3 * This file is part of LyX, the document processor.
4 * Licence details can be found in the file COPYING.
6 * \author Vincent van Ravesteijn
8 * Full author contact details are available in file CREDITS.
16 #include "BufferParams.h"
18 #include "CutAndPaste.h"
19 #include "ErrorList.h"
22 #include "insets/InsetText.h"
24 #include "support/lassert.h"
25 #include "support/lyxalgo.h"
26 #include "support/qstring_helpers.h"
29 using namespace lyx::support;
41 static void step(DocIterator & dit, Direction direction)
43 if (direction == Forward)
44 dit.top().forwardPos();
46 dit.top().backwardPos();
50 static void step(DocIterator & dit, DocIterator const & end, Direction direction)
58 * A pair of two DocIterators that form a range.
62 DocRange(DocIterator const & from_, DocIterator const & to_)
63 : from(from_), to(to_)
66 DocRange(Buffer const * buf)
68 from = doc_iterator_begin(buf);
69 to = doc_iterator_end(buf);
74 Text * text() const { return from.text(); }
76 bool empty() const { return to <= from; }
78 size_t length() const;
80 /// The begin of the range
82 /// The end of the range
87 size_t DocRange::length() const
89 ParagraphList const & ps = from.text()->paragraphs();
91 pit_type pit = from.pit();
92 pit_type const endpit = to.pit();
93 for (; pit < endpit; ++pit)
94 length += ps[pit].size() + 1;
95 length += to.pos() - from.pos();
105 DocPair(DocIterator o_, DocIterator n_)
109 bool operator!=(DocPair const & rhs)
111 // this might not be intuitive but correct for our purpose
112 return o != rhs.o && n != rhs.n;
116 DocPair & operator++()
123 DocPair & operator--()
136 * A pair of two DocRanges.
140 DocRangePair(DocRange const & o_, DocRange const & n_)
144 DocRangePair(DocPair const & from, DocPair const & to)
145 : o(from.o, to.o), n(from.n, to.n)
148 DocRangePair(Buffer const * o_buf, Buffer const * n_buf)
152 /// Returns the from pair
155 return DocPair(o.from, n.from);
158 /// Returns the to pair
161 return DocPair(o.to, n.to);
169 static DocRangePair stepIntoInset(DocPair const & inset_location)
171 DocRangePair rp(inset_location, inset_location);
172 rp.o.from.forwardPos();
173 rp.n.from.forwardPos();
174 step(rp.o.to, Forward);
175 step(rp.n.to, Forward);
176 rp.o.to.backwardPos();
177 rp.n.to.backwardPos();
183 * This class is designed to hold a vector that has both positive as
184 * negative indices. It is internally represented as two vectors, one
185 * for non-zero indices and one for negative indices. In this way, the
186 * vector can grow in both directions.
187 * If an index is not available in the vector, the default value is
188 * returned. If an object is put in the vector beyond its size, the
189 * empty spots in between are also filled with the default value.
197 void reset(T const & def)
204 /// Gets the value at index. If it is not in the vector
205 /// the default value is inserted and returned.
206 T & operator[](int index) {
207 vector<T> & V = index >= 0 ? Vp_ : Vn_;
208 unsigned int const ii = index >= 0 ? index : -index - 1;
209 while (ii >= V.size())
210 V.push_back(default_);
215 /// The vector for positive indices
217 /// The vector for negative indices
219 /// The default value that is inserted in the vector
220 /// if more space is needed
226 * The implementation of the algorithm that does the comparison
227 * between two documents.
229 class Compare::Impl {
232 Impl(Compare const & compare)
233 : abort_(false), n_(0), m_(0), offset_reverse_diagonal_(0),
234 odd_offset_(0), compare_(compare),
235 old_buf_(0), new_buf_(0), dest_buf_(0), dest_pars_(0),
236 recursion_level_(0), nested_inset_level_(0), D_(0)
243 // Algorithm to find the shortest edit string. This algorithm
244 // only needs a linear amount of memory (linear with the sum
245 // of the number of characters in the two paragraph-lists).
246 bool diff(Buffer const * new_buf, Buffer const * old_buf,
247 Buffer const * dest_buf);
249 /// Set to true to cancel the algorithm
256 status += toqstr("recursion level:") + " " + QString::number(recursion_level_)
257 + " " + toqstr("differences:") + " " + QString::number(D_);
262 /// Finds the middle snake and returns the length of the
263 /// shortest edit script.
264 int findMiddleSnake(DocRangePair const & rp, DocPair & middle_snake);
272 /// Retrieve the middle snake when there is overlap between
273 /// the forward and backward path.
274 SnakeResult retrieveMiddleSnake(int k, int D, Direction direction,
275 DocPair & middle_snake);
277 /// Find the furthest reaching D-path (number of horizontal
278 /// and vertical steps; differences between the old and new
279 /// document) in the k-diagonal (vertical minus horizontal steps).
280 void furthestDpathKdiagonal(int D, int k,
281 DocRangePair const & rp, Direction direction);
283 /// Is there overlap between the forward and backward path
284 bool overlap(int k, int D);
286 /// This function is called recursively by a divide and conquer
287 /// algorithm. Each time, the string is divided into two split
288 /// around the middle snake.
289 void diff_i(DocRangePair const & rp);
291 /// Processes the split chunks. It either adds them as deleted,
292 /// as added, or call diff_i for further processing.
293 void diffPart(DocRangePair const & rp);
295 /// Runs the algorithm for the inset located at /c it and /c it_n
296 /// and adds the result to /c pars.
297 void diffInset(Inset * inset, DocPair const & p);
299 /// Adds the snake to the destination buffer. The algorithm will
300 /// recursively be applied to any InsetTexts that are within the snake.
301 void processSnake(DocRangePair const & rp);
303 /// Writes the range to the destination buffer
304 void writeToDestBuffer(DocRange const & range,
305 Change::Type type = Change::UNCHANGED);
307 /// Writes the paragraph list to the destination buffer
308 void writeToDestBuffer(ParagraphList const & copy_pars) const;
310 /// The length of the old chunk currently processed
312 /// The length of the new chunk currently processed
314 /// The offset diagonal of the reverse path of the
315 /// currently processed chunk
316 int offset_reverse_diagonal_;
317 /// Is the offset odd or even ?
320 /// The thread object, used to emit signals to the GUI
321 Compare const & compare_;
323 /// The buffer containing text that will be marked as old
324 Buffer const * old_buf_;
325 /// The buffer containing text that will be marked as new
326 Buffer const * new_buf_;
327 /// The buffer containing text that will be marked as new
328 Buffer const * dest_buf_;
330 /// The paragraph list of the destination buffer
331 ParagraphList * dest_pars_;
333 /// The level of recursion
334 int recursion_level_;
336 /// The number of nested insets at this level
337 int nested_inset_level_;
339 /// The position/snake in the old/new document
340 /// of the forward/reverse search
341 compl_vector<DocIterator> ofp;
342 compl_vector<DocIterator> nfp;
343 compl_vector<DocIterator> ofs;
344 compl_vector<DocIterator> nfs;
345 compl_vector<DocIterator> orp;
346 compl_vector<DocIterator> nrp;
347 compl_vector<DocIterator> ors;
348 compl_vector<DocIterator> nrs;
350 /// The number of differences in the path the algorithm
351 /// is currently processing.
355 /////////////////////////////////////////////////////////////////////
359 /////////////////////////////////////////////////////////////////////
361 Compare::Compare(Buffer const * new_buf, Buffer const * old_buf,
362 Buffer * const dest_buf, CompareOptions const & options)
363 : new_buffer(new_buf), old_buffer(old_buf), dest_buffer(dest_buf),
364 options_(options), pimpl_(new Impl(*this))
366 connect(&status_timer_, SIGNAL(timeout()),
367 this, SLOT(doStatusMessage()));
368 status_timer_.start(1000);
372 void Compare::doStatusMessage()
374 statusMessage(pimpl_->status());
380 if (!dest_buffer || !new_buffer || !old_buffer)
383 // Copy the buffer params to the destination buffer
384 dest_buffer->params() = options_.settings_from_new
385 ? new_buffer->params() : old_buffer->params();
386 // Copy extra authors to the destination buffer
387 AuthorList const & extra_authors = options_.settings_from_new ?
388 old_buffer->params().authors() : new_buffer->params().authors();
389 AuthorList::Authors::const_iterator it = extra_authors.begin();
390 for (; it != extra_authors.end(); ++it)
391 dest_buffer->params().authors().record(*it);
393 // We will need this later
394 DocumentClassConstPtr const olddc =
395 dest_buffer->params().documentClassPtr();
396 // We do not want to share the DocumentClass with the other Buffer.
398 dest_buffer->params().makeDocumentClass();
405 // The comparison routine simply copies the paragraphs over into the
406 // new buffer with the document class from wherever they came from.
407 // So we need to reset the document class of all the paragraphs.
410 cap::switchBetweenClasses(
411 olddc, dest_buffer->params().documentClassPtr(),
412 static_cast<InsetText &>(dest_buffer->inset()), el);
414 finished(pimpl_->abort_);
419 int Compare::doCompare()
421 return pimpl_->diff(new_buffer, old_buffer, dest_buffer);
425 void Compare::abort()
427 pimpl_->abort_ = true;
428 condition_.wakeOne();
430 pimpl_->abort_ = false;
434 static void getParagraphList(DocRange const & range,
435 ParagraphList & pars)
437 // Clone the paragraphs within the selection.
438 pit_type startpit = range.from.pit();
439 pit_type endpit = range.to.pit();
440 ParagraphList const & ps_ = range.text()->paragraphs();
441 ParagraphList tmp_pars(lyx::next(ps_.begin(), startpit),
442 lyx::next(ps_.begin(), endpit + 1));
444 // Remove the end of the last paragraph; afterwards, remove the
445 // beginning of the first paragraph. Keep this order - there may only
447 Paragraph & back = tmp_pars.back();
448 back.eraseChars(range.to.pos(), back.size(), false);
449 Paragraph & front = tmp_pars.front();
450 front.eraseChars(0, range.from.pos(), false);
452 pars.insert(pars.begin(), tmp_pars.begin(), tmp_pars.end());
456 static bool equal(Inset const * i_o, Inset const * i_n)
461 // Different types of insets
462 if (i_o->lyxCode() != i_n->lyxCode())
465 // Editable insets are assumed to be the same as they are of the
466 // same type. If we later on decide that we insert them in the
467 // document as being unchanged, we will run the algorithm on the
468 // contents of the two insets.
469 // FIXME: This fails if the parameters of the insets differ.
470 // FIXME: We do not recurse into InsetTabulars.
471 // FIXME: We need methods inset->equivalent(inset).
472 if (i_o->editable() && !i_o->asInsetMath()
473 && i_o->asInsetText())
480 return o_os.str() == n_os.str();
484 static bool equal(DocIterator & o, DocIterator & n)
486 // Explicitly check for this, so we won't call
487 // Paragraph::getChar for the last pos.
488 bool const o_lastpos = o.pos() == o.lastpos();
489 bool const n_lastpos = n.pos() == n.lastpos();
490 if (o_lastpos || n_lastpos)
491 return o_lastpos && n_lastpos;
493 Paragraph const & old_par = o.text()->getPar(o.pit());
494 Paragraph const & new_par = n.text()->getPar(n.pit());
496 char_type const c_o = old_par.getChar(o.pos());
497 char_type const c_n = new_par.getChar(n.pos());
501 if (old_par.isInset(o.pos())) {
502 Inset const * i_o = old_par.getInset(o.pos());
503 Inset const * i_n = new_par.getInset(n.pos());
506 return equal(i_o, i_n);
509 Font fo = old_par.getFontSettings(o.buffer()->params(), o.pos());
510 Font fn = new_par.getFontSettings(n.buffer()->params(), n.pos());
515 /// Traverses a snake in a certain direction. p points to a
516 /// position in the old and new file and they are synchronously
517 /// moved along the snake. The function returns true if a snake
519 static bool traverseSnake(DocPair & p, DocRangePair const & range,
523 DocPair const & p_end =
524 direction == Forward ? range.to() : range.from();
527 if (direction == Backward)
529 if (!equal(p.o, p.n)) {
530 if (direction == Backward)
534 if (direction == Forward)
542 /////////////////////////////////////////////////////////////////////
546 /////////////////////////////////////////////////////////////////////
549 void Compare::Impl::furthestDpathKdiagonal(int D, int k,
550 DocRangePair const & rp, Direction direction)
552 compl_vector<DocIterator> & op = direction == Forward ? ofp : orp;
553 compl_vector<DocIterator> & np = direction == Forward ? nfp : nrp;
554 compl_vector<DocIterator> & os = direction == Forward ? ofs : ors;
555 compl_vector<DocIterator> & ns = direction == Forward ? nfs : nrs;
557 // A vertical step means stepping one character in the new document.
558 bool vertical_step = k == -D;
559 if (!vertical_step && k != D) {
560 vertical_step = direction == Forward
561 ? op[k - 1] < op[k + 1] : op[k - 1] > op[k + 1];
564 // Where do we take the step from ?
565 int const kk = vertical_step ? k + 1 : k - 1;
566 DocPair p(op[kk], np[kk]);
567 DocPair const s(os[kk], ns[kk]);
569 // If D==0 we simulate a vertical step from (0,-1) by doing nothing.
572 if (vertical_step && direction == Forward)
573 step(p.n, rp.n.to, direction);
574 else if (vertical_step && direction == Backward)
575 step(p.n, rp.n.from, direction);
576 else if (!vertical_step && direction == Forward)
577 step(p.o, rp.o.to, direction);
578 else if (!vertical_step && direction == Backward)
579 step(p.o, rp.o.from, direction);
583 if (traverseSnake(p, rp, direction)) {
588 // Copy last snake from the previous step
593 //Record new position
599 bool Compare::Impl::overlap(int k, int D)
601 // To generalize for the forward and reverse checks
602 int kk = offset_reverse_diagonal_ - k;
604 // Can we have overlap ?
605 if (kk <= D && kk >= -D) {
606 // Do we have overlap ?
608 return ofp[k] >= orp[kk] && nfp[k] >= nrp[kk];
610 return ofp[kk] >= orp[k] && nfp[kk] >= nrp[k];
616 Compare::Impl::SnakeResult Compare::Impl::retrieveMiddleSnake(
617 int k, int D, Direction direction, DocPair & middle_snake)
619 compl_vector<DocIterator> & os = direction == Forward ? ofs : ors;
620 compl_vector<DocIterator> & ns = direction == Forward ? nfs : nrs;
621 compl_vector<DocIterator> & os_r = direction == Forward ? ors : ofs;
622 compl_vector<DocIterator> & ns_r = direction == Forward ? nrs : nfs;
624 // The diagonal while doing the backward search
625 int kk = -k + offset_reverse_diagonal_;
627 // Did we find a snake ?
628 if (os[k].empty() && os_r[kk].empty()) {
629 // No, there is no snake at all, in which case
630 // the length of the shortest edit script is M+N.
631 LATTEST(2 * D - odd_offset_ == m_ + n_);
636 // Yes, but there is only 1 snake and we found it in the
638 middle_snake.o = os_r[kk];
639 middle_snake.n = ns_r[kk];
643 middle_snake.o = os[k];
644 middle_snake.n = ns[k];
649 int Compare::Impl::findMiddleSnake(DocRangePair const & rp,
650 DocPair & middle_snake)
652 // The lengths of the old and new chunks.
656 // Forward paths are centered around the 0-diagonal; reverse paths
657 // are centered around the diagonal N - M. (Delta in the article)
658 offset_reverse_diagonal_ = n_ - m_;
660 // If the offset is odd, only check for overlap while extending forward
661 // paths, otherwise only check while extending reverse paths.
662 odd_offset_ = (offset_reverse_diagonal_ % 2 != 0);
664 ofp.reset(rp.o.from);
665 nfp.reset(rp.n.from);
666 ofs.reset(DocIterator());
667 nfs.reset(DocIterator());
670 ors.reset(DocIterator());
671 nrs.reset(DocIterator());
673 // In the formula below, the "+ 1" ensures we round like ceil()
674 int const D_max = (m_ + n_ + 1)/2;
675 // D is the number of horizontal and vertical steps, i.e.
676 // different characters in the old and new chunk.
677 for (int D = 0; D <= D_max; ++D) {
678 // to be used in the status messages
681 // Forward and reverse paths
682 for (int f = 0; f < 2; ++f) {
683 Direction direction = f == 0 ? Forward : Backward;
685 // Diagonals between -D and D can be reached by a D-path
686 for (int k = -D; k <= D; k += 2) {
687 // Find the furthest reaching D-path on this diagonal
688 furthestDpathKdiagonal(D, k, rp, direction);
690 // Only check for overlap for forward paths if the offset is odd
691 // and only for reverse paths if the offset is even.
692 if (odd_offset_ == (direction == Forward)) {
694 // Do the forward and backward paths overlap ?
695 if (overlap(k, D - odd_offset_)) {
696 retrieveMiddleSnake(k, D, direction, middle_snake);
697 return 2 * D - odd_offset_;
705 // This should never be reached
710 bool Compare::Impl::diff(Buffer const * new_buf, Buffer const * old_buf,
711 Buffer const * dest_buf)
713 if (!new_buf || !old_buf || !dest_buf)
718 dest_buf_ = dest_buf;
719 dest_pars_ = &dest_buf->inset().asInsetText()->paragraphs();
722 recursion_level_ = 0;
723 nested_inset_level_ = 0;
725 DocRangePair rp(old_buf_, new_buf_);
727 DocPair from = rp.from();
728 traverseSnake(from, rp, Forward);
729 DocRangePair const snake(rp.from(), from);
732 // Start the recursive algorithm
733 DocRangePair rp_new(from, rp.to());
734 if (!rp_new.o.empty() || !rp_new.n.empty())
737 for (pit_type p = 0; p < (pit_type)dest_pars_->size(); ++p) {
738 (*dest_pars_)[p].setBuffer(const_cast<Buffer &>(*dest_buf));
739 (*dest_pars_)[p].setInsetOwner(&dest_buf_->inset());
746 void Compare::Impl::diff_i(DocRangePair const & rp)
752 DocPair middle_snake;
754 // Divides the problem into two smaller problems, split around
755 // the snake in the middle.
756 int const L_ses = findMiddleSnake(rp, middle_snake);
758 // Set maximum of progress bar
759 if (++recursion_level_ == 1)
760 compare_.progressMax(L_ses);
762 // There are now three possibilities: the strings were the same,
763 // the strings were completely different, or we found a middle
764 // snake and we can split the string into two parts to process.
766 // Two the same strings (this must be a very rare case, because
767 // usually this will be part of a snake adjacent to these strings).
768 writeToDestBuffer(rp.o);
770 else if (middle_snake.o.empty()) {
771 // Two totally different strings
772 writeToDestBuffer(rp.o, Change::DELETED);
773 writeToDestBuffer(rp.n, Change::INSERTED);
776 // Retrieve the complete snake
777 DocPair first_part_end = middle_snake;
778 traverseSnake(first_part_end, rp, Backward);
779 DocRangePair first_part(rp.from(), first_part_end);
781 DocPair second_part_begin = middle_snake;
782 traverseSnake(second_part_begin, rp, Forward);
783 DocRangePair second_part(second_part_begin, rp.to());
785 // Split the string in three parts:
786 // 1. in front of the snake
787 diffPart(first_part);
789 // 2. the snake itself, and
790 DocRangePair const snake(first_part.to(), second_part.from());
793 // 3. behind the snake.
794 diffPart(second_part);
800 void Compare::Impl::diffPart(DocRangePair const & rp)
802 // Is there a finite length string in both buffers, if not there
803 // is an empty string and we write the other one to the buffer.
804 if (!rp.o.empty() && !rp.n.empty())
807 else if (!rp.o.empty())
808 writeToDestBuffer(rp.o, Change::DELETED);
810 else if (!rp.n.empty())
811 writeToDestBuffer(rp.n, Change::INSERTED);
815 void Compare::Impl::diffInset(Inset * inset, DocPair const & p)
817 // Find the dociterators for the beginning and the
818 // end of the inset, for the old and new document.
819 DocRangePair const rp = stepIntoInset(p);
821 // Recurse into the inset. Temporarily replace the dest_pars
822 // paragraph list by the paragraph list of the nested inset.
823 ParagraphList * backup_dest_pars = dest_pars_;
824 dest_pars_ = &inset->asInsetText()->text().paragraphs();
827 ++nested_inset_level_;
829 --nested_inset_level_;
831 dest_pars_ = backup_dest_pars;
835 void Compare::Impl::processSnake(DocRangePair const & rp)
838 getParagraphList(rp.o, pars);
840 // Find insets in this paragaph list
841 DocPair it = rp.from();
842 for (; it.o < rp.o.to; ++it) {
843 Inset * inset = it.o.text()->getPar(it.o.pit()).getInset(it.o.pos());
844 if (inset && inset->editable() && inset->asInsetText()) {
845 // Find the inset in the paragraph list that will be pasted into
846 // the final document. The contents of the inset will be replaced
847 // by the output of the algorithm below.
848 pit_type const pit = it.o.pit() - rp.o.from.pit();
849 pos_type const pos = pit ? it.o.pos() : it.o.pos() - rp.o.from.pos();
850 inset = pars[pit].getInset(pos);
851 LASSERT(inset, continue);
852 diffInset(inset, it);
855 writeToDestBuffer(pars);
859 void Compare::Impl::writeToDestBuffer(DocRange const & range,
863 getParagraphList(range, pars);
868 ParagraphList::iterator it = pars.begin();
869 for (; it != pars.end(); ++it) {
870 it->setChange(Change(type));
874 writeToDestBuffer(pars);
876 if (nested_inset_level_ == 0)
877 compare_.progress(size);
881 void Compare::Impl::writeToDestBuffer(ParagraphList const & pars) const
883 pit_type const pit = dest_pars_->size() - 1;
884 dest_pars_->insert(dest_pars_->end(), pars.begin(), pars.end());
886 mergeParagraph(dest_buf_->params(), *dest_pars_, pit);
890 #include "moc_Compare.cpp"