1 //===-- llvm/ADT/edit_distance.h - Array edit distance function --- C++ -*-===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file defines a Levenshtein distance function that works for any two
11 // sequences, with each element of each sequence being analogous to a character
12 // in a string.
13 //
14 //===----------------------------------------------------------------------===//
15 
16 #ifndef LLVM_ADT_EDIT_DISTANCE_H
17 #define LLVM_ADT_EDIT_DISTANCE_H
18 
19 #include "llvm/ADT/ArrayRef.h"
20 #include <algorithm>
21 #include <memory>
22 
23 namespace llvm {
24 
25 /// \brief Determine the edit distance between two sequences.
26 ///
27 /// \param FromArray the first sequence to compare.
28 ///
29 /// \param ToArray the second sequence to compare.
30 ///
31 /// \param AllowReplacements whether to allow element replacements (change one
32 /// element into another) as a single operation, rather than as two operations
33 /// (an insertion and a removal).
34 ///
35 /// \param MaxEditDistance If non-zero, the maximum edit distance that this
36 /// routine is allowed to compute. If the edit distance will exceed that
37 /// maximum, returns \c MaxEditDistance+1.
38 ///
39 /// \returns the minimum number of element insertions, removals, or (if
40 /// \p AllowReplacements is \c true) replacements needed to transform one of
41 /// the given sequences into the other. If zero, the sequences are identical.
42 template<typename T>
43 unsigned ComputeEditDistance(ArrayRef<T> FromArray, ArrayRef<T> ToArray,
44                              bool AllowReplacements = true,
45                              unsigned MaxEditDistance = 0) {
46   // The algorithm implemented below is the "classic"
47   // dynamic-programming algorithm for computing the Levenshtein
48   // distance, which is described here:
49   //
50   //   http://en.wikipedia.org/wiki/Levenshtein_distance
51   //
52   // Although the algorithm is typically described using an m x n
53   // array, only two rows are used at a time, so this implemenation
54   // just keeps two separate vectors for those two rows.
55   typename ArrayRef<T>::size_type m = FromArray.size();
56   typename ArrayRef<T>::size_type n = ToArray.size();
57 
58   const unsigned SmallBufferSize = 64;
59   unsigned SmallBuffer[SmallBufferSize];
60   std::unique_ptr<unsigned[]> Allocated;
61   unsigned *Previous = SmallBuffer;
62   if (2*(n + 1) > SmallBufferSize) {
63     Previous = new unsigned [2*(n+1)];
64     Allocated.reset(Previous);
65   }
66   unsigned *Current = Previous + (n + 1);
67 
68   for (unsigned i = 0; i <= n; ++i)
69     Previous[i] = i;
70 
71   for (typename ArrayRef<T>::size_type y = 1; y <= m; ++y) {
72     Current[0] = y;
73     unsigned BestThisRow = Current[0];
74 
75     for (typename ArrayRef<T>::size_type x = 1; x <= n; ++x) {
76       if (AllowReplacements) {
77         Current[x] = std::min(
78             Previous[x-1] + (FromArray[y-1] == ToArray[x-1] ? 0u : 1u),
79             std::min(Current[x-1], Previous[x])+1);
80       }
81       else {
82         if (FromArray[y-1] == ToArray[x-1]) Current[x] = Previous[x-1];
83         else Current[x] = std::min(Current[x-1], Previous[x]) + 1;
84       }
85       BestThisRow = std::min(BestThisRow, Current[x]);
86     }
87 
88     if (MaxEditDistance && BestThisRow > MaxEditDistance)
89       return MaxEditDistance + 1;
90 
91     unsigned *tmp = Current;
92     Current = Previous;
93     Previous = tmp;
94   }
95 
96   unsigned Result = Previous[n];
97   return Result;
98 }
99 
100 } // End llvm namespace
101 
102 #endif
103