xdiff / xdiffi.con commit commit-reach: move walk methods from commit.c (5227c38)
   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
  25
  26
  27#define XDL_MAX_COST_MIN 256
  28#define XDL_HEUR_MIN_COST 256
  29#define XDL_LINE_MAX (long)((1UL << (CHAR_BIT * sizeof(long) - 1)) - 1)
  30#define XDL_SNAKE_CNT 20
  31#define XDL_K_HEUR 4
  32
  33
  34
  35typedef struct s_xdpsplit {
  36        long i1, i2;
  37        int min_lo, min_hi;
  38} xdpsplit_t;
  39
  40
  41
  42
  43static 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
  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 */
  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
  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.
  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
  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.
 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
 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
 222                        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
 234                        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
 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        }
 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 */
 268int 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 {
 296                xdpsplit_t spl;
 297                spl.i1 = spl.i2 = 0;
 298
 299                /*
 300                 * Divide ...
 301                 */
 302                if (xdl_split(ha1, off1, lim1, ha2, off2, lim2, kvdf, kvdb,
 303                              need_min, &spl, xenv) < 0) {
 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
 324int 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
 331        if (XDF_DIFF_ALG(xpp->flags) == XDF_PATIENCE_DIFF)
 332                return xdl_do_patience_diff(mf1, mf2, xpp, xe);
 333
 334        if (XDF_DIFF_ALG(xpp->flags) == XDF_HISTOGRAM_DIFF)
 335                return xdl_do_histogram_diff(mf1, mf2, xpp, xe);
 336
 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
 357        xenv.mxcost = xdl_bogosqrt(ndiags);
 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
 386static 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;
 397        xch->ignore = 0;
 398
 399        return xch;
 400}
 401
 402
 403static int recs_match(xrecord_t *rec1, xrecord_t *rec2, long flags)
 404{
 405        return (rec1->ha == rec2->ha &&
 406                xdl_recmatch(rec1->ptr, rec1->size,
 407                             rec2->ptr, rec2->size,
 408                             flags));
 409}
 410
 411/*
 412 * If a line is indented more than this, get_indent() just returns this value.
 413 * This avoids having to do absurd amounts of work for data that are not
 414 * human-readable text, and also ensures that the output of get_indent fits within
 415 * an int.
 416 */
 417#define MAX_INDENT 200
 418
 419/*
 420 * Return the amount of indentation of the specified line, treating TAB as 8
 421 * columns. Return -1 if line is empty or contains only whitespace. Clamp the
 422 * output value at MAX_INDENT.
 423 */
 424static int get_indent(xrecord_t *rec)
 425{
 426        long i;
 427        int ret = 0;
 428
 429        for (i = 0; i < rec->size; i++) {
 430                char c = rec->ptr[i];
 431
 432                if (!XDL_ISSPACE(c))
 433                        return ret;
 434                else if (c == ' ')
 435                        ret += 1;
 436                else if (c == '\t')
 437                        ret += 8 - ret % 8;
 438                /* ignore other whitespace characters */
 439
 440                if (ret >= MAX_INDENT)
 441                        return MAX_INDENT;
 442        }
 443
 444        /* The line contains only whitespace. */
 445        return -1;
 446}
 447
 448/*
 449 * If more than this number of consecutive blank rows are found, just return this
 450 * value. This avoids requiring O(N^2) work for pathological cases, and also
 451 * ensures that the output of score_split fits in an int.
 452 */
 453#define MAX_BLANKS 20
 454
 455/* Characteristics measured about a hypothetical split position. */
 456struct split_measurement {
 457        /*
 458         * Is the split at the end of the file (aside from any blank lines)?
 459         */
 460        int end_of_file;
 461
 462        /*
 463         * How much is the line immediately following the split indented (or -1 if
 464         * the line is blank):
 465         */
 466        int indent;
 467
 468        /*
 469         * How many consecutive lines above the split are blank?
 470         */
 471        int pre_blank;
 472
 473        /*
 474         * How much is the nearest non-blank line above the split indented (or -1
 475         * if there is no such line)?
 476         */
 477        int pre_indent;
 478
 479        /*
 480         * How many lines after the line following the split are blank?
 481         */
 482        int post_blank;
 483
 484        /*
 485         * How much is the nearest non-blank line after the line following the
 486         * split indented (or -1 if there is no such line)?
 487         */
 488        int post_indent;
 489};
 490
 491struct split_score {
 492        /* The effective indent of this split (smaller is preferred). */
 493        int effective_indent;
 494
 495        /* Penalty for this split (smaller is preferred). */
 496        int penalty;
 497};
 498
 499/*
 500 * Fill m with information about a hypothetical split of xdf above line split.
 501 */
 502static void measure_split(const xdfile_t *xdf, long split,
 503                          struct split_measurement *m)
 504{
 505        long i;
 506
 507        if (split >= xdf->nrec) {
 508                m->end_of_file = 1;
 509                m->indent = -1;
 510        } else {
 511                m->end_of_file = 0;
 512                m->indent = get_indent(xdf->recs[split]);
 513        }
 514
 515        m->pre_blank = 0;
 516        m->pre_indent = -1;
 517        for (i = split - 1; i >= 0; i--) {
 518                m->pre_indent = get_indent(xdf->recs[i]);
 519                if (m->pre_indent != -1)
 520                        break;
 521                m->pre_blank += 1;
 522                if (m->pre_blank == MAX_BLANKS) {
 523                        m->pre_indent = 0;
 524                        break;
 525                }
 526        }
 527
 528        m->post_blank = 0;
 529        m->post_indent = -1;
 530        for (i = split + 1; i < xdf->nrec; i++) {
 531                m->post_indent = get_indent(xdf->recs[i]);
 532                if (m->post_indent != -1)
 533                        break;
 534                m->post_blank += 1;
 535                if (m->post_blank == MAX_BLANKS) {
 536                        m->post_indent = 0;
 537                        break;
 538                }
 539        }
 540}
 541
 542/*
 543 * The empirically-determined weight factors used by score_split() below.
 544 * Larger values means that the position is a less favorable place to split.
 545 *
 546 * Note that scores are only ever compared against each other, so multiplying
 547 * all of these weight/penalty values by the same factor wouldn't change the
 548 * heuristic's behavior. Still, we need to set that arbitrary scale *somehow*.
 549 * In practice, these numbers are chosen to be large enough that they can be
 550 * adjusted relative to each other with sufficient precision despite using
 551 * integer math.
 552 */
 553
 554/* Penalty if there are no non-blank lines before the split */
 555#define START_OF_FILE_PENALTY 1
 556
 557/* Penalty if there are no non-blank lines after the split */
 558#define END_OF_FILE_PENALTY 21
 559
 560/* Multiplier for the number of blank lines around the split */
 561#define TOTAL_BLANK_WEIGHT (-30)
 562
 563/* Multiplier for the number of blank lines after the split */
 564#define POST_BLANK_WEIGHT 6
 565
 566/*
 567 * Penalties applied if the line is indented more than its predecessor
 568 */
 569#define RELATIVE_INDENT_PENALTY (-4)
 570#define RELATIVE_INDENT_WITH_BLANK_PENALTY 10
 571
 572/*
 573 * Penalties applied if the line is indented less than both its predecessor and
 574 * its successor
 575 */
 576#define RELATIVE_OUTDENT_PENALTY 24
 577#define RELATIVE_OUTDENT_WITH_BLANK_PENALTY 17
 578
 579/*
 580 * Penalties applied if the line is indented less than its predecessor but not
 581 * less than its successor
 582 */
 583#define RELATIVE_DEDENT_PENALTY 23
 584#define RELATIVE_DEDENT_WITH_BLANK_PENALTY 17
 585
 586/*
 587 * We only consider whether the sum of the effective indents for splits are
 588 * less than (-1), equal to (0), or greater than (+1) each other. The resulting
 589 * value is multiplied by the following weight and combined with the penalty to
 590 * determine the better of two scores.
 591 */
 592#define INDENT_WEIGHT 60
 593
 594/*
 595 * Compute a badness score for the hypothetical split whose measurements are
 596 * stored in m. The weight factors were determined empirically using the tools and
 597 * corpus described in
 598 *
 599 *     https://github.com/mhagger/diff-slider-tools
 600 *
 601 * Also see that project if you want to improve the weights based on, for example,
 602 * a larger or more diverse corpus.
 603 */
 604static void score_add_split(const struct split_measurement *m, struct split_score *s)
 605{
 606        /*
 607         * A place to accumulate penalty factors (positive makes this index more
 608         * favored):
 609         */
 610        int post_blank, total_blank, indent, any_blanks;
 611
 612        if (m->pre_indent == -1 && m->pre_blank == 0)
 613                s->penalty += START_OF_FILE_PENALTY;
 614
 615        if (m->end_of_file)
 616                s->penalty += END_OF_FILE_PENALTY;
 617
 618        /*
 619         * Set post_blank to the number of blank lines following the split,
 620         * including the line immediately after the split:
 621         */
 622        post_blank = (m->indent == -1) ? 1 + m->post_blank : 0;
 623        total_blank = m->pre_blank + post_blank;
 624
 625        /* Penalties based on nearby blank lines: */
 626        s->penalty += TOTAL_BLANK_WEIGHT * total_blank;
 627        s->penalty += POST_BLANK_WEIGHT * post_blank;
 628
 629        if (m->indent != -1)
 630                indent = m->indent;
 631        else
 632                indent = m->post_indent;
 633
 634        any_blanks = (total_blank != 0);
 635
 636        /* Note that the effective indent is -1 at the end of the file: */
 637        s->effective_indent += indent;
 638
 639        if (indent == -1) {
 640                /* No additional adjustments needed. */
 641        } else if (m->pre_indent == -1) {
 642                /* No additional adjustments needed. */
 643        } else if (indent > m->pre_indent) {
 644                /*
 645                 * The line is indented more than its predecessor.
 646                 */
 647                s->penalty += any_blanks ?
 648                        RELATIVE_INDENT_WITH_BLANK_PENALTY :
 649                        RELATIVE_INDENT_PENALTY;
 650        } else if (indent == m->pre_indent) {
 651                /*
 652                 * The line has the same indentation level as its predecessor.
 653                 * No additional adjustments needed.
 654                 */
 655        } else {
 656                /*
 657                 * The line is indented less than its predecessor. It could be
 658                 * the block terminator of the previous block, but it could
 659                 * also be the start of a new block (e.g., an "else" block, or
 660                 * maybe the previous block didn't have a block terminator).
 661                 * Try to distinguish those cases based on what comes next:
 662                 */
 663                if (m->post_indent != -1 && m->post_indent > indent) {
 664                        /*
 665                         * The following line is indented more. So it is likely
 666                         * that this line is the start of a block.
 667                         */
 668                        s->penalty += any_blanks ?
 669                                RELATIVE_OUTDENT_WITH_BLANK_PENALTY :
 670                                RELATIVE_OUTDENT_PENALTY;
 671                } else {
 672                        /*
 673                         * That was probably the end of a block.
 674                         */
 675                        s->penalty += any_blanks ?
 676                                RELATIVE_DEDENT_WITH_BLANK_PENALTY :
 677                                RELATIVE_DEDENT_PENALTY;
 678                }
 679        }
 680}
 681
 682static int score_cmp(struct split_score *s1, struct split_score *s2)
 683{
 684        /* -1 if s1.effective_indent < s2->effective_indent, etc. */
 685        int cmp_indents = ((s1->effective_indent > s2->effective_indent) -
 686                           (s1->effective_indent < s2->effective_indent));
 687
 688        return INDENT_WEIGHT * cmp_indents + (s1->penalty - s2->penalty);
 689}
 690
 691/*
 692 * Represent a group of changed lines in an xdfile_t (i.e., a contiguous group
 693 * of lines that was inserted or deleted from the corresponding version of the
 694 * file). We consider there to be such a group at the beginning of the file, at
 695 * the end of the file, and between any two unchanged lines, though most such
 696 * groups will usually be empty.
 697 *
 698 * If the first line in a group is equal to the line following the group, then
 699 * the group can be slid down. Similarly, if the last line in a group is equal
 700 * to the line preceding the group, then the group can be slid up. See
 701 * group_slide_down() and group_slide_up().
 702 *
 703 * Note that loops that are testing for changed lines in xdf->rchg do not need
 704 * index bounding since the array is prepared with a zero at position -1 and N.
 705 */
 706struct xdlgroup {
 707        /*
 708         * The index of the first changed line in the group, or the index of
 709         * the unchanged line above which the (empty) group is located.
 710         */
 711        long start;
 712
 713        /*
 714         * The index of the first unchanged line after the group. For an empty
 715         * group, end is equal to start.
 716         */
 717        long end;
 718};
 719
 720/*
 721 * Initialize g to point at the first group in xdf.
 722 */
 723static void group_init(xdfile_t *xdf, struct xdlgroup *g)
 724{
 725        g->start = g->end = 0;
 726        while (xdf->rchg[g->end])
 727                g->end++;
 728}
 729
 730/*
 731 * Move g to describe the next (possibly empty) group in xdf and return 0. If g
 732 * is already at the end of the file, do nothing and return -1.
 733 */
 734static inline int group_next(xdfile_t *xdf, struct xdlgroup *g)
 735{
 736        if (g->end == xdf->nrec)
 737                return -1;
 738
 739        g->start = g->end + 1;
 740        for (g->end = g->start; xdf->rchg[g->end]; g->end++)
 741                ;
 742
 743        return 0;
 744}
 745
 746/*
 747 * Move g to describe the previous (possibly empty) group in xdf and return 0.
 748 * If g is already at the beginning of the file, do nothing and return -1.
 749 */
 750static inline int group_previous(xdfile_t *xdf, struct xdlgroup *g)
 751{
 752        if (g->start == 0)
 753                return -1;
 754
 755        g->end = g->start - 1;
 756        for (g->start = g->end; xdf->rchg[g->start - 1]; g->start--)
 757                ;
 758
 759        return 0;
 760}
 761
 762/*
 763 * If g can be slid toward the end of the file, do so, and if it bumps into a
 764 * following group, expand this group to include it. Return 0 on success or -1
 765 * if g cannot be slid down.
 766 */
 767static int group_slide_down(xdfile_t *xdf, struct xdlgroup *g, long flags)
 768{
 769        if (g->end < xdf->nrec &&
 770            recs_match(xdf->recs[g->start], xdf->recs[g->end], flags)) {
 771                xdf->rchg[g->start++] = 0;
 772                xdf->rchg[g->end++] = 1;
 773
 774                while (xdf->rchg[g->end])
 775                        g->end++;
 776
 777                return 0;
 778        } else {
 779                return -1;
 780        }
 781}
 782
 783/*
 784 * If g can be slid toward the beginning of the file, do so, and if it bumps
 785 * into a previous group, expand this group to include it. Return 0 on success
 786 * or -1 if g cannot be slid up.
 787 */
 788static int group_slide_up(xdfile_t *xdf, struct xdlgroup *g, long flags)
 789{
 790        if (g->start > 0 &&
 791            recs_match(xdf->recs[g->start - 1], xdf->recs[g->end - 1], flags)) {
 792                xdf->rchg[--g->start] = 1;
 793                xdf->rchg[--g->end] = 0;
 794
 795                while (xdf->rchg[g->start - 1])
 796                        g->start--;
 797
 798                return 0;
 799        } else {
 800                return -1;
 801        }
 802}
 803
 804static void xdl_bug(const char *msg)
 805{
 806        fprintf(stderr, "BUG: %s\n", msg);
 807        exit(1);
 808}
 809
 810/*
 811 * Move back and forward change groups for a consistent and pretty diff output.
 812 * This also helps in finding joinable change groups and reducing the diff
 813 * size.
 814 */
 815int xdl_change_compact(xdfile_t *xdf, xdfile_t *xdfo, long flags) {
 816        struct xdlgroup g, go;
 817        long earliest_end, end_matching_other;
 818        long groupsize;
 819
 820        group_init(xdf, &g);
 821        group_init(xdfo, &go);
 822
 823        while (1) {
 824                /* If the group is empty in the to-be-compacted file, skip it: */
 825                if (g.end == g.start)
 826                        goto next;
 827
 828                /*
 829                 * Now shift the change up and then down as far as possible in
 830                 * each direction. If it bumps into any other changes, merge them.
 831                 */
 832                do {
 833                        groupsize = g.end - g.start;
 834
 835                        /*
 836                         * Keep track of the last "end" index that causes this
 837                         * group to align with a group of changed lines in the
 838                         * other file. -1 indicates that we haven't found such
 839                         * a match yet:
 840                         */
 841                        end_matching_other = -1;
 842
 843                        /* Shift the group backward as much as possible: */
 844                        while (!group_slide_up(xdf, &g, flags))
 845                                if (group_previous(xdfo, &go))
 846                                        xdl_bug("group sync broken sliding up");
 847
 848                        /*
 849                         * This is this highest that this group can be shifted.
 850                         * Record its end index:
 851                         */
 852                        earliest_end = g.end;
 853
 854                        if (go.end > go.start)
 855                                end_matching_other = g.end;
 856
 857                        /* Now shift the group forward as far as possible: */
 858                        while (1) {
 859                                if (group_slide_down(xdf, &g, flags))
 860                                        break;
 861                                if (group_next(xdfo, &go))
 862                                        xdl_bug("group sync broken sliding down");
 863
 864                                if (go.end > go.start)
 865                                        end_matching_other = g.end;
 866                        }
 867                } while (groupsize != g.end - g.start);
 868
 869                /*
 870                 * If the group can be shifted, then we can possibly use this
 871                 * freedom to produce a more intuitive diff.
 872                 *
 873                 * The group is currently shifted as far down as possible, so the
 874                 * heuristics below only have to handle upwards shifts.
 875                 */
 876
 877                if (g.end == earliest_end) {
 878                        /* no shifting was possible */
 879                } else if (end_matching_other != -1) {
 880                        /*
 881                         * Move the possibly merged group of changes back to line
 882                         * up with the last group of changes from the other file
 883                         * that it can align with.
 884                         */
 885                        while (go.end == go.start) {
 886                                if (group_slide_up(xdf, &g, flags))
 887                                        xdl_bug("match disappeared");
 888                                if (group_previous(xdfo, &go))
 889                                        xdl_bug("group sync broken sliding to match");
 890                        }
 891                } else if (flags & XDF_INDENT_HEURISTIC) {
 892                        /*
 893                         * Indent heuristic: a group of pure add/delete lines
 894                         * implies two splits, one between the end of the "before"
 895                         * context and the start of the group, and another between
 896                         * the end of the group and the beginning of the "after"
 897                         * context. Some splits are aesthetically better and some
 898                         * are worse. We compute a badness "score" for each split,
 899                         * and add the scores for the two splits to define a
 900                         * "score" for each position that the group can be shifted
 901                         * to. Then we pick the shift with the lowest score.
 902                         */
 903                        long shift, best_shift = -1;
 904                        struct split_score best_score;
 905
 906                        for (shift = earliest_end; shift <= g.end; shift++) {
 907                                struct split_measurement m;
 908                                struct split_score score = {0, 0};
 909
 910                                measure_split(xdf, shift, &m);
 911                                score_add_split(&m, &score);
 912                                measure_split(xdf, shift - groupsize, &m);
 913                                score_add_split(&m, &score);
 914                                if (best_shift == -1 ||
 915                                    score_cmp(&score, &best_score) <= 0) {
 916                                        best_score.effective_indent = score.effective_indent;
 917                                        best_score.penalty = score.penalty;
 918                                        best_shift = shift;
 919                                }
 920                        }
 921
 922                        while (g.end > best_shift) {
 923                                if (group_slide_up(xdf, &g, flags))
 924                                        xdl_bug("best shift unreached");
 925                                if (group_previous(xdfo, &go))
 926                                        xdl_bug("group sync broken sliding to blank line");
 927                        }
 928                }
 929
 930        next:
 931                /* Move past the just-processed group: */
 932                if (group_next(xdf, &g))
 933                        break;
 934                if (group_next(xdfo, &go))
 935                        xdl_bug("group sync broken moving to next group");
 936        }
 937
 938        if (!group_next(xdfo, &go))
 939                xdl_bug("group sync broken at end of file");
 940
 941        return 0;
 942}
 943
 944
 945int xdl_build_script(xdfenv_t *xe, xdchange_t **xscr) {
 946        xdchange_t *cscr = NULL, *xch;
 947        char *rchg1 = xe->xdf1.rchg, *rchg2 = xe->xdf2.rchg;
 948        long i1, i2, l1, l2;
 949
 950        /*
 951         * Trivial. Collects "groups" of changes and creates an edit script.
 952         */
 953        for (i1 = xe->xdf1.nrec, i2 = xe->xdf2.nrec; i1 >= 0 || i2 >= 0; i1--, i2--)
 954                if (rchg1[i1 - 1] || rchg2[i2 - 1]) {
 955                        for (l1 = i1; rchg1[i1 - 1]; i1--);
 956                        for (l2 = i2; rchg2[i2 - 1]; i2--);
 957
 958                        if (!(xch = xdl_add_change(cscr, i1, i2, l1 - i1, l2 - i2))) {
 959                                xdl_free_script(cscr);
 960                                return -1;
 961                        }
 962                        cscr = xch;
 963                }
 964
 965        *xscr = cscr;
 966
 967        return 0;
 968}
 969
 970
 971void xdl_free_script(xdchange_t *xscr) {
 972        xdchange_t *xch;
 973
 974        while ((xch = xscr) != NULL) {
 975                xscr = xscr->next;
 976                xdl_free(xch);
 977        }
 978}
 979
 980static int xdl_call_hunk_func(xdfenv_t *xe, xdchange_t *xscr, xdemitcb_t *ecb,
 981                              xdemitconf_t const *xecfg)
 982{
 983        xdchange_t *xch, *xche;
 984
 985        for (xch = xscr; xch; xch = xche->next) {
 986                xche = xdl_get_hunk(&xch, xecfg);
 987                if (!xch)
 988                        break;
 989                if (xecfg->hunk_func(xch->i1, xche->i1 + xche->chg1 - xch->i1,
 990                                     xch->i2, xche->i2 + xche->chg2 - xch->i2,
 991                                     ecb->priv) < 0)
 992                        return -1;
 993        }
 994        return 0;
 995}
 996
 997static void xdl_mark_ignorable(xdchange_t *xscr, xdfenv_t *xe, long flags)
 998{
 999        xdchange_t *xch;
1000
1001        for (xch = xscr; xch; xch = xch->next) {
1002                int ignore = 1;
1003                xrecord_t **rec;
1004                long i;
1005
1006                rec = &xe->xdf1.recs[xch->i1];
1007                for (i = 0; i < xch->chg1 && ignore; i++)
1008                        ignore = xdl_blankline(rec[i]->ptr, rec[i]->size, flags);
1009
1010                rec = &xe->xdf2.recs[xch->i2];
1011                for (i = 0; i < xch->chg2 && ignore; i++)
1012                        ignore = xdl_blankline(rec[i]->ptr, rec[i]->size, flags);
1013
1014                xch->ignore = ignore;
1015        }
1016}
1017
1018int xdl_diff(mmfile_t *mf1, mmfile_t *mf2, xpparam_t const *xpp,
1019             xdemitconf_t const *xecfg, xdemitcb_t *ecb) {
1020        xdchange_t *xscr;
1021        xdfenv_t xe;
1022        emit_func_t ef = xecfg->hunk_func ? xdl_call_hunk_func : xdl_emit_diff;
1023
1024        if (xdl_do_diff(mf1, mf2, xpp, &xe) < 0) {
1025
1026                return -1;
1027        }
1028        if (xdl_change_compact(&xe.xdf1, &xe.xdf2, xpp->flags) < 0 ||
1029            xdl_change_compact(&xe.xdf2, &xe.xdf1, xpp->flags) < 0 ||
1030            xdl_build_script(&xe, &xscr) < 0) {
1031
1032                xdl_free_env(&xe);
1033                return -1;
1034        }
1035        if (xscr) {
1036                if (xpp->flags & XDF_IGNORE_BLANK_LINES)
1037                        xdl_mark_ignorable(xscr, &xe, xpp->flags);
1038
1039                if (ef(&xe, xscr, ecb, xecfg) < 0) {
1040
1041                        xdl_free_script(xscr);
1042                        xdl_free_env(&xe);
1043                        return -1;
1044                }
1045                xdl_free_script(xscr);
1046        }
1047        xdl_free_env(&xe);
1048
1049        return 0;
1050}