1 /* vi: set sw=4 ts=4: */
3 * Mini diff implementation for busybox, adapted from OpenBSD diff.
5 * Copyright (C) 2010 by Matheus Izvekov <mizvekov@gmail.com>
6 * Copyright (C) 2006 by Robert Sullivan <cogito.ergo.cogito@hotmail.com>
7 * Copyright (c) 2003 Todd C. Miller <Todd.Miller@courtesan.com>
9 * Sponsored in part by the Defense Advanced Research Projects
10 * Agency (DARPA) and Air Force Research Laboratory, Air Force
11 * Materiel Command, USAF, under agreement number F39502-99-1-0512.
13 * Licensed under GPLv2 or later, see file LICENSE in this tarball for details.
17 * The following code uses an algorithm due to Harold Stone,
18 * which finds a pair of longest identical subsequences in
21 * The major goal is to generate the match vector J.
22 * J[i] is the index of the line in file1 corresponding
23 * to line i in file0. J[i] = 0 if there is no
26 * Lines are hashed so as to work in core. All potential
27 * matches are located by sorting the lines of each file
28 * on the hash (called "value"). In particular, this
29 * collects the equivalence classes in file1 together.
30 * Subroutine equiv replaces the value of each line in
31 * file0 by the index of the first element of its
32 * matching equivalence in (the reordered) file1.
33 * To save space equiv squeezes file1 into a single
34 * array member in which the equivalence classes
35 * are simply concatenated, except that their first
36 * members are flagged by changing sign.
38 * Next the indices that point into member are unsorted into
39 * array class according to the original order of file0.
41 * The cleverness lies in routine stone. This marches
42 * through the lines of file0, developing a vector klist
43 * of "k-candidates". At step i a k-candidate is a matched
44 * pair of lines x,y (x in file0, y in file1) such that
45 * there is a common subsequence of length k
46 * between the first i lines of file0 and the first y
47 * lines of file1, but there is no such subsequence for
48 * any smaller y. x is the earliest possible mate to y
49 * that occurs in such a subsequence.
51 * Whenever any of the members of the equivalence class of
52 * lines in file1 matable to a line in file0 has serial number
53 * less than the y of some k-candidate, that k-candidate
54 * with the smallest such y is replaced. The new
55 * k-candidate is chained (via pred) to the current
56 * k-1 candidate so that the actual subsequence can
57 * be recovered. When a member has serial number greater
58 * that the y of all k-candidates, the klist is extended.
59 * At the end, the longest subsequence is pulled out
60 * and placed in the array J by unravel
62 * With J in hand, the matches there recorded are
63 * checked against reality to assure that no spurious
64 * matches have crept in due to hashing. If they have,
65 * they are broken, and "jackpot" is recorded--a harmless
66 * matter except that a true match for a spuriously
67 * mated line may now be unnecessarily reported as a change.
69 * Much of the complexity of the program comes simply
70 * from trying to minimize core utilization and
71 * maximize the range of doable problems by dynamically
72 * allocating what is needed and reusing what is not.
73 * The core requirements for problems larger than somewhat
74 * are (in words) 2*length(file0) + length(file1) +
75 * 3*(number of k-candidates installed), typically about
76 * 6n words for files of length n.
82 //#define dbg_error_msg(...) bb_error_msg(__VA_ARGS__)
84 #define dbg_error_msg(...) ((void)0)
87 enum { /* print_status() and diffreg() return values */
88 STATUS_SAME, /* files are the same */
89 STATUS_DIFFER, /* files differ */
90 STATUS_BINARY, /* binary files differ */
93 enum { /* Commandline flags */
98 FLAG_L, /* never used, handled by getopt32 */
103 FLAG_S, /* never used, handled by getopt32 */
106 FLAG_U, /* never used, handled by getopt32 */
108 FLAG_u, /* ignored, this is the default */
109 FLAG_p, /* not implemented */
111 FLAG_E, /* not implemented */
113 #define FLAG(x) (1 << FLAG_##x)
115 /* We cache file position to avoid excessive seeking */
116 typedef struct FILE_and_pos_t {
122 smallint exit_status;
127 #define G (*ptr_to_globals)
128 #define exit_status (G.exit_status )
129 #define opt_U_context (G.opt_U_context )
130 #define label (G.label )
132 #define INIT_G() do { \
133 SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
141 TOK_EMPTY = 1 << 9, /* Line fully processed, you can proceed to the next */
142 TOK_EOF = 1 << 10, /* File ended */
143 /* Private (Only to be used by read_token() */
144 TOK_EOL = 1 << 11, /* we saw EOL (sticky) */
145 TOK_SPACE = 1 << 12, /* used -b code, means we are skipping spaces */
146 SHIFT_EOF = (sizeof(token_t)*8 - 8) - 1,
147 CHAR_MASK = 0x1ff, /* 8th bit is used to distinguish EOF from 0xff */
150 /* Restores full EOF from one 8th bit: */
151 //#define TOK2CHAR(t) (((t) << SHIFT_EOF) >> SHIFT_EOF)
152 /* We don't really need the above, we only need to have EOF != any_real_char: */
153 #define TOK2CHAR(t) ((t) & CHAR_MASK)
155 static void seek_ft(FILE_and_pos_t *ft, off_t pos)
157 if (ft->ft_pos != pos) {
159 fseeko(ft->ft_fp, pos, SEEK_SET);
163 /* Reads tokens from given fp, handling -b and -w flags
164 * The user must reset tok every line start
166 static int read_token(FILE_and_pos_t *ft, token_t tok)
169 while (!(tok & TOK_EOL)) {
173 t = fgetc(ft->ft_fp);
176 is_space = (t == EOF || isspace(t));
178 /* If t == EOF (-1), set both TOK_EOF and TOK_EOL */
179 tok |= (t & (TOK_EOF + TOK_EOL));
184 if (option_mask32 & FLAG(i)) /* Handcoded tolower() */
185 t = (t >= 'A' && t <= 'Z') ? t - ('A' - 'a') : t;
187 if ((option_mask32 & FLAG(w)) && is_space)
190 /* Trim char value to low 9 bits */
193 if (option_mask32 & FLAG(b)) {
194 /* Was prev char whitespace? */
195 if (tok & TOK_SPACE) { /* yes */
196 if (is_space) /* this one too, ignore it */
199 } else if (is_space) {
200 /* 1st whitespace char.
201 * Set TOK_SPACE and replace char by ' ' */
206 tok &= ~(TOK_EMPTY + CHAR_MASK);
207 /* Assign char value (low 9 bits) and maybe set TOK_SPACE */
212 bb_error_msg("fp:%p tok:%x '%c'%s%s%s%s", fp, tok, tok & 0xff
213 , tok & TOK_EOF ? " EOF" : ""
214 , tok & TOK_EOL ? " EOL" : ""
215 , tok & TOK_EMPTY ? " EMPTY" : ""
216 , tok & TOK_SPACE ? " SPACE" : ""
228 static int search(const int *c, int k, int y, const struct cand *list)
230 if (list[c[k]].y < y) /* quick look for typical case */
233 for (int i = 0, j = k + 1;;) {
234 const int l = (i + j) >> 1;
236 const int t = list[c[l]].y;
248 static unsigned isqrt(unsigned n)
252 const unsigned y = x;
253 x = ((n / x) + x) >> 1;
254 if (x <= (y + 1) && x >= (y - 1))
259 static void stone(const int *a, int n, const int *b, int *J, int pref)
261 const unsigned isq = isqrt(n);
262 const unsigned bound =
263 (option_mask32 & FLAG(d)) ? UINT_MAX : MAX(256, isq);
267 struct cand *clist = xzalloc(clistlen * sizeof(clist[0]));
268 int *klist = xzalloc((n + 2) * sizeof(klist[0]));
269 /*clist[0] = (struct cand){0}; - xzalloc did it */
272 for (struct cand cand = {1}; cand.x <= n; cand.x++) {
273 int j = a[cand.x], oldl = 0;
274 unsigned numtries = 0;
278 cand.pred = klist[0];
281 if (cand.y <= clist[cand.pred].y)
283 l = search(klist, k, cand.y, clist);
285 cand.pred = klist[l - 1];
286 if (l <= k && clist[klist[l]].y <= cand.y)
288 if (clen == clistlen) {
289 clistlen = clistlen * 11 / 10;
290 clist = xrealloc(clist, clistlen * sizeof(clist[0]));
303 } while ((cand.y = b[++j]) > 0 && numtries < bound);
306 for (struct cand *q = clist + klist[k]; q->y; q = clist + q->pred)
307 J[q->x + pref] = q->y + pref;
313 /* 'serial' is not used in the begining, so we reuse it
314 * to store line offsets, thus reducing memory pressure
323 static void equiv(struct line *a, int n, struct line *b, int m, int *c)
327 while (i <= n && j <= m) {
328 if (a[i].value < b[j].value)
330 else if (a[i].value == b[j].value)
341 while (b[j + 1].value == b[j].value) {
349 static void unsort(const struct line *f, int l, int *b)
351 int *a = xmalloc((l + 1) * sizeof(a[0]));
352 for (int i = 1; i <= l; i++)
353 a[f[i].serial] = f[i].value;
354 for (int i = 1; i <= l; i++)
359 static int line_compar(const void *a, const void *b)
361 #define l0 ((const struct line*)a)
362 #define l1 ((const struct line*)b)
363 int r = l0->value - l1->value;
366 return l0->serial - l1->serial;
371 static void uni_range(int a, int b)
374 printf("%d,%d", a, b - a + 1);
381 static void fetch(FILE_and_pos_t *ft, const off_t *ix, int a, int b, int ch)
383 for (int i = a; i <= b; i++) {
384 seek_ft(ft, ix[i - 1]);
386 if (option_mask32 & FLAG(T))
388 for (int j = 0, col = 0; j < ix[i] - ix[i - 1]; j++) {
389 int c = fgetc(ft->ft_fp);
391 printf("\n\\ No newline at end of file\n");
395 if (c == '\t' && (option_mask32 & FLAG(t)))
396 do putchar(' '); while (++col & 7);
405 /* Creates the match vector J, where J[i] is the index
406 * of the line in the new file corresponding to the line i
407 * in the old file. Lines start at 1 instead of 0, that value
408 * being used instead to denote no corresponding line.
409 * This vector is dynamically allocated and must be freed by the caller.
411 * * fp is an input parameter, where fp[0] and fp[1] are the open
412 * old file and new file respectively.
413 * * nlen is an output variable, where nlen[0] and nlen[1]
414 * gets the number of lines in the old and new file respectively.
415 * * ix is an output variable, where ix[0] and ix[1] gets
416 * assigned dynamically allocated vectors of the offsets of the lines
417 * of the old and new file respectively. These must be freed by the caller.
419 static NOINLINE int *create_J(FILE_and_pos_t ft[2], int nlen[2], off_t *ix[2])
421 int *J, slen[2], *class, *member;
422 struct line *nfile[2], *sfile[2];
423 int pref = 0, suff = 0;
425 /* Lines of both files are hashed, and in the process
426 * their offsets are stored in the array ix[fileno]
427 * where fileno == 0 points to the old file, and
428 * fileno == 1 points to the new one.
430 for (int i = 0; i < 2; i++) {
434 nfile[i] = xmalloc((sz + 3) * sizeof(nfile[i][0]));
435 fseeko(ft[i].ft_fp, 0, SEEK_SET); /* ft gets here without the correct position */
438 /* We could zalloc nfile, but then zalloc starts showing in gprof at ~1% */
439 nfile[i][0].offset = 0;
440 goto start; /* saves code */
442 tok = read_token(&ft[i], tok);
443 if (!(tok & TOK_EMPTY)) {
444 /* Hash algorithm taken from Robert Sedgewick, Algorithms in C, 3d ed., p 578. */
445 /*hash = hash * 128 - hash + TOK2CHAR(tok);
446 * gcc insists on optimizing above to "hash * 127 + ...", thus... */
447 unsigned o = hash - TOK2CHAR(tok);
448 hash = hash * 128 - o; /* we want SPEED here */
451 if (nlen[i]++ == sz) {
453 nfile[i] = xrealloc(nfile[i], (sz + 3) * sizeof(nfile[i][0]));
455 /* line_compar needs hashes fit into positive int */
456 nfile[i][nlen[i]].value = hash & INT_MAX;
457 /* like ftello(ft[i].ft_fp) but faster (avoids lseek syscall) */
458 nfile[i][nlen[i]].offset = ft[i].ft_pos;
460 /* EOF counts as a token, so we have to adjust it here */
461 nfile[i][nlen[i]].offset++;
467 /* Exclude lone EOF line from the end of the file, to make fetch()'s job easier */
468 if (nfile[i][nlen[i]].offset - nfile[i][nlen[i] - 1].offset == 1)
470 /* Now we copy the line offsets into ix */
471 ix[i] = xmalloc((nlen[i] + 2) * sizeof(ix[i][0]));
472 for (int j = 0; j < nlen[i] + 1; j++)
473 ix[i][j] = nfile[i][j].offset;
476 /* lenght of prefix and suffix is calculated */
477 for (; pref < nlen[0] && pref < nlen[1] &&
478 nfile[0][pref + 1].value == nfile[1][pref + 1].value;
480 for (; suff < nlen[0] - pref && suff < nlen[1] - pref &&
481 nfile[0][nlen[0] - suff].value == nfile[1][nlen[1] - suff].value;
483 /* Arrays are pruned by the suffix and prefix lenght,
484 * the result being sorted and stored in sfile[fileno],
485 * and their sizes are stored in slen[fileno]
487 for (int j = 0; j < 2; j++) {
488 sfile[j] = nfile[j] + pref;
489 slen[j] = nlen[j] - pref - suff;
490 for (int i = 0; i <= slen[j]; i++)
491 sfile[j][i].serial = i;
492 qsort(sfile[j] + 1, slen[j], sizeof(*sfile[j]), line_compar);
494 /* nfile arrays are reused to reduce memory pressure
495 * The #if zeroed out section performs the same task as the
496 * one in the #else section.
497 * Peak memory usage is higher, but one array copy is avoided
498 * by not using unsort()
501 member = xmalloc((slen[1] + 2) * sizeof(member[0]));
502 equiv(sfile[0], slen[0], sfile[1], slen[1], member);
505 class = xmalloc((slen[0] + 1) * sizeof(class[0]));
506 for (int i = 1; i <= slen[0]; i++) /* Unsorting */
507 class[sfile[0][i].serial] = sfile[0][i].value;
510 member = (int *)nfile[1];
511 equiv(sfile[0], slen[0], sfile[1], slen[1], member);
512 member = xrealloc(member, (slen[1] + 2) * sizeof(member[0]));
514 class = (int *)nfile[0];
515 unsort(sfile[0], slen[0], (int *)nfile[0]);
516 class = xrealloc(class, (slen[0] + 2) * sizeof(class[0]));
518 J = xmalloc((nlen[0] + 2) * sizeof(J[0]));
519 /* The elements of J which fall inside the prefix and suffix regions
520 * are marked as unchanged, while the ones which fall outside
521 * are initialized with 0 (no matches), so that function stone can
522 * then assign them their right values
524 for (int i = 0, delta = nlen[1] - nlen[0]; i <= nlen[0]; i++)
525 J[i] = i <= pref ? i :
526 i > (nlen[0] - suff) ? (i + delta) : 0;
527 /* Here the magic is performed */
528 stone(class, slen[0], member, J, pref);
529 J[nlen[0] + 1] = nlen[1] + 1;
534 /* Both files are rescanned, in an effort to find any lines
535 * which, due to limitations intrinsic to any hashing algorithm,
536 * are different but ended up confounded as the same
538 for (int i = 1; i <= nlen[0]; i++) {
542 seek_ft(&ft[0], ix[0][i - 1]);
543 seek_ft(&ft[1], ix[1][J[i] - 1]);
545 for (int j = J[i]; i <= nlen[0] && J[i] == j; i++, j++) {
546 token_t tok0 = 0, tok1 = 0;
548 tok0 = read_token(&ft[0], tok0);
549 tok1 = read_token(&ft[1], tok1);
551 if (((tok0 ^ tok1) & TOK_EMPTY) != 0 /* one is empty (not both) */
552 || (!(tok0 & TOK_EMPTY) && TOK2CHAR(tok0) != TOK2CHAR(tok1))
554 J[i] = 0; /* Break the correspondence */
556 } while (!(tok0 & tok1 & TOK_EMPTY));
564 * The following struct is used to record change information
565 * doing a "context" or "unified" diff.
568 int a; /* start line in old file */
569 int b; /* end line in old file */
570 int c; /* start line in new file */
571 int d; /* end line in new file */
574 static bool diff(FILE* fp[2], char *file[2])
578 FILE_and_pos_t ft[2] = { { fp[0] }, { fp[1] } };
579 int *J = create_J(ft, nlen, ix);
581 bool anychange = false;
582 struct context_vec *vec = NULL;
586 bool nonempty = false;
589 struct context_vec v;
591 for (v.a = i; v.a <= nlen[0] && J[v.a] == J[v.a - 1] + 1; v.a++)
593 v.c = J[v.a - 1] + 1;
595 for (v.b = v.a - 1; v.b < nlen[0] && !J[v.b + 1]; v.b++)
597 v.d = J[v.b + 1] - 1;
599 * Indicate that there is a difference between lines a and b of the 'from' file
600 * to get to lines c to d of the 'to' file. If a is greater than b then there
601 * are no lines in the 'from' file involved and this means that there were
602 * lines appended (beginning at b). If c is greater than d then there are
603 * lines missing from the 'to' file.
605 if (v.a <= v.b || v.c <= v.d) {
607 * If this change is more than 'context' lines from the
608 * previous change, dump the record and reset it.
611 && v.a > vec[idx].b + (2 * opt_U_context) + 1
612 && v.c > vec[idx].d + (2 * opt_U_context) + 1
616 nonempty |= (v.a >= v.b) && (v.c >= v.d);
617 vec = xrealloc_vector(vec, 6, ++idx);
628 if (!(option_mask32 & FLAG(q)) && !((option_mask32 & FLAG(B)) && !nonempty)) {
629 struct context_vec *cvp = vec;
630 int lowa = MAX(1, cvp->a - opt_U_context);
631 int upb = MIN(nlen[0], vec[idx].b + opt_U_context);
632 int lowc = MAX(1, cvp->c - opt_U_context);
633 int upd = MIN(nlen[1], vec[idx].d + opt_U_context);
636 /* Print the context/unidiff header first time through */
637 printf("--- %s\n", label[0] ?: file[0]);
638 printf("+++ %s\n", label[1] ?: file[1]);
642 uni_range(lowa, upb);
644 uni_range(lowc, upd);
648 * Output changes in "unified" diff format--the old and new lines
649 * are printed together.
652 bool end = cvp > &vec[idx];
653 fetch(&ft[0], ix[0], lowa, end ? upb : cvp->a - 1, ' ');
656 fetch(&ft[0], ix[0], cvp->a, cvp->b, '-');
657 fetch(&ft[1], ix[1], cvp->c, cvp->d, '+');
663 } while (i <= nlen[0]);
672 static int diffreg(char *file[2])
675 bool binary = false, differ = false;
676 int status = STATUS_SAME;
678 for (int i = 0; i < 2; i++) {
679 int fd = open_or_warn_stdin(file[i]);
682 /* Our diff implementation is using seek.
683 * When we meet non-seekable file, we must make a temp copy.
685 if (lseek(fd, 0, SEEK_SET) == -1 && errno == ESPIPE) {
686 char name[] = "/tmp/difXXXXXX";
687 int fd_tmp = mkstemp(name);
689 bb_perror_msg_and_die("mkstemp");
691 if (bb_copyfd_eof(fd, fd_tmp) < 0)
693 if (fd) /* Prevents closing of stdin */
697 fp[i] = fdopen(fd, "r");
701 const size_t sz = COMMON_BUFSIZE / 2;
702 char *const buf0 = bb_common_bufsiz1;
703 char *const buf1 = buf0 + sz;
705 i = fread(buf0, 1, sz, fp[0]);
706 j = fread(buf1, 1, sz, fp[1]);
713 for (int k = 0; k < i; k++) {
714 if (!buf0[k] || !buf1[k])
716 if (buf0[k] != buf1[k])
721 if (binary && !(option_mask32 & FLAG(a)))
722 status = STATUS_BINARY;
723 else if (diff(fp, file))
724 status = STATUS_DIFFER;
726 if (status != STATUS_SAME)
729 fclose_if_not_stdin(fp[0]);
730 fclose_if_not_stdin(fp[1]);
735 static void print_status(int status, char *path[2])
740 if ((option_mask32 & FLAG(q)) || status == STATUS_BINARY)
741 printf("Files %s and %s differ\n", path[0], path[1]);
744 if (option_mask32 & FLAG(s))
745 printf("Files %s and %s are identical\n", path[0], path[1]);
750 #if ENABLE_FEATURE_DIFF_DIR
757 /* This function adds a filename to dl, the directory listing. */
758 static int FAST_FUNC add_to_dirlist(const char *filename,
759 struct stat *sb UNUSED_PARAM,
760 void *userdata, int depth UNUSED_PARAM)
762 struct dlist *const l = userdata;
763 l->dl = xrealloc_vector(l->dl, 6, l->e);
764 /* + 1 skips "/" after dirname */
765 l->dl[l->e] = xstrdup(filename + l->len + 1);
770 /* If recursion is not set, this function adds the directory
771 * to the list and prevents recursive_action from recursing into it.
773 static int FAST_FUNC skip_dir(const char *filename,
774 struct stat *sb, void *userdata,
777 if (!(option_mask32 & FLAG(r)) && depth) {
778 add_to_dirlist(filename, sb, userdata, depth);
784 static void diffdir(char *p[2], const char *s_start)
786 struct dlist list[2];
788 memset(&list, 0, sizeof(list));
789 for (int i = 0; i < 2; i++) {
790 /*list[i].s = list[i].e = 0; - memset did it */
791 /*list[i].dl = NULL; */
793 /* We need to trim root directory prefix.
794 * Using list.len to specify its length,
795 * add_to_dirlist will remove it. */
796 list[i].len = strlen(p[i]);
797 recursive_action(p[i], ACTION_RECURSE | ACTION_FOLLOWLINKS,
798 add_to_dirlist, skip_dir, &list[i], 0);
799 /* Sort dl alphabetically.
800 * GNU diff does this ignoring any number of trailing dots.
801 * We don't, so for us dotted files almost always are
804 qsort_string_vector(list[i].dl, list[i].e);
805 /* If -S was set, find the starting point. */
808 while (list[i].s < list[i].e && strcmp(list[i].dl[list[i].s], s_start) < 0)
811 /* Now that both dirlist1 and dirlist2 contain sorted directory
812 * listings, we can start to go through dirlist1. If both listings
813 * contain the same file, then do a normal diff. Otherwise, behaviour
814 * is determined by whether the -N flag is set. */
820 dp[0] = list[0].s < list[0].e ? list[0].dl[list[0].s] : NULL;
821 dp[1] = list[1].s < list[1].e ? list[1].dl[list[1].s] : NULL;
822 if (!dp[0] && !dp[1])
824 pos = !dp[0] ? 1 : (!dp[1] ? -1 : strcmp(dp[0], dp[1]));
826 if (pos && !(option_mask32 & FLAG(N)))
827 printf("Only in %s: %s\n", p[k], dp[k]);
829 char *fullpath[2], *path[2]; /* if -N */
831 for (int i = 0; i < 2; i++) {
832 if (pos == 0 || i == k) {
833 path[i] = fullpath[i] = concat_path_file(p[i], dp[i]);
834 stat(fullpath[i], &stb[i]);
836 fullpath[i] = concat_path_file(p[i], dp[1 - i]);
837 path[i] = (char *)bb_dev_null;
841 stat(fullpath[k], &stb[1 - k]);
843 if (S_ISDIR(stb[0].st_mode) && S_ISDIR(stb[1].st_mode))
844 printf("Common subdirectories: %s and %s\n", fullpath[0], fullpath[1]);
845 else if (!S_ISREG(stb[0].st_mode) && !S_ISDIR(stb[0].st_mode))
846 printf("File %s is not a regular file or directory and was skipped\n", fullpath[0]);
847 else if (!S_ISREG(stb[1].st_mode) && !S_ISDIR(stb[1].st_mode))
848 printf("File %s is not a regular file or directory and was skipped\n", fullpath[1]);
849 else if (S_ISDIR(stb[0].st_mode) != S_ISDIR(stb[1].st_mode)) {
850 if (S_ISDIR(stb[0].st_mode))
851 printf("File %s is a %s while file %s is a %s\n", fullpath[0], "directory", fullpath[1], "regular file");
853 printf("File %s is a %s while file %s is a %s\n", fullpath[0], "regular file", fullpath[1], "directory");
855 print_status(diffreg(path), fullpath);
867 if (ENABLE_FEATURE_CLEAN_UP) {
874 #if ENABLE_FEATURE_DIFF_LONG_OPTIONS
875 static const char diff_longopts[] ALIGN1 =
876 "ignore-case\0" No_argument "i"
877 "ignore-tab-expansion\0" No_argument "E"
878 "ignore-space-change\0" No_argument "b"
879 "ignore-all-space\0" No_argument "w"
880 "ignore-blank-lines\0" No_argument "B"
881 "text\0" No_argument "a"
882 "unified\0" Required_argument "U"
883 "label\0" Required_argument "L"
884 "show-c-function\0" No_argument "p"
885 "brief\0" No_argument "q"
886 "expand-tabs\0" No_argument "t"
887 "initial-tab\0" No_argument "T"
888 "recursive\0" No_argument "r"
889 "new-file\0" No_argument "N"
890 "report-identical-files\0" No_argument "s"
891 "starting-file\0" Required_argument "S"
892 "minimal\0" No_argument "d"
896 int diff_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
897 int diff_main(int argc UNUSED_PARAM, char **argv)
900 char *file[2], *s_start = NULL;
901 llist_t *L_arg = NULL;
905 /* exactly 2 params; collect multiple -L <label>; -U N */
906 opt_complementary = "=2:L::U+";
907 #if ENABLE_FEATURE_DIFF_LONG_OPTIONS
908 applet_long_options = diff_longopts;
910 getopt32(argv, "abdiL:NqrsS:tTU:wupBE",
911 &L_arg, &s_start, &opt_U_context);
914 label[!!label[0]] = llist_pop(&L_arg);
915 xfunc_error_retval = 2;
916 for (int i = 0; i < 2; i++) {
918 /* Compat: "diff file name_which_doesnt_exist" exits with 2 */
919 if (LONE_DASH(file[i])) {
920 fstat(STDIN_FILENO, &stb[i]);
923 xstat(file[i], &stb[i]);
925 xfunc_error_retval = 1;
926 if (gotstdin && (S_ISDIR(stb[0].st_mode) || S_ISDIR(stb[1].st_mode)))
927 bb_error_msg_and_die("can't compare stdin to a directory");
929 if (S_ISDIR(stb[0].st_mode) && S_ISDIR(stb[1].st_mode)) {
930 #if ENABLE_FEATURE_DIFF_DIR
931 diffdir(file, s_start);
933 bb_error_msg_and_die("no support for directory comparison");
936 bool dirfile = S_ISDIR(stb[0].st_mode) || S_ISDIR(stb[1].st_mode);
937 bool dir = S_ISDIR(stb[1].st_mode);
939 const char *slash = strrchr(file[!dir], '/');
940 file[dir] = concat_path_file(file[dir], slash ? slash + 1 : file[!dir]);
941 xstat(file[dir], &stb[dir]);
943 /* diffreg can get non-regular files here */
944 print_status(gotstdin > 1 ? STATUS_SAME : diffreg(file), file);