The diffing algorithm is not accurate about google-diff-match-patch HOT 2 CLOSED

Incidentally, I did play around with the parameters at the beginning of the 
file;
aside from causing the differencing operation to take much longer, they had no
significant change to the results.  The final settings I used were:

  // Number of seconds to map a diff before giving up.  (0 for infinity)
  public float Diff_Timeout = 40.0f;
  // Cost of an empty edit operation in terms of edit characters.
  public short Diff_EditCost = 4;
  // The size beyond which the double-ended diff activates.
  // Double-ending is twice as fast, but less accurate.
  public short Diff_DualThreshold = 32;
  // Tweak the relative importance (0.0 = accuracy, 1.0 = proximity)
  public float Match_Balance = 0.0f;
  // At what point is no match declared (0.0 = perfection, 1.0 = very loose)
  public float Match_Threshold = 0.0f;
  // The min and max cutoffs used when computing text lengths.
  public int Match_MinLength = 10;
  public int Match_MaxLength = 1000;
  // Chunk size for context length.
  public short Patch_Margin = 4;

At this point, the differencing operation (diff_main) took much longer than the 
GNU
diff algorithm in Java (http://www.bmsi.com/java/#diff); the latter produces 
results
similar to GNU diff in 26 seconds; the former produced the usual, inaccurate, 
"the
whole file changed" results in 41 seconds.

The accuracy of the diff is not in question (it's simply reaching the timeout 
and correctly returning a non-optimal solution).  Merging this bug with the 
performance bug since that's the real issue.