/// <summary>
/// Find the differences between two texts. Assumes that the texts do not
/// have any common prefix or suffix.
/// </summary>
/// <param name="text1">Old string to be diffed.</param>
/// <param name="text2">New string to be diffed.</param>
/// <param name="checklines">Speedup flag. If false, then don't run a line-level diff first to identify the changed areas. If true, then run a faster slightly less optimal diff.</param>
/// <param name="token">Cancellation token for cooperative cancellation</param>
/// <param name="optimizeForSpeed">Should optimizations be enabled?</param>
/// <returns></returns>
private static List <Diff> ComputeImpl(
string text1,
string text2,
bool checklines, CancellationToken token, bool optimizeForSpeed)
{
var diffs = new List <Diff>();
if (text1.Length == 0)
{
// Just add some text (speedup).
diffs.Add(Diff.Insert(text2));
return(diffs);
}
if (text2.Length == 0)
{
// Just delete some text (speedup).
diffs.Add(Diff.Delete(text1));
return(diffs);
}
var longtext = text1.Length > text2.Length ? text1 : text2;
var shorttext = text1.Length > text2.Length ? text2 : text1;
var i = longtext.IndexOf(shorttext, StringComparison.Ordinal);
if (i != -1)
{
// Shorter text is inside the longer text (speedup).
var op = text1.Length > text2.Length ? Operation.Delete : Operation.Insert;
diffs.Add(Diff.Create(op, longtext.Substring(0, i)));
diffs.Add(Diff.Equal(shorttext));
diffs.Add(Diff.Create(op, longtext.Substring(i + shorttext.Length)));
return(diffs);
}
if (shorttext.Length == 1)
{
// Single character string.
// After the previous speedup, the character can't be an equality.
diffs.Add(Diff.Delete(text1));
diffs.Add(Diff.Insert(text2));
return(diffs);
}
// Don't risk returning a non-optimal diff if we have unlimited time.
if (optimizeForSpeed)
{
// Check to see if the problem can be split in two.
var result = TextUtil.HalfMatch(text1, text2);
if (!result.IsEmpty)
{
// A half-match was found, sort out the return data.
// Send both pairs off for separate processing.
var diffsA = Compute(result.Prefix1, result.Prefix2, checklines, token, optimizeForSpeed);
var diffsB = Compute(result.Suffix1, result.Suffix2, checklines, token, optimizeForSpeed);
// Merge the results.
diffs = diffsA;
diffs.Add(Diff.Equal(result.CommonMiddle));
diffs.AddRange(diffsB);
return(diffs);
}
}
if (checklines && text1.Length > 100 && text2.Length > 100)
{
return(LineDiff(text1, text2, token, optimizeForSpeed));
}
return(MyersDiffBisect(text1, text2, token, optimizeForSpeed));
}
/// <summary>
/// Find the 'middle snake' of a diff, split the problem in two
/// and return the recursively constructed diff.
/// See Myers 1986 paper: An O(ND) Difference Algorithm and Its Variations.
/// </summary>
/// <param name="text1"></param>
/// <param name="text2"></param>
/// <param name="token"></param>
/// <param name="optimizeForSpeed"></param>
/// <returns></returns>
internal static List <Diff> MyersDiffBisect(string text1, string text2, CancellationToken token, bool optimizeForSpeed)
{
// Cache the text lengths to prevent multiple calls.
var text1Length = text1.Length;
var text2Length = text2.Length;
var maxD = (text1Length + text2Length + 1) / 2;
var vOffset = maxD;
var vLength = 2 * maxD;
var v1 = new int[vLength];
var v2 = new int[vLength];
for (var x = 0; x < vLength; x++)
{
v1[x] = -1;
}
for (var x = 0; x < vLength; x++)
{
v2[x] = -1;
}
v1[vOffset + 1] = 0;
v2[vOffset + 1] = 0;
var delta = text1Length - text2Length;
// If the total number of characters is odd, then the front path will
// collide with the reverse path.
var front = delta % 2 != 0;
// Offsets for start and end of k loop.
// Prevents mapping of space beyond the grid.
var k1Start = 0;
var k1End = 0;
var k2Start = 0;
var k2End = 0;
for (var d = 0; d < maxD; d++)
{
// Bail out if cancelled.
if (token.IsCancellationRequested)
{
break;
}
// Walk the front path one step.
for (var k1 = -d + k1Start; k1 <= d - k1End; k1 += 2)
{
var k1Offset = vOffset + k1;
int x1;
if (k1 == -d || k1 != d && v1[k1Offset - 1] < v1[k1Offset + 1])
{
x1 = v1[k1Offset + 1];
}
else
{
x1 = v1[k1Offset - 1] + 1;
}
var y1 = x1 - k1;
while (x1 < text1Length && y1 < text2Length &&
text1[x1] == text2[y1])
{
x1++;
y1++;
}
v1[k1Offset] = x1;
if (x1 > text1Length)
{
// Ran off the right of the graph.
k1End += 2;
}
else if (y1 > text2Length)
{
// Ran off the bottom of the graph.
k1Start += 2;
}
else if (front)
{
var k2Offset = vOffset + delta - k1;
if (k2Offset >= 0 && k2Offset < vLength && v2[k2Offset] != -1)
{
// Mirror x2 onto top-left coordinate system.
var x2 = text1Length - v2[k2Offset];
if (x1 >= x2)
{
// Overlap detected.
return(BisectSplit(text1, text2, x1, y1, token, optimizeForSpeed));
}
}
}
}
// Walk the reverse path one step.
for (var k2 = -d + k2Start; k2 <= d - k2End; k2 += 2)
{
var k2Offset = vOffset + k2;
int x2;
if (k2 == -d || k2 != d && v2[k2Offset - 1] < v2[k2Offset + 1])
{
x2 = v2[k2Offset + 1];
}
else
{
x2 = v2[k2Offset - 1] + 1;
}
var y2 = x2 - k2;
while (x2 < text1Length && y2 < text2Length &&
text1[text1Length - x2 - 1]
== text2[text2Length - y2 - 1])
{
x2++;
y2++;
}
v2[k2Offset] = x2;
if (x2 > text1Length)
{
// Ran off the left of the graph.
k2End += 2;
}
else if (y2 > text2Length)
{
// Ran off the top of the graph.
k2Start += 2;
}
else if (!front)
{
var k1Offset = vOffset + delta - k2;
if (k1Offset >= 0 && k1Offset < vLength && v1[k1Offset] != -1)
{
var x1 = v1[k1Offset];
var y1 = vOffset + x1 - k1Offset;
// Mirror x2 onto top-left coordinate system.
x2 = text1Length - v2[k2Offset];
if (x1 >= x2)
{
// Overlap detected.
return(BisectSplit(text1, text2, x1, y1, token, optimizeForSpeed));
}
}
}
}
}
// Diff took too long and hit the deadline or
// number of Diffs equals number of characters, no commonality at all.
var diffs = new List <Diff> {
Diff.Delete(text1), Diff.Insert(text2)
};
return(diffs);
}
/// <summary>
/// Reduce the number of edits by eliminating operationally trivial equalities.
/// </summary>
/// <param name="diffs"></param>
/// <param name="diffEditCost"></param>
public static void CleanupEfficiency(this List <Diff> diffs, short diffEditCost = 4)
{
var changes = false;
// Stack of indices where equalities are found.
var equalities = new Stack <int>();
// Always equal to equalities[equalitiesLength-1][1]
var lastEquality = string.Empty;
// Is there an insertion operation before the last equality.
var preIns = false;
// Is there a deletion operation before the last equality.
var preDel = false;
// Is there an insertion operation after the last equality.
var postIns = false;
// Is there a deletion operation after the last equality.
var postDel = false;
for (var i = 0; i < diffs.Count; i++)
{
if (diffs[i].Operation == Operation.Equal)
{ // Equality found.
if (diffs[i].Text.Length < diffEditCost && (postIns || postDel))
{
// Candidate found.
equalities.Push(i);
preIns = postIns;
preDel = postDel;
lastEquality = diffs[i].Text;
}
else
{
// Not a candidate, and can never become one.
equalities.Clear();
lastEquality = string.Empty;
}
postIns = postDel = false;
}
else
{ // An insertion or deletion.
if (diffs[i].Operation == Operation.Delete)
{
postDel = true;
}
else
{
postIns = true;
}
/*
* Five types to be split:
* <ins>A</ins><del>B</del>XY<ins>C</ins><del>D</del>
* <ins>A</ins>X<ins>C</ins><del>D</del>
* <ins>A</ins><del>B</del>X<ins>C</ins>
* <ins>A</del>X<ins>C</ins><del>D</del>
* <ins>A</ins><del>B</del>X<del>C</del>
*/
if ((lastEquality.Length != 0) &&
((preIns && preDel && postIns && postDel) ||
((lastEquality.Length < diffEditCost / 2) &&
(preIns ? 1 : 0) + (preDel ? 1 : 0) + (postIns ? 1 : 0)
+ (postDel ? 1 : 0) == 3)))
{
diffs.Splice(equalities.Peek(), 1, Diff.Delete(lastEquality), Diff.Insert(lastEquality));
equalities.Pop(); // Throw away the equality we just deleted.
lastEquality = string.Empty;
if (preIns && preDel)
{
// No changes made which could affect previous entry, keep going.
postIns = postDel = true;
equalities.Clear();
}
else
{
if (equalities.Count > 0)
{
equalities.Pop();
}
i = equalities.Count > 0 ? equalities.Peek() : -1;
postIns = postDel = false;
}
changes = true;
}
}
}
if (changes)
{
diffs.CleanupMerge();
}
}
/// <summary>
/// Reduce the number of edits by eliminating semantically trivial equalities.
/// </summary>
/// <param name="diffs"></param>
public static void CleanupSemantic(this List <Diff> diffs)
{
// Stack of indices where equalities are found.
var equalities = new Stack <int>();
// Always equal to equalities[equalitiesLength-1][1]
string lastEquality = null;
var pointer = 0; // Index of current position.
// Number of characters that changed prior to the equality.
var lengthInsertions1 = 0;
var lengthDeletions1 = 0;
// Number of characters that changed after the equality.
var lengthInsertions2 = 0;
var lengthDeletions2 = 0;
while (pointer < diffs.Count)
{
if (diffs[pointer].Operation == Operation.Equal)
{ // Equality found.
equalities.Push(pointer);
lengthInsertions1 = lengthInsertions2;
lengthDeletions1 = lengthDeletions2;
lengthInsertions2 = 0;
lengthDeletions2 = 0;
lastEquality = diffs[pointer].Text;
}
else
{ // an insertion or deletion
if (diffs[pointer].Operation == Operation.Insert)
{
lengthInsertions2 += diffs[pointer].Text.Length;
}
else
{
lengthDeletions2 += diffs[pointer].Text.Length;
}
// Eliminate an equality that is smaller or equal to the edits on both
// sides of it.
if (lastEquality != null && (lastEquality.Length
<= Math.Max(lengthInsertions1, lengthDeletions1)) &&
(lastEquality.Length
<= Math.Max(lengthInsertions2, lengthDeletions2)))
{
// Duplicate record.
diffs.Splice(equalities.Peek(), 1, Diff.Delete(lastEquality), Diff.Insert(lastEquality));
// Throw away the equality we just deleted.
equalities.Pop();
if (equalities.Count > 0)
{
equalities.Pop();
}
pointer = equalities.Count > 0 ? equalities.Peek() : -1;
lengthInsertions1 = 0; // Reset the counters.
lengthDeletions1 = 0;
lengthInsertions2 = 0;
lengthDeletions2 = 0;
lastEquality = null;
}
}
pointer++;
}
diffs.CleanupMerge();
diffs.CleanupSemanticLossless();
// Find any overlaps between deletions and insertions.
// e.g: <del>abcxxx</del><ins>xxxdef</ins>
// -> <del>abc</del>xxx<ins>def</ins>
// e.g: <del>xxxabc</del><ins>defxxx</ins>
// -> <ins>def</ins>xxx<del>abc</del>
// Only extract an overlap if it is as big as the edit ahead or behind it.
pointer = 1;
while (pointer < diffs.Count)
{
if (diffs[pointer - 1].Operation == Operation.Delete &&
diffs[pointer].Operation == Operation.Insert)
{
var deletion = diffs[pointer - 1].Text;
var insertion = diffs[pointer].Text;
var overlapLength1 = TextUtil.CommonOverlap(deletion, insertion);
var overlapLength2 = TextUtil.CommonOverlap(insertion, deletion);
if (overlapLength1 >= overlapLength2)
{
if (overlapLength1 >= deletion.Length / 2.0 ||
overlapLength1 >= insertion.Length / 2.0)
{
// Overlap found.
// Insert an equality and trim the surrounding edits.
var newDiffs = new[]
{
Diff.Delete(deletion.Substring(0, deletion.Length - overlapLength1)),
Diff.Equal(insertion.Substring(0, overlapLength1)),
Diff.Insert(insertion.Substring(overlapLength1))
};
diffs.Splice(pointer - 1, 2, newDiffs);
pointer++;
}
}
else
{
if (overlapLength2 >= deletion.Length / 2.0 ||
overlapLength2 >= insertion.Length / 2.0)
{
// Reverse overlap found.
// Insert an equality and swap and trim the surrounding edits.
diffs.Splice(pointer - 1, 2,
Diff.Insert(insertion.Substring(0, insertion.Length - overlapLength2)),
Diff.Equal(deletion.Substring(0, overlapLength2)),
Diff.Delete(deletion.Substring(overlapLength2)
));
pointer++;
}
}
pointer++;
}
pointer++;
}
}
/// <summary>
/// Reorder and merge like edit sections. Merge equalities.
/// Any edit section can move as long as it doesn't cross an equality.
/// </summary>
/// <param name="diffs">list of Diffs</param>
internal static void CleanupMerge(this List <Diff> diffs)
{
// Add a dummy entry at the end.
diffs.Add(Diff.Equal(string.Empty));
var nofdiffs = 0;
var sbDelete = new StringBuilder();
var sbInsert = new StringBuilder();
var pointer = 0;
while (pointer < diffs.Count)
{
switch (diffs[pointer].Operation)
{
case Operation.Insert:
nofdiffs++;
sbInsert.Append(diffs[pointer].Text);
pointer++;
break;
case Operation.Delete:
nofdiffs++;
sbDelete.Append(diffs[pointer].Text);
pointer++;
break;
case Operation.Equal:
// Upon reaching an equality, check for prior redundancies.
if (nofdiffs > 1)
{
if (sbDelete.Length > 0 && sbInsert.Length > 0)
{
// Factor out any common prefixies.
var commonlength = TextUtil.CommonPrefix(sbInsert, sbDelete);
if (commonlength != 0)
{
var commonprefix = sbInsert.ToString(0, commonlength);
sbInsert.Remove(0, commonlength);
sbDelete.Remove(0, commonlength);
var index = pointer - nofdiffs - 1;
if (index >= 0 && diffs[index].Operation == Operation.Equal)
{
diffs[index] = diffs[index].Replace(diffs[index].Text + commonprefix);
}
else
{
diffs.Insert(0, Diff.Equal(commonprefix));
pointer++;
}
}
// Factor out any common suffixies.
commonlength = TextUtil.CommonSuffix(sbInsert, sbDelete);
if (commonlength != 0)
{
var commonsuffix = sbInsert.ToString(sbInsert.Length - commonlength, commonlength);
sbInsert.Remove(sbInsert.Length - commonlength, commonlength);
sbDelete.Remove(sbDelete.Length - commonlength, commonlength);
diffs[pointer] = diffs[pointer].Replace(commonsuffix + diffs[pointer].Text);
}
}
// Delete the offending records and add the merged ones.
IEnumerable <Diff> Replacements()
{
if (sbDelete.Length > 0)
{
yield return(Diff.Delete(sbDelete.ToString()));
}
if (sbInsert.Length > 0)
{
yield return(Diff.Insert(sbInsert.ToString()));
}
}
var replacements = Replacements().ToList();
diffs.Splice(pointer - nofdiffs, nofdiffs, replacements);
pointer = pointer - nofdiffs + replacements.Count + 1;
}
else if (pointer > 0 && diffs[pointer - 1].Operation == Operation.Equal)
{
// Merge this equality with the previous one.
diffs[pointer - 1] = diffs[pointer - 1].Replace(diffs[pointer - 1].Text + diffs[pointer].Text);
diffs.RemoveAt(pointer);
}
else
{
pointer++;
}
nofdiffs = 0;
sbDelete.Clear();
sbInsert.Clear();
break;
}
}
if (diffs.Last().Text.Length == 0)
{
diffs.RemoveAt(diffs.Count - 1); // Remove the dummy entry at the end.
}
// Second pass: look for single edits surrounded on both sides by
// equalities which can be shifted sideways to eliminate an equality.
// e.g: A<ins>BA</ins>C -> <ins>AB</ins>AC
var changes = false;
// Intentionally ignore the first and last element (don't need checking).
for (var i = 1; i < diffs.Count - 1; i++)
{
var previous = diffs[i - 1];
var current = diffs[i];
var next = diffs[i + 1];
if (previous.Operation == Operation.Equal && next.Operation == Operation.Equal)
{
// This is a single edit surrounded by equalities.
if (current.Text.EndsWith(previous.Text, StringComparison.Ordinal))
{
// Shift the edit over the previous equality.
var text = previous.Text + current.Text.Substring(0, current.Text.Length - previous.Text.Length);
diffs[i] = current.Replace(text);
diffs[i + 1] = next.Replace(previous.Text + next.Text);
diffs.Splice(i - 1, 1);
changes = true;
}
else if (current.Text.StartsWith(next.Text, StringComparison.Ordinal))
{
// Shift the edit over the next equality.
diffs[i - 1] = previous.Replace(previous.Text + next.Text);
diffs[i] = current.Replace(current.Text.Substring(next.Text.Length) + next.Text);
diffs.Splice(i + 1, 1);
changes = true;
}
}
}
// If shifts were made, the diff needs reordering and another shift sweep.
if (changes)
{
diffs.CleanupMerge();
}
}
