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#include "xinclude.h"
24#define XDL_MAX_COST_MIN 256
28#define XDL_HEUR_MIN_COST 256
29#define XDL_LINE_MAX (long)((1UL << (8 * sizeof(long) - 1)) - 1)
30#define XDL_SNAKE_CNT 20
31#define XDL_K_HEUR 4
32typedef struct s_xdpsplit {
36 long i1, i2;
37 int min_lo, min_hi;
38} xdpsplit_t;
39static 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);
47static xdchange_t *xdl_add_change(xdchange_t *xscr, long i1, long i2, long chg1, long chg2);
48/*
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 */
62static 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 /*
74 * Set initial diagonal values for both forward and backward path.
75 */
76 kvdf[fmid] = off1;
77 kvdb[bmid] = lim1;
78 for (ec = 1;; ec++) {
80 int got_snake = 0;
81 /*
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.
86 * Also we initialize the external K value to -1 so that we can
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 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 /*
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.
121 * Also we initialize the external K value to -1 so that we can
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 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 if (need_min)
153 continue;
154 /*
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 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 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 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 /*
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 fbest = fbest1 = -1;
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 bbest = bbest1 = XDL_LINE_MAX;
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 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 }
260 return -1;
262}
263/*
266 * Rule: "Divide et Impera". Recursively split the box in sub-boxes by calling
267 * the box splitting function. Note that the real job (marking changed lines)
268 * is done in the two boundary reaching checks.
269 */
270int xdl_recs_cmp(diffdata_t *dd1, long off1, long lim1,
271 diffdata_t *dd2, long off2, long lim2,
272 long *kvdf, long *kvdb, int need_min, xdalgoenv_t *xenv) {
273 unsigned long const *ha1 = dd1->ha, *ha2 = dd2->ha;
274 /*
276 * Shrink the box by walking through each diagonal snake (SW and NE).
277 */
278 for (; off1 < lim1 && off2 < lim2 && ha1[off1] == ha2[off2]; off1++, off2++);
279 for (; off1 < lim1 && off2 < lim2 && ha1[lim1 - 1] == ha2[lim2 - 1]; lim1--, lim2--);
280 /*
282 * If one dimension is empty, then all records on the other one must
283 * be obviously changed.
284 */
285 if (off1 == lim1) {
286 char *rchg2 = dd2->rchg;
287 long *rindex2 = dd2->rindex;
288 for (; off2 < lim2; off2++)
290 rchg2[rindex2[off2]] = 1;
291 } else if (off2 == lim2) {
292 char *rchg1 = dd1->rchg;
293 long *rindex1 = dd1->rindex;
294 for (; off1 < lim1; off1++)
296 rchg1[rindex1[off1]] = 1;
297 } else {
298 long ec;
299 xdpsplit_t spl;
300 spl.i1 = spl.i2 = 0;
301 /*
303 * Divide ...
304 */
305 if ((ec = xdl_split(ha1, off1, lim1, ha2, off2, lim2, kvdf, kvdb,
306 need_min, &spl, xenv)) < 0) {
307 return -1;
309 }
310 /*
312 * ... et Impera.
313 */
314 if (xdl_recs_cmp(dd1, off1, spl.i1, dd2, off2, spl.i2,
315 kvdf, kvdb, spl.min_lo, xenv) < 0 ||
316 xdl_recs_cmp(dd1, spl.i1, lim1, dd2, spl.i2, lim2,
317 kvdf, kvdb, spl.min_hi, xenv) < 0) {
318 return -1;
320 }
321 }
322 return 0;
324}
325int xdl_do_diff(mmfile_t *mf1, mmfile_t *mf2, xpparam_t const *xpp,
328 xdfenv_t *xe) {
329 long ndiags;
330 long *kvd, *kvdf, *kvdb;
331 xdalgoenv_t xenv;
332 diffdata_t dd1, dd2;
333 if (xdl_prepare_env(mf1, mf2, xpp, xe) < 0) {
335 return -1;
337 }
338 /*
340 * Allocate and setup K vectors to be used by the differential algorithm.
341 * One is to store the forward path and one to store the backward path.
342 */
343 ndiags = xe->xdf1.nreff + xe->xdf2.nreff + 3;
344 if (!(kvd = (long *) xdl_malloc((2 * ndiags + 2) * sizeof(long)))) {
345 xdl_free_env(xe);
347 return -1;
348 }
349 kvdf = kvd;
350 kvdb = kvdf + ndiags;
351 kvdf += xe->xdf2.nreff + 1;
352 kvdb += xe->xdf2.nreff + 1;
353 xenv.mxcost = xdl_bogosqrt(ndiags);
355 if (xenv.mxcost < XDL_MAX_COST_MIN)
356 xenv.mxcost = XDL_MAX_COST_MIN;
357 xenv.snake_cnt = XDL_SNAKE_CNT;
358 xenv.heur_min = XDL_HEUR_MIN_COST;
359 dd1.nrec = xe->xdf1.nreff;
361 dd1.ha = xe->xdf1.ha;
362 dd1.rchg = xe->xdf1.rchg;
363 dd1.rindex = xe->xdf1.rindex;
364 dd2.nrec = xe->xdf2.nreff;
365 dd2.ha = xe->xdf2.ha;
366 dd2.rchg = xe->xdf2.rchg;
367 dd2.rindex = xe->xdf2.rindex;
368 if (xdl_recs_cmp(&dd1, 0, dd1.nrec, &dd2, 0, dd2.nrec,
370 kvdf, kvdb, (xpp->flags & XDF_NEED_MINIMAL) != 0, &xenv) < 0) {
371 xdl_free(kvd);
373 xdl_free_env(xe);
374 return -1;
375 }
376 xdl_free(kvd);
378 return 0;
380}
381static xdchange_t *xdl_add_change(xdchange_t *xscr, long i1, long i2, long chg1, long chg2) {
384 xdchange_t *xch;
385 if (!(xch = (xdchange_t *) xdl_malloc(sizeof(xdchange_t))))
387 return NULL;
388 xch->next = xscr;
390 xch->i1 = i1;
391 xch->i2 = i2;
392 xch->chg1 = chg1;
393 xch->chg2 = chg2;
394 return xch;
396}
397int xdl_change_compact(xdfile_t *xdf, xdfile_t *xdfo, long flags) {
400 long ix, ixo, ixs, ixref, grpsiz, nrec = xdf->nrec;
401 char *rchg = xdf->rchg, *rchgo = xdfo->rchg;
402 xrecord_t **recs = xdf->recs;
403 /*
405 * This is the same of what GNU diff does. Move back and forward
406 * change groups for a consistent and pretty diff output. This also
407 * helps in finding joinable change groups and reduce the diff size.
408 */
409 for (ix = ixo = 0;;) {
410 /*
411 * Find the first changed line in the to-be-compacted file.
412 * We need to keep track of both indexes, so if we find a
413 * changed lines group on the other file, while scanning the
414 * to-be-compacted file, we need to skip it properly. Note
415 * that loops that are testing for changed lines on rchg* do
416 * not need index bounding since the array is prepared with
417 * a zero at position -1 and N.
418 */
419 for (; ix < nrec && !rchg[ix]; ix++)
420 while (rchgo[ixo++]);
421 if (ix == nrec)
422 break;
423 /*
425 * Record the start of a changed-group in the to-be-compacted file
426 * and find the end of it, on both to-be-compacted and other file
427 * indexes (ix and ixo).
428 */
429 ixs = ix;
430 for (ix++; rchg[ix]; ix++);
431 for (; rchgo[ixo]; ixo++);
432 do {
434 grpsiz = ix - ixs;
435 /*
437 * If the line before the current change group, is equal to
438 * the last line of the current change group, shift backward
439 * the group.
440 */
441 while (ixs > 0 && recs[ixs - 1]->ha == recs[ix - 1]->ha &&
442 xdl_recmatch(recs[ixs - 1]->ptr, recs[ixs - 1]->size, recs[ix - 1]->ptr, recs[ix - 1]->size, flags)) {
443 rchg[--ixs] = 1;
444 rchg[--ix] = 0;
445 /*
447 * This change might have joined two change groups,
448 * so we try to take this scenario in account by moving
449 * the start index accordingly (and so the other-file
450 * end-of-group index).
451 */
452 for (; rchg[ixs - 1]; ixs--);
453 while (rchgo[--ixo]);
454 }
455 /*
457 * Record the end-of-group position in case we are matched
458 * with a group of changes in the other file (that is, the
459 * change record before the enf-of-group index in the other
460 * file is set).
461 */
462 ixref = rchgo[ixo - 1] ? ix: nrec;
463 /*
465 * If the first line of the current change group, is equal to
466 * the line next of the current change group, shift forward
467 * the group.
468 */
469 while (ix < nrec && recs[ixs]->ha == recs[ix]->ha &&
470 xdl_recmatch(recs[ixs]->ptr, recs[ixs]->size, recs[ix]->ptr, recs[ix]->size, flags)) {
471 rchg[ixs++] = 0;
472 rchg[ix++] = 1;
473 /*
475 * This change might have joined two change groups,
476 * so we try to take this scenario in account by moving
477 * the start index accordingly (and so the other-file
478 * end-of-group index). Keep tracking the reference
479 * index in case we are shifting together with a
480 * corresponding group of changes in the other file.
481 */
482 for (; rchg[ix]; ix++);
483 while (rchgo[++ixo])
484 ixref = ix;
485 }
486 } while (grpsiz != ix - ixs);
487 /*
489 * Try to move back the possibly merged group of changes, to match
490 * the recorded postion in the other file.
491 */
492 while (ixref < ix) {
493 rchg[--ixs] = 1;
494 rchg[--ix] = 0;
495 while (rchgo[--ixo]);
496 }
497 }
498 return 0;
500}
501int xdl_build_script(xdfenv_t *xe, xdchange_t **xscr) {
504 xdchange_t *cscr = NULL, *xch;
505 char *rchg1 = xe->xdf1.rchg, *rchg2 = xe->xdf2.rchg;
506 long i1, i2, l1, l2;
507 /*
509 * Trivial. Collects "groups" of changes and creates an edit script.
510 */
511 for (i1 = xe->xdf1.nrec, i2 = xe->xdf2.nrec; i1 >= 0 || i2 >= 0; i1--, i2--)
512 if (rchg1[i1 - 1] || rchg2[i2 - 1]) {
513 for (l1 = i1; rchg1[i1 - 1]; i1--);
514 for (l2 = i2; rchg2[i2 - 1]; i2--);
515 if (!(xch = xdl_add_change(cscr, i1, i2, l1 - i1, l2 - i2))) {
517 xdl_free_script(cscr);
518 return -1;
519 }
520 cscr = xch;
521 }
522 *xscr = cscr;
524 return 0;
526}
527void xdl_free_script(xdchange_t *xscr) {
530 xdchange_t *xch;
531 while ((xch = xscr) != NULL) {
533 xscr = xscr->next;
534 xdl_free(xch);
535 }
536}
537int xdl_diff(mmfile_t *mf1, mmfile_t *mf2, xpparam_t const *xpp,
540 xdemitconf_t const *xecfg, xdemitcb_t *ecb) {
541 xdchange_t *xscr;
542 xdfenv_t xe;
543 if (xdl_do_diff(mf1, mf2, xpp, &xe) < 0) {
545 return -1;
547 }
548 if (xdl_change_compact(&xe.xdf1, &xe.xdf2, xpp->flags) < 0 ||
549 xdl_change_compact(&xe.xdf2, &xe.xdf1, xpp->flags) < 0 ||
550 xdl_build_script(&xe, &xscr) < 0) {
551 xdl_free_env(&xe);
553 return -1;
554 }
555 if (xscr) {
556 if (xdl_emit_diff(&xe, xscr, ecb, xecfg) < 0) {
557 xdl_free_script(xscr);
559 xdl_free_env(&xe);
560 return -1;
561 }
562 xdl_free_script(xscr);
563 }
564 xdl_free_env(&xe);
565 return 0;
567}