Practical Fast Searching in Strings

The problem is that of searching a large block of text to find the first occurrence of a substring (which we will call the ‘pattern’). This particular operation is provided in most text editing systems and it also has applications in bibliographic retrieval systems. Since the text to be searched can be overwhelmingly large — perhaps hundreds of thousands of characters — it is important to use efficient techniques. Simple programs for searching text typically require a worst-case running time of O(mn), where m is the length of the pattern and n is the length of the text. However, Knuth et al.5 showed that this time can be reduced to O(n) with a fairly complicated algorithm. Later, Boyer and Moore published a practical and simpler algorithm1 that also has this linear worst case running time. In the average case, only a small fraction of the n characters are actually inspected. (A recent paper by Galil3 reports some improvements to the Boyer and Moore algorithm for its worst case behaviour.) Many programmers may not believe that the Boyer and Moore algorithm (if they have heard of it) is a truly practical approach. It is the purpose of this paper to demonstrate that it is and to show the circumstances under which it should be employed. Many computers, particularly. the larger machines, possess instructions to search for individual characters within main memory. One might think that these instructions would permit codings of routines that could beat the Boyer and Moore algorithm. However, we will experimentally show that this is not always the case — even when the search instructions are used in the most efficient manner imaginable.