freshlybakedca.ke
Git fork
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, see
17 * <http://www.gnu.org/licenses/>.
18 *
19 * Davide Libenzi <davidel@xmailserver.org>
20 *
21 */
22
23#include "xinclude.h"
24
25static unsigned long get_hash(xdfile_t *xdf, long index)
26{
27 return xdf->recs[xdf->rindex[index]].ha;
28}
29
30#define XDL_MAX_COST_MIN 256
31#define XDL_HEUR_MIN_COST 256
32#define XDL_LINE_MAX (long)((1UL << (CHAR_BIT * sizeof(long) - 1)) - 1)
33#define XDL_SNAKE_CNT 20
34#define XDL_K_HEUR 4
35
36typedef struct s_xdpsplit {
37 long i1, i2;
38 int min_lo, min_hi;
39} xdpsplit_t;
40
41/*
42 * See "An O(ND) Difference Algorithm and its Variations", by Eugene Myers.
43 * Basically considers a "box" (off1, off2, lim1, lim2) and scan from both
44 * the forward diagonal starting from (off1, off2) and the backward diagonal
45 * starting from (lim1, lim2). If the K values on the same diagonal crosses
46 * returns the furthest point of reach. We might encounter expensive edge cases
47 * using this algorithm, so a little bit of heuristic is needed to cut the
48 * search and to return a suboptimal point.
49 */
50static long xdl_split(xdfile_t *xdf1, long off1, long lim1,
51 xdfile_t *xdf2, long off2, long lim2,
52 long *kvdf, long *kvdb, int need_min, xdpsplit_t *spl,
53 xdalgoenv_t *xenv) {
54 long dmin = off1 - lim2, dmax = lim1 - off2;
55 long fmid = off1 - off2, bmid = lim1 - lim2;
56 long odd = (fmid - bmid) & 1;
57 long fmin = fmid, fmax = fmid;
58 long bmin = bmid, bmax = bmid;
59 long ec, d, i1, i2, prev1, best, dd, v, k;
60
61 /*
62 * Set initial diagonal values for both forward and backward path.
63 */
64 kvdf[fmid] = off1;
65 kvdb[bmid] = lim1;
66
67 for (ec = 1;; ec++) {
68 int got_snake = 0;
69
70 /*
71 * We need to extend the diagonal "domain" by one. If the next
72 * values exits the box boundaries we need to change it in the
73 * opposite direction because (max - min) must be a power of
74 * two.
75 *
76 * Also we initialize the external K value to -1 so that we can
77 * avoid extra conditions in the check inside the core loop.
78 */
79 if (fmin > dmin)
80 kvdf[--fmin - 1] = -1;
81 else
82 ++fmin;
83 if (fmax < dmax)
84 kvdf[++fmax + 1] = -1;
85 else
86 --fmax;
87
88 for (d = fmax; d >= fmin; d -= 2) {
89 if (kvdf[d - 1] >= kvdf[d + 1])
90 i1 = kvdf[d - 1] + 1;
91 else
92 i1 = kvdf[d + 1];
93 prev1 = i1;
94 i2 = i1 - d;
95 for (; i1 < lim1 && i2 < lim2 && get_hash(xdf1, i1) == get_hash(xdf2, i2); i1++, i2++);
96 if (i1 - prev1 > xenv->snake_cnt)
97 got_snake = 1;
98 kvdf[d] = i1;
99 if (odd && bmin <= d && d <= bmax && kvdb[d] <= i1) {
100 spl->i1 = i1;
101 spl->i2 = i2;
102 spl->min_lo = spl->min_hi = 1;
103 return ec;
104 }
105 }
106
107 /*
108 * We need to extend the diagonal "domain" by one. If the next
109 * values exits the box boundaries we need to change it in the
110 * opposite direction because (max - min) must be a power of
111 * two.
112 *
113 * Also we initialize the external K value to -1 so that we can
114 * avoid extra conditions in the check inside the core loop.
115 */
116 if (bmin > dmin)
117 kvdb[--bmin - 1] = XDL_LINE_MAX;
118 else
119 ++bmin;
120 if (bmax < dmax)
121 kvdb[++bmax + 1] = XDL_LINE_MAX;
122 else
123 --bmax;
124
125 for (d = bmax; d >= bmin; d -= 2) {
126 if (kvdb[d - 1] < kvdb[d + 1])
127 i1 = kvdb[d - 1];
128 else
129 i1 = kvdb[d + 1] - 1;
130 prev1 = i1;
131 i2 = i1 - d;
132 for (; i1 > off1 && i2 > off2 && get_hash(xdf1, i1 - 1) == get_hash(xdf2, i2 - 1); i1--, i2--);
133 if (prev1 - i1 > xenv->snake_cnt)
134 got_snake = 1;
135 kvdb[d] = i1;
136 if (!odd && fmin <= d && d <= fmax && i1 <= kvdf[d]) {
137 spl->i1 = i1;
138 spl->i2 = i2;
139 spl->min_lo = spl->min_hi = 1;
140 return ec;
141 }
142 }
143
144 if (need_min)
145 continue;
146
147 /*
148 * If the edit cost is above the heuristic trigger and if
149 * we got a good snake, we sample current diagonals to see
150 * if some of them have reached an "interesting" path. Our
151 * measure is a function of the distance from the diagonal
152 * corner (i1 + i2) penalized with the distance from the
153 * mid diagonal itself. If this value is above the current
154 * edit cost times a magic factor (XDL_K_HEUR) we consider
155 * it interesting.
156 */
157 if (got_snake && ec > xenv->heur_min) {
158 for (best = 0, d = fmax; d >= fmin; d -= 2) {
159 dd = d > fmid ? d - fmid: fmid - d;
160 i1 = kvdf[d];
161 i2 = i1 - d;
162 v = (i1 - off1) + (i2 - off2) - dd;
163
164 if (v > XDL_K_HEUR * ec && v > best &&
165 off1 + xenv->snake_cnt <= i1 && i1 < lim1 &&
166 off2 + xenv->snake_cnt <= i2 && i2 < lim2) {
167 for (k = 1; get_hash(xdf1, i1 - k) == get_hash(xdf2, i2 - k); k++)
168 if (k == xenv->snake_cnt) {
169 best = v;
170 spl->i1 = i1;
171 spl->i2 = i2;
172 break;
173 }
174 }
175 }
176 if (best > 0) {
177 spl->min_lo = 1;
178 spl->min_hi = 0;
179 return ec;
180 }
181
182 for (best = 0, d = bmax; d >= bmin; d -= 2) {
183 dd = d > bmid ? d - bmid: bmid - d;
184 i1 = kvdb[d];
185 i2 = i1 - d;
186 v = (lim1 - i1) + (lim2 - i2) - dd;
187
188 if (v > XDL_K_HEUR * ec && v > best &&
189 off1 < i1 && i1 <= lim1 - xenv->snake_cnt &&
190 off2 < i2 && i2 <= lim2 - xenv->snake_cnt) {
191 for (k = 0; get_hash(xdf1, i1 + k) == get_hash(xdf2, i2 + k); k++)
192 if (k == xenv->snake_cnt - 1) {
193 best = v;
194 spl->i1 = i1;
195 spl->i2 = i2;
196 break;
197 }
198 }
199 }
200 if (best > 0) {
201 spl->min_lo = 0;
202 spl->min_hi = 1;
203 return ec;
204 }
205 }
206
207 /*
208 * Enough is enough. We spent too much time here and now we
209 * collect the furthest reaching path using the (i1 + i2)
210 * measure.
211 */
212 if (ec >= xenv->mxcost) {
213 long fbest, fbest1, bbest, bbest1;
214
215 fbest = fbest1 = -1;
216 for (d = fmax; d >= fmin; d -= 2) {
217 i1 = XDL_MIN(kvdf[d], lim1);
218 i2 = i1 - d;
219 if (lim2 < i2) {
220 i1 = lim2 + d;
221 i2 = lim2;
222 }
223 if (fbest < i1 + i2) {
224 fbest = i1 + i2;
225 fbest1 = i1;
226 }
227 }
228
229 bbest = bbest1 = XDL_LINE_MAX;
230 for (d = bmax; d >= bmin; d -= 2) {
231 i1 = XDL_MAX(off1, kvdb[d]);
232 i2 = i1 - d;
233 if (i2 < off2) {
234 i1 = off2 + d;
235 i2 = off2;
236 }
237 if (i1 + i2 < bbest) {
238 bbest = i1 + i2;
239 bbest1 = i1;
240 }
241 }
242
243 if ((lim1 + lim2) - bbest < fbest - (off1 + off2)) {
244 spl->i1 = fbest1;
245 spl->i2 = fbest - fbest1;
246 spl->min_lo = 1;
247 spl->min_hi = 0;
248 } else {
249 spl->i1 = bbest1;
250 spl->i2 = bbest - bbest1;
251 spl->min_lo = 0;
252 spl->min_hi = 1;
253 }
254 return ec;
255 }
256 }
257}
258
259
260/*
261 * Rule: "Divide et Impera" (divide & conquer). Recursively split the box in
262 * sub-boxes by calling the box splitting function. Note that the real job
263 * (marking changed lines) is done in the two boundary reaching checks.
264 */
265int xdl_recs_cmp(xdfile_t *xdf1, long off1, long lim1,
266 xdfile_t *xdf2, long off2, long lim2,
267 long *kvdf, long *kvdb, int need_min, xdalgoenv_t *xenv) {
268
269 /*
270 * Shrink the box by walking through each diagonal snake (SW and NE).
271 */
272 for (; off1 < lim1 && off2 < lim2 && get_hash(xdf1, off1) == get_hash(xdf2, off2); off1++, off2++);
273 for (; off1 < lim1 && off2 < lim2 && get_hash(xdf1, lim1 - 1) == get_hash(xdf2, lim2 - 1); lim1--, lim2--);
274
275 /*
276 * If one dimension is empty, then all records on the other one must
277 * be obviously changed.
278 */
279 if (off1 == lim1) {
280 for (; off2 < lim2; off2++)
281 xdf2->changed[xdf2->rindex[off2]] = true;
282 } else if (off2 == lim2) {
283 for (; off1 < lim1; off1++)
284 xdf1->changed[xdf1->rindex[off1]] = true;
285 } else {
286 xdpsplit_t spl;
287 spl.i1 = spl.i2 = 0;
288
289 /*
290 * Divide ...
291 */
292 if (xdl_split(xdf1, off1, lim1, xdf2, off2, lim2, kvdf, kvdb,
293 need_min, &spl, xenv) < 0) {
294
295 return -1;
296 }
297
298 /*
299 * ... et Impera.
300 */
301 if (xdl_recs_cmp(xdf1, off1, spl.i1, xdf2, off2, spl.i2,
302 kvdf, kvdb, spl.min_lo, xenv) < 0 ||
303 xdl_recs_cmp(xdf1, spl.i1, lim1, xdf2, spl.i2, lim2,
304 kvdf, kvdb, spl.min_hi, xenv) < 0) {
305
306 return -1;
307 }
308 }
309
310 return 0;
311}
312
313
314int xdl_do_diff(mmfile_t *mf1, mmfile_t *mf2, xpparam_t const *xpp,
315 xdfenv_t *xe) {
316 long ndiags;
317 long *kvd, *kvdf, *kvdb;
318 xdalgoenv_t xenv;
319 int res;
320
321 if (xdl_prepare_env(mf1, mf2, xpp, xe) < 0)
322 return -1;
323
324 if (XDF_DIFF_ALG(xpp->flags) == XDF_PATIENCE_DIFF) {
325 res = xdl_do_patience_diff(xpp, xe);
326 goto out;
327 }
328
329 if (XDF_DIFF_ALG(xpp->flags) == XDF_HISTOGRAM_DIFF) {
330 res = xdl_do_histogram_diff(xpp, xe);
331 goto out;
332 }
333
334 /*
335 * Allocate and setup K vectors to be used by the differential
336 * algorithm.
337 *
338 * One is to store the forward path and one to store the backward path.
339 */
340 ndiags = xe->xdf1.nreff + xe->xdf2.nreff + 3;
341 if (!XDL_ALLOC_ARRAY(kvd, 2 * ndiags + 2)) {
342
343 xdl_free_env(xe);
344 return -1;
345 }
346 kvdf = kvd;
347 kvdb = kvdf + ndiags;
348 kvdf += xe->xdf2.nreff + 1;
349 kvdb += xe->xdf2.nreff + 1;
350
351 xenv.mxcost = xdl_bogosqrt(ndiags);
352 if (xenv.mxcost < XDL_MAX_COST_MIN)
353 xenv.mxcost = XDL_MAX_COST_MIN;
354 xenv.snake_cnt = XDL_SNAKE_CNT;
355 xenv.heur_min = XDL_HEUR_MIN_COST;
356
357 res = xdl_recs_cmp(&xe->xdf1, 0, xe->xdf1.nreff, &xe->xdf2, 0, xe->xdf2.nreff,
358 kvdf, kvdb, (xpp->flags & XDF_NEED_MINIMAL) != 0,
359 &xenv);
360 xdl_free(kvd);
361 out:
362 if (res < 0)
363 xdl_free_env(xe);
364
365 return res;
366}
367
368
369static xdchange_t *xdl_add_change(xdchange_t *xscr, long i1, long i2, long chg1, long chg2) {
370 xdchange_t *xch;
371
372 if (!(xch = (xdchange_t *) xdl_malloc(sizeof(xdchange_t))))
373 return NULL;
374
375 xch->next = xscr;
376 xch->i1 = i1;
377 xch->i2 = i2;
378 xch->chg1 = chg1;
379 xch->chg2 = chg2;
380 xch->ignore = 0;
381
382 return xch;
383}
384
385
386static int recs_match(xrecord_t *rec1, xrecord_t *rec2)
387{
388 return (rec1->ha == rec2->ha);
389}
390
391/*
392 * If a line is indented more than this, get_indent() just returns this value.
393 * This avoids having to do absurd amounts of work for data that are not
394 * human-readable text, and also ensures that the output of get_indent fits
395 * within an int.
396 */
397#define MAX_INDENT 200
398
399/*
400 * Return the amount of indentation of the specified line, treating TAB as 8
401 * columns. Return -1 if line is empty or contains only whitespace. Clamp the
402 * output value at MAX_INDENT.
403 */
404static int get_indent(xrecord_t *rec)
405{
406 long i;
407 int ret = 0;
408
409 for (i = 0; i < rec->size; i++) {
410 char c = rec->ptr[i];
411
412 if (!XDL_ISSPACE(c))
413 return ret;
414 else if (c == ' ')
415 ret += 1;
416 else if (c == '\t')
417 ret += 8 - ret % 8;
418 /* ignore other whitespace characters */
419
420 if (ret >= MAX_INDENT)
421 return MAX_INDENT;
422 }
423
424 /* The line contains only whitespace. */
425 return -1;
426}
427
428/*
429 * If more than this number of consecutive blank rows are found, just return
430 * this value. This avoids requiring O(N^2) work for pathological cases, and
431 * also ensures that the output of score_split fits in an int.
432 */
433#define MAX_BLANKS 20
434
435/* Characteristics measured about a hypothetical split position. */
436struct split_measurement {
437 /*
438 * Is the split at the end of the file (aside from any blank lines)?
439 */
440 int end_of_file;
441
442 /*
443 * How much is the line immediately following the split indented (or -1
444 * if the line is blank):
445 */
446 int indent;
447
448 /*
449 * How many consecutive lines above the split are blank?
450 */
451 int pre_blank;
452
453 /*
454 * How much is the nearest non-blank line above the split indented (or
455 * -1 if there is no such line)?
456 */
457 int pre_indent;
458
459 /*
460 * How many lines after the line following the split are blank?
461 */
462 int post_blank;
463
464 /*
465 * How much is the nearest non-blank line after the line following the
466 * split indented (or -1 if there is no such line)?
467 */
468 int post_indent;
469};
470
471struct split_score {
472 /* The effective indent of this split (smaller is preferred). */
473 int effective_indent;
474
475 /* Penalty for this split (smaller is preferred). */
476 int penalty;
477};
478
479/*
480 * Fill m with information about a hypothetical split of xdf above line split.
481 */
482static void measure_split(const xdfile_t *xdf, long split,
483 struct split_measurement *m)
484{
485 long i;
486
487 if (split >= xdf->nrec) {
488 m->end_of_file = 1;
489 m->indent = -1;
490 } else {
491 m->end_of_file = 0;
492 m->indent = get_indent(&xdf->recs[split]);
493 }
494
495 m->pre_blank = 0;
496 m->pre_indent = -1;
497 for (i = split - 1; i >= 0; i--) {
498 m->pre_indent = get_indent(&xdf->recs[i]);
499 if (m->pre_indent != -1)
500 break;
501 m->pre_blank += 1;
502 if (m->pre_blank == MAX_BLANKS) {
503 m->pre_indent = 0;
504 break;
505 }
506 }
507
508 m->post_blank = 0;
509 m->post_indent = -1;
510 for (i = split + 1; i < xdf->nrec; i++) {
511 m->post_indent = get_indent(&xdf->recs[i]);
512 if (m->post_indent != -1)
513 break;
514 m->post_blank += 1;
515 if (m->post_blank == MAX_BLANKS) {
516 m->post_indent = 0;
517 break;
518 }
519 }
520}
521
522/*
523 * The empirically-determined weight factors used by score_split() below.
524 * Larger values means that the position is a less favorable place to split.
525 *
526 * Note that scores are only ever compared against each other, so multiplying
527 * all of these weight/penalty values by the same factor wouldn't change the
528 * heuristic's behavior. Still, we need to set that arbitrary scale *somehow*.
529 * In practice, these numbers are chosen to be large enough that they can be
530 * adjusted relative to each other with sufficient precision despite using
531 * integer math.
532 */
533
534/* Penalty if there are no non-blank lines before the split */
535#define START_OF_FILE_PENALTY 1
536
537/* Penalty if there are no non-blank lines after the split */
538#define END_OF_FILE_PENALTY 21
539
540/* Multiplier for the number of blank lines around the split */
541#define TOTAL_BLANK_WEIGHT (-30)
542
543/* Multiplier for the number of blank lines after the split */
544#define POST_BLANK_WEIGHT 6
545
546/*
547 * Penalties applied if the line is indented more than its predecessor
548 */
549#define RELATIVE_INDENT_PENALTY (-4)
550#define RELATIVE_INDENT_WITH_BLANK_PENALTY 10
551
552/*
553 * Penalties applied if the line is indented less than both its predecessor and
554 * its successor
555 */
556#define RELATIVE_OUTDENT_PENALTY 24
557#define RELATIVE_OUTDENT_WITH_BLANK_PENALTY 17
558
559/*
560 * Penalties applied if the line is indented less than its predecessor but not
561 * less than its successor
562 */
563#define RELATIVE_DEDENT_PENALTY 23
564#define RELATIVE_DEDENT_WITH_BLANK_PENALTY 17
565
566/*
567 * We only consider whether the sum of the effective indents for splits are
568 * less than (-1), equal to (0), or greater than (+1) each other. The resulting
569 * value is multiplied by the following weight and combined with the penalty to
570 * determine the better of two scores.
571 */
572#define INDENT_WEIGHT 60
573
574/*
575 * How far do we slide a hunk at most?
576 */
577#define INDENT_HEURISTIC_MAX_SLIDING 100
578
579/*
580 * Compute a badness score for the hypothetical split whose measurements are
581 * stored in m. The weight factors were determined empirically using the tools
582 * and corpus described in
583 *
584 * https://github.com/mhagger/diff-slider-tools
585 *
586 * Also see that project if you want to improve the weights based on, for
587 * example, a larger or more diverse corpus.
588 */
589static void score_add_split(const struct split_measurement *m, struct split_score *s)
590{
591 /*
592 * A place to accumulate penalty factors (positive makes this index more
593 * favored):
594 */
595 int post_blank, total_blank, indent, any_blanks;
596
597 if (m->pre_indent == -1 && m->pre_blank == 0)
598 s->penalty += START_OF_FILE_PENALTY;
599
600 if (m->end_of_file)
601 s->penalty += END_OF_FILE_PENALTY;
602
603 /*
604 * Set post_blank to the number of blank lines following the split,
605 * including the line immediately after the split:
606 */
607 post_blank = (m->indent == -1) ? 1 + m->post_blank : 0;
608 total_blank = m->pre_blank + post_blank;
609
610 /* Penalties based on nearby blank lines: */
611 s->penalty += TOTAL_BLANK_WEIGHT * total_blank;
612 s->penalty += POST_BLANK_WEIGHT * post_blank;
613
614 if (m->indent != -1)
615 indent = m->indent;
616 else
617 indent = m->post_indent;
618
619 any_blanks = (total_blank != 0);
620
621 /* Note that the effective indent is -1 at the end of the file: */
622 s->effective_indent += indent;
623
624 if (indent == -1) {
625 /* No additional adjustments needed. */
626 } else if (m->pre_indent == -1) {
627 /* No additional adjustments needed. */
628 } else if (indent > m->pre_indent) {
629 /*
630 * The line is indented more than its predecessor.
631 */
632 s->penalty += any_blanks ?
633 RELATIVE_INDENT_WITH_BLANK_PENALTY :
634 RELATIVE_INDENT_PENALTY;
635 } else if (indent == m->pre_indent) {
636 /*
637 * The line has the same indentation level as its predecessor.
638 * No additional adjustments needed.
639 */
640 } else {
641 /*
642 * The line is indented less than its predecessor. It could be
643 * the block terminator of the previous block, but it could
644 * also be the start of a new block (e.g., an "else" block, or
645 * maybe the previous block didn't have a block terminator).
646 * Try to distinguish those cases based on what comes next:
647 */
648 if (m->post_indent != -1 && m->post_indent > indent) {
649 /*
650 * The following line is indented more. So it is likely
651 * that this line is the start of a block.
652 */
653 s->penalty += any_blanks ?
654 RELATIVE_OUTDENT_WITH_BLANK_PENALTY :
655 RELATIVE_OUTDENT_PENALTY;
656 } else {
657 /*
658 * That was probably the end of a block.
659 */
660 s->penalty += any_blanks ?
661 RELATIVE_DEDENT_WITH_BLANK_PENALTY :
662 RELATIVE_DEDENT_PENALTY;
663 }
664 }
665}
666
667static int score_cmp(struct split_score *s1, struct split_score *s2)
668{
669 /* -1 if s1.effective_indent < s2->effective_indent, etc. */
670 int cmp_indents = ((s1->effective_indent > s2->effective_indent) -
671 (s1->effective_indent < s2->effective_indent));
672
673 return INDENT_WEIGHT * cmp_indents + (s1->penalty - s2->penalty);
674}
675
676/*
677 * Represent a group of changed lines in an xdfile_t (i.e., a contiguous group
678 * of lines that was inserted or deleted from the corresponding version of the
679 * file). We consider there to be such a group at the beginning of the file, at
680 * the end of the file, and between any two unchanged lines, though most such
681 * groups will usually be empty.
682 *
683 * If the first line in a group is equal to the line following the group, then
684 * the group can be slid down. Similarly, if the last line in a group is equal
685 * to the line preceding the group, then the group can be slid up. See
686 * group_slide_down() and group_slide_up().
687 *
688 * Note that loops that are testing for changed lines in xdf->rchg do not need
689 * index bounding since the array is prepared with a zero at position -1 and N.
690 */
691struct xdlgroup {
692 /*
693 * The index of the first changed line in the group, or the index of
694 * the unchanged line above which the (empty) group is located.
695 */
696 long start;
697
698 /*
699 * The index of the first unchanged line after the group. For an empty
700 * group, end is equal to start.
701 */
702 long end;
703};
704
705/*
706 * Initialize g to point at the first group in xdf.
707 */
708static void group_init(xdfile_t *xdf, struct xdlgroup *g)
709{
710 g->start = g->end = 0;
711 while (xdf->changed[g->end])
712 g->end++;
713}
714
715/*
716 * Move g to describe the next (possibly empty) group in xdf and return 0. If g
717 * is already at the end of the file, do nothing and return -1.
718 */
719static inline int group_next(xdfile_t *xdf, struct xdlgroup *g)
720{
721 if (g->end == xdf->nrec)
722 return -1;
723
724 g->start = g->end + 1;
725 for (g->end = g->start; xdf->changed[g->end]; g->end++)
726 ;
727
728 return 0;
729}
730
731/*
732 * Move g to describe the previous (possibly empty) group in xdf and return 0.
733 * If g is already at the beginning of the file, do nothing and return -1.
734 */
735static inline int group_previous(xdfile_t *xdf, struct xdlgroup *g)
736{
737 if (g->start == 0)
738 return -1;
739
740 g->end = g->start - 1;
741 for (g->start = g->end; xdf->changed[g->start - 1]; g->start--)
742 ;
743
744 return 0;
745}
746
747/*
748 * If g can be slid toward the end of the file, do so, and if it bumps into a
749 * following group, expand this group to include it. Return 0 on success or -1
750 * if g cannot be slid down.
751 */
752static int group_slide_down(xdfile_t *xdf, struct xdlgroup *g)
753{
754 if (g->end < xdf->nrec &&
755 recs_match(&xdf->recs[g->start], &xdf->recs[g->end])) {
756 xdf->changed[g->start++] = false;
757 xdf->changed[g->end++] = true;
758
759 while (xdf->changed[g->end])
760 g->end++;
761
762 return 0;
763 } else {
764 return -1;
765 }
766}
767
768/*
769 * If g can be slid toward the beginning of the file, do so, and if it bumps
770 * into a previous group, expand this group to include it. Return 0 on success
771 * or -1 if g cannot be slid up.
772 */
773static int group_slide_up(xdfile_t *xdf, struct xdlgroup *g)
774{
775 if (g->start > 0 &&
776 recs_match(&xdf->recs[g->start - 1], &xdf->recs[g->end - 1])) {
777 xdf->changed[--g->start] = true;
778 xdf->changed[--g->end] = false;
779
780 while (xdf->changed[g->start - 1])
781 g->start--;
782
783 return 0;
784 } else {
785 return -1;
786 }
787}
788
789/*
790 * Move back and forward change groups for a consistent and pretty diff output.
791 * This also helps in finding joinable change groups and reducing the diff
792 * size.
793 */
794int xdl_change_compact(xdfile_t *xdf, xdfile_t *xdfo, long flags) {
795 struct xdlgroup g, go;
796 long earliest_end, end_matching_other;
797 long groupsize;
798
799 group_init(xdf, &g);
800 group_init(xdfo, &go);
801
802 while (1) {
803 /*
804 * If the group is empty in the to-be-compacted file, skip it:
805 */
806 if (g.end == g.start)
807 goto next;
808
809 /*
810 * Now shift the change up and then down as far as possible in
811 * each direction. If it bumps into any other changes, merge
812 * them.
813 */
814 do {
815 groupsize = g.end - g.start;
816
817 /*
818 * Keep track of the last "end" index that causes this
819 * group to align with a group of changed lines in the
820 * other file. -1 indicates that we haven't found such
821 * a match yet:
822 */
823 end_matching_other = -1;
824
825 /* Shift the group backward as much as possible: */
826 while (!group_slide_up(xdf, &g))
827 if (group_previous(xdfo, &go))
828 BUG("group sync broken sliding up");
829
830 /*
831 * This is this highest that this group can be shifted.
832 * Record its end index:
833 */
834 earliest_end = g.end;
835
836 if (go.end > go.start)
837 end_matching_other = g.end;
838
839 /* Now shift the group forward as far as possible: */
840 while (1) {
841 if (group_slide_down(xdf, &g))
842 break;
843 if (group_next(xdfo, &go))
844 BUG("group sync broken sliding down");
845
846 if (go.end > go.start)
847 end_matching_other = g.end;
848 }
849 } while (groupsize != g.end - g.start);
850
851 /*
852 * If the group can be shifted, then we can possibly use this
853 * freedom to produce a more intuitive diff.
854 *
855 * The group is currently shifted as far down as possible, so
856 * the heuristics below only have to handle upwards shifts.
857 */
858
859 if (g.end == earliest_end) {
860 /* no shifting was possible */
861 } else if (end_matching_other != -1) {
862 /*
863 * Move the possibly merged group of changes back to
864 * line up with the last group of changes from the
865 * other file that it can align with.
866 */
867 while (go.end == go.start) {
868 if (group_slide_up(xdf, &g))
869 BUG("match disappeared");
870 if (group_previous(xdfo, &go))
871 BUG("group sync broken sliding to match");
872 }
873 } else if (flags & XDF_INDENT_HEURISTIC) {
874 /*
875 * Indent heuristic: a group of pure add/delete lines
876 * implies two splits, one between the end of the
877 * "before" context and the start of the group, and
878 * another between the end of the group and the
879 * beginning of the "after" context. Some splits are
880 * aesthetically better and some are worse. We compute
881 * a badness "score" for each split, and add the scores
882 * for the two splits to define a "score" for each
883 * position that the group can be shifted to. Then we
884 * pick the shift with the lowest score.
885 */
886 long shift, best_shift = -1;
887 struct split_score best_score;
888
889 shift = earliest_end;
890 if (g.end - groupsize - 1 > shift)
891 shift = g.end - groupsize - 1;
892 if (g.end - INDENT_HEURISTIC_MAX_SLIDING > shift)
893 shift = g.end - INDENT_HEURISTIC_MAX_SLIDING;
894 for (; shift <= g.end; shift++) {
895 struct split_measurement m;
896 struct split_score score = {0, 0};
897
898 measure_split(xdf, shift, &m);
899 score_add_split(&m, &score);
900 measure_split(xdf, shift - groupsize, &m);
901 score_add_split(&m, &score);
902 if (best_shift == -1 ||
903 score_cmp(&score, &best_score) <= 0) {
904 best_score.effective_indent = score.effective_indent;
905 best_score.penalty = score.penalty;
906 best_shift = shift;
907 }
908 }
909
910 while (g.end > best_shift) {
911 if (group_slide_up(xdf, &g))
912 BUG("best shift unreached");
913 if (group_previous(xdfo, &go))
914 BUG("group sync broken sliding to blank line");
915 }
916 }
917
918 next:
919 /* Move past the just-processed group: */
920 if (group_next(xdf, &g))
921 break;
922 if (group_next(xdfo, &go))
923 BUG("group sync broken moving to next group");
924 }
925
926 if (!group_next(xdfo, &go))
927 BUG("group sync broken at end of file");
928
929 return 0;
930}
931
932
933int xdl_build_script(xdfenv_t *xe, xdchange_t **xscr) {
934 xdchange_t *cscr = NULL, *xch;
935 bool *changed1 = xe->xdf1.changed, *changed2 = xe->xdf2.changed;
936 long i1, i2, l1, l2;
937
938 /*
939 * Trivial. Collects "groups" of changes and creates an edit script.
940 */
941 for (i1 = xe->xdf1.nrec, i2 = xe->xdf2.nrec; i1 >= 0 || i2 >= 0; i1--, i2--)
942 if (changed1[i1 - 1] || changed2[i2 - 1]) {
943 for (l1 = i1; changed1[i1 - 1]; i1--);
944 for (l2 = i2; changed2[i2 - 1]; i2--);
945
946 if (!(xch = xdl_add_change(cscr, i1, i2, l1 - i1, l2 - i2))) {
947 xdl_free_script(cscr);
948 return -1;
949 }
950 cscr = xch;
951 }
952
953 *xscr = cscr;
954
955 return 0;
956}
957
958
959void xdl_free_script(xdchange_t *xscr) {
960 xdchange_t *xch;
961
962 while ((xch = xscr) != NULL) {
963 xscr = xscr->next;
964 xdl_free(xch);
965 }
966}
967
968static int xdl_call_hunk_func(xdfenv_t *xe UNUSED, xdchange_t *xscr, xdemitcb_t *ecb,
969 xdemitconf_t const *xecfg)
970{
971 xdchange_t *xch, *xche;
972
973 for (xch = xscr; xch; xch = xche->next) {
974 xche = xdl_get_hunk(&xch, xecfg);
975 if (!xch)
976 break;
977 if (xecfg->hunk_func(xch->i1, xche->i1 + xche->chg1 - xch->i1,
978 xch->i2, xche->i2 + xche->chg2 - xch->i2,
979 ecb->priv) < 0)
980 return -1;
981 }
982 return 0;
983}
984
985static void xdl_mark_ignorable_lines(xdchange_t *xscr, xdfenv_t *xe, long flags)
986{
987 xdchange_t *xch;
988
989 for (xch = xscr; xch; xch = xch->next) {
990 int ignore = 1;
991 xrecord_t *rec;
992 long i;
993
994 rec = &xe->xdf1.recs[xch->i1];
995 for (i = 0; i < xch->chg1 && ignore; i++)
996 ignore = xdl_blankline(rec[i].ptr, rec[i].size, flags);
997
998 rec = &xe->xdf2.recs[xch->i2];
999 for (i = 0; i < xch->chg2 && ignore; i++)
1000 ignore = xdl_blankline(rec[i].ptr, rec[i].size, flags);
1001
1002 xch->ignore = ignore;
1003 }
1004}
1005
1006static int record_matches_regex(xrecord_t *rec, xpparam_t const *xpp) {
1007 regmatch_t regmatch;
1008 size_t i;
1009
1010 for (i = 0; i < xpp->ignore_regex_nr; i++)
1011 if (!regexec_buf(xpp->ignore_regex[i], rec->ptr, rec->size, 1,
1012 ®match, 0))
1013 return 1;
1014
1015 return 0;
1016}
1017
1018static void xdl_mark_ignorable_regex(xdchange_t *xscr, const xdfenv_t *xe,
1019 xpparam_t const *xpp)
1020{
1021 xdchange_t *xch;
1022
1023 for (xch = xscr; xch; xch = xch->next) {
1024 xrecord_t *rec;
1025 int ignore = 1;
1026 long i;
1027
1028 /*
1029 * Do not override --ignore-blank-lines.
1030 */
1031 if (xch->ignore)
1032 continue;
1033
1034 rec = &xe->xdf1.recs[xch->i1];
1035 for (i = 0; i < xch->chg1 && ignore; i++)
1036 ignore = record_matches_regex(&rec[i], xpp);
1037
1038 rec = &xe->xdf2.recs[xch->i2];
1039 for (i = 0; i < xch->chg2 && ignore; i++)
1040 ignore = record_matches_regex(&rec[i], xpp);
1041
1042 xch->ignore = ignore;
1043 }
1044}
1045
1046int xdl_diff(mmfile_t *mf1, mmfile_t *mf2, xpparam_t const *xpp,
1047 xdemitconf_t const *xecfg, xdemitcb_t *ecb) {
1048 xdchange_t *xscr;
1049 xdfenv_t xe;
1050 emit_func_t ef = xecfg->hunk_func ? xdl_call_hunk_func : xdl_emit_diff;
1051
1052 if (xdl_do_diff(mf1, mf2, xpp, &xe) < 0) {
1053
1054 return -1;
1055 }
1056 if (xdl_change_compact(&xe.xdf1, &xe.xdf2, xpp->flags) < 0 ||
1057 xdl_change_compact(&xe.xdf2, &xe.xdf1, xpp->flags) < 0 ||
1058 xdl_build_script(&xe, &xscr) < 0) {
1059
1060 xdl_free_env(&xe);
1061 return -1;
1062 }
1063 if (xscr) {
1064 if (xpp->flags & XDF_IGNORE_BLANK_LINES)
1065 xdl_mark_ignorable_lines(xscr, &xe, xpp->flags);
1066
1067 if (xpp->ignore_regex)
1068 xdl_mark_ignorable_regex(xscr, &xe, xpp);
1069
1070 if (ef(&xe, xscr, ecb, xecfg) < 0) {
1071
1072 xdl_free_script(xscr);
1073 xdl_free_env(&xe);
1074 return -1;
1075 }
1076 xdl_free_script(xscr);
1077 }
1078 xdl_free_env(&xe);
1079
1080 return 0;
1081}