Copyright (c) 2000, 2014 IBM Corporation and others. This program and the accompanying materials are made available under the terms of the Eclipse Public License 2.0 which accompanies this distribution, and is available at https://www.eclipse.org/legal/epl-2.0/ SPDX-License-Identifier: EPL-2.0 Contributors: IBM Corporation - initial API and implementation
/******************************************************************************* * Copyright (c) 2000, 2014 IBM Corporation and others. * * This program and the accompanying materials * are made available under the terms of the Eclipse Public License 2.0 * which accompanies this distribution, and is available at * https://www.eclipse.org/legal/epl-2.0/ * * SPDX-License-Identifier: EPL-2.0 * * Contributors: * IBM Corporation - initial API and implementation *******************************************************************************/
package org.eclipse.compare.internal.core; import org.eclipse.core.runtime.OperationCanceledException; import org.eclipse.core.runtime.SubMonitor; /* Used to determine the change set responsible for each line */ public abstract class LCS { public static final double TOO_LONG = 100000000.0; // 10^8, the value of N*M when // to start binding the // run time private static final double POW_LIMIT = 1.5; // limit the time to // D^POW_LIMIT private int max_differences; // the maximum number of differences from // each end to consider private int length;
Myers' algorithm for longest common subsequence. O((M + N)D) worst case time, O(M + N + D^2) expected time, O(M + N) space (http://citeseer.ist.psu.edu/myers86ond.html) Note: Beyond implementing the algorithm as described in the paper I have added diagonal range compression which helps when finding the LCS of a very long and a very short sequence, also bound the running time to (N + M)^1.5 when both sequences are very long. After this method is called, the longest common subsequence is available by calling getResult() where result[0] is composed of entries from l1 and result[1] is composed of entries from l2
Params:
  • subMonitor –
/** * Myers' algorithm for longest common subsequence. O((M + N)D) worst case * time, O(M + N + D^2) expected time, O(M + N) space * (http://citeseer.ist.psu.edu/myers86ond.html) * * Note: Beyond implementing the algorithm as described in the paper I have * added diagonal range compression which helps when finding the LCS of a * very long and a very short sequence, also bound the running time to (N + * M)^1.5 when both sequences are very long. * * After this method is called, the longest common subsequence is available * by calling getResult() where result[0] is composed of * entries from l1 and result[1] is composed of entries from l2 * @param subMonitor */
public void longestCommonSubsequence(SubMonitor subMonitor) { int length1 = getLength1(); int length2 = getLength2(); if (length1 == 0 || length2 == 0) { this.length = 0; return; } this.max_differences = (length1 + length2 + 1) / 2; // ceil((N+M)/2) if (!isCappingDisabled() && (double) length1 * (double) length2 > TOO_LONG) { // limit complexity to D^POW_LIMIT for long sequences this.max_differences = (int) Math.pow(this.max_differences, POW_LIMIT - 1.0); } initializeLcs(length1); subMonitor.beginTask(null, length1); /* * The common prefixes and suffixes are always part of some LCS, include * them now to reduce our search space */ int forwardBound; int max = Math.min(length1, length2); for (forwardBound = 0; forwardBound < max && isRangeEqual(forwardBound, forwardBound); forwardBound++) { setLcs(forwardBound, forwardBound); worked(subMonitor, 1); } int backBoundL1 = length1 - 1; int backBoundL2 = length2 - 1; while (backBoundL1 >= forwardBound && backBoundL2 >= forwardBound && isRangeEqual(backBoundL1, backBoundL2)) { setLcs(backBoundL1, backBoundL2); backBoundL1--; backBoundL2--; worked(subMonitor, 1); } this.length = forwardBound + length1 - backBoundL1 - 1 + lcs_rec(forwardBound, backBoundL1, forwardBound, backBoundL2, new int[2][length1 + length2 + 1], new int[3], subMonitor); } private boolean isCappingDisabled() { return ComparePlugin.getDefault().isCappingDisabled(); }
The recursive helper function for Myers' LCS. Computes the LCS of l1[bottoml1 .. topl1] and l2[bottoml2 .. topl2] fills in the appropriate location in lcs and returns the length
Params:
  • l1 – The 1st sequence
  • bottoml1 – Index in the 1st sequence to start from (inclusive)
  • topl1 – Index in the 1st sequence to end on (inclusive)
  • l2 – The 2nd sequence
  • bottoml2 – Index in the 2nd sequence to start from (inclusive)
  • topl2 – Index in the 2nd sequence to end on (inclusive)
  • V – should be allocated as int[2][l1.length + l2.length + 1], used to store furthest reaching D-paths
  • snake – should be allocated as int[3], used to store the beginning x, y coordinates and the length of the latest snake traversed
  • subMonitor –
  • lcs – should be allocated as TextLine[2][l1.length], used to store the common points found to be part of the LCS where lcs[0] references lines of l1 and lcs[1] references lines of l2.
Returns:the length of the LCS
/** * The recursive helper function for Myers' LCS. Computes the LCS of * l1[bottoml1 .. topl1] and l2[bottoml2 .. topl2] fills in the appropriate * location in lcs and returns the length * * @param l1 The 1st sequence * @param bottoml1 Index in the 1st sequence to start from (inclusive) * @param topl1 Index in the 1st sequence to end on (inclusive) * @param l2 The 2nd sequence * @param bottoml2 Index in the 2nd sequence to start from (inclusive) * @param topl2 Index in the 2nd sequence to end on (inclusive) * @param V should be allocated as int[2][l1.length + l2.length + 1], used * to store furthest reaching D-paths * @param snake should be allocated as int[3], used to store the beginning * x, y coordinates and the length of the latest snake traversed * @param subMonitor * @param lcs should be allocated as TextLine[2][l1.length], used to store * the common points found to be part of the LCS where lcs[0] * references lines of l1 and lcs[1] references lines of l2. * * @return the length of the LCS */
private int lcs_rec( int bottoml1, int topl1, int bottoml2, int topl2, int[][] V, int[] snake, SubMonitor subMonitor) { // check that both sequences are non-empty if (bottoml1 > topl1 || bottoml2 > topl2) { return 0; } int d = find_middle_snake(bottoml1, topl1, bottoml2, topl2, V, snake, subMonitor); // System.out.println(snake[0] + " " + snake[1] + " " + snake[2]); // need to store these so we don't lose them when they're overwritten by // the recursion int len = snake[2]; int startx = snake[0]; int starty = snake[1]; // the middle snake is part of the LCS, store it for (int i = 0; i < len; i++) { setLcs(startx + i, starty + i); worked(subMonitor, 1); } if (d > 1) { return len + lcs_rec(bottoml1, startx - 1, bottoml2, starty - 1, V, snake, subMonitor) + lcs_rec(startx + len, topl1, starty + len, topl2, V, snake, subMonitor); } else if (d == 1) { /* * In this case the sequences differ by exactly 1 line. We have * already saved all the lines after the difference in the for loop * above, now we need to save all the lines before the difference. */ int max = Math.min(startx - bottoml1, starty - bottoml2); for (int i = 0; i < max; i++) { setLcs(bottoml1 + i, bottoml2 + i); worked(subMonitor, 1); } return max + len; } return len; } private void worked(SubMonitor subMonitor, int work) { if (subMonitor.isCanceled()) throw new OperationCanceledException(); subMonitor.worked(work); }
Helper function for Myers' LCS algorithm to find the middle snake for l1[bottoml1..topl1] and l2[bottoml2..topl2] The x, y coordinates of the start of the middle snake are saved in snake[0], snake[1] respectively and the length of the snake is saved in s[2].
Params:
  • l1 – The 1st sequence
  • bottoml1 – Index in the 1st sequence to start from (inclusive)
  • topl1 – Index in the 1st sequence to end on (inclusive)
  • l2 – The 2nd sequence
  • bottoml2 – Index in the 2nd sequence to start from (inclusive)
  • topl2 – Index in the 2nd sequence to end on (inclusive)
  • V – should be allocated as int[2][l1.length + l2.length + 1], used to store furthest reaching D-paths
  • snake – should be allocated as int[3], used to store the beginning x, y coordinates and the length of the middle snake
  • subMonitor –
Returns:The number of differences (SES) between l1[bottoml1..topl1] and l2[bottoml2..topl2]
/** * Helper function for Myers' LCS algorithm to find the middle snake for * l1[bottoml1..topl1] and l2[bottoml2..topl2] The x, y coordinates of the * start of the middle snake are saved in snake[0], snake[1] respectively * and the length of the snake is saved in s[2]. * * @param l1 The 1st sequence * @param bottoml1 Index in the 1st sequence to start from (inclusive) * @param topl1 Index in the 1st sequence to end on (inclusive) * @param l2 The 2nd sequence * @param bottoml2 Index in the 2nd sequence to start from (inclusive) * @param topl2 Index in the 2nd sequence to end on (inclusive) * @param V should be allocated as int[2][l1.length + l2.length + 1], used * to store furthest reaching D-paths * @param snake should be allocated as int[3], used to store the beginning * x, y coordinates and the length of the middle snake * @param subMonitor * * @return The number of differences (SES) between l1[bottoml1..topl1] and * l2[bottoml2..topl2] */
private int find_middle_snake( int bottoml1, int topl1, int bottoml2, int topl2, int[][] V, int[] snake, SubMonitor subMonitor) { int N = topl1 - bottoml1 + 1; int M = topl2 - bottoml2 + 1; // System.out.println("N: " + N + " M: " + M + " bottom: " + bottoml1 + // ", " + // bottoml2 + " top: " + topl1 + ", " + topl2); int delta = N - M; boolean isEven; if ((delta & 1) == 1) { isEven = false; } else { isEven = true; } int limit = Math.min(this.max_differences, (N + M + 1) / 2); // ceil((N+M)/2) int value_to_add_forward; // a 0 or 1 that we add to the start offset // to make it odd/even if ((M & 1) == 1) { value_to_add_forward = 1; } else { value_to_add_forward = 0; } int value_to_add_backward; if ((N & 1) == 1) { value_to_add_backward = 1; } else { value_to_add_backward = 0; } int start_forward = -M; int end_forward = N; int start_backward = -N; int end_backward = M; V[0][limit + 1] = 0; V[1][limit - 1] = N; for (int d = 0; d <= limit; d++) { int start_diag = Math.max(value_to_add_forward + start_forward, -d); int end_diag = Math.min(end_forward, d); value_to_add_forward = 1 - value_to_add_forward; // compute forward furthest reaching paths for (int k = start_diag; k <= end_diag; k += 2) { int x; if (k == -d || (k < d && V[0][limit + k - 1] < V[0][limit + k + 1])) { x = V[0][limit + k + 1]; } else { x = V[0][limit + k - 1] + 1; } int y = x - k; snake[0] = x + bottoml1; snake[1] = y + bottoml2; snake[2] = 0; // System.out.println("1 x: " + x + " y: " + y + " k: " + k + " // d: " + d ); while (x < N && y < M && isRangeEqual(x + bottoml1, y + bottoml2)) { x++; y++; snake[2]++; } V[0][limit + k] = x; // System.out.println(x + " " + V[1][limit+k -delta] + " " + k + // " " + delta); if (!isEven && k >= delta - d + 1 && k <= delta + d - 1 && x >= V[1][limit + k - delta]) { // System.out.println("Returning: " + (2*d-1)); return 2 * d - 1; } // check to see if we can cut down the diagonal range if (x >= N && end_forward > k - 1) { end_forward = k - 1; } else if (y >= M) { start_forward = k + 1; value_to_add_forward = 0; } } start_diag = Math.max(value_to_add_backward + start_backward, -d); end_diag = Math.min(end_backward, d); value_to_add_backward = 1 - value_to_add_backward; // compute backward furthest reaching paths for (int k = start_diag; k <= end_diag; k += 2) { int x; if (k == d || (k != -d && V[1][limit + k - 1] < V[1][limit + k + 1])) { x = V[1][limit + k - 1]; } else { x = V[1][limit + k + 1] - 1; } int y = x - k - delta; snake[2] = 0; // System.out.println("2 x: " + x + " y: " + y + " k: " + k + " // d: " + d); while (x > 0 && y > 0 && isRangeEqual(x - 1 + bottoml1, y - 1 + bottoml2)) { x--; y--; snake[2]++; } V[1][limit + k] = x; if (isEven && k >= -delta - d && k <= d - delta && x <= V[0][limit + k + delta]) { // System.out.println("Returning: " + 2*d); snake[0] = bottoml1 + x; snake[1] = bottoml2 + y; return 2 * d; } // check to see if we can cut down our diagonal range if (x <= 0) { start_backward = k + 1; value_to_add_backward = 0; } else if (y <= 0 && end_backward > k - 1) { end_backward = k - 1; } } worked(subMonitor, 1); } /* * computing the true LCS is too expensive, instead find the diagonal * with the most progress and pretend a midle snake of length 0 occurs * there. */ int[] most_progress = findMostProgress(M, N, limit, V); snake[0] = bottoml1 + most_progress[0]; snake[1] = bottoml2 + most_progress[1]; snake[2] = 0; return 5; /* * HACK: since we didn't really finish the LCS computation * we don't really know the length of the SES. We don't do * anything with the result anyway, unless it's <=1. We know * for a fact SES > 1 so 5 is as good a number as any to * return here */ }
Takes the array with furthest reaching D-paths from an LCS computation and returns the x,y coordinates and progress made in the middle diagonal among those with maximum progress, both from the front and from the back.
Params:
  • M – the length of the 1st sequence for which LCS is being computed
  • N – the length of the 2nd sequence for which LCS is being computed
  • limit – the number of steps made in an attempt to find the LCS from the front and back
  • V – the array storing the furthest reaching D-paths for the LCS computation
Returns:The result as an array of 3 integers where result[0] is the x coordinate of the current location in the diagonal with the most progress, result[1] is the y coordinate of the current location in the diagonal with the most progress and result[2] is the amount of progress made in that diagonal
/** * Takes the array with furthest reaching D-paths from an LCS computation * and returns the x,y coordinates and progress made in the middle diagonal * among those with maximum progress, both from the front and from the back. * * @param M the length of the 1st sequence for which LCS is being computed * @param N the length of the 2nd sequence for which LCS is being computed * @param limit the number of steps made in an attempt to find the LCS from * the front and back * @param V the array storing the furthest reaching D-paths for the LCS * computation * @return The result as an array of 3 integers where result[0] is the x * coordinate of the current location in the diagonal with the most * progress, result[1] is the y coordinate of the current location * in the diagonal with the most progress and result[2] is the * amount of progress made in that diagonal */
private static int[] findMostProgress(int M, int N, int limit, int[][] V) { int delta = N - M; int forward_start_diag; if ((M & 1) == (limit & 1)) { forward_start_diag = Math.max(-M, -limit); } else { forward_start_diag = Math.max(1 - M, -limit); } int forward_end_diag = Math.min(N, limit); int backward_start_diag; if ((N & 1) == (limit & 1)) { backward_start_diag = Math.max(-N, -limit); } else { backward_start_diag = Math.max(1 - N, -limit); } int backward_end_diag = Math.min(M, limit); int[][] max_progress = new int[Math.max(forward_end_diag - forward_start_diag, backward_end_diag - backward_start_diag) / 2 + 1][3]; int num_progress = 0; // the 1st entry is current, it is initialized // with 0s // first search the forward diagonals for (int k = forward_start_diag; k <= forward_end_diag; k += 2) { int x = V[0][limit + k]; int y = x - k; if (x > N || y > M) { continue; } int progress = x + y; if (progress > max_progress[0][2]) { num_progress = 0; max_progress[0][0] = x; max_progress[0][1] = y; max_progress[0][2] = progress; } else if (progress == max_progress[0][2]) { num_progress++; max_progress[num_progress][0] = x; max_progress[num_progress][1] = y; max_progress[num_progress][2] = progress; } } boolean max_progress_forward = true; // initially the maximum // progress is in the forward // direction // now search the backward diagonals for (int k = backward_start_diag; k <= backward_end_diag; k += 2) { int x = V[1][limit + k]; int y = x - k - delta; if (x < 0 || y < 0) { continue; } int progress = N - x + M - y; if (progress > max_progress[0][2]) { num_progress = 0; max_progress_forward = false; max_progress[0][0] = x; max_progress[0][1] = y; max_progress[0][2] = progress; } else if (progress == max_progress[0][2] && !max_progress_forward) { num_progress++; max_progress[num_progress][0] = x; max_progress[num_progress][1] = y; max_progress[num_progress][2] = progress; } } // return the middle diagonal with maximal progress. return max_progress[num_progress / 2]; } protected abstract int getLength2(); protected abstract int getLength1(); protected abstract boolean isRangeEqual(int i1, int i2); protected abstract void setLcs(int sl1, int sl2); protected abstract void initializeLcs(int lcsLength); public int getLength() { return this.length; } }