The algorithm is optimised for situations where the input sequences have few repeated objects. If it is given input with many repeated objects it will report sub-optimal changes. However, given appropriate input, it is fast, and linear in memory usage.
The algorithm consists of the following steps:
The first step is to compute a an equivalence set for the input data. The equivalence set is computed from objects that are in the original input sequence
eq(x) = the index of the first occurence of x in the original sequence.
With this equivalence function, the algorithm can compare integers rather than strings, which is considerably more efficient.
The second step is to compute the datastructure on which the algorithm will operate. Having computed the equivalence function in the previous step, we can compute two arrays where indx[i] = eqs(orig[i]) and jndx[i] = eqs(rev[i]). The algorithm can now operate on indx and jndx instead of orig and rev. Thus, comparisons are then on O(int == int) instead of O(Object.equals(Object)).
The algorithm now matches indx and jndx. Whilst indx[i] == jndx[i] it skips matching objects in the sequence. In seeking to match objects in the input sequence it assumes that each object is likely to be unique. It uses the known characteristics of the unique equivalence function. It can tell from the eq value if this object appeared in the other sequence at all. If it did not, there is no point in searching for a match.
Recall that the eq function value is the index earliest occurrence in the orig sequence. This information is used to search efficiently for the next match. The algorithm is perfect when all input objects are unique, but degrades when input objects are not unique. When input objects are not unique an optimal match may not be found, but a correct match will be.
Having identified common matching objects in the orig and revised sequences, the differences between them are easily computed.
@see DeltaImpl @see RevisionImpl Modifications: 27/Apr/2003 bwm Added some comments whilsttrying to figure out the algorithm 03 May 2003 bwm Created this implementation class by refactoring it out of the Diff class to enable plug in difference algorithms