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: Enable use of long options.
94 //config:config FEATURE_DIFF_DIR
95 //config: bool "Enable directory support"
97 //config: depends on DIFF
99 //config: This option enables support for directory and subdirectory
100 //config: comparison.
102 //kbuild:lib-$(CONFIG_DIFF) += diff.o
104 //applet:IF_DIFF(APPLET(diff, BB_DIR_USR_BIN, BB_SUID_DROP))
106 //usage:#define diff_trivial_usage
107 //usage: "[-abBdiNqrTstw] [-L LABEL] [-S FILE] [-U LINES] FILE1 FILE2"
108 //usage:#define diff_full_usage "\n\n"
109 //usage: "Compare files line by line and output the differences between them.\n"
110 //usage: "This implementation supports unified diffs only.\n"
111 //usage: "\n -a Treat all files as text"
112 //usage: "\n -b Ignore changes in the amount of whitespace"
113 //usage: "\n -B Ignore changes whose lines are all blank"
114 //usage: "\n -d Try hard to find a smaller set of changes"
115 //usage: "\n -i Ignore case differences"
116 //usage: "\n -L Use LABEL instead of the filename in the unified header"
117 //usage: "\n -N Treat absent files as empty"
118 //usage: "\n -q Output only whether files differ"
119 //usage: "\n -r Recurse"
120 //usage: "\n -S Start with FILE when comparing directories"
121 //usage: "\n -T Make tabs line up by prefixing a tab when necessary"
122 //usage: "\n -s Report when two files are the same"
123 //usage: "\n -t Expand tabs to spaces in output"
124 //usage: "\n -U Output LINES lines of context"
125 //usage: "\n -w Ignore all whitespace"
128 #include "common_bufsiz.h"
131 # define dbg_error_msg(...) bb_error_msg(__VA_ARGS__)
133 # define dbg_error_msg(...) ((void)0)
136 enum { /* print_status() and diffreg() return values */
137 STATUS_SAME, /* files are the same */
138 STATUS_DIFFER, /* files differ */
139 STATUS_BINARY, /* binary files differ */
142 enum { /* Commandline flags */
147 FLAG_L, /* never used, handled by getopt32 */
152 FLAG_S, /* never used, handled by getopt32 */
155 FLAG_U, /* never used, handled by getopt32 */
157 FLAG_u, /* ignored, this is the default */
158 FLAG_p, /* not implemented */
160 FLAG_E, /* not implemented */
162 #define FLAG(x) (1 << FLAG_##x)
164 /* We cache file position to avoid excessive seeking */
165 typedef struct FILE_and_pos_t {
171 smallint exit_status;
173 const char *other_dir;
177 #define G (*ptr_to_globals)
178 #define exit_status (G.exit_status )
179 #define opt_U_context (G.opt_U_context )
180 #define label (G.label )
182 #define INIT_G() do { \
183 SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
191 TOK_EMPTY = 1 << 9, /* Line fully processed, you can proceed to the next */
192 TOK_EOF = 1 << 10, /* File ended */
193 /* Private (Only to be used by read_token() */
194 TOK_EOL = 1 << 11, /* we saw EOL (sticky) */
195 TOK_SPACE = 1 << 12, /* used -b code, means we are skipping spaces */
196 SHIFT_EOF = (sizeof(token_t)*8 - 8) - 1,
197 CHAR_MASK = 0x1ff, /* 8th bit is used to distinguish EOF from 0xff */
200 /* Restores full EOF from one 8th bit: */
201 //#define TOK2CHAR(t) (((t) << SHIFT_EOF) >> SHIFT_EOF)
202 /* We don't really need the above, we only need to have EOF != any_real_char: */
203 #define TOK2CHAR(t) ((t) & CHAR_MASK)
205 static void seek_ft(FILE_and_pos_t *ft, off_t pos)
207 if (ft->ft_pos != pos) {
209 fseeko(ft->ft_fp, pos, SEEK_SET);
213 /* Reads tokens from given fp, handling -b and -w flags
214 * The user must reset tok every line start
216 static int read_token(FILE_and_pos_t *ft, token_t tok)
219 while (!(tok & TOK_EOL)) {
223 t = fgetc(ft->ft_fp);
226 is_space = (t == EOF || isspace(t));
228 /* If t == EOF (-1), set both TOK_EOF and TOK_EOL */
229 tok |= (t & (TOK_EOF + TOK_EOL));
234 if (option_mask32 & FLAG(i)) /* Handcoded tolower() */
235 t = (t >= 'A' && t <= 'Z') ? t - ('A' - 'a') : t;
237 if ((option_mask32 & FLAG(w)) && is_space)
240 /* Trim char value to low 9 bits */
243 if (option_mask32 & FLAG(b)) {
244 /* Was prev char whitespace? */
245 if (tok & TOK_SPACE) { /* yes */
246 if (is_space) /* this one too, ignore it */
249 } else if (is_space) {
250 /* 1st whitespace char.
251 * Set TOK_SPACE and replace char by ' ' */
256 tok &= ~(TOK_EMPTY + CHAR_MASK);
257 /* Assign char value (low 9 bits) and maybe set TOK_SPACE */
262 bb_error_msg("fp:%p tok:%x '%c'%s%s%s%s", fp, tok, tok & 0xff
263 , tok & TOK_EOF ? " EOF" : ""
264 , tok & TOK_EOL ? " EOL" : ""
265 , tok & TOK_EMPTY ? " EMPTY" : ""
266 , tok & TOK_SPACE ? " SPACE" : ""
278 static int search(const int *c, int k, int y, const struct cand *list)
282 if (list[c[k]].y < y) /* quick look for typical case */
285 for (i = 0, j = k + 1;;) {
286 const int l = (i + j) >> 1;
288 const int t = list[c[l]].y;
300 static unsigned isqrt(unsigned n)
304 const unsigned y = x;
305 x = ((n / x) + x) >> 1;
306 if (x <= (y + 1) && x >= (y - 1))
311 static void stone(const int *a, int n, const int *b, int *J, int pref)
313 const unsigned isq = isqrt(n);
314 const unsigned bound =
315 (option_mask32 & FLAG(d)) ? UINT_MAX : MAX(256, isq);
319 struct cand *clist = xzalloc(clistlen * sizeof(clist[0]));
322 int *klist = xzalloc((n + 2) * sizeof(klist[0]));
323 /*clist[0] = (struct cand){0}; - xzalloc did it */
326 for (cand.x = 1; cand.x <= n; cand.x++) {
327 int j = a[cand.x], oldl = 0;
328 unsigned numtries = 0;
332 cand.pred = klist[0];
335 if (cand.y <= clist[cand.pred].y)
337 l = search(klist, k, cand.y, clist);
339 cand.pred = klist[l - 1];
340 if (l <= k && clist[klist[l]].y <= cand.y)
342 if (clen == clistlen) {
343 clistlen = clistlen * 11 / 10;
344 clist = xrealloc(clist, clistlen * sizeof(clist[0]));
357 } while ((cand.y = b[++j]) > 0 && numtries < bound);
360 for (q = clist + klist[k]; q->y; q = clist + q->pred)
361 J[q->x + pref] = q->y + pref;
367 /* 'serial' is not used in the beginning, so we reuse it
368 * to store line offsets, thus reducing memory pressure
377 static void equiv(struct line *a, int n, struct line *b, int m, int *c)
381 while (i <= n && j <= m) {
382 if (a[i].value < b[j].value)
384 else if (a[i].value == b[j].value)
395 while (b[j + 1].value == b[j].value) {
403 static void unsort(const struct line *f, int l, int *b)
406 int *a = xmalloc((l + 1) * sizeof(a[0]));
407 for (i = 1; i <= l; i++)
408 a[f[i].serial] = f[i].value;
409 for (i = 1; i <= l; i++)
414 static int line_compar(const void *a, const void *b)
416 #define l0 ((const struct line*)a)
417 #define l1 ((const struct line*)b)
418 int r = l0->value - l1->value;
421 return l0->serial - l1->serial;
426 static void fetch(FILE_and_pos_t *ft, const off_t *ix, int a, int b, int ch)
429 for (i = a; i <= b; i++) {
430 seek_ft(ft, ix[i - 1]);
432 if (option_mask32 & FLAG(T))
434 for (j = 0, col = 0; j < ix[i] - ix[i - 1]; j++) {
435 int c = fgetc(ft->ft_fp);
437 puts("\n\\ No newline at end of file");
441 if (c == '\t' && (option_mask32 & FLAG(t)))
442 do putchar(' '); while (++col & 7);
451 /* Creates the match vector J, where J[i] is the index
452 * of the line in the new file corresponding to the line i
453 * in the old file. Lines start at 1 instead of 0, that value
454 * being used instead to denote no corresponding line.
455 * This vector is dynamically allocated and must be freed by the caller.
457 * * fp is an input parameter, where fp[0] and fp[1] are the open
458 * old file and new file respectively.
459 * * nlen is an output variable, where nlen[0] and nlen[1]
460 * gets the number of lines in the old and new file respectively.
461 * * ix is an output variable, where ix[0] and ix[1] gets
462 * assigned dynamically allocated vectors of the offsets of the lines
463 * of the old and new file respectively. These must be freed by the caller.
465 static NOINLINE int *create_J(FILE_and_pos_t ft[2], int nlen[2], off_t *ix[2])
467 int *J, slen[2], *class, *member;
468 struct line *nfile[2], *sfile[2];
469 int pref = 0, suff = 0, i, j, delta;
471 /* Lines of both files are hashed, and in the process
472 * their offsets are stored in the array ix[fileno]
473 * where fileno == 0 points to the old file, and
474 * fileno == 1 points to the new one.
476 for (i = 0; i < 2; i++) {
480 nfile[i] = xmalloc((sz + 3) * sizeof(nfile[i][0]));
481 /* ft gets here without the correct position, cant use seek_ft */
483 fseeko(ft[i].ft_fp, 0, SEEK_SET);
486 /* We could zalloc nfile, but then zalloc starts showing in gprof at ~1% */
487 nfile[i][0].offset = 0;
488 goto start; /* saves code */
490 tok = read_token(&ft[i], tok);
491 if (!(tok & TOK_EMPTY)) {
492 /* Hash algorithm taken from Robert Sedgewick, Algorithms in C, 3d ed., p 578. */
493 /*hash = hash * 128 - hash + TOK2CHAR(tok);
494 * gcc insists on optimizing above to "hash * 127 + ...", thus... */
495 unsigned o = hash - TOK2CHAR(tok);
496 hash = hash * 128 - o; /* we want SPEED here */
499 if (nlen[i]++ == sz) {
501 nfile[i] = xrealloc(nfile[i], (sz + 3) * sizeof(nfile[i][0]));
503 /* line_compar needs hashes fit into positive int */
504 nfile[i][nlen[i]].value = hash & INT_MAX;
505 /* like ftello(ft[i].ft_fp) but faster (avoids lseek syscall) */
506 nfile[i][nlen[i]].offset = ft[i].ft_pos;
508 /* EOF counts as a token, so we have to adjust it here */
509 nfile[i][nlen[i]].offset++;
515 /* Exclude lone EOF line from the end of the file, to make fetch()'s job easier */
516 if (nfile[i][nlen[i]].offset - nfile[i][nlen[i] - 1].offset == 1)
518 /* Now we copy the line offsets into ix */
519 ix[i] = xmalloc((nlen[i] + 2) * sizeof(ix[i][0]));
520 for (j = 0; j < nlen[i] + 1; j++)
521 ix[i][j] = nfile[i][j].offset;
524 /* length of prefix and suffix is calculated */
525 for (; pref < nlen[0] && pref < nlen[1] &&
526 nfile[0][pref + 1].value == nfile[1][pref + 1].value;
528 for (; suff < nlen[0] - pref && suff < nlen[1] - pref &&
529 nfile[0][nlen[0] - suff].value == nfile[1][nlen[1] - suff].value;
531 /* Arrays are pruned by the suffix and prefix length,
532 * the result being sorted and stored in sfile[fileno],
533 * and their sizes are stored in slen[fileno]
535 for (j = 0; j < 2; j++) {
536 sfile[j] = nfile[j] + pref;
537 slen[j] = nlen[j] - pref - suff;
538 for (i = 0; i <= slen[j]; i++)
539 sfile[j][i].serial = i;
540 qsort(sfile[j] + 1, slen[j], sizeof(*sfile[j]), line_compar);
542 /* nfile arrays are reused to reduce memory pressure
543 * The #if zeroed out section performs the same task as the
544 * one in the #else section.
545 * Peak memory usage is higher, but one array copy is avoided
546 * by not using unsort()
549 member = xmalloc((slen[1] + 2) * sizeof(member[0]));
550 equiv(sfile[0], slen[0], sfile[1], slen[1], member);
553 class = xmalloc((slen[0] + 1) * sizeof(class[0]));
554 for (i = 1; i <= slen[0]; i++) /* Unsorting */
555 class[sfile[0][i].serial] = sfile[0][i].value;
558 member = (int *)nfile[1];
559 equiv(sfile[0], slen[0], sfile[1], slen[1], member);
560 member = xrealloc(member, (slen[1] + 2) * sizeof(member[0]));
562 class = (int *)nfile[0];
563 unsort(sfile[0], slen[0], (int *)nfile[0]);
564 class = xrealloc(class, (slen[0] + 2) * sizeof(class[0]));
566 J = xmalloc((nlen[0] + 2) * sizeof(J[0]));
567 /* The elements of J which fall inside the prefix and suffix regions
568 * are marked as unchanged, while the ones which fall outside
569 * are initialized with 0 (no matches), so that function stone can
570 * then assign them their right values
572 for (i = 0, delta = nlen[1] - nlen[0]; i <= nlen[0]; i++)
573 J[i] = i <= pref ? i :
574 i > (nlen[0] - suff) ? (i + delta) : 0;
575 /* Here the magic is performed */
576 stone(class, slen[0], member, J, pref);
577 J[nlen[0] + 1] = nlen[1] + 1;
582 /* Both files are rescanned, in an effort to find any lines
583 * which, due to limitations intrinsic to any hashing algorithm,
584 * are different but ended up confounded as the same
586 for (i = 1; i <= nlen[0]; i++) {
590 seek_ft(&ft[0], ix[0][i - 1]);
591 seek_ft(&ft[1], ix[1][J[i] - 1]);
593 for (j = J[i]; i <= nlen[0] && J[i] == j; i++, j++) {
594 token_t tok0 = 0, tok1 = 0;
596 tok0 = read_token(&ft[0], tok0);
597 tok1 = read_token(&ft[1], tok1);
599 if (((tok0 ^ tok1) & TOK_EMPTY) != 0 /* one is empty (not both) */
600 || (!(tok0 & TOK_EMPTY) && TOK2CHAR(tok0) != TOK2CHAR(tok1))
602 J[i] = 0; /* Break the correspondence */
604 } while (!(tok0 & tok1 & TOK_EMPTY));
611 static bool diff(FILE* fp[2], char *file[2])
615 FILE_and_pos_t ft[2];
616 typedef struct { int a, b; } vec_t[2];
618 int i = 1, j, k, idx = -1;
619 bool anychange = false;
624 /* note that ft[i].ft_pos is unintitalized, create_J()
625 * must not assume otherwise */
626 J = create_J(ft, nlen, ix);
629 bool nonempty = false;
634 for (v[0].a = i; v[0].a <= nlen[0] && J[v[0].a] == J[v[0].a - 1] + 1; v[0].a++)
636 v[1].a = J[v[0].a - 1] + 1;
638 for (v[0].b = v[0].a - 1; v[0].b < nlen[0] && !J[v[0].b + 1]; v[0].b++)
640 v[1].b = J[v[0].b + 1] - 1;
642 * Indicate that there is a difference between lines a and b of the 'from' file
643 * to get to lines c to d of the 'to' file. If a is greater than b then there
644 * are no lines in the 'from' file involved and this means that there were
645 * lines appended (beginning at b). If c is greater than d then there are
646 * lines missing from the 'to' file.
648 if (v[0].a <= v[0].b || v[1].a <= v[1].b) {
650 * If this change is more than 'context' lines from the
651 * previous change, dump the record and reset it.
653 int ct = (2 * opt_U_context) + 1;
655 && v[0].a > vec[idx][0].b + ct
656 && v[1].a > vec[idx][1].b + ct
661 for (j = 0; j < 2; j++)
662 for (k = v[j].a; k <= v[j].b; k++)
663 nonempty |= (ix[j][k] - ix[j][k - 1] != 1);
665 vec = xrealloc_vector(vec, 6, ++idx);
666 memcpy(vec[idx], v, sizeof(v));
674 if (idx < 0 || ((option_mask32 & FLAG(B)) && !nonempty))
676 if (!(option_mask32 & FLAG(q))) {
678 vec_t span, *cvp = vec;
681 /* Print the context/unidiff header first time through */
682 printf("--- %s\n", label[0] ? label[0] : file[0]);
683 printf("+++ %s\n", label[1] ? label[1] : file[1]);
687 for (j = 0; j < 2; j++) {
688 int a = span[j].a = MAX(1, (*cvp)[j].a - opt_U_context);
689 int b = span[j].b = MIN(nlen[j], vec[idx][j].b + opt_U_context);
691 printf(" %c%d", j ? '+' : '-', MIN(a, b));
694 printf(",%d", (a < b) ? b - a + 1 : 0);
698 * Output changes in "unified" diff format--the old and new lines
699 * are printed together.
701 for (lowa = span[0].a; ; lowa = (*cvp++)[0].b + 1) {
702 bool end = cvp > &vec[idx];
703 fetch(&ft[0], ix[0], lowa, end ? span[0].b : (*cvp)[0].a - 1, ' ');
706 for (j = 0; j < 2; j++)
707 fetch(&ft[j], ix[j], (*cvp)[j].a, (*cvp)[j].b, j ? '+' : '-');
713 } while (i <= nlen[0]);
722 static int diffreg(char *file[2])
725 bool binary = false, differ = false;
726 int status = STATUS_SAME, i;
730 for (i = 0; i < 2; i++) {
731 int fd = open_or_warn_stdin(file[i]);
734 /* Our diff implementation is using seek.
735 * When we meet non-seekable file, we must make a temp copy.
737 if (lseek(fd, 0, SEEK_SET) == -1 && errno == ESPIPE) {
738 char name[] = "/tmp/difXXXXXX";
739 int fd_tmp = xmkstemp(name);
742 if (bb_copyfd_eof(fd, fd_tmp) < 0)
744 if (fd != STDIN_FILENO)
747 xlseek(fd, 0, SEEK_SET);
749 fp[i] = fdopen(fd, "r");
753 const size_t sz = COMMON_BUFSIZE / 2;
754 char *const buf0 = bb_common_bufsiz1;
755 char *const buf1 = buf0 + sz;
757 i = fread(buf0, 1, sz, fp[0]);
758 j = fread(buf1, 1, sz, fp[1]);
765 for (k = 0; k < i; k++) {
766 if (!buf0[k] || !buf1[k])
768 if (buf0[k] != buf1[k])
773 if (binary && !(option_mask32 & FLAG(a)))
774 status = STATUS_BINARY;
775 else if (diff(fp, file))
776 status = STATUS_DIFFER;
778 if (status != STATUS_SAME)
781 fclose_if_not_stdin(fp[0]);
782 fclose_if_not_stdin(fp[1]);
787 static void print_status(int status, char *path[2])
792 if ((option_mask32 & FLAG(q)) || status == STATUS_BINARY)
793 printf("Files %s and %s differ\n", path[0], path[1]);
796 if (option_mask32 & FLAG(s))
797 printf("Files %s and %s are identical\n", path[0], path[1]);
802 #if ENABLE_FEATURE_DIFF_DIR
809 /* This function adds a filename to dl, the directory listing. */
810 static int FAST_FUNC add_to_dirlist(const char *filename,
811 struct stat *sb UNUSED_PARAM,
812 void *userdata, int depth UNUSED_PARAM)
814 struct dlist *const l = userdata;
815 const char *file = filename + l->len;
818 l->dl = xrealloc_vector(l->dl, 6, l->e);
819 l->dl[l->e] = xstrdup(file);
824 /* If recursion is not set, this function adds the directory
825 * to the list and prevents recursive_action from recursing into it.
827 static int FAST_FUNC skip_dir(const char *filename,
828 struct stat *sb, void *userdata,
831 if (!(option_mask32 & FLAG(r)) && depth) {
832 add_to_dirlist(filename, sb, userdata, depth);
835 if (!(option_mask32 & FLAG(N))) {
836 /* -r without -N: no need to recurse into dirs
837 * which do not exist on the "other side".
838 * Testcase: diff -r /tmp /
839 * (it would recurse deep into /proc without this code) */
840 struct dlist *const l = userdata;
844 char *othername = concat_path_file(G.other_dir, filename);
845 int r = stat(othername, &osb);
847 if (r != 0 || !S_ISDIR(osb.st_mode)) {
848 /* other dir doesn't have similarly named
849 * directory, don't recurse; return 1 upon
850 * exit, just like diffutils' diff */
859 static void diffdir(char *p[2], const char *s_start)
861 struct dlist list[2];
864 memset(&list, 0, sizeof(list));
865 for (i = 0; i < 2; i++) {
866 /*list[i].s = list[i].e = 0; - memset did it */
867 /*list[i].dl = NULL; */
869 G.other_dir = p[1 - i];
870 /* We need to trim root directory prefix.
871 * Using list.len to specify its length,
872 * add_to_dirlist will remove it. */
873 list[i].len = strlen(p[i]);
874 recursive_action(p[i], ACTION_RECURSE | ACTION_FOLLOWLINKS,
875 add_to_dirlist, skip_dir, &list[i], 0);
876 /* Sort dl alphabetically.
877 * GNU diff does this ignoring any number of trailing dots.
878 * We don't, so for us dotted files almost always are
881 qsort_string_vector(list[i].dl, list[i].e);
882 /* If -S was set, find the starting point. */
885 while (list[i].s < list[i].e && strcmp(list[i].dl[list[i].s], s_start) < 0)
888 /* Now that both dirlist1 and dirlist2 contain sorted directory
889 * listings, we can start to go through dirlist1. If both listings
890 * contain the same file, then do a normal diff. Otherwise, behaviour
891 * is determined by whether the -N flag is set. */
897 dp[0] = list[0].s < list[0].e ? list[0].dl[list[0].s] : NULL;
898 dp[1] = list[1].s < list[1].e ? list[1].dl[list[1].s] : NULL;
899 if (!dp[0] && !dp[1])
901 pos = !dp[0] ? 1 : (!dp[1] ? -1 : strcmp(dp[0], dp[1]));
903 if (pos && !(option_mask32 & FLAG(N))) {
904 printf("Only in %s: %s\n", p[k], dp[k]);
907 char *fullpath[2], *path[2]; /* if -N */
909 for (i = 0; i < 2; i++) {
910 if (pos == 0 || i == k) {
911 path[i] = fullpath[i] = concat_path_file(p[i], dp[i]);
912 stat(fullpath[i], &stb[i]);
914 fullpath[i] = concat_path_file(p[i], dp[1 - i]);
915 path[i] = (char *)bb_dev_null;
919 stat(fullpath[k], &stb[1 - k]);
921 if (S_ISDIR(stb[0].st_mode) && S_ISDIR(stb[1].st_mode))
922 printf("Common subdirectories: %s and %s\n", fullpath[0], fullpath[1]);
923 else if (!S_ISREG(stb[0].st_mode) && !S_ISDIR(stb[0].st_mode))
924 printf("File %s is not a regular file or directory and was skipped\n", fullpath[0]);
925 else if (!S_ISREG(stb[1].st_mode) && !S_ISDIR(stb[1].st_mode))
926 printf("File %s is not a regular file or directory and was skipped\n", fullpath[1]);
927 else if (S_ISDIR(stb[0].st_mode) != S_ISDIR(stb[1].st_mode)) {
928 if (S_ISDIR(stb[0].st_mode))
929 printf("File %s is a %s while file %s is a %s\n", fullpath[0], "directory", fullpath[1], "regular file");
931 printf("File %s is a %s while file %s is a %s\n", fullpath[0], "regular file", fullpath[1], "directory");
933 print_status(diffreg(path), fullpath);
945 if (ENABLE_FEATURE_CLEAN_UP) {
952 #if ENABLE_FEATURE_DIFF_LONG_OPTIONS
953 static const char diff_longopts[] ALIGN1 =
954 "ignore-case\0" No_argument "i"
955 "ignore-tab-expansion\0" No_argument "E"
956 "ignore-space-change\0" No_argument "b"
957 "ignore-all-space\0" No_argument "w"
958 "ignore-blank-lines\0" No_argument "B"
959 "text\0" No_argument "a"
960 "unified\0" Required_argument "U"
961 "label\0" Required_argument "L"
962 "show-c-function\0" No_argument "p"
963 "brief\0" No_argument "q"
964 "expand-tabs\0" No_argument "t"
965 "initial-tab\0" No_argument "T"
966 "recursive\0" No_argument "r"
967 "new-file\0" No_argument "N"
968 "report-identical-files\0" No_argument "s"
969 "starting-file\0" Required_argument "S"
970 "minimal\0" No_argument "d"
974 int diff_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
975 int diff_main(int argc UNUSED_PARAM, char **argv)
978 char *file[2], *s_start = NULL;
979 llist_t *L_arg = NULL;
983 /* exactly 2 params; collect multiple -L <label>; -U N */
984 opt_complementary = "=2:L::U+";
985 #if ENABLE_FEATURE_DIFF_LONG_OPTIONS
986 applet_long_options = diff_longopts;
988 getopt32(argv, "abdiL:NqrsS:tTU:wupBE",
989 &L_arg, &s_start, &opt_U_context);
992 label[!!label[0]] = llist_pop(&L_arg);
993 xfunc_error_retval = 2;
994 for (i = 0; i < 2; i++) {
996 /* Compat: "diff file name_which_doesnt_exist" exits with 2 */
997 if (LONE_DASH(file[i])) {
998 fstat(STDIN_FILENO, &stb[i]);
1001 xstat(file[i], &stb[i]);
1003 xfunc_error_retval = 1;
1004 if (gotstdin && (S_ISDIR(stb[0].st_mode) || S_ISDIR(stb[1].st_mode)))
1005 bb_error_msg_and_die("can't compare stdin to a directory");
1007 /* Compare metadata to check if the files are the same physical file.
1009 * Comment from diffutils source says:
1010 * POSIX says that two files are identical if st_ino and st_dev are
1011 * the same, but many file systems incorrectly assign the same (device,
1012 * inode) pair to two distinct files, including:
1013 * GNU/Linux NFS servers that export all local file systems as a
1014 * single NFS file system, if a local device number (st_dev) exceeds
1015 * 255, or if a local inode number (st_ino) exceeds 16777215.
1018 && stb[0].st_ino == stb[1].st_ino
1019 && stb[0].st_dev == stb[1].st_dev
1020 && stb[0].st_size == stb[1].st_size
1021 && stb[0].st_mtime == stb[1].st_mtime
1022 && stb[0].st_ctime == stb[1].st_ctime
1023 && stb[0].st_mode == stb[1].st_mode
1024 && stb[0].st_nlink == stb[1].st_nlink
1025 && stb[0].st_uid == stb[1].st_uid
1026 && stb[0].st_gid == stb[1].st_gid
1028 /* files are physically the same; no need to compare them */
1032 if (S_ISDIR(stb[0].st_mode) && S_ISDIR(stb[1].st_mode)) {
1033 #if ENABLE_FEATURE_DIFF_DIR
1034 diffdir(file, s_start);
1036 bb_error_msg_and_die("no support for directory comparison");
1039 bool dirfile = S_ISDIR(stb[0].st_mode) || S_ISDIR(stb[1].st_mode);
1040 bool dir = S_ISDIR(stb[1].st_mode);
1042 const char *slash = strrchr(file[!dir], '/');
1043 file[dir] = concat_path_file(file[dir], slash ? slash + 1 : file[!dir]);
1044 xstat(file[dir], &stb[dir]);
1046 /* diffreg can get non-regular files here */
1047 print_status(gotstdin > 1 ? STATUS_SAME : diffreg(file), file);