// Determines the longest common subsequence between two sequences and populates the
// list of diffs derived from the result as we go. Each recursive step identifies
// a middle "snake" (a common sequence) then splits the problem until nothing remains.
private void ComputeLCS(Substring left, Substring right)
{
int n = left.Length;
int m = right.Length;
if (n != 0 && m != 0)
{
int middleSnakeLeftStartIndex, middleSnakeRightStartIndex, middleSnakeLength;
int ses = FindMiddleSnake(left, right, out middleSnakeLeftStartIndex, out middleSnakeRightStartIndex, out middleSnakeLength);
if (ses > 1)
{
// If SES >= 2 then the edit script includes at least 2 differences, so we divide the problem.
ComputeLCS(
left.Extract(0, middleSnakeLeftStartIndex),
right.Extract(0, middleSnakeRightStartIndex));
EmitDiffsFromCommonSequence(
left.Range.StartIndex + middleSnakeLeftStartIndex,
right.Range.StartIndex + middleSnakeRightStartIndex,
middleSnakeLength);
ComputeLCS(
left.Extract(middleSnakeLeftStartIndex + middleSnakeLength),
right.Extract(middleSnakeRightStartIndex + middleSnakeLength));
}
else
{
// If SES = 1, then exactly one symbol needs to be added or deleted from either sequence.
// If SES = 0, then both sequences are equal.
if (ses != 0)
{
// The middle snake is the common part after the change so we just need to grab the
// common part before the change.
EmitDiffsFromCommonSequence(
left.Range.StartIndex,
right.Range.StartIndex,
Math.Min(middleSnakeLeftStartIndex, middleSnakeRightStartIndex));
}
EmitDiffsFromCommonSequence(
left.Range.StartIndex + middleSnakeLeftStartIndex,
right.Range.StartIndex + middleSnakeRightStartIndex,
middleSnakeLength);
}
}
}
public static void PopulateDiffs(IList <Diff> diffs, Substring left, Substring right, bool boundRuntime)
{
int n = left.Length;
int m = right.Length;
int max = CeilNPlusMOverTwo(n, m);
if (boundRuntime && ((long)n) * ((long)m) > TooLong)
{
max = (int)Math.Pow(max, PowLimit - 1.0);
}
DiffAlgorithm algorithm = new DiffAlgorithm(diffs, left.Range.StartIndex, right.Range.StartIndex, max);
algorithm.ComputeLCS(left, right);
algorithm.FlushDiffs(left.Range.EndIndex, right.Range.EndIndex);
}
private static void WriteContext(MarkupStreamWriter writer, Substring context, int maxContextLength)
{
if (context.Length < maxContextLength)
{
writer.Write(context.ToString());
}
else
{
int split = maxContextLength / 2;
if (split > 0)
{
writer.Write(context.Extract(0, split).ToString());
writer.WriteEllipsis();
writer.Write(context.Extract(context.Length - split));
}
}
}
private static void FastDiff(IList <Diff> diffs, Substring left, Substring right, bool boundRuntime)
{
// If either document is empty, then the change covers the whole document.
if (left.Length == 0 || right.Length == 0)
{
diffs.Add(new Diff(DiffKind.Change, left.Range, right.Range));
return;
}
// Reduce the problem size by identifying a common prefix and suffix, if any.
int commonPrefixLength = left.FindCommonPrefixLength(right);
if (commonPrefixLength != 0)
{
if (commonPrefixLength == left.Length && commonPrefixLength == right.Length)
{
diffs.Add(new Diff(DiffKind.NoChange, left.Range, right.Range));
return;
}
diffs.Add(new Diff(DiffKind.NoChange, new Range(left.Range.StartIndex, commonPrefixLength),
new Range(right.Range.StartIndex, commonPrefixLength)));
}
int commonSuffixLength = left.Extract(commonPrefixLength).FindCommonSuffixLength(right.Extract(commonPrefixLength));
// Now work on the middle part.
Substring leftMiddle = left.Extract(commonPrefixLength, left.Length - commonPrefixLength - commonSuffixLength);
Substring rightMiddle = right.Extract(commonPrefixLength, right.Length - commonPrefixLength - commonSuffixLength);
SlowDiff(diffs, leftMiddle, rightMiddle, boundRuntime);
// Tack on the final diff for the common suffix, if any.
if (commonSuffixLength != 0)
{
diffs.Add(new Diff(DiffKind.NoChange,
new Range(leftMiddle.Range.EndIndex, commonSuffixLength),
new Range(rightMiddle.Range.EndIndex, commonSuffixLength)));
}
}
// Finds a middle "snake", which is a (possibly empty) sequence of diagonal edges in the edit
// graph. Thus it directly represents a common sequence.
//
// In essence, this function searches D-paths forward and backward in the sequence until it
// finds the middle snake. The middle snake informs us about a common sequence sandwiched
// between two other sequences that may contain changes. By definition, the left and right
// middle snakes must be of equal length.
private int FindMiddleSnake(Substring left, Substring right, out int middleSnakeLeftStartIndex, out int middleSnakeRightStartIndex, out int middleSnakeLength)
{
int n = left.Length;
int m = right.Length;
int delta = n - m;
bool isDeltaOdd = (delta & 1) != 0;
leftVector[max + 1] = 0;
rightVector[max - 1] = n;
int end = Math.Min(CeilNPlusMOverTwo(n, m), max);
for (int d = 0; d <= end; d++)
{
// Search forward D-paths.
for (int k = -d; k <= d; k += 2)
{
// Find the end of the furthest reaching forward D-path in diagonal k.
int x = k == -d || k != d && leftVector[max + k - 1] < leftVector[max + k + 1]
? leftVector[max + k + 1]
: leftVector[max + k - 1] + 1;
int origX = x;
for (int y = x - k; x < n && y < m && left[x] == right[y];)
{
x += 1;
y += 1;
}
leftVector[max + k] = x;
// If the D-path is feasible and overlaps the furthest reaching reverse (D-1)-Path in diagonal k
// then we have found the middle snake.
if (isDeltaOdd && k >= delta - d + 1 && k <= delta + d - 1)
{
int u = rightVector[max + k - delta];
if (x >= u)
{
middleSnakeLeftStartIndex = origX;
middleSnakeRightStartIndex = origX - k;
middleSnakeLength = x - origX;
return(d * 2 - 1);
}
}
}
// Search reverse D-paths.
for (int k = -d; k <= d; k += 2)
{
// Find the end of the furthest reaching reverse D-path in diagonal k + delta.
int u = k == d || k != -d && rightVector[max + k - 1] < rightVector[max + k + 1]
? rightVector[max + k - 1]
: rightVector[max + k + 1] - 1;
int kPlusDelta = k + delta;
int origU = u;
int v;
for (v = u - kPlusDelta; u > 0 && v > 0 && left[u - 1] == right[v - 1];)
{
u -= 1;
v -= 1;
}
rightVector[max + k] = u;
// If the D-path is feasible and overlaps the furthest reaching forward D-Path in diagonal k
// then we have found the middle snake.
if (!isDeltaOdd && kPlusDelta >= -d && kPlusDelta <= d)
{
int x = leftVector[max + kPlusDelta];
if (u <= x)
{
middleSnakeLeftStartIndex = u;
middleSnakeRightStartIndex = v;
middleSnakeLength = origU - u;
return(d * 2);
}
}
}
}
// We have exceeded the maximum effort we are willing to expend finding a diff.
//
// So we artificially divide the problem by finding the snakes in the forward / reverse
// direction that have the most progress toward (N, M) / (0, 0). These are the
// ones that maximize x + y / minimize u + v. The snake we return will not actually
// be the middle snake (since we haven't found it yet) but it will be good enough
// to reduce the problem.
//
// These snakes all begin on the same diagonal as the others of equal
// progress in the same direction. As there may be several of them, we need a way
// to decide which one to pursue.
//
// The Eclipse LCS implementation chooses the median of these snakes with respect to k.
// Intuitively this is the one that is nearer the direct line between (0, 0) and (N, M)
// so it has a good chance of forming a path with more balanced changes between A and B
// than the snakes that consist of significantly more changes to A than B or vice-versa.
// Consequently the median of theses snakes should yield a pretty good approximation. -- Jeff.
int bestProgress = 0;
Dictionary <int, bool> bestKs = new Dictionary <int, bool>(); // with the forward direction indicated by value true
for (int k = -end; k <= end; k += 2)
{
// Forward direction.
int x = leftVector[max + k];
int y = x - k;
if (x < n && y < m)
{
int progress = x + y;
if (progress >= bestProgress)
{
if (progress > bestProgress)
{
bestProgress = progress;
bestKs.Clear();
}
bestKs[k] = true;
}
}
// Reverse direction.
int u = rightVector[max + k];
int v = u - k - delta;
if (u >= 0 && v >= 0)
{
int progress = n + m - u - v;
if (progress >= bestProgress)
{
if (progress > bestProgress)
{
bestProgress = progress;
bestKs.Clear();
}
bestKs[k] = false;
}
}
}
int[] sortedKs = new int[bestKs.Count];
bestKs.Keys.CopyTo(sortedKs, 0);
Array.Sort(sortedKs);
int medianK = sortedKs[sortedKs.Length / 2];
if (bestKs[medianK])
{
int x = leftVector[max + medianK];
int y = x - medianK;
middleSnakeLeftStartIndex = x;
middleSnakeRightStartIndex = y;
}
else
{
int u = rightVector[max + medianK];
int v = u - medianK - delta;
middleSnakeLeftStartIndex = u;
middleSnakeRightStartIndex = v;
}
middleSnakeLength = 0;
// We need to return the length of the shortest edit script but we don't actually know
// what it is. Fortunately the caller does not care as long as it's greater than 2, which
// it must be since d > end >= max > 2.
return(int.MaxValue);
}
private static void SlowDiff(IList <Diff> diffs, Substring left, Substring right, bool boundRuntime)
{
DiffAlgorithm.PopulateDiffs(diffs, left, right, boundRuntime);
}