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 @@ -54,6 +54,7 @@ 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 @@ -77,6 +78,8 @@ static dev_t tmp_rdev;
  46  static struct idev tmp_idev;
  47  static char tmp_sum[MD4_SUM_LENGTH];
  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 @@ -259,6 +262,44 @@ 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(F_SUM(f1), F_SUM(f2), 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 @@ -1411,6 +1452,25 @@ 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 +                       out_of_memory("recv_file_list");
 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 @@ -91,16 +92,19 @@ 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  static BOOL solo_file = 0;
 149
 150 -/* For calling delete_item() and delete_dir_contents(). */
 151 +/* For calling delete_item(), delete_dir_contents(), and delete_in_dir(). */
 152 +#define DEL_NO_DELETIONS       (1<<0)
 153  #define DEL_RECURSE            (1<<1) /* recurse */
 154  #define DEL_DIR_IS_EMPTY       (1<<2) /* internal delete_FUNCTIONS use only */
 155
 156 @@ -122,11 +126,121 @@ static int is_backup_file(char *fn)
 157         return k > 0 && strcmp(fn+k, backup_suffix) == 0;
 158  }
 159
 160 +/* Search for a regular file that matches either (1) the size & modified
 161 + * time (plus the basename, if possible) or (2) the size & checksum.  If
 162 + * we find an exact match down to the dirname, return -1 because we found
 163 + * an up-to-date file in the transfer, not a renamed file. */
 164 +static int fattr_find(struct file_struct *f, char *fname, alloc_pool_t pool)
 165 +{
 166 +       int low = the_fattr_list.low, high = the_fattr_list.high;
 167 +       int mid, ok_match = -1, good_match = -1;
 168 +       struct file_struct *fmid;
 169 +       int diff;
 170 +
 171 +       while (low <= high) {
 172 +               mid = (low + high) / 2;
 173 +               fmid = the_fattr_list.files[mid];
 174 +               if (fmid->length != f->length) {
 175 +                       if (fmid->length < f->length)
 176 +                               low = mid + 1;
 177 +                       else
 178 +                               high = mid - 1;
 179 +                       continue;
 180 +               }
 181 +               if (always_checksum) {
 182 +                       if (!F_SUM(f)) {
 183 +                               if (fmid->modtime == f->modtime
 184 +                                && f_name_cmp(fmid, f) == 0)
 185 +                                       return -1; /* assume we can't help */
 186 +                               /* XXX update this to new checksum var idiom! */
 187 +                               F_SUM(f) = pool_alloc(pool, MD4_SUM_LENGTH,
 188 +                                                     "fattr_find");
 189 +                               file_checksum(fname, F_SUM(f), f->length);
 190 +                       }
 191 +                       diff = u_memcmp(F_SUM(fmid), F_SUM(f), checksum_len);
 192 +                       if (diff) {
 193 +                               if (diff < 0)
 194 +                                       low = mid + 1;
 195 +                               else
 196 +                                       high = mid - 1;
 197 +                               continue;
 198 +                       }
 199 +               } else {
 200 +                       if (fmid->modtime != f->modtime) {
 201 +                               if (fmid->modtime < f->modtime)
 202 +                                       low = mid + 1;
 203 +                               else
 204 +                                       high = mid - 1;
 205 +                               continue;
 206 +                       }
 207 +               }
 208 +               ok_match = mid;
 209 +               diff = u_strcmp(fmid->basename, f->basename);
 210 +               if (diff == 0) {
 211 +                       good_match = mid;
 212 +                       if (fmid->dirname == f->dirname)
 213 +                               return -1; /* file is up-to-date */
 214 +                       if (!fmid->dirname) {
 215 +                               low = mid + 1;
 216 +                               continue;
 217 +                       }
 218 +                       if (!f->dirname) {
 219 +                               high = mid - 1;
 220 +                               continue;
 221 +                       }
 222 +                       diff = u_strcmp(fmid->dirname, f->dirname);
 223 +                       if (diff == 0)
 224 +                               return -1; /* file is up-to-date */
 225 +               }
 226 +               if (diff < 0)
 227 +                       low = mid + 1;
 228 +               else
 229 +                       high = mid - 1;
 230 +       }
 231 +
 232 +       return good_match >= 0 ? good_match : ok_match;
 233 +}
 234 +
 235 +static void look_for_rename(struct file_struct *file, char *fname,
 236 +                           alloc_pool_t pool)
 237 +{
 238 +       struct file_struct *fp;
 239 +       char *partialptr, *fn;
 240 +       STRUCT_STAT st;
 241 +       int ndx;
 242 +
 243 +       if ((ndx = fattr_find(file, fname, pool)) < 0)
 244 +               return;
 245 +
 246 +       fp = the_fattr_list.files[ndx];
 247 +       fn = f_name(fp, NULL);
 248 +       /* We don't provide an alternate-basis file if there is a basis file. */
 249 +       if (link_stat(fn, &st, 0) == 0)
 250 +               return;
 251 +       if ((partialptr = partial_dir_fname(fn)) == NULL
 252 +        || !handle_partial_dir(partialptr, PDIR_CREATE))
 253 +               return;
 254 +
 255 +       /* We only use the file if we can hard-link it into our tmp dir. */
 256 +       if (link(fname, partialptr) == 0) {
 257 +               if (verbose > 2) {
 258 +                       rprintf(FINFO, "found renamed: %s => %s\n",
 259 +                               fname, partialptr);
 260 +               }
 261 +               return;
 262 +       }
 263 +
 264 +       if (errno != EEXIST)
 265 +               handle_partial_dir(partialptr, PDIR_DELETE);
 266 +}
 267 +
 268  /* Delete a file or directory.  If DEL_RECURSE is set in the flags, this will
 269   * delete recursively.
 270   *
 271   * Note that fbuf must point to a MAXPATHLEN buffer if the mode indicates it's
 272   * a directory! (The buffer is used for recursion, but returned unchanged.)
 273 + *
 274 + * Also note: --detect-rename may use this routine with DEL_NO_DELETIONS set!
 275   */
 276  static enum delret delete_item(char *fbuf, int mode, char *replace, int flags)
 277  {
 278 @@ -148,6 +262,8 @@ static enum delret delete_item(char *fbu
 279                         goto check_ret;
 280                 /* OK: try to delete the directory. */
 281         }
 282 +       if (flags & DEL_NO_DELETIONS)
 283 +               return DR_SUCCESS;
 284
 285         if (!replace && max_delete >= 0 && ++deletion_count > max_delete)
 286                 return DR_AT_LIMIT;
 287 @@ -194,6 +310,8 @@ static enum delret delete_item(char *fbu
 288   * its contents, otherwise just checks for content.  Returns DR_SUCCESS or
 289   * DR_NOT_EMPTY.  Note that fname must point to a MAXPATHLEN buffer!  (The
 290   * buffer is used for recursion, but returned unchanged.)
 291 + *
 292 + * Note: --detect-rename may use this routine with DEL_NO_DELETIONS set!
 293   */
 294  static enum delret delete_dir_contents(char *fname, int flags)
 295  {
 296 @@ -250,6 +368,8 @@ static enum delret delete_dir_contents(c
 297                 if (S_ISDIR(fp->mode)
 298                  && delete_dir_contents(fname, flags | DEL_RECURSE) != DR_SUCCESS)
 299                         ret = DR_NOT_EMPTY;
 300 +               if (detect_renamed && S_ISREG(fp->mode))
 301 +                       look_for_rename(fp, fname, dirlist->file_pool);
 302                 if (delete_item(fname, fp->mode, NULL, flags) != DR_SUCCESS)
 303                         ret = DR_NOT_EMPTY;
 304         }
 305 @@ -334,15 +454,19 @@ static void delayed_deletions(char *delb
 306   * all the --delete-WHEN options.  Note that the fbuf pointer must point to a
 307   * MAXPATHLEN buffer with the name of the directory in it (the functions we
 308   * call will append names onto the end, but the old dir value will be restored
 309 - * on exit). */
 310 + * on exit).
 311 + *
 312 + * Note:  --detect-rename may use this routine with DEL_NO_DELETIONS set!
 313 + */
 314  static void delete_in_dir(struct file_list *flist, char *fbuf,
 315 -                         struct file_struct *file, STRUCT_STAT *stp)
 316 +                         struct file_struct *file, STRUCT_STAT *stp, int flags)
 317  {
 318         static int min_depth = MAXPATHLEN, cur_depth = -1;
 319         static void *filt_array[MAXPATHLEN/2+1];
 320         static int already_warned = 0;
 321         struct file_list *dirlist;
 322 -       char delbuf[MAXPATHLEN];
 323 +       char *p, delbuf[MAXPATHLEN];
 324 +       unsigned remainder;
 325         int dlen, i;
 326
 327         if (!flist) {
 328 @@ -356,6 +480,8 @@ static void delete_in_dir(struct file_li
 329         if (verbose > 2)
 330                 rprintf(FINFO, "delete_in_dir(%s)\n", fbuf);
 331
 332 +       flags |= DEL_RECURSE;
 333 +
 334         if (allowed_lull)
 335                 maybe_send_keepalive();
 336
 337 @@ -363,12 +489,14 @@ static void delete_in_dir(struct file_li
 338                 return; /* Impossible... */
 339
 340         if (io_error && !(lp_ignore_errors(module_id) || ignore_errors)) {
 341 -               if (already_warned)
 342 +               if (!already_warned) {
 343 +                       rprintf(FINFO,
 344 +                           "IO error encountered -- skipping file deletion\n");
 345 +                       already_warned = 1;
 346 +               }
 347 +               if (!detect_renamed)
 348                         return;
 349 -               rprintf(FINFO,
 350 -                       "IO error encountered -- skipping file deletion\n");
 351 -               already_warned = 1;
 352 -               return;
 353 +               flags |= DEL_NO_DELETIONS;
 354         }
 355
 356         while (cur_depth >= file->dir.depth && cur_depth >= min_depth)
 357 @@ -379,6 +507,9 @@ static void delete_in_dir(struct file_li
 358         dlen = strlen(fbuf);
 359         filt_array[cur_depth] = push_local_filters(fbuf, dlen);
 360
 361 +       if (detect_renamed)
 362 +               unexplored_dirs--;
 363 +
 364         if (one_file_system) {
 365                 if (file->flags & FLAG_TOP_DIR)
 366                         filesystem_dev = stp->st_dev;
 367 @@ -388,6 +519,11 @@ static void delete_in_dir(struct file_li
 368
 369         dirlist = get_dirlist(fbuf, dlen, 0);
 370
 371 +       p = fbuf + dlen;
 372 +       if (dlen != 1 || *fbuf != '/')
 373 +               *p++ = '/';
 374 +       remainder = MAXPATHLEN - (p - fbuf);
 375 +
 376         /* If an item in dirlist is not found in flist, delete it
 377          * from the filesystem. */
 378         for (i = dirlist->count; i--; ) {
 379 @@ -400,15 +536,22 @@ static void delete_in_dir(struct file_li
 380                                         f_name(fp, NULL));
 381                         continue;
 382                 }
 383 +               if (detect_renamed && S_ISREG(fp->mode)) {
 384 +                       strlcpy(p, fp->basename, remainder);
 385 +                       look_for_rename(fp, fbuf, dirlist->file_pool);
 386 +               }
 387                 if (flist_find(flist, fp) < 0) {
 388                         f_name(fp, delbuf);
 389 -                       if (delete_delay_fp)
 390 +                       if (delete_delay_fp && !(flags & DEL_NO_DELETIONS))
 391                                 fprintf(delete_delay_fp, "%o %s%c", (short)fp->mode, delbuf, '\0');
 392                         else
 393 -                               delete_item(delbuf, fp->mode, NULL, DEL_RECURSE);
 394 -               }
 395 +                               delete_item(delbuf, fp->mode, NULL, flags);
 396 +               } else if (detect_renamed && S_ISDIR(fp->mode))
 397 +                       unexplored_dirs++;
 398         }
 399
 400 +       fbuf[dlen] = '\0';
 401 +
 402         flist_free(dirlist);
 403  }
 404
 405 @@ -438,9 +581,9 @@ static void do_delete_pass(struct file_l
 406                  || !S_ISDIR(st.st_mode))
 407                         continue;
 408
 409 -               delete_in_dir(flist, fbuf, file, &st);
 410 +               delete_in_dir(flist, fbuf, file, &st, 0);
 411         }
 412 -       delete_in_dir(NULL, NULL, NULL, NULL);
 413 +       delete_in_dir(NULL, NULL, NULL, NULL, 0);
 414
 415         if (do_progress && !am_server)
 416                 rprintf(FINFO, "                    \r");
 417 @@ -971,6 +1114,7 @@ static int try_dests_non(struct file_str
 418         return j;
 419  }
 420
 421 +static struct bitbag *delayed_bits = NULL;
 422  static int phase = 0;
 423
 424  /* Acts on the_file_list->file's ndx'th item, whose name is fname.  If a dir,
 425 @@ -1157,8 +1301,12 @@ static void recv_generator(char *fname,
 426                 if (real_ret != 0 && one_file_system)
 427                         real_st.st_dev = filesystem_dev;
 428                 if (delete_during && f_out != -1 && !phase && dry_run < 2
 429 -                   && (file->flags & FLAG_XFER_DIR))
 430 -                       delete_in_dir(the_file_list, fname, file, &real_st);
 431 +                   && (file->flags & FLAG_XFER_DIR)) {
 432 +                       if (detect_renamed && real_ret != 0)
 433 +                               unexplored_dirs++;
 434 +                       delete_in_dir(the_file_list, fname, file, &real_st,
 435 +                                     delete_during < 0 ? DEL_NO_DELETIONS : 0);
 436 +               }
 437                 return;
 438         }
 439
 440 @@ -1429,8 +1577,14 @@ static void recv_generator(char *fname,
 441                                        itemizing, code, HL_SKIP))
 442                         return;
 443  #endif
 444 -               if (stat_errno == ENOENT)
 445 +               if (stat_errno == ENOENT) {
 446 +                       if (detect_renamed && unexplored_dirs > 0
 447 +                        && file->length) {
 448 +                               bitbag_set_bit(delayed_bits, ndx);
 449 +                               return;
 450 +                       }
 451                         goto notify_others;
 452 +               }
 453                 rsyserr(FERROR, stat_errno, "recv_generator: failed to stat %s",
 454                         full_fname(fname));
 455                 return;
 456 @@ -1619,13 +1773,19 @@ void generate_files(int f_out, struct fi
 457                         (long)getpid(), flist->count);
 458         }
 459
 460 +       if (detect_renamed) {
 461 +               delayed_bits = bitbag_create(flist->count);
 462 +               if (!delete_before && !delete_during)
 463 +                       delete_during = -1;
 464 +       }
 465 +
 466         if (delete_before && !local_name && flist->count > 0)
 467                 do_delete_pass(flist);
 468         if (delete_during == 2)
 469                 start_delete_temp();
 470         do_progress = 0;
 471
 472 -       if (append_mode || whole_file < 0)
 473 +       if (append_mode || detect_renamed || whole_file < 0)
 474                 whole_file = 0;
 475         if (verbose >= 2) {
 476                 rprintf(FINFO, "delta-transmission %s\n",
 477 @@ -1682,7 +1842,23 @@ void generate_files(int f_out, struct fi
 478         }
 479         recv_generator(NULL, NULL, 0, 0, 0, code, -1);
 480         if (delete_during)
 481 -               delete_in_dir(NULL, NULL, NULL, NULL);
 482 +               delete_in_dir(NULL, NULL, NULL, NULL, 0);
 483 +
 484 +       if (detect_renamed) {
 485 +               if (delete_during < 0)
 486 +                       delete_during = 0;
 487 +               detect_renamed = 0;
 488 +
 489 +               for (i = -1; (i = bitbag_next_bit(delayed_bits, i)) >= 0; ) {
 490 +                       struct file_struct *file = flist->files[i];
 491 +                       if (local_name)
 492 +                               strlcpy(fbuf, local_name, sizeof fbuf);
 493 +                       else
 494 +                               f_name(file, fbuf);
 495 +                       recv_generator(fbuf, file, i, itemizing,
 496 +                                      maybe_ATTRS_REPORT, code, f_out);
 497 +               }
 498 +       }
 499
 500         phase++;
 501         csum_length = SUM_LENGTH;
 502 --- old/options.c
 503 +++ new/options.c
 504 @@ -78,6 +78,7 @@ int am_generator = 0;
 505  int am_starting_up = 1;
 506  int relative_paths = -1;
 507  int implied_dirs = 1;
 508 +int detect_renamed = 0;
 509  int numeric_ids = 0;
 510  int allow_8bit_chars = 0;
 511  int force_delete = 0;
 512 @@ -343,6 +344,7 @@ void usage(enum logcode F)
 513    rprintf(F,"     --modify-window=NUM     compare mod-times with reduced accuracy\n");
 514    rprintf(F," -T, --temp-dir=DIR          create temporary files in directory DIR\n");
 515    rprintf(F," -y, --fuzzy                 find similar file for basis if no dest file\n");
 516 +  rprintf(F,"     --detect-renamed        try to find renamed files to speed up the transfer\n");
 517    rprintf(F,"     --compare-dest=DIR      also compare destination files relative to DIR\n");
 518    rprintf(F,"     --copy-dest=DIR         ... and include copies of unchanged files\n");
 519    rprintf(F,"     --link-dest=DIR         hardlink to files in DIR when unchanged\n");
 520 @@ -497,6 +499,7 @@ static struct poptOption long_options[]
 521    {"compare-dest",     0,  POPT_ARG_STRING, 0, OPT_COMPARE_DEST, 0, 0 },
 522    {"copy-dest",        0,  POPT_ARG_STRING, 0, OPT_COPY_DEST, 0, 0 },
 523    {"link-dest",        0,  POPT_ARG_STRING, 0, OPT_LINK_DEST, 0, 0 },
 524 +  {"detect-renamed",   0,  POPT_ARG_NONE,   &detect_renamed, 0, 0, 0 },
 525    {"fuzzy",           'y', POPT_ARG_NONE,   &fuzzy_basis, 0, 0, 0 },
 526    {"compress",        'z', POPT_ARG_NONE,   0, 'z', 0, 0 },
 527    {"compress-level",   0,  POPT_ARG_INT,    &def_compress_level, 'z', 0, 0 },
 528 @@ -1361,7 +1364,7 @@ int parse_arguments(int *argc, const cha
 529                 inplace = 1;
 530         }
 531
 532 -       if (delay_updates && !partial_dir)
 533 +       if ((delay_updates || detect_renamed) && !partial_dir)
 534                 partial_dir = tmp_partialdir;
 535
 536         if (inplace) {
 537 @@ -1370,6 +1373,7 @@ int parse_arguments(int *argc, const cha
 538                         snprintf(err_buf, sizeof err_buf,
 539                                  "--%s cannot be used with --%s\n",
 540                                  append_mode ? "append" : "inplace",
 541 +                                detect_renamed ? "detect-renamed" :
 542                                  delay_updates ? "delay-updates" : "partial-dir");
 543                         return 0;
 544                 }
 545 @@ -1680,6 +1684,8 @@ void server_options(char **args,int *arg
 546                         args[ac++] = "--super";
 547                 if (size_only)
 548                         args[ac++] = "--size-only";
 549 +               if (detect_renamed)
 550 +                       args[ac++] = "--detect-renamed";
 551         }
 552
 553         if (modify_window_set) {
 554 --- old/rsync.yo
 555 +++ new/rsync.yo
 556 @@ -364,6 +364,7 @@ to the detailed description below for a
 557       --modify-window=NUM     compare mod-times with reduced accuracy
 558   -T, --temp-dir=DIR          create temporary files in directory DIR
 559   -y, --fuzzy                 find similar file for basis if no dest file
 560 +     --detect-renamed        try to find renamed files to speed the xfer
 561       --compare-dest=DIR      also compare received files relative to DIR
 562       --copy-dest=DIR         ... and include copies of unchanged files
 563       --link-dest=DIR         hardlink to files in DIR when unchanged
 564 @@ -1272,6 +1273,15 @@ Note that the use of the bf(--delete) op
 565  fuzzy-match files, so either use bf(--delete-after) or specify some
 566  filename exclusions if you need to prevent this.
 567
 568 +dit(bf(--detect-renamed)) This option tells rsync to scan the receiving
 569 +side for files that have been renamed, and to use any that are found as
 570 +alternate basis files to help speed up the transfer.
 571 +By default, alternate-basis files are hard-linked into a directory named
 572 +".~tmp~" in each file's destination directory, but if you've specified
 573 +the bf(--partial-dir) option, that directory will be used instead.  These
 574 +potential alternate-basis files will be removed as the transfer progresses.
 575 +This option conflicts with bf(--inplace) and bf(--append).
 576 +
 577  dit(bf(--compare-dest=DIR)) This option instructs rsync to use em(DIR) on
 578  the destination machine as an additional hierarchy to compare destination
 579  files against doing transfers (if the files are missing in the destination
 580 --- old/util.c
 581 +++ new/util.c
 582 @@ -1027,6 +1027,32 @@ int handle_partial_dir(const char *fname
 583         return 1;
 584  }
 585
 586 +/* We need to supply our own strcmp function for file list comparisons
 587 + * to ensure that signed/unsigned usage is consistent between machines. */
 588 +int u_strcmp(const char *p1, const char *p2)
 589 +{
 590 +        for ( ; *p1; p1++, p2++) {
 591 +               if (*p1 != *p2)
 592 +                       break;
 593 +       }
 594 +
 595 +       return (int)*(uchar*)p1 - (int)*(uchar*)p2;
 596 +}
 597 +
 598 +/* We need a memcmp function compares unsigned-byte values. */
 599 +int u_memcmp(const void *p1, const void *p2, size_t len)
 600 +{
 601 +       const uchar *u1 = p1;
 602 +       const uchar *u2 = p2;
 603 +
 604 +       while (len--) {
 605 +               if (*u1 != *u2)
 606 +                       return (int)*u1 - (int)*u2;
 607 +       }
 608 +
 609 +       return 0;
 610 +}
 611 +
 612  /**
 613   * Determine if a symlink points outside the current directory tree.
 614   * This is considered "unsafe" because e.g. when mirroring somebody