
			case '}':
				os << "\\braceright ";
				return;
			case '$':
				os << "\\lyxdollar ";
				return;
			case '~':
				os << "\\lyxtilde ";
				return;
			case ' ':
			case '\0':
				break;
			default:
				os.put(c);
				return;
			}
		}
		else if (c != '\0') {
			Encoding const * const enc = runparams.encoding;
			if (enc && !enc->encodable(c))
				throw EncodingException(c);
			os.put(c);
			return;
		}
	}

	// TIPA uses its own T3 encoding
	if (runparams.inIPA && latexSpecialT3(c, os, i, column))
		return;
	// If T1 font encoding is used, use the special
	// characters it provides.
	// NOTE: Some languages reset the font encoding internally to a
	//       non-standard font encoding. If we are using such a language,
	//       we do not output special T1 chars.
	if (!runparams.inIPA && !running_font.language()->internalFontEncoding()
	    && !runparams.isFullUnicode() && runparams.main_fontenc == "T1"
	    && latexSpecialT1(c, os, i, column))
		return;
	// NOTE: "fontspec" (non-TeX fonts) sets the font encoding to "TU" (untill 2017 "EU1" or "EU2")
	else if (!runparams.inIPA && !running_font.language()->internalFontEncoding()
		 && runparams.isFullUnicode() && latexSpecialTU(c, os, i, column))
		     return;

	// Otherwise, we use what LaTeX provides us.
	switch (c) {
	case '\\':
		os << "\\textbackslash" << termcmd;
		column += 15;
		break;
	case '<':
		os << "\\textless" << termcmd;
		column += 10;
		break;
	case '>':
		os << "\\textgreater" << termcmd;
		column += 13;
		break;
	case '|':
		os << "\\textbar" << termcmd;
		column += 9;
		break;
	case '-':
		os << '-';
		if (i + 1 < static_cast<pos_type>(text_.size()) &&
		    (end_pos == -1 || i + 1 < end_pos) &&
		    text_[i+1] == '-') {
			// Prevent "--" becoming an en dash and "---" an em dash.
			// (Within \ttfamily, "---" is merged to en dash + hyphen.)
			os << "{}";
			column += 2;
		}
		break;
	case '\"':
		os << "\\textquotedbl" << termcmd;
		column += 14;
		break;

	case '$': case '&':
	case '%': case '#': case '{':
	case '}': case '_':
		os << '\\';
		os.put(c);
		column += 1;
		break;

	case '~':
		os << "\\textasciitilde" << termcmd;
		column += 16;
		break;

	case '^':
		os << "\\textasciicircum" << termcmd;
		column += 17;
		break;

	case '*':
	case '[':
	case ']':
		// avoid being mistaken for optional arguments
		os << '{';
		os.put(c);
		os << '}';
		column += 2;
		break;

	case ' ':
		// Blanks are printed before font switching.
		// Sure? I am not! (try nice-latex)
		// I am sure it's correct. LyX might be smarter
		// in the future, but for now, nothing wrong is
		// written. (Asger)
		break;

	case 0x2013:
	case 0x2014:
		// XeTeX's dash behaviour is determined via a global setting
		if (bparams.use_dash_ligatures
		    && owner_->getFontSettings(bparams, i).fontInfo().family() != TYPEWRITER_FAMILY
		    && !runparams.inIPA
			// TODO #10961: && not in inset Flex Code
			// TODO #10961: && not in layout LyXCode
		    && (!bparams.useNonTeXFonts || runparams.flavor != Flavor::XeTeX)) {
			if (c == 0x2013) {
				// en-dash
				os << "--";
				column +=2;
			} else {
				// em-dash
				os << "---";
				column +=3;
			}
			break;
		}
		// fall through
	default:
		if (c == '\0')
			return;

		Encoding const & encoding = *(runparams.encoding);
		char_type next = '\0';
		if (i + 1 < int(text_.size())) {
			next = text_[i + 1];
			if (Encodings::isCombiningChar(next)) {
				column += latexSurrogatePair(bparams, os, c, next, runparams) - 1;
				++i;
				break;
			}
		}
		pair<docstring, bool> latex = encoding.latexChar(c);
		docstring nextlatex;
		bool nexttipas = false;
		string nexttipashortcut;
		if (next != '\0' && next != META_INSET && !encoding.encodable(next)) {
			nextlatex = encoding.latexChar(next).first;
			if (runparams.inIPA) {
				nexttipashortcut = Encodings::TIPAShortcut(next);
				nexttipas = !nexttipashortcut.empty();
			}
		}
		bool tipas = false;
		if (runparams.inIPA) {
			string const tipashortcut = Encodings::TIPAShortcut(c);
			if (!tipashortcut.empty()) {
				latex.first = from_ascii(tipashortcut);
				latex.second = false;
				tipas = true;
			}
		}
		// eventually close "script wrapper" command (see `Paragraph::latex`)
		if (!alien_script.empty()
			&& alien_script != Encodings::isKnownScriptChar(next)) {
			column += latex.first.length();
			alien_script.clear();
			os << latex.first << "}";
			break;
		}
		if (latex.second
			 && ((!prefixIs(nextlatex, '\\')
			       && !prefixIs(nextlatex, '{')
			       && !prefixIs(nextlatex, '}'))
			     || (nexttipas
			         && !prefixIs(from_ascii(nexttipashortcut), '\\')))
			 && !tipas) {
			// Prevent eating of a following space or command corruption by
			// following characters
			if (next == ' ' || next == '\0') {
				column += latex.first.length() + 1;
				os << latex.first << "{}";
			} else {
				column += latex.first.length();
				os << latex.first << " ";
			}
		} else {
			column += latex.first.length() - 1;
			os << latex.first;
		}
		break;
	}
}


bool Paragraph::Private::latexSpecialT1(char_type const c, otexstream & os,
	pos_type i, unsigned int & column) const
{
	switch (c) {
	case '>':
	case '<':
		os.put(c);
		// In T1 encoding, these characters exist
		// but we should avoid ligatures
		if (i + 1 >= int(text_.size()) || text_[i + 1] != c)
			return true;
		os << "\\textcompwordmark" << termcmd;
		column += 19;
		return true;
	case '|':
		os.put(c);
		return true;
	case '\"':
		// soul.sty breaks with \char`\"
		os << "\\textquotedbl" << termcmd;
		column += 14;
		return true;
	default:
		return false;
	}
}


bool Paragraph::Private::latexSpecialTU(char_type const c, otexstream & os,
	pos_type i, unsigned int & column) const
{
	// TU encoding is currently on par with T1.
	return latexSpecialT1(c, os, i, column);
}


bool Paragraph::Private::latexSpecialT3(char_type const c, otexstream & os,
	pos_type /*i*/, unsigned int & column) const
{
	switch (c) {
	case '*':
	case '[':
	case ']':
	case '\"':
		os.put(c);
		return true;
	case '|':
		os << "\\textvertline" << termcmd;
		column += 14;
		return true;
	default:
		return false;
	}
}


void Paragraph::Private::validate(LaTeXFeatures & features) const
{
	if (layout_->inpreamble && inset_owner_) {
		// FIXME: Using a string stream here circumvents the encoding
		// switching machinery of odocstream. Therefore the
		// output is wrong if this paragraph contains content
		// that needs to switch encoding.
		Buffer const & buf = inset_owner_->buffer();
		otexstringstream os;
		os << layout_->preamble();
		size_t const length = os.length();
		TeXOnePar(buf, *inset_owner_->getText(int(buf.getParFromID(owner_->id()).idx())),
			  buf.getParFromID(owner_->id()).pit(), os,
			  features.runparams(), string(), 0, -1, true);
		if (os.length() > length)
			features.addPreambleSnippet(os.release(), true);
	}

	if (features.runparams().flavor == Flavor::Html
	    && layout_->htmltitle()) {
		features.setHTMLTitle(owner_->asString(AS_STR_INSETS | AS_STR_SKIPDELETE));
	}

	// check the params.
	if (!params_.spacing().isDefault())
		features.require("setspace");

	// then the layouts
	features.useLayout(layout_->name());

	// then the fonts
	fontlist_.validate(features);

	// then the indentation
	if (!params_.leftIndent().zero())
		features.require("ParagraphLeftIndent");

	// then the insets
	InsetList::const_iterator icit = insetlist_.begin();
	InsetList::const_iterator iend = insetlist_.end();
	for (; icit != iend; ++icit) {
		if (icit->inset) {
			features.inDeletedInset(owner_->isDeleted(icit->pos));
			if (icit->inset->lyxCode() == FOOT_CODE) {
				// FIXME: an item inset would make things much easier.
				if ((layout_->latextype == LATEX_LIST_ENVIRONMENT
				     || (layout_->latextype == LATEX_ITEM_ENVIRONMENT
					 && layout_->margintype == MARGIN_FIRST_DYNAMIC))
				    && (icit->pos < begin_of_body_
					|| (icit->pos == begin_of_body_
					    && (icit->pos == 0 || text_[icit->pos - 1] != ' '))))
					features.saveNoteEnv("description");
			}
			icit->inset->validate(features);
			features.inDeletedInset(false);
			if (layout_->needprotect &&
			    icit->inset->lyxCode() == FOOT_CODE)
				features.require("NeedLyXFootnoteCode");
		}
	}

	// then the contents
	BufferParams const bp = features.runparams().is_child
		? features.buffer().masterParams() : features.buffer().params();
	for (pos_type i = 0; i < int(text_.size()) ; ++i) {
		char_type c = text_[i];
		CharInfo const & ci = Encodings::unicodeCharInfo(c);
		if (c == 0x0022) {
			if (features.runparams().isFullUnicode() && bp.useNonTeXFonts)
				features.require("textquotedblp");
			else if (features.runparams().main_fontenc != "T1"
				 || ((&owner_->getFontSettings(bp, i))->language()->internalFontEncoding()))
				features.require("textquotedbl");
		} else if (ci.textFeature() && contains(ci.textPreamble(), '=')) {
			// features that depend on the font or input encoding
			string feats = ci.textPreamble();
			string fontenc = (&owner_->getFontSettings(bp, i))->language()->fontenc(bp);
			if (fontenc.empty())
				fontenc = features.runparams().main_fontenc;
			while (!feats.empty()) {
				string feat;
				feats = split(feats, feat, ',');
				if (contains(feat, "!=")) {
					// a feature that is required except for the spcified
					// font or input encodings
					string realfeature;
					string const contexts = ltrim(split(feat, realfeature, '!'), "=");
					// multiple encodings are separated by semicolon
					vector<string> context = getVectorFromString(contexts, ";");
					// require feature if the context matches neither current font
					// nor input encoding
					if (std::find(context.begin(), context.end(), fontenc) == context.end()
					    && std::find(context.begin(), context.end(),
							 features.runparams().encoding->name()) == context.end())
						features.require(realfeature);
				} else if (contains(feat, '=')) {
					// a feature that is required only for the spcified
					// font or input encodings
					string realfeature;
					string const contexts = split(feat, realfeature, '=');
					// multiple encodings are separated by semicolon
					vector<string> context = getVectorFromString(contexts, ";");
					// require feature if the context matches either current font
					// or input encoding
					if (std::find(context.begin(), context.end(), fontenc) != context.end()
					    || std::find(context.begin(), context.end(),
							 features.runparams().encoding->name()) != context.end())
						features.require(realfeature);
				}
			}
		} else if (!bp.use_dash_ligatures
			   && (c == 0x2013 || c == 0x2014)
			   && bp.useNonTeXFonts
			   && features.runparams().flavor == Flavor::XeTeX)
			// XeTeX's dash behaviour is determined via a global setting
			features.require("xetexdashbreakstate");
		BufferEncodings::validate(c, features);
	}
}

/////////////////////////////////////////////////////////////////////
//
// Paragraph
//
/////////////////////////////////////////////////////////////////////

namespace {
	Layout const emptyParagraphLayout;
}

Paragraph::Paragraph()
	: d(new Paragraph::Private(this, emptyParagraphLayout))
{
	itemdepth = 0;
	d->params_.clear();
}


Paragraph::Paragraph(Paragraph const & par)
	: itemdepth(par.itemdepth),
	d(new Paragraph::Private(*par.d, this))
{
	registerWords();
}


Paragraph::Paragraph(Paragraph const & par, pos_type beg, pos_type end)
	: itemdepth(par.itemdepth),
	d(new Paragraph::Private(*par.d, this, beg, end))
{
	registerWords();
}


Paragraph & Paragraph::operator=(Paragraph const & par)
{
	// needed as we will destroy the private part before copying it
	if (&par != this) {
		itemdepth = par.itemdepth;

		deregisterWords();
		delete d;
		d = new Private(*par.d, this);
		registerWords();
	}
	return *this;
}


Paragraph::~Paragraph()
{
	deregisterWords();
	delete d;
}


namespace {

// this shall be called just before every "os << ..." action.
void flushString(ostream & os, docstring & s)
{
	os << to_utf8(s);
	s.erase();
}

} // namespace


void Paragraph::write(ostream & os, BufferParams const & bparams,
	depth_type & depth) const
{
	// The beginning or end of a deeper (i.e. nested) area?
	if (depth != d->params_.depth()) {
		if (d->params_.depth() > depth) {
			while (d->params_.depth() > depth) {
				os << "\n\\begin_deeper";
				++depth;
			}
		} else {
			while (d->params_.depth() < depth) {
				os << "\n\\end_deeper";
				--depth;
			}
		}
	}

	// First write the layout
	os << "\n\\begin_layout " << to_utf8(d->layout_->name()) << '\n';

	d->params_.write(os);

	Font font1(inherit_font, bparams.language);

	Change running_change = Change(Change::UNCHANGED);

	// this string is used as a buffer to avoid repetitive calls
	// to to_utf8(), which turn out to be expensive (JMarc)
	docstring write_buffer;

	int column = 0;
	for (pos_type i = 0; i <= size(); ++i) {

		Change const & change = lookupChange(i);
		if (change != running_change)
			flushString(os, write_buffer);
		Changes::lyxMarkChange(os, bparams, column, running_change, change);
		running_change = change;

		if (i == size())
			break;

		// Write font changes
		Font font2 = getFontSettings(bparams, i);
		if (font2 != font1) {
			flushString(os, write_buffer);
			font2.lyxWriteChanges(font1, os);
			column = 0;
			font1 = font2;
		}

		char_type const c = d->text_[i];
		switch (c) {
		case META_INSET:
			if (Inset const * inset = getInset(i)) {
				flushString(os, write_buffer);
				if (inset->directWrite()) {
					// international char, let it write
					// code directly so it's shorter in
					// the file
					inset->write(os);
				} else {
					if (i)
						os << '\n';
					os << "\\begin_inset ";
					inset->write(os);
					os << "\n\\end_inset\n\n";
					column = 0;
				}
				// FIXME This can be removed again once the mystery
				// crash has been resolved.
				os << flush;
			}
			break;
		case '\\':
			flushString(os, write_buffer);
			os << "\n\\backslash\n";
			column = 0;
			break;
		case '.':
		case '!':
		case '?':
		case ':':
		case ';':
		case ',':
		case 0x061F:  // ؟ U+061F  ARABIC QUESTION MARK
		case 0x061B:  // ؛ U+061B  ARABIC SEMICOLON
		case 0x060C:  // ، U+060C  ARABIC COMMA
			flushString(os, write_buffer);
			if (i + 1 < size() && d->text_[i + 1] == ' ') {
				os << to_utf8(docstring(1, c)) << '\n';
				column = 0;
			} else
				os << to_utf8(docstring(1, c));
			break;
		case 0x2014:  // — U+2014  EM DASH
		case 0x3002:  // 。 U+3002  IDEOGRAPHIC FULL STOP
		case 0xFF01:  // ！ U+FF01  FULLWIDTH EXCLAMATION MARK
		case 0xFF1F:  // ？ U+FF1F  FULLWIDTH QUESTION MARK
		case 0xFF1A:  // ： U+FF1A  FULLWIDTH COLON
		case 0xFF1B:  // ； U+FF1B  FULLWIDTH SEMICOLON
		case 0xFF0C:  // ， U+FF0C  FULLWIDTH COMMA
			flushString(os, write_buffer);
			os << to_utf8(docstring(1, c)) << '\n';
			column = 0;
			break;
		default:
			if (column > 500) {
				flushString(os, write_buffer);
				os << '\n';
				column = 0;
			}
			// this check is to amend a bug. LyX sometimes
			// inserts '\0' this could cause problems.
			if (c != '\0')
				write_buffer.push_back(c);
			else
				LYXERR0("NUL char in structure.");
			++column;
			break;
		}
	}

	flushString(os, write_buffer);
	os << "\n\\end_layout\n";
	// FIXME This can be removed again once the mystery
	// crash has been resolved.
	os << flush;
}


void Paragraph::validate(LaTeXFeatures & features) const
{
	d->validate(features);
	bool fragile = features.runparams().moving_arg;
	fragile |= layout().needprotect;
	if (inInset().getLayout().isNeedProtect())
		fragile = true;
	if (needsCProtection(fragile))
		features.require("cprotect");
}


void Paragraph::insert(pos_type pos, docstring const & str,
		       Font const & font, Change const & change)
{
	for (size_t i = 0, n = str.size(); i != n ; ++i)
		insertChar(pos + i, str[i], font, change);
}


void Paragraph::appendChar(char_type c, Font const & font,
		Change const & change)
{
	// Make sure that Buffer::hasChangesPresent is updated
	ChangesMonitor cm(*this);

	// track change
	d->changes_.insert(change, d->text_.size());
	// when appending characters, no need to update tables
	d->text_.push_back(c);
	setFont(d->text_.size() - 1, font);
	d->requestSpellCheck(d->text_.size() - 1);
}


void Paragraph::appendString(docstring const & s, Font const & font,
		Change const & change)
{
	// Make sure that Buffer::hasChangesPresent is updated
	ChangesMonitor cm(*this);

	pos_type end = s.size();
	size_t oldsize = d->text_.size();
	size_t newsize = oldsize + end;
	size_t capacity = d->text_.capacity();
	if (newsize >= capacity)
		d->text_.reserve(max(capacity + 100, newsize));

	// when appending characters, no need to update tables
	d->text_.append(s);

	// FIXME: Optimize this!
	for (size_t i = oldsize; i != newsize; ++i) {
		// track change
		d->changes_.insert(change, i);
		d->requestSpellCheck(i);
	}
	d->fontlist_.set(oldsize, font);
	d->fontlist_.set(newsize - 1, font);
}


void Paragraph::insertChar(pos_type pos, char_type c,
			   bool trackChanges)
{
	d->insertChar(pos, c, Change(trackChanges ?
			   Change::INSERTED : Change::UNCHANGED));
}


void Paragraph::insertChar(pos_type pos, char_type c,
			   Font const & font, bool trackChanges)
{
	d->insertChar(pos, c, Change(trackChanges ?
			   Change::INSERTED : Change::UNCHANGED));
	setFont(pos, font);
}


void Paragraph::insertChar(pos_type pos, char_type c,
			   Font const & font, Change const & change)
{
	d->insertChar(pos, c, change);
	setFont(pos, font);
}


void Paragraph::resetFonts(Font const & font)
{
	d->fontlist_.clear();
	d->fontlist_.set(0, font);
	d->fontlist_.set(d->text_.size() - 1, font);
}

// Gets uninstantiated font setting at position.
Font const & Paragraph::getFontSettings(BufferParams const & bparams,
					 pos_type pos) const
{
	if (pos > size()) {
		LYXERR0("pos: " << pos << " size: " << size());
		LBUFERR(false);
	}

	FontList::const_iterator cit = d->fontlist_.fontIterator(pos);
	if (cit != d->fontlist_.end())
		return cit->font();

	if (pos == size() && !empty())
		return getFontSettings(bparams, pos - 1);

	// Optimisation: avoid a full font instantiation if there is no
	// language change from previous call.
	static Font previous_font;
	static Language const * previous_lang = nullptr;
	Language const * lang = getParLanguage(bparams);
	if (lang != previous_lang) {
		previous_lang = lang;
		previous_font = Font(inherit_font, lang);
	}
	return previous_font;
}


FontSpan Paragraph::fontSpan(pos_type pos) const
{
	LBUFERR(pos <= size());

	if (pos == size())
		return FontSpan(pos, pos);

	pos_type start = 0;
	FontList::const_iterator cit = d->fontlist_.begin();
	FontList::const_iterator end = d->fontlist_.end();
	for (; cit != end; ++cit) {
		if (cit->pos() >= pos) {
			if (pos >= beginOfBody())
				return FontSpan(max(start, beginOfBody()),
						cit->pos());
			else
				return FontSpan(start,
						min(beginOfBody() - 1,
							 cit->pos()));
		}
		start = cit->pos() + 1;
	}

	// This should not happen, but if so, we take no chances.
	LYXERR0("Paragraph::fontSpan: position not found in fontinfo table!");
	LASSERT(false, /**/);
	return FontSpan(pos, pos);
}


// Gets uninstantiated font setting at position 0
Font const & Paragraph::getFirstFontSettings(BufferParams const & bparams) const
{
	if (!empty() && !d->fontlist_.empty())
		return d->fontlist_.begin()->font();

	// Optimisation: avoid a full font instantiation if there is no
	// language change from previous call.
	static Font previous_font;
	static Language const * previous_lang = nullptr;
	if (bparams.language != previous_lang) {
		previous_lang = bparams.language;
		previous_font = Font(inherit_font, bparams.language);
	}

	return previous_font;
}


// Gets the fully instantiated font at a given position in a paragraph
// This is basically the same function as TextMetrics::displayFont().
// The difference is that this one is used for generating the LaTeX file,
// and thus cosmetic "improvements" are disallowed: This has to deliver
// the true picture of the buffer. (Asger)
Font const Paragraph::getFont(BufferParams const & bparams, pos_type pos,
				 Font const & outerfont) const
{
	LBUFERR(pos >= 0);

	Font font = getFontSettings(bparams, pos);

	pos_type const body_pos = beginOfBody();
	FontInfo & fi = font.fontInfo();
	if (pos < body_pos)
		fi.realize(d->layout_->labelfont);
	else
		fi.realize(d->layout_->font);

	fi.realize(outerfont.fontInfo());
	fi.realize(bparams.getFont().fontInfo());

	return font;
}


Font const Paragraph::getLabelFont
	(BufferParams const & bparams, Font const & outerfont) const
{
	FontInfo tmpfont = d->layout_->labelfont;
	tmpfont.realize(outerfont.fontInfo());
	tmpfont.realize(bparams.getFont().fontInfo());
	return Font(tmpfont, getParLanguage(bparams));
}


Font const Paragraph::getLayoutFont
	(BufferParams const & bparams, Font const & outerfont) const
{
	FontInfo tmpfont = d->layout_->font;
	tmpfont.realize(outerfont.fontInfo());
	tmpfont.realize(bparams.getFont().fontInfo());
	return Font(tmpfont, getParLanguage(bparams));
}


char_type Paragraph::getUChar(BufferParams const & bparams,
			      OutputParams const & rp,
			      pos_type pos) const
{
	char_type c = d->text_[pos];

	// Return unchanged character
	// 1. in all LTR languages
	// 2. if we use XeTeX (both with babel and polyglossia)
	// 3. if we use LuaTeX with babel
	if (!getFontSettings(bparams, pos).isRightToLeft()
	    || rp.flavor == Flavor::XeTeX
	    || (rp.use_babel && rp.flavor == Flavor::LuaTeX))
		return c;

	// For the remaining cases, we need to account for some special cases.
	// FIXME This needs to be audited!
	// Check if:
	// * The input is as expected for all delimiters
	//   => checked for Hebrew!
	// * The output matches the display in the LyX workarea
	//   => checked for Hebrew!
	// * The special cases below are really necessary
	//   => checked for Hebrew!
	// * In arabic_arabi, brackets are transformed to Arabic
	//   Ornate Parentheses. Is this is really wanted?
	//   => Yes, in file ararabeyes.enc from the arabi bundle
	//      the slot of the left bracket (slot 91) is encoded as
	//      "ornaterightparenthesis". This is also the reason
	//      brackets don't need to be mirrored with arabi

	string const & lang = getFontSettings(bparams, pos).language()->lang();
	char_type uc = c;

	// These are the cases where we need to mirror delimiters in RTL context
	// in the remaining cases (polyglossia + LuaTeX or classic [pdf]latex):
	// 1. With polyglossia and LuaTeX (luabidi) parentheses and brackets
	//    need to be mirrored in RTL, regardless of the language, or script.
	// 2. In the languages that follow, parentheses need to be mirrored
	//    in classic (pdf)latex
	bool const reverseparens = (rp.use_polyglossia || lang == "hebrew");
	// 3. In all RTL languages except for those that follow, brackets
	//    need to be mirrored in classic (pdf)latex
	bool const reversebrackets = rp.use_polyglossia
			|| (lang != "arabic_arabtex"
			    && lang != "arabic_arabi"
			    && lang != "farsi");

	// Now swap delimiters if needed.
	switch (c) {
	case '(':
		if (reverseparens)
			uc = ')';
		break;
	case ')':
		if (reverseparens)
			uc = '(';
		break;
	case '[':
		if (reversebrackets)
			uc = ']';
		break;
	case ']':
		if (reversebrackets)
			uc = '[';
		break;
	case '{':
		uc = '}';
		break;
	case '}':
		uc = '{';
		break;
	case '<':
		uc = '>';
		break;
	case '>':
		uc = '<';
		break;
	}

	return uc;
}


void Paragraph::setFont(pos_type pos, Font const & font)
{
	LASSERT(pos <= size(), return);

	// Text::setCharFont() already reduces font against layout/label
	// font, so we don't need to do that here. (Asger)
	d->fontlist_.set(pos, font);
}


void Paragraph::makeSameLayout(Paragraph const & par)
{
	d->layout_ = par.d->layout_;
	d->params_ = par.d->params_;
}


bool Paragraph::stripLeadingSpaces(bool trackChanges)
{
	if (isFreeSpacing())
		return false;

	int pos = 0;
	int count = 0;

	while (pos < size() && (isNewline(pos) || isLineSeparator(pos))) {
		if (eraseChar(pos, trackChanges))
			++count;
		else
			++pos;
	}

	return count > 0 || pos > 0;
}


bool Paragraph::hasSameLayout(Paragraph const & par) const
{
	return par.d->layout_ == d->layout_
		&& d->params_.sameLayout(par.d->params_);
}


depth_type Paragraph::getDepth() const
{
	return d->params_.depth();
}


depth_type Paragraph::getMaxDepthAfter() const
{
	if (d->layout_->isEnvironment())
		return d->params_.depth() + 1;
	else
		return d->params_.depth();
}


LyXAlignment Paragraph::getAlign(BufferParams const & bparams) const
{
	if (d->params_.align() == LYX_ALIGN_LAYOUT)
		return getDefaultAlign(bparams);
	else
		return d->params_.align();
}


LyXAlignment Paragraph::getDefaultAlign(BufferParams const & bparams) const
{
	LyXAlignment res = layout().align;
	if (isRTL(bparams)) {
		// Swap sides
		if (res == LYX_ALIGN_LEFT)
			res = LYX_ALIGN_RIGHT;
		else if  (res == LYX_ALIGN_RIGHT)
			res = LYX_ALIGN_LEFT;
	}
	return res;
}


docstring const & Paragraph::labelString() const
{
	return d->params_.labelString();
}


// the next two functions are for the manual labels
docstring const Paragraph::getLabelWidthString() const
{
	if (d->layout_->margintype == MARGIN_MANUAL
	    || d->layout_->latextype == LATEX_BIB_ENVIRONMENT)
		return d->params_.labelWidthString();
	else
		return _("Senseless with this layout!");
}


void Paragraph::setLabelWidthString(docstring const & s)
{
	d->params_.labelWidthString(s);
}


docstring Paragraph::expandLabel(Layout const & layout,
		BufferParams const & bparams) const
{
	return expandParagraphLabel(layout, bparams, true);
}


docstring Paragraph::expandParagraphLabel(Layout const & layout,
		BufferParams const & bparams, bool process_appendix) const
{
	DocumentClass const & tclass = bparams.documentClass();
	string const & lang = getParLanguage(bparams)->code();
	bool const in_appendix = process_appendix && d->params_.appendix();
	docstring fmt = translateIfPossible(layout.labelstring(in_appendix), lang);

	if (fmt.empty() && !layout.counter.empty())
		return tclass.counters().theCounter(layout.counter, lang);

	// handle 'inherited level parts' in 'fmt',
	// i.e. the stuff between '@' in   '@Section@.\arabic{subsection}'
	size_t const i = fmt.find('@', 0);
	if (i != docstring::npos) {
		size_t const j = fmt.find('@', i + 1);
		if (j != docstring::npos) {
			docstring parent(fmt, i + 1, j - i - 1);
			docstring label = from_ascii("??");
			if (tclass.hasLayout(parent))
				label = expandParagraphLabel(tclass[parent], bparams,
						      process_appendix);
			fmt = docstring(fmt, 0, i) + label
				+ docstring(fmt, j + 1, docstring::npos);
		}
	}

	return tclass.counters().counterLabel(fmt, lang);
}


void Paragraph::applyLayout(Layout const & new_layout)
{
	d->layout_ = &new_layout;
	LyXAlignment const oldAlign = d->params_.align();

	if (!(oldAlign & d->layout_->alignpossible)) {
		frontend::Alert::warning(_("Alignment not permitted"),
			_("The new layout does not permit the alignment previously used.\nSetting to default."));
		d->params_.align(LYX_ALIGN_LAYOUT);
	}
}


pos_type Paragraph::beginOfBody() const
{
	return d->begin_of_body_;
}


void Paragraph::setBeginOfBody()
{
	if (d->layout_->labeltype != LABEL_MANUAL) {
		d->begin_of_body_ = 0;
		return;
	}

	// Unroll the first two cycles of the loop
	// and remember the previous character to
	// remove unnecessary getChar() calls
	pos_type i = 0;
	pos_type end = size();
	bool prev_char_deleted = false;
	if (i < end && (!(isNewline(i) || isEnvSeparator(i)) || isDeleted(i))) {
		++i;
		if (i < end) {
			char_type previous_char = d->text_[i];
			if (!(isNewline(i) || isEnvSeparator(i))) {
				++i;
				while (i < end && (previous_char != ' ' || prev_char_deleted)) {
					char_type temp = d->text_[i];
					prev_char_deleted = isDeleted(i);
					if (!isDeleted(i) && (isNewline(i) || isEnvSeparator(i)))
						break;
					++i;
					previous_char = temp;
				}
			}
		}
	}

	d->begin_of_body_ = i;
}


bool Paragraph::allowParagraphCustomization() const
{
	return inInset().allowParagraphCustomization();
}


bool Paragraph::usePlainLayout() const
{
	return inInset().usePlainLayout();
}


bool Paragraph::isPassThru() const
{
	return inInset().isPassThru() || d->layout_->pass_thru;
}


bool Paragraph::parbreakIsNewline() const
{
	return inInset().getLayout().parbreakIsNewline() || d->layout_->parbreak_is_newline;
}


bool Paragraph::allowedInContext(Cursor const & cur, InsetLayout const & il) const
{
	set<docstring> const & allowed_insets = il.allowedInInsets();
	set<docstring> const & allowed_layouts = il.allowedInLayouts();

	bool in_allowed_inset =
		allowed_insets.find(inInset().getLayout().name()) != allowed_insets.end();

	bool in_allowed_layout =
		allowed_layouts.find(d->layout_->name()) != allowed_layouts.end();

	if (!in_allowed_inset && inInset().asInsetArgument()) {
		// check if the argument allows the inset in question
		if (cur.depth() > 1) {
			docstring parlayout = cur[cur.depth() - 2].inset().getLayout().name()
					+ from_ascii("@") + from_ascii(inInset().asInsetArgument()->name());
			if (allowed_insets.find(parlayout) != allowed_insets.end())
				in_allowed_inset = true;
		}
	}
	
	int have_ins = 0;
	// check if we exceed the number of allowed insets in this inset
	if (in_allowed_inset && inInset().asInsetText() && il.allowedOccurrences() != -1) {
		ParagraphList & pars = cur.text()->paragraphs();
			for (Paragraph const & par : pars) {
				for (auto const & elem : par.insetList())
				if (elem.inset->getLayout().name() == il.name())
					++have_ins;
			}
		if (have_ins >= il.allowedOccurrences())
			return false;
	}
	
	have_ins = 0;
	// check if we exceed the number of allowed insets in the layout group
	if (in_allowed_layout && il.allowedOccurrences() != -1) {
		pit_type pit = cur.pit();
		pit_type lastpit = cur.pit();
		ParagraphList & pars = cur.text()->paragraphs();
		// If we are not on a list-type environment or AllowedOccurrencesPerItem
		// is false, we check the whole paragraph group
		if (d->layout_->isEnvironment()
		    && !(il.allowedOccurrencesPerItem()
			 && (d->layout_->latextype == LATEX_LIST_ENVIRONMENT
			     || d->layout_->latextype == LATEX_ITEM_ENVIRONMENT))) {
			lastpit = cur.lastpit();
			// get the first paragraph in sequence with this layout
			depth_type const current_depth = params().depth();
			while (true) {
				if (pit == 0)
					break;
				Paragraph cpar = pars[pit - 1];
				if (&cpar.layout() == d->layout_
				    && cpar.params().depth() == current_depth)
					--pit;
				else
					break;
			}
		}
		for (; pit <= lastpit; ++pit) {
			if (&pars[pit].layout() != d->layout_)
				break;
			for (auto const & elem : pars[pit].insetList())
				if (elem.inset->getLayout().name() == il.name())
					++have_ins;
		}
		if (have_ins >= il.allowedOccurrences())
			return false;
	}
	
	if (in_allowed_layout || in_allowed_inset)
		return true;

	return (allowed_insets.empty() && allowed_layouts.empty());
}


bool Paragraph::isPartOfTextSequence() const
{
	for (pos_type i = 0; i < size(); ++i) {
		if (!isInset(i) || getInset(i)->isPartOfTextSequence())
			return true;
	}
	return false;
}

namespace {

// paragraphs inside floats need different alignment tags to avoid
// unwanted space

bool noTrivlistCentering(InsetCode code)
{
	return code == FLOAT_CODE
	       || code == WRAP_CODE
	       || code == CELL_CODE;
}


string correction(string const & orig)
{
	if (orig == "flushleft")
		return "raggedright";
	if (orig == "flushright")
		return "raggedleft";
	if (orig == "center")
		return "centering";
	return orig;
}


bool corrected_env(otexstream & os, string const & suffix, string const & env,
	InsetCode code, bool const lastpar, int & col)
{
	string macro = suffix + "{";
	if (noTrivlistCentering(code)) {
		if (lastpar) {
			// the last paragraph in non-trivlist-aligned
			// context is special (to avoid unwanted whitespace)
			if (suffix == "\\begin") {
				macro = "\\" + correction(env) + "{}";
				os << from_ascii(macro);
				col += macro.size();
				return true;
			}
			return false;
		}
		macro += correction(env);
	} else
		macro += env;
	macro += "}";
	if (suffix == "\\par\\end") {
		os << breakln;
		col = 0;
	}
	os << from_ascii(macro);
	col += macro.size();
	if (suffix == "\\begin") {
		os << breakln;
		col = 0;
	}
	return true;
}

} // namespace


int Paragraph::Private::startTeXParParams(BufferParams const & bparams,
			otexstream & os, OutputParams const & runparams) const
{
	int column = 0;

	bool canindent =
		(bparams.paragraph_separation == BufferParams::ParagraphIndentSeparation) ?
			(layout_->toggle_indent != ITOGGLE_NEVER) :
			(layout_->toggle_indent == ITOGGLE_ALWAYS);

	LyXAlignment const curAlign = params_.align();

	// Do not output \\noindent for paragraphs
	// 1. that cannot have indentation or are indented always,
	// 2. that are not part of the immediate text sequence (e.g., contain only floats),
	// 3. that are PassThru,
	// 4. or that are centered.
	if (canindent && params_.noindent()
	    && owner_->isPartOfTextSequence()
	    && !layout_->pass_thru
	    && curAlign != LYX_ALIGN_CENTER) {
		if (!owner_->empty()
		    && owner_->getInset(0)
		    && owner_->getInset(0)->lyxCode() == VSPACE_CODE)
			// If the paragraph starts with a vspace, the \\noindent
			// needs to come after that (as it leaves vmode).
			// If the paragraph consists only of the vspace,
			// \\noindent is not needed at all.
			runparams.need_noindent = owner_->size() > 1;
		else {
			os << "\\noindent" << termcmd;
			column += 10;
		}
	}

	if (curAlign == layout_->align)
		return column;

	switch (curAlign) {
	case LYX_ALIGN_NONE:
	case LYX_ALIGN_BLOCK:
	case LYX_ALIGN_LAYOUT:
	case LYX_ALIGN_SPECIAL:
	case LYX_ALIGN_DECIMAL:
		break;
	case LYX_ALIGN_LEFT:
	case LYX_ALIGN_RIGHT:
	case LYX_ALIGN_CENTER:
		if (runparams.moving_arg) {
			os << "\\protect";
			column += 8;
		}
		break;
	}

	string const begin_tag = "\\begin";
	InsetCode code = ownerCode();
	bool const lastpar = runparams.isLastPar;
	// RTL in classic (PDF)LaTeX (without the Bidi package)
	// Luabibdi (used by LuaTeX) behaves like classic
	bool const rtl_classic = owner_->getParLanguage(bparams)->rightToLeft()
		&& !bparams.useBidiPackage(runparams);

	switch (curAlign) {
	case LYX_ALIGN_NONE:
	case LYX_ALIGN_BLOCK:
	case LYX_ALIGN_LAYOUT:
	case LYX_ALIGN_SPECIAL:
	case LYX_ALIGN_DECIMAL:
		break;
	case LYX_ALIGN_LEFT: {
		if (rtl_classic)
			// Classic (PDF)LaTeX switches the left/right logic in RTL mode
			corrected_env(os, begin_tag, "flushright", code, lastpar, column);
		else
			corrected_env(os, begin_tag, "flushleft", code, lastpar, column);
		break;
	} case LYX_ALIGN_RIGHT: {
		if (rtl_classic)
			// Classic (PDF)LaTeX switches the left/right logic in RTL mode
			corrected_env(os, begin_tag, "flushleft", code, lastpar, column);
		else
			corrected_env(os, begin_tag, "flushright", code, lastpar, column);
		break;
	} case LYX_ALIGN_CENTER: {
		corrected_env(os, begin_tag, "center", code, lastpar, column);
		break;
	}
	}

	return column;
}


bool Paragraph::Private::endTeXParParams(BufferParams const & bparams,
			otexstream & os, OutputParams const & runparams) const
{
	LyXAlignment const curAlign = params_.align();

	if (curAlign == layout_->align)
		return false;

	switch (curAlign) {
	case LYX_ALIGN_NONE:
	case LYX_ALIGN_BLOCK:
	case LYX_ALIGN_LAYOUT:
	case LYX_ALIGN_SPECIAL:
	case LYX_ALIGN_DECIMAL:
		break;
	case LYX_ALIGN_LEFT:
	case LYX_ALIGN_RIGHT:
	case LYX_ALIGN_CENTER:
		if (runparams.moving_arg)
			os << "\\protect";
		break;
	}

	bool output = false;
	int col = 0;
	string const end_tag = "\\par\\end";
	InsetCode code = ownerCode();
	bool const lastpar = runparams.isLastPar;
	// RTL in classic (PDF)LaTeX (without the Bidi package)
	// Luabibdi (used by LuaTeX) behaves like classic
	bool const rtl_classic = owner_->getParLanguage(bparams)->rightToLeft()
		&& !bparams.useBidiPackage(runparams);

	switch (curAlign) {
	case LYX_ALIGN_NONE:
	case LYX_ALIGN_BLOCK:
	case LYX_ALIGN_LAYOUT:
	case LYX_ALIGN_SPECIAL:
	case LYX_ALIGN_DECIMAL:
		break;
	case LYX_ALIGN_LEFT: {
		if (rtl_classic)
			// Classic (PDF)LaTeX switches the left/right logic in RTL mode
			output = corrected_env(os, end_tag, "flushright", code, lastpar, col);
		else
			output = corrected_env(os, end_tag, "flushleft", code, lastpar, col);
		break;
	} case LYX_ALIGN_RIGHT: {
		if (rtl_classic)
			// Classic (PDF)LaTeX switches the left/right logic in RTL mode
			output = corrected_env(os, end_tag, "flushleft", code, lastpar, col);
		else
			output = corrected_env(os, end_tag, "flushright", code, lastpar, col);
		break;
	} case LYX_ALIGN_CENTER: {
		corrected_env(os, end_tag, "center", code, lastpar, col);
		break;
	}
	}

	return output || lastpar;
}


// This one spits out the text of the paragraph
void Paragraph::latex(BufferParams const & bparams,
	Font const & outerfont,
	otexstream & os,
	OutputParams const & runparams,
	int start_pos, int end_pos, bool force) const
{
	LYXERR(Debug::OUTFILE, "Paragraph::latex...     " << this);

	// FIXME This check should not be needed. Perhaps issue an
	// error if it triggers.
	Layout const & style = inInset().forcePlainLayout() ?
		bparams.documentClass().plainLayout() : *d->layout_;

	if (!force && style.inpreamble)
		return;

	bool const allowcust = allowParagraphCustomization();

	// Current base font for all inherited font changes, without any
	// change caused by an individual character, except for the language:
	// It is set to the language of the first character.
	// As long as we are in the label, this font is the base font of the
	// label. Before the first body character it is set to the base font
	// of the body.
	Font basefont;

	// If there is an open font-encoding changing command (script wrapper),
	// alien_script is set to its name
	string alien_script;
	string script;

	// Maybe we have to create a optional argument.
	pos_type body_pos = beginOfBody();
	unsigned int column = 0;

	// If we are inside an non inheritFont() inset,
	// the outerfont is the buffer's main font
	Font const real_outerfont =
		inInset().inheritFont() ? outerfont : Font(bparams.getFont());

	if (body_pos > 0) {
		// the optional argument is kept in curly brackets in
		// case it contains a ']'
		// This is not strictly needed, but if this is changed it
		// would be a file format change, and tex2lyx would need
		// to be adjusted, since it unconditionally removes the
		// braces when it parses \item.
		os << "[{";
		column += 2;
		basefont = getLabelFont(bparams, real_outerfont);
	} else {
		basefont = getLayoutFont(bparams, real_outerfont);
	}

	// Which font is currently active?
	Font running_font(basefont);
	// Do we have an open font change?
	bool open_font = false;

	Change runningChange =
	    runparams.inDeletedInset && !inInset().canTrackChanges()
	    ? runparams.changeOfDeletedInset : Change(Change::UNCHANGED);

	Encoding const * const prev_encoding = runparams.encoding;

	os.texrow().start(id(), 0);

	// if the paragraph is empty, the loop will not be entered at all
	if (empty()) {
		// For InTitle commands, we have already opened a group
		// in output_latex::TeXOnePar.
		if (style.isCommand() && !style.intitle) {
			os << '{';
			++column;
		}
		if (!style.leftdelim().empty()) {
			os << style.leftdelim();
			column += style.leftdelim().size();
		}
		if (allowcust)
			column += d->startTeXParParams(bparams, os, runparams);
	}

	// Whether a \par can be issued for insets typeset inline with text.
	// Yes if greater than 0. This has to be static.
	THREAD_LOCAL_STATIC int parInline = 0;

	for (pos_type i = 0; i < size(); ++i) {
		// First char in paragraph or after label?
		if (i == body_pos) {
			if (body_pos > 0) {
				if (open_font) {
					bool needPar = false;
					column += running_font.latexWriteEndChanges(
						os, bparams, runparams,
						basefont, basefont, needPar);
					open_font = false;
				}
				basefont = getLayoutFont(bparams, real_outerfont);
				running_font = basefont;

				column += Changes::latexMarkChange(os, bparams,
						runningChange, Change(Change::UNCHANGED),
						runparams);
				runningChange = Change(Change::UNCHANGED);

				os << ((isEnvSeparator(i) && !runparams.find_effective()) ? "}]~" : "}] ");
				column +=3;
			}
			// For InTitle commands, we have already opened a group
			// in output_latex::TeXOnePar.
			if (style.isCommand() && !style.intitle) {
				os << '{';
				++column;
			}

			if (!style.leftdelim().empty()) {
				os << style.leftdelim();
				column += style.leftdelim().size();
			}

			if (allowcust)
				column += d->startTeXParParams(bparams, os,
							    runparams);
		}

		runparams.wasDisplayMath = runparams.inDisplayMath;
		runparams.inDisplayMath = false;
		bool deleted_display_math = false;
		Change const & change = runparams.inDeletedInset
			? runparams.changeOfDeletedInset : lookupChange(i);

		char_type const c = d->text_[i];

		// Check whether a display math inset follows
		bool output_changes;
		if (!runparams.find_effective())
			output_changes = bparams.output_changes;
		else
			output_changes = runparams.find_with_deleted();
		if (c == META_INSET
		    && i >= start_pos && (end_pos == -1 || i < end_pos)) {
			if (isDeleted(i))
				runparams.ctObject = getInset(i)->getCtObject(runparams);
	
			InsetMath const * im = getInset(i)->asInsetMath();
			if (im && im->asHullInset()
			    && im->asHullInset()->outerDisplay()) {
				runparams.inDisplayMath = true;
				// runparams.inDeletedInset will be set by
				// latexInset later, but we need this info
				// before it is called. On the other hand, we
				// cannot set it here because it is a counter.
				deleted_display_math = isDeleted(i);
			}
			if (output_changes && deleted_display_math
			    && runningChange == change
			    && change.type == Change::DELETED
			    && !os.afterParbreak()) {
				// A display math in the same paragraph follows.
				// We have to close and then reopen \lyxdeleted,
				// otherwise the math will be shifted up.
				OutputParams rp = runparams;
				if (open_font) {
					bool needPar = false;
					column += running_font.latexWriteEndChanges(
						os, bparams, rp, basefont,
						basefont, needPar);
					open_font = false;
				}
				basefont = (body_pos > i) ? getLabelFont(bparams, real_outerfont)
							  : getLayoutFont(bparams, real_outerfont);
				running_font = basefont;
				column += Changes::latexMarkChange(os, bparams,
					Change(Change::INSERTED), change, rp);
			}
		}

		if (output_changes && runningChange != change) {
			if (!alien_script.empty()) {
				column += 1;
				os << "}";
				alien_script.clear();
			}
			if (open_font) {
				bool needPar = false;
				column += running_font.latexWriteEndChanges(
						os, bparams, runparams,
						basefont, basefont, needPar);
				open_font = false;
			}
			basefont = (body_pos > i) ? getLabelFont(bparams, real_outerfont)
						  : getLayoutFont(bparams, real_outerfont);
			running_font = basefont;
			column += Changes::latexMarkChange(os, bparams, runningChange,
							   change, runparams);
			runningChange = change;
		}

		// do not output text which is marked deleted
		// if change tracking output is disabled
		if (!output_changes && change.deleted()) {
			continue;
		}

		++column;

		// Fully instantiated font
		Font current_font = getFont(bparams, i, outerfont);
		// Previous font
		Font const prev_font = (i > 0) ?
					getFont(bparams, i - 1, outerfont)
				      : current_font;

		Font const last_font = running_font;
		bool const in_ct_deletion = (output_changes
					     && runningChange == change
					     && change.type == Change::DELETED
					     && !os.afterParbreak());
		// Insets where font switches are used (rather than font commands)
		bool const fontswitch_inset =
				c == META_INSET
				&& getInset(i)
				&& getInset(i)->allowMultiPar()
				&& getInset(i)->lyxCode() != ERT_CODE
				&& (getInset(i)->producesOutput()
				    // FIXME Something more general?
				    // Comments do not "produce output" but are still
				    // part of the TeX source and require font switches
				    // to be closed (otherwise LaTeX fails).
				    || getInset(i)->layoutName() == "Note:Comment");

		bool closeLanguage = false;
		bool lang_switched_at_inset = false;
		if (fontswitch_inset) {
			// Some insets cannot be inside a font change command.
			// However, even such insets *can* be placed in \L or \R
			// or their equivalents (for RTL language switches),
			// so we don't close the language in those cases
			// (= differing isRightToLeft()).
			// ArabTeX, though, doesn't seem to handle this special behavior.
			closeLanguage = basefont.isRightToLeft() == current_font.isRightToLeft()
					|| basefont.language()->lang() == "arabic_arabtex"
					|| current_font.language()->lang() == "arabic_arabtex";
			// We need to check prev_font as language changes directly at inset
			// will only be started inside the inset.
			lang_switched_at_inset = prev_font.language() != current_font.language();
		}

		// Do we need to close the previous font?
		bool langClosed = false;
		if (open_font &&
		    ((current_font != running_font
		      || current_font.language() != running_font.language())
		     || (fontswitch_inset
			 && (current_font == prev_font))))
		{
			// ensure there is no open script-wrapper
			if (!alien_script.empty()) {
				column += 1;
				os << "}";
				alien_script.clear();
			}
			if (in_ct_deletion) {
				// We have to close and then reopen \lyxdeleted,
				// as strikeout needs to be on lowest level.
				os << '}';
				column += 1;
			}
			if (closeLanguage) {
				// Force language closing
				current_font.setLanguage(basefont.language());
				langClosed = true;
			}
			Font const nextfont = (i == body_pos-1) ? basefont : current_font;
			bool needPar = false;
			column += running_font.latexWriteEndChanges(
				    os, bparams, runparams, basefont,
				    nextfont, needPar);
			if (in_ct_deletion) {
				// We have to close and then reopen \lyxdeleted,
				// as strikeout needs to be on lowest level.
				OutputParams rp = runparams;
				column += Changes::latexMarkChange(os, bparams,
					Change(Change::UNCHANGED), Change(Change::DELETED), rp);
			}
			// Has the language been closed in the latexWriteEndChanges() call above?
			langClosed |= running_font.language() != basefont.language()
					&& running_font.language() != nextfont.language()
					&& (running_font.language()->encoding()->package() != Encoding::CJK);
			running_font = basefont;
			open_font &= !langClosed;
		}

		// if necessary, close language environment before opening CJK
		string const running_lang = running_font.language()->babel();
		string const lang_end_command = lyxrc.language_command_end;
		if (!lang_end_command.empty() && !bparams.useNonTeXFonts
			&& !running_lang.empty()
			&& running_lang == openLanguageName()
			&& current_font.language()->encoding()->package() == Encoding::CJK) {
			string end_tag = subst(lang_end_command, "$$lang", running_lang);
			os << from_ascii(end_tag);
			column += end_tag.length();
			if (!languageStackEmpty())
				popLanguageName();
		}

		// Switch file encoding if necessary (and allowed)
		if ((!fontswitch_inset || closeLanguage)
		    && !runparams.pass_thru && !style.pass_thru &&
		    runparams.encoding->package() != Encoding::none &&
		    current_font.language()->encoding()->package() != Encoding::none) {
			pair<bool, int> const enc_switch =
			    	switchEncoding(os.os(), bparams, runparams,
					*(current_font.language()->encoding()));
			if (enc_switch.first) {
				column += enc_switch.second;
				runparams.encoding = current_font.language()->encoding();
			}
		}

		// A display math inset inside an ulem command will be output
		// as a box of width \linewidth, so we have to either disable
		// indentation if the inset starts a paragraph, or start a new
		// line to accommodate such box. This has to be done before
		// writing any font changing commands.
		if (runparams.inDisplayMath && !deleted_display_math
		    && runparams.inulemcmd) {
			if (os.afterParbreak())
				os << "\\noindent";
			else
				os << "\\\\\n";
		}

		// Do we need to change font?
		if ((current_font != running_font ||
		     current_font.language() != running_font.language())
		    && i != body_pos - 1)
		{
			if (!fontswitch_inset) {
				if (in_ct_deletion) {
					// We have to close and then reopen \lyxdeleted,
					// as strikeout needs to be on lowest level.
					OutputParams rp = runparams;
					bool needPar = false;
					column += running_font.latexWriteEndChanges(
						os, bparams, rp, basefont,
						basefont, needPar);
					os << '}';
					column += 1;
				}
				otexstringstream ots;
				InsetText const * textinset = inInset().asInsetText();
				bool const cprotect = textinset
					? textinset->hasCProtectContent(runparams.moving_arg)
					  && !textinset->text().isMainText()
					  && inInset().lyxCode() != BRANCH_CODE
					  && !runparams.no_cprotect
					: false;
				column += current_font.latexWriteStartChanges(ots, bparams,
									      runparams, basefont, last_font, false,
									      cprotect);
				// Check again for display math in ulem commands as a
				// font change may also occur just before a math inset.
				if (runparams.inDisplayMath && !deleted_display_math
				    && runparams.inulemcmd) {
					if (os.afterParbreak())
						os << "\\noindent";
					else
						os << "\\\\\n";
				}
				running_font = current_font;
				open_font = true;
				docstring fontchange = ots.str();
				os << fontchange;
				// check whether the fontchange ends with a \\textcolor
				// modifier and the text starts with a space. If so we
				// need to add } in order to prevent \\textcolor from gobbling
				// the space (bug 4473).
				docstring const last_modifier = rsplit(fontchange, '\\');
				if (prefixIs(last_modifier, from_ascii("textcolor")) && c == ' ')
					os << from_ascii("{}");
				else if (ots.terminateCommand())
					os << termcmd;
				if (in_ct_deletion) {
					// We have to close and then reopen \lyxdeleted,
					// as strikeout needs to be on lowest level.
					OutputParams rp = runparams;
					column += Changes::latexMarkChange(os, bparams,
						Change(Change::UNCHANGED), change, rp);
				}
			} else {// if fontswitch_inset
				if (current_font != running_font || !langClosed)
					// font is still open in fontswitch_insets if we have
					// a non-lang font difference or if the language
					// is the only difference but has not been forcedly
					// closed meanwhile
					open_font = true;
				running_font = current_font;
			}
		}

		// FIXME: think about end_pos implementation...
		if (c == ' ' && i >= start_pos && (end_pos == -1 || i < end_pos)) {
			// FIXME: integrate this case in latexSpecialChar
			// Do not print the separation of the optional argument
			// if style.pass_thru is false. This works because
			// latexSpecialChar ignores spaces if
			// style.pass_thru is false.
			if (i != body_pos - 1) {
				if (d->simpleTeXBlanks(bparams, runparams, os,
						i, column, current_font, style)) {
					// A surrogate pair was output. We
					// must not call latexSpecialChar
					// in this iteration, since it would output
					// the combining character again.
					++i;
					continue;
				}
			}
		}

		OutputParams rp = runparams;
		rp.free_spacing = style.free_spacing;
		rp.local_font = &current_font;
		rp.intitle |= style.intitle;

		// Two major modes:  LaTeX or plain
		// Handle here those cases common to both modes
		// and then split to handle the two modes separately.
		if (c == META_INSET) {
			if (i >= start_pos && (end_pos == -1 || i < end_pos)) {
				// Greyedout notes and, in general, all insets
				// with InsetLayout::isDisplay() == false,
				// are typeset inline with the text. So, we
				// can add a \par to the last paragraph of
				// such insets only if nothing else follows.
				bool incremented = false;
				Inset const * inset = getInset(i);
				InsetText const * textinset = inset
							? inset->asInsetText()
							: nullptr;
				if (i + 1 == size() && textinset
				    && !inset->getLayout().isDisplay()) {
					ParagraphList const & pars =
						textinset->text().paragraphs();
					pit_type const pit = pars.size() - 1;
					Font const lastfont =
						pit < 0 || pars[pit].empty()
						? pars[pit].getLayoutFont(
								bparams,
								real_outerfont)
						: pars[pit].getFont(bparams,
							pars[pit].size() - 1,
							real_outerfont);
					if (lastfont.fontInfo().size() !=
					    basefont.fontInfo().size()) {
						++parInline;
						incremented = true;
					}
				}
				// We need to restore parts of this after insets with
				// allowMultiPar() true
				Font const save_basefont = basefont;
				d->latexInset(bparams, os, rp, running_font,
						basefont, real_outerfont, open_font,
						runningChange, style, i, column, fontswitch_inset,
						closeLanguage, (lang_switched_at_inset || langClosed));
				if (fontswitch_inset) {
					if (open_font) {
						bool needPar = false;
						column += running_font.latexWriteEndChanges(
							os, bparams, runparams,
							basefont, basefont, needPar);
						open_font = false;
					}
					basefont.fontInfo().setSize(save_basefont.fontInfo().size());
					basefont.fontInfo().setFamily(save_basefont.fontInfo().family());
					basefont.fontInfo().setSeries(save_basefont.fontInfo().series());
				}
				if (incremented)
					--parInline;

				if (runparams.ctObject == CtObject::DisplayObject
				    || runparams.ctObject == CtObject::UDisplayObject) {
					// Close \lyx*deleted and force its
					// reopening (if needed)
					os << '}';
					column++;
					runningChange = Change(Change::UNCHANGED);
					runparams.ctObject = CtObject::Normal;
				}
			}
		} else if (i >= start_pos && (end_pos == -1 || i < end_pos)) {
			if (!bparams.useNonTeXFonts && !runparams.pass_thru
			    && !contains(runparams.pass_thru_chars, c))
				script = Encodings::isKnownScriptChar(c);
			if (script != alien_script) {
				if (!alien_script.empty()) {
					os << "}";
					alien_script.clear();
				}
				string fontenc = running_font.language()->fontenc(bparams);
				if (!script.empty()
					&& !Encodings::fontencSupportsScript(fontenc, script)) {
					column += script.length() + 2;
					os << "\\" << script << "{";
					alien_script = script;
				}
			}
			try {
				d->latexSpecialChar(os, bparams, rp, running_font, alien_script,
						    style, inInset().getLayout(), i, end_pos, column);
			} catch (EncodingException & e) {
				if (runparams.dryrun) {
					os << "<" << _("LyX Warning: ")
					   << _("uncodable character") << " '";
					os.put(c);
					os << "'>";
				} else {
					// add location information and throw again.
					e.par_id = id();
					e.pos = i;
					throw;
				}
			}
		}

		// Set the encoding to that returned from latexSpecialChar (see
		// comment for encoding member in OutputParams.h)
		runparams.encoding = rp.encoding;

		// Also carry on the info on a closed ulem command for insets
		// such as Note that do not produce any output, so that no
		// command is ever executed but its opening was recorded.
		runparams.inulemcmd = rp.inulemcmd;

		// These need to be passed upstream as well
		runparams.need_maketitle = rp.need_maketitle;
		runparams.have_maketitle = rp.have_maketitle;

		// And finally, pass the post_macros upstream
		runparams.post_macro = rp.post_macro;
	}

	// Close wrapper for alien script
	if (!alien_script.empty()) {
		os << "}";
		alien_script.clear();
	}

	Font const font = empty()
		? getLayoutFont(bparams, real_outerfont)
		: getFont(bparams, size() - 1, real_outerfont);

	InsetText const * textinset = inInset().asInsetText();

	bool const maintext = textinset
		? textinset->text().isMainText()
		: false;

	size_t const numpars = textinset
		? textinset->text().paragraphs().size()
		: 0;

	bool needPar = false;

	if (style.resfont.size() != font.fontInfo().size()
	    && (!runparams.isLastPar || maintext
		|| (numpars > 1 && d->ownerCode() != CELL_CODE
		    && (inInset().getLayout().isDisplay()
			|| parInline)))
	    && !style.isCommand()) {
		needPar = true;
	}

	// If we have an open font definition, we have to close it
	if (open_font) {
		// Make sure that \\par is done with the font of the last
		// character if this has another size as the default.
		// This is necessary because LaTeX (and LyX on the screen)
		// calculates the space between the baselines according
		// to this font. (Matthias)
		//
		// We must not change the font for the last paragraph
		// of non-multipar insets, tabular cells or commands,
		// since this produces unwanted whitespace.
#ifdef FIXED_LANGUAGE_END_DETECTION
		if (next_) {
			running_font.latexWriteEndChanges(os, bparams,
					runparams, basefont,
					next_->getFont(bparams, 0, outerfont),
						       needPar);
		} else {
			running_font.latexWriteEndChanges(os, bparams,
					runparams, basefont, basefont, needPar);
		}
#else
//FIXME: For now we ALWAYS have to close the foreign font settings if they are
//FIXME: there as we start another \selectlanguage with the next paragraph if
//FIXME: we are in need of this. This should be fixed sometime (Jug)
		running_font.latexWriteEndChanges(os, bparams, runparams,
				basefont, basefont, needPar);
#endif
	}
	if (needPar) {
		// The \par could not be inserted at the same nesting
		// level of the font size change, so do it now.
		os << "{\\" << font.latexSize() << "\\par}";
	}

	if (!runparams.inDeletedInset || inInset().canTrackChanges())
		column += Changes::latexMarkChange(os, bparams, runningChange,
					Change(Change::UNCHANGED), runparams);

	// Needed if there is an optional argument but no contents.
	if (body_pos > 0 && body_pos == size()) {
		os << "}]~";
	}

	if (!style.rightdelim().empty()) {
		os << style.rightdelim();
		column += style.rightdelim().size();
	}

	if (allowcust && d->endTeXParParams(bparams, os, runparams)
	    && runparams.encoding != prev_encoding) {
		runparams.encoding = prev_encoding;
		os << setEncoding(prev_encoding->iconvName());
	}

	LYXERR(Debug::OUTFILE, "Paragraph::latex... done " << this);
}


bool Paragraph::emptyTag() const
{
	for (pos_type i = 0; i < size(); ++i) {
		if (Inset const * inset = getInset(i)) {
			InsetCode lyx_code = inset->lyxCode();
			// FIXME testing like that is wrong. What is
			// the intent?
			if (lyx_code != TOC_CODE &&
			    lyx_code != INCLUDE_CODE &&
			    lyx_code != GRAPHICS_CODE &&
			    lyx_code != ERT_CODE &&
			    lyx_code != LISTINGS_CODE &&
			    lyx_code != FLOAT_CODE &&
			    lyx_code != TABULAR_CODE) {
				return false;
			}
		} else {
			char_type c = d->text_[i];
			if (c != ' ' && c != '\t')
				return false;
		}
	}
	return true;
}


string Paragraph::getID(Buffer const &, OutputParams const &)
	const
{
	for (pos_type i = 0; i < size(); ++i) {
		if (Inset const * inset = getInset(i)) {
			InsetCode lyx_code = inset->lyxCode();
			if (lyx_code == LABEL_CODE) {
				InsetLabel const * const il = static_cast<InsetLabel const *>(inset);
				docstring const & id = il->getParam("name");
				return "id='" + to_utf8(xml::cleanID(id)) + "'";
			}
		}
	}
	return string();
}


string Paragraph::getLabelForXRef() const
{
	for (pos_type i = 0; i < size(); ++i) {
		if (Inset const * inset = getInset(i)) {
			InsetCode lyx_code = inset->lyxCode();
			if (lyx_code == LABEL_CODE) {
				InsetLabel const * const il = static_cast<InsetLabel const *>(inset);
				return to_utf8(il->getParam("name"));
			}
		}
	}
	return string();
}


pos_type Paragraph::firstWordDocBook(XMLStream & xs, OutputParams const & runparams) const
{
	pos_type i;
	for (i = 0; i < size(); ++i) {
		if (Inset const * inset = getInset(i)) {
			inset->docbook(xs, runparams);
		} else {
			char_type c = d->text_[i];
			if (c == ' ')
				break;
			xs << c;
		}
	}
	return i;
}


pos_type Paragraph::firstWordLyXHTML(XMLStream & xs, OutputParams const & runparams)
	const
{
	pos_type i;
	for (i = 0; i < size(); ++i) {
		if (Inset const * inset = getInset(i)) {
			inset->xhtml(xs, runparams);
		} else {
			char_type c = d->text_[i];
			if (c == ' ')
				break;
			xs << c;
		}
	}
	return i;
}


bool Paragraph::Private::onlyText(Buffer const & buf, Font const & outerfont, pos_type initial) const
{
	Font font_old;
	pos_type size = text_.size();
	for (pos_type i = initial; i < size; ++i) {
		Font font = owner_->getFont(buf.params(), i, outerfont);
		if (text_[i] == META_INSET)
			return false;
		if (i != initial && font != font_old)
			return false;
		font_old = font;
	}

	return true;
}


namespace {

void doFontSwitchDocBook(vector<xml::FontTag> & tagsToOpen,
                  vector<xml::EndFontTag> & tagsToClose,
                  bool & flag, FontState curstate, xml::FontTypes type)
{
	if (curstate == FONT_ON) {
		tagsToOpen.push_back(docbookStartFontTag(type));
		flag = true;
	} else if (flag) {
		tagsToClose.push_back(docbookEndFontTag(type));
		flag = false;
	}
}

class OptionalFontType {
public:
	xml::FontTypes ft;
	bool has_value;

	OptionalFontType(): ft(xml::FT_EMPH), has_value(false) {} // A possible value at random for ft.
	OptionalFontType(xml::FontTypes ft): ft(ft), has_value(true) {}
};

OptionalFontType fontShapeToXml(FontShape fs)
{
	switch (fs) {
	case ITALIC_SHAPE:
		return {xml::FT_ITALIC};
	case SLANTED_SHAPE:
		return {xml::FT_SLANTED};
	case SMALLCAPS_SHAPE:
		return {xml::FT_SMALLCAPS};
	case UP_SHAPE:
	case INHERIT_SHAPE:
		return {};
	default:
		// the other tags are for internal use
		LATTEST(false);
		return {};
	}
}

OptionalFontType fontFamilyToXml(FontFamily fm)
{
	switch (fm) {
	case ROMAN_FAMILY:
		return {xml::FT_ROMAN};
	case SANS_FAMILY:
		return {xml::FT_SANS};
	case TYPEWRITER_FAMILY:
		return {xml::FT_TYPE};
	case INHERIT_FAMILY:
		return {};
	default:
		// the other tags are for internal use
		LATTEST(false);
		return {};
	}
}

OptionalFontType fontSizeToXml(FontSize fs)
{
	switch (fs) {
	case TINY_SIZE:
		return {xml::FT_SIZE_TINY};
	case SCRIPT_SIZE:
		return {xml::FT_SIZE_SCRIPT};
	case FOOTNOTE_SIZE:
		return {xml::FT_SIZE_FOOTNOTE};
	case SMALL_SIZE:
		return {xml::FT_SIZE_SMALL};
	case LARGE_SIZE:
		return {xml::FT_SIZE_LARGE};
	case LARGER_SIZE:
		return {xml::FT_SIZE_LARGER};
	case LARGEST_SIZE:
		return {xml::FT_SIZE_LARGEST};
	case HUGE_SIZE:
		return {xml::FT_SIZE_HUGE};
	case HUGER_SIZE:
		return {xml::FT_SIZE_HUGER};
	case INCREASE_SIZE:
		return {xml::FT_SIZE_INCREASE};
	case DECREASE_SIZE:
		return {xml::FT_SIZE_DECREASE};
	case INHERIT_SIZE:
	case NORMAL_SIZE:
		return {};
	default:
		// the other tags are for internal use
		LATTEST(false);
		return {};
	}
}

struct DocBookFontState
{
	FontShape  curr_fs   = INHERIT_SHAPE;
	FontFamily curr_fam  = INHERIT_FAMILY;
	FontSize   curr_size = INHERIT_SIZE;

	// track whether we have opened these tags
	bool emph_flag = false;
	bool bold_flag = false;
	bool noun_flag = false;
	bool ubar_flag = false;
	bool dbar_flag = false;
	bool sout_flag = false;
	bool xout_flag = false;
	bool wave_flag = false;
	// shape tags
	bool shap_flag = false;
	// family tags
	bool faml_flag = false;
	// size tags
	bool size_flag = false;
};

std::tuple<vector<xml::FontTag>, vector<xml::EndFontTag>> computeDocBookFontSwitch(FontInfo const & font_old,
		                                                                           Font const & font,
		                                                                           std::string const & default_family,
		                                                                           DocBookFontState & fs)
{
	vector<xml::FontTag> tagsToOpen;
	vector<xml::EndFontTag> tagsToClose;

	// emphasis
	FontState curstate = font.fontInfo().emph();
	if (font_old.emph() != curstate)
		doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.emph_flag, curstate, xml::FT_EMPH);

	// noun
	curstate = font.fontInfo().noun();
	if (font_old.noun() != curstate)
		doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.noun_flag, curstate, xml::FT_NOUN);

	// underbar
	curstate = font.fontInfo().underbar();
	if (font_old.underbar() != curstate)
		doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.ubar_flag, curstate, xml::FT_UBAR);

	// strikeout
	curstate = font.fontInfo().strikeout();
	if (font_old.strikeout() != curstate)
		doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.sout_flag, curstate, xml::FT_SOUT);

	// xout
	curstate = font.fontInfo().xout();
	if (font_old.xout() != curstate)
		doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.xout_flag, curstate, xml::FT_XOUT);

	// double underbar
	curstate = font.fontInfo().uuline();
	if (font_old.uuline() != curstate)
		doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.dbar_flag, curstate, xml::FT_DBAR);

	// wavy line
	curstate = font.fontInfo().uwave();
	if (font_old.uwave() != curstate)
		doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.wave_flag, curstate, xml::FT_WAVE);

	// bold
	// a little hackish, but allows us to reuse what we have.
	curstate = (font.fontInfo().series() == BOLD_SERIES ? FONT_ON : FONT_OFF);
	if (font_old.series() != font.fontInfo().series())
		doFontSwitchDocBook(tagsToOpen, tagsToClose, fs.bold_flag, curstate, xml::FT_BOLD);

	// Font shape
	fs.curr_fs = font.fontInfo().shape();
	FontShape old_fs = font_old.shape();
	if (old_fs != fs.curr_fs) {
		if (fs.shap_flag) {
			OptionalFontType tag = fontShapeToXml(old_fs);
			if (tag.has_value)
				tagsToClose.push_back(docbookEndFontTag(tag.ft));
			fs.shap_flag = false;
		}

		OptionalFontType tag = fontShapeToXml(fs.curr_fs);
		if (tag.has_value)
			tagsToOpen.push_back(docbookStartFontTag(tag.ft));
	}

	// Font family
	fs.curr_fam = font.fontInfo().family();
	FontFamily old_fam = font_old.family();
	if (old_fam != fs.curr_fam) {
		if (fs.faml_flag) {
			OptionalFontType tag = fontFamilyToXml(old_fam);
			if (tag.has_value)
				tagsToClose.push_back(docbookEndFontTag(tag.ft));
			fs.faml_flag = false;
		}
		switch (fs.curr_fam) {
			case ROMAN_FAMILY:
				// we will treat a "default" font family as roman, since we have
				// no other idea what to do.
				if (default_family != "rmdefault" && default_family != "default") {
					tagsToOpen.push_back(docbookStartFontTag(xml::FT_ROMAN));
					fs.faml_flag = true;
				}
				break;
			case SANS_FAMILY:
				if (default_family != "sfdefault") {
					tagsToOpen.push_back(docbookStartFontTag(xml::FT_SANS));
					fs.faml_flag = true;
				}
				break;
			case TYPEWRITER_FAMILY:
				if (default_family != "ttdefault") {
					tagsToOpen.push_back(docbookStartFontTag(xml::FT_TYPE));
					fs.faml_flag = true;
				}
				break;
			case INHERIT_FAMILY:
				break;
			default:
				// the other tags are for internal use
				LATTEST(false);
				break;
		}
	}

	// Font size
	fs.curr_size = font.fontInfo().size();
	FontSize old_size = font_old.size();
	if (old_size != fs.curr_size) {
		if (fs.size_flag) {
			OptionalFontType tag = fontSizeToXml(old_size);
			if (tag.has_value)
				tagsToClose.push_back(docbookEndFontTag(tag.ft));
			fs.size_flag = false;
		}

		OptionalFontType tag = fontSizeToXml(fs.curr_size);
		if (tag.has_value) {
			tagsToOpen.push_back(docbookStartFontTag(tag.ft));
			fs.size_flag = true;
		}
	}

	return std::tuple<vector<xml::FontTag>, vector<xml::EndFontTag>>(tagsToOpen, tagsToClose);
}

} // anonymous namespace


std::tuple<std::vector<docstring>, std::vector<docstring>, std::vector<docstring>>
    Paragraph::simpleDocBookOnePar(Buffer const & buf,
                                   OutputParams const & runparams,
                                   Font const & outerfont,
                                   pos_type initial,
                                   bool is_last_par,
                                   bool ignore_fonts) const
{
	// Return values: segregation of the content of this paragraph.
	std::vector<docstring> prependedParagraphs; // Anything that must be output before the main tag of this paragraph.
	std::vector<docstring> generatedParagraphs; // The main content of the paragraph.
	std::vector<docstring> appendedParagraphs;  // Anything that must be output after the main tag of this paragraph.

	// Internal string stream to store the output before being added to one of the previous lists.
	odocstringstream os;

	// If there is an argument that must be output before the main tag, do it before handling the rest of the paragraph.
	// Also tag all arguments that shouldn't go in the main content right now, so that they are never generated at the
	// wrong place.
	OutputParams rp = runparams;
    for (pos_type i = initial; i < size(); ++i) {
        if (getInset(i) && getInset(i)->lyxCode() == ARG_CODE) {
            const InsetArgument * arg = getInset(i)->asInsetArgument();
            if (arg->docbookargumentbeforemaintag()) {
                auto xs_local = XMLStream(os);
                arg->docbook(xs_local, rp);

                prependedParagraphs.push_back(os.str());
                os.str(from_ascii(""));

                rp.docbook_prepended_arguments.insert(arg);
            } else if (arg->docbookargumentaftermaintag()) {
                rp.docbook_appended_arguments.insert(arg);
            }
        }
    }
	rp.lastid = id();

    // State variables for the main loop.
    auto xs = new XMLStream(os); // XMLStream has no copy constructor: to create a new object, the only solution
    // is to hold a pointer to the XMLStream (xs = XMLStream(os) is not allowed once the first object is built).
	xs->startDivision(false);
    std::vector<docstring> delayedChars; // When a font tag ends with a space, output it after the closing font tag.
    // This requires to store delayed characters at some point.

	// Track whether we have opened font tags
    DocBookFontState fs;
    DocBookFontState old_fs = fs;

    Layout const & style = *d->layout_;

	// Conversion of the font opening/closing into DocBook tags.
    vector<xml::FontTag> tagsToOpen;
    vector<xml::EndFontTag> tagsToClose;

	// Parsing main loop.
	for (pos_type i = initial; i < size(); ++i) {
	    bool ignore_fonts_i = ignore_fonts
                              || style.docbooknofontinside()
                              || (getInset(i) && getInset(i)->getLayout().docbooknofontinside());

		// Don't show deleted material in the output.
		if (isDeleted(i))
			continue;

		// If this is an InsetNewline, generate a new paragraph (this is the reason why generatedParagraphs is a list
		// of paragraphs). Also reset the fonts, so that tags are closed in this paragraph.
		if (getInset(i) && getInset(i)->lyxCode() == NEWLINE_CODE) {
			if (!ignore_fonts_i)
				xs->closeFontTags();

			// Output one paragraph (i.e. one string entry in generatedParagraphs).
			generatedParagraphs.push_back(os.str());

			xs->endDivision();

			// Create a new XMLStream for the new paragraph, completely independent of the previous one. This implies
			// that the string stream must be reset.
			os.str(from_ascii(""));
			delete xs;
			xs = new XMLStream(os);
			xs->startDivision(false);

			// Restore the fonts for the new paragraph, so that the right tags are opened for the new entry.
			if (!ignore_fonts_i) {
				fs = old_fs;
			}
		}

		// Determine which tags should be opened or closed regarding fonts. Consider the last output character (i.e. not
		// deleted).
		int last_output_char = (i == 0) ? 0 : i - 1;
		if (i > 0) {
			while (last_output_char > 0 && isDeleted(last_output_char))
				--last_output_char;
		}
		FontInfo const font_old = (i == 0 ?
				(style.labeltype == LABEL_MANUAL ? style.labelfont : style.font) :
				getFont(buf.masterBuffer()->params(), last_output_char, outerfont).fontInfo());
		Font const font = getFont(buf.masterBuffer()->params(), i, outerfont);
        tie(tagsToOpen, tagsToClose) = computeDocBookFontSwitch(
				font_old, font, buf.masterBuffer()->params().fonts_default_family, fs);

		if (!ignore_fonts_i) {
            vector<xml::EndFontTag>::const_iterator cit = tagsToClose.begin();
            vector<xml::EndFontTag>::const_iterator cen = tagsToClose.end();
            for (; cit != cen; ++cit)
                *xs << *cit;
        }

        // Deal with the delayed characters *after* closing font tags.
        if (!delayedChars.empty()) {
            for (const docstring& c: delayedChars)
                *xs << XMLStream::ESCAPE_NONE << c;
            delayedChars.clear();
        }

        if (!ignore_fonts_i) {
			vector<xml::FontTag>::const_iterator sit = tagsToOpen.begin();
			vector<xml::FontTag>::const_iterator sen = tagsToOpen.end();
			for (; sit != sen; ++sit)
				*xs << *sit;
		}

		// The font tags are no longer useful; free their memory right now.
		tagsToClose.clear();
		tagsToOpen.clear();

        // Finally, write the next character or inset.
		if (Inset const * inset = getInset(i)) {
		    bool inset_is_argument_elsewhere = getInset(i)->asInsetArgument() &&
		            rp.docbook_appended_arguments.find(inset->asInsetArgument()) != rp.docbook_appended_arguments.end() &&
		            rp.docbook_prepended_arguments.find(inset->asInsetArgument()) != rp.docbook_prepended_arguments.end();

			if ((!rp.for_toc || inset->isInToc()) && !inset_is_argument_elsewhere) {
			    // Arguments may need to be output
				OutputParams np = rp;
				np.local_font = &font;

				// TODO: special case will bite here.
				np.docbook_in_par = true;
				inset->docbook(*xs, np);
			}
		} else {
			char_type c = getUChar(buf.masterBuffer()->params(), rp, i);
			if (lyx::isSpace(c) && !ignore_fonts) { // Delay spaces *after* the font-tag closure for cleaner output.
				if (c == ' ' && (style.free_spacing || rp.free_spacing)) {
					delayedChars.push_back(from_ascii("&#160;"));
				} else {
					delayedChars.emplace_back(1, c);
				}
			} else { // No need to delay the character.
				if (c == '\'' && !ignore_fonts)
					*xs << XMLStream::ESCAPE_NONE << "&#8217;";
				else
					*xs << c;
			}
		}
	}

	// Ensure that the tags are closed at the right place. Otherwise, there might be an open font tag with no content
	// that no other code cares to close.
	*xs << xml::NullTag();

	// FIXME, this code is just imported from XHTML
	// I'm worried about what happens if a branch, say, is itself
	// wrapped in some font stuff. I think that will not work.
	if (!ignore_fonts)
		xs->closeFontTags();

	// Close the potentially remaining tags, like pending font tags.
	// There is no need to check for ignore_fonts, as these tags won't be
	// inserted in the stack in the first place if ignore_fonts is false.
	xs->endDivision();

	// Deal with the delayed characters *after* closing font tags.
	if (!delayedChars.empty()) {
		for (const docstring &c: delayedChars)
			*xs << XMLStream::ESCAPE_NONE << c;
		delayedChars.clear();
	}

	// In listings, new lines (i.e. \n characters in the output) are very important. Avoid generating one for the
	// last line to get a clean output.
	if (rp.docbook_in_listing && !is_last_par)
		*xs << xml::CR();

	// Finalise the last (and most likely only) paragraph.
	generatedParagraphs.push_back(os.str());
    os.str(from_ascii(""));
    delete xs;

    // If there is an argument that must be output after the main tag, do it after handling the rest of the paragraph.
    for (pos_type i = initial; i < size(); ++i) {
        if (getInset(i) && getInset(i)->lyxCode() == ARG_CODE) {
            const InsetArgument * arg = getInset(i)->asInsetArgument();
            if (arg->docbookargumentaftermaintag()) {
                // Don't use rp, as this argument would not generate anything.
                auto xs_local = XMLStream(os);
                arg->docbook(xs_local, runparams);

                appendedParagraphs.push_back(os.str());
                os.str(from_ascii(""));
            }
        }
    }

	return std::make_tuple(prependedParagraphs, generatedParagraphs, appendedParagraphs);
}


namespace {

void doFontSwitchXHTML(vector<xml::FontTag> & tagsToOpen,
                  vector<xml::EndFontTag> & tagsToClose,
                  bool & flag, FontState curstate, xml::FontTypes type)
{
	if (curstate == FONT_ON) {
		tagsToOpen.push_back(xhtmlStartFontTag(type));
		flag = true;
	} else if (flag) {
		tagsToClose.push_back(xhtmlEndFontTag(type));
		flag = false;
	}
}

} // anonymous namespace


docstring Paragraph::simpleLyXHTMLOnePar(Buffer const & buf,
				    XMLStream & xs,
				    OutputParams const & runparams,
				    Font const & outerfont,
				    bool start_paragraph, bool close_paragraph,
				    pos_type initial) const
{
	docstring retval;

	// track whether we have opened these tags
	bool emph_flag = false;
	bool bold_flag = false;
	bool noun_flag = false;
	bool ubar_flag = false;
	bool dbar_flag = false;
	bool sout_flag = false;
	bool xout_flag = false;
	bool wave_flag = false;
	// shape tags
	bool shap_flag = false;
	// family tags
	bool faml_flag = false;
	// size tags
	bool size_flag = false;

	Layout const & style = *d->layout_;

	if (start_paragraph)
		xs.startDivision(allowEmpty());

	FontInfo font_old =
		style.labeltype == LABEL_MANUAL ? style.labelfont : style.font;

	FontShape  curr_fs   = INHERIT_SHAPE;
	FontFamily curr_fam  = INHERIT_FAMILY;
	FontSize   curr_size = INHERIT_SIZE;

	string const default_family =
		buf.masterBuffer()->params().fonts_default_family;

	vector<xml::FontTag> tagsToOpen;
	vector<xml::EndFontTag> tagsToClose;

	// parsing main loop
	for (pos_type i = initial; i < size(); ++i) {
		// let's not show deleted material in the output
		if (isDeleted(i))
			continue;

		Font const font = getFont(buf.masterBuffer()->params(), i, outerfont);

		// emphasis
		FontState curstate = font.fontInfo().emph();
		if (font_old.emph() != curstate)
			doFontSwitchXHTML(tagsToOpen, tagsToClose, emph_flag, curstate, xml::FT_EMPH);

		// noun
		curstate = font.fontInfo().noun();
		if (font_old.noun() != curstate)
			doFontSwitchXHTML(tagsToOpen, tagsToClose, noun_flag, curstate, xml::FT_NOUN);

		// underbar
		curstate = font.fontInfo().underbar();
		if (font_old.underbar() != curstate)
			doFontSwitchXHTML(tagsToOpen, tagsToClose, ubar_flag, curstate, xml::FT_UBAR);

		// strikeout
		curstate = font.fontInfo().strikeout();
		if (font_old.strikeout() != curstate)
			doFontSwitchXHTML(tagsToOpen, tagsToClose, sout_flag, curstate, xml::FT_SOUT);

		// xout
		curstate = font.fontInfo().xout();
		if (font_old.xout() != curstate)
			doFontSwitchXHTML(tagsToOpen, tagsToClose, xout_flag, curstate, xml::FT_XOUT);

		// double underbar
		curstate = font.fontInfo().uuline();
		if (font_old.uuline() != curstate)
			doFontSwitchXHTML(tagsToOpen, tagsToClose, dbar_flag, curstate, xml::FT_DBAR);

		// wavy line
		curstate = font.fontInfo().uwave();
		if (font_old.uwave() != curstate)
			doFontSwitchXHTML(tagsToOpen, tagsToClose, wave_flag, curstate, xml::FT_WAVE);

		// bold
		// a little hackish, but allows us to reuse what we have.
		curstate = (font.fontInfo().series() == BOLD_SERIES ? FONT_ON : FONT_OFF);
		if (font_old.series() != font.fontInfo().series())
			doFontSwitchXHTML(tagsToOpen, tagsToClose, bold_flag, curstate, xml::FT_BOLD);

		// Font shape
		curr_fs = font.fontInfo().shape();
		FontShape old_fs = font_old.shape();
		if (old_fs != curr_fs) {
			if (shap_flag) {
				switch (old_fs) {
				case ITALIC_SHAPE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_ITALIC));
					break;
				case SLANTED_SHAPE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SLANTED));
					break;
				case SMALLCAPS_SHAPE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SMALLCAPS));
					break;
				case UP_SHAPE:
				case INHERIT_SHAPE:
					break;
				default:
					// the other tags are for internal use
					LATTEST(false);
					break;
				}
				shap_flag = false;
			}
			switch (curr_fs) {
			case ITALIC_SHAPE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_ITALIC));
				shap_flag = true;
				break;
			case SLANTED_SHAPE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SLANTED));
				shap_flag = true;
				break;
			case SMALLCAPS_SHAPE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SMALLCAPS));
				shap_flag = true;
				break;
			case UP_SHAPE:
			case INHERIT_SHAPE:
				break;
			default:
				// the other tags are for internal use
				LATTEST(false);
				break;
			}
		}

		// Font family
		curr_fam = font.fontInfo().family();
		FontFamily old_fam = font_old.family();
		if (old_fam != curr_fam) {
			if (faml_flag) {
				switch (old_fam) {
				case ROMAN_FAMILY:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_ROMAN));
					break;
				case SANS_FAMILY:
				    tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SANS));
				    break;
				case TYPEWRITER_FAMILY:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_TYPE));
					break;
				case INHERIT_FAMILY:
					break;
				default:
					// the other tags are for internal use
					LATTEST(false);
					break;
				}
				faml_flag = false;
			}
			switch (curr_fam) {
			case ROMAN_FAMILY:
				// we will treat a "default" font family as roman, since we have
				// no other idea what to do.
				if (default_family != "rmdefault" && default_family != "default") {
					tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_ROMAN));
					faml_flag = true;
				}
				break;
			case SANS_FAMILY:
				if (default_family != "sfdefault") {
					tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SANS));
					faml_flag = true;
				}
				break;
			case TYPEWRITER_FAMILY:
				if (default_family != "ttdefault") {
					tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_TYPE));
					faml_flag = true;
				}
				break;
			case INHERIT_FAMILY:
				break;
			default:
				// the other tags are for internal use
				LATTEST(false);
				break;
			}
		}

		// Font size
		curr_size = font.fontInfo().size();
		FontSize old_size = font_old.size();
		if (old_size != curr_size) {
			if (size_flag) {
				switch (old_size) {
				case TINY_SIZE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_TINY));
					break;
				case SCRIPT_SIZE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_SCRIPT));
					break;
				case FOOTNOTE_SIZE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_FOOTNOTE));
					break;
				case SMALL_SIZE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_SMALL));
					break;
				case LARGE_SIZE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_LARGE));
					break;
				case LARGER_SIZE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_LARGER));
					break;
				case LARGEST_SIZE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_LARGEST));
					break;
				case HUGE_SIZE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_HUGE));
					break;
				case HUGER_SIZE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_HUGER));
					break;
				case INCREASE_SIZE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_INCREASE));
					break;
				case DECREASE_SIZE:
					tagsToClose.emplace_back(xhtmlEndFontTag(xml::FT_SIZE_DECREASE));
					break;
				case INHERIT_SIZE:
				case NORMAL_SIZE:
					break;
				default:
					// the other tags are for internal use
					LATTEST(false);
					break;
				}
				size_flag = false;
			}
			switch (curr_size) {
			case TINY_SIZE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_TINY));
				size_flag = true;
				break;
			case SCRIPT_SIZE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_SCRIPT));
				size_flag = true;
				break;
			case FOOTNOTE_SIZE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_FOOTNOTE));
				size_flag = true;
				break;
			case SMALL_SIZE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_SMALL));
				size_flag = true;
				break;
			case LARGE_SIZE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_LARGE));
				size_flag = true;
				break;
			case LARGER_SIZE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_LARGER));
				size_flag = true;
				break;
			case LARGEST_SIZE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_LARGEST));
				size_flag = true;
				break;
			case HUGE_SIZE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_HUGE));
				size_flag = true;
				break;
			case HUGER_SIZE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_HUGER));
				size_flag = true;
				break;
			case INCREASE_SIZE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_INCREASE));
				size_flag = true;
				break;
			case DECREASE_SIZE:
				tagsToOpen.emplace_back(xhtmlStartFontTag(xml::FT_SIZE_DECREASE));
				size_flag = true;
				break;
			case INHERIT_SIZE:
			case NORMAL_SIZE:
				break;
			default:
				// the other tags are for internal use
				LATTEST(false);
				break;
			}
		}

		// FIXME XHTML
		// Other such tags? What about the other text ranges?

		for (auto const & t : tagsToClose)
			xs << t;
		for (auto const & t : tagsToOpen)
			xs << t;
		tagsToClose.clear();
		tagsToOpen.clear();

		Inset const * inset = getInset(i);
		if (inset) {
			if (!runparams.for_toc || inset->isInToc()) {
				OutputParams np = runparams;
				np.local_font = &font;
				// If the paragraph has size 1, then we are in the "special
				// case" where we do not output the containing paragraph info
				if (!inset->getLayout().htmlisblock() && size() != 1)
					np.html_in_par = true;
				retval += inset->xhtml(xs, np);
			}
		} else {
			char_type c = getUChar(buf.masterBuffer()->params(),
					       runparams, i);
			if (c == ' ' && (style.free_spacing || runparams.free_spacing))
				xs << XMLStream::ESCAPE_NONE << "&#160;";
			else if (c == '\'')
				xs << XMLStream::ESCAPE_NONE << "&#8217;";
			else
				xs << c;
		}
		font_old = font.fontInfo();
	}

	// FIXME XHTML
	// I'm worried about what happens if a branch, say, is itself
	// wrapped in some font stuff. I think that will not work.
	xs.closeFontTags();
	if (close_paragraph)
		xs.endDivision();

	return retval;
}


bool Paragraph::isHfill(pos_type pos) const
{
	Inset const * inset = getInset(pos);
	return inset && inset->isHfill();
}


bool Paragraph::isNewline(pos_type pos) const
{
	// U+2028 LINE SEPARATOR
	// U+2029 PARAGRAPH SEPARATOR
	char_type const c = d->text_[pos];
	if (c == 0x2028 || c == 0x2029)
		return true;
	Inset const * inset = getInset(pos);
	return inset && inset->lyxCode() == NEWLINE_CODE;
}


bool Paragraph::isEnvSeparator(pos_type pos) const
{
	Inset const * inset = getInset(pos);
	return inset && inset->lyxCode() == SEPARATOR_CODE;
}


bool Paragraph::isLineSeparator(pos_type pos) const
{
	char_type const c = d->text_[pos];
	if (isLineSeparatorChar(c))
		return true;
	Inset const * inset = getInset(pos);
	return inset && inset->isLineSeparator();
}


bool Paragraph::isWordSeparator(pos_type pos, bool const ignore_deleted) const
{
	if (pos == size())
		return true;
	if (ignore_deleted && isDeleted(pos))
		return false;
	if (Inset const * inset = getInset(pos))
		return !inset->isLetter();
	// if we have a hard hyphen (no en- or emdash) or apostrophe
	// we pass this to the spell checker
	// FIXME: this method is subject to change, visit
	// https://bugzilla.mozilla.org/show_bug.cgi?id=355178
	// to get an impression how complex this is.
	if (isHardHyphenOrApostrophe(pos))
		return false;
	char_type const c = d->text_[pos];
	// We want to pass the escape chars to the spellchecker
	return !isLetterChar(c) && !isDigitASCII(c)
	       && !contains(lyxrc.spellchecker_esc_chars, c);
}


bool Paragraph::isHardHyphenOrApostrophe(pos_type pos) const
{
	pos_type const psize = size();
	if (pos >= psize)
		return false;
	char_type const c = d->text_[pos];
	if (c != '-' && c != '\'')
		return false;
	pos_type nextpos = pos + 1;
	pos_type prevpos = pos > 0 ? pos - 1 : 0;
	if ((nextpos == psize || isSpace(nextpos))
		&& (pos == 0 || isSpace(prevpos)))
		return false;
	return true;
}


bool Paragraph::needsCProtection(bool const fragile) const
{
	// first check the layout of the paragraph, but only in insets
	// and not in tables
	InsetText const * textinset = inInset().asInsetText();
	bool const maintext = textinset
		? textinset->text().isMainText() || inInset().lyxCode() == CELL_CODE
		: false;

	if (!maintext && layout().needcprotect) {
		// Environments need cprotection regardless the content
		if (layout().latextype == LATEX_ENVIRONMENT)
			return true;

		// Commands need cprotection if they contain specific chars
		int const nchars_escape = 9;
		static char_type const chars_escape[nchars_escape] = {
			'&', '_', '$', '%', '#', '^', '{', '}', '\\'};

		docstring const pars = asString();
		for (int k = 0; k < nchars_escape; k++) {
			if (contains(pars, chars_escape[k]))
				return true;
		}
	}

	// now check whether we have insets that need cprotection
	for (auto const & icit : d->insetlist_) {
		Inset const * ins = icit.inset;
		if (!ins)
			continue;
		if (ins->needsCProtection(maintext, fragile))
			return true;
	}

	return false;
}


FontSpan const & Paragraph::getSpellRange(pos_type pos) const
{
	return d->speller_state_.getRange(pos);
}


bool Paragraph::isChar(pos_type pos) const
{
	if (Inset const * inset = getInset(pos))
		return inset->isChar();
	char_type const c = d->text_[pos];
	return !isLetterChar(c) && !isDigitASCII(c) && !lyx::isSpace(c);
}


bool Paragraph::isSpace(pos_type pos) const
{
	if (Inset const * inset = getInset(pos))
		return inset->isSpace();
	char_type const c = d->text_[pos];
	return lyx::isSpace(c);
}


Language const *
Paragraph::getParLanguage(BufferParams const & bparams) const
{
	if (!empty())
		return getFirstFontSettings(bparams).language();
	// FIXME: we should check the prev par as well (Lgb)
	return bparams.language;
}


bool Paragraph::isRTL(BufferParams const & bparams) const
{
	return getParLanguage(bparams)->rightToLeft()
		&& !inInset().getLayout().forceLTR();
}


void Paragraph::changeLanguage(BufferParams const & bparams,
			       Language const * from, Language const * to)
{
	// change language including dummy font change at the end
	for (pos_type i = 0; i <= size(); ++i) {
		Font font = getFontSettings(bparams, i);
		if (font.language() == from) {
			font.setLanguage(to);
			setFont(i, font);
			d->requestSpellCheck(i);
		}
	}
}


bool Paragraph::isMultiLingual(BufferParams const & bparams) const
{
	Language const * doc_language = bparams.language;
	for (auto const & f : d->fontlist_)
		if (f.font().language() != ignore_language &&
		    f.font().language() != latex_language &&
		    f.font().language() != doc_language)
			return true;
	return false;
}


void Paragraph::getLanguages(std::set<Language const *> & langs) const
{
	for (auto const & f : d->fontlist_) {
		Language const * lang = f.font().language();
		if (lang != ignore_language &&
		    lang != latex_language)
			langs.insert(lang);
	}
}


docstring Paragraph::asString(int options) const
{
	return asString(0, size(), options);
}


docstring Paragraph::asString(pos_type beg, pos_type end, int options, const OutputParams *runparams) const
{
	odocstringstream os;

	if (beg == 0
	    && options & AS_STR_LABEL
	    && d->layout_->labeltype != LABEL_MANUAL
	    && !d->params_.labelString().empty())
		os << d->params_.labelString() << ' ';

	for (pos_type i = beg; i < end; ++i) {
		if ((options & AS_STR_SKIPDELETE) && isDeleted(i))
			continue;
		char_type const c = d->text_[i];
		if (isPrintable(c) || c == '\t'
		    || (c == '\n' && (options & AS_STR_NEWLINES)))
			os.put(c);
		else if (c == META_INSET && (options & AS_STR_INSETS)) {
			if (c == META_INSET && (options & AS_STR_PLAINTEXT)) {
				LASSERT(runparams != nullptr, return docstring());
				if (runparams->find_effective() && getInset(i)->findUsesToString())
					getInset(i)->toString(os);
				else
					getInset(i)->plaintext(os, *runparams);
			} else if (c == META_INSET && (options & AS_STR_MATHED)
				   && getInset(i)->lyxCode() == REF_CODE) {
				Buffer const & buf = getInset(i)->buffer();
				OutputParams rp(&buf.params().encoding());
				Font const font(inherit_font, buf.params().language);
				rp.local_font = &font;
				otexstream ots(os);
				getInset(i)->latex(ots, rp);
			} else {
				getInset(i)->toString(os);
			}
		}
	}

	return os.str();
}


void Paragraph::forOutliner(docstring & os, size_t const maxlen,
                            bool const shorten, bool const label) const
{
	size_t tmplen = shorten ? maxlen + 1 : maxlen;
	if (label && !labelString().empty())
		os += labelString() + ' ';
	if (!layout().isTocCaption())
		return;
	for (pos_type i = 0; i < size() && os.length() < tmplen; ++i) {
		if (isDeleted(i))
			continue;
		char_type const c = d->text_[i];
		if (isPrintable(c))
			os += c;
		else if (c == META_INSET)
			getInset(i)->forOutliner(os, tmplen, false);
	}
	if (shorten)
		Text::shortenForOutliner(os, maxlen);
}


void Paragraph::setInsetOwner(Inset const * inset)
{
	d->inset_owner_ = inset;
}


int Paragraph::id() const
{
	return d->id_;
}


void Paragraph::setId(int id)
{
	d->id_ = id;
}


Layout const & Paragraph::layout() const
{
	return *d->layout_;
}


void Paragraph::setLayout(Layout const & layout)
{
	d->layout_ = &layout;
}


void Paragraph::setDefaultLayout(DocumentClass const & tc)
{
	setLayout(tc.defaultLayout());
}


void Paragraph::setPlainLayout(DocumentClass const & tc)
{
	setLayout(tc.plainLayout());
}


void Paragraph::setPlainOrDefaultLayout(DocumentClass const & tc)
{
	if (usePlainLayout())
		setPlainLayout(tc);
	else
		setDefaultLayout(tc);
}


Inset const & Paragraph::inInset() const
{
	LBUFERR(d->inset_owner_);
	return *d->inset_owner_;
}


ParagraphParameters & Paragraph::params()
{
	return d->params_;
}


ParagraphParameters const & Paragraph::params() const
{
	return d->params_;
}


bool Paragraph::isFreeSpacing() const
{
	if (d->layout_->free_spacing)
		return true;
	return d->inset_owner_ && d->inset_owner_->isFreeSpacing();
}


bool Paragraph::allowEmpty() const
{
	if (d->layout_->keepempty)
		return true;
	return d->inset_owner_ && d->inset_owner_->allowEmpty();
}


int Paragraph::getInsetPos(InsetCode const code, int startpos,
			   bool ignore_deleted) const
{
	while (startpos != -1) {
		int found_pos = d->insetlist_.find(code, startpos);
		if (found_pos == -1)
			// nothing found
			return -1;
		if (isDeleted(found_pos) && ignore_deleted) {
			// we're not interested in deleted insets
			if (found_pos + 1 == size())
				return -1;
			startpos = found_pos + 1;
			continue;
		} else
			return found_pos;
	}
	return -1;
}


bool Paragraph::brokenBiblio() const
{
	// There is a problem if there is no bibitem at position 0 in
	// paragraphs that need one, if there is another bibitem in the
	// paragraph or if this paragraph is not supposed to have
	// a bibitem inset at all.
	return ((d->layout_->labeltype == LABEL_BIBLIO
		&& (getInsetPos(BIBITEM_CODE, 0, true) != 0
		    || getInsetPos(BIBITEM_CODE, 1, true) > 0))
		|| (d->layout_->labeltype != LABEL_BIBLIO
		    && getInsetPos(BIBITEM_CODE, 0, true) != -1));
}


int Paragraph::fixBiblio(Buffer const & buffer)
{
	// FIXME: when there was already an inset at 0, the return value is 1,
	// which does not tell whether another inset has been removed; the
	// cursor cannot be correctly updated.

	bool const track_changes = buffer.params().track_changes;
	int bibitem_pos = getInsetPos(BIBITEM_CODE, 0, true);

	// The case where paragraph is not BIBLIO
	if (d->layout_->labeltype != LABEL_BIBLIO) {
		if (bibitem_pos == -1)
			// No InsetBibitem => OK
			return 0;
		// There is an InsetBibitem: remove it!
		d->insetlist_.release(bibitem_pos);
		eraseChar(bibitem_pos, track_changes);
		return (bibitem_pos == 0) ? -1 : -bibitem_pos;
	}

	bool const hasbibitem0 = bibitem_pos == 0;
	if (hasbibitem0) {
		bibitem_pos = getInsetPos(BIBITEM_CODE, 0, true);
		// There was an InsetBibitem at pos 0,
		// and no other one => OK
		if (bibitem_pos == -1)
			return 0;
		// there is a bibitem at the 0 position, but since
		// there is a second one, we copy the second on the
		// first. We're assuming there are at most two of
		// these, which there should be.
		// FIXME: why does it make sense to do that rather
		// than keep the first? (JMarc)
		Inset * inset = releaseInset(bibitem_pos);
		d->insetlist_.begin()->inset = inset;
		// This needs to be done to update the counter (#8499)
		buffer.updateBuffer();
		return -bibitem_pos;
	}

	// We need to create an inset at the beginning
	Inset * inset = nullptr;
	if (bibitem_pos > 0) {
		// There was one somewhere in the paragraph, let's move it
		// * With change tracking, we use a clone
		//   and leave the old inset at its position
		//   (marked deleted)
		// * Without change tracking, we release the inset
		//   from its previous InsetList position
		inset = track_changes
				? new InsetBibitem(const_cast<Buffer *>(&buffer),
						   getInset(bibitem_pos)->asInsetCommand()->params())
				: d->insetlist_.release(bibitem_pos);
		eraseChar(bibitem_pos, track_changes);
	} else
		// No inset found -- make a fresh one
		inset = new InsetBibitem(const_cast<Buffer *>(&buffer),
					 InsetCommandParams(BIBITEM_CODE));

	Font font(inherit_font, buffer.params().language);
	insertInset(0, inset, font, Change(track_changes ? Change::INSERTED
				                   : Change::UNCHANGED));

	// This is needed to get the counters right
	buffer.updateBuffer();
	return 1;
}


void Paragraph::checkAuthors(AuthorList const & authorList)
{
	d->changes_.checkAuthors(authorList);
}


bool Paragraph::isChanged(pos_type pos) const
{
	return lookupChange(pos).changed();
}


bool Paragraph::isInserted(pos_type pos) const
{
	return lookupChange(pos).inserted();
}


bool Paragraph::isDeleted(pos_type pos) const
{
	return lookupChange(pos).deleted();
}


InsetList const & Paragraph::insetList() const
{
	return d->insetlist_;
}


void Paragraph::setInsetBuffers(Buffer & b)
{
	d->insetlist_.setBuffer(b);
}


void Paragraph::resetBuffer()
{
	d->insetlist_.resetBuffer();
}


Inset * Paragraph::releaseInset(pos_type pos)
{
	Inset * inset = d->insetlist_.release(pos);
	/// does not honour change tracking!
	eraseChar(pos, false);
	return inset;
}


Inset * Paragraph::getInset(pos_type pos)
{
	return (pos < pos_type(d->text_.size()) && d->text_[pos] == META_INSET)
		 ? d->insetlist_.get(pos) : nullptr;
}


Inset const * Paragraph::getInset(pos_type pos) const
{
	return (pos < pos_type(d->text_.size()) && d->text_[pos] == META_INSET)
		 ? d->insetlist_.get(pos) : nullptr;
}


void Paragraph::changeCase(BufferParams const & bparams, pos_type pos,
		pos_type & right, TextCase action)
{
	// process sequences of modified characters; in change
	// tracking mode, this approach results in much better
	// usability than changing case on a char-by-char basis
	// We also need to track the current font, since font
	// changes within sequences can occur.
	vector<pair<char_type, Font> > changes;

	bool const trackChanges = bparams.track_changes;

	bool capitalize = true;

	for (; pos < right; ++pos) {
		char_type oldChar = d->text_[pos];
		char_type newChar = oldChar;

		// ignore insets and don't play with deleted text!
		if (oldChar != META_INSET && !isDeleted(pos)) {
			switch (action) {
			case text_lowercase:
				newChar = lowercase(oldChar);
				break;
			case text_capitalization:
				if (capitalize) {
					newChar = uppercase(oldChar);
					capitalize = false;
				}
				break;
			case text_uppercase:
				newChar = uppercase(oldChar);
				break;
			case text_invertcase:
				if (isUpperCase(oldChar))
					newChar = lowercase(oldChar);
				else if (isLowerCase(oldChar))
					newChar = uppercase(oldChar);
			}
		}

		if (isWordSeparator(pos) || isDeleted(pos)) {
			// permit capitalization again
			capitalize = true;
		}

		if (oldChar != newChar) {
			changes.push_back(make_pair(newChar, getFontSettings(bparams, pos)));
			if (pos != right - 1)
				continue;
			// step behind the changing area
			pos++;
		}

		int erasePos = pos - changes.size();
		for (auto const & change : changes) {
			insertChar(pos, change.first, change.second, trackChanges);
			if (!eraseChar(erasePos, trackChanges)) {
				++erasePos;
				++pos; // advance
				++right; // expand selection
			}
		}
		changes.clear();
	}
}

int Paragraph::find(docstring const & str, bool cs, bool mw,
		pos_type start_pos, bool del) const
{
	pos_type pos = start_pos;
	int const strsize = str.length();
	int i = 0;
	pos_type const parsize = d->text_.size();
	for (i = 0; i < strsize && pos < parsize; ++i, ++pos) {
		// ignore deleted matter
		if (!del && isDeleted(pos)) {
			if (pos == parsize - 1)
				break;
			pos++;
			--i;
			continue;
		}
		// Ignore "invisible" letters such as ligature breaks
		// and hyphenation chars while searching
		bool nonmatch = false;
		while (pos < parsize && isInset(pos)) {
			Inset const * inset = getInset(pos);
			if (!inset->isLetter() && !inset->isChar())
				break;
			odocstringstream os;
			if (inset->lyxCode() == lyx::QUOTE_CODE || inset->lyxCode() == lyx::SPACE_CODE) {
				OutputParams op(0);
				op.find_set_feature(OutputParams::SearchQuick);
				inset->plaintext(os, op);
			}
			else {
				inset->toString(os);
			}
			docstring const insetstring = os.str();
			if (!insetstring.empty()) {
				int const insetstringsize = insetstring.length();
				for (int j = 0; j < insetstringsize && pos < parsize; ++i, ++j) {
					if ((cs && str[i] != insetstring[j])
					    || (!cs && uppercase(str[i]) != uppercase(insetstring[j]))) {
						nonmatch = true;
						break;
					}
				}
			}
			pos++;
		}
		if (nonmatch || i == strsize)
			break;
		char_type dp = d->text_[pos];
		if (cs && str[i] != dp)
			break;
		if (!cs && uppercase(str[i]) != uppercase(dp))
			break;
	}

	if (i != strsize)
		return 0;

	// if necessary, check whether string matches word
	if (mw) {
		if (start_pos > 0 && !isWordSeparator(start_pos - 1))
			return 0;
		if (pos < parsize
			&& !isWordSeparator(pos))
			return 0;
	}

	return pos - start_pos;
}


char_type Paragraph::getChar(pos_type pos) const
{
	return d->text_[pos];
}


pos_type Paragraph::size() const
{
	return d->text_.size();
}


bool Paragraph::empty() const
{
	return d->text_.empty();
}


bool Paragraph::isInset(pos_type pos) const
{
	return d->text_[pos] == META_INSET;
}


bool Paragraph::isSeparator(pos_type pos) const
{
	//FIXME: Are we sure this can be the only separator?
	return d->text_[pos] == ' ';
}


void Paragraph::deregisterWords()
{
	Private::LangWordsMap::const_iterator itl = d->words_.begin();
	Private::LangWordsMap::const_iterator ite = d->words_.end();
	for (; itl != ite; ++itl) {
		WordList & wl = theWordList(itl->first);
		Private::Words::const_iterator it = (itl->second).begin();
		Private::Words::const_iterator et = (itl->second).end();
		for (; it != et; ++it)
			wl.remove(*it);
	}
	d->words_.clear();
}


void Paragraph::locateWord(pos_type & from, pos_type & to,
	word_location const loc, bool const ignore_deleted) const
{
	switch (loc) {
	case WHOLE_WORD_STRICT:
		if (from == 0 || from == size()
		    || isWordSeparator(from, ignore_deleted)
		    || isWordSeparator(from - 1, ignore_deleted)) {
			to = from;
			return;
		}
		// fall through

	case WHOLE_WORD:
		// If we are already at the beginning of a word, do nothing
		if (!from || isWordSeparator(from - 1, ignore_deleted))
			break;
		// fall through

	case PREVIOUS_WORD:
		// always move the cursor to the beginning of previous word
		while (from && !isWordSeparator(from - 1, ignore_deleted))
			--from;
		break;
	case NEXT_WORD:
		LYXERR0("Paragraph::locateWord: NEXT_WORD not implemented yet");
		break;
	case PARTIAL_WORD:
		// no need to move the 'from' cursor
		break;
	}
	to = from;
	while (to < size() && !isWordSeparator(to, ignore_deleted))
		++to;
}


void Paragraph::collectWords()
{
	for (pos_type pos = 0; pos < size(); ++pos) {
		if (isWordSeparator(pos))
			continue;
		pos_type from = pos;
		locateWord(from, pos, WHOLE_WORD);
		// Work around MSVC warning: The statement
		// if (pos < from + lyxrc.completion_minlength)
		// triggers a signed vs. unsigned warning.
		// I don't know why this happens, it could be a MSVC bug, or
		// related to LLP64 (windows) vs. LP64 (unix) programming
		// model, or the C++ standard might be ambigous in the section
		// defining the "usual arithmetic conversions". However, using
		// a temporary variable is safe and works on all compilers.
		pos_type const endpos = from + lyxrc.completion_minlength;
		if (pos < endpos)
			continue;
		FontList::const_iterator cit = d->fontlist_.fontIterator(from);
		if (cit == d->fontlist_.end())
			return;
		Language const * lang = cit->font().language();
		docstring const word = asString(from, pos, AS_STR_NONE);
		d->words_[lang->lang()].insert(word);
	}
}


void Paragraph::registerWords()
{
	Private::LangWordsMap::const_iterator itl = d->words_.begin();
	Private::LangWordsMap::const_iterator ite = d->words_.end();
	for (; itl != ite; ++itl) {
		WordList & wl = theWordList(itl->first);
		Private::Words::const_iterator it = (itl->second).begin();
		Private::Words::const_iterator et = (itl->second).end();
		for (; it != et; ++it)
			wl.insert(*it);
	}
}


void Paragraph::updateWords()
{
	deregisterWords();
	collectWords();
	registerWords();
}


void Paragraph::Private::appendSkipPosition(SkipPositions & skips, pos_type const pos) const
{
	SkipPositionsIterator begin = skips.begin();
	SkipPositions::iterator end = skips.end();
	if (pos > 0 && begin < end) {
		--end;
		if (end->last == pos - 1) {
			end->last = pos;
			return;
		}
	}
	skips.insert(end, FontSpan(pos, pos));
}


Language * Paragraph::Private::locateSpellRange(
	pos_type & from, pos_type & to,
	SkipPositions & skips) const
{
	// skip leading white space
	while (from < to && owner_->isWordSeparator(from))
		++from;
	// don't check empty range
	if (from >= to)
		return nullptr;
	// get current language
	Language * lang = getSpellLanguage(from);
	pos_type last = from;
	bool samelang = true;
	bool sameinset = true;
	while (last < to && samelang && sameinset) {
		// hop to end of word
		while (last < to && !owner_->isWordSeparator(last)) {
			Inset const * inset = owner_->getInset(last);
			if (inset && dynamic_cast<const InsetSpecialChar *>(inset)) {
				// check for "invisible" letters such as ligature breaks
				odocstringstream os;
				inset->toString(os);
				if (os.str().length() != 0) {
					// avoid spell check of visible special char insets
					// stop the loop in front of the special char inset
					sameinset = false;
					break;
				}
			} else if (inset) {
				appendSkipPosition(skips, last);
			} else if (owner_->isDeleted(last)) {
				appendSkipPosition(skips, last);
			}
			++last;
		}
		// hop to next word while checking for insets
		while (sameinset && last < to && owner_->isWordSeparator(last)) {
			if (Inset const * inset = owner_->getInset(last))
				sameinset = inset->isChar() && inset->isLetter();
			if (sameinset && owner_->isDeleted(last)) {
				appendSkipPosition(skips, last);
			}
			if (sameinset)
				last++;
		}
		if (sameinset && last < to) {
			// now check for language change
			samelang = lang == getSpellLanguage(last);
		}
	}
	// if language change detected backstep is needed
	if (!samelang)
		--last;
	to = last;
	return lang;
}


Language * Paragraph::Private::getSpellLanguage(pos_type const from) const
{
	Language * lang =
		const_cast<Language *>(owner_->getFontSettings(
			inset_owner_->buffer().params(), from).language());
	if (lang == inset_owner_->buffer().params().language
		&& !lyxrc.spellchecker_alt_lang.empty()) {
		string lang_code;
		string const lang_variety =
			split(lyxrc.spellchecker_alt_lang, lang_code, '-');
		lang->setCode(lang_code);
		lang->setVariety(lang_variety);
	}
	return lang;
}


void Paragraph::requestSpellCheck(pos_type pos)
{
	d->requestSpellCheck(pos);
	if (pos == -1) {
		// Also request spellcheck within (text) insets
		for (auto const & insets : insetList()) {
			if (!insets.inset->asInsetText())
				continue;
			ParagraphList & inset_pars =
				insets.inset->asInsetText()->paragraphs();
			ParagraphList::iterator pit = inset_pars.begin();
			ParagraphList::iterator pend = inset_pars.end();
			for (; pit != pend; ++pit)
				pit->requestSpellCheck();
		}
	}
}


bool Paragraph::needsSpellCheck() const
{
	SpellChecker::ChangeNumber speller_change_number = 0;
	if (theSpellChecker())
		speller_change_number = theSpellChecker()->changeNumber();
	if (speller_change_number > d->speller_state_.currentChangeNumber()) {
		d->speller_state_.needsCompleteRefresh(speller_change_number);
	}
	return d->needsSpellCheck();
}


bool Paragraph::Private::ignoreWord(docstring const & word) const
{
	// Ignore words with digits
	// FIXME: make this customizable
	// (note that some checkers ignore words with digits by default)
	docstring::const_iterator cit = word.begin();
	docstring::const_iterator const end = word.end();
	for (; cit != end; ++cit) {
		if (isNumber((*cit)))
			return true;
	}
	return false;
}


SpellChecker::Result Paragraph::spellCheck(pos_type & from, pos_type & to,
	WordLangTuple & wl, docstring_list & suggestions,
	bool do_suggestion, bool check_learned) const
{
	SpellChecker::Result result = SpellChecker::WORD_OK;
	SpellChecker * speller = theSpellChecker();
	if (!speller)
		return result;

	if (!d->layout_->spellcheck || !inInset().allowSpellCheck())
		return result;

	locateWord(from, to, WHOLE_WORD, true);
	if (from == to || from >= size())
		return result;

	docstring word = asString(from, to, AS_STR_INSETS | AS_STR_SKIPDELETE);
	Language * lang = d->getSpellLanguage(from);

	BufferParams const & bparams = d->inset_owner_->buffer().params();

	if (getFontSettings(bparams, from).fontInfo().nospellcheck() == FONT_ON)
		return result;

	wl = WordLangTuple(word, lang);

	if (word.empty())
		return result;

	if (needsSpellCheck() || check_learned) {
		pos_type end = to;
		if (!d->ignoreWord(word)) {
			bool const trailing_dot = to < size() && d->text_[to] == '.';
			result = speller->check(wl, bparams.spellignore());
			if (SpellChecker::misspelled(result) && trailing_dot) {
				wl = WordLangTuple(word.append(from_ascii(".")), lang);
				result = speller->check(wl, bparams.spellignore());
				if (!SpellChecker::misspelled(result)) {
					LYXERR(Debug::GUI, "misspelled word is correct with dot: \"" <<
					   word << "\" [" <<
					   from << ".." << to << "]");
				} else {
					// spell check with dot appended failed too
					// restore original word/lang value
					word = asString(from, to, AS_STR_INSETS | AS_STR_SKIPDELETE);
					wl = WordLangTuple(word, lang);
				}
			}
		}
		if (!SpellChecker::misspelled(result)) {
			// area up to the begin of the next word is not misspelled
			while (end < size() && isWordSeparator(end))
				++end;
		}
		d->setMisspelled(from, end, result);
	} else {
		result = d->speller_state_.getState(from);
	}

	if (do_suggestion)
		suggestions.clear();

	if (SpellChecker::misspelled(result)) {
		LYXERR(Debug::GUI, "misspelled word: \"" <<
			   word << "\" [" <<
			   from << ".." << to << "]");
		if (do_suggestion)
			speller->suggest(wl, suggestions);
	}
	return result;
}


void Paragraph::anonymize()
{
	// This is a very crude anonymization for now
	for (char_type & c : d->text_)
		if (isLetterChar(c) || isNumber(c))
			c = 'a';
}


void Paragraph::Private::markMisspelledWords(
	Language const * lang,
	pos_type const & first, pos_type const & last,
	SpellChecker::Result result,
	docstring const & word,
	SkipPositions const & skips)
{
	if (!SpellChecker::misspelled(result)) {
		setMisspelled(first, last, SpellChecker::WORD_OK);
		return;
	}
	int snext = first;
	SpellChecker * speller = theSpellChecker();
	// locate and enumerate the error positions
	int nerrors = speller->numMisspelledWords();
	int numskipped = 0;
	SkipPositionsIterator it = skips.begin();
	SkipPositionsIterator et = skips.end();
	for (int index = 0; index < nerrors; ++index) {
		int wstart;
		int wlen = 0;
		speller->misspelledWord(index, wstart, wlen);
		/// should not happen if speller supports range checks
		if (!wlen)
			continue;
		WordLangTuple const candidate(word.substr(wstart, wlen), lang);
		wstart += first + numskipped;
		if (snext < wstart) {
			/// mark the range of correct spelling
			numskipped += countSkips(it, et, wstart);
			setMisspelled(snext,
				wstart - 1, SpellChecker::WORD_OK);
		}
		snext = wstart + wlen;
		// Check whether the candidate is in the document's local dict
		SpellChecker::Result actresult = result;
		if (inset_owner_->buffer().params().spellignored(candidate))
			actresult = SpellChecker::DOCUMENT_LEARNED_WORD;
		numskipped += countSkips(it, et, snext);
		/// mark the range of misspelling
		setMisspelled(wstart, snext, actresult);
		if (actresult == SpellChecker::DOCUMENT_LEARNED_WORD)
			LYXERR(Debug::GUI, "local dictionary word: \"" <<
				   candidate.word() << "\" [" <<
				   wstart << ".." << (snext-1) << "]");
		else
			LYXERR(Debug::GUI, "misspelled word: \"" <<
				   candidate.word() << "\" [" <<
				   wstart << ".." << (snext-1) << "]");
		++snext;
	}
	if (snext <= last) {
		/// mark the range of correct spelling at end
		setMisspelled(snext, last, SpellChecker::WORD_OK);
	}
}


void Paragraph::spellCheck() const
{
	SpellChecker * speller = theSpellChecker();
	if (!speller || empty() ||!needsSpellCheck())
		return;
	pos_type start;
	pos_type endpos;
	d->rangeOfSpellCheck(start, endpos);
	if (speller->canCheckParagraph()) {
		// loop until we leave the range
		for (pos_type first = start; first < endpos; ) {
			pos_type last = endpos;
			Private::SkipPositions skips;
			Language * lang = d->locateSpellRange(first, last, skips);
			if (first >= endpos)
				break;
			// start the spell checker on the unit of meaning
			docstring word = asString(first, last, AS_STR_INSETS + AS_STR_SKIPDELETE);
			WordLangTuple wl = WordLangTuple(word, lang);
			BufferParams const & bparams = d->inset_owner_->buffer().params();
			SpellChecker::Result result = !word.empty() ?
				speller->check(wl, bparams.spellignore()) : SpellChecker::WORD_OK;
			d->markMisspelledWords(lang, first, last, result, word, skips);
			first = ++last;
		}
	} else {
		static docstring_list suggestions;
		pos_type to = endpos;
		while (start < endpos) {
			WordLangTuple wl;
			spellCheck(start, to, wl, suggestions, false);
			start = to + 1;
		}
	}
	d->readySpellCheck();
}


bool Paragraph::isMisspelled(pos_type pos, bool check_boundary) const
{
	bool result = SpellChecker::misspelled(d->speller_state_.getState(pos));
	if (result || pos <= 0 || pos > size())
		return result;
	if (check_boundary && (pos == size() || isWordSeparator(pos)))
		result = SpellChecker::misspelled(d->speller_state_.getState(pos - 1));
	return result;
}


string Paragraph::magicLabel() const
{
	stringstream ss;
	ss << "magicparlabel-" << id();
	return ss.str();
}


} // namespace lyx