blob: 1961de607ebaac9efb0e73ed5dbddf86939bfedd
1 | /* vi: set sw=4 ts=4: */ |
2 | /* |
3 | * Mini diff implementation for busybox, adapted from OpenBSD diff. |
4 | * |
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> |
8 | * |
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. |
12 | * |
13 | * Licensed under GPLv2 or later, see file LICENSE in this source tree. |
14 | */ |
15 | |
16 | /* |
17 | * The following code uses an algorithm due to Harold Stone, |
18 | * which finds a pair of longest identical subsequences in |
19 | * the two files. |
20 | * |
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 |
24 | * such line in file1. |
25 | * |
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. |
37 | * |
38 | * Next the indices that point into member are unsorted into |
39 | * array class according to the original order of file0. |
40 | * |
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. |
50 | * |
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 |
61 | * |
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. |
68 | * |
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. |
77 | */ |
78 | |
79 | //config:config DIFF |
80 | //config: bool "diff" |
81 | //config: default y |
82 | //config: help |
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. |
86 | //config: |
87 | //config:config FEATURE_DIFF_LONG_OPTIONS |
88 | //config: bool "Enable long options" |
89 | //config: default y |
90 | //config: depends on DIFF && LONG_OPTS |
91 | //config: help |
92 | //config: Enable use of long options. |
93 | //config: |
94 | //config:config FEATURE_DIFF_DIR |
95 | //config: bool "Enable directory support" |
96 | //config: default y |
97 | //config: depends on DIFF |
98 | //config: help |
99 | //config: This option enables support for directory and subdirectory |
100 | //config: comparison. |
101 | |
102 | //kbuild:lib-$(CONFIG_DIFF) += diff.o |
103 | |
104 | //applet:IF_DIFF(APPLET(diff, BB_DIR_USR_BIN, BB_SUID_DROP)) |
105 | |
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" |
126 | |
127 | #include "libbb.h" |
128 | |
129 | #if 0 |
130 | # define dbg_error_msg(...) bb_error_msg(__VA_ARGS__) |
131 | #else |
132 | # define dbg_error_msg(...) ((void)0) |
133 | #endif |
134 | |
135 | enum { /* print_status() and diffreg() return values */ |
136 | STATUS_SAME, /* files are the same */ |
137 | STATUS_DIFFER, /* files differ */ |
138 | STATUS_BINARY, /* binary files differ */ |
139 | }; |
140 | |
141 | enum { /* Commandline flags */ |
142 | FLAG_a, |
143 | FLAG_b, |
144 | FLAG_d, |
145 | FLAG_i, |
146 | FLAG_L, /* never used, handled by getopt32 */ |
147 | FLAG_N, |
148 | FLAG_q, |
149 | FLAG_r, |
150 | FLAG_s, |
151 | FLAG_S, /* never used, handled by getopt32 */ |
152 | FLAG_t, |
153 | FLAG_T, |
154 | FLAG_U, /* never used, handled by getopt32 */ |
155 | FLAG_w, |
156 | FLAG_u, /* ignored, this is the default */ |
157 | FLAG_p, /* not implemented */ |
158 | FLAG_B, |
159 | FLAG_E, /* not implemented */ |
160 | }; |
161 | #define FLAG(x) (1 << FLAG_##x) |
162 | |
163 | /* We cache file position to avoid excessive seeking */ |
164 | typedef struct FILE_and_pos_t { |
165 | FILE *ft_fp; |
166 | off_t ft_pos; |
167 | } FILE_and_pos_t; |
168 | |
169 | struct globals { |
170 | smallint exit_status; |
171 | int opt_U_context; |
172 | const char *other_dir; |
173 | char *label[2]; |
174 | struct stat stb[2]; |
175 | }; |
176 | #define G (*ptr_to_globals) |
177 | #define exit_status (G.exit_status ) |
178 | #define opt_U_context (G.opt_U_context ) |
179 | #define label (G.label ) |
180 | #define stb (G.stb ) |
181 | #define INIT_G() do { \ |
182 | SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \ |
183 | opt_U_context = 3; \ |
184 | } while (0) |
185 | |
186 | typedef int token_t; |
187 | |
188 | enum { |
189 | /* Public */ |
190 | TOK_EMPTY = 1 << 9, /* Line fully processed, you can proceed to the next */ |
191 | TOK_EOF = 1 << 10, /* File ended */ |
192 | /* Private (Only to be used by read_token() */ |
193 | TOK_EOL = 1 << 11, /* we saw EOL (sticky) */ |
194 | TOK_SPACE = 1 << 12, /* used -b code, means we are skipping spaces */ |
195 | SHIFT_EOF = (sizeof(token_t)*8 - 8) - 1, |
196 | CHAR_MASK = 0x1ff, /* 8th bit is used to distinguish EOF from 0xff */ |
197 | }; |
198 | |
199 | /* Restores full EOF from one 8th bit: */ |
200 | //#define TOK2CHAR(t) (((t) << SHIFT_EOF) >> SHIFT_EOF) |
201 | /* We don't really need the above, we only need to have EOF != any_real_char: */ |
202 | #define TOK2CHAR(t) ((t) & CHAR_MASK) |
203 | |
204 | static void seek_ft(FILE_and_pos_t *ft, off_t pos) |
205 | { |
206 | if (ft->ft_pos != pos) { |
207 | ft->ft_pos = pos; |
208 | fseeko(ft->ft_fp, pos, SEEK_SET); |
209 | } |
210 | } |
211 | |
212 | /* Reads tokens from given fp, handling -b and -w flags |
213 | * The user must reset tok every line start |
214 | */ |
215 | static int read_token(FILE_and_pos_t *ft, token_t tok) |
216 | { |
217 | tok |= TOK_EMPTY; |
218 | while (!(tok & TOK_EOL)) { |
219 | bool is_space; |
220 | int t; |
221 | |
222 | t = fgetc(ft->ft_fp); |
223 | if (t != EOF) |
224 | ft->ft_pos++; |
225 | is_space = (t == EOF || isspace(t)); |
226 | |
227 | /* If t == EOF (-1), set both TOK_EOF and TOK_EOL */ |
228 | tok |= (t & (TOK_EOF + TOK_EOL)); |
229 | /* Only EOL? */ |
230 | if (t == '\n') |
231 | tok |= TOK_EOL; |
232 | |
233 | if (option_mask32 & FLAG(i)) /* Handcoded tolower() */ |
234 | t = (t >= 'A' && t <= 'Z') ? t - ('A' - 'a') : t; |
235 | |
236 | if ((option_mask32 & FLAG(w)) && is_space) |
237 | continue; |
238 | |
239 | /* Trim char value to low 9 bits */ |
240 | t &= CHAR_MASK; |
241 | |
242 | if (option_mask32 & FLAG(b)) { |
243 | /* Was prev char whitespace? */ |
244 | if (tok & TOK_SPACE) { /* yes */ |
245 | if (is_space) /* this one too, ignore it */ |
246 | continue; |
247 | tok &= ~TOK_SPACE; |
248 | } else if (is_space) { |
249 | /* 1st whitespace char. |
250 | * Set TOK_SPACE and replace char by ' ' */ |
251 | t = TOK_SPACE + ' '; |
252 | } |
253 | } |
254 | /* Clear EMPTY */ |
255 | tok &= ~(TOK_EMPTY + CHAR_MASK); |
256 | /* Assign char value (low 9 bits) and maybe set TOK_SPACE */ |
257 | tok |= t; |
258 | break; |
259 | } |
260 | #if 0 |
261 | bb_error_msg("fp:%p tok:%x '%c'%s%s%s%s", fp, tok, tok & 0xff |
262 | , tok & TOK_EOF ? " EOF" : "" |
263 | , tok & TOK_EOL ? " EOL" : "" |
264 | , tok & TOK_EMPTY ? " EMPTY" : "" |
265 | , tok & TOK_SPACE ? " SPACE" : "" |
266 | ); |
267 | #endif |
268 | return tok; |
269 | } |
270 | |
271 | struct cand { |
272 | int x; |
273 | int y; |
274 | int pred; |
275 | }; |
276 | |
277 | static int search(const int *c, int k, int y, const struct cand *list) |
278 | { |
279 | int i, j; |
280 | |
281 | if (list[c[k]].y < y) /* quick look for typical case */ |
282 | return k + 1; |
283 | |
284 | for (i = 0, j = k + 1;;) { |
285 | const int l = (i + j) >> 1; |
286 | if (l > i) { |
287 | const int t = list[c[l]].y; |
288 | if (t > y) |
289 | j = l; |
290 | else if (t < y) |
291 | i = l; |
292 | else |
293 | return l; |
294 | } else |
295 | return l + 1; |
296 | } |
297 | } |
298 | |
299 | static unsigned isqrt(unsigned n) |
300 | { |
301 | unsigned x = 1; |
302 | while (1) { |
303 | const unsigned y = x; |
304 | x = ((n / x) + x) >> 1; |
305 | if (x <= (y + 1) && x >= (y - 1)) |
306 | return x; |
307 | } |
308 | } |
309 | |
310 | static void stone(const int *a, int n, const int *b, int *J, int pref) |
311 | { |
312 | const unsigned isq = isqrt(n); |
313 | const unsigned bound = |
314 | (option_mask32 & FLAG(d)) ? UINT_MAX : MAX(256, isq); |
315 | int clen = 1; |
316 | int clistlen = 100; |
317 | int k = 0; |
318 | struct cand *clist = xzalloc(clistlen * sizeof(clist[0])); |
319 | struct cand cand; |
320 | struct cand *q; |
321 | int *klist = xzalloc((n + 2) * sizeof(klist[0])); |
322 | /*clist[0] = (struct cand){0}; - xzalloc did it */ |
323 | /*klist[0] = 0; */ |
324 | |
325 | for (cand.x = 1; cand.x <= n; cand.x++) { |
326 | int j = a[cand.x], oldl = 0; |
327 | unsigned numtries = 0; |
328 | if (j == 0) |
329 | continue; |
330 | cand.y = -b[j]; |
331 | cand.pred = klist[0]; |
332 | do { |
333 | int l, tc; |
334 | if (cand.y <= clist[cand.pred].y) |
335 | continue; |
336 | l = search(klist, k, cand.y, clist); |
337 | if (l != oldl + 1) |
338 | cand.pred = klist[l - 1]; |
339 | if (l <= k && clist[klist[l]].y <= cand.y) |
340 | continue; |
341 | if (clen == clistlen) { |
342 | clistlen = clistlen * 11 / 10; |
343 | clist = xrealloc(clist, clistlen * sizeof(clist[0])); |
344 | } |
345 | clist[clen] = cand; |
346 | tc = klist[l]; |
347 | klist[l] = clen++; |
348 | if (l <= k) { |
349 | cand.pred = tc; |
350 | oldl = l; |
351 | numtries++; |
352 | } else { |
353 | k++; |
354 | break; |
355 | } |
356 | } while ((cand.y = b[++j]) > 0 && numtries < bound); |
357 | } |
358 | /* Unravel */ |
359 | for (q = clist + klist[k]; q->y; q = clist + q->pred) |
360 | J[q->x + pref] = q->y + pref; |
361 | free(klist); |
362 | free(clist); |
363 | } |
364 | |
365 | struct line { |
366 | /* 'serial' is not used in the begining, so we reuse it |
367 | * to store line offsets, thus reducing memory pressure |
368 | */ |
369 | union { |
370 | unsigned serial; |
371 | off_t offset; |
372 | }; |
373 | unsigned value; |
374 | }; |
375 | |
376 | static void equiv(struct line *a, int n, struct line *b, int m, int *c) |
377 | { |
378 | int i = 1, j = 1; |
379 | |
380 | while (i <= n && j <= m) { |
381 | if (a[i].value < b[j].value) |
382 | a[i++].value = 0; |
383 | else if (a[i].value == b[j].value) |
384 | a[i++].value = j; |
385 | else |
386 | j++; |
387 | } |
388 | while (i <= n) |
389 | a[i++].value = 0; |
390 | b[m + 1].value = 0; |
391 | j = 0; |
392 | while (++j <= m) { |
393 | c[j] = -b[j].serial; |
394 | while (b[j + 1].value == b[j].value) { |
395 | j++; |
396 | c[j] = b[j].serial; |
397 | } |
398 | } |
399 | c[j] = -1; |
400 | } |
401 | |
402 | static void unsort(const struct line *f, int l, int *b) |
403 | { |
404 | int i; |
405 | int *a = xmalloc((l + 1) * sizeof(a[0])); |
406 | for (i = 1; i <= l; i++) |
407 | a[f[i].serial] = f[i].value; |
408 | for (i = 1; i <= l; i++) |
409 | b[i] = a[i]; |
410 | free(a); |
411 | } |
412 | |
413 | static int line_compar(const void *a, const void *b) |
414 | { |
415 | #define l0 ((const struct line*)a) |
416 | #define l1 ((const struct line*)b) |
417 | int r = l0->value - l1->value; |
418 | if (r) |
419 | return r; |
420 | return l0->serial - l1->serial; |
421 | #undef l0 |
422 | #undef l1 |
423 | } |
424 | |
425 | static void fetch(FILE_and_pos_t *ft, const off_t *ix, int a, int b, int ch) |
426 | { |
427 | int i, j, col; |
428 | for (i = a; i <= b; i++) { |
429 | seek_ft(ft, ix[i - 1]); |
430 | putchar(ch); |
431 | if (option_mask32 & FLAG(T)) |
432 | putchar('\t'); |
433 | for (j = 0, col = 0; j < ix[i] - ix[i - 1]; j++) { |
434 | int c = fgetc(ft->ft_fp); |
435 | if (c == EOF) { |
436 | printf("\n\\ No newline at end of file\n"); |
437 | return; |
438 | } |
439 | ft->ft_pos++; |
440 | if (c == '\t' && (option_mask32 & FLAG(t))) |
441 | do putchar(' '); while (++col & 7); |
442 | else { |
443 | putchar(c); |
444 | col++; |
445 | } |
446 | } |
447 | } |
448 | } |
449 | |
450 | /* Creates the match vector J, where J[i] is the index |
451 | * of the line in the new file corresponding to the line i |
452 | * in the old file. Lines start at 1 instead of 0, that value |
453 | * being used instead to denote no corresponding line. |
454 | * This vector is dynamically allocated and must be freed by the caller. |
455 | * |
456 | * * fp is an input parameter, where fp[0] and fp[1] are the open |
457 | * old file and new file respectively. |
458 | * * nlen is an output variable, where nlen[0] and nlen[1] |
459 | * gets the number of lines in the old and new file respectively. |
460 | * * ix is an output variable, where ix[0] and ix[1] gets |
461 | * assigned dynamically allocated vectors of the offsets of the lines |
462 | * of the old and new file respectively. These must be freed by the caller. |
463 | */ |
464 | static NOINLINE int *create_J(FILE_and_pos_t ft[2], int nlen[2], off_t *ix[2]) |
465 | { |
466 | int *J, slen[2], *class, *member; |
467 | struct line *nfile[2], *sfile[2]; |
468 | int pref = 0, suff = 0, i, j, delta; |
469 | |
470 | /* Lines of both files are hashed, and in the process |
471 | * their offsets are stored in the array ix[fileno] |
472 | * where fileno == 0 points to the old file, and |
473 | * fileno == 1 points to the new one. |
474 | */ |
475 | for (i = 0; i < 2; i++) { |
476 | unsigned hash; |
477 | token_t tok; |
478 | int sz = 100; |
479 | nfile[i] = xmalloc((sz + 3) * sizeof(nfile[i][0])); |
480 | /* ft gets here without the correct position, cant use seek_ft */ |
481 | ft[i].ft_pos = 0; |
482 | fseeko(ft[i].ft_fp, 0, SEEK_SET); |
483 | |
484 | nlen[i] = 0; |
485 | /* We could zalloc nfile, but then zalloc starts showing in gprof at ~1% */ |
486 | nfile[i][0].offset = 0; |
487 | goto start; /* saves code */ |
488 | while (1) { |
489 | tok = read_token(&ft[i], tok); |
490 | if (!(tok & TOK_EMPTY)) { |
491 | /* Hash algorithm taken from Robert Sedgewick, Algorithms in C, 3d ed., p 578. */ |
492 | /*hash = hash * 128 - hash + TOK2CHAR(tok); |
493 | * gcc insists on optimizing above to "hash * 127 + ...", thus... */ |
494 | unsigned o = hash - TOK2CHAR(tok); |
495 | hash = hash * 128 - o; /* we want SPEED here */ |
496 | continue; |
497 | } |
498 | if (nlen[i]++ == sz) { |
499 | sz = sz * 3 / 2; |
500 | nfile[i] = xrealloc(nfile[i], (sz + 3) * sizeof(nfile[i][0])); |
501 | } |
502 | /* line_compar needs hashes fit into positive int */ |
503 | nfile[i][nlen[i]].value = hash & INT_MAX; |
504 | /* like ftello(ft[i].ft_fp) but faster (avoids lseek syscall) */ |
505 | nfile[i][nlen[i]].offset = ft[i].ft_pos; |
506 | if (tok & TOK_EOF) { |
507 | /* EOF counts as a token, so we have to adjust it here */ |
508 | nfile[i][nlen[i]].offset++; |
509 | break; |
510 | } |
511 | start: |
512 | hash = tok = 0; |
513 | } |
514 | /* Exclude lone EOF line from the end of the file, to make fetch()'s job easier */ |
515 | if (nfile[i][nlen[i]].offset - nfile[i][nlen[i] - 1].offset == 1) |
516 | nlen[i]--; |
517 | /* Now we copy the line offsets into ix */ |
518 | ix[i] = xmalloc((nlen[i] + 2) * sizeof(ix[i][0])); |
519 | for (j = 0; j < nlen[i] + 1; j++) |
520 | ix[i][j] = nfile[i][j].offset; |
521 | } |
522 | |
523 | /* length of prefix and suffix is calculated */ |
524 | for (; pref < nlen[0] && pref < nlen[1] && |
525 | nfile[0][pref + 1].value == nfile[1][pref + 1].value; |
526 | pref++); |
527 | for (; suff < nlen[0] - pref && suff < nlen[1] - pref && |
528 | nfile[0][nlen[0] - suff].value == nfile[1][nlen[1] - suff].value; |
529 | suff++); |
530 | /* Arrays are pruned by the suffix and prefix length, |
531 | * the result being sorted and stored in sfile[fileno], |
532 | * and their sizes are stored in slen[fileno] |
533 | */ |
534 | for (j = 0; j < 2; j++) { |
535 | sfile[j] = nfile[j] + pref; |
536 | slen[j] = nlen[j] - pref - suff; |
537 | for (i = 0; i <= slen[j]; i++) |
538 | sfile[j][i].serial = i; |
539 | qsort(sfile[j] + 1, slen[j], sizeof(*sfile[j]), line_compar); |
540 | } |
541 | /* nfile arrays are reused to reduce memory pressure |
542 | * The #if zeroed out section performs the same task as the |
543 | * one in the #else section. |
544 | * Peak memory usage is higher, but one array copy is avoided |
545 | * by not using unsort() |
546 | */ |
547 | #if 0 |
548 | member = xmalloc((slen[1] + 2) * sizeof(member[0])); |
549 | equiv(sfile[0], slen[0], sfile[1], slen[1], member); |
550 | free(nfile[1]); |
551 | |
552 | class = xmalloc((slen[0] + 1) * sizeof(class[0])); |
553 | for (i = 1; i <= slen[0]; i++) /* Unsorting */ |
554 | class[sfile[0][i].serial] = sfile[0][i].value; |
555 | free(nfile[0]); |
556 | #else |
557 | member = (int *)nfile[1]; |
558 | equiv(sfile[0], slen[0], sfile[1], slen[1], member); |
559 | member = xrealloc(member, (slen[1] + 2) * sizeof(member[0])); |
560 | |
561 | class = (int *)nfile[0]; |
562 | unsort(sfile[0], slen[0], (int *)nfile[0]); |
563 | class = xrealloc(class, (slen[0] + 2) * sizeof(class[0])); |
564 | #endif |
565 | J = xmalloc((nlen[0] + 2) * sizeof(J[0])); |
566 | /* The elements of J which fall inside the prefix and suffix regions |
567 | * are marked as unchanged, while the ones which fall outside |
568 | * are initialized with 0 (no matches), so that function stone can |
569 | * then assign them their right values |
570 | */ |
571 | for (i = 0, delta = nlen[1] - nlen[0]; i <= nlen[0]; i++) |
572 | J[i] = i <= pref ? i : |
573 | i > (nlen[0] - suff) ? (i + delta) : 0; |
574 | /* Here the magic is performed */ |
575 | stone(class, slen[0], member, J, pref); |
576 | J[nlen[0] + 1] = nlen[1] + 1; |
577 | |
578 | free(class); |
579 | free(member); |
580 | |
581 | /* Both files are rescanned, in an effort to find any lines |
582 | * which, due to limitations intrinsic to any hashing algorithm, |
583 | * are different but ended up confounded as the same |
584 | */ |
585 | for (i = 1; i <= nlen[0]; i++) { |
586 | if (!J[i]) |
587 | continue; |
588 | |
589 | seek_ft(&ft[0], ix[0][i - 1]); |
590 | seek_ft(&ft[1], ix[1][J[i] - 1]); |
591 | |
592 | for (j = J[i]; i <= nlen[0] && J[i] == j; i++, j++) { |
593 | token_t tok0 = 0, tok1 = 0; |
594 | do { |
595 | tok0 = read_token(&ft[0], tok0); |
596 | tok1 = read_token(&ft[1], tok1); |
597 | |
598 | if (((tok0 ^ tok1) & TOK_EMPTY) != 0 /* one is empty (not both) */ |
599 | || (!(tok0 & TOK_EMPTY) && TOK2CHAR(tok0) != TOK2CHAR(tok1)) |
600 | ) { |
601 | J[i] = 0; /* Break the correspondence */ |
602 | } |
603 | } while (!(tok0 & tok1 & TOK_EMPTY)); |
604 | } |
605 | } |
606 | |
607 | return J; |
608 | } |
609 | |
610 | static bool diff(FILE* fp[2], char *file[2]) |
611 | { |
612 | int nlen[2]; |
613 | off_t *ix[2]; |
614 | FILE_and_pos_t ft[2]; |
615 | typedef struct { int a, b; } vec_t[2]; |
616 | vec_t *vec = NULL; |
617 | int i = 1, j, k, idx = -1; |
618 | bool anychange = false; |
619 | int *J; |
620 | |
621 | ft[0].ft_fp = fp[0]; |
622 | ft[1].ft_fp = fp[1]; |
623 | /* note that ft[i].ft_pos is unintitalized, create_J() |
624 | * must not assume otherwise */ |
625 | J = create_J(ft, nlen, ix); |
626 | |
627 | do { |
628 | bool nonempty = false; |
629 | |
630 | while (1) { |
631 | vec_t v; |
632 | |
633 | for (v[0].a = i; v[0].a <= nlen[0] && J[v[0].a] == J[v[0].a - 1] + 1; v[0].a++) |
634 | continue; |
635 | v[1].a = J[v[0].a - 1] + 1; |
636 | |
637 | for (v[0].b = v[0].a - 1; v[0].b < nlen[0] && !J[v[0].b + 1]; v[0].b++) |
638 | continue; |
639 | v[1].b = J[v[0].b + 1] - 1; |
640 | /* |
641 | * Indicate that there is a difference between lines a and b of the 'from' file |
642 | * to get to lines c to d of the 'to' file. If a is greater than b then there |
643 | * are no lines in the 'from' file involved and this means that there were |
644 | * lines appended (beginning at b). If c is greater than d then there are |
645 | * lines missing from the 'to' file. |
646 | */ |
647 | if (v[0].a <= v[0].b || v[1].a <= v[1].b) { |
648 | /* |
649 | * If this change is more than 'context' lines from the |
650 | * previous change, dump the record and reset it. |
651 | */ |
652 | int ct = (2 * opt_U_context) + 1; |
653 | if (idx >= 0 |
654 | && v[0].a > vec[idx][0].b + ct |
655 | && v[1].a > vec[idx][1].b + ct |
656 | ) { |
657 | break; |
658 | } |
659 | |
660 | for (j = 0; j < 2; j++) |
661 | for (k = v[j].a; k < v[j].b; k++) |
662 | nonempty |= (ix[j][k+1] - ix[j][k] != 1); |
663 | |
664 | vec = xrealloc_vector(vec, 6, ++idx); |
665 | memcpy(vec[idx], v, sizeof(v)); |
666 | } |
667 | |
668 | i = v[0].b + 1; |
669 | if (i > nlen[0]) |
670 | break; |
671 | J[v[0].b] = v[1].b; |
672 | } |
673 | if (idx < 0 || ((option_mask32 & FLAG(B)) && !nonempty)) |
674 | goto cont; |
675 | if (!(option_mask32 & FLAG(q))) { |
676 | int lowa; |
677 | vec_t span, *cvp = vec; |
678 | |
679 | if (!anychange) { |
680 | /* Print the context/unidiff header first time through */ |
681 | printf("--- %s\n", label[0] ? label[0] : file[0]); |
682 | printf("+++ %s\n", label[1] ? label[1] : file[1]); |
683 | } |
684 | |
685 | printf("@@"); |
686 | for (j = 0; j < 2; j++) { |
687 | int a = span[j].a = MAX(1, (*cvp)[j].a - opt_U_context); |
688 | int b = span[j].b = MIN(nlen[j], vec[idx][j].b + opt_U_context); |
689 | |
690 | printf(" %c%d", j ? '+' : '-', MIN(a, b)); |
691 | if (a == b) |
692 | continue; |
693 | printf(",%d", (a < b) ? b - a + 1 : 0); |
694 | } |
695 | printf(" @@\n"); |
696 | /* |
697 | * Output changes in "unified" diff format--the old and new lines |
698 | * are printed together. |
699 | */ |
700 | for (lowa = span[0].a; ; lowa = (*cvp++)[0].b + 1) { |
701 | bool end = cvp > &vec[idx]; |
702 | fetch(&ft[0], ix[0], lowa, end ? span[0].b : (*cvp)[0].a - 1, ' '); |
703 | if (end) |
704 | break; |
705 | for (j = 0; j < 2; j++) |
706 | fetch(&ft[j], ix[j], (*cvp)[j].a, (*cvp)[j].b, j ? '+' : '-'); |
707 | } |
708 | } |
709 | anychange = true; |
710 | cont: |
711 | idx = -1; |
712 | } while (i <= nlen[0]); |
713 | |
714 | free(vec); |
715 | free(ix[0]); |
716 | free(ix[1]); |
717 | free(J); |
718 | return anychange; |
719 | } |
720 | |
721 | static int diffreg(char *file[2]) |
722 | { |
723 | FILE *fp[2]; |
724 | bool binary = false, differ = false; |
725 | int status = STATUS_SAME, i; |
726 | |
727 | fp[0] = stdin; |
728 | fp[1] = stdin; |
729 | for (i = 0; i < 2; i++) { |
730 | int fd = open_or_warn_stdin(file[i]); |
731 | if (fd == -1) |
732 | goto out; |
733 | /* Our diff implementation is using seek. |
734 | * When we meet non-seekable file, we must make a temp copy. |
735 | */ |
736 | if (lseek(fd, 0, SEEK_SET) == -1 && errno == ESPIPE) { |
737 | /* really should use $TMPDIR, but not usually set on android anyway |
738 | here with ifdef, android will use "/data/local/tmp/difXXXXXX" |
739 | */ |
740 | char name[] = |
741 | #ifdef __BIONIC__ |
742 | "/data/local" |
743 | #endif |
744 | "/tmp/difXXXXXX"; |
745 | int fd_tmp = xmkstemp(name); |
746 | |
747 | unlink(name); |
748 | if (bb_copyfd_eof(fd, fd_tmp) < 0) |
749 | xfunc_die(); |
750 | if (fd) /* Prevents closing of stdin */ |
751 | close(fd); |
752 | fd = fd_tmp; |
753 | } |
754 | fp[i] = fdopen(fd, "r"); |
755 | } |
756 | |
757 | while (1) { |
758 | const size_t sz = COMMON_BUFSIZE / 2; |
759 | char *const buf0 = bb_common_bufsiz1; |
760 | char *const buf1 = buf0 + sz; |
761 | int j, k; |
762 | i = fread(buf0, 1, sz, fp[0]); |
763 | j = fread(buf1, 1, sz, fp[1]); |
764 | if (i != j) { |
765 | differ = true; |
766 | i = MIN(i, j); |
767 | } |
768 | if (i == 0) |
769 | break; |
770 | for (k = 0; k < i; k++) { |
771 | if (!buf0[k] || !buf1[k]) |
772 | binary = true; |
773 | if (buf0[k] != buf1[k]) |
774 | differ = true; |
775 | } |
776 | } |
777 | if (differ) { |
778 | if (binary && !(option_mask32 & FLAG(a))) |
779 | status = STATUS_BINARY; |
780 | else if (diff(fp, file)) |
781 | status = STATUS_DIFFER; |
782 | } |
783 | if (status != STATUS_SAME) |
784 | exit_status |= 1; |
785 | out: |
786 | fclose_if_not_stdin(fp[0]); |
787 | fclose_if_not_stdin(fp[1]); |
788 | |
789 | return status; |
790 | } |
791 | |
792 | static void print_status(int status, char *path[2]) |
793 | { |
794 | switch (status) { |
795 | case STATUS_BINARY: |
796 | case STATUS_DIFFER: |
797 | if ((option_mask32 & FLAG(q)) || status == STATUS_BINARY) |
798 | printf("Files %s and %s differ\n", path[0], path[1]); |
799 | break; |
800 | case STATUS_SAME: |
801 | if (option_mask32 & FLAG(s)) |
802 | printf("Files %s and %s are identical\n", path[0], path[1]); |
803 | break; |
804 | } |
805 | } |
806 | |
807 | #if ENABLE_FEATURE_DIFF_DIR |
808 | struct dlist { |
809 | size_t len; |
810 | int s, e; |
811 | char **dl; |
812 | }; |
813 | |
814 | /* This function adds a filename to dl, the directory listing. */ |
815 | static int FAST_FUNC add_to_dirlist(const char *filename, |
816 | struct stat *sb UNUSED_PARAM, |
817 | void *userdata, int depth UNUSED_PARAM) |
818 | { |
819 | struct dlist *const l = userdata; |
820 | const char *file = filename + l->len; |
821 | while (*file == '/') |
822 | file++; |
823 | l->dl = xrealloc_vector(l->dl, 6, l->e); |
824 | l->dl[l->e] = xstrdup(file); |
825 | l->e++; |
826 | return TRUE; |
827 | } |
828 | |
829 | /* If recursion is not set, this function adds the directory |
830 | * to the list and prevents recursive_action from recursing into it. |
831 | */ |
832 | static int FAST_FUNC skip_dir(const char *filename, |
833 | struct stat *sb, void *userdata, |
834 | int depth) |
835 | { |
836 | if (!(option_mask32 & FLAG(r)) && depth) { |
837 | add_to_dirlist(filename, sb, userdata, depth); |
838 | return SKIP; |
839 | } |
840 | if (!(option_mask32 & FLAG(N))) { |
841 | /* -r without -N: no need to recurse into dirs |
842 | * which do not exist on the "other side". |
843 | * Testcase: diff -r /tmp / |
844 | * (it would recurse deep into /proc without this code) */ |
845 | struct dlist *const l = userdata; |
846 | filename += l->len; |
847 | if (filename[0]) { |
848 | struct stat osb; |
849 | char *othername = concat_path_file(G.other_dir, filename); |
850 | int r = stat(othername, &osb); |
851 | free(othername); |
852 | if (r != 0 || !S_ISDIR(osb.st_mode)) { |
853 | /* other dir doesn't have similarly named |
854 | * directory, don't recurse; return 1 upon |
855 | * exit, just like diffutils' diff */ |
856 | exit_status |= 1; |
857 | return SKIP; |
858 | } |
859 | } |
860 | } |
861 | return TRUE; |
862 | } |
863 | |
864 | static void diffdir(char *p[2], const char *s_start) |
865 | { |
866 | struct dlist list[2]; |
867 | int i; |
868 | |
869 | memset(&list, 0, sizeof(list)); |
870 | for (i = 0; i < 2; i++) { |
871 | /*list[i].s = list[i].e = 0; - memset did it */ |
872 | /*list[i].dl = NULL; */ |
873 | |
874 | G.other_dir = p[1 - i]; |
875 | /* We need to trim root directory prefix. |
876 | * Using list.len to specify its length, |
877 | * add_to_dirlist will remove it. */ |
878 | list[i].len = strlen(p[i]); |
879 | recursive_action(p[i], ACTION_RECURSE | ACTION_FOLLOWLINKS, |
880 | add_to_dirlist, skip_dir, &list[i], 0); |
881 | /* Sort dl alphabetically. |
882 | * GNU diff does this ignoring any number of trailing dots. |
883 | * We don't, so for us dotted files almost always are |
884 | * first on the list. |
885 | */ |
886 | qsort_string_vector(list[i].dl, list[i].e); |
887 | /* If -S was set, find the starting point. */ |
888 | if (!s_start) |
889 | continue; |
890 | while (list[i].s < list[i].e && strcmp(list[i].dl[list[i].s], s_start) < 0) |
891 | list[i].s++; |
892 | } |
893 | /* Now that both dirlist1 and dirlist2 contain sorted directory |
894 | * listings, we can start to go through dirlist1. If both listings |
895 | * contain the same file, then do a normal diff. Otherwise, behaviour |
896 | * is determined by whether the -N flag is set. */ |
897 | while (1) { |
898 | char *dp[2]; |
899 | int pos; |
900 | int k; |
901 | |
902 | dp[0] = list[0].s < list[0].e ? list[0].dl[list[0].s] : NULL; |
903 | dp[1] = list[1].s < list[1].e ? list[1].dl[list[1].s] : NULL; |
904 | if (!dp[0] && !dp[1]) |
905 | break; |
906 | pos = !dp[0] ? 1 : (!dp[1] ? -1 : strcmp(dp[0], dp[1])); |
907 | k = pos > 0; |
908 | if (pos && !(option_mask32 & FLAG(N))) { |
909 | printf("Only in %s: %s\n", p[k], dp[k]); |
910 | exit_status |= 1; |
911 | } else { |
912 | char *fullpath[2], *path[2]; /* if -N */ |
913 | |
914 | for (i = 0; i < 2; i++) { |
915 | if (pos == 0 || i == k) { |
916 | path[i] = fullpath[i] = concat_path_file(p[i], dp[i]); |
917 | stat(fullpath[i], &stb[i]); |
918 | } else { |
919 | fullpath[i] = concat_path_file(p[i], dp[1 - i]); |
920 | path[i] = (char *)bb_dev_null; |
921 | } |
922 | } |
923 | if (pos) |
924 | stat(fullpath[k], &stb[1 - k]); |
925 | |
926 | if (S_ISDIR(stb[0].st_mode) && S_ISDIR(stb[1].st_mode)) |
927 | printf("Common subdirectories: %s and %s\n", fullpath[0], fullpath[1]); |
928 | else if (!S_ISREG(stb[0].st_mode) && !S_ISDIR(stb[0].st_mode)) |
929 | printf("File %s is not a regular file or directory and was skipped\n", fullpath[0]); |
930 | else if (!S_ISREG(stb[1].st_mode) && !S_ISDIR(stb[1].st_mode)) |
931 | printf("File %s is not a regular file or directory and was skipped\n", fullpath[1]); |
932 | else if (S_ISDIR(stb[0].st_mode) != S_ISDIR(stb[1].st_mode)) { |
933 | if (S_ISDIR(stb[0].st_mode)) |
934 | printf("File %s is a %s while file %s is a %s\n", fullpath[0], "directory", fullpath[1], "regular file"); |
935 | else |
936 | printf("File %s is a %s while file %s is a %s\n", fullpath[0], "regular file", fullpath[1], "directory"); |
937 | } else |
938 | print_status(diffreg(path), fullpath); |
939 | |
940 | free(fullpath[0]); |
941 | free(fullpath[1]); |
942 | } |
943 | free(dp[k]); |
944 | list[k].s++; |
945 | if (pos == 0) { |
946 | free(dp[1 - k]); |
947 | list[1 - k].s++; |
948 | } |
949 | } |
950 | if (ENABLE_FEATURE_CLEAN_UP) { |
951 | free(list[0].dl); |
952 | free(list[1].dl); |
953 | } |
954 | } |
955 | #endif |
956 | |
957 | #if ENABLE_FEATURE_DIFF_LONG_OPTIONS |
958 | static const char diff_longopts[] ALIGN1 = |
959 | "ignore-case\0" No_argument "i" |
960 | "ignore-tab-expansion\0" No_argument "E" |
961 | "ignore-space-change\0" No_argument "b" |
962 | "ignore-all-space\0" No_argument "w" |
963 | "ignore-blank-lines\0" No_argument "B" |
964 | "text\0" No_argument "a" |
965 | "unified\0" Required_argument "U" |
966 | "label\0" Required_argument "L" |
967 | "show-c-function\0" No_argument "p" |
968 | "brief\0" No_argument "q" |
969 | "expand-tabs\0" No_argument "t" |
970 | "initial-tab\0" No_argument "T" |
971 | "recursive\0" No_argument "r" |
972 | "new-file\0" No_argument "N" |
973 | "report-identical-files\0" No_argument "s" |
974 | "starting-file\0" Required_argument "S" |
975 | "minimal\0" No_argument "d" |
976 | ; |
977 | #endif |
978 | |
979 | int diff_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE; |
980 | int diff_main(int argc UNUSED_PARAM, char **argv) |
981 | { |
982 | int gotstdin = 0, i; |
983 | char *file[2], *s_start = NULL; |
984 | llist_t *L_arg = NULL; |
985 | |
986 | INIT_G(); |
987 | |
988 | /* exactly 2 params; collect multiple -L <label>; -U N */ |
989 | opt_complementary = "=2:L::U+"; |
990 | #if ENABLE_FEATURE_DIFF_LONG_OPTIONS |
991 | applet_long_options = diff_longopts; |
992 | #endif |
993 | getopt32(argv, "abdiL:NqrsS:tTU:wupBE", |
994 | &L_arg, &s_start, &opt_U_context); |
995 | argv += optind; |
996 | while (L_arg) |
997 | label[!!label[0]] = llist_pop(&L_arg); |
998 | xfunc_error_retval = 2; |
999 | for (i = 0; i < 2; i++) { |
1000 | file[i] = argv[i]; |
1001 | /* Compat: "diff file name_which_doesnt_exist" exits with 2 */ |
1002 | if (LONE_DASH(file[i])) { |
1003 | fstat(STDIN_FILENO, &stb[i]); |
1004 | gotstdin++; |
1005 | } else |
1006 | xstat(file[i], &stb[i]); |
1007 | } |
1008 | xfunc_error_retval = 1; |
1009 | if (gotstdin && (S_ISDIR(stb[0].st_mode) || S_ISDIR(stb[1].st_mode))) |
1010 | bb_error_msg_and_die("can't compare stdin to a directory"); |
1011 | |
1012 | /* Compare metadata to check if the files are the same physical file. |
1013 | * |
1014 | * Comment from diffutils source says: |
1015 | * POSIX says that two files are identical if st_ino and st_dev are |
1016 | * the same, but many file systems incorrectly assign the same (device, |
1017 | * inode) pair to two distinct files, including: |
1018 | * GNU/Linux NFS servers that export all local file systems as a |
1019 | * single NFS file system, if a local device number (st_dev) exceeds |
1020 | * 255, or if a local inode number (st_ino) exceeds 16777215. |
1021 | */ |
1022 | if (ENABLE_DESKTOP |
1023 | && stb[0].st_ino == stb[1].st_ino |
1024 | && stb[0].st_dev == stb[1].st_dev |
1025 | && stb[0].st_size == stb[1].st_size |
1026 | && stb[0].st_mtime == stb[1].st_mtime |
1027 | && stb[0].st_ctime == stb[1].st_ctime |
1028 | && stb[0].st_mode == stb[1].st_mode |
1029 | && stb[0].st_nlink == stb[1].st_nlink |
1030 | && stb[0].st_uid == stb[1].st_uid |
1031 | && stb[0].st_gid == stb[1].st_gid |
1032 | ) { |
1033 | /* files are physically the same; no need to compare them */ |
1034 | return STATUS_SAME; |
1035 | } |
1036 | |
1037 | if (S_ISDIR(stb[0].st_mode) && S_ISDIR(stb[1].st_mode)) { |
1038 | #if ENABLE_FEATURE_DIFF_DIR |
1039 | diffdir(file, s_start); |
1040 | #else |
1041 | bb_error_msg_and_die("no support for directory comparison"); |
1042 | #endif |
1043 | } else { |
1044 | bool dirfile = S_ISDIR(stb[0].st_mode) || S_ISDIR(stb[1].st_mode); |
1045 | bool dir = S_ISDIR(stb[1].st_mode); |
1046 | if (dirfile) { |
1047 | const char *slash = strrchr(file[!dir], '/'); |
1048 | file[dir] = concat_path_file(file[dir], slash ? slash + 1 : file[!dir]); |
1049 | xstat(file[dir], &stb[dir]); |
1050 | } |
1051 | /* diffreg can get non-regular files here */ |
1052 | print_status(gotstdin > 1 ? STATUS_SAME : diffreg(file), file); |
1053 | |
1054 | if (dirfile) |
1055 | free(file[dir]); |
1056 | } |
1057 | |
1058 | return exit_status; |
1059 | } |
1060 |