1/* Package difflib is a partial port of Python difflib module.
2
3Original source: https://github.com/pmezard/go-difflib
4
5This file is trimmed to only the parts used by this repository.
6*/
7package difflib // import "gotest.tools/internal/difflib"
8
9func min(a, b int) int {
10 if a < b {
11 return a
12 }
13 return b
14}
15
16func max(a, b int) int {
17 if a > b {
18 return a
19 }
20 return b
21}
22
23type Match struct {
24 A int
25 B int
26 Size int
27}
28
29type OpCode struct {
30 Tag byte
31 I1 int
32 I2 int
33 J1 int
34 J2 int
35}
36
37// SequenceMatcher compares sequence of strings. The basic
38// algorithm predates, and is a little fancier than, an algorithm
39// published in the late 1980's by Ratcliff and Obershelp under the
40// hyperbolic name "gestalt pattern matching". The basic idea is to find
41// the longest contiguous matching subsequence that contains no "junk"
42// elements (R-O doesn't address junk). The same idea is then applied
43// recursively to the pieces of the sequences to the left and to the right
44// of the matching subsequence. This does not yield minimal edit
45// sequences, but does tend to yield matches that "look right" to people.
46//
47// SequenceMatcher tries to compute a "human-friendly diff" between two
48// sequences. Unlike e.g. UNIX(tm) diff, the fundamental notion is the
49// longest *contiguous* & junk-free matching subsequence. That's what
50// catches peoples' eyes. The Windows(tm) windiff has another interesting
51// notion, pairing up elements that appear uniquely in each sequence.
52// That, and the method here, appear to yield more intuitive difference
53// reports than does diff. This method appears to be the least vulnerable
54// to synching up on blocks of "junk lines", though (like blank lines in
55// ordinary text files, or maybe "<P>" lines in HTML files). That may be
56// because this is the only method of the 3 that has a *concept* of
57// "junk" <wink>.
58//
59// Timing: Basic R-O is cubic time worst case and quadratic time expected
60// case. SequenceMatcher is quadratic time for the worst case and has
61// expected-case behavior dependent in a complicated way on how many
62// elements the sequences have in common; best case time is linear.
63type SequenceMatcher struct {
64 a []string
65 b []string
66 b2j map[string][]int
67 IsJunk func(string) bool
68 autoJunk bool
69 bJunk map[string]struct{}
70 matchingBlocks []Match
71 fullBCount map[string]int
72 bPopular map[string]struct{}
73 opCodes []OpCode
74}
75
76func NewMatcher(a, b []string) *SequenceMatcher {
77 m := SequenceMatcher{autoJunk: true}
78 m.SetSeqs(a, b)
79 return &m
80}
81
82// Set two sequences to be compared.
83func (m *SequenceMatcher) SetSeqs(a, b []string) {
84 m.SetSeq1(a)
85 m.SetSeq2(b)
86}
87
88// Set the first sequence to be compared. The second sequence to be compared is
89// not changed.
90//
91// SequenceMatcher computes and caches detailed information about the second
92// sequence, so if you want to compare one sequence S against many sequences,
93// use .SetSeq2(s) once and call .SetSeq1(x) repeatedly for each of the other
94// sequences.
95//
96// See also SetSeqs() and SetSeq2().
97func (m *SequenceMatcher) SetSeq1(a []string) {
98 if &a == &m.a {
99 return
100 }
101 m.a = a
102 m.matchingBlocks = nil
103 m.opCodes = nil
104}
105
106// Set the second sequence to be compared. The first sequence to be compared is
107// not changed.
108func (m *SequenceMatcher) SetSeq2(b []string) {
109 if &b == &m.b {
110 return
111 }
112 m.b = b
113 m.matchingBlocks = nil
114 m.opCodes = nil
115 m.fullBCount = nil
116 m.chainB()
117}
118
119func (m *SequenceMatcher) chainB() {
120 // Populate line -> index mapping
121 b2j := map[string][]int{}
122 for i, s := range m.b {
123 indices := b2j[s]
124 indices = append(indices, i)
125 b2j[s] = indices
126 }
127
128 // Purge junk elements
129 m.bJunk = map[string]struct{}{}
130 if m.IsJunk != nil {
131 junk := m.bJunk
132 for s, _ := range b2j {
133 if m.IsJunk(s) {
134 junk[s] = struct{}{}
135 }
136 }
137 for s, _ := range junk {
138 delete(b2j, s)
139 }
140 }
141
142 // Purge remaining popular elements
143 popular := map[string]struct{}{}
144 n := len(m.b)
145 if m.autoJunk && n >= 200 {
146 ntest := n/100 + 1
147 for s, indices := range b2j {
148 if len(indices) > ntest {
149 popular[s] = struct{}{}
150 }
151 }
152 for s, _ := range popular {
153 delete(b2j, s)
154 }
155 }
156 m.bPopular = popular
157 m.b2j = b2j
158}
159
160func (m *SequenceMatcher) isBJunk(s string) bool {
161 _, ok := m.bJunk[s]
162 return ok
163}
164
165// Find longest matching block in a[alo:ahi] and b[blo:bhi].
166//
167// If IsJunk is not defined:
168//
169// Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
170// alo <= i <= i+k <= ahi
171// blo <= j <= j+k <= bhi
172// and for all (i',j',k') meeting those conditions,
173// k >= k'
174// i <= i'
175// and if i == i', j <= j'
176//
177// In other words, of all maximal matching blocks, return one that
178// starts earliest in a, and of all those maximal matching blocks that
179// start earliest in a, return the one that starts earliest in b.
180//
181// If IsJunk is defined, first the longest matching block is
182// determined as above, but with the additional restriction that no
183// junk element appears in the block. Then that block is extended as
184// far as possible by matching (only) junk elements on both sides. So
185// the resulting block never matches on junk except as identical junk
186// happens to be adjacent to an "interesting" match.
187//
188// If no blocks match, return (alo, blo, 0).
189func (m *SequenceMatcher) findLongestMatch(alo, ahi, blo, bhi int) Match {
190 // CAUTION: stripping common prefix or suffix would be incorrect.
191 // E.g.,
192 // ab
193 // acab
194 // Longest matching block is "ab", but if common prefix is
195 // stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
196 // strip, so ends up claiming that ab is changed to acab by
197 // inserting "ca" in the middle. That's minimal but unintuitive:
198 // "it's obvious" that someone inserted "ac" at the front.
199 // Windiff ends up at the same place as diff, but by pairing up
200 // the unique 'b's and then matching the first two 'a's.
201 besti, bestj, bestsize := alo, blo, 0
202
203 // find longest junk-free match
204 // during an iteration of the loop, j2len[j] = length of longest
205 // junk-free match ending with a[i-1] and b[j]
206 j2len := map[int]int{}
207 for i := alo; i != ahi; i++ {
208 // look at all instances of a[i] in b; note that because
209 // b2j has no junk keys, the loop is skipped if a[i] is junk
210 newj2len := map[int]int{}
211 for _, j := range m.b2j[m.a[i]] {
212 // a[i] matches b[j]
213 if j < blo {
214 continue
215 }
216 if j >= bhi {
217 break
218 }
219 k := j2len[j-1] + 1
220 newj2len[j] = k
221 if k > bestsize {
222 besti, bestj, bestsize = i-k+1, j-k+1, k
223 }
224 }
225 j2len = newj2len
226 }
227
228 // Extend the best by non-junk elements on each end. In particular,
229 // "popular" non-junk elements aren't in b2j, which greatly speeds
230 // the inner loop above, but also means "the best" match so far
231 // doesn't contain any junk *or* popular non-junk elements.
232 for besti > alo && bestj > blo && !m.isBJunk(m.b[bestj-1]) &&
233 m.a[besti-1] == m.b[bestj-1] {
234 besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
235 }
236 for besti+bestsize < ahi && bestj+bestsize < bhi &&
237 !m.isBJunk(m.b[bestj+bestsize]) &&
238 m.a[besti+bestsize] == m.b[bestj+bestsize] {
239 bestsize += 1
240 }
241
242 // Now that we have a wholly interesting match (albeit possibly
243 // empty!), we may as well suck up the matching junk on each
244 // side of it too. Can't think of a good reason not to, and it
245 // saves post-processing the (possibly considerable) expense of
246 // figuring out what to do with it. In the case of an empty
247 // interesting match, this is clearly the right thing to do,
248 // because no other kind of match is possible in the regions.
249 for besti > alo && bestj > blo && m.isBJunk(m.b[bestj-1]) &&
250 m.a[besti-1] == m.b[bestj-1] {
251 besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
252 }
253 for besti+bestsize < ahi && bestj+bestsize < bhi &&
254 m.isBJunk(m.b[bestj+bestsize]) &&
255 m.a[besti+bestsize] == m.b[bestj+bestsize] {
256 bestsize += 1
257 }
258
259 return Match{A: besti, B: bestj, Size: bestsize}
260}
261
262// Return list of triples describing matching subsequences.
263//
264// Each triple is of the form (i, j, n), and means that
265// a[i:i+n] == b[j:j+n]. The triples are monotonically increasing in
266// i and in j. It's also guaranteed that if (i, j, n) and (i', j', n') are
267// adjacent triples in the list, and the second is not the last triple in the
268// list, then i+n != i' or j+n != j'. IOW, adjacent triples never describe
269// adjacent equal blocks.
270//
271// The last triple is a dummy, (len(a), len(b), 0), and is the only
272// triple with n==0.
273func (m *SequenceMatcher) GetMatchingBlocks() []Match {
274 if m.matchingBlocks != nil {
275 return m.matchingBlocks
276 }
277
278 var matchBlocks func(alo, ahi, blo, bhi int, matched []Match) []Match
279 matchBlocks = func(alo, ahi, blo, bhi int, matched []Match) []Match {
280 match := m.findLongestMatch(alo, ahi, blo, bhi)
281 i, j, k := match.A, match.B, match.Size
282 if match.Size > 0 {
283 if alo < i && blo < j {
284 matched = matchBlocks(alo, i, blo, j, matched)
285 }
286 matched = append(matched, match)
287 if i+k < ahi && j+k < bhi {
288 matched = matchBlocks(i+k, ahi, j+k, bhi, matched)
289 }
290 }
291 return matched
292 }
293 matched := matchBlocks(0, len(m.a), 0, len(m.b), nil)
294
295 // It's possible that we have adjacent equal blocks in the
296 // matching_blocks list now.
297 nonAdjacent := []Match{}
298 i1, j1, k1 := 0, 0, 0
299 for _, b := range matched {
300 // Is this block adjacent to i1, j1, k1?
301 i2, j2, k2 := b.A, b.B, b.Size
302 if i1+k1 == i2 && j1+k1 == j2 {
303 // Yes, so collapse them -- this just increases the length of
304 // the first block by the length of the second, and the first
305 // block so lengthened remains the block to compare against.
306 k1 += k2
307 } else {
308 // Not adjacent. Remember the first block (k1==0 means it's
309 // the dummy we started with), and make the second block the
310 // new block to compare against.
311 if k1 > 0 {
312 nonAdjacent = append(nonAdjacent, Match{i1, j1, k1})
313 }
314 i1, j1, k1 = i2, j2, k2
315 }
316 }
317 if k1 > 0 {
318 nonAdjacent = append(nonAdjacent, Match{i1, j1, k1})
319 }
320
321 nonAdjacent = append(nonAdjacent, Match{len(m.a), len(m.b), 0})
322 m.matchingBlocks = nonAdjacent
323 return m.matchingBlocks
324}
325
326// Return list of 5-tuples describing how to turn a into b.
327//
328// Each tuple is of the form (tag, i1, i2, j1, j2). The first tuple
329// has i1 == j1 == 0, and remaining tuples have i1 == the i2 from the
330// tuple preceding it, and likewise for j1 == the previous j2.
331//
332// The tags are characters, with these meanings:
333//
334// 'r' (replace): a[i1:i2] should be replaced by b[j1:j2]
335//
336// 'd' (delete): a[i1:i2] should be deleted, j1==j2 in this case.
337//
338// 'i' (insert): b[j1:j2] should be inserted at a[i1:i1], i1==i2 in this case.
339//
340// 'e' (equal): a[i1:i2] == b[j1:j2]
341func (m *SequenceMatcher) GetOpCodes() []OpCode {
342 if m.opCodes != nil {
343 return m.opCodes
344 }
345 i, j := 0, 0
346 matching := m.GetMatchingBlocks()
347 opCodes := make([]OpCode, 0, len(matching))
348 for _, m := range matching {
349 // invariant: we've pumped out correct diffs to change
350 // a[:i] into b[:j], and the next matching block is
351 // a[ai:ai+size] == b[bj:bj+size]. So we need to pump
352 // out a diff to change a[i:ai] into b[j:bj], pump out
353 // the matching block, and move (i,j) beyond the match
354 ai, bj, size := m.A, m.B, m.Size
355 tag := byte(0)
356 if i < ai && j < bj {
357 tag = 'r'
358 } else if i < ai {
359 tag = 'd'
360 } else if j < bj {
361 tag = 'i'
362 }
363 if tag > 0 {
364 opCodes = append(opCodes, OpCode{tag, i, ai, j, bj})
365 }
366 i, j = ai+size, bj+size
367 // the list of matching blocks is terminated by a
368 // sentinel with size 0
369 if size > 0 {
370 opCodes = append(opCodes, OpCode{'e', ai, i, bj, j})
371 }
372 }
373 m.opCodes = opCodes
374 return m.opCodes
375}
376
377// Isolate change clusters by eliminating ranges with no changes.
378//
379// Return a generator of groups with up to n lines of context.
380// Each group is in the same format as returned by GetOpCodes().
381func (m *SequenceMatcher) GetGroupedOpCodes(n int) [][]OpCode {
382 if n < 0 {
383 n = 3
384 }
385 codes := m.GetOpCodes()
386 if len(codes) == 0 {
387 codes = []OpCode{OpCode{'e', 0, 1, 0, 1}}
388 }
389 // Fixup leading and trailing groups if they show no changes.
390 if codes[0].Tag == 'e' {
391 c := codes[0]
392 i1, i2, j1, j2 := c.I1, c.I2, c.J1, c.J2
393 codes[0] = OpCode{c.Tag, max(i1, i2-n), i2, max(j1, j2-n), j2}
394 }
395 if codes[len(codes)-1].Tag == 'e' {
396 c := codes[len(codes)-1]
397 i1, i2, j1, j2 := c.I1, c.I2, c.J1, c.J2
398 codes[len(codes)-1] = OpCode{c.Tag, i1, min(i2, i1+n), j1, min(j2, j1+n)}
399 }
400 nn := n + n
401 groups := [][]OpCode{}
402 group := []OpCode{}
403 for _, c := range codes {
404 i1, i2, j1, j2 := c.I1, c.I2, c.J1, c.J2
405 // End the current group and start a new one whenever
406 // there is a large range with no changes.
407 if c.Tag == 'e' && i2-i1 > nn {
408 group = append(group, OpCode{c.Tag, i1, min(i2, i1+n),
409 j1, min(j2, j1+n)})
410 groups = append(groups, group)
411 group = []OpCode{}
412 i1, j1 = max(i1, i2-n), max(j1, j2-n)
413 }
414 group = append(group, OpCode{c.Tag, i1, i2, j1, j2})
415 }
416 if len(group) > 0 && !(len(group) == 1 && group[0].Tag == 'e') {
417 groups = append(groups, group)
418 }
419 return groups
420}