]>
Commit | Line | Data |
---|---|---|
3443546f LT |
1 | /* |
2 | * LibXDiff by Davide Libenzi ( File Differential Library ) | |
3 | * Copyright (C) 2003 Davide Libenzi | |
4 | * | |
5 | * This library is free software; you can redistribute it and/or | |
6 | * modify it under the terms of the GNU Lesser General Public | |
7 | * License as published by the Free Software Foundation; either | |
8 | * version 2.1 of the License, or (at your option) any later version. | |
9 | * | |
10 | * This library is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 | * Lesser General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU Lesser General Public | |
16 | * License along with this library; if not, write to the Free Software | |
17 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
18 | * | |
19 | * Davide Libenzi <davidel@xmailserver.org> | |
20 | * | |
21 | */ | |
22 | ||
23 | #include "xinclude.h" | |
24 | ||
25 | ||
26 | ||
27 | #define XDL_MAX_COST_MIN 256 | |
28 | #define XDL_HEUR_MIN_COST 256 | |
f630cfda | 29 | #define XDL_LINE_MAX (long)((1UL << (CHAR_BIT * sizeof(long) - 1)) - 1) |
3443546f LT |
30 | #define XDL_SNAKE_CNT 20 |
31 | #define XDL_K_HEUR 4 | |
32 | ||
33 | ||
34 | ||
35 | typedef struct s_xdpsplit { | |
36 | long i1, i2; | |
37 | int min_lo, min_hi; | |
38 | } xdpsplit_t; | |
39 | ||
40 | ||
41 | ||
42 | ||
43 | static long xdl_split(unsigned long const *ha1, long off1, long lim1, | |
44 | unsigned long const *ha2, long off2, long lim2, | |
45 | long *kvdf, long *kvdb, int need_min, xdpsplit_t *spl, | |
46 | xdalgoenv_t *xenv); | |
47 | static xdchange_t *xdl_add_change(xdchange_t *xscr, long i1, long i2, long chg1, long chg2); | |
295ba2fb | 48 | |
3443546f LT |
49 | |
50 | ||
51 | ||
52 | ||
53 | /* | |
54 | * See "An O(ND) Difference Algorithm and its Variations", by Eugene Myers. | |
55 | * Basically considers a "box" (off1, off2, lim1, lim2) and scan from both | |
56 | * the forward diagonal starting from (off1, off2) and the backward diagonal | |
57 | * starting from (lim1, lim2). If the K values on the same diagonal crosses | |
58 | * returns the furthest point of reach. We might end up having to expensive | |
59 | * cases using this algorithm is full, so a little bit of heuristic is needed | |
60 | * to cut the search and to return a suboptimal point. | |
61 | */ | |
62 | static long xdl_split(unsigned long const *ha1, long off1, long lim1, | |
63 | unsigned long const *ha2, long off2, long lim2, | |
64 | long *kvdf, long *kvdb, int need_min, xdpsplit_t *spl, | |
65 | xdalgoenv_t *xenv) { | |
66 | long dmin = off1 - lim2, dmax = lim1 - off2; | |
67 | long fmid = off1 - off2, bmid = lim1 - lim2; | |
68 | long odd = (fmid - bmid) & 1; | |
69 | long fmin = fmid, fmax = fmid; | |
70 | long bmin = bmid, bmax = bmid; | |
71 | long ec, d, i1, i2, prev1, best, dd, v, k; | |
72 | ||
73 | /* | |
74 | * Set initial diagonal values for both forward and backward path. | |
75 | */ | |
76 | kvdf[fmid] = off1; | |
77 | kvdb[bmid] = lim1; | |
78 | ||
79 | for (ec = 1;; ec++) { | |
80 | int got_snake = 0; | |
81 | ||
82 | /* | |
83 | * We need to extent the diagonal "domain" by one. If the next | |
84 | * values exits the box boundaries we need to change it in the | |
85 | * opposite direction because (max - min) must be a power of two. | |
82e5a82f | 86 | * Also we initialize the external K value to -1 so that we can |
3443546f LT |
87 | * avoid extra conditions check inside the core loop. |
88 | */ | |
89 | if (fmin > dmin) | |
90 | kvdf[--fmin - 1] = -1; | |
91 | else | |
92 | ++fmin; | |
93 | if (fmax < dmax) | |
94 | kvdf[++fmax + 1] = -1; | |
95 | else | |
96 | --fmax; | |
97 | ||
98 | for (d = fmax; d >= fmin; d -= 2) { | |
99 | if (kvdf[d - 1] >= kvdf[d + 1]) | |
100 | i1 = kvdf[d - 1] + 1; | |
101 | else | |
102 | i1 = kvdf[d + 1]; | |
103 | prev1 = i1; | |
104 | i2 = i1 - d; | |
105 | for (; i1 < lim1 && i2 < lim2 && ha1[i1] == ha2[i2]; i1++, i2++); | |
106 | if (i1 - prev1 > xenv->snake_cnt) | |
107 | got_snake = 1; | |
108 | kvdf[d] = i1; | |
109 | if (odd && bmin <= d && d <= bmax && kvdb[d] <= i1) { | |
110 | spl->i1 = i1; | |
111 | spl->i2 = i2; | |
112 | spl->min_lo = spl->min_hi = 1; | |
113 | return ec; | |
114 | } | |
115 | } | |
116 | ||
117 | /* | |
118 | * We need to extent the diagonal "domain" by one. If the next | |
119 | * values exits the box boundaries we need to change it in the | |
120 | * opposite direction because (max - min) must be a power of two. | |
82e5a82f | 121 | * Also we initialize the external K value to -1 so that we can |
3443546f LT |
122 | * avoid extra conditions check inside the core loop. |
123 | */ | |
124 | if (bmin > dmin) | |
125 | kvdb[--bmin - 1] = XDL_LINE_MAX; | |
126 | else | |
127 | ++bmin; | |
128 | if (bmax < dmax) | |
129 | kvdb[++bmax + 1] = XDL_LINE_MAX; | |
130 | else | |
131 | --bmax; | |
132 | ||
133 | for (d = bmax; d >= bmin; d -= 2) { | |
134 | if (kvdb[d - 1] < kvdb[d + 1]) | |
135 | i1 = kvdb[d - 1]; | |
136 | else | |
137 | i1 = kvdb[d + 1] - 1; | |
138 | prev1 = i1; | |
139 | i2 = i1 - d; | |
140 | for (; i1 > off1 && i2 > off2 && ha1[i1 - 1] == ha2[i2 - 1]; i1--, i2--); | |
141 | if (prev1 - i1 > xenv->snake_cnt) | |
142 | got_snake = 1; | |
143 | kvdb[d] = i1; | |
144 | if (!odd && fmin <= d && d <= fmax && i1 <= kvdf[d]) { | |
145 | spl->i1 = i1; | |
146 | spl->i2 = i2; | |
147 | spl->min_lo = spl->min_hi = 1; | |
148 | return ec; | |
149 | } | |
150 | } | |
151 | ||
152 | if (need_min) | |
153 | continue; | |
154 | ||
155 | /* | |
156 | * If the edit cost is above the heuristic trigger and if | |
157 | * we got a good snake, we sample current diagonals to see | |
158 | * if some of the, have reached an "interesting" path. Our | |
159 | * measure is a function of the distance from the diagonal | |
160 | * corner (i1 + i2) penalized with the distance from the | |
161 | * mid diagonal itself. If this value is above the current | |
162 | * edit cost times a magic factor (XDL_K_HEUR) we consider | |
163 | * it interesting. | |
164 | */ | |
165 | if (got_snake && ec > xenv->heur_min) { | |
166 | for (best = 0, d = fmax; d >= fmin; d -= 2) { | |
167 | dd = d > fmid ? d - fmid: fmid - d; | |
168 | i1 = kvdf[d]; | |
169 | i2 = i1 - d; | |
170 | v = (i1 - off1) + (i2 - off2) - dd; | |
171 | ||
172 | if (v > XDL_K_HEUR * ec && v > best && | |
173 | off1 + xenv->snake_cnt <= i1 && i1 < lim1 && | |
174 | off2 + xenv->snake_cnt <= i2 && i2 < lim2) { | |
175 | for (k = 1; ha1[i1 - k] == ha2[i2 - k]; k++) | |
176 | if (k == xenv->snake_cnt) { | |
177 | best = v; | |
178 | spl->i1 = i1; | |
179 | spl->i2 = i2; | |
180 | break; | |
181 | } | |
182 | } | |
183 | } | |
184 | if (best > 0) { | |
185 | spl->min_lo = 1; | |
186 | spl->min_hi = 0; | |
187 | return ec; | |
188 | } | |
189 | ||
190 | for (best = 0, d = bmax; d >= bmin; d -= 2) { | |
191 | dd = d > bmid ? d - bmid: bmid - d; | |
192 | i1 = kvdb[d]; | |
193 | i2 = i1 - d; | |
194 | v = (lim1 - i1) + (lim2 - i2) - dd; | |
195 | ||
196 | if (v > XDL_K_HEUR * ec && v > best && | |
197 | off1 < i1 && i1 <= lim1 - xenv->snake_cnt && | |
198 | off2 < i2 && i2 <= lim2 - xenv->snake_cnt) { | |
199 | for (k = 0; ha1[i1 + k] == ha2[i2 + k]; k++) | |
200 | if (k == xenv->snake_cnt - 1) { | |
201 | best = v; | |
202 | spl->i1 = i1; | |
203 | spl->i2 = i2; | |
204 | break; | |
205 | } | |
206 | } | |
207 | } | |
208 | if (best > 0) { | |
209 | spl->min_lo = 0; | |
210 | spl->min_hi = 1; | |
211 | return ec; | |
212 | } | |
213 | } | |
214 | ||
215 | /* | |
216 | * Enough is enough. We spent too much time here and now we collect | |
217 | * the furthest reaching path using the (i1 + i2) measure. | |
218 | */ | |
219 | if (ec >= xenv->mxcost) { | |
220 | long fbest, fbest1, bbest, bbest1; | |
221 | ||
0ed49a3e | 222 | fbest = fbest1 = -1; |
3443546f LT |
223 | for (d = fmax; d >= fmin; d -= 2) { |
224 | i1 = XDL_MIN(kvdf[d], lim1); | |
225 | i2 = i1 - d; | |
226 | if (lim2 < i2) | |
227 | i1 = lim2 + d, i2 = lim2; | |
228 | if (fbest < i1 + i2) { | |
229 | fbest = i1 + i2; | |
230 | fbest1 = i1; | |
231 | } | |
232 | } | |
233 | ||
0ed49a3e | 234 | bbest = bbest1 = XDL_LINE_MAX; |
3443546f LT |
235 | for (d = bmax; d >= bmin; d -= 2) { |
236 | i1 = XDL_MAX(off1, kvdb[d]); | |
237 | i2 = i1 - d; | |
238 | if (i2 < off2) | |
239 | i1 = off2 + d, i2 = off2; | |
240 | if (i1 + i2 < bbest) { | |
241 | bbest = i1 + i2; | |
242 | bbest1 = i1; | |
243 | } | |
244 | } | |
245 | ||
246 | if ((lim1 + lim2) - bbest < fbest - (off1 + off2)) { | |
247 | spl->i1 = fbest1; | |
248 | spl->i2 = fbest - fbest1; | |
249 | spl->min_lo = 1; | |
250 | spl->min_hi = 0; | |
251 | } else { | |
252 | spl->i1 = bbest1; | |
253 | spl->i2 = bbest - bbest1; | |
254 | spl->min_lo = 0; | |
255 | spl->min_hi = 1; | |
256 | } | |
257 | return ec; | |
258 | } | |
259 | } | |
3443546f LT |
260 | } |
261 | ||
262 | ||
263 | /* | |
264 | * Rule: "Divide et Impera". Recursively split the box in sub-boxes by calling | |
265 | * the box splitting function. Note that the real job (marking changed lines) | |
266 | * is done in the two boundary reaching checks. | |
267 | */ | |
268 | int xdl_recs_cmp(diffdata_t *dd1, long off1, long lim1, | |
269 | diffdata_t *dd2, long off2, long lim2, | |
270 | long *kvdf, long *kvdb, int need_min, xdalgoenv_t *xenv) { | |
271 | unsigned long const *ha1 = dd1->ha, *ha2 = dd2->ha; | |
272 | ||
273 | /* | |
274 | * Shrink the box by walking through each diagonal snake (SW and NE). | |
275 | */ | |
276 | for (; off1 < lim1 && off2 < lim2 && ha1[off1] == ha2[off2]; off1++, off2++); | |
277 | for (; off1 < lim1 && off2 < lim2 && ha1[lim1 - 1] == ha2[lim2 - 1]; lim1--, lim2--); | |
278 | ||
279 | /* | |
280 | * If one dimension is empty, then all records on the other one must | |
281 | * be obviously changed. | |
282 | */ | |
283 | if (off1 == lim1) { | |
284 | char *rchg2 = dd2->rchg; | |
285 | long *rindex2 = dd2->rindex; | |
286 | ||
287 | for (; off2 < lim2; off2++) | |
288 | rchg2[rindex2[off2]] = 1; | |
289 | } else if (off2 == lim2) { | |
290 | char *rchg1 = dd1->rchg; | |
291 | long *rindex1 = dd1->rindex; | |
292 | ||
293 | for (; off1 < lim1; off1++) | |
294 | rchg1[rindex1[off1]] = 1; | |
295 | } else { | |
3443546f | 296 | xdpsplit_t spl; |
0ed49a3e | 297 | spl.i1 = spl.i2 = 0; |
3443546f LT |
298 | |
299 | /* | |
300 | * Divide ... | |
301 | */ | |
8e24cbae BK |
302 | if (xdl_split(ha1, off1, lim1, ha2, off2, lim2, kvdf, kvdb, |
303 | need_min, &spl, xenv) < 0) { | |
3443546f LT |
304 | |
305 | return -1; | |
306 | } | |
307 | ||
308 | /* | |
309 | * ... et Impera. | |
310 | */ | |
311 | if (xdl_recs_cmp(dd1, off1, spl.i1, dd2, off2, spl.i2, | |
312 | kvdf, kvdb, spl.min_lo, xenv) < 0 || | |
313 | xdl_recs_cmp(dd1, spl.i1, lim1, dd2, spl.i2, lim2, | |
314 | kvdf, kvdb, spl.min_hi, xenv) < 0) { | |
315 | ||
316 | return -1; | |
317 | } | |
318 | } | |
319 | ||
320 | return 0; | |
321 | } | |
322 | ||
323 | ||
324 | int xdl_do_diff(mmfile_t *mf1, mmfile_t *mf2, xpparam_t const *xpp, | |
325 | xdfenv_t *xe) { | |
326 | long ndiags; | |
327 | long *kvd, *kvdf, *kvdb; | |
328 | xdalgoenv_t xenv; | |
329 | diffdata_t dd1, dd2; | |
330 | ||
307ab20b | 331 | if (XDF_DIFF_ALG(xpp->flags) == XDF_PATIENCE_DIFF) |
92b7de93 JS |
332 | return xdl_do_patience_diff(mf1, mf2, xpp, xe); |
333 | ||
307ab20b | 334 | if (XDF_DIFF_ALG(xpp->flags) == XDF_HISTOGRAM_DIFF) |
8c912eea TRC |
335 | return xdl_do_histogram_diff(mf1, mf2, xpp, xe); |
336 | ||
3443546f LT |
337 | if (xdl_prepare_env(mf1, mf2, xpp, xe) < 0) { |
338 | ||
339 | return -1; | |
340 | } | |
341 | ||
342 | /* | |
343 | * Allocate and setup K vectors to be used by the differential algorithm. | |
344 | * One is to store the forward path and one to store the backward path. | |
345 | */ | |
346 | ndiags = xe->xdf1.nreff + xe->xdf2.nreff + 3; | |
347 | if (!(kvd = (long *) xdl_malloc((2 * ndiags + 2) * sizeof(long)))) { | |
348 | ||
349 | xdl_free_env(xe); | |
350 | return -1; | |
351 | } | |
352 | kvdf = kvd; | |
353 | kvdb = kvdf + ndiags; | |
354 | kvdf += xe->xdf2.nreff + 1; | |
355 | kvdb += xe->xdf2.nreff + 1; | |
356 | ||
ca557aff | 357 | xenv.mxcost = xdl_bogosqrt(ndiags); |
3443546f LT |
358 | if (xenv.mxcost < XDL_MAX_COST_MIN) |
359 | xenv.mxcost = XDL_MAX_COST_MIN; | |
360 | xenv.snake_cnt = XDL_SNAKE_CNT; | |
361 | xenv.heur_min = XDL_HEUR_MIN_COST; | |
362 | ||
363 | dd1.nrec = xe->xdf1.nreff; | |
364 | dd1.ha = xe->xdf1.ha; | |
365 | dd1.rchg = xe->xdf1.rchg; | |
366 | dd1.rindex = xe->xdf1.rindex; | |
367 | dd2.nrec = xe->xdf2.nreff; | |
368 | dd2.ha = xe->xdf2.ha; | |
369 | dd2.rchg = xe->xdf2.rchg; | |
370 | dd2.rindex = xe->xdf2.rindex; | |
371 | ||
372 | if (xdl_recs_cmp(&dd1, 0, dd1.nrec, &dd2, 0, dd2.nrec, | |
373 | kvdf, kvdb, (xpp->flags & XDF_NEED_MINIMAL) != 0, &xenv) < 0) { | |
374 | ||
375 | xdl_free(kvd); | |
376 | xdl_free_env(xe); | |
377 | return -1; | |
378 | } | |
379 | ||
380 | xdl_free(kvd); | |
381 | ||
382 | return 0; | |
383 | } | |
384 | ||
385 | ||
386 | static xdchange_t *xdl_add_change(xdchange_t *xscr, long i1, long i2, long chg1, long chg2) { | |
387 | xdchange_t *xch; | |
388 | ||
389 | if (!(xch = (xdchange_t *) xdl_malloc(sizeof(xdchange_t)))) | |
390 | return NULL; | |
391 | ||
392 | xch->next = xscr; | |
393 | xch->i1 = i1; | |
394 | xch->i2 = i2; | |
395 | xch->chg1 = chg1; | |
396 | xch->chg2 = chg2; | |
36617af7 | 397 | xch->ignore = 0; |
3443546f LT |
398 | |
399 | return xch; | |
400 | } | |
401 | ||
402 | ||
c06c0b63 | 403 | static int is_blank_line(xrecord_t *rec, long flags) |
d634d61e | 404 | { |
c06c0b63 | 405 | return xdl_blankline(rec->ptr, rec->size, flags); |
d634d61e SB |
406 | } |
407 | ||
152598cb | 408 | static int recs_match(xrecord_t *rec1, xrecord_t *rec2, long flags) |
92e5b62f | 409 | { |
152598cb MH |
410 | return (rec1->ha == rec2->ha && |
411 | xdl_recmatch(rec1->ptr, rec1->size, | |
412 | rec2->ptr, rec2->size, | |
92e5b62f JK |
413 | flags)); |
414 | } | |
415 | ||
433860f3 MH |
416 | /* |
417 | * If a line is indented more than this, get_indent() just returns this value. | |
418 | * This avoids having to do absurd amounts of work for data that are not | |
419 | * human-readable text, and also ensures that the output of get_indent fits within | |
420 | * an int. | |
421 | */ | |
422 | #define MAX_INDENT 200 | |
423 | ||
424 | /* | |
425 | * Return the amount of indentation of the specified line, treating TAB as 8 | |
426 | * columns. Return -1 if line is empty or contains only whitespace. Clamp the | |
427 | * output value at MAX_INDENT. | |
428 | */ | |
429 | static int get_indent(xrecord_t *rec) | |
430 | { | |
431 | long i; | |
432 | int ret = 0; | |
433 | ||
434 | for (i = 0; i < rec->size; i++) { | |
435 | char c = rec->ptr[i]; | |
436 | ||
437 | if (!XDL_ISSPACE(c)) | |
438 | return ret; | |
439 | else if (c == ' ') | |
440 | ret += 1; | |
441 | else if (c == '\t') | |
442 | ret += 8 - ret % 8; | |
443 | /* ignore other whitespace characters */ | |
444 | ||
445 | if (ret >= MAX_INDENT) | |
446 | return MAX_INDENT; | |
447 | } | |
448 | ||
449 | /* The line contains only whitespace. */ | |
450 | return -1; | |
451 | } | |
452 | ||
453 | /* | |
454 | * If more than this number of consecutive blank rows are found, just return this | |
455 | * value. This avoids requiring O(N^2) work for pathological cases, and also | |
456 | * ensures that the output of score_split fits in an int. | |
457 | */ | |
458 | #define MAX_BLANKS 20 | |
459 | ||
460 | /* Characteristics measured about a hypothetical split position. */ | |
461 | struct split_measurement { | |
462 | /* | |
463 | * Is the split at the end of the file (aside from any blank lines)? | |
464 | */ | |
465 | int end_of_file; | |
466 | ||
467 | /* | |
468 | * How much is the line immediately following the split indented (or -1 if | |
469 | * the line is blank): | |
470 | */ | |
471 | int indent; | |
472 | ||
473 | /* | |
474 | * How many consecutive lines above the split are blank? | |
475 | */ | |
476 | int pre_blank; | |
477 | ||
478 | /* | |
479 | * How much is the nearest non-blank line above the split indented (or -1 | |
480 | * if there is no such line)? | |
481 | */ | |
482 | int pre_indent; | |
483 | ||
484 | /* | |
485 | * How many lines after the line following the split are blank? | |
486 | */ | |
487 | int post_blank; | |
488 | ||
489 | /* | |
490 | * How much is the nearest non-blank line after the line following the | |
491 | * split indented (or -1 if there is no such line)? | |
492 | */ | |
493 | int post_indent; | |
494 | }; | |
495 | ||
496 | struct split_score { | |
497 | /* The effective indent of this split (smaller is preferred). */ | |
498 | int effective_indent; | |
499 | ||
500 | /* Penalty for this split (smaller is preferred). */ | |
501 | int penalty; | |
502 | }; | |
503 | ||
504 | /* | |
505 | * Fill m with information about a hypothetical split of xdf above line split. | |
506 | */ | |
507 | static void measure_split(const xdfile_t *xdf, long split, | |
508 | struct split_measurement *m) | |
509 | { | |
510 | long i; | |
511 | ||
512 | if (split >= xdf->nrec) { | |
513 | m->end_of_file = 1; | |
514 | m->indent = -1; | |
515 | } else { | |
516 | m->end_of_file = 0; | |
517 | m->indent = get_indent(xdf->recs[split]); | |
518 | } | |
519 | ||
520 | m->pre_blank = 0; | |
521 | m->pre_indent = -1; | |
522 | for (i = split - 1; i >= 0; i--) { | |
523 | m->pre_indent = get_indent(xdf->recs[i]); | |
524 | if (m->pre_indent != -1) | |
525 | break; | |
526 | m->pre_blank += 1; | |
527 | if (m->pre_blank == MAX_BLANKS) { | |
528 | m->pre_indent = 0; | |
529 | break; | |
530 | } | |
531 | } | |
532 | ||
533 | m->post_blank = 0; | |
534 | m->post_indent = -1; | |
535 | for (i = split + 1; i < xdf->nrec; i++) { | |
536 | m->post_indent = get_indent(xdf->recs[i]); | |
537 | if (m->post_indent != -1) | |
538 | break; | |
539 | m->post_blank += 1; | |
540 | if (m->post_blank == MAX_BLANKS) { | |
541 | m->post_indent = 0; | |
542 | break; | |
543 | } | |
544 | } | |
545 | } | |
546 | ||
547 | /* | |
548 | * The empirically-determined weight factors used by score_split() below. | |
549 | * Larger values means that the position is a less favorable place to split. | |
550 | * | |
551 | * Note that scores are only ever compared against each other, so multiplying | |
552 | * all of these weight/penalty values by the same factor wouldn't change the | |
553 | * heuristic's behavior. Still, we need to set that arbitrary scale *somehow*. | |
554 | * In practice, these numbers are chosen to be large enough that they can be | |
555 | * adjusted relative to each other with sufficient precision despite using | |
556 | * integer math. | |
557 | */ | |
558 | ||
559 | /* Penalty if there are no non-blank lines before the split */ | |
560 | #define START_OF_FILE_PENALTY 1 | |
561 | ||
562 | /* Penalty if there are no non-blank lines after the split */ | |
563 | #define END_OF_FILE_PENALTY 21 | |
564 | ||
565 | /* Multiplier for the number of blank lines around the split */ | |
566 | #define TOTAL_BLANK_WEIGHT (-30) | |
567 | ||
568 | /* Multiplier for the number of blank lines after the split */ | |
569 | #define POST_BLANK_WEIGHT 6 | |
570 | ||
571 | /* | |
572 | * Penalties applied if the line is indented more than its predecessor | |
573 | */ | |
574 | #define RELATIVE_INDENT_PENALTY (-4) | |
575 | #define RELATIVE_INDENT_WITH_BLANK_PENALTY 10 | |
576 | ||
577 | /* | |
578 | * Penalties applied if the line is indented less than both its predecessor and | |
579 | * its successor | |
580 | */ | |
581 | #define RELATIVE_OUTDENT_PENALTY 24 | |
582 | #define RELATIVE_OUTDENT_WITH_BLANK_PENALTY 17 | |
583 | ||
584 | /* | |
585 | * Penalties applied if the line is indented less than its predecessor but not | |
586 | * less than its successor | |
587 | */ | |
588 | #define RELATIVE_DEDENT_PENALTY 23 | |
589 | #define RELATIVE_DEDENT_WITH_BLANK_PENALTY 17 | |
590 | ||
591 | /* | |
592 | * We only consider whether the sum of the effective indents for splits are | |
593 | * less than (-1), equal to (0), or greater than (+1) each other. The resulting | |
594 | * value is multiplied by the following weight and combined with the penalty to | |
595 | * determine the better of two scores. | |
596 | */ | |
597 | #define INDENT_WEIGHT 60 | |
598 | ||
599 | /* | |
600 | * Compute a badness score for the hypothetical split whose measurements are | |
601 | * stored in m. The weight factors were determined empirically using the tools and | |
602 | * corpus described in | |
603 | * | |
604 | * https://github.com/mhagger/diff-slider-tools | |
605 | * | |
606 | * Also see that project if you want to improve the weights based on, for example, | |
607 | * a larger or more diverse corpus. | |
608 | */ | |
609 | static void score_add_split(const struct split_measurement *m, struct split_score *s) | |
610 | { | |
611 | /* | |
612 | * A place to accumulate penalty factors (positive makes this index more | |
613 | * favored): | |
614 | */ | |
615 | int post_blank, total_blank, indent, any_blanks; | |
616 | ||
617 | if (m->pre_indent == -1 && m->pre_blank == 0) | |
618 | s->penalty += START_OF_FILE_PENALTY; | |
619 | ||
620 | if (m->end_of_file) | |
621 | s->penalty += END_OF_FILE_PENALTY; | |
622 | ||
623 | /* | |
624 | * Set post_blank to the number of blank lines following the split, | |
625 | * including the line immediately after the split: | |
626 | */ | |
627 | post_blank = (m->indent == -1) ? 1 + m->post_blank : 0; | |
628 | total_blank = m->pre_blank + post_blank; | |
629 | ||
630 | /* Penalties based on nearby blank lines: */ | |
631 | s->penalty += TOTAL_BLANK_WEIGHT * total_blank; | |
632 | s->penalty += POST_BLANK_WEIGHT * post_blank; | |
633 | ||
634 | if (m->indent != -1) | |
635 | indent = m->indent; | |
636 | else | |
637 | indent = m->post_indent; | |
638 | ||
639 | any_blanks = (total_blank != 0); | |
640 | ||
641 | /* Note that the effective indent is -1 at the end of the file: */ | |
642 | s->effective_indent += indent; | |
643 | ||
644 | if (indent == -1) { | |
645 | /* No additional adjustments needed. */ | |
646 | } else if (m->pre_indent == -1) { | |
647 | /* No additional adjustments needed. */ | |
648 | } else if (indent > m->pre_indent) { | |
649 | /* | |
650 | * The line is indented more than its predecessor. | |
651 | */ | |
652 | s->penalty += any_blanks ? | |
653 | RELATIVE_INDENT_WITH_BLANK_PENALTY : | |
654 | RELATIVE_INDENT_PENALTY; | |
655 | } else if (indent == m->pre_indent) { | |
656 | /* | |
657 | * The line has the same indentation level as its predecessor. | |
658 | * No additional adjustments needed. | |
659 | */ | |
660 | } else { | |
661 | /* | |
662 | * The line is indented less than its predecessor. It could be | |
663 | * the block terminator of the previous block, but it could | |
664 | * also be the start of a new block (e.g., an "else" block, or | |
665 | * maybe the previous block didn't have a block terminator). | |
666 | * Try to distinguish those cases based on what comes next: | |
667 | */ | |
668 | if (m->post_indent != -1 && m->post_indent > indent) { | |
669 | /* | |
670 | * The following line is indented more. So it is likely | |
671 | * that this line is the start of a block. | |
672 | */ | |
673 | s->penalty += any_blanks ? | |
674 | RELATIVE_OUTDENT_WITH_BLANK_PENALTY : | |
675 | RELATIVE_OUTDENT_PENALTY; | |
676 | } else { | |
677 | /* | |
678 | * That was probably the end of a block. | |
679 | */ | |
680 | s->penalty += any_blanks ? | |
681 | RELATIVE_DEDENT_WITH_BLANK_PENALTY : | |
682 | RELATIVE_DEDENT_PENALTY; | |
683 | } | |
684 | } | |
685 | } | |
686 | ||
687 | static int score_cmp(struct split_score *s1, struct split_score *s2) | |
688 | { | |
689 | /* -1 if s1.effective_indent < s2->effective_indent, etc. */ | |
690 | int cmp_indents = ((s1->effective_indent > s2->effective_indent) - | |
691 | (s1->effective_indent < s2->effective_indent)); | |
692 | ||
693 | return INDENT_WEIGHT * cmp_indents + (s1->penalty - s2->penalty); | |
694 | } | |
695 | ||
e8adf23d MH |
696 | /* |
697 | * Represent a group of changed lines in an xdfile_t (i.e., a contiguous group | |
698 | * of lines that was inserted or deleted from the corresponding version of the | |
699 | * file). We consider there to be such a group at the beginning of the file, at | |
700 | * the end of the file, and between any two unchanged lines, though most such | |
701 | * groups will usually be empty. | |
702 | * | |
703 | * If the first line in a group is equal to the line following the group, then | |
704 | * the group can be slid down. Similarly, if the last line in a group is equal | |
705 | * to the line preceding the group, then the group can be slid up. See | |
706 | * group_slide_down() and group_slide_up(). | |
707 | * | |
708 | * Note that loops that are testing for changed lines in xdf->rchg do not need | |
709 | * index bounding since the array is prepared with a zero at position -1 and N. | |
710 | */ | |
711 | struct group { | |
712 | /* | |
713 | * The index of the first changed line in the group, or the index of | |
714 | * the unchanged line above which the (empty) group is located. | |
715 | */ | |
716 | long start; | |
295ba2fb DL |
717 | |
718 | /* | |
e8adf23d MH |
719 | * The index of the first unchanged line after the group. For an empty |
720 | * group, end is equal to start. | |
295ba2fb | 721 | */ |
e8adf23d MH |
722 | long end; |
723 | }; | |
724 | ||
725 | /* | |
726 | * Initialize g to point at the first group in xdf. | |
727 | */ | |
728 | static void group_init(xdfile_t *xdf, struct group *g) | |
729 | { | |
730 | g->start = g->end = 0; | |
731 | while (xdf->rchg[g->end]) | |
732 | g->end++; | |
733 | } | |
734 | ||
735 | /* | |
736 | * Move g to describe the next (possibly empty) group in xdf and return 0. If g | |
737 | * is already at the end of the file, do nothing and return -1. | |
738 | */ | |
739 | static inline int group_next(xdfile_t *xdf, struct group *g) | |
740 | { | |
741 | if (g->end == xdf->nrec) | |
742 | return -1; | |
743 | ||
744 | g->start = g->end + 1; | |
745 | for (g->end = g->start; xdf->rchg[g->end]; g->end++) | |
746 | ; | |
747 | ||
748 | return 0; | |
749 | } | |
750 | ||
751 | /* | |
752 | * Move g to describe the previous (possibly empty) group in xdf and return 0. | |
753 | * If g is already at the beginning of the file, do nothing and return -1. | |
754 | */ | |
755 | static inline int group_previous(xdfile_t *xdf, struct group *g) | |
756 | { | |
757 | if (g->start == 0) | |
758 | return -1; | |
759 | ||
760 | g->end = g->start - 1; | |
761 | for (g->start = g->end; xdf->rchg[g->start - 1]; g->start--) | |
762 | ; | |
763 | ||
764 | return 0; | |
765 | } | |
766 | ||
767 | /* | |
768 | * If g can be slid toward the end of the file, do so, and if it bumps into a | |
769 | * following group, expand this group to include it. Return 0 on success or -1 | |
770 | * if g cannot be slid down. | |
771 | */ | |
772 | static int group_slide_down(xdfile_t *xdf, struct group *g, long flags) | |
773 | { | |
774 | if (g->end < xdf->nrec && | |
775 | recs_match(xdf->recs[g->start], xdf->recs[g->end], flags)) { | |
776 | xdf->rchg[g->start++] = 0; | |
777 | xdf->rchg[g->end++] = 1; | |
778 | ||
779 | while (xdf->rchg[g->end]) | |
780 | g->end++; | |
781 | ||
782 | return 0; | |
783 | } else { | |
784 | return -1; | |
785 | } | |
786 | } | |
787 | ||
788 | /* | |
789 | * If g can be slid toward the beginning of the file, do so, and if it bumps | |
790 | * into a previous group, expand this group to include it. Return 0 on success | |
791 | * or -1 if g cannot be slid up. | |
792 | */ | |
793 | static int group_slide_up(xdfile_t *xdf, struct group *g, long flags) | |
794 | { | |
795 | if (g->start > 0 && | |
796 | recs_match(xdf->recs[g->start - 1], xdf->recs[g->end - 1], flags)) { | |
797 | xdf->rchg[--g->start] = 1; | |
798 | xdf->rchg[--g->end] = 0; | |
799 | ||
800 | while (xdf->rchg[g->start - 1]) | |
801 | g->start--; | |
802 | ||
803 | return 0; | |
804 | } else { | |
805 | return -1; | |
806 | } | |
807 | } | |
808 | ||
809 | static void xdl_bug(const char *msg) | |
810 | { | |
811 | fprintf(stderr, "BUG: %s\n", msg); | |
812 | exit(1); | |
813 | } | |
814 | ||
815 | /* | |
816 | * Move back and forward change groups for a consistent and pretty diff output. | |
817 | * This also helps in finding joinable change groups and reducing the diff | |
818 | * size. | |
819 | */ | |
820 | int xdl_change_compact(xdfile_t *xdf, xdfile_t *xdfo, long flags) { | |
821 | struct group g, go; | |
822 | long earliest_end, end_matching_other; | |
823 | long groupsize; | |
824 | unsigned int blank_lines; | |
825 | ||
826 | group_init(xdf, &g); | |
827 | group_init(xdfo, &go); | |
828 | ||
829 | while (1) { | |
830 | /* If the group is empty in the to-be-compacted file, skip it: */ | |
831 | if (g.end == g.start) | |
832 | goto next; | |
295ba2fb DL |
833 | |
834 | /* | |
e8adf23d MH |
835 | * Now shift the change up and then down as far as possible in |
836 | * each direction. If it bumps into any other changes, merge them. | |
295ba2fb | 837 | */ |
295ba2fb | 838 | do { |
e8adf23d | 839 | groupsize = g.end - g.start; |
295ba2fb DL |
840 | |
841 | /* | |
e8adf23d MH |
842 | * Keep track of the last "end" index that causes this |
843 | * group to align with a group of changed lines in the | |
844 | * other file. -1 indicates that we haven't found such | |
845 | * a match yet: | |
295ba2fb | 846 | */ |
e8adf23d | 847 | end_matching_other = -1; |
295ba2fb DL |
848 | |
849 | /* | |
e8adf23d MH |
850 | * Boolean value that records whether there are any blank |
851 | * lines that could be made to be the last line of this | |
852 | * group. | |
295ba2fb | 853 | */ |
e8adf23d MH |
854 | blank_lines = 0; |
855 | ||
856 | /* Shift the group backward as much as possible: */ | |
857 | while (!group_slide_up(xdf, &g, flags)) | |
858 | if (group_previous(xdfo, &go)) | |
859 | xdl_bug("group sync broken sliding up"); | |
295ba2fb DL |
860 | |
861 | /* | |
e8adf23d MH |
862 | * This is this highest that this group can be shifted. |
863 | * Record its end index: | |
295ba2fb | 864 | */ |
e8adf23d MH |
865 | earliest_end = g.end; |
866 | ||
867 | if (go.end > go.start) | |
868 | end_matching_other = g.end; | |
869 | ||
870 | /* Now shift the group forward as far as possible: */ | |
871 | while (1) { | |
872 | if (!blank_lines) | |
873 | blank_lines = is_blank_line( | |
874 | xdf->recs[g.end - 1], | |
875 | flags); | |
876 | ||
877 | if (group_slide_down(xdf, &g, flags)) | |
878 | break; | |
879 | if (group_next(xdfo, &go)) | |
880 | xdl_bug("group sync broken sliding down"); | |
881 | ||
882 | if (go.end > go.start) | |
883 | end_matching_other = g.end; | |
295ba2fb | 884 | } |
e8adf23d | 885 | } while (groupsize != g.end - g.start); |
295ba2fb | 886 | |
433860f3 MH |
887 | /* |
888 | * If the group can be shifted, then we can possibly use this | |
889 | * freedom to produce a more intuitive diff. | |
890 | * | |
891 | * The group is currently shifted as far down as possible, so the | |
892 | * heuristics below only have to handle upwards shifts. | |
893 | */ | |
894 | ||
e8adf23d MH |
895 | if (g.end == earliest_end) { |
896 | /* no shifting was possible */ | |
897 | } else if (end_matching_other != -1) { | |
cb0eded8 | 898 | /* |
e8adf23d MH |
899 | * Move the possibly merged group of changes back to line |
900 | * up with the last group of changes from the other file | |
901 | * that it can align with. | |
cb0eded8 | 902 | */ |
e8adf23d MH |
903 | while (go.end == go.start) { |
904 | if (group_slide_up(xdf, &g, flags)) | |
905 | xdl_bug("match disappeared"); | |
906 | if (group_previous(xdfo, &go)) | |
907 | xdl_bug("group sync broken sliding to match"); | |
cb0eded8 MH |
908 | } |
909 | } else if ((flags & XDF_COMPACTION_HEURISTIC) && blank_lines) { | |
910 | /* | |
e8adf23d MH |
911 | * Compaction heuristic: if it is possible to shift the |
912 | * group to make its bottom line a blank line, do so. | |
cb0eded8 MH |
913 | * |
914 | * As we already shifted the group forward as far as | |
e8adf23d MH |
915 | * possible in the earlier loop, we only need to handle |
916 | * backward shifts, not forward ones. | |
cb0eded8 | 917 | */ |
e8adf23d MH |
918 | while (!is_blank_line(xdf->recs[g.end - 1], flags)) { |
919 | if (group_slide_up(xdf, &g, flags)) | |
920 | xdl_bug("blank line disappeared"); | |
921 | if (group_previous(xdfo, &go)) | |
922 | xdl_bug("group sync broken sliding to blank line"); | |
d634d61e | 923 | } |
433860f3 MH |
924 | } else if (flags & XDF_INDENT_HEURISTIC) { |
925 | /* | |
926 | * Indent heuristic: a group of pure add/delete lines | |
927 | * implies two splits, one between the end of the "before" | |
928 | * context and the start of the group, and another between | |
929 | * the end of the group and the beginning of the "after" | |
930 | * context. Some splits are aesthetically better and some | |
931 | * are worse. We compute a badness "score" for each split, | |
932 | * and add the scores for the two splits to define a | |
933 | * "score" for each position that the group can be shifted | |
934 | * to. Then we pick the shift with the lowest score. | |
935 | */ | |
936 | long shift, best_shift = -1; | |
937 | struct split_score best_score; | |
938 | ||
939 | for (shift = earliest_end; shift <= g.end; shift++) { | |
940 | struct split_measurement m; | |
941 | struct split_score score = {0, 0}; | |
942 | ||
943 | measure_split(xdf, shift, &m); | |
944 | score_add_split(&m, &score); | |
945 | measure_split(xdf, shift - groupsize, &m); | |
946 | score_add_split(&m, &score); | |
947 | if (best_shift == -1 || | |
948 | score_cmp(&score, &best_score) <= 0) { | |
949 | best_score.effective_indent = score.effective_indent; | |
950 | best_score.penalty = score.penalty; | |
951 | best_shift = shift; | |
952 | } | |
953 | } | |
954 | ||
955 | while (g.end > best_shift) { | |
956 | if (group_slide_up(xdf, &g, flags)) | |
957 | xdl_bug("best shift unreached"); | |
958 | if (group_previous(xdfo, &go)) | |
959 | xdl_bug("group sync broken sliding to blank line"); | |
960 | } | |
d634d61e | 961 | } |
e8adf23d MH |
962 | |
963 | next: | |
964 | /* Move past the just-processed group: */ | |
965 | if (group_next(xdf, &g)) | |
966 | break; | |
967 | if (group_next(xdfo, &go)) | |
968 | xdl_bug("group sync broken moving to next group"); | |
295ba2fb DL |
969 | } |
970 | ||
e8adf23d MH |
971 | if (!group_next(xdfo, &go)) |
972 | xdl_bug("group sync broken at end of file"); | |
973 | ||
295ba2fb DL |
974 | return 0; |
975 | } | |
976 | ||
977 | ||
3443546f LT |
978 | int xdl_build_script(xdfenv_t *xe, xdchange_t **xscr) { |
979 | xdchange_t *cscr = NULL, *xch; | |
980 | char *rchg1 = xe->xdf1.rchg, *rchg2 = xe->xdf2.rchg; | |
981 | long i1, i2, l1, l2; | |
982 | ||
983 | /* | |
984 | * Trivial. Collects "groups" of changes and creates an edit script. | |
985 | */ | |
986 | for (i1 = xe->xdf1.nrec, i2 = xe->xdf2.nrec; i1 >= 0 || i2 >= 0; i1--, i2--) | |
987 | if (rchg1[i1 - 1] || rchg2[i2 - 1]) { | |
988 | for (l1 = i1; rchg1[i1 - 1]; i1--); | |
989 | for (l2 = i2; rchg2[i2 - 1]; i2--); | |
990 | ||
991 | if (!(xch = xdl_add_change(cscr, i1, i2, l1 - i1, l2 - i2))) { | |
992 | xdl_free_script(cscr); | |
993 | return -1; | |
994 | } | |
995 | cscr = xch; | |
996 | } | |
997 | ||
998 | *xscr = cscr; | |
999 | ||
1000 | return 0; | |
1001 | } | |
1002 | ||
1003 | ||
1004 | void xdl_free_script(xdchange_t *xscr) { | |
1005 | xdchange_t *xch; | |
1006 | ||
1007 | while ((xch = xscr) != NULL) { | |
1008 | xscr = xscr->next; | |
1009 | xdl_free(xch); | |
1010 | } | |
1011 | } | |
1012 | ||
467d348c RS |
1013 | static int xdl_call_hunk_func(xdfenv_t *xe, xdchange_t *xscr, xdemitcb_t *ecb, |
1014 | xdemitconf_t const *xecfg) | |
1015 | { | |
1016 | xdchange_t *xch, *xche; | |
1017 | ||
1018 | for (xch = xscr; xch; xch = xche->next) { | |
36617af7 AP |
1019 | xche = xdl_get_hunk(&xch, xecfg); |
1020 | if (!xch) | |
1021 | break; | |
467d348c RS |
1022 | if (xecfg->hunk_func(xch->i1, xche->i1 + xche->chg1 - xch->i1, |
1023 | xch->i2, xche->i2 + xche->chg2 - xch->i2, | |
1024 | ecb->priv) < 0) | |
1025 | return -1; | |
1026 | } | |
1027 | return 0; | |
1028 | } | |
3443546f | 1029 | |
36617af7 AP |
1030 | static void xdl_mark_ignorable(xdchange_t *xscr, xdfenv_t *xe, long flags) |
1031 | { | |
1032 | xdchange_t *xch; | |
1033 | ||
1034 | for (xch = xscr; xch; xch = xch->next) { | |
1035 | int ignore = 1; | |
1036 | xrecord_t **rec; | |
1037 | long i; | |
1038 | ||
1039 | rec = &xe->xdf1.recs[xch->i1]; | |
1040 | for (i = 0; i < xch->chg1 && ignore; i++) | |
1041 | ignore = xdl_blankline(rec[i]->ptr, rec[i]->size, flags); | |
1042 | ||
1043 | rec = &xe->xdf2.recs[xch->i2]; | |
1044 | for (i = 0; i < xch->chg2 && ignore; i++) | |
1045 | ignore = xdl_blankline(rec[i]->ptr, rec[i]->size, flags); | |
1046 | ||
1047 | xch->ignore = ignore; | |
1048 | } | |
1049 | } | |
1050 | ||
3443546f LT |
1051 | int xdl_diff(mmfile_t *mf1, mmfile_t *mf2, xpparam_t const *xpp, |
1052 | xdemitconf_t const *xecfg, xdemitcb_t *ecb) { | |
1053 | xdchange_t *xscr; | |
1054 | xdfenv_t xe; | |
3319e606 | 1055 | emit_func_t ef = xecfg->hunk_func ? xdl_call_hunk_func : xdl_emit_diff; |
467d348c | 1056 | |
3443546f LT |
1057 | if (xdl_do_diff(mf1, mf2, xpp, &xe) < 0) { |
1058 | ||
1059 | return -1; | |
1060 | } | |
0d21efa5 JS |
1061 | if (xdl_change_compact(&xe.xdf1, &xe.xdf2, xpp->flags) < 0 || |
1062 | xdl_change_compact(&xe.xdf2, &xe.xdf1, xpp->flags) < 0 || | |
295ba2fb | 1063 | xdl_build_script(&xe, &xscr) < 0) { |
3443546f LT |
1064 | |
1065 | xdl_free_env(&xe); | |
1066 | return -1; | |
1067 | } | |
3443546f | 1068 | if (xscr) { |
36617af7 AP |
1069 | if (xpp->flags & XDF_IGNORE_BLANK_LINES) |
1070 | xdl_mark_ignorable(xscr, &xe, xpp->flags); | |
1071 | ||
ef2e62fe | 1072 | if (ef(&xe, xscr, ecb, xecfg) < 0) { |
3443546f LT |
1073 | |
1074 | xdl_free_script(xscr); | |
1075 | xdl_free_env(&xe); | |
1076 | return -1; | |
1077 | } | |
3443546f LT |
1078 | xdl_free_script(xscr); |
1079 | } | |
3443546f LT |
1080 | xdl_free_env(&xe); |
1081 | ||
1082 | return 0; | |
1083 | } |