combine-diff.con commit Merge branch 'jk/diff-algo-finishing-touches' into maint (e4d1595)
   1#include "cache.h"
   2#include "commit.h"
   3#include "blob.h"
   4#include "diff.h"
   5#include "diffcore.h"
   6#include "quote.h"
   7#include "xdiff-interface.h"
   8#include "xdiff/xmacros.h"
   9#include "log-tree.h"
  10#include "refs.h"
  11#include "userdiff.h"
  12#include "sha1-array.h"
  13
  14static struct combine_diff_path *intersect_paths(struct combine_diff_path *curr, int n, int num_parent)
  15{
  16        struct diff_queue_struct *q = &diff_queued_diff;
  17        struct combine_diff_path *p;
  18        int i;
  19
  20        if (!n) {
  21                struct combine_diff_path *list = NULL, **tail = &list;
  22                for (i = 0; i < q->nr; i++) {
  23                        int len;
  24                        const char *path;
  25                        if (diff_unmodified_pair(q->queue[i]))
  26                                continue;
  27                        path = q->queue[i]->two->path;
  28                        len = strlen(path);
  29                        p = xmalloc(combine_diff_path_size(num_parent, len));
  30                        p->path = (char *) &(p->parent[num_parent]);
  31                        memcpy(p->path, path, len);
  32                        p->path[len] = 0;
  33                        p->len = len;
  34                        p->next = NULL;
  35                        memset(p->parent, 0,
  36                               sizeof(p->parent[0]) * num_parent);
  37
  38                        hashcpy(p->sha1, q->queue[i]->two->sha1);
  39                        p->mode = q->queue[i]->two->mode;
  40                        hashcpy(p->parent[n].sha1, q->queue[i]->one->sha1);
  41                        p->parent[n].mode = q->queue[i]->one->mode;
  42                        p->parent[n].status = q->queue[i]->status;
  43                        *tail = p;
  44                        tail = &p->next;
  45                }
  46                return list;
  47        }
  48
  49        for (p = curr; p; p = p->next) {
  50                int found = 0;
  51                if (!p->len)
  52                        continue;
  53                for (i = 0; i < q->nr; i++) {
  54                        const char *path;
  55                        int len;
  56
  57                        if (diff_unmodified_pair(q->queue[i]))
  58                                continue;
  59                        path = q->queue[i]->two->path;
  60                        len = strlen(path);
  61                        if (len == p->len && !memcmp(path, p->path, len)) {
  62                                found = 1;
  63                                hashcpy(p->parent[n].sha1, q->queue[i]->one->sha1);
  64                                p->parent[n].mode = q->queue[i]->one->mode;
  65                                p->parent[n].status = q->queue[i]->status;
  66                                break;
  67                        }
  68                }
  69                if (!found)
  70                        p->len = 0;
  71        }
  72        return curr;
  73}
  74
  75/* Lines lost from parent */
  76struct lline {
  77        struct lline *next;
  78        int len;
  79        unsigned long parent_map;
  80        char line[FLEX_ARRAY];
  81};
  82
  83/* Lines surviving in the merge result */
  84struct sline {
  85        struct lline *lost_head, **lost_tail;
  86        struct lline *next_lost;
  87        char *bol;
  88        int len;
  89        /* bit 0 up to (N-1) are on if the parent has this line (i.e.
  90         * we did not change it).
  91         * bit N is used for "interesting" lines, including context.
  92         * bit (N+1) is used for "do not show deletion before this".
  93         */
  94        unsigned long flag;
  95        unsigned long *p_lno;
  96};
  97
  98static char *grab_blob(const unsigned char *sha1, unsigned int mode,
  99                       unsigned long *size, struct userdiff_driver *textconv,
 100                       const char *path)
 101{
 102        char *blob;
 103        enum object_type type;
 104
 105        if (S_ISGITLINK(mode)) {
 106                blob = xmalloc(100);
 107                *size = snprintf(blob, 100,
 108                                 "Subproject commit %s\n", sha1_to_hex(sha1));
 109        } else if (is_null_sha1(sha1)) {
 110                /* deleted blob */
 111                *size = 0;
 112                return xcalloc(1, 1);
 113        } else if (textconv) {
 114                struct diff_filespec *df = alloc_filespec(path);
 115                fill_filespec(df, sha1, 1, mode);
 116                *size = fill_textconv(textconv, df, &blob);
 117                free_filespec(df);
 118        } else {
 119                blob = read_sha1_file(sha1, &type, size);
 120                if (type != OBJ_BLOB)
 121                        die("object '%s' is not a blob!", sha1_to_hex(sha1));
 122        }
 123        return blob;
 124}
 125
 126static int match_string_spaces(const char *line1, int len1,
 127                               const char *line2, int len2,
 128                               long flags)
 129{
 130        if (flags & XDF_WHITESPACE_FLAGS) {
 131                for (; len1 > 0 && XDL_ISSPACE(line1[len1 - 1]); len1--);
 132                for (; len2 > 0 && XDL_ISSPACE(line2[len2 - 1]); len2--);
 133        }
 134
 135        if (!(flags & (XDF_IGNORE_WHITESPACE | XDF_IGNORE_WHITESPACE_CHANGE)))
 136                return (len1 == len2 && !memcmp(line1, line2, len1));
 137
 138        while (len1 > 0 && len2 > 0) {
 139                len1--;
 140                len2--;
 141                if (XDL_ISSPACE(line1[len1]) || XDL_ISSPACE(line2[len2])) {
 142                        if ((flags & XDF_IGNORE_WHITESPACE_CHANGE) &&
 143                            (!XDL_ISSPACE(line1[len1]) || !XDL_ISSPACE(line2[len2])))
 144                                return 0;
 145
 146                        for (; len1 > 0 && XDL_ISSPACE(line1[len1]); len1--);
 147                        for (; len2 > 0 && XDL_ISSPACE(line2[len2]); len2--);
 148                }
 149                if (line1[len1] != line2[len2])
 150                        return 0;
 151        }
 152
 153        if (flags & XDF_IGNORE_WHITESPACE) {
 154                /* Consume remaining spaces */
 155                for (; len1 > 0 && XDL_ISSPACE(line1[len1 - 1]); len1--);
 156                for (; len2 > 0 && XDL_ISSPACE(line2[len2 - 1]); len2--);
 157        }
 158
 159        /* We matched full line1 and line2 */
 160        if (!len1 && !len2)
 161                return 1;
 162
 163        return 0;
 164}
 165
 166static void append_lost(struct sline *sline, int n, const char *line, int len, long flags)
 167{
 168        struct lline *lline;
 169        unsigned long this_mask = (1UL<<n);
 170        if (line[len-1] == '\n')
 171                len--;
 172
 173        /* Check to see if we can squash things */
 174        if (sline->lost_head) {
 175                lline = sline->next_lost;
 176                while (lline) {
 177                        if (match_string_spaces(lline->line, lline->len,
 178                                                line, len, flags)) {
 179                                lline->parent_map |= this_mask;
 180                                sline->next_lost = lline->next;
 181                                return;
 182                        }
 183                        lline = lline->next;
 184                }
 185        }
 186
 187        lline = xmalloc(sizeof(*lline) + len + 1);
 188        lline->len = len;
 189        lline->next = NULL;
 190        lline->parent_map = this_mask;
 191        memcpy(lline->line, line, len);
 192        lline->line[len] = 0;
 193        *sline->lost_tail = lline;
 194        sline->lost_tail = &lline->next;
 195        sline->next_lost = NULL;
 196}
 197
 198struct combine_diff_state {
 199        unsigned int lno;
 200        int ob, on, nb, nn;
 201        unsigned long nmask;
 202        int num_parent;
 203        int n;
 204        struct sline *sline;
 205        struct sline *lost_bucket;
 206        long flags;
 207};
 208
 209static void consume_line(void *state_, char *line, unsigned long len)
 210{
 211        struct combine_diff_state *state = state_;
 212        if (5 < len && !memcmp("@@ -", line, 4)) {
 213                if (parse_hunk_header(line, len,
 214                                      &state->ob, &state->on,
 215                                      &state->nb, &state->nn))
 216                        return;
 217                state->lno = state->nb;
 218                if (state->nn == 0) {
 219                        /* @@ -X,Y +N,0 @@ removed Y lines
 220                         * that would have come *after* line N
 221                         * in the result.  Our lost buckets hang
 222                         * to the line after the removed lines,
 223                         *
 224                         * Note that this is correct even when N == 0,
 225                         * in which case the hunk removes the first
 226                         * line in the file.
 227                         */
 228                        state->lost_bucket = &state->sline[state->nb];
 229                        if (!state->nb)
 230                                state->nb = 1;
 231                } else {
 232                        state->lost_bucket = &state->sline[state->nb-1];
 233                }
 234                if (!state->sline[state->nb-1].p_lno)
 235                        state->sline[state->nb-1].p_lno =
 236                                xcalloc(state->num_parent,
 237                                        sizeof(unsigned long));
 238                state->sline[state->nb-1].p_lno[state->n] = state->ob;
 239                state->lost_bucket->next_lost = state->lost_bucket->lost_head;
 240                return;
 241        }
 242        if (!state->lost_bucket)
 243                return; /* not in any hunk yet */
 244        switch (line[0]) {
 245        case '-':
 246                append_lost(state->lost_bucket, state->n, line+1, len-1, state->flags);
 247                break;
 248        case '+':
 249                state->sline[state->lno-1].flag |= state->nmask;
 250                state->lno++;
 251                break;
 252        }
 253}
 254
 255static void combine_diff(const unsigned char *parent, unsigned int mode,
 256                         mmfile_t *result_file,
 257                         struct sline *sline, unsigned int cnt, int n,
 258                         int num_parent, int result_deleted,
 259                         struct userdiff_driver *textconv,
 260                         const char *path, long flags)
 261{
 262        unsigned int p_lno, lno;
 263        unsigned long nmask = (1UL << n);
 264        xpparam_t xpp;
 265        xdemitconf_t xecfg;
 266        mmfile_t parent_file;
 267        struct combine_diff_state state;
 268        unsigned long sz;
 269
 270        if (result_deleted)
 271                return; /* result deleted */
 272
 273        parent_file.ptr = grab_blob(parent, mode, &sz, textconv, path);
 274        parent_file.size = sz;
 275        memset(&xpp, 0, sizeof(xpp));
 276        xpp.flags = flags;
 277        memset(&xecfg, 0, sizeof(xecfg));
 278        memset(&state, 0, sizeof(state));
 279        state.flags = flags;
 280        state.nmask = nmask;
 281        state.sline = sline;
 282        state.lno = 1;
 283        state.num_parent = num_parent;
 284        state.n = n;
 285
 286        xdi_diff_outf(&parent_file, result_file, consume_line, &state,
 287                      &xpp, &xecfg);
 288        free(parent_file.ptr);
 289
 290        /* Assign line numbers for this parent.
 291         *
 292         * sline[lno].p_lno[n] records the first line number
 293         * (counting from 1) for parent N if the final hunk display
 294         * started by showing sline[lno] (possibly showing the lost
 295         * lines attached to it first).
 296         */
 297        for (lno = 0,  p_lno = 1; lno <= cnt; lno++) {
 298                struct lline *ll;
 299                sline[lno].p_lno[n] = p_lno;
 300
 301                /* How many lines would this sline advance the p_lno? */
 302                ll = sline[lno].lost_head;
 303                while (ll) {
 304                        if (ll->parent_map & nmask)
 305                                p_lno++; /* '-' means parent had it */
 306                        ll = ll->next;
 307                }
 308                if (lno < cnt && !(sline[lno].flag & nmask))
 309                        p_lno++; /* no '+' means parent had it */
 310        }
 311        sline[lno].p_lno[n] = p_lno; /* trailer */
 312}
 313
 314static unsigned long context = 3;
 315static char combine_marker = '@';
 316
 317static int interesting(struct sline *sline, unsigned long all_mask)
 318{
 319        /* If some parents lost lines here, or if we have added to
 320         * some parent, it is interesting.
 321         */
 322        return ((sline->flag & all_mask) || sline->lost_head);
 323}
 324
 325static unsigned long adjust_hunk_tail(struct sline *sline,
 326                                      unsigned long all_mask,
 327                                      unsigned long hunk_begin,
 328                                      unsigned long i)
 329{
 330        /* i points at the first uninteresting line.  If the last line
 331         * of the hunk was interesting only because it has some
 332         * deletion, then it is not all that interesting for the
 333         * purpose of giving trailing context lines.  This is because
 334         * we output '-' line and then unmodified sline[i-1] itself in
 335         * that case which gives us one extra context line.
 336         */
 337        if ((hunk_begin + 1 <= i) && !(sline[i-1].flag & all_mask))
 338                i--;
 339        return i;
 340}
 341
 342static unsigned long find_next(struct sline *sline,
 343                               unsigned long mark,
 344                               unsigned long i,
 345                               unsigned long cnt,
 346                               int look_for_uninteresting)
 347{
 348        /* We have examined up to i-1 and are about to look at i.
 349         * Find next interesting or uninteresting line.  Here,
 350         * "interesting" does not mean interesting(), but marked by
 351         * the give_context() function below (i.e. it includes context
 352         * lines that are not interesting to interesting() function
 353         * that are surrounded by interesting() ones.
 354         */
 355        while (i <= cnt)
 356                if (look_for_uninteresting
 357                    ? !(sline[i].flag & mark)
 358                    : (sline[i].flag & mark))
 359                        return i;
 360                else
 361                        i++;
 362        return i;
 363}
 364
 365static int give_context(struct sline *sline, unsigned long cnt, int num_parent)
 366{
 367        unsigned long all_mask = (1UL<<num_parent) - 1;
 368        unsigned long mark = (1UL<<num_parent);
 369        unsigned long no_pre_delete = (2UL<<num_parent);
 370        unsigned long i;
 371
 372        /* Two groups of interesting lines may have a short gap of
 373         * uninteresting lines.  Connect such groups to give them a
 374         * bit of context.
 375         *
 376         * We first start from what the interesting() function says,
 377         * and mark them with "mark", and paint context lines with the
 378         * mark.  So interesting() would still say false for such context
 379         * lines but they are treated as "interesting" in the end.
 380         */
 381        i = find_next(sline, mark, 0, cnt, 0);
 382        if (cnt < i)
 383                return 0;
 384
 385        while (i <= cnt) {
 386                unsigned long j = (context < i) ? (i - context) : 0;
 387                unsigned long k;
 388
 389                /* Paint a few lines before the first interesting line. */
 390                while (j < i)
 391                        sline[j++].flag |= mark | no_pre_delete;
 392
 393        again:
 394                /* we know up to i is to be included.  where does the
 395                 * next uninteresting one start?
 396                 */
 397                j = find_next(sline, mark, i, cnt, 1);
 398                if (cnt < j)
 399                        break; /* the rest are all interesting */
 400
 401                /* lookahead context lines */
 402                k = find_next(sline, mark, j, cnt, 0);
 403                j = adjust_hunk_tail(sline, all_mask, i, j);
 404
 405                if (k < j + context) {
 406                        /* k is interesting and [j,k) are not, but
 407                         * paint them interesting because the gap is small.
 408                         */
 409                        while (j < k)
 410                                sline[j++].flag |= mark;
 411                        i = k;
 412                        goto again;
 413                }
 414
 415                /* j is the first uninteresting line and there is
 416                 * no overlap beyond it within context lines.  Paint
 417                 * the trailing edge a bit.
 418                 */
 419                i = k;
 420                k = (j + context < cnt+1) ? j + context : cnt+1;
 421                while (j < k)
 422                        sline[j++].flag |= mark;
 423        }
 424        return 1;
 425}
 426
 427static int make_hunks(struct sline *sline, unsigned long cnt,
 428                       int num_parent, int dense)
 429{
 430        unsigned long all_mask = (1UL<<num_parent) - 1;
 431        unsigned long mark = (1UL<<num_parent);
 432        unsigned long i;
 433        int has_interesting = 0;
 434
 435        for (i = 0; i <= cnt; i++) {
 436                if (interesting(&sline[i], all_mask))
 437                        sline[i].flag |= mark;
 438                else
 439                        sline[i].flag &= ~mark;
 440        }
 441        if (!dense)
 442                return give_context(sline, cnt, num_parent);
 443
 444        /* Look at each hunk, and if we have changes from only one
 445         * parent, or the changes are the same from all but one
 446         * parent, mark that uninteresting.
 447         */
 448        i = 0;
 449        while (i <= cnt) {
 450                unsigned long j, hunk_begin, hunk_end;
 451                unsigned long same_diff;
 452                while (i <= cnt && !(sline[i].flag & mark))
 453                        i++;
 454                if (cnt < i)
 455                        break; /* No more interesting hunks */
 456                hunk_begin = i;
 457                for (j = i + 1; j <= cnt; j++) {
 458                        if (!(sline[j].flag & mark)) {
 459                                /* Look beyond the end to see if there
 460                                 * is an interesting line after this
 461                                 * hunk within context span.
 462                                 */
 463                                unsigned long la; /* lookahead */
 464                                int contin = 0;
 465                                la = adjust_hunk_tail(sline, all_mask,
 466                                                     hunk_begin, j);
 467                                la = (la + context < cnt + 1) ?
 468                                        (la + context) : cnt + 1;
 469                                while (la && j <= --la) {
 470                                        if (sline[la].flag & mark) {
 471                                                contin = 1;
 472                                                break;
 473                                        }
 474                                }
 475                                if (!contin)
 476                                        break;
 477                                j = la;
 478                        }
 479                }
 480                hunk_end = j;
 481
 482                /* [i..hunk_end) are interesting.  Now is it really
 483                 * interesting?  We check if there are only two versions
 484                 * and the result matches one of them.  That is, we look
 485                 * at:
 486                 *   (+) line, which records lines added to which parents;
 487                 *       this line appears in the result.
 488                 *   (-) line, which records from what parents the line
 489                 *       was removed; this line does not appear in the result.
 490                 * then check the set of parents the result has difference
 491                 * from, from all lines.  If there are lines that has
 492                 * different set of parents that the result has differences
 493                 * from, that means we have more than two versions.
 494                 *
 495                 * Even when we have only two versions, if the result does
 496                 * not match any of the parents, the it should be considered
 497                 * interesting.  In such a case, we would have all '+' line.
 498                 * After passing the above "two versions" test, that would
 499                 * appear as "the same set of parents" to be "all parents".
 500                 */
 501                same_diff = 0;
 502                has_interesting = 0;
 503                for (j = i; j < hunk_end && !has_interesting; j++) {
 504                        unsigned long this_diff = sline[j].flag & all_mask;
 505                        struct lline *ll = sline[j].lost_head;
 506                        if (this_diff) {
 507                                /* This has some changes.  Is it the
 508                                 * same as others?
 509                                 */
 510                                if (!same_diff)
 511                                        same_diff = this_diff;
 512                                else if (same_diff != this_diff) {
 513                                        has_interesting = 1;
 514                                        break;
 515                                }
 516                        }
 517                        while (ll && !has_interesting) {
 518                                /* Lost this line from these parents;
 519                                 * who are they?  Are they the same?
 520                                 */
 521                                this_diff = ll->parent_map;
 522                                if (!same_diff)
 523                                        same_diff = this_diff;
 524                                else if (same_diff != this_diff) {
 525                                        has_interesting = 1;
 526                                }
 527                                ll = ll->next;
 528                        }
 529                }
 530
 531                if (!has_interesting && same_diff != all_mask) {
 532                        /* This hunk is not that interesting after all */
 533                        for (j = hunk_begin; j < hunk_end; j++)
 534                                sline[j].flag &= ~mark;
 535                }
 536                i = hunk_end;
 537        }
 538
 539        has_interesting = give_context(sline, cnt, num_parent);
 540        return has_interesting;
 541}
 542
 543static void show_parent_lno(struct sline *sline, unsigned long l0, unsigned long l1, int n, unsigned long null_context)
 544{
 545        l0 = sline[l0].p_lno[n];
 546        l1 = sline[l1].p_lno[n];
 547        printf(" -%lu,%lu", l0, l1-l0-null_context);
 548}
 549
 550static int hunk_comment_line(const char *bol)
 551{
 552        int ch;
 553
 554        if (!bol)
 555                return 0;
 556        ch = *bol & 0xff;
 557        return (isalpha(ch) || ch == '_' || ch == '$');
 558}
 559
 560static void show_line_to_eol(const char *line, int len, const char *reset)
 561{
 562        int saw_cr_at_eol = 0;
 563        if (len < 0)
 564                len = strlen(line);
 565        saw_cr_at_eol = (len && line[len-1] == '\r');
 566
 567        printf("%.*s%s%s\n", len - saw_cr_at_eol, line,
 568               reset,
 569               saw_cr_at_eol ? "\r" : "");
 570}
 571
 572static void dump_sline(struct sline *sline, const char *line_prefix,
 573                       unsigned long cnt, int num_parent,
 574                       int use_color, int result_deleted)
 575{
 576        unsigned long mark = (1UL<<num_parent);
 577        unsigned long no_pre_delete = (2UL<<num_parent);
 578        int i;
 579        unsigned long lno = 0;
 580        const char *c_frag = diff_get_color(use_color, DIFF_FRAGINFO);
 581        const char *c_func = diff_get_color(use_color, DIFF_FUNCINFO);
 582        const char *c_new = diff_get_color(use_color, DIFF_FILE_NEW);
 583        const char *c_old = diff_get_color(use_color, DIFF_FILE_OLD);
 584        const char *c_plain = diff_get_color(use_color, DIFF_PLAIN);
 585        const char *c_reset = diff_get_color(use_color, DIFF_RESET);
 586
 587        if (result_deleted)
 588                return; /* result deleted */
 589
 590        while (1) {
 591                unsigned long hunk_end;
 592                unsigned long rlines;
 593                const char *hunk_comment = NULL;
 594                unsigned long null_context = 0;
 595
 596                while (lno <= cnt && !(sline[lno].flag & mark)) {
 597                        if (hunk_comment_line(sline[lno].bol))
 598                                hunk_comment = sline[lno].bol;
 599                        lno++;
 600                }
 601                if (cnt < lno)
 602                        break;
 603                else {
 604                        for (hunk_end = lno + 1; hunk_end <= cnt; hunk_end++)
 605                                if (!(sline[hunk_end].flag & mark))
 606                                        break;
 607                }
 608                rlines = hunk_end - lno;
 609                if (cnt < hunk_end)
 610                        rlines--; /* pointing at the last delete hunk */
 611
 612                if (!context) {
 613                        /*
 614                         * Even when running with --unified=0, all
 615                         * lines in the hunk needs to be processed in
 616                         * the loop below in order to show the
 617                         * deletion recorded in lost_head.  However,
 618                         * we do not want to show the resulting line
 619                         * with all blank context markers in such a
 620                         * case.  Compensate.
 621                         */
 622                        unsigned long j;
 623                        for (j = lno; j < hunk_end; j++)
 624                                if (!(sline[j].flag & (mark-1)))
 625                                        null_context++;
 626                        rlines -= null_context;
 627                }
 628
 629                printf("%s%s", line_prefix, c_frag);
 630                for (i = 0; i <= num_parent; i++) putchar(combine_marker);
 631                for (i = 0; i < num_parent; i++)
 632                        show_parent_lno(sline, lno, hunk_end, i, null_context);
 633                printf(" +%lu,%lu ", lno+1, rlines);
 634                for (i = 0; i <= num_parent; i++) putchar(combine_marker);
 635
 636                if (hunk_comment) {
 637                        int comment_end = 0;
 638                        for (i = 0; i < 40; i++) {
 639                                int ch = hunk_comment[i] & 0xff;
 640                                if (!ch || ch == '\n')
 641                                        break;
 642                                if (!isspace(ch))
 643                                    comment_end = i;
 644                        }
 645                        if (comment_end)
 646                                printf("%s%s %s%s", c_reset,
 647                                                    c_plain, c_reset,
 648                                                    c_func);
 649                        for (i = 0; i < comment_end; i++)
 650                                putchar(hunk_comment[i]);
 651                }
 652
 653                printf("%s\n", c_reset);
 654                while (lno < hunk_end) {
 655                        struct lline *ll;
 656                        int j;
 657                        unsigned long p_mask;
 658                        struct sline *sl = &sline[lno++];
 659                        ll = (sl->flag & no_pre_delete) ? NULL : sl->lost_head;
 660                        while (ll) {
 661                                printf("%s%s", line_prefix, c_old);
 662                                for (j = 0; j < num_parent; j++) {
 663                                        if (ll->parent_map & (1UL<<j))
 664                                                putchar('-');
 665                                        else
 666                                                putchar(' ');
 667                                }
 668                                show_line_to_eol(ll->line, -1, c_reset);
 669                                ll = ll->next;
 670                        }
 671                        if (cnt < lno)
 672                                break;
 673                        p_mask = 1;
 674                        fputs(line_prefix, stdout);
 675                        if (!(sl->flag & (mark-1))) {
 676                                /*
 677                                 * This sline was here to hang the
 678                                 * lost lines in front of it.
 679                                 */
 680                                if (!context)
 681                                        continue;
 682                                fputs(c_plain, stdout);
 683                        }
 684                        else
 685                                fputs(c_new, stdout);
 686                        for (j = 0; j < num_parent; j++) {
 687                                if (p_mask & sl->flag)
 688                                        putchar('+');
 689                                else
 690                                        putchar(' ');
 691                                p_mask <<= 1;
 692                        }
 693                        show_line_to_eol(sl->bol, sl->len, c_reset);
 694                }
 695        }
 696}
 697
 698static void reuse_combine_diff(struct sline *sline, unsigned long cnt,
 699                               int i, int j)
 700{
 701        /* We have already examined parent j and we know parent i
 702         * and parent j are the same, so reuse the combined result
 703         * of parent j for parent i.
 704         */
 705        unsigned long lno, imask, jmask;
 706        imask = (1UL<<i);
 707        jmask = (1UL<<j);
 708
 709        for (lno = 0; lno <= cnt; lno++) {
 710                struct lline *ll = sline->lost_head;
 711                sline->p_lno[i] = sline->p_lno[j];
 712                while (ll) {
 713                        if (ll->parent_map & jmask)
 714                                ll->parent_map |= imask;
 715                        ll = ll->next;
 716                }
 717                if (sline->flag & jmask)
 718                        sline->flag |= imask;
 719                sline++;
 720        }
 721        /* the overall size of the file (sline[cnt]) */
 722        sline->p_lno[i] = sline->p_lno[j];
 723}
 724
 725static void dump_quoted_path(const char *head,
 726                             const char *prefix,
 727                             const char *path,
 728                             const char *line_prefix,
 729                             const char *c_meta, const char *c_reset)
 730{
 731        static struct strbuf buf = STRBUF_INIT;
 732
 733        strbuf_reset(&buf);
 734        strbuf_addstr(&buf, line_prefix);
 735        strbuf_addstr(&buf, c_meta);
 736        strbuf_addstr(&buf, head);
 737        quote_two_c_style(&buf, prefix, path, 0);
 738        strbuf_addstr(&buf, c_reset);
 739        puts(buf.buf);
 740}
 741
 742static void show_combined_header(struct combine_diff_path *elem,
 743                                 int num_parent,
 744                                 int dense,
 745                                 struct rev_info *rev,
 746                                 const char *line_prefix,
 747                                 int mode_differs,
 748                                 int show_file_header)
 749{
 750        struct diff_options *opt = &rev->diffopt;
 751        int abbrev = DIFF_OPT_TST(opt, FULL_INDEX) ? 40 : DEFAULT_ABBREV;
 752        const char *a_prefix = opt->a_prefix ? opt->a_prefix : "a/";
 753        const char *b_prefix = opt->b_prefix ? opt->b_prefix : "b/";
 754        const char *c_meta = diff_get_color_opt(opt, DIFF_METAINFO);
 755        const char *c_reset = diff_get_color_opt(opt, DIFF_RESET);
 756        const char *abb;
 757        int added = 0;
 758        int deleted = 0;
 759        int i;
 760
 761        if (rev->loginfo && !rev->no_commit_id)
 762                show_log(rev);
 763
 764        dump_quoted_path(dense ? "diff --cc " : "diff --combined ",
 765                         "", elem->path, line_prefix, c_meta, c_reset);
 766        printf("%s%sindex ", line_prefix, c_meta);
 767        for (i = 0; i < num_parent; i++) {
 768                abb = find_unique_abbrev(elem->parent[i].sha1,
 769                                         abbrev);
 770                printf("%s%s", i ? "," : "", abb);
 771        }
 772        abb = find_unique_abbrev(elem->sha1, abbrev);
 773        printf("..%s%s\n", abb, c_reset);
 774
 775        if (mode_differs) {
 776                deleted = !elem->mode;
 777
 778                /* We say it was added if nobody had it */
 779                added = !deleted;
 780                for (i = 0; added && i < num_parent; i++)
 781                        if (elem->parent[i].status !=
 782                            DIFF_STATUS_ADDED)
 783                                added = 0;
 784                if (added)
 785                        printf("%s%snew file mode %06o",
 786                               line_prefix, c_meta, elem->mode);
 787                else {
 788                        if (deleted)
 789                                printf("%s%sdeleted file ",
 790                                       line_prefix, c_meta);
 791                        printf("mode ");
 792                        for (i = 0; i < num_parent; i++) {
 793                                printf("%s%06o", i ? "," : "",
 794                                       elem->parent[i].mode);
 795                        }
 796                        if (elem->mode)
 797                                printf("..%06o", elem->mode);
 798                }
 799                printf("%s\n", c_reset);
 800        }
 801
 802        if (!show_file_header)
 803                return;
 804
 805        if (added)
 806                dump_quoted_path("--- ", "", "/dev/null",
 807                                 line_prefix, c_meta, c_reset);
 808        else
 809                dump_quoted_path("--- ", a_prefix, elem->path,
 810                                 line_prefix, c_meta, c_reset);
 811        if (deleted)
 812                dump_quoted_path("+++ ", "", "/dev/null",
 813                                 line_prefix, c_meta, c_reset);
 814        else
 815                dump_quoted_path("+++ ", b_prefix, elem->path,
 816                                 line_prefix, c_meta, c_reset);
 817}
 818
 819static void show_patch_diff(struct combine_diff_path *elem, int num_parent,
 820                            int dense, int working_tree_file,
 821                            struct rev_info *rev)
 822{
 823        struct diff_options *opt = &rev->diffopt;
 824        unsigned long result_size, cnt, lno;
 825        int result_deleted = 0;
 826        char *result, *cp;
 827        struct sline *sline; /* survived lines */
 828        int mode_differs = 0;
 829        int i, show_hunks;
 830        mmfile_t result_file;
 831        struct userdiff_driver *userdiff;
 832        struct userdiff_driver *textconv = NULL;
 833        int is_binary;
 834        const char *line_prefix = diff_line_prefix(opt);
 835
 836        context = opt->context;
 837        userdiff = userdiff_find_by_path(elem->path);
 838        if (!userdiff)
 839                userdiff = userdiff_find_by_name("default");
 840        if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV))
 841                textconv = userdiff_get_textconv(userdiff);
 842
 843        /* Read the result of merge first */
 844        if (!working_tree_file)
 845                result = grab_blob(elem->sha1, elem->mode, &result_size,
 846                                   textconv, elem->path);
 847        else {
 848                /* Used by diff-tree to read from the working tree */
 849                struct stat st;
 850                int fd = -1;
 851
 852                if (lstat(elem->path, &st) < 0)
 853                        goto deleted_file;
 854
 855                if (S_ISLNK(st.st_mode)) {
 856                        struct strbuf buf = STRBUF_INIT;
 857
 858                        if (strbuf_readlink(&buf, elem->path, st.st_size) < 0) {
 859                                error("readlink(%s): %s", elem->path,
 860                                      strerror(errno));
 861                                return;
 862                        }
 863                        result_size = buf.len;
 864                        result = strbuf_detach(&buf, NULL);
 865                        elem->mode = canon_mode(st.st_mode);
 866                } else if (S_ISDIR(st.st_mode)) {
 867                        unsigned char sha1[20];
 868                        if (resolve_gitlink_ref(elem->path, "HEAD", sha1) < 0)
 869                                result = grab_blob(elem->sha1, elem->mode,
 870                                                   &result_size, NULL, NULL);
 871                        else
 872                                result = grab_blob(sha1, elem->mode,
 873                                                   &result_size, NULL, NULL);
 874                } else if (textconv) {
 875                        struct diff_filespec *df = alloc_filespec(elem->path);
 876                        fill_filespec(df, null_sha1, 0, st.st_mode);
 877                        result_size = fill_textconv(textconv, df, &result);
 878                        free_filespec(df);
 879                } else if (0 <= (fd = open(elem->path, O_RDONLY))) {
 880                        size_t len = xsize_t(st.st_size);
 881                        ssize_t done;
 882                        int is_file, i;
 883
 884                        elem->mode = canon_mode(st.st_mode);
 885                        /* if symlinks don't work, assume symlink if all parents
 886                         * are symlinks
 887                         */
 888                        is_file = has_symlinks;
 889                        for (i = 0; !is_file && i < num_parent; i++)
 890                                is_file = !S_ISLNK(elem->parent[i].mode);
 891                        if (!is_file)
 892                                elem->mode = canon_mode(S_IFLNK);
 893
 894                        result_size = len;
 895                        result = xmalloc(len + 1);
 896
 897                        done = read_in_full(fd, result, len);
 898                        if (done < 0)
 899                                die_errno("read error '%s'", elem->path);
 900                        else if (done < len)
 901                                die("early EOF '%s'", elem->path);
 902
 903                        result[len] = 0;
 904
 905                        /* If not a fake symlink, apply filters, e.g. autocrlf */
 906                        if (is_file) {
 907                                struct strbuf buf = STRBUF_INIT;
 908
 909                                if (convert_to_git(elem->path, result, len, &buf, safe_crlf)) {
 910                                        free(result);
 911                                        result = strbuf_detach(&buf, &len);
 912                                        result_size = len;
 913                                }
 914                        }
 915                }
 916                else {
 917                deleted_file:
 918                        result_deleted = 1;
 919                        result_size = 0;
 920                        elem->mode = 0;
 921                        result = xcalloc(1, 1);
 922                }
 923
 924                if (0 <= fd)
 925                        close(fd);
 926        }
 927
 928        for (i = 0; i < num_parent; i++) {
 929                if (elem->parent[i].mode != elem->mode) {
 930                        mode_differs = 1;
 931                        break;
 932                }
 933        }
 934
 935        if (textconv)
 936                is_binary = 0;
 937        else if (userdiff->binary != -1)
 938                is_binary = userdiff->binary;
 939        else {
 940                is_binary = buffer_is_binary(result, result_size);
 941                for (i = 0; !is_binary && i < num_parent; i++) {
 942                        char *buf;
 943                        unsigned long size;
 944                        buf = grab_blob(elem->parent[i].sha1,
 945                                        elem->parent[i].mode,
 946                                        &size, NULL, NULL);
 947                        if (buffer_is_binary(buf, size))
 948                                is_binary = 1;
 949                        free(buf);
 950                }
 951        }
 952        if (is_binary) {
 953                show_combined_header(elem, num_parent, dense, rev,
 954                                     line_prefix, mode_differs, 0);
 955                printf("Binary files differ\n");
 956                free(result);
 957                return;
 958        }
 959
 960        for (cnt = 0, cp = result; cp < result + result_size; cp++) {
 961                if (*cp == '\n')
 962                        cnt++;
 963        }
 964        if (result_size && result[result_size-1] != '\n')
 965                cnt++; /* incomplete line */
 966
 967        sline = xcalloc(cnt+2, sizeof(*sline));
 968        sline[0].bol = result;
 969        for (lno = 0; lno <= cnt + 1; lno++) {
 970                sline[lno].lost_tail = &sline[lno].lost_head;
 971                sline[lno].flag = 0;
 972        }
 973        for (lno = 0, cp = result; cp < result + result_size; cp++) {
 974                if (*cp == '\n') {
 975                        sline[lno].len = cp - sline[lno].bol;
 976                        lno++;
 977                        if (lno < cnt)
 978                                sline[lno].bol = cp + 1;
 979                }
 980        }
 981        if (result_size && result[result_size-1] != '\n')
 982                sline[cnt-1].len = result_size - (sline[cnt-1].bol - result);
 983
 984        result_file.ptr = result;
 985        result_file.size = result_size;
 986
 987        /* Even p_lno[cnt+1] is valid -- that is for the end line number
 988         * for deletion hunk at the end.
 989         */
 990        sline[0].p_lno = xcalloc((cnt+2) * num_parent, sizeof(unsigned long));
 991        for (lno = 0; lno <= cnt; lno++)
 992                sline[lno+1].p_lno = sline[lno].p_lno + num_parent;
 993
 994        for (i = 0; i < num_parent; i++) {
 995                int j;
 996                for (j = 0; j < i; j++) {
 997                        if (!hashcmp(elem->parent[i].sha1,
 998                                     elem->parent[j].sha1)) {
 999                                reuse_combine_diff(sline, cnt, i, j);
1000                                break;
1001                        }
1002                }
1003                if (i <= j)
1004                        combine_diff(elem->parent[i].sha1,
1005                                     elem->parent[i].mode,
1006                                     &result_file, sline,
1007                                     cnt, i, num_parent, result_deleted,
1008                                     textconv, elem->path, opt->xdl_opts);
1009        }
1010
1011        show_hunks = make_hunks(sline, cnt, num_parent, dense);
1012
1013        if (show_hunks || mode_differs || working_tree_file) {
1014                show_combined_header(elem, num_parent, dense, rev,
1015                                     line_prefix, mode_differs, 1);
1016                dump_sline(sline, line_prefix, cnt, num_parent,
1017                           opt->use_color, result_deleted);
1018        }
1019        free(result);
1020
1021        for (lno = 0; lno < cnt; lno++) {
1022                if (sline[lno].lost_head) {
1023                        struct lline *ll = sline[lno].lost_head;
1024                        while (ll) {
1025                                struct lline *tmp = ll;
1026                                ll = ll->next;
1027                                free(tmp);
1028                        }
1029                }
1030        }
1031        free(sline[0].p_lno);
1032        free(sline);
1033}
1034
1035static void show_raw_diff(struct combine_diff_path *p, int num_parent, struct rev_info *rev)
1036{
1037        struct diff_options *opt = &rev->diffopt;
1038        int line_termination, inter_name_termination, i;
1039        const char *line_prefix = diff_line_prefix(opt);
1040
1041        line_termination = opt->line_termination;
1042        inter_name_termination = '\t';
1043        if (!line_termination)
1044                inter_name_termination = 0;
1045
1046        if (rev->loginfo && !rev->no_commit_id)
1047                show_log(rev);
1048
1049
1050        if (opt->output_format & DIFF_FORMAT_RAW) {
1051                printf("%s", line_prefix);
1052
1053                /* As many colons as there are parents */
1054                for (i = 0; i < num_parent; i++)
1055                        putchar(':');
1056
1057                /* Show the modes */
1058                for (i = 0; i < num_parent; i++)
1059                        printf("%06o ", p->parent[i].mode);
1060                printf("%06o", p->mode);
1061
1062                /* Show sha1's */
1063                for (i = 0; i < num_parent; i++)
1064                        printf(" %s", diff_unique_abbrev(p->parent[i].sha1,
1065                                                         opt->abbrev));
1066                printf(" %s ", diff_unique_abbrev(p->sha1, opt->abbrev));
1067        }
1068
1069        if (opt->output_format & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS)) {
1070                for (i = 0; i < num_parent; i++)
1071                        putchar(p->parent[i].status);
1072                putchar(inter_name_termination);
1073        }
1074
1075        write_name_quoted(p->path, stdout, line_termination);
1076}
1077
1078/*
1079 * The result (p->elem) is from the working tree and their
1080 * parents are typically from multiple stages during a merge
1081 * (i.e. diff-files) or the state in HEAD and in the index
1082 * (i.e. diff-index).
1083 */
1084void show_combined_diff(struct combine_diff_path *p,
1085                       int num_parent,
1086                       int dense,
1087                       struct rev_info *rev)
1088{
1089        struct diff_options *opt = &rev->diffopt;
1090
1091        if (!p->len)
1092                return;
1093        if (opt->output_format & (DIFF_FORMAT_RAW |
1094                                  DIFF_FORMAT_NAME |
1095                                  DIFF_FORMAT_NAME_STATUS))
1096                show_raw_diff(p, num_parent, rev);
1097        else if (opt->output_format & DIFF_FORMAT_PATCH)
1098                show_patch_diff(p, num_parent, dense, 1, rev);
1099}
1100
1101static void free_combined_pair(struct diff_filepair *pair)
1102{
1103        free(pair->two);
1104        free(pair);
1105}
1106
1107/*
1108 * A combine_diff_path expresses N parents on the LHS against 1 merge
1109 * result. Synthesize a diff_filepair that has N entries on the "one"
1110 * side and 1 entry on the "two" side.
1111 *
1112 * In the future, we might want to add more data to combine_diff_path
1113 * so that we can fill fields we are ignoring (most notably, size) here,
1114 * but currently nobody uses it, so this should suffice for now.
1115 */
1116static struct diff_filepair *combined_pair(struct combine_diff_path *p,
1117                                           int num_parent)
1118{
1119        int i;
1120        struct diff_filepair *pair;
1121        struct diff_filespec *pool;
1122
1123        pair = xmalloc(sizeof(*pair));
1124        pool = xcalloc(num_parent + 1, sizeof(struct diff_filespec));
1125        pair->one = pool + 1;
1126        pair->two = pool;
1127
1128        for (i = 0; i < num_parent; i++) {
1129                pair->one[i].path = p->path;
1130                pair->one[i].mode = p->parent[i].mode;
1131                hashcpy(pair->one[i].sha1, p->parent[i].sha1);
1132                pair->one[i].sha1_valid = !is_null_sha1(p->parent[i].sha1);
1133                pair->one[i].has_more_entries = 1;
1134        }
1135        pair->one[num_parent - 1].has_more_entries = 0;
1136
1137        pair->two->path = p->path;
1138        pair->two->mode = p->mode;
1139        hashcpy(pair->two->sha1, p->sha1);
1140        pair->two->sha1_valid = !is_null_sha1(p->sha1);
1141        return pair;
1142}
1143
1144static void handle_combined_callback(struct diff_options *opt,
1145                                     struct combine_diff_path *paths,
1146                                     int num_parent,
1147                                     int num_paths)
1148{
1149        struct combine_diff_path *p;
1150        struct diff_queue_struct q;
1151        int i;
1152
1153        q.queue = xcalloc(num_paths, sizeof(struct diff_filepair *));
1154        q.alloc = num_paths;
1155        q.nr = num_paths;
1156        for (i = 0, p = paths; p; p = p->next) {
1157                if (!p->len)
1158                        continue;
1159                q.queue[i++] = combined_pair(p, num_parent);
1160        }
1161        opt->format_callback(&q, opt, opt->format_callback_data);
1162        for (i = 0; i < num_paths; i++)
1163                free_combined_pair(q.queue[i]);
1164        free(q.queue);
1165}
1166
1167void diff_tree_combined(const unsigned char *sha1,
1168                        const struct sha1_array *parents,
1169                        int dense,
1170                        struct rev_info *rev)
1171{
1172        struct diff_options *opt = &rev->diffopt;
1173        struct diff_options diffopts;
1174        struct combine_diff_path *p, *paths = NULL;
1175        int i, num_paths, needsep, show_log_first, num_parent = parents->nr;
1176
1177        diffopts = *opt;
1178        diffopts.output_format = DIFF_FORMAT_NO_OUTPUT;
1179        DIFF_OPT_SET(&diffopts, RECURSIVE);
1180        DIFF_OPT_CLR(&diffopts, ALLOW_EXTERNAL);
1181
1182        show_log_first = !!rev->loginfo && !rev->no_commit_id;
1183        needsep = 0;
1184        /* find set of paths that everybody touches */
1185        for (i = 0; i < num_parent; i++) {
1186                /* show stat against the first parent even
1187                 * when doing combined diff.
1188                 */
1189                int stat_opt = (opt->output_format &
1190                                (DIFF_FORMAT_NUMSTAT|DIFF_FORMAT_DIFFSTAT));
1191                if (i == 0 && stat_opt)
1192                        diffopts.output_format = stat_opt;
1193                else
1194                        diffopts.output_format = DIFF_FORMAT_NO_OUTPUT;
1195                diff_tree_sha1(parents->sha1[i], sha1, "", &diffopts);
1196                diffcore_std(&diffopts);
1197                paths = intersect_paths(paths, i, num_parent);
1198
1199                if (show_log_first && i == 0) {
1200                        show_log(rev);
1201
1202                        if (rev->verbose_header && opt->output_format)
1203                                printf("%s%c", diff_line_prefix(opt),
1204                                       opt->line_termination);
1205                }
1206                diff_flush(&diffopts);
1207        }
1208
1209        /* find out surviving paths */
1210        for (num_paths = 0, p = paths; p; p = p->next) {
1211                if (p->len)
1212                        num_paths++;
1213        }
1214        if (num_paths) {
1215                if (opt->output_format & (DIFF_FORMAT_RAW |
1216                                          DIFF_FORMAT_NAME |
1217                                          DIFF_FORMAT_NAME_STATUS)) {
1218                        for (p = paths; p; p = p->next) {
1219                                if (p->len)
1220                                        show_raw_diff(p, num_parent, rev);
1221                        }
1222                        needsep = 1;
1223                }
1224                else if (opt->output_format &
1225                         (DIFF_FORMAT_NUMSTAT|DIFF_FORMAT_DIFFSTAT))
1226                        needsep = 1;
1227                else if (opt->output_format & DIFF_FORMAT_CALLBACK)
1228                        handle_combined_callback(opt, paths, num_parent, num_paths);
1229
1230                if (opt->output_format & DIFF_FORMAT_PATCH) {
1231                        if (needsep)
1232                                printf("%s%c", diff_line_prefix(opt),
1233                                       opt->line_termination);
1234                        for (p = paths; p; p = p->next) {
1235                                if (p->len)
1236                                        show_patch_diff(p, num_parent, dense,
1237                                                        0, rev);
1238                        }
1239                }
1240        }
1241
1242        /* Clean things up */
1243        while (paths) {
1244                struct combine_diff_path *tmp = paths;
1245                paths = paths->next;
1246                free(tmp);
1247        }
1248}
1249
1250void diff_tree_combined_merge(const struct commit *commit, int dense,
1251                              struct rev_info *rev)
1252{
1253        struct commit_list *parent = commit->parents;
1254        struct sha1_array parents = SHA1_ARRAY_INIT;
1255
1256        while (parent) {
1257                sha1_array_append(&parents, parent->item->object.sha1);
1258                parent = parent->next;
1259        }
1260        diff_tree_combined(commit->object.sha1, &parents, dense, rev);
1261        sha1_array_clear(&parents);
1262}