#include "Compare.h"
-#include "Buffer.h"
#include "BufferParams.h"
#include "Changes.h"
#include <boost/next_prior.hpp>
+#include <cmath>
+
using namespace std;
using namespace lyx::support;
namespace lyx {
-void step_forward(DocIterator & dit)
-{
- dit.top().forwardPos();
-}
-
-void step_backward(DocIterator & dit)
-{
- dit.top().backwardPos();
-}
+enum Direction {
+ Forward = 0,
+ Backward
+};
-bool step_forward(DocIterator & dit, DocIterator const & end)
+static void step(DocIterator & dit, Direction direction)
{
- if (dit == end)
- return false;
- step_forward(dit);
- return true;
+ if (direction == Forward)
+ dit.top().forwardPos();
+ else
+ dit.top().backwardPos();
}
-bool step_backward(DocIterator & dit, DocIterator const & beg)
+static void step(DocIterator & dit, DocIterator const & end, Direction direction)
{
- if (dit == beg)
- return false;
- step_backward(dit);
- return true;
+ if (dit != end)
+ step(dit, direction);
}
+
/**
* A pair of two DocIterators that form a range.
*/
size_t DocRange::length() const
{
- pit_type startpit = from.pit();
- pit_type endpit = to.pit();
- ParagraphList const & ps_ = from.text()->paragraphs();
-
- ParagraphList pars(boost::next(ps_.begin(), startpit),
- boost::next(ps_.begin(), endpit + 1));
-
- // Remove the end of the last paragraph; afterwards, remove the
- // beginning of the first paragraph.
- Paragraph & back = pars.back();
- back.eraseChars(to.pos(), back.size(), false);
- Paragraph & front = pars.front();
- front.eraseChars(0, from.pos(), false);
-
- ParagraphList::const_iterator pit = pars.begin();
- ParagraphList::const_iterator end_it = pars.end();
-
+ ParagraphList const & ps = from.text()->paragraphs();
size_t length = 0;
- for (; pit != end_it; ++pit)
- length += pit->size() + 1;
-
- // The last paragraph has no paragraph-end
- --length;
- return length;
+ pit_type pit = from.pit();
+ pit_type const endpit = to.pit();
+ for (; pit < endpit; ++pit)
+ length += ps[pit].size() + 1;
+ length += to.pos() - from.pos();
+ return length;
}
: o(o_), n(n_)
{}
+ bool operator!=(DocPair const & rhs) {
+ // this might not be intuitive but correct for our purpose
+ return o != rhs.o && n != rhs.n;
+ }
+
+
DocPair & operator++()
{
- step_forward(o);
- step_forward(n);
+ step(o, Forward);
+ step(n, Forward);
+ return *this;
+ }
+
+ DocPair & operator--()
+ {
+ step(o, Backward);
+ step(n, Backward);
return *this;
}
///
};
-DocRangePair stepIntoInset(DocPair const & inset_location)
+static DocRangePair stepIntoInset(DocPair const & inset_location)
{
DocRangePair rp(inset_location, inset_location);
rp.o.from.forwardPos();
rp.n.from.forwardPos();
- step_forward(rp.o.to);
- step_forward(rp.n.to);
+ step(rp.o.to, Forward);
+ step(rp.n.to, Forward);
rp.o.to.backwardPos();
rp.n.to.backwardPos();
return rp;
}
+/**
+ * This class is designed to hold a vector that has both positive as
+ * negative indices. It is internally represented as two vectors, one
+ * for non-zero indices and one for negative indices. In this way, the
+ * vector can grow in both directions.
+ * If an index is not available in the vector, the default value is
+ * returned. If an object is put in the vector beyond its size, the
+ * empty spots in between are also filled with the default value.
+ */
+template<class T>
+class compl_vector {
+public:
+ compl_vector() {}
+
+ void reset(T const & def)
+ {
+ default_ = def;
+ Vp_.clear();
+ Vn_.clear();
+ }
+
+ /// Gets the value at index. If it is not in the vector
+ /// the default value is inserted and returned.
+ T & operator[](int index) {
+ vector<T> & V = index >= 0 ? Vp_ : Vn_;
+ unsigned int const ii = index >= 0 ? index : -index - 1;
+ while (ii >= V.size())
+ V.push_back(default_);
+ return V[ii];
+ }
+
+private:
+ /// The vector for positive indices
+ vector<T> Vp_;
+ /// The vector for negative indices
+ vector<T> Vn_;
+ /// The default value that is inserted in the vector
+ /// if more space is needed
+ T default_;
+};
+
+
/**
* The implementation of the algorithm that does the comparison
* between two documents.
bool abort_;
private:
- // Finds the middle snake and returns the length of the
- // shortest edit script.
+ /// Finds the middle snake and returns the length of the
+ /// shortest edit script.
int find_middle_snake(DocRangePair const & rp, DocPair & middle_snake);
- // This function is called recursively by a divide and conquer
- // algorithm. Each time, the string is divided into two split
- // around the middle snake.
+ enum SnakeResult {
+ NoSnake,
+ SingleSnake,
+ NormalSnake
+ };
+
+ /// Retrieve the middle snake when there is overlap between
+ /// the forward and backward path.
+ SnakeResult retrieve_middle_snake(int k, int D, Direction direction,
+ DocPair & middle_snake);
+
+ /// Find the the furthest reaching D-path (number of horizontal
+ /// and vertical steps; differences between the old and new
+ /// document) in the k-diagonal (vertical minus horizontal steps).
+ void furthest_Dpath_kdiagonal(int D, int k,
+ DocRangePair const & rp, Direction direction);
+
+ /// Is there overlap between the forward and backward path
+ bool overlap(int k, int D);
+
+ /// This function is called recursively by a divide and conquer
+ /// algorithm. Each time, the string is divided into two split
+ /// around the middle snake.
void diff_i(DocRangePair const & rp);
/// Processes the splitted chunks. It either adds them as deleted,
/// and adds the result to /c pars.
void diff_inset(Inset * inset, DocPair const & p);
- // Adds the snake to the destination buffer. The algorithm will
- // recursively be applied to any InsetTexts that are within the snake.
+ /// Adds the snake to the destination buffer. The algorithm will
+ /// recursively be applied to any InsetTexts that are within the snake.
void process_snake(DocRangePair const & rp);
/// Writes the range to the destination buffer
/// Writes the paragraph list to the destination buffer
void writeToDestBuffer(ParagraphList const & copy_pars) const;
- /// The length of the first chunk currently processed
- int N;
- /// The length of the second chunk currently processed
- int M;
+ /// The length of the old chunk currently processed
+ int N_;
+ /// The length of the new chunk currently processed
+ int M_;
+ /// The offset diagonal of the reverse path of the
+ /// currently processed chunk
+ int offset_reverse_diagonal_;
+ /// Is the offset odd or even ?
+ bool odd_offset_;
/// The thread object, used to emit signals to the GUI
Compare const & compare_;
/// The number of nested insets at this level
int nested_inset_level_;
+
+ /// The position/snake in the old/new document
+ /// of the forward/reverse search
+ compl_vector<DocIterator> ofp;
+ compl_vector<DocIterator> nfp;
+ compl_vector<DocIterator> ofs;
+ compl_vector<DocIterator> nfs;
+ compl_vector<DocIterator> orp;
+ compl_vector<DocIterator> nrp;
+ compl_vector<DocIterator> ors;
+ compl_vector<DocIterator> nrs;
};
/////////////////////////////////////////////////////////////////////
}
-void get_paragraph_list(DocRange const & range,
+static void get_paragraph_list(DocRange const & range,
ParagraphList & pars)
{
// Clone the paragraphs within the selection.
}
-bool equal(Inset const * i_o, Inset const * i_n)
+static bool equal(Inset const * i_o, Inset const * i_n)
{
if (!i_o || !i_n)
return false;
}
-bool equal(DocIterator & o, DocIterator & n) {
+static bool equal(DocIterator & o, DocIterator & n) {
Paragraph const & old_par = o.text()->getPar(o.pit());
Paragraph const & new_par = n.text()->getPar(n.pit());
- Inset const * i_o = old_par.getInset(o.pos());
- Inset const * i_n = new_par.getInset(n.pos());
+ char_type const c_o = old_par.getChar(o.pos());
+ char_type const c_n = new_par.getChar(n.pos());
+ if (c_o != c_n)
+ return false;
+
+ if (old_par.isInset(o.pos())) {
+ Inset const * i_o = old_par.getInset(o.pos());
+ Inset const * i_n = new_par.getInset(n.pos());
+
+ if (i_o && i_n)
+ return equal(i_o, i_n);
+ }
- if (i_o && i_n)
- return equal(i_o, i_n);
-
- char_type c_o = old_par.getChar(o.pos());
- char_type c_n = new_par.getChar(n.pos());
Font fo = old_par.getFontSettings(o.buffer()->params(), o.pos());
Font fn = new_par.getFontSettings(n.buffer()->params(), n.pos());
- return c_o == c_n && fo == fn;
+ return fo == fn;
}
-void traverse_snake_back(DocRangePair & rp)
+/// Traverses a snake in a certain direction. p points to a
+/// position in the old and new file and they are synchronously
+/// moved along the snake. The function returns true if a snake
+/// was found.
+static bool traverse_snake(DocPair & p, DocRangePair const & range,
+ Direction direction)
{
- while (true) {
- // Traverse snake
- if (!step_backward(rp.o.to, rp.o.from))
- break;
-
- if (!step_backward(rp.n.to, rp.n.from)) {
- step_forward(rp.o.to);
- break;
+ bool ret = false;
+ DocPair const & p_end =
+ direction == Forward ? range.to() : range.from();
+
+ while (p != p_end) {
+ if (direction == Backward)
+ --p;
+ if (!equal(p.o, p.n)) {
+ if (direction == Backward)
+ ++p;
+ return ret;
}
+ if (direction == Forward)
+ ++p;
+ ret = true;
+ }
+ return ret;
+}
- if (!equal(rp.o.to, rp.n.to)) {
- step_forward(rp.o.to);
- step_forward(rp.n.to);
- break;
- }
+
+/////////////////////////////////////////////////////////////////////
+//
+// Compare::Impl
+//
+/////////////////////////////////////////////////////////////////////
+
+
+void Compare::Impl::furthest_Dpath_kdiagonal(int D, int k,
+ DocRangePair const & rp, Direction direction)
+{
+ compl_vector<DocIterator> & op = direction == Forward ? ofp : orp;
+ compl_vector<DocIterator> & np = direction == Forward ? nfp : nrp;
+ compl_vector<DocIterator> & os = direction == Forward ? ofs : ors;
+ compl_vector<DocIterator> & ns = direction == Forward ? nfs : nrs;
+
+ // A vertical step means stepping one character in the new document.
+ bool vertical_step = k == -D;
+ if (!vertical_step && k != D) {
+ vertical_step = direction == Forward
+ ? op[k - 1] < op[k + 1] : op[k - 1] > op[k + 1];
}
+
+ // Where do we take the step from ?
+ int const kk = vertical_step ? k + 1 : k - 1;
+ DocPair p(op[kk], np[kk]);
+
+ // If D==0 we simulate a vertical step from (0,-1) by doing nothing.
+ if (D != 0) {
+ // Take a step
+ if (vertical_step && direction == Forward)
+ step(p.n, rp.n.to, direction);
+ else if (vertical_step && direction == Backward)
+ step(p.n, rp.n.from, direction);
+ else if (!vertical_step && direction == Forward)
+ step(p.o, rp.o.to, direction);
+ else if (!vertical_step && direction == Backward)
+ step(p.o, rp.o.from, direction);
+ }
+
+ // Traverse snake
+ if (traverse_snake(p, rp, direction)) {
+ // Record last snake
+ os[k] = p.o;
+ ns[k] = p.n;
+ } else {
+ // Copy last snake from the previous step
+ os[k] = os[kk];
+ ns[k] = ns[kk];
+ }
+
+ //Record new position
+ op[k] = p.o;
+ np[k] = p.n;
}
-void traverse_snake_forw(DocRangePair & rp)
+bool Compare::Impl::overlap(int k, int D)
{
- while (equal(rp.o.from, rp.n.from)) {
- if (!step_forward(rp.o.from, rp.o.to))
- break;
+ // To generalize for the forward and reverse checks
+ int kk = offset_reverse_diagonal_ - k;
+
+ // Can we have overlap ?
+ if (kk <= D && kk >= -D) {
+ // Do we have overlap ?
+ if (odd_offset_)
+ return ofp[k] >= orp[kk] && nfp[k] >= nrp[kk];
+ else
+ return ofp[kk] >= orp[k] && nfp[kk] >= nrp[k];
+ }
+ return false;
+}
- if (!step_forward(rp.n.from, rp.n.to)) {
- step_backward(rp.o.from);
- break;
- }
+
+Compare::Impl::SnakeResult Compare::Impl::retrieve_middle_snake(
+ int k, int D, Direction direction, DocPair & middle_snake)
+{
+ compl_vector<DocIterator> & os = direction == Forward ? ofs : ors;
+ compl_vector<DocIterator> & ns = direction == Forward ? nfs : nrs;
+ compl_vector<DocIterator> & os_r = direction == Forward ? ors : ofs;
+ compl_vector<DocIterator> & ns_r = direction == Forward ? nrs : nfs;
+
+ // The diagonal while doing the backward search
+ int kk = -k + offset_reverse_diagonal_;
+
+ // Did we find a snake ?
+ if (os[k].empty() && os_r[kk].empty()) {
+ // No, there is no snake at all, in which case
+ // the length of the shortest edit script is M+N.
+ LASSERT(2 * D - odd_offset_ == M_ + N_, /**/);
+ return NoSnake;
+ }
+
+ if (os[k].empty()) {
+ // Yes, but there is only 1 snake and we found it in the
+ // reverse path.
+ middle_snake.o = os_r[kk];
+ middle_snake.n = ns_r[kk];
+ return SingleSnake;
}
+
+ middle_snake.o = os[k];
+ middle_snake.n = ns[k];
+ return NormalSnake;
}
-/////////////////////////////////////////////////////////////////////
-//
-// Compare::Impl
-//
-/////////////////////////////////////////////////////////////////////
int Compare::Impl::find_middle_snake(DocRangePair const & rp,
DocPair & middle_snake)
{
- N = rp.o.length();
- M = rp.n.length();
- return M+N;
+ // The lengths of the old and new chunks.
+ N_ = rp.o.length();
+ M_ = rp.n.length();
+
+ // Forward paths are centered around the 0-diagonal; reverse paths
+ // are centered around the diagonal N - M. (Delta in the article)
+ offset_reverse_diagonal_ = N_ - M_;
+
+ // If the offset is odd, only check for overlap while extending forward
+ // paths, otherwise only check while extending reverse paths.
+ odd_offset_ = (offset_reverse_diagonal_ % 2 != 0);
+
+ ofp.reset(rp.o.from);
+ nfp.reset(rp.n.from);
+ ofs.reset(DocIterator());
+ nfs.reset(DocIterator());
+ orp.reset(rp.o.to);
+ nrp.reset(rp.n.to);
+ ors.reset(DocIterator());
+ nrs.reset(DocIterator());
+
+ // D is the number of horizontal and vertical steps, i.e.
+ // different characters in the old and new chunk.
+ int const D_max = ceil(((double)M_ + N_)/2);
+ for (int D = 0; D <= D_max; ++D) {
+
+ // Forward and reverse paths
+ for (int f = 0; f < 2; ++f) {
+ Direction direction = f == 0 ? Forward : Backward;
+
+ // Diagonals between -D and D can be reached by a D-path
+ for (int k = -D; k <= D; k += 2) {
+ // Find the furthest reaching D-path on this diagonal
+ furthest_Dpath_kdiagonal(D, k, rp, direction);
+
+ // Only check for overlap for forward paths if the offset is odd
+ // and only for reverse paths if the offset is even.
+ if (odd_offset_ == (direction == Forward)) {
+
+ // Do the forward and backward paths overlap ?
+ if (overlap(k, D - odd_offset_)) {
+ retrieve_middle_snake(k, D, direction, middle_snake);
+ return 2 * D - odd_offset_;
+ }
+ }
+ }
+ }
+ }
+ // This should never be reached
+ return -2;
}
DocRangePair rp(old_buf_, new_buf_);
DocPair from = rp.from();
- traverse_snake_forw(rp);
- DocRangePair const snake(from, rp.from());
+ traverse_snake(from, rp, Forward);
+ DocRangePair const snake(rp.from(), from);
process_snake(snake);
// Start the recursive algorithm
} else {
// Retrieve the complete snake
- DocRangePair first_part(rp.from(), middle_snake);
- traverse_snake_back(first_part);
-
- DocRangePair second_part(middle_snake, rp.to());
- traverse_snake_forw(second_part);
+ DocPair first_part_end = middle_snake;
+ traverse_snake(first_part_end, rp, Backward);
+ DocRangePair first_part(rp.from(), first_part_end);
+ DocPair second_part_begin = middle_snake;
+ traverse_snake(second_part_begin, rp, Forward);
+ DocRangePair second_part(second_part_begin, rp.to());
+
// Split the string in three parts:
// 1. in front of the snake
diff_part(first_part);
projects / lyx.git / blobdiff