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 @@ -78,6 +79,8 @@ 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 @@ -260,6 +263,44 @@ 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 @@ -1395,6 +1436,25 @@ 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 @@ -77,6 +77,7 @@ 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 @@ -93,15 +94,18 @@ 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 @@ -123,11 +127,121 @@ 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_SUM(f)) {
 182 +                               if (fmid->modtime == f->modtime
 183 +                                && f_name_cmp(fmid, f) == 0)
 184 +                                       return -1; /* assume we can't help */
 185 +                               /* XXX update this to new checksum var idiom! */
 186 +                               F_SUM(f) = pool_alloc(pool, MD4_SUM_LENGTH,
 187 +                                                     "fattr_find");
 188 +                               file_checksum(fname, F_SUM(f), f->length);
 189 +                       }
 190 +                       diff = u_memcmp(F_SUM(fmid), F_SUM(f), checksum_len);
 191 +                       if (diff) {
 192 +                               if (diff < 0)
 193 +                                       low = mid + 1;
 194 +                               else
 195 +                                       high = mid - 1;
 196 +                               continue;
 197 +                       }
 198 +               } else {
 199 +                       if (fmid->modtime != f->modtime) {
 200 +                               if (fmid->modtime < f->modtime)
 201 +                                       low = mid + 1;
 202 +                               else
 203 +                                       high = mid - 1;
 204 +                               continue;
 205 +                       }
 206 +               }
 207 +               ok_match = mid;
 208 +               diff = u_strcmp(fmid->basename, f->basename);
 209 +               if (diff == 0) {
 210 +                       good_match = mid;
 211 +                       if (fmid->dirname == f->dirname)
 212 +                               return -1; /* file is up-to-date */
 213 +                       if (!fmid->dirname) {
 214 +                               low = mid + 1;
 215 +                               continue;
 216 +                       }
 217 +                       if (!f->dirname) {
 218 +                               high = mid - 1;
 219 +                               continue;
 220 +                       }
 221 +                       diff = u_strcmp(fmid->dirname, f->dirname);
 222 +                       if (diff == 0)
 223 +                               return -1; /* file is up-to-date */
 224 +               }
 225 +               if (diff < 0)
 226 +                       low = mid + 1;
 227 +               else
 228 +                       high = mid - 1;
 229 +       }
 230 +
 231 +       return good_match >= 0 ? good_match : ok_match;
 232 +}
 233 +
 234 +static void look_for_rename(struct file_struct *file, char *fname,
 235 +                           alloc_pool_t pool)
 236 +{
 237 +       struct file_struct *fp;
 238 +       char *partialptr, *fn;
 239 +       STRUCT_STAT st;
 240 +       int ndx;
 241 +
 242 +       if ((ndx = fattr_find(file, fname, pool)) < 0)
 243 +               return;
 244 +
 245 +       fp = the_fattr_list.files[ndx];
 246 +       fn = f_name(fp, NULL);
 247 +       /* We don't provide an alternate-basis file if there is a basis file. */
 248 +       if (link_stat(fn, &st, 0) == 0)
 249 +               return;
 250 +       if ((partialptr = partial_dir_fname(fn)) == NULL
 251 +        || !handle_partial_dir(partialptr, PDIR_CREATE))
 252 +               return;
 253 +
 254 +       /* We only use the file if we can hard-link it into our tmp dir. */
 255 +       if (link(fname, partialptr) == 0) {
 256 +               if (verbose > 2) {
 257 +                       rprintf(FINFO, "found renamed: %s => %s\n",
 258 +                               fname, partialptr);
 259 +               }
 260 +               return;
 261 +       }
 262 +
 263 +       if (errno != EEXIST)
 264 +               handle_partial_dir(partialptr, PDIR_DELETE);
 265 +}
 266 +
 267  /* Delete a file or directory.  If DEL_RECURSE is set in the flags, this will
 268   * delete recursively.
 269   *
 270   * Note that fbuf must point to a MAXPATHLEN buffer if the mode indicates it's
 271   * a directory! (The buffer is used for recursion, but returned unchanged.)
 272 + *
 273 + * Also note: --detect-rename may use this routine with DEL_NO_DELETIONS set!
 274   */
 275  static enum delret delete_item(char *fbuf, int mode, char *replace, int flags)
 276  {
 277 @@ -149,6 +263,8 @@ static enum delret delete_item(char *fbu
 278                         goto check_ret;
 279                 /* OK: try to delete the directory. */
 280         }
 281 +       if (flags & DEL_NO_DELETIONS)
 282 +               return DR_SUCCESS;
 283
 284         if (!replace && max_delete >= 0 && ++deletion_count > max_delete)
 285                 return DR_AT_LIMIT;
 286 @@ -195,6 +311,8 @@ static enum delret delete_item(char *fbu
 287   * its contents, otherwise just checks for content.  Returns DR_SUCCESS or
 288   * DR_NOT_EMPTY.  Note that fname must point to a MAXPATHLEN buffer!  (The
 289   * buffer is used for recursion, but returned unchanged.)
 290 + *
 291 + * Note: --detect-rename may use this routine with DEL_NO_DELETIONS set!
 292   */
 293  static enum delret delete_dir_contents(char *fname, int flags)
 294  {
 295 @@ -251,6 +369,8 @@ static enum delret delete_dir_contents(c
 296                 if (S_ISDIR(fp->mode)
 297                  && delete_dir_contents(fname, flags | DEL_RECURSE) != DR_SUCCESS)
 298                         ret = DR_NOT_EMPTY;
 299 +               if (detect_renamed && S_ISREG(fp->mode))
 300 +                       look_for_rename(fp, fname, dirlist->file_pool);
 301                 if (delete_item(fname, fp->mode, NULL, flags) != DR_SUCCESS)
 302                         ret = DR_NOT_EMPTY;
 303         }
 304 @@ -335,15 +455,19 @@ static void delayed_deletions(char *delb
 305   * all the --delete-WHEN options.  Note that the fbuf pointer must point to a
 306   * MAXPATHLEN buffer with the name of the directory in it (the functions we
 307   * call will append names onto the end, but the old dir value will be restored
 308 - * on exit). */
 309 + * on exit).
 310 + *
 311 + * Note:  --detect-rename may use this routine with DEL_NO_DELETIONS set!
 312 + */
 313  static void delete_in_dir(struct file_list *flist, char *fbuf,
 314 -                         struct file_struct *file, STRUCT_STAT *stp)
 315 +                         struct file_struct *file, STRUCT_STAT *stp, int flags)
 316  {
 317         static int min_depth = MAXPATHLEN, cur_depth = -1;
 318         static void *filt_array[MAXPATHLEN/2+1];
 319         static int already_warned = 0;
 320         struct file_list *dirlist;
 321 -       char delbuf[MAXPATHLEN];
 322 +       char *p, delbuf[MAXPATHLEN];
 323 +       unsigned remainder;
 324         int dlen, i;
 325
 326         if (!flist) {
 327 @@ -357,6 +481,8 @@ static void delete_in_dir(struct file_li
 328         if (verbose > 2)
 329                 rprintf(FINFO, "delete_in_dir(%s)\n", fbuf);
 330
 331 +       flags |= DEL_RECURSE;
 332 +
 333         if (allowed_lull)
 334                 maybe_send_keepalive();
 335
 336 @@ -364,12 +490,14 @@ static void delete_in_dir(struct file_li
 337                 return; /* Impossible... */
 338
 339         if (io_error && !(lp_ignore_errors(module_id) || ignore_errors)) {
 340 -               if (already_warned)
 341 +               if (!already_warned) {
 342 +                       rprintf(FINFO,
 343 +                           "IO error encountered -- skipping file deletion\n");
 344 +                       already_warned = 1;
 345 +               }
 346 +               if (!detect_renamed)
 347                         return;
 348 -               rprintf(FINFO,
 349 -                       "IO error encountered -- skipping file deletion\n");
 350 -               already_warned = 1;
 351 -               return;
 352 +               flags |= DEL_NO_DELETIONS;
 353         }
 354
 355         while (cur_depth >= file->dir.depth && cur_depth >= min_depth)
 356 @@ -380,6 +508,9 @@ static void delete_in_dir(struct file_li
 357         dlen = strlen(fbuf);
 358         filt_array[cur_depth] = push_local_filters(fbuf, dlen);
 359
 360 +       if (detect_renamed)
 361 +               unexplored_dirs--;
 362 +
 363         if (one_file_system) {
 364                 if (file->flags & FLAG_TOP_DIR)
 365                         filesystem_dev = stp->st_dev;
 366 @@ -389,6 +520,11 @@ static void delete_in_dir(struct file_li
 367
 368         dirlist = get_dirlist(fbuf, dlen, 0);
 369
 370 +       p = fbuf + dlen;
 371 +       if (dlen != 1 || *fbuf != '/')
 372 +               *p++ = '/';
 373 +       remainder = MAXPATHLEN - (p - fbuf);
 374 +
 375         /* If an item in dirlist is not found in flist, delete it
 376          * from the filesystem. */
 377         for (i = dirlist->count; i--; ) {
 378 @@ -401,15 +537,22 @@ static void delete_in_dir(struct file_li
 379                                         f_name(fp, NULL));
 380                         continue;
 381                 }
 382 +               if (detect_renamed && S_ISREG(fp->mode)) {
 383 +                       strlcpy(p, fp->basename, remainder);
 384 +                       look_for_rename(fp, fbuf, dirlist->file_pool);
 385 +               }
 386                 if (flist_find(flist, fp) < 0) {
 387                         f_name(fp, delbuf);
 388 -                       if (delete_delay_fp)
 389 +                       if (delete_delay_fp && !(flags & DEL_NO_DELETIONS))
 390                                 fprintf(delete_delay_fp, "%o %s%c", (short)fp->mode, delbuf, '\0');
 391                         else
 392 -                               delete_item(delbuf, fp->mode, NULL, DEL_RECURSE);
 393 -               }
 394 +                               delete_item(delbuf, fp->mode, NULL, flags);
 395 +               } else if (detect_renamed && S_ISDIR(fp->mode))
 396 +                       unexplored_dirs++;
 397         }
 398
 399 +       fbuf[dlen] = '\0';
 400 +
 401         flist_free(dirlist);
 402  }
 403
 404 @@ -439,9 +582,9 @@ static void do_delete_pass(struct file_l
 405                  || !S_ISDIR(st.st_mode))
 406                         continue;
 407
 408 -               delete_in_dir(flist, fbuf, file, &st);
 409 +               delete_in_dir(flist, fbuf, file, &st, 0);
 410         }
 411 -       delete_in_dir(NULL, NULL, NULL, NULL);
 412 +       delete_in_dir(NULL, NULL, NULL, NULL, 0);
 413
 414         if (do_progress && !am_server)
 415                 rprintf(FINFO, "                    \r");
 416 @@ -966,6 +1109,7 @@ static int try_dests_non(struct file_str
 417         return j;
 418  }
 419
 420 +static struct bitbag *delayed_bits = NULL;
 421  static int phase = 0;
 422
 423  /* Acts on the_file_list->file's ndx'th item, whose name is fname.  If a dir,
 424 @@ -1152,8 +1296,12 @@ static void recv_generator(char *fname,
 425                 if (real_ret != 0 && one_file_system)
 426                         real_st.st_dev = filesystem_dev;
 427                 if (delete_during && f_out != -1 && !phase && dry_run < 2
 428 -                   && (file->flags & FLAG_XFER_DIR))
 429 -                       delete_in_dir(the_file_list, fname, file, &real_st);
 430 +                   && (file->flags & FLAG_XFER_DIR)) {
 431 +                       if (detect_renamed && real_ret != 0)
 432 +                               unexplored_dirs++;
 433 +                       delete_in_dir(the_file_list, fname, file, &real_st,
 434 +                                     delete_during < 0 ? DEL_NO_DELETIONS : 0);
 435 +               }
 436                 return;
 437         }
 438
 439 @@ -1406,8 +1554,14 @@ static void recv_generator(char *fname,
 440                     && hard_link_check(file, ndx, fname, statret, &st,
 441                                        itemizing, code, HL_SKIP))
 442                         return;
 443 -               if (stat_errno == ENOENT)
 444 +               if (stat_errno == ENOENT) {
 445 +                       if (detect_renamed && unexplored_dirs > 0
 446 +                        && file->length) {
 447 +                               bitbag_set_bit(delayed_bits, ndx);
 448 +                               return;
 449 +                       }
 450                         goto notify_others;
 451 +               }
 452                 rsyserr(FERROR, stat_errno, "recv_generator: failed to stat %s",
 453                         full_fname(fname));
 454                 return;
 455 @@ -1593,13 +1747,19 @@ void generate_files(int f_out, struct fi
 456                         (long)getpid(), flist->count);
 457         }
 458
 459 +       if (detect_renamed) {
 460 +               delayed_bits = bitbag_create(flist->count);
 461 +               if (!delete_before && !delete_during)
 462 +                       delete_during = -1;
 463 +       }
 464 +
 465         if (delete_before && !local_name && flist->count > 0)
 466                 do_delete_pass(flist);
 467         if (delete_during == 2)
 468                 start_delete_temp();
 469         do_progress = 0;
 470
 471 -       if (append_mode || whole_file < 0)
 472 +       if (append_mode || detect_renamed || whole_file < 0)
 473                 whole_file = 0;
 474         if (verbose >= 2) {
 475                 rprintf(FINFO, "delta-transmission %s\n",
 476 @@ -1654,7 +1814,23 @@ void generate_files(int f_out, struct fi
 477         }
 478         recv_generator(NULL, NULL, 0, 0, 0, code, -1);
 479         if (delete_during)
 480 -               delete_in_dir(NULL, NULL, NULL, NULL);
 481 +               delete_in_dir(NULL, NULL, NULL, NULL, 0);
 482 +
 483 +       if (detect_renamed) {
 484 +               if (delete_during < 0)
 485 +                       delete_during = 0;
 486 +               detect_renamed = 0;
 487 +
 488 +               for (i = -1; (i = bitbag_next_bit(delayed_bits, i)) >= 0; ) {
 489 +                       struct file_struct *file = flist->files[i];
 490 +                       if (local_name)
 491 +                               strlcpy(fbuf, local_name, sizeof fbuf);
 492 +                       else
 493 +                               f_name(file, fbuf);
 494 +                       recv_generator(fbuf, file, i, itemizing,
 495 +                                      maybe_ATTRS_REPORT, code, f_out);
 496 +               }
 497 +       }
 498
 499         phase++;
 500         csum_length = SUM_LENGTH;
 501 --- old/options.c
 502 +++ new/options.c
 503 @@ -78,6 +78,7 @@ int am_generator = 0;
 504  int am_starting_up = 1;
 505  int relative_paths = -1;
 506  int implied_dirs = 1;
 507 +int detect_renamed = 0;
 508  int numeric_ids = 0;
 509  int allow_8bit_chars = 0;
 510  int force_delete = 0;
 511 @@ -343,6 +344,7 @@ void usage(enum logcode F)
 512    rprintf(F,"     --modify-window=NUM     compare mod-times with reduced accuracy\n");
 513    rprintf(F," -T, --temp-dir=DIR          create temporary files in directory DIR\n");
 514    rprintf(F," -y, --fuzzy                 find similar file for basis if no dest file\n");
 515 +  rprintf(F,"     --detect-renamed        try to find renamed files to speed up the transfer\n");
 516    rprintf(F,"     --compare-dest=DIR      also compare destination files relative to DIR\n");
 517    rprintf(F,"     --copy-dest=DIR         ... and include copies of unchanged files\n");
 518    rprintf(F,"     --link-dest=DIR         hardlink to files in DIR when unchanged\n");
 519 @@ -497,6 +499,7 @@ static struct poptOption long_options[]
 520    {"compare-dest",     0,  POPT_ARG_STRING, 0, OPT_COMPARE_DEST, 0, 0 },
 521    {"copy-dest",        0,  POPT_ARG_STRING, 0, OPT_COPY_DEST, 0, 0 },
 522    {"link-dest",        0,  POPT_ARG_STRING, 0, OPT_LINK_DEST, 0, 0 },
 523 +  {"detect-renamed",   0,  POPT_ARG_NONE,   &detect_renamed, 0, 0, 0 },
 524    {"fuzzy",           'y', POPT_ARG_NONE,   &fuzzy_basis, 0, 0, 0 },
 525    {"compress",        'z', POPT_ARG_NONE,   0, 'z', 0, 0 },
 526    {"compress-level",   0,  POPT_ARG_INT,    &def_compress_level, 'z', 0, 0 },
 527 @@ -1361,7 +1364,7 @@ int parse_arguments(int *argc, const cha
 528                 inplace = 1;
 529         }
 530
 531 -       if (delay_updates && !partial_dir)
 532 +       if ((delay_updates || detect_renamed) && !partial_dir)
 533                 partial_dir = tmp_partialdir;
 534
 535         if (inplace) {
 536 @@ -1370,6 +1373,7 @@ int parse_arguments(int *argc, const cha
 537                         snprintf(err_buf, sizeof err_buf,
 538                                  "--%s cannot be used with --%s\n",
 539                                  append_mode ? "append" : "inplace",
 540 +                                detect_renamed ? "detect-renamed" :
 541                                  delay_updates ? "delay-updates" : "partial-dir");
 542                         return 0;
 543                 }
 544 @@ -1680,6 +1684,8 @@ void server_options(char **args,int *arg
 545                         args[ac++] = "--super";
 546                 if (size_only)
 547                         args[ac++] = "--size-only";
 548 +               if (detect_renamed)
 549 +                       args[ac++] = "--detect-renamed";
 550         }
 551
 552         if (modify_window_set) {
 553 --- old/rsync.yo
 554 +++ new/rsync.yo
 555 @@ -364,6 +364,7 @@ to the detailed description below for a
 556       --modify-window=NUM     compare mod-times with reduced accuracy
 557   -T, --temp-dir=DIR          create temporary files in directory DIR
 558   -y, --fuzzy                 find similar file for basis if no dest file
 559 +     --detect-renamed        try to find renamed files to speed the xfer
 560       --compare-dest=DIR      also compare received files relative to DIR
 561       --copy-dest=DIR         ... and include copies of unchanged files
 562       --link-dest=DIR         hardlink to files in DIR when unchanged
 563 @@ -1272,6 +1273,15 @@ Note that the use of the bf(--delete) op
 564  fuzzy-match files, so either use bf(--delete-after) or specify some
 565  filename exclusions if you need to prevent this.
 566
 567 +dit(bf(--detect-renamed)) This option tells rsync to scan the receiving
 568 +side for files that have been renamed, and to use any that are found as
 569 +alternate basis files to help speed up the transfer.
 570 +By default, alternate-basis files are hard-linked into a directory named
 571 +".~tmp~" in each file's destination directory, but if you've specified
 572 +the bf(--partial-dir) option, that directory will be used instead.  These
 573 +potential alternate-basis files will be removed as the transfer progresses.
 574 +This option conflicts with bf(--inplace) and bf(--append).
 575 +
 576  dit(bf(--compare-dest=DIR)) This option instructs rsync to use em(DIR) on
 577  the destination machine as an additional hierarchy to compare destination
 578  files against doing transfers (if the files are missing in the destination
 579 --- old/util.c
 580 +++ new/util.c
 581 @@ -1027,6 +1027,32 @@ int handle_partial_dir(const char *fname
 582         return 1;
 583  }
 584
 585 +/* We need to supply our own strcmp function for file list comparisons
 586 + * to ensure that signed/unsigned usage is consistent between machines. */
 587 +int u_strcmp(const char *p1, const char *p2)
 588 +{
 589 +        for ( ; *p1; p1++, p2++) {
 590 +               if (*p1 != *p2)
 591 +                       break;
 592 +       }
 593 +
 594 +       return (int)*(uchar*)p1 - (int)*(uchar*)p2;
 595 +}
 596 +
 597 +/* We need a memcmp function compares unsigned-byte values. */
 598 +int u_memcmp(const void *p1, const void *p2, size_t len)
 599 +{
 600 +       const uchar *u1 = p1;
 601 +       const uchar *u2 = p2;
 602 +
 603 +       while (len--) {
 604 +               if (*u1 != *u2)
 605 +                       return (int)*u1 - (int)*u2;
 606 +       }
 607 +
 608 +       return 0;
 609 +}
 610 +
 611  /**
 612   * Determine if a symlink points outside the current directory tree.
 613   * This is considered "unsafe" because e.g. when mirroring somebody