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 source tree.
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.
83 //config: diff compares two files or directories and outputs the
84 //config: differences between them in a form that can be given to
85 //config: the patch command.
87 //config:config FEATURE_DIFF_LONG_OPTIONS
88 //config: bool "Enable long options"
90 //config: depends on DIFF && LONG_OPTS
92 //config:config FEATURE_DIFF_DIR
93 //config: bool "Enable directory support"
95 //config: depends on DIFF
97 //config: This option enables support for directory and subdirectory
100 //kbuild:lib-$(CONFIG_DIFF) += diff.o
102 //applet:IF_DIFF(APPLET(diff, BB_DIR_USR_BIN, BB_SUID_DROP))
104 //usage:#define diff_trivial_usage
105 //usage: "[-abBdiNqrTstw] [-L LABEL] [-S FILE] [-U LINES] FILE1 FILE2"
106 //usage:#define diff_full_usage "\n\n"
107 //usage: "Compare files line by line and output the differences between them.\n"
108 //usage: "This implementation supports unified diffs only.\n"
109 //usage: "\n -a Treat all files as text"
110 //usage: "\n -b Ignore changes in the amount of whitespace"
111 //usage: "\n -B Ignore changes whose lines are all blank"
112 //usage: "\n -d Try hard to find a smaller set of changes"
113 //usage: "\n -i Ignore case differences"
114 //usage: "\n -L Use LABEL instead of the filename in the unified header"
115 //usage: "\n -N Treat absent files as empty"
116 //usage: "\n -q Output only whether files differ"
117 //usage: "\n -r Recurse"
118 //usage: "\n -S Start with FILE when comparing directories"
119 //usage: "\n -T Make tabs line up by prefixing a tab when necessary"
120 //usage: "\n -s Report when two files are the same"
121 //usage: "\n -t Expand tabs to spaces in output"
122 //usage: "\n -U Output LINES lines of context"
123 //usage: "\n -w Ignore all whitespace"
126 #include "common_bufsiz.h"
129 # define dbg_error_msg(...) bb_error_msg(__VA_ARGS__)
131 # define dbg_error_msg(...) ((void)0)
134 enum { /* print_status() and diffreg() return values */
135 STATUS_SAME, /* files are the same */
136 STATUS_DIFFER, /* files differ */
137 STATUS_BINARY, /* binary files differ */
140 enum { /* Commandline flags */
145 FLAG_L, /* never used, handled by getopt32 */
150 FLAG_S, /* never used, handled by getopt32 */
153 FLAG_U, /* never used, handled by getopt32 */
155 FLAG_u, /* ignored, this is the default */
156 FLAG_p, /* not implemented */
158 FLAG_E, /* not implemented */
160 #define FLAG(x) (1 << FLAG_##x)
162 /* We cache file position to avoid excessive seeking */
163 typedef struct FILE_and_pos_t {
169 smallint exit_status;
171 const char *other_dir;
175 #define G (*ptr_to_globals)
176 #define exit_status (G.exit_status )
177 #define opt_U_context (G.opt_U_context )
178 #define label (G.label )
180 #define INIT_G() do { \
181 SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
189 TOK_EMPTY = 1 << 9, /* Line fully processed, you can proceed to the next */
190 TOK_EOF = 1 << 10, /* File ended */
191 /* Private (Only to be used by read_token() */
192 TOK_EOL = 1 << 11, /* we saw EOL (sticky) */
193 TOK_SPACE = 1 << 12, /* used -b code, means we are skipping spaces */
194 SHIFT_EOF = (sizeof(token_t)*8 - 8) - 1,
195 CHAR_MASK = 0x1ff, /* 8th bit is used to distinguish EOF from 0xff */
198 /* Restores full EOF from one 8th bit: */
199 //#define TOK2CHAR(t) (((t) << SHIFT_EOF) >> SHIFT_EOF)
200 /* We don't really need the above, we only need to have EOF != any_real_char: */
201 #define TOK2CHAR(t) ((t) & CHAR_MASK)
203 static void seek_ft(FILE_and_pos_t *ft, off_t pos)
205 if (ft->ft_pos != pos) {
207 fseeko(ft->ft_fp, pos, SEEK_SET);
211 /* Reads tokens from given fp, handling -b and -w flags
212 * The user must reset tok every line start
214 static int read_token(FILE_and_pos_t *ft, token_t tok)
217 while (!(tok & TOK_EOL)) {
221 t = fgetc(ft->ft_fp);
224 is_space = (t == EOF || isspace(t));
226 /* If t == EOF (-1), set both TOK_EOF and TOK_EOL */
227 tok |= (t & (TOK_EOF + TOK_EOL));
232 if (option_mask32 & FLAG(i)) /* Handcoded tolower() */
233 t = (t >= 'A' && t <= 'Z') ? t - ('A' - 'a') : t;
235 if ((option_mask32 & FLAG(w)) && is_space)
238 /* Trim char value to low 9 bits */
241 if (option_mask32 & FLAG(b)) {
242 /* Was prev char whitespace? */
243 if (tok & TOK_SPACE) { /* yes */
244 if (is_space) /* this one too, ignore it */
247 } else if (is_space) {
248 /* 1st whitespace char.
249 * Set TOK_SPACE and replace char by ' ' */
254 tok &= ~(TOK_EMPTY + CHAR_MASK);
255 /* Assign char value (low 9 bits) and maybe set TOK_SPACE */
260 bb_error_msg("fp:%p tok:%x '%c'%s%s%s%s", fp, tok, tok & 0xff
261 , tok & TOK_EOF ? " EOF" : ""
262 , tok & TOK_EOL ? " EOL" : ""
263 , tok & TOK_EMPTY ? " EMPTY" : ""
264 , tok & TOK_SPACE ? " SPACE" : ""
276 static int search(const int *c, int k, int y, const struct cand *list)
280 if (list[c[k]].y < y) /* quick look for typical case */
283 for (i = 0, j = k + 1;;) {
284 const int l = (i + j) >> 1;
286 const int t = list[c[l]].y;
298 static unsigned isqrt(unsigned n)
302 const unsigned y = x;
303 x = ((n / x) + x) >> 1;
304 if (x <= (y + 1) && x >= (y - 1))
309 static void stone(const int *a, int n, const int *b, int *J, int pref)
311 const unsigned isq = isqrt(n);
312 const unsigned bound =
313 (option_mask32 & FLAG(d)) ? UINT_MAX : MAX(256, isq);
317 struct cand *clist = xzalloc(clistlen * sizeof(clist[0]));
320 int *klist = xzalloc((n + 2) * sizeof(klist[0]));
321 /*clist[0] = (struct cand){0}; - xzalloc did it */
324 for (cand.x = 1; cand.x <= n; cand.x++) {
325 int j = a[cand.x], oldl = 0;
326 unsigned numtries = 0;
330 cand.pred = klist[0];
333 if (cand.y <= clist[cand.pred].y)
335 l = search(klist, k, cand.y, clist);
337 cand.pred = klist[l - 1];
338 if (l <= k && clist[klist[l]].y <= cand.y)
340 if (clen == clistlen) {
341 clistlen = clistlen * 11 / 10;
342 clist = xrealloc(clist, clistlen * sizeof(clist[0]));
355 } while ((cand.y = b[++j]) > 0 && numtries < bound);
358 for (q = clist + klist[k]; q->y; q = clist + q->pred)
359 J[q->x + pref] = q->y + pref;
365 /* 'serial' is not used in the beginning, so we reuse it
366 * to store line offsets, thus reducing memory pressure
375 static void equiv(struct line *a, int n, struct line *b, int m, int *c)
379 while (i <= n && j <= m) {
380 if (a[i].value < b[j].value)
382 else if (a[i].value == b[j].value)
393 while (b[j + 1].value == b[j].value) {
401 static void unsort(const struct line *f, int l, int *b)
404 int *a = xmalloc((l + 1) * sizeof(a[0]));
405 for (i = 1; i <= l; i++)
406 a[f[i].serial] = f[i].value;
407 for (i = 1; i <= l; i++)
412 static int line_compar(const void *a, const void *b)
414 #define l0 ((const struct line*)a)
415 #define l1 ((const struct line*)b)
416 int r = l0->value - l1->value;
419 return l0->serial - l1->serial;
424 static void fetch(FILE_and_pos_t *ft, const off_t *ix, int a, int b, int ch)
427 for (i = a; i <= b; i++) {
428 seek_ft(ft, ix[i - 1]);
430 if (option_mask32 & FLAG(T))
432 for (j = 0, col = 0; j < ix[i] - ix[i - 1]; j++) {
433 int c = fgetc(ft->ft_fp);
435 puts("\n\\ No newline at end of file");
439 if (c == '\t' && (option_mask32 & FLAG(t)))
440 do putchar(' '); while (++col & 7);
449 /* Creates the match vector J, where J[i] is the index
450 * of the line in the new file corresponding to the line i
451 * in the old file. Lines start at 1 instead of 0, that value
452 * being used instead to denote no corresponding line.
453 * This vector is dynamically allocated and must be freed by the caller.
455 * * fp is an input parameter, where fp[0] and fp[1] are the open
456 * old file and new file respectively.
457 * * nlen is an output variable, where nlen[0] and nlen[1]
458 * gets the number of lines in the old and new file respectively.
459 * * ix is an output variable, where ix[0] and ix[1] gets
460 * assigned dynamically allocated vectors of the offsets of the lines
461 * of the old and new file respectively. These must be freed by the caller.
463 static NOINLINE int *create_J(FILE_and_pos_t ft[2], int nlen[2], off_t *ix[2])
465 int *J, slen[2], *class, *member;
466 struct line *nfile[2], *sfile[2];
467 int pref = 0, suff = 0, i, j, delta;
469 /* Lines of both files are hashed, and in the process
470 * their offsets are stored in the array ix[fileno]
471 * where fileno == 0 points to the old file, and
472 * fileno == 1 points to the new one.
474 for (i = 0; i < 2; i++) {
478 nfile[i] = xmalloc((sz + 3) * sizeof(nfile[i][0]));
479 /* ft gets here without the correct position, cant use seek_ft */
481 fseeko(ft[i].ft_fp, 0, SEEK_SET);
484 /* We could zalloc nfile, but then zalloc starts showing in gprof at ~1% */
485 nfile[i][0].offset = 0;
486 goto start; /* saves code */
488 tok = read_token(&ft[i], tok);
489 if (!(tok & TOK_EMPTY)) {
490 /* Hash algorithm taken from Robert Sedgewick, Algorithms in C, 3d ed., p 578. */
491 /*hash = hash * 128 - hash + TOK2CHAR(tok);
492 * gcc insists on optimizing above to "hash * 127 + ...", thus... */
493 unsigned o = hash - TOK2CHAR(tok);
494 hash = hash * 128 - o; /* we want SPEED here */
497 if (nlen[i]++ == sz) {
499 nfile[i] = xrealloc(nfile[i], (sz + 3) * sizeof(nfile[i][0]));
501 /* line_compar needs hashes fit into positive int */
502 nfile[i][nlen[i]].value = hash & INT_MAX;
503 /* like ftello(ft[i].ft_fp) but faster (avoids lseek syscall) */
504 nfile[i][nlen[i]].offset = ft[i].ft_pos;
506 /* EOF counts as a token, so we have to adjust it here */
507 nfile[i][nlen[i]].offset++;
513 /* Exclude lone EOF line from the end of the file, to make fetch()'s job easier */
514 if (nfile[i][nlen[i]].offset - nfile[i][nlen[i] - 1].offset == 1)
516 /* Now we copy the line offsets into ix */
517 ix[i] = xmalloc((nlen[i] + 2) * sizeof(ix[i][0]));
518 for (j = 0; j < nlen[i] + 1; j++)
519 ix[i][j] = nfile[i][j].offset;
522 /* length of prefix and suffix is calculated */
523 for (; pref < nlen[0] && pref < nlen[1] &&
524 nfile[0][pref + 1].value == nfile[1][pref + 1].value;
526 for (; suff < nlen[0] - pref && suff < nlen[1] - pref &&
527 nfile[0][nlen[0] - suff].value == nfile[1][nlen[1] - suff].value;
529 /* Arrays are pruned by the suffix and prefix length,
530 * the result being sorted and stored in sfile[fileno],
531 * and their sizes are stored in slen[fileno]
533 for (j = 0; j < 2; j++) {
534 sfile[j] = nfile[j] + pref;
535 slen[j] = nlen[j] - pref - suff;
536 for (i = 0; i <= slen[j]; i++)
537 sfile[j][i].serial = i;
538 qsort(sfile[j] + 1, slen[j], sizeof(*sfile[j]), line_compar);
540 /* nfile arrays are reused to reduce memory pressure
541 * The #if zeroed out section performs the same task as the
542 * one in the #else section.
543 * Peak memory usage is higher, but one array copy is avoided
544 * by not using unsort()
547 member = xmalloc((slen[1] + 2) * sizeof(member[0]));
548 equiv(sfile[0], slen[0], sfile[1], slen[1], member);
551 class = xmalloc((slen[0] + 1) * sizeof(class[0]));
552 for (i = 1; i <= slen[0]; i++) /* Unsorting */
553 class[sfile[0][i].serial] = sfile[0][i].value;
556 member = (int *)nfile[1];
557 equiv(sfile[0], slen[0], sfile[1], slen[1], member);
558 member = xrealloc(member, (slen[1] + 2) * sizeof(member[0]));
560 class = (int *)nfile[0];
561 unsort(sfile[0], slen[0], (int *)nfile[0]);
562 class = xrealloc(class, (slen[0] + 2) * sizeof(class[0]));
564 J = xmalloc((nlen[0] + 2) * sizeof(J[0]));
565 /* The elements of J which fall inside the prefix and suffix regions
566 * are marked as unchanged, while the ones which fall outside
567 * are initialized with 0 (no matches), so that function stone can
568 * then assign them their right values
570 for (i = 0, delta = nlen[1] - nlen[0]; i <= nlen[0]; i++)
571 J[i] = i <= pref ? i :
572 i > (nlen[0] - suff) ? (i + delta) : 0;
573 /* Here the magic is performed */
574 stone(class, slen[0], member, J, pref);
575 J[nlen[0] + 1] = nlen[1] + 1;
580 /* Both files are rescanned, in an effort to find any lines
581 * which, due to limitations intrinsic to any hashing algorithm,
582 * are different but ended up confounded as the same
584 for (i = 1; i <= nlen[0]; i++) {
588 seek_ft(&ft[0], ix[0][i - 1]);
589 seek_ft(&ft[1], ix[1][J[i] - 1]);
591 for (j = J[i]; i <= nlen[0] && J[i] == j; i++, j++) {
592 token_t tok0 = 0, tok1 = 0;
594 tok0 = read_token(&ft[0], tok0);
595 tok1 = read_token(&ft[1], tok1);
597 if (((tok0 ^ tok1) & TOK_EMPTY) != 0 /* one is empty (not both) */
598 || (!(tok0 & TOK_EMPTY) && TOK2CHAR(tok0) != TOK2CHAR(tok1))
600 J[i] = 0; /* Break the correspondence */
602 } while (!(tok0 & tok1 & TOK_EMPTY));
609 static bool diff(FILE* fp[2], char *file[2])
613 FILE_and_pos_t ft[2];
614 typedef struct { int a, b; } vec_t[2];
616 int i = 1, j, k, idx = -1;
617 bool anychange = false;
622 /* note that ft[i].ft_pos is unintitalized, create_J()
623 * must not assume otherwise */
624 J = create_J(ft, nlen, ix);
627 bool nonempty = false;
632 for (v[0].a = i; v[0].a <= nlen[0] && J[v[0].a] == J[v[0].a - 1] + 1; v[0].a++)
634 v[1].a = J[v[0].a - 1] + 1;
636 for (v[0].b = v[0].a - 1; v[0].b < nlen[0] && !J[v[0].b + 1]; v[0].b++)
638 v[1].b = J[v[0].b + 1] - 1;
640 * Indicate that there is a difference between lines a and b of the 'from' file
641 * to get to lines c to d of the 'to' file. If a is greater than b then there
642 * are no lines in the 'from' file involved and this means that there were
643 * lines appended (beginning at b). If c is greater than d then there are
644 * lines missing from the 'to' file.
646 if (v[0].a <= v[0].b || v[1].a <= v[1].b) {
648 * If this change is more than 'context' lines from the
649 * previous change, dump the record and reset it.
651 int ct = (2 * opt_U_context) + 1;
653 && v[0].a > vec[idx][0].b + ct
654 && v[1].a > vec[idx][1].b + ct
659 for (j = 0; j < 2; j++)
660 for (k = v[j].a; k <= v[j].b; k++)
661 nonempty |= (ix[j][k] - ix[j][k - 1] != 1);
663 vec = xrealloc_vector(vec, 6, ++idx);
664 memcpy(vec[idx], v, sizeof(v));
672 if (idx < 0 || ((option_mask32 & FLAG(B)) && !nonempty))
674 if (!(option_mask32 & FLAG(q))) {
676 vec_t span, *cvp = vec;
679 /* Print the context/unidiff header first time through */
680 printf("--- %s\n", label[0] ? label[0] : file[0]);
681 printf("+++ %s\n", label[1] ? label[1] : file[1]);
685 for (j = 0; j < 2; j++) {
686 int a = span[j].a = MAX(1, (*cvp)[j].a - opt_U_context);
687 int b = span[j].b = MIN(nlen[j], vec[idx][j].b + opt_U_context);
689 printf(" %c%d", j ? '+' : '-', MIN(a, b));
692 printf(",%d", (a < b) ? b - a + 1 : 0);
696 * Output changes in "unified" diff format--the old and new lines
697 * are printed together.
699 for (lowa = span[0].a; ; lowa = (*cvp++)[0].b + 1) {
700 bool end = cvp > &vec[idx];
701 fetch(&ft[0], ix[0], lowa, end ? span[0].b : (*cvp)[0].a - 1, ' ');
704 for (j = 0; j < 2; j++)
705 fetch(&ft[j], ix[j], (*cvp)[j].a, (*cvp)[j].b, j ? '+' : '-');
711 } while (i <= nlen[0]);
720 static int diffreg(char *file[2])
723 bool binary = false, differ = false;
724 int status = STATUS_SAME, i;
728 for (i = 0; i < 2; i++) {
729 int fd = open_or_warn_stdin(file[i]);
732 /* Our diff implementation is using seek.
733 * When we meet non-seekable file, we must make a temp copy.
735 if (lseek(fd, 0, SEEK_SET) == -1 && errno == ESPIPE) {
736 char name[] = "/tmp/difXXXXXX";
737 int fd_tmp = xmkstemp(name);
740 if (bb_copyfd_eof(fd, fd_tmp) < 0)
742 if (fd != STDIN_FILENO)
745 xlseek(fd, 0, SEEK_SET);
747 fp[i] = fdopen(fd, "r");
750 setup_common_bufsiz();
752 const size_t sz = COMMON_BUFSIZE / 2;
753 char *const buf0 = bb_common_bufsiz1;
754 char *const buf1 = buf0 + sz;
756 i = fread(buf0, 1, sz, fp[0]);
757 j = fread(buf1, 1, sz, fp[1]);
764 for (k = 0; k < i; k++) {
765 if (!buf0[k] || !buf1[k])
767 if (buf0[k] != buf1[k])
772 if (binary && !(option_mask32 & FLAG(a)))
773 status = STATUS_BINARY;
774 else if (diff(fp, file))
775 status = STATUS_DIFFER;
777 if (status != STATUS_SAME)
780 fclose_if_not_stdin(fp[0]);
781 fclose_if_not_stdin(fp[1]);
786 static void print_status(int status, char *path[2])
791 if ((option_mask32 & FLAG(q)) || status == STATUS_BINARY)
792 printf("Files %s and %s differ\n", path[0], path[1]);
795 if (option_mask32 & FLAG(s))
796 printf("Files %s and %s are identical\n", path[0], path[1]);
801 #if ENABLE_FEATURE_DIFF_DIR
808 /* This function adds a filename to dl, the directory listing. */
809 static int FAST_FUNC add_to_dirlist(const char *filename,
810 struct stat *sb UNUSED_PARAM,
811 void *userdata, int depth UNUSED_PARAM)
813 struct dlist *const l = userdata;
814 const char *file = filename + l->len;
817 l->dl = xrealloc_vector(l->dl, 6, l->e);
818 l->dl[l->e] = xstrdup(file);
823 /* If recursion is not set, this function adds the directory
824 * to the list and prevents recursive_action from recursing into it.
826 static int FAST_FUNC skip_dir(const char *filename,
827 struct stat *sb, void *userdata,
830 if (!(option_mask32 & FLAG(r)) && depth) {
831 add_to_dirlist(filename, sb, userdata, depth);
834 if (!(option_mask32 & FLAG(N))) {
835 /* -r without -N: no need to recurse into dirs
836 * which do not exist on the "other side".
837 * Testcase: diff -r /tmp /
838 * (it would recurse deep into /proc without this code) */
839 struct dlist *const l = userdata;
843 char *othername = concat_path_file(G.other_dir, filename);
844 int r = stat(othername, &osb);
846 if (r != 0 || !S_ISDIR(osb.st_mode)) {
847 /* other dir doesn't have similarly named
848 * directory, don't recurse; return 1 upon
849 * exit, just like diffutils' diff */
858 static void diffdir(char *p[2], const char *s_start)
860 struct dlist list[2];
863 memset(&list, 0, sizeof(list));
864 for (i = 0; i < 2; i++) {
865 /*list[i].s = list[i].e = 0; - memset did it */
866 /*list[i].dl = NULL; */
868 G.other_dir = p[1 - i];
869 /* We need to trim root directory prefix.
870 * Using list.len to specify its length,
871 * add_to_dirlist will remove it. */
872 list[i].len = strlen(p[i]);
873 recursive_action(p[i], ACTION_RECURSE | ACTION_FOLLOWLINKS,
874 add_to_dirlist, skip_dir, &list[i], 0);
875 /* Sort dl alphabetically.
876 * GNU diff does this ignoring any number of trailing dots.
877 * We don't, so for us dotted files almost always are
880 qsort_string_vector(list[i].dl, list[i].e);
881 /* If -S was set, find the starting point. */
884 while (list[i].s < list[i].e && strcmp(list[i].dl[list[i].s], s_start) < 0)
887 /* Now that both dirlist1 and dirlist2 contain sorted directory
888 * listings, we can start to go through dirlist1. If both listings
889 * contain the same file, then do a normal diff. Otherwise, behaviour
890 * is determined by whether the -N flag is set. */
896 dp[0] = list[0].s < list[0].e ? list[0].dl[list[0].s] : NULL;
897 dp[1] = list[1].s < list[1].e ? list[1].dl[list[1].s] : NULL;
898 if (!dp[0] && !dp[1])
900 pos = !dp[0] ? 1 : (!dp[1] ? -1 : strcmp(dp[0], dp[1]));
902 if (pos && !(option_mask32 & FLAG(N))) {
903 printf("Only in %s: %s\n", p[k], dp[k]);
906 char *fullpath[2], *path[2]; /* if -N */
908 for (i = 0; i < 2; i++) {
909 if (pos == 0 || i == k) {
910 path[i] = fullpath[i] = concat_path_file(p[i], dp[i]);
911 stat(fullpath[i], &stb[i]);
913 fullpath[i] = concat_path_file(p[i], dp[1 - i]);
914 path[i] = (char *)bb_dev_null;
918 stat(fullpath[k], &stb[1 - k]);
920 if (S_ISDIR(stb[0].st_mode) && S_ISDIR(stb[1].st_mode))
921 printf("Common subdirectories: %s and %s\n", fullpath[0], fullpath[1]);
922 else if (!S_ISREG(stb[0].st_mode) && !S_ISDIR(stb[0].st_mode))
923 printf("File %s is not a regular file or directory and was skipped\n", fullpath[0]);
924 else if (!S_ISREG(stb[1].st_mode) && !S_ISDIR(stb[1].st_mode))
925 printf("File %s is not a regular file or directory and was skipped\n", fullpath[1]);
926 else if (S_ISDIR(stb[0].st_mode) != S_ISDIR(stb[1].st_mode)) {
927 if (S_ISDIR(stb[0].st_mode))
928 printf("File %s is a %s while file %s is a %s\n", fullpath[0], "directory", fullpath[1], "regular file");
930 printf("File %s is a %s while file %s is a %s\n", fullpath[0], "regular file", fullpath[1], "directory");
932 print_status(diffreg(path), fullpath);
944 if (ENABLE_FEATURE_CLEAN_UP) {
951 #if ENABLE_FEATURE_DIFF_LONG_OPTIONS
952 static const char diff_longopts[] ALIGN1 =
953 "ignore-case\0" No_argument "i"
954 "ignore-tab-expansion\0" No_argument "E"
955 "ignore-space-change\0" No_argument "b"
956 "ignore-all-space\0" No_argument "w"
957 "ignore-blank-lines\0" No_argument "B"
958 "text\0" No_argument "a"
959 "unified\0" Required_argument "U"
960 "label\0" Required_argument "L"
961 "show-c-function\0" No_argument "p"
962 "brief\0" No_argument "q"
963 "expand-tabs\0" No_argument "t"
964 "initial-tab\0" No_argument "T"
965 "recursive\0" No_argument "r"
966 "new-file\0" No_argument "N"
967 "report-identical-files\0" No_argument "s"
968 "starting-file\0" Required_argument "S"
969 "minimal\0" No_argument "d"
973 int diff_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
974 int diff_main(int argc UNUSED_PARAM, char **argv)
977 char *file[2], *s_start = NULL;
978 llist_t *L_arg = NULL;
982 /* exactly 2 params; collect multiple -L <label>; -U N */
983 opt_complementary = "=2";
984 #if ENABLE_FEATURE_DIFF_LONG_OPTIONS
985 applet_long_options = diff_longopts;
987 getopt32(argv, "abdiL:*NqrsS:tTU:+wupBE",
988 &L_arg, &s_start, &opt_U_context);
991 label[!!label[0]] = llist_pop(&L_arg);
992 xfunc_error_retval = 2;
993 for (i = 0; i < 2; i++) {
995 /* Compat: "diff file name_which_doesnt_exist" exits with 2 */
996 if (LONE_DASH(file[i])) {
997 fstat(STDIN_FILENO, &stb[i]);
1000 xstat(file[i], &stb[i]);
1002 xfunc_error_retval = 1;
1003 if (gotstdin && (S_ISDIR(stb[0].st_mode) || S_ISDIR(stb[1].st_mode)))
1004 bb_error_msg_and_die("can't compare stdin to a directory");
1006 /* Compare metadata to check if the files are the same physical file.
1008 * Comment from diffutils source says:
1009 * POSIX says that two files are identical if st_ino and st_dev are
1010 * the same, but many file systems incorrectly assign the same (device,
1011 * inode) pair to two distinct files, including:
1012 * GNU/Linux NFS servers that export all local file systems as a
1013 * single NFS file system, if a local device number (st_dev) exceeds
1014 * 255, or if a local inode number (st_ino) exceeds 16777215.
1017 && stb[0].st_ino == stb[1].st_ino
1018 && stb[0].st_dev == stb[1].st_dev
1019 && stb[0].st_size == stb[1].st_size
1020 && stb[0].st_mtime == stb[1].st_mtime
1021 && stb[0].st_ctime == stb[1].st_ctime
1022 && stb[0].st_mode == stb[1].st_mode
1023 && stb[0].st_nlink == stb[1].st_nlink
1024 && stb[0].st_uid == stb[1].st_uid
1025 && stb[0].st_gid == stb[1].st_gid
1027 /* files are physically the same; no need to compare them */
1031 if (S_ISDIR(stb[0].st_mode) && S_ISDIR(stb[1].st_mode)) {
1032 #if ENABLE_FEATURE_DIFF_DIR
1033 diffdir(file, s_start);
1035 bb_error_msg_and_die("no support for directory comparison");
1038 bool dirfile = S_ISDIR(stb[0].st_mode) || S_ISDIR(stb[1].st_mode);
1039 bool dir = S_ISDIR(stb[1].st_mode);
1041 const char *slash = strrchr(file[!dir], '/');
1042 file[dir] = concat_path_file(file[dir], slash ? slash + 1 : file[!dir]);
1043 xstat(file[dir], &stb[dir]);
1045 /* diffreg can get non-regular files here */
1046 print_status(gotstdin > 1 ? STATUS_SAME : diffreg(file), file);