Esempio n. 1
0
		/// <summary>
		/// Find the difference in 2 text documents, comparing by textlines.
		/// The algorithm itself is comparing 2 arrays of numbers so when comparing 2 text documents
		/// each line is converted into a (hash) number. This hash-value is computed by storing all
		/// textlines into a common hashtable so i can find dublicates in there, and generating a 
		/// new number each time a new textline is inserted.
		/// </summary>
		/// <param name="TextA">A-version of the text (usualy the old one)</param>
		/// <param name="TextB">B-version of the text (usualy the new one)</param>
		/// <param name="trimSpace">When set to true, all leading and trailing whitespace characters are stripped out before the comparation is done.</param>
		/// <param name="ignoreSpace">When set to true, all whitespace characters are converted to a single space character before the comparation is done.</param>
		/// <param name="ignoreCase">When set to true, all characters are converted to their lowercase equivivalence before the comparation is done.</param>
		/// <returns>Returns a array of Items that describe the differences.</returns>
		public static Item[] DiffText(string TextA, string TextB, bool trimSpace, bool ignoreSpace, bool ignoreCase) {
			// prepare the input-text and convert to comparable numbers.
			Hashtable h = new Hashtable(TextA.Length + TextB.Length);
			
			// The A-Version of the data (original data) to be compared.
			DiffData DataA = new DiffData(DiffCodes(TextA, h, trimSpace, ignoreSpace, ignoreCase));
			
			// The B-Version of the data (modified data) to be compared.
			DiffData DataB = new DiffData(DiffCodes(TextB, h, trimSpace, ignoreSpace, ignoreCase));
			
			h = null; // free up hashtable memory (maybe)
			
			int MAX = DataA.Length + DataB.Length + 1;
			/// vector for the (0,0) to (x,y) search
			int[] DownVector = new int[2 * MAX + 2];
			/// vector for the (u,v) to (N,M) search
			int[] UpVector = new int[2 * MAX + 2];
			
			LCS(DataA, 0, DataA.Length, DataB, 0, DataB.Length, DownVector, UpVector);
			
			Optimize(DataA);
			Optimize(DataB);
			return CreateDiffs(DataA, DataB);
		}
Esempio n. 2
0
		/// <summary>
		/// If a sequence of modified lines starts with a line that contains the same content
		/// as the line that appends the changes, the difference sequence is modified so that the
		/// appended line and not the starting line is marked as modified.
		/// This leads to more readable diff sequences when comparing text files.
		/// </summary>
		/// <param name="Data">A Diff data buffer containing the identified changes.</param>
		private static void Optimize(DiffData Data) {
			int StartPos, EndPos;
			
			StartPos = 0;
			while (StartPos < Data.Length) {
				while ((StartPos < Data.Length) && (Data.modified[StartPos] == false))
					StartPos++;
				EndPos = StartPos;
				while ((EndPos < Data.Length) && (Data.modified[EndPos] == true))
					EndPos++;
				
				if ((EndPos < Data.Length) && (Data.data[StartPos] == Data.data[EndPos])) {
					Data.modified[StartPos] = false;
					Data.modified[EndPos] = true;
				} else {
					StartPos = EndPos;
				}
			}
		}
Esempio n. 3
0
		/// <summary>Scan the tables of which lines are inserted and deleted,
		/// producing an edit script in forward order.  
		/// </summary>
		/// dynamic array
		private static Item[] CreateDiffs (DiffData DataA, DiffData DataB) {
			ArrayList a = new ArrayList ();
			Item aItem;
			Item[] result;
			
			int StartA, StartB;
			int LineA, LineB;
			
			LineA = 0;
			LineB = 0;
			while (LineA < DataA.Length || LineB < DataB.Length) {
				if ((LineA < DataA.Length) && (!DataA.modified[LineA])
						&& (LineB < DataB.Length) && (!DataB.modified[LineB])) {
					// equal lines
					LineA++;
					LineB++;
					
				} else {
					// maybe deleted and/or inserted lines
					StartA = LineA;
					StartB = LineB;
					
					while (LineA < DataA.Length && (LineB >= DataB.Length || DataA.modified[LineA]))
						// while (LineA < DataA.Length && DataA.modified[LineA])
						LineA++;
					
					while (LineB < DataB.Length && (LineA >= DataA.Length || DataB.modified[LineB]))
						// while (LineB < DataB.Length && DataB.modified[LineB])
						LineB++;
					
					if ((StartA < LineA) || (StartB < LineB)) {
						// store a new difference-item
						aItem = new Item();
						aItem.StartA = StartA;
						aItem.StartB = StartB;
						aItem.deletedA = LineA - StartA;
						aItem.insertedB = LineB - StartB;
						a.Add(aItem);
					}
				}
			}
			
			result = new Item[a.Count];
			a.CopyTo(result);
			
			return (result);
		}
Esempio n. 4
0
		/// <summary>
		/// This is the divide-and-conquer implementation of the longes common-subsequence (LCS) 
		/// algorithm.
		/// The published algorithm passes recursively parts of the A and B sequences.
		/// To avoid copying these arrays the lower and upper bounds are passed while the sequences stay constant.
		/// </summary>
		/// <param name="DataA">sequence A</param>
		/// <param name="LowerA">lower bound of the actual range in DataA</param>
		/// <param name="UpperA">upper bound of the actual range in DataA (exclusive)</param>
		/// <param name="DataB">sequence B</param>
		/// <param name="LowerB">lower bound of the actual range in DataB</param>
		/// <param name="UpperB">upper bound of the actual range in DataB (exclusive)</param>
		/// <param name="DownVector">a vector for the (0,0) to (x,y) search. Passed as a parameter for speed reasons.</param>
		/// <param name="UpVector">a vector for the (u,v) to (N,M) search. Passed as a parameter for speed reasons.</param>
		private static void LCS(DiffData DataA, int LowerA, int UpperA, DiffData DataB, int LowerB, int UpperB, int[] DownVector, int[] UpVector) {
			// Fast walkthrough equal lines at the start
			while (LowerA < UpperA && LowerB < UpperB && DataA.data[LowerA] == DataB.data[LowerB]) {
				LowerA++; LowerB++;
			}
			
			// Fast walkthrough equal lines at the end
			while (LowerA < UpperA && LowerB < UpperB && DataA.data[UpperA - 1] == DataB.data[UpperB - 1]) {
				--UpperA; --UpperB;
			}
			
			if (LowerA == UpperA) {
				// mark as inserted lines.
				while (LowerB < UpperB)
					DataB.modified[LowerB++] = true;
				
			} else if (LowerB == UpperB) {
				// mark as deleted lines.
				while (LowerA < UpperA)
					DataA.modified[LowerA++] = true;
				
			} else {
				// Find the middle snakea and length of an optimal path for A and B
				SMSRD smsrd = SMS(DataA, LowerA, UpperA, DataB, LowerB, UpperB, DownVector, UpVector);
				
				LCS(DataA, LowerA, smsrd.x, DataB, LowerB, smsrd.y, DownVector, UpVector);
				LCS(DataA, smsrd.x, UpperA, DataB, smsrd.y, UpperB, DownVector, UpVector); 
			}
		}
Esempio n. 5
0
		/// <summary>
		/// This is the algorithm to find the Shortest Middle Snake (SMS).
		/// </summary>
		/// <param name="DataA">sequence A</param>
		/// <param name="LowerA">lower bound of the actual range in DataA</param>
		/// <param name="UpperA">upper bound of the actual range in DataA (exclusive)</param>
		/// <param name="DataB">sequence B</param>
		/// <param name="LowerB">lower bound of the actual range in DataB</param>
		/// <param name="UpperB">upper bound of the actual range in DataB (exclusive)</param>
		/// <param name="DownVector">a vector for the (0,0) to (x,y) search. Passed as a parameter for speed reasons.</param>
		/// <param name="UpVector">a vector for the (u,v) to (N,M) search. Passed as a parameter for speed reasons.</param>
		/// <returns>a MiddleSnakeData record containing x,y and u,v</returns>
		private static SMSRD SMS(DiffData DataA, int LowerA, int UpperA, DiffData DataB, int LowerB, int UpperB, int[] DownVector, int[] UpVector) {
			SMSRD ret;
			int MAX = DataA.Length + DataB.Length + 1;
			
			int DownK = LowerA - LowerB; // the k-line to start the forward search
			int UpK = UpperA - UpperB; // the k-line to start the reverse search
			
			int Delta = (UpperA - LowerA) - (UpperB - LowerB);
			bool oddDelta = (Delta & 1) != 0;
			
			// The vectors in the publication accepts negative indexes. the vectors implemented here are 0-based
			// and are access using a specific offset: UpOffset UpVector and DownOffset for DownVektor
			int DownOffset = MAX - DownK;
			int UpOffset = MAX - UpK;
			
			int MaxD = ((UpperA - LowerA + UpperB - LowerB) / 2) + 1;
			
			// init vectors
			DownVector[DownOffset + DownK + 1] = LowerA;
			UpVector[UpOffset + UpK - 1] = UpperA;
			
			for (int D = 0; D <= MaxD; D++) {
				// Extend the forward path.
				for (int k = DownK - D; k <= DownK + D; k += 2) {
					// find the only or better starting point
					int x, y;
					if (k == DownK - D) {
						x = DownVector[DownOffset + k + 1]; // down
					} else {
						x = DownVector[DownOffset + k - 1] + 1; // a step to the right
						if ((k < DownK + D) && (DownVector[DownOffset + k + 1] >= x))
							x = DownVector[DownOffset + k + 1]; // down
					}
					y = x - k;
					
					// find the end of the furthest reaching forward D-path in diagonal k.
					while ((x < UpperA) && (y < UpperB) && (DataA.data[x] == DataB.data[y])) {
						x++; y++;
					}
					DownVector[DownOffset + k] = x;
					
					// overlap ?
					if (oddDelta && (UpK - D < k) && (k < UpK + D)) {
						if (UpVector[UpOffset + k] <= DownVector[DownOffset + k]) {
							ret.x = DownVector[DownOffset + k];
							ret.y = DownVector[DownOffset + k] - k;
							
							return (ret);
						}
					}
				}
				
				// Extend the reverse path.
				for (int k = UpK - D; k <= UpK + D; k += 2) {
					// find the only or better starting point
					int x, y;
					if (k == UpK + D) {
						x = UpVector[UpOffset + k - 1];
					} else {
						x = UpVector[UpOffset + k + 1] - 1;
						if ((k > UpK - D) && (UpVector[UpOffset + k - 1] < x))
							x = UpVector[UpOffset + k - 1];
					}
					y = x - k;
					
					while ((x > LowerA) && (y > LowerB) && (DataA.data[x - 1] == DataB.data[y - 1])) {
						x--; y--; // diagonal
					}
					UpVector[UpOffset + k] = x;
					
					// overlap ?
					if (!oddDelta && (DownK - D <= k) && (k <= DownK + D)) {
						if (UpVector[UpOffset + k] <= DownVector[DownOffset + k]) {
							ret.x = DownVector[DownOffset + k];
							ret.y = DownVector[DownOffset + k] - k;
							
							return (ret);
						}
					}
				}
			}
			
			throw new Exception ("function return path");
		}
Esempio n. 6
0
		/// <summary>
		/// Find the difference in 2 arrays of integers.
		/// </summary>
		/// <param name="ArrayA">A-version of the numbers (usualy the old one)</param>
		/// <param name="ArrayB">B-version of the numbers (usualy the new one)</param>
		/// <returns>Returns a array of Items that describe the differences.</returns>
		public static Item[] DiffInt(int[] ArrayA, int[] ArrayB) {
			// The A-Version of the data (original data) to be compared.
			DiffData DataA = new DiffData(ArrayA);
			
			// The B-Version of the data (modified data) to be compared.
			DiffData DataB = new DiffData(ArrayB);
			
			int MAX = DataA.Length + DataB.Length + 1;
			/// vector for the (0,0) to (x,y) search
			int[] DownVector = new int[2 * MAX + 2];
			/// vector for the (u,v) to (N,M) search
			int[] UpVector = new int[2 * MAX + 2];
			
			LCS(DataA, 0, DataA.Length, DataB, 0, DataB.Length, DownVector, UpVector);
			return CreateDiffs(DataA, DataB);
		}