detect-renamed.diff

   1 This patch adds the --detect-renamed option which makes rsync notice files
   2 that either (1) match in size & modify-time (plus the basename, if possible)
   3 or (2) match in size & checksum (when --checksum was also specified) and use
   4 each match as an alternate basis file to speed up the transfer.
   5
   6 The algorithm attempts to scan the receiving-side's files in an efficient
   7 manner.  If --delete[-before] is enabled, we'll take advantage of the
   8 pre-transfer delete pass to prepare any alternate-basis-file matches we
   9 might find.  If --delete-before is not enabled, rsync does the rename scan
  10 during the regular file-sending scan (scanning each directory right before
  11 the generator starts updating files from that dir).  In this latter mode,
  12 rsync might delay the updating of a file (if no alternate-basis match was
  13 yet found) until the full scan of the receiving side is complete, at which
  14 point any delayed files are processed.
  15
  16 I chose to hard-link the alternate-basis files into a ".~tmp~" subdir that
  17 takes advantage of rsync's pre-existing partial-dir logic.  This uses less
  18 memory than trying to keep track of the matches internally, and also allows
  19 any deletions or file-updates to occur normally without interfering with
  20 these alternate-basis discoveries.
  21
  22 To use this patch, run these commands for a successful build:
  23
  24     patch -p1 <patches/detect-renamed.diff
  25     ./configure                                 (optional if already run)
  26     make
  27
  28 TODO:
  29
  30   We need to never return a match from fattr_find() that has a basis
  31   file.  This will ensure that we don't try to give a renamed file to
  32   a file that can't use it, while missing out on giving it to a file
  33   that could use it.
  34
  35 --- old/flist.c
  36 +++ new/flist.c
  37 @@ -53,6 +53,7 @@ extern int non_perishable_cnt;
  38  extern int prune_empty_dirs;
  39  extern int copy_links;
  40  extern int copy_unsafe_links;
  41 +extern int detect_renamed;
  42  extern int protocol_version;
  43  extern int sanitize_paths;
  44  extern struct stats stats;
  45 @@ -70,6 +71,8 @@ int checksum_len;
  46  dev_t filesystem_dev; /* used to implement -x */
  47  unsigned int file_struct_len;
  48
  49 +struct file_list the_fattr_list;
  50 +
  51  static char empty_sum[MD4_SUM_LENGTH];
  52  static int flist_count_offset;
  53
  54 @@ -252,6 +255,44 @@ static mode_t from_wire_mode(int mode)
  55         return mode;
  56  }
  57
  58 +static int fattr_compare(struct file_struct **file1, struct file_struct **file2)
  59 +{
  60 +       struct file_struct *f1 = *file1;
  61 +       struct file_struct *f2 = *file2;
  62 +       int diff;
  63 +
  64 +       if (!f1->basename || !S_ISREG(f1->mode) || !f1->length) {
  65 +               if (!f2->basename || !S_ISREG(f2->mode) || !f2->length)
  66 +                       return 0;
  67 +               return 1;
  68 +       }
  69 +       if (!f2->basename || !S_ISREG(f2->mode) || !f2->length)
  70 +               return -1;
  71 +
  72 +       /* Don't use diff for values that are longer than an int. */
  73 +       if (f1->length != f2->length)
  74 +               return f1->length < f2->length ? -1 : 1;
  75 +
  76 +       if (always_checksum) {
  77 +               diff = u_memcmp(f1->u.sum, f2->u.sum, checksum_len);
  78 +               if (diff)
  79 +                       return diff;
  80 +       } else if (f1->modtime != f2->modtime)
  81 +               return f1->modtime < f2->modtime ? -1 : 1;
  82 +
  83 +       diff = u_strcmp(f1->basename, f2->basename);
  84 +       if (diff)
  85 +               return diff;
  86 +
  87 +       if (f1->dirname == f2->dirname)
  88 +               return 0;
  89 +       if (!f1->dirname)
  90 +               return -1;
  91 +       if (!f2->dirname)
  92 +               return 1;
  93 +       return u_strcmp(f1->dirname, f2->dirname);
  94 +}
  95 +
  96  static void send_directory(int f, struct file_list *flist,
  97                            char *fbuf, int len);
  98
  99 @@ -1388,6 +1429,25 @@ struct file_list *recv_file_list(int f)
 100
 101         clean_flist(flist, relative_paths, 1);
 102
 103 +       if (detect_renamed) {
 104 +               int j = flist->count;
 105 +               the_fattr_list.count = j;
 106 +               the_fattr_list.files = new_array(struct file_struct *, j);
 107 +               if (!the_fattr_list.files)
 108 +                       goto oom;
 109 +               memcpy(the_fattr_list.files, flist->files,
 110 +                      j * sizeof (struct file_struct *));
 111 +               qsort(the_fattr_list.files, j,
 112 +                   sizeof the_fattr_list.files[0], (int (*)())fattr_compare);
 113 +               the_fattr_list.low = 0;
 114 +               while (j-- > 0) {
 115 +                       struct file_struct *fp = the_fattr_list.files[j];
 116 +                       if (fp->basename && S_ISREG(fp->mode) && fp->length)
 117 +                               break;
 118 +               }
 119 +               the_fattr_list.high = j;
 120 +       }
 121 +
 122         if (f >= 0) {
 123                 recv_uid_list(f, flist);
 124
 125 --- old/generator.c
 126 +++ new/generator.c
 127 @@ -76,6 +76,7 @@ extern char *basis_dir[];
 128  extern int compare_dest;
 129  extern int copy_dest;
 130  extern int link_dest;
 131 +extern int detect_renamed;
 132  extern int whole_file;
 133  extern int list_only;
 134  extern int new_root_dir;
 135 @@ -91,15 +92,18 @@ extern char *backup_dir;
 136  extern char *backup_suffix;
 137  extern int backup_suffix_len;
 138  extern struct file_list *the_file_list;
 139 +extern struct file_list the_fattr_list;
 140  extern struct filter_list_struct server_filter_list;
 141
 142  int ignore_perishable = 0;
 143  int non_perishable_cnt = 0;
 144
 145  static int deletion_count = 0; /* used to implement --max-delete */
 146 +static int unexplored_dirs = 1;
 147  static FILE *delete_delay_fp = NULL;
 148
 149 -/* For calling delete_item() and delete_dir_contents(). */
 150 +/* For calling delete_item(), delete_dir_contents(), and delete_in_dir(). */
 151 +#define DEL_NO_DELETIONS       (1<<0)
 152  #define DEL_RECURSE            (1<<1) /* recurse */
 153  #define DEL_DIR_IS_EMPTY       (1<<2) /* internal delete_FUNCTIONS use only */
 154
 155 @@ -121,11 +125,120 @@ static int is_backup_file(char *fn)
 156         return k > 0 && strcmp(fn+k, backup_suffix) == 0;
 157  }
 158
 159 +/* Search for a regular file that matches either (1) the size & modified
 160 + * time (plus the basename, if possible) or (2) the size & checksum.  If
 161 + * we find an exact match down to the dirname, return -1 because we found
 162 + * an up-to-date file in the transfer, not a renamed file. */
 163 +static int fattr_find(struct file_struct *f, char *fname, alloc_pool_t pool)
 164 +{
 165 +       int low = the_fattr_list.low, high = the_fattr_list.high;
 166 +       int mid, ok_match = -1, good_match = -1;
 167 +       struct file_struct *fmid;
 168 +       int diff;
 169 +
 170 +       while (low <= high) {
 171 +               mid = (low + high) / 2;
 172 +               fmid = the_fattr_list.files[mid];
 173 +               if (fmid->length != f->length) {
 174 +                       if (fmid->length < f->length)
 175 +                               low = mid + 1;
 176 +                       else
 177 +                               high = mid - 1;
 178 +                       continue;
 179 +               }
 180 +               if (always_checksum) {
 181 +                       if (!f->u.sum) {
 182 +                               if (fmid->modtime == f->modtime
 183 +                                && f_name_cmp(fmid, f) == 0)
 184 +                                       return -1; /* assume we can't help */
 185 +                               f->u.sum = pool_alloc(pool, MD4_SUM_LENGTH,
 186 +                                                     "fattr_find");
 187 +                               file_checksum(fname, f->u.sum, f->length);
 188 +                       }
 189 +                       diff = u_memcmp(fmid->u.sum, f->u.sum, checksum_len);
 190 +                       if (diff) {
 191 +                               if (diff < 0)
 192 +                                       low = mid + 1;
 193 +                               else
 194 +                                       high = mid - 1;
 195 +                               continue;
 196 +                       }
 197 +               } else {
 198 +                       if (fmid->modtime != f->modtime) {
 199 +                               if (fmid->modtime < f->modtime)
 200 +                                       low = mid + 1;
 201 +                               else
 202 +                                       high = mid - 1;
 203 +                               continue;
 204 +                       }
 205 +               }
 206 +               ok_match = mid;
 207 +               diff = u_strcmp(fmid->basename, f->basename);
 208 +               if (diff == 0) {
 209 +                       good_match = mid;
 210 +                       if (fmid->dirname == f->dirname)
 211 +                               return -1; /* file is up-to-date */
 212 +                       if (!fmid->dirname) {
 213 +                               low = mid + 1;
 214 +                               continue;
 215 +                       }
 216 +                       if (!f->dirname) {
 217 +                               high = mid - 1;
 218 +                               continue;
 219 +                       }
 220 +                       diff = u_strcmp(fmid->dirname, f->dirname);
 221 +                       if (diff == 0)
 222 +                               return -1; /* file is up-to-date */
 223 +               }
 224 +               if (diff < 0)
 225 +                       low = mid + 1;
 226 +               else
 227 +                       high = mid - 1;
 228 +       }
 229 +
 230 +       return good_match >= 0 ? good_match : ok_match;
 231 +}
 232 +
 233 +static void look_for_rename(struct file_struct *file, char *fname,
 234 +                           alloc_pool_t pool)
 235 +{
 236 +       struct file_struct *fp;
 237 +       char *partialptr, *fn;
 238 +       STRUCT_STAT st;
 239 +       int ndx;
 240 +
 241 +       if ((ndx = fattr_find(file, fname, pool)) < 0)
 242 +               return;
 243 +
 244 +       fp = the_fattr_list.files[ndx];
 245 +       fn = f_name(fp, NULL);
 246 +       /* We don't provide an alternate-basis file if there is a basis file. */
 247 +       if (link_stat(fn, &st, 0) == 0)
 248 +               return;
 249 +       if ((partialptr = partial_dir_fname(fn)) == NULL
 250 +        || !handle_partial_dir(partialptr, PDIR_CREATE))
 251 +               return;
 252 +
 253 +       /* We only use the file if we can hard-link it into our tmp dir. */
 254 +       if (link(fname, partialptr) == 0) {
 255 +               if (verbose > 2) {
 256 +                       rprintf(FINFO, "found renamed: %s => %s\n",
 257 +                               fname, partialptr);
 258 +               }
 259 +               return;
 260 +       }
 261 +
 262 +       if (errno != EEXIST)
 263 +               handle_partial_dir(partialptr, PDIR_DELETE);
 264 +}
 265 +
 266  /* Delete a file or directory.  If DEL_RECURSE is set in the flags, this will
 267   * delete recursively.
 268   *
 269   * Note that fbuf must point to a MAXPATHLEN buffer if the mode indicates it's
 270   * a directory! (The buffer is used for recursion, but returned unchanged.)
 271 + *
 272 + * Also note: --detect-rename may use this routine with DEL_NO_DELETIONS set!
 273   */
 274  static enum delret delete_item(char *fbuf, int mode, char *replace, int flags)
 275  {
 276 @@ -147,6 +260,8 @@ static enum delret delete_item(char *fbu
 277                         goto check_ret;
 278                 /* OK: try to delete the directory. */
 279         }
 280 +       if (flags & DEL_NO_DELETIONS)
 281 +               return DR_SUCCESS;
 282
 283         if (!replace && max_delete >= 0 && ++deletion_count > max_delete)
 284                 return DR_AT_LIMIT;
 285 @@ -193,6 +308,8 @@ static enum delret delete_item(char *fbu
 286   * its contents, otherwise just checks for content.  Returns DR_SUCCESS or
 287   * DR_NOT_EMPTY.  Note that fname must point to a MAXPATHLEN buffer!  (The
 288   * buffer is used for recursion, but returned unchanged.)
 289 + *
 290 + * Note: --detect-rename may use this routine with DEL_NO_DELETIONS set!
 291   */
 292  static enum delret delete_dir_contents(char *fname, int flags)
 293  {
 294 @@ -249,6 +366,8 @@ static enum delret delete_dir_contents(c
 295                 if (S_ISDIR(fp->mode)
 296                  && delete_dir_contents(fname, flags | DEL_RECURSE) != DR_SUCCESS)
 297                         ret = DR_NOT_EMPTY;
 298 +               if (detect_renamed && S_ISREG(fp->mode))
 299 +                       look_for_rename(fp, fname, dirlist->file_pool);
 300                 if (delete_item(fname, fp->mode, NULL, flags) != DR_SUCCESS)
 301                         ret = DR_NOT_EMPTY;
 302         }
 303 @@ -333,15 +452,19 @@ static void delayed_deletions(char *delb
 304   * all the --delete-WHEN options.  Note that the fbuf pointer must point to a
 305   * MAXPATHLEN buffer with the name of the directory in it (the functions we
 306   * call will append names onto the end, but the old dir value will be restored
 307 - * on exit). */
 308 + * on exit).
 309 + *
 310 + * Note:  --detect-rename may use this routine with DEL_NO_DELETIONS set!
 311 + */
 312  static void delete_in_dir(struct file_list *flist, char *fbuf,
 313 -                         struct file_struct *file, STRUCT_STAT *stp)
 314 +                         struct file_struct *file, STRUCT_STAT *stp, int flags)
 315  {
 316         static int min_depth = MAXPATHLEN, cur_depth = -1;
 317         static void *filt_array[MAXPATHLEN/2+1];
 318         static int already_warned = 0;
 319         struct file_list *dirlist;
 320 -       char delbuf[MAXPATHLEN];
 321 +       char *p, delbuf[MAXPATHLEN];
 322 +       unsigned remainder;
 323         int dlen, i;
 324
 325         if (!flist) {
 326 @@ -355,6 +478,8 @@ static void delete_in_dir(struct file_li
 327         if (verbose > 2)
 328                 rprintf(FINFO, "delete_in_dir(%s)\n", fbuf);
 329
 330 +       flags |= DEL_RECURSE;
 331 +
 332         if (allowed_lull)
 333                 maybe_send_keepalive();
 334
 335 @@ -362,12 +487,14 @@ static void delete_in_dir(struct file_li
 336                 return; /* Impossible... */
 337
 338         if (io_error && !(lp_ignore_errors(module_id) || ignore_errors)) {
 339 -               if (already_warned)
 340 +               if (!already_warned) {
 341 +                       rprintf(FINFO,
 342 +                           "IO error encountered -- skipping file deletion\n");
 343 +                       already_warned = 1;
 344 +               }
 345 +               if (!detect_renamed)
 346                         return;
 347 -               rprintf(FINFO,
 348 -                       "IO error encountered -- skipping file deletion\n");
 349 -               already_warned = 1;
 350 -               return;
 351 +               flags |= DEL_NO_DELETIONS;
 352         }
 353
 354         while (cur_depth >= file->dir.depth && cur_depth >= min_depth)
 355 @@ -378,6 +505,9 @@ static void delete_in_dir(struct file_li
 356         dlen = strlen(fbuf);
 357         filt_array[cur_depth] = push_local_filters(fbuf, dlen);
 358
 359 +       if (detect_renamed)
 360 +               unexplored_dirs--;
 361 +
 362         if (one_file_system) {
 363                 if (file->flags & FLAG_TOP_DIR)
 364                         filesystem_dev = stp->st_dev;
 365 @@ -387,6 +517,11 @@ static void delete_in_dir(struct file_li
 366
 367         dirlist = get_dirlist(fbuf, dlen, 0);
 368
 369 +       p = fbuf + dlen;
 370 +       if (dlen != 1 || *fbuf != '/')
 371 +               *p++ = '/';
 372 +       remainder = MAXPATHLEN - (p - fbuf);
 373 +
 374         /* If an item in dirlist is not found in flist, delete it
 375          * from the filesystem. */
 376         for (i = dirlist->count; i--; ) {
 377 @@ -399,15 +534,22 @@ static void delete_in_dir(struct file_li
 378                                         f_name(fp, NULL));
 379                         continue;
 380                 }
 381 +               if (detect_renamed && S_ISREG(fp->mode)) {
 382 +                       strlcpy(p, fp->basename, remainder);
 383 +                       look_for_rename(fp, fbuf, dirlist->file_pool);
 384 +               }
 385                 if (flist_find(flist, fp) < 0) {
 386                         f_name(fp, delbuf);
 387 -                       if (delete_delay_fp)
 388 +                       if (delete_delay_fp && !(flags & DEL_NO_DELETIONS))
 389                                 fprintf(delete_delay_fp, "%o %s%c", (short)fp->mode, delbuf, '\0');
 390                         else
 391 -                               delete_item(delbuf, fp->mode, NULL, DEL_RECURSE);
 392 -               }
 393 +                               delete_item(delbuf, fp->mode, NULL, flags);
 394 +               } else if (detect_renamed && S_ISDIR(fp->mode))
 395 +                       unexplored_dirs++;
 396         }
 397
 398 +       fbuf[dlen] = '\0';
 399 +
 400         flist_free(dirlist);
 401  }
 402
 403 @@ -437,9 +579,9 @@ static void do_delete_pass(struct file_l
 404                  || !S_ISDIR(st.st_mode))
 405                         continue;
 406
 407 -               delete_in_dir(flist, fbuf, file, &st);
 408 +               delete_in_dir(flist, fbuf, file, &st, 0);
 409         }
 410 -       delete_in_dir(NULL, NULL, NULL, NULL);
 411 +       delete_in_dir(NULL, NULL, NULL, NULL, 0);
 412
 413         if (do_progress && !am_server)
 414                 rprintf(FINFO, "                    \r");
 415 @@ -968,6 +1110,7 @@ static int try_dests_non(struct file_str
 416         return j;
 417  }
 418
 419 +static struct bitbag *delayed_bits = NULL;
 420  static int phase = 0;
 421
 422  /* Acts on the_file_list->file's ndx'th item, whose name is fname.  If a dir,
 423 @@ -1154,8 +1297,12 @@ static void recv_generator(char *fname,
 424                 if (real_ret != 0 && one_file_system)
 425                         real_st.st_dev = filesystem_dev;
 426                 if (delete_during && f_out != -1 && !phase && dry_run < 2
 427 -                   && (file->flags & FLAG_DEL_HERE))
 428 -                       delete_in_dir(the_file_list, fname, file, &real_st);
 429 +                   && (file->flags & FLAG_DEL_HERE)) {
 430 +                       if (detect_renamed && real_ret != 0)
 431 +                               unexplored_dirs++;
 432 +                       delete_in_dir(the_file_list, fname, file, &real_st,
 433 +                                     delete_during < 0 ? DEL_NO_DELETIONS : 0);
 434 +               }
 435                 return;
 436         }
 437
 438 @@ -1407,8 +1554,14 @@ static void recv_generator(char *fname,
 439                     && hard_link_check(file, ndx, fname, statret, &st,
 440                                        itemizing, code, HL_SKIP))
 441                         return;
 442 -               if (stat_errno == ENOENT)
 443 +               if (stat_errno == ENOENT) {
 444 +                       if (detect_renamed && unexplored_dirs > 0
 445 +                        && file->length) {
 446 +                               bitbag_set_bit(delayed_bits, ndx);
 447 +                               return;
 448 +                       }
 449                         goto notify_others;
 450 +               }
 451                 rsyserr(FERROR, stat_errno, "recv_generator: failed to stat %s",
 452                         full_fname(fname));
 453                 return;
 454 @@ -1594,13 +1747,19 @@ void generate_files(int f_out, struct fi
 455                         (long)getpid(), flist->count);
 456         }
 457
 458 +       if (detect_renamed) {
 459 +               delayed_bits = bitbag_create(flist->count);
 460 +               if (!delete_before && !delete_during)
 461 +                       delete_during = -1;
 462 +       }
 463 +
 464         if (delete_before && !local_name && flist->count > 0)
 465                 do_delete_pass(flist);
 466         if (delete_during == 2)
 467                 start_delete_temp();
 468         do_progress = 0;
 469
 470 -       if (append_mode || whole_file < 0)
 471 +       if (append_mode || detect_renamed || whole_file < 0)
 472                 whole_file = 0;
 473         if (verbose >= 2) {
 474                 rprintf(FINFO, "delta-transmission %s\n",
 475 @@ -1655,7 +1814,23 @@ void generate_files(int f_out, struct fi
 476         }
 477         recv_generator(NULL, NULL, 0, 0, 0, code, -1);
 478         if (delete_during)
 479 -               delete_in_dir(NULL, NULL, NULL, NULL);
 480 +               delete_in_dir(NULL, NULL, NULL, NULL, 0);
 481 +
 482 +       if (detect_renamed) {
 483 +               if (delete_during < 0)
 484 +                       delete_during = 0;
 485 +               detect_renamed = 0;
 486 +
 487 +               for (i = -1; (i = bitbag_next_bit(delayed_bits, i)) >= 0; ) {
 488 +                       struct file_struct *file = flist->files[i];
 489 +                       if (local_name)
 490 +                               strlcpy(fbuf, local_name, sizeof fbuf);
 491 +                       else
 492 +                               f_name(file, fbuf);
 493 +                       recv_generator(fbuf, file, i, itemizing,
 494 +                                      maybe_ATTRS_REPORT, code, f_out);
 495 +               }
 496 +       }
 497
 498         phase++;
 499         csum_length = SUM_LENGTH;
 500 --- old/options.c
 501 +++ new/options.c
 502 @@ -78,6 +78,7 @@ int am_generator = 0;
 503  int am_starting_up = 1;
 504  int relative_paths = -1;
 505  int implied_dirs = 1;
 506 +int detect_renamed = 0;
 507  int numeric_ids = 0;
 508  int allow_8bit_chars = 0;
 509  int force_delete = 0;
 510 @@ -347,6 +348,7 @@ void usage(enum logcode F)
 511    rprintf(F,"     --modify-window=NUM     compare mod-times with reduced accuracy\n");
 512    rprintf(F," -T, --temp-dir=DIR          create temporary files in directory DIR\n");
 513    rprintf(F," -y, --fuzzy                 find similar file for basis if no dest file\n");
 514 +  rprintf(F,"     --detect-renamed        try to find renamed files to speed up the transfer\n");
 515    rprintf(F,"     --compare-dest=DIR      also compare destination files relative to DIR\n");
 516    rprintf(F,"     --copy-dest=DIR         ... and include copies of unchanged files\n");
 517    rprintf(F,"     --link-dest=DIR         hardlink to files in DIR when unchanged\n");
 518 @@ -501,6 +503,7 @@ static struct poptOption long_options[]
 519    {"compare-dest",     0,  POPT_ARG_STRING, 0, OPT_COMPARE_DEST, 0, 0 },
 520    {"copy-dest",        0,  POPT_ARG_STRING, 0, OPT_COPY_DEST, 0, 0 },
 521    {"link-dest",        0,  POPT_ARG_STRING, 0, OPT_LINK_DEST, 0, 0 },
 522 +  {"detect-renamed",   0,  POPT_ARG_NONE,   &detect_renamed, 0, 0, 0 },
 523    {"fuzzy",           'y', POPT_ARG_NONE,   &fuzzy_basis, 0, 0, 0 },
 524    {"compress",        'z', POPT_ARG_NONE,   0, 'z', 0, 0 },
 525    {"compress-level",   0,  POPT_ARG_INT,    &def_compress_level, 'z', 0, 0 },
 526 @@ -1360,7 +1363,7 @@ int parse_arguments(int *argc, const cha
 527                 inplace = 1;
 528         }
 529
 530 -       if (delay_updates && !partial_dir)
 531 +       if ((delay_updates || detect_renamed) && !partial_dir)
 532                 partial_dir = tmp_partialdir;
 533
 534         if (inplace) {
 535 @@ -1369,6 +1372,7 @@ int parse_arguments(int *argc, const cha
 536                         snprintf(err_buf, sizeof err_buf,
 537                                  "--%s cannot be used with --%s\n",
 538                                  append_mode ? "append" : "inplace",
 539 +                                detect_renamed ? "detect-renamed" :
 540                                  delay_updates ? "delay-updates" : "partial-dir");
 541                         return 0;
 542                 }
 543 @@ -1679,6 +1683,8 @@ void server_options(char **args,int *arg
 544                         args[ac++] = "--super";
 545                 if (size_only)
 546                         args[ac++] = "--size-only";
 547 +               if (detect_renamed)
 548 +                       args[ac++] = "--detect-renamed";
 549         }
 550
 551         if (modify_window_set) {
 552 --- old/rsync.yo
 553 +++ new/rsync.yo
 554 @@ -364,6 +364,7 @@ to the detailed description below for a
 555       --modify-window=NUM     compare mod-times with reduced accuracy
 556   -T, --temp-dir=DIR          create temporary files in directory DIR
 557   -y, --fuzzy                 find similar file for basis if no dest file
 558 +     --detect-renamed        try to find renamed files to speed the xfer
 559       --compare-dest=DIR      also compare received files relative to DIR
 560       --copy-dest=DIR         ... and include copies of unchanged files
 561       --link-dest=DIR         hardlink to files in DIR when unchanged
 562 @@ -1272,6 +1273,15 @@ Note that the use of the bf(--delete) op
 563  fuzzy-match files, so either use bf(--delete-after) or specify some
 564  filename exclusions if you need to prevent this.
 565
 566 +dit(bf(--detect-renamed)) This option tells rsync to scan the receiving
 567 +side for files that have been renamed, and to use any that are found as
 568 +alternate basis files to help speed up the transfer.
 569 +By default, alternate-basis files are hard-linked into a directory named
 570 +".~tmp~" in each file's destination directory, but if you've specified
 571 +the bf(--partial-dir) option, that directory will be used instead.  These
 572 +potential alternate-basis files will be removed as the transfer progresses.
 573 +This option conflicts with bf(--inplace) and bf(--append).
 574 +
 575  dit(bf(--compare-dest=DIR)) This option instructs rsync to use em(DIR) on
 576  the destination machine as an additional hierarchy to compare destination
 577  files against doing transfers (if the files are missing in the destination
 578 --- old/util.c
 579 +++ new/util.c
 580 @@ -1027,6 +1027,32 @@ int handle_partial_dir(const char *fname
 581         return 1;
 582  }
 583
 584 +/* We need to supply our own strcmp function for file list comparisons
 585 + * to ensure that signed/unsigned usage is consistent between machines. */
 586 +int u_strcmp(const char *p1, const char *p2)
 587 +{
 588 +        for ( ; *p1; p1++, p2++) {
 589 +               if (*p1 != *p2)
 590 +                       break;
 591 +       }
 592 +
 593 +       return (int)*(uchar*)p1 - (int)*(uchar*)p2;
 594 +}
 595 +
 596 +/* We need a memcmp function compares unsigned-byte values. */
 597 +int u_memcmp(const void *p1, const void *p2, size_t len)
 598 +{
 599 +       const uchar *u1 = p1;
 600 +       const uchar *u2 = p2;
 601 +
 602 +       while (len--) {
 603 +               if (*u1 != *u2)
 604 +                       return (int)*u1 - (int)*u2;
 605 +       }
 606 +
 607 +       return 0;
 608 +}
 609 +
 610  /**
 611   * Determine if a symlink points outside the current directory tree.
 612   * This is considered "unsafe" because e.g. when mirroring somebody