Contents of /trunk/mkinitrd-magellan/busybox/editors/diff.c
Parent Directory | Revision Log
Revision 1123 -
(show annotations)
(download)
Wed Aug 18 21:56:57 2010 UTC (14 years, 1 month ago) by niro
File MIME type: text/plain
File size: 27827 byte(s)
Wed Aug 18 21:56:57 2010 UTC (14 years, 1 month ago) by niro
File MIME type: text/plain
File size: 27827 byte(s)
-updated to busybox-1.17.1
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 tarball for details. |
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 | #include "libbb.h" |
80 | |
81 | #if 0 |
82 | //#define dbg_error_msg(...) bb_error_msg(__VA_ARGS__) |
83 | #else |
84 | #define dbg_error_msg(...) ((void)0) |
85 | #endif |
86 | |
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 */ |
91 | }; |
92 | |
93 | enum { /* Commandline flags */ |
94 | FLAG_a, |
95 | FLAG_b, |
96 | FLAG_d, |
97 | FLAG_i, |
98 | FLAG_L, /* never used, handled by getopt32 */ |
99 | FLAG_N, |
100 | FLAG_q, |
101 | FLAG_r, |
102 | FLAG_s, |
103 | FLAG_S, /* never used, handled by getopt32 */ |
104 | FLAG_t, |
105 | FLAG_T, |
106 | FLAG_U, /* never used, handled by getopt32 */ |
107 | FLAG_w, |
108 | FLAG_u, /* ignored, this is the default */ |
109 | FLAG_p, /* not implemented */ |
110 | FLAG_B, |
111 | FLAG_E, /* not implemented */ |
112 | }; |
113 | #define FLAG(x) (1 << FLAG_##x) |
114 | |
115 | /* We cache file position to avoid excessive seeking */ |
116 | typedef struct FILE_and_pos_t { |
117 | FILE *ft_fp; |
118 | off_t ft_pos; |
119 | } FILE_and_pos_t; |
120 | |
121 | struct globals { |
122 | smallint exit_status; |
123 | int opt_U_context; |
124 | const char *other_dir; |
125 | char *label[2]; |
126 | struct stat stb[2]; |
127 | }; |
128 | #define G (*ptr_to_globals) |
129 | #define exit_status (G.exit_status ) |
130 | #define opt_U_context (G.opt_U_context ) |
131 | #define label (G.label ) |
132 | #define stb (G.stb ) |
133 | #define INIT_G() do { \ |
134 | SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \ |
135 | opt_U_context = 3; \ |
136 | } while (0) |
137 | |
138 | typedef int token_t; |
139 | |
140 | enum { |
141 | /* Public */ |
142 | TOK_EMPTY = 1 << 9, /* Line fully processed, you can proceed to the next */ |
143 | TOK_EOF = 1 << 10, /* File ended */ |
144 | /* Private (Only to be used by read_token() */ |
145 | TOK_EOL = 1 << 11, /* we saw EOL (sticky) */ |
146 | TOK_SPACE = 1 << 12, /* used -b code, means we are skipping spaces */ |
147 | SHIFT_EOF = (sizeof(token_t)*8 - 8) - 1, |
148 | CHAR_MASK = 0x1ff, /* 8th bit is used to distinguish EOF from 0xff */ |
149 | }; |
150 | |
151 | /* Restores full EOF from one 8th bit: */ |
152 | //#define TOK2CHAR(t) (((t) << SHIFT_EOF) >> SHIFT_EOF) |
153 | /* We don't really need the above, we only need to have EOF != any_real_char: */ |
154 | #define TOK2CHAR(t) ((t) & CHAR_MASK) |
155 | |
156 | static void seek_ft(FILE_and_pos_t *ft, off_t pos) |
157 | { |
158 | if (ft->ft_pos != pos) { |
159 | ft->ft_pos = pos; |
160 | fseeko(ft->ft_fp, pos, SEEK_SET); |
161 | } |
162 | } |
163 | |
164 | /* Reads tokens from given fp, handling -b and -w flags |
165 | * The user must reset tok every line start |
166 | */ |
167 | static int read_token(FILE_and_pos_t *ft, token_t tok) |
168 | { |
169 | tok |= TOK_EMPTY; |
170 | while (!(tok & TOK_EOL)) { |
171 | bool is_space; |
172 | int t; |
173 | |
174 | t = fgetc(ft->ft_fp); |
175 | if (t != EOF) |
176 | ft->ft_pos++; |
177 | is_space = (t == EOF || isspace(t)); |
178 | |
179 | /* If t == EOF (-1), set both TOK_EOF and TOK_EOL */ |
180 | tok |= (t & (TOK_EOF + TOK_EOL)); |
181 | /* Only EOL? */ |
182 | if (t == '\n') |
183 | tok |= TOK_EOL; |
184 | |
185 | if (option_mask32 & FLAG(i)) /* Handcoded tolower() */ |
186 | t = (t >= 'A' && t <= 'Z') ? t - ('A' - 'a') : t; |
187 | |
188 | if ((option_mask32 & FLAG(w)) && is_space) |
189 | continue; |
190 | |
191 | /* Trim char value to low 9 bits */ |
192 | t &= CHAR_MASK; |
193 | |
194 | if (option_mask32 & FLAG(b)) { |
195 | /* Was prev char whitespace? */ |
196 | if (tok & TOK_SPACE) { /* yes */ |
197 | if (is_space) /* this one too, ignore it */ |
198 | continue; |
199 | tok &= ~TOK_SPACE; |
200 | } else if (is_space) { |
201 | /* 1st whitespace char. |
202 | * Set TOK_SPACE and replace char by ' ' */ |
203 | t = TOK_SPACE + ' '; |
204 | } |
205 | } |
206 | /* Clear EMPTY */ |
207 | tok &= ~(TOK_EMPTY + CHAR_MASK); |
208 | /* Assign char value (low 9 bits) and maybe set TOK_SPACE */ |
209 | tok |= t; |
210 | break; |
211 | } |
212 | #if 0 |
213 | bb_error_msg("fp:%p tok:%x '%c'%s%s%s%s", fp, tok, tok & 0xff |
214 | , tok & TOK_EOF ? " EOF" : "" |
215 | , tok & TOK_EOL ? " EOL" : "" |
216 | , tok & TOK_EMPTY ? " EMPTY" : "" |
217 | , tok & TOK_SPACE ? " SPACE" : "" |
218 | ); |
219 | #endif |
220 | return tok; |
221 | } |
222 | |
223 | struct cand { |
224 | int x; |
225 | int y; |
226 | int pred; |
227 | }; |
228 | |
229 | static int search(const int *c, int k, int y, const struct cand *list) |
230 | { |
231 | int i, j; |
232 | |
233 | if (list[c[k]].y < y) /* quick look for typical case */ |
234 | return k + 1; |
235 | |
236 | for (i = 0, j = k + 1;;) { |
237 | const int l = (i + j) >> 1; |
238 | if (l > i) { |
239 | const int t = list[c[l]].y; |
240 | if (t > y) |
241 | j = l; |
242 | else if (t < y) |
243 | i = l; |
244 | else |
245 | return l; |
246 | } else |
247 | return l + 1; |
248 | } |
249 | } |
250 | |
251 | static unsigned isqrt(unsigned n) |
252 | { |
253 | unsigned x = 1; |
254 | while (1) { |
255 | const unsigned y = x; |
256 | x = ((n / x) + x) >> 1; |
257 | if (x <= (y + 1) && x >= (y - 1)) |
258 | return x; |
259 | } |
260 | } |
261 | |
262 | static void stone(const int *a, int n, const int *b, int *J, int pref) |
263 | { |
264 | const unsigned isq = isqrt(n); |
265 | const unsigned bound = |
266 | (option_mask32 & FLAG(d)) ? UINT_MAX : MAX(256, isq); |
267 | int clen = 1; |
268 | int clistlen = 100; |
269 | int k = 0; |
270 | struct cand *clist = xzalloc(clistlen * sizeof(clist[0])); |
271 | struct cand cand; |
272 | struct cand *q; |
273 | int *klist = xzalloc((n + 2) * sizeof(klist[0])); |
274 | /*clist[0] = (struct cand){0}; - xzalloc did it */ |
275 | /*klist[0] = 0; */ |
276 | |
277 | for (cand.x = 1; cand.x <= n; cand.x++) { |
278 | int j = a[cand.x], oldl = 0; |
279 | unsigned numtries = 0; |
280 | if (j == 0) |
281 | continue; |
282 | cand.y = -b[j]; |
283 | cand.pred = klist[0]; |
284 | do { |
285 | int l, tc; |
286 | if (cand.y <= clist[cand.pred].y) |
287 | continue; |
288 | l = search(klist, k, cand.y, clist); |
289 | if (l != oldl + 1) |
290 | cand.pred = klist[l - 1]; |
291 | if (l <= k && clist[klist[l]].y <= cand.y) |
292 | continue; |
293 | if (clen == clistlen) { |
294 | clistlen = clistlen * 11 / 10; |
295 | clist = xrealloc(clist, clistlen * sizeof(clist[0])); |
296 | } |
297 | clist[clen] = cand; |
298 | tc = klist[l]; |
299 | klist[l] = clen++; |
300 | if (l <= k) { |
301 | cand.pred = tc; |
302 | oldl = l; |
303 | numtries++; |
304 | } else { |
305 | k++; |
306 | break; |
307 | } |
308 | } while ((cand.y = b[++j]) > 0 && numtries < bound); |
309 | } |
310 | /* Unravel */ |
311 | for (q = clist + klist[k]; q->y; q = clist + q->pred) |
312 | J[q->x + pref] = q->y + pref; |
313 | free(klist); |
314 | free(clist); |
315 | } |
316 | |
317 | struct line { |
318 | /* 'serial' is not used in the begining, so we reuse it |
319 | * to store line offsets, thus reducing memory pressure |
320 | */ |
321 | union { |
322 | unsigned serial; |
323 | off_t offset; |
324 | }; |
325 | unsigned value; |
326 | }; |
327 | |
328 | static void equiv(struct line *a, int n, struct line *b, int m, int *c) |
329 | { |
330 | int i = 1, j = 1; |
331 | |
332 | while (i <= n && j <= m) { |
333 | if (a[i].value < b[j].value) |
334 | a[i++].value = 0; |
335 | else if (a[i].value == b[j].value) |
336 | a[i++].value = j; |
337 | else |
338 | j++; |
339 | } |
340 | while (i <= n) |
341 | a[i++].value = 0; |
342 | b[m + 1].value = 0; |
343 | j = 0; |
344 | while (++j <= m) { |
345 | c[j] = -b[j].serial; |
346 | while (b[j + 1].value == b[j].value) { |
347 | j++; |
348 | c[j] = b[j].serial; |
349 | } |
350 | } |
351 | c[j] = -1; |
352 | } |
353 | |
354 | static void unsort(const struct line *f, int l, int *b) |
355 | { |
356 | int i; |
357 | int *a = xmalloc((l + 1) * sizeof(a[0])); |
358 | for (i = 1; i <= l; i++) |
359 | a[f[i].serial] = f[i].value; |
360 | for (i = 1; i <= l; i++) |
361 | b[i] = a[i]; |
362 | free(a); |
363 | } |
364 | |
365 | static int line_compar(const void *a, const void *b) |
366 | { |
367 | #define l0 ((const struct line*)a) |
368 | #define l1 ((const struct line*)b) |
369 | int r = l0->value - l1->value; |
370 | if (r) |
371 | return r; |
372 | return l0->serial - l1->serial; |
373 | #undef l0 |
374 | #undef l1 |
375 | } |
376 | |
377 | static void fetch(FILE_and_pos_t *ft, const off_t *ix, int a, int b, int ch) |
378 | { |
379 | int i, j, col; |
380 | for (i = a; i <= b; i++) { |
381 | seek_ft(ft, ix[i - 1]); |
382 | putchar(ch); |
383 | if (option_mask32 & FLAG(T)) |
384 | putchar('\t'); |
385 | for (j = 0, col = 0; j < ix[i] - ix[i - 1]; j++) { |
386 | int c = fgetc(ft->ft_fp); |
387 | if (c == EOF) { |
388 | printf("\n\\ No newline at end of file\n"); |
389 | return; |
390 | } |
391 | ft->ft_pos++; |
392 | if (c == '\t' && (option_mask32 & FLAG(t))) |
393 | do putchar(' '); while (++col & 7); |
394 | else { |
395 | putchar(c); |
396 | col++; |
397 | } |
398 | } |
399 | } |
400 | } |
401 | |
402 | /* Creates the match vector J, where J[i] is the index |
403 | * of the line in the new file corresponding to the line i |
404 | * in the old file. Lines start at 1 instead of 0, that value |
405 | * being used instead to denote no corresponding line. |
406 | * This vector is dynamically allocated and must be freed by the caller. |
407 | * |
408 | * * fp is an input parameter, where fp[0] and fp[1] are the open |
409 | * old file and new file respectively. |
410 | * * nlen is an output variable, where nlen[0] and nlen[1] |
411 | * gets the number of lines in the old and new file respectively. |
412 | * * ix is an output variable, where ix[0] and ix[1] gets |
413 | * assigned dynamically allocated vectors of the offsets of the lines |
414 | * of the old and new file respectively. These must be freed by the caller. |
415 | */ |
416 | static NOINLINE int *create_J(FILE_and_pos_t ft[2], int nlen[2], off_t *ix[2]) |
417 | { |
418 | int *J, slen[2], *class, *member; |
419 | struct line *nfile[2], *sfile[2]; |
420 | int pref = 0, suff = 0, i, j, delta; |
421 | |
422 | /* Lines of both files are hashed, and in the process |
423 | * their offsets are stored in the array ix[fileno] |
424 | * where fileno == 0 points to the old file, and |
425 | * fileno == 1 points to the new one. |
426 | */ |
427 | for (i = 0; i < 2; i++) { |
428 | unsigned hash; |
429 | token_t tok; |
430 | size_t sz = 100; |
431 | nfile[i] = xmalloc((sz + 3) * sizeof(nfile[i][0])); |
432 | /* ft gets here without the correct position, cant use seek_ft */ |
433 | ft[i].ft_pos = 0; |
434 | fseeko(ft[i].ft_fp, 0, SEEK_SET); |
435 | |
436 | nlen[i] = 0; |
437 | /* We could zalloc nfile, but then zalloc starts showing in gprof at ~1% */ |
438 | nfile[i][0].offset = 0; |
439 | goto start; /* saves code */ |
440 | while (1) { |
441 | tok = read_token(&ft[i], tok); |
442 | if (!(tok & TOK_EMPTY)) { |
443 | /* Hash algorithm taken from Robert Sedgewick, Algorithms in C, 3d ed., p 578. */ |
444 | /*hash = hash * 128 - hash + TOK2CHAR(tok); |
445 | * gcc insists on optimizing above to "hash * 127 + ...", thus... */ |
446 | unsigned o = hash - TOK2CHAR(tok); |
447 | hash = hash * 128 - o; /* we want SPEED here */ |
448 | continue; |
449 | } |
450 | if (nlen[i]++ == sz) { |
451 | sz = sz * 3 / 2; |
452 | nfile[i] = xrealloc(nfile[i], (sz + 3) * sizeof(nfile[i][0])); |
453 | } |
454 | /* line_compar needs hashes fit into positive int */ |
455 | nfile[i][nlen[i]].value = hash & INT_MAX; |
456 | /* like ftello(ft[i].ft_fp) but faster (avoids lseek syscall) */ |
457 | nfile[i][nlen[i]].offset = ft[i].ft_pos; |
458 | if (tok & TOK_EOF) { |
459 | /* EOF counts as a token, so we have to adjust it here */ |
460 | nfile[i][nlen[i]].offset++; |
461 | break; |
462 | } |
463 | start: |
464 | hash = tok = 0; |
465 | } |
466 | /* Exclude lone EOF line from the end of the file, to make fetch()'s job easier */ |
467 | if (nfile[i][nlen[i]].offset - nfile[i][nlen[i] - 1].offset == 1) |
468 | nlen[i]--; |
469 | /* Now we copy the line offsets into ix */ |
470 | ix[i] = xmalloc((nlen[i] + 2) * sizeof(ix[i][0])); |
471 | for (j = 0; j < nlen[i] + 1; j++) |
472 | ix[i][j] = nfile[i][j].offset; |
473 | } |
474 | |
475 | /* length of prefix and suffix is calculated */ |
476 | for (; pref < nlen[0] && pref < nlen[1] && |
477 | nfile[0][pref + 1].value == nfile[1][pref + 1].value; |
478 | pref++); |
479 | for (; suff < nlen[0] - pref && suff < nlen[1] - pref && |
480 | nfile[0][nlen[0] - suff].value == nfile[1][nlen[1] - suff].value; |
481 | suff++); |
482 | /* Arrays are pruned by the suffix and prefix lenght, |
483 | * the result being sorted and stored in sfile[fileno], |
484 | * and their sizes are stored in slen[fileno] |
485 | */ |
486 | for (j = 0; j < 2; j++) { |
487 | sfile[j] = nfile[j] + pref; |
488 | slen[j] = nlen[j] - pref - suff; |
489 | for (i = 0; i <= slen[j]; i++) |
490 | sfile[j][i].serial = i; |
491 | qsort(sfile[j] + 1, slen[j], sizeof(*sfile[j]), line_compar); |
492 | } |
493 | /* nfile arrays are reused to reduce memory pressure |
494 | * The #if zeroed out section performs the same task as the |
495 | * one in the #else section. |
496 | * Peak memory usage is higher, but one array copy is avoided |
497 | * by not using unsort() |
498 | */ |
499 | #if 0 |
500 | member = xmalloc((slen[1] + 2) * sizeof(member[0])); |
501 | equiv(sfile[0], slen[0], sfile[1], slen[1], member); |
502 | free(nfile[1]); |
503 | |
504 | class = xmalloc((slen[0] + 1) * sizeof(class[0])); |
505 | for (i = 1; i <= slen[0]; i++) /* Unsorting */ |
506 | class[sfile[0][i].serial] = sfile[0][i].value; |
507 | free(nfile[0]); |
508 | #else |
509 | member = (int *)nfile[1]; |
510 | equiv(sfile[0], slen[0], sfile[1], slen[1], member); |
511 | member = xrealloc(member, (slen[1] + 2) * sizeof(member[0])); |
512 | |
513 | class = (int *)nfile[0]; |
514 | unsort(sfile[0], slen[0], (int *)nfile[0]); |
515 | class = xrealloc(class, (slen[0] + 2) * sizeof(class[0])); |
516 | #endif |
517 | J = xmalloc((nlen[0] + 2) * sizeof(J[0])); |
518 | /* The elements of J which fall inside the prefix and suffix regions |
519 | * are marked as unchanged, while the ones which fall outside |
520 | * are initialized with 0 (no matches), so that function stone can |
521 | * then assign them their right values |
522 | */ |
523 | for (i = 0, delta = nlen[1] - nlen[0]; i <= nlen[0]; i++) |
524 | J[i] = i <= pref ? i : |
525 | i > (nlen[0] - suff) ? (i + delta) : 0; |
526 | /* Here the magic is performed */ |
527 | stone(class, slen[0], member, J, pref); |
528 | J[nlen[0] + 1] = nlen[1] + 1; |
529 | |
530 | free(class); |
531 | free(member); |
532 | |
533 | /* Both files are rescanned, in an effort to find any lines |
534 | * which, due to limitations intrinsic to any hashing algorithm, |
535 | * are different but ended up confounded as the same |
536 | */ |
537 | for (i = 1; i <= nlen[0]; i++) { |
538 | if (!J[i]) |
539 | continue; |
540 | |
541 | seek_ft(&ft[0], ix[0][i - 1]); |
542 | seek_ft(&ft[1], ix[1][J[i] - 1]); |
543 | |
544 | for (j = J[i]; i <= nlen[0] && J[i] == j; i++, j++) { |
545 | token_t tok0 = 0, tok1 = 0; |
546 | do { |
547 | tok0 = read_token(&ft[0], tok0); |
548 | tok1 = read_token(&ft[1], tok1); |
549 | |
550 | if (((tok0 ^ tok1) & TOK_EMPTY) != 0 /* one is empty (not both) */ |
551 | || (!(tok0 & TOK_EMPTY) && TOK2CHAR(tok0) != TOK2CHAR(tok1)) |
552 | ) { |
553 | J[i] = 0; /* Break the correspondence */ |
554 | } |
555 | } while (!(tok0 & tok1 & TOK_EMPTY)); |
556 | } |
557 | } |
558 | |
559 | return J; |
560 | } |
561 | |
562 | static bool diff(FILE* fp[2], char *file[2]) |
563 | { |
564 | int nlen[2]; |
565 | off_t *ix[2]; |
566 | FILE_and_pos_t ft[2]; |
567 | typedef struct { int a, b; } vec_t[2]; |
568 | vec_t *vec = NULL; |
569 | int i = 1, j, k, idx = -1; |
570 | bool anychange = false; |
571 | int *J; |
572 | |
573 | ft[0].ft_fp = fp[0]; |
574 | ft[1].ft_fp = fp[1]; |
575 | /* note that ft[i].ft_pos is unintitalized, create_J() |
576 | * must not assume otherwise */ |
577 | J = create_J(ft, nlen, ix); |
578 | |
579 | do { |
580 | bool nonempty = false; |
581 | |
582 | while (1) { |
583 | vec_t v; |
584 | |
585 | for (v[0].a = i; v[0].a <= nlen[0] && J[v[0].a] == J[v[0].a - 1] + 1; v[0].a++) |
586 | continue; |
587 | v[1].a = J[v[0].a - 1] + 1; |
588 | |
589 | for (v[0].b = v[0].a - 1; v[0].b < nlen[0] && !J[v[0].b + 1]; v[0].b++) |
590 | continue; |
591 | v[1].b = J[v[0].b + 1] - 1; |
592 | /* |
593 | * Indicate that there is a difference between lines a and b of the 'from' file |
594 | * to get to lines c to d of the 'to' file. If a is greater than b then there |
595 | * are no lines in the 'from' file involved and this means that there were |
596 | * lines appended (beginning at b). If c is greater than d then there are |
597 | * lines missing from the 'to' file. |
598 | */ |
599 | if (v[0].a <= v[0].b || v[1].a <= v[1].b) { |
600 | /* |
601 | * If this change is more than 'context' lines from the |
602 | * previous change, dump the record and reset it. |
603 | */ |
604 | int ct = (2 * opt_U_context) + 1; |
605 | if (idx >= 0 |
606 | && v[0].a > vec[idx][0].b + ct |
607 | && v[1].a > vec[idx][1].b + ct |
608 | ) { |
609 | break; |
610 | } |
611 | |
612 | for (j = 0; j < 2; j++) |
613 | for (k = v[j].a; k < v[j].b; k++) |
614 | nonempty |= (ix[j][k+1] - ix[j][k] != 1); |
615 | |
616 | vec = xrealloc_vector(vec, 6, ++idx); |
617 | memcpy(vec[idx], v, sizeof(v)); |
618 | } |
619 | |
620 | i = v[0].b + 1; |
621 | if (i > nlen[0]) |
622 | break; |
623 | J[v[0].b] = v[1].b; |
624 | } |
625 | if (idx < 0 || ((option_mask32 & FLAG(B)) && !nonempty)) |
626 | goto cont; |
627 | if (!(option_mask32 & FLAG(q))) { |
628 | int lowa; |
629 | vec_t span, *cvp = vec; |
630 | |
631 | if (!anychange) { |
632 | /* Print the context/unidiff header first time through */ |
633 | printf("--- %s\n", label[0] ? label[0] : file[0]); |
634 | printf("+++ %s\n", label[1] ? label[1] : file[1]); |
635 | } |
636 | |
637 | printf("@@"); |
638 | for (j = 0; j < 2; j++) { |
639 | int a = span[j].a = MAX(1, (*cvp)[j].a - opt_U_context); |
640 | int b = span[j].b = MIN(nlen[j], vec[idx][j].b + opt_U_context); |
641 | |
642 | printf(" %c%d", j ? '+' : '-', MIN(a, b)); |
643 | if (a == b) |
644 | continue; |
645 | printf(",%d", (a < b) ? b - a + 1 : 0); |
646 | } |
647 | printf(" @@\n"); |
648 | /* |
649 | * Output changes in "unified" diff format--the old and new lines |
650 | * are printed together. |
651 | */ |
652 | for (lowa = span[0].a; ; lowa = (*cvp++)[0].b + 1) { |
653 | bool end = cvp > &vec[idx]; |
654 | fetch(&ft[0], ix[0], lowa, end ? span[0].b : (*cvp)[0].a - 1, ' '); |
655 | if (end) |
656 | break; |
657 | for (j = 0; j < 2; j++) |
658 | fetch(&ft[j], ix[j], (*cvp)[j].a, (*cvp)[j].b, j ? '+' : '-'); |
659 | } |
660 | } |
661 | anychange = true; |
662 | cont: |
663 | idx = -1; |
664 | } while (i <= nlen[0]); |
665 | |
666 | free(vec); |
667 | free(ix[0]); |
668 | free(ix[1]); |
669 | free(J); |
670 | return anychange; |
671 | } |
672 | |
673 | static int diffreg(char *file[2]) |
674 | { |
675 | FILE *fp[2] = { stdin, stdin }; |
676 | bool binary = false, differ = false; |
677 | int status = STATUS_SAME, i; |
678 | |
679 | for (i = 0; i < 2; i++) { |
680 | int fd = open_or_warn_stdin(file[i]); |
681 | if (fd == -1) |
682 | goto out; |
683 | /* Our diff implementation is using seek. |
684 | * When we meet non-seekable file, we must make a temp copy. |
685 | */ |
686 | if (lseek(fd, 0, SEEK_SET) == -1 && errno == ESPIPE) { |
687 | char name[] = "/tmp/difXXXXXX"; |
688 | int fd_tmp = mkstemp(name); |
689 | if (fd_tmp < 0) |
690 | bb_perror_msg_and_die("mkstemp"); |
691 | unlink(name); |
692 | if (bb_copyfd_eof(fd, fd_tmp) < 0) |
693 | xfunc_die(); |
694 | if (fd) /* Prevents closing of stdin */ |
695 | close(fd); |
696 | fd = fd_tmp; |
697 | } |
698 | fp[i] = fdopen(fd, "r"); |
699 | } |
700 | |
701 | while (1) { |
702 | const size_t sz = COMMON_BUFSIZE / 2; |
703 | char *const buf0 = bb_common_bufsiz1; |
704 | char *const buf1 = buf0 + sz; |
705 | int j, k; |
706 | i = fread(buf0, 1, sz, fp[0]); |
707 | j = fread(buf1, 1, sz, fp[1]); |
708 | if (i != j) { |
709 | differ = true; |
710 | i = MIN(i, j); |
711 | } |
712 | if (i == 0) |
713 | break; |
714 | for (k = 0; k < i; k++) { |
715 | if (!buf0[k] || !buf1[k]) |
716 | binary = true; |
717 | if (buf0[k] != buf1[k]) |
718 | differ = true; |
719 | } |
720 | } |
721 | if (differ) { |
722 | if (binary && !(option_mask32 & FLAG(a))) |
723 | status = STATUS_BINARY; |
724 | else if (diff(fp, file)) |
725 | status = STATUS_DIFFER; |
726 | } |
727 | if (status != STATUS_SAME) |
728 | exit_status |= 1; |
729 | out: |
730 | fclose_if_not_stdin(fp[0]); |
731 | fclose_if_not_stdin(fp[1]); |
732 | |
733 | return status; |
734 | } |
735 | |
736 | static void print_status(int status, char *path[2]) |
737 | { |
738 | switch (status) { |
739 | case STATUS_BINARY: |
740 | case STATUS_DIFFER: |
741 | if ((option_mask32 & FLAG(q)) || status == STATUS_BINARY) |
742 | printf("Files %s and %s differ\n", path[0], path[1]); |
743 | break; |
744 | case STATUS_SAME: |
745 | if (option_mask32 & FLAG(s)) |
746 | printf("Files %s and %s are identical\n", path[0], path[1]); |
747 | break; |
748 | } |
749 | } |
750 | |
751 | #if ENABLE_FEATURE_DIFF_DIR |
752 | struct dlist { |
753 | size_t len; |
754 | int s, e; |
755 | char **dl; |
756 | }; |
757 | |
758 | /* This function adds a filename to dl, the directory listing. */ |
759 | static int FAST_FUNC add_to_dirlist(const char *filename, |
760 | struct stat *sb UNUSED_PARAM, |
761 | void *userdata, int depth UNUSED_PARAM) |
762 | { |
763 | struct dlist *const l = userdata; |
764 | const char *file = filename + l->len; |
765 | while (*file == '/') |
766 | file++; |
767 | l->dl = xrealloc_vector(l->dl, 6, l->e); |
768 | l->dl[l->e] = xstrdup(file); |
769 | l->e++; |
770 | return TRUE; |
771 | } |
772 | |
773 | /* If recursion is not set, this function adds the directory |
774 | * to the list and prevents recursive_action from recursing into it. |
775 | */ |
776 | static int FAST_FUNC skip_dir(const char *filename, |
777 | struct stat *sb, void *userdata, |
778 | int depth) |
779 | { |
780 | if (!(option_mask32 & FLAG(r)) && depth) { |
781 | add_to_dirlist(filename, sb, userdata, depth); |
782 | return SKIP; |
783 | } |
784 | if (!(option_mask32 & FLAG(N))) { |
785 | /* -r without -N: no need to recurse into dirs |
786 | * which do not exist on the "other side". |
787 | * Testcase: diff -r /tmp / |
788 | * (it would recurse deep into /proc without this code) */ |
789 | struct dlist *const l = userdata; |
790 | filename += l->len; |
791 | if (filename[0]) { |
792 | struct stat osb; |
793 | char *othername = concat_path_file(G.other_dir, filename); |
794 | int r = stat(othername, &osb); |
795 | free(othername); |
796 | if (r != 0 || !S_ISDIR(osb.st_mode)) { |
797 | /* other dir doesn't have similarly named |
798 | * directory, don't recurse */ |
799 | return SKIP; |
800 | } |
801 | } |
802 | } |
803 | return TRUE; |
804 | } |
805 | |
806 | static void diffdir(char *p[2], const char *s_start) |
807 | { |
808 | struct dlist list[2]; |
809 | int i; |
810 | |
811 | memset(&list, 0, sizeof(list)); |
812 | for (i = 0; i < 2; i++) { |
813 | /*list[i].s = list[i].e = 0; - memset did it */ |
814 | /*list[i].dl = NULL; */ |
815 | |
816 | G.other_dir = p[1 - i]; |
817 | /* We need to trim root directory prefix. |
818 | * Using list.len to specify its length, |
819 | * add_to_dirlist will remove it. */ |
820 | list[i].len = strlen(p[i]); |
821 | recursive_action(p[i], ACTION_RECURSE | ACTION_FOLLOWLINKS, |
822 | add_to_dirlist, skip_dir, &list[i], 0); |
823 | /* Sort dl alphabetically. |
824 | * GNU diff does this ignoring any number of trailing dots. |
825 | * We don't, so for us dotted files almost always are |
826 | * first on the list. |
827 | */ |
828 | qsort_string_vector(list[i].dl, list[i].e); |
829 | /* If -S was set, find the starting point. */ |
830 | if (!s_start) |
831 | continue; |
832 | while (list[i].s < list[i].e && strcmp(list[i].dl[list[i].s], s_start) < 0) |
833 | list[i].s++; |
834 | } |
835 | /* Now that both dirlist1 and dirlist2 contain sorted directory |
836 | * listings, we can start to go through dirlist1. If both listings |
837 | * contain the same file, then do a normal diff. Otherwise, behaviour |
838 | * is determined by whether the -N flag is set. */ |
839 | while (1) { |
840 | char *dp[2]; |
841 | int pos; |
842 | int k; |
843 | |
844 | dp[0] = list[0].s < list[0].e ? list[0].dl[list[0].s] : NULL; |
845 | dp[1] = list[1].s < list[1].e ? list[1].dl[list[1].s] : NULL; |
846 | if (!dp[0] && !dp[1]) |
847 | break; |
848 | pos = !dp[0] ? 1 : (!dp[1] ? -1 : strcmp(dp[0], dp[1])); |
849 | k = pos > 0; |
850 | if (pos && !(option_mask32 & FLAG(N))) |
851 | printf("Only in %s: %s\n", p[k], dp[k]); |
852 | else { |
853 | char *fullpath[2], *path[2]; /* if -N */ |
854 | |
855 | for (i = 0; i < 2; i++) { |
856 | if (pos == 0 || i == k) { |
857 | path[i] = fullpath[i] = concat_path_file(p[i], dp[i]); |
858 | stat(fullpath[i], &stb[i]); |
859 | } else { |
860 | fullpath[i] = concat_path_file(p[i], dp[1 - i]); |
861 | path[i] = (char *)bb_dev_null; |
862 | } |
863 | } |
864 | if (pos) |
865 | stat(fullpath[k], &stb[1 - k]); |
866 | |
867 | if (S_ISDIR(stb[0].st_mode) && S_ISDIR(stb[1].st_mode)) |
868 | printf("Common subdirectories: %s and %s\n", fullpath[0], fullpath[1]); |
869 | else if (!S_ISREG(stb[0].st_mode) && !S_ISDIR(stb[0].st_mode)) |
870 | printf("File %s is not a regular file or directory and was skipped\n", fullpath[0]); |
871 | else if (!S_ISREG(stb[1].st_mode) && !S_ISDIR(stb[1].st_mode)) |
872 | printf("File %s is not a regular file or directory and was skipped\n", fullpath[1]); |
873 | else if (S_ISDIR(stb[0].st_mode) != S_ISDIR(stb[1].st_mode)) { |
874 | if (S_ISDIR(stb[0].st_mode)) |
875 | printf("File %s is a %s while file %s is a %s\n", fullpath[0], "directory", fullpath[1], "regular file"); |
876 | else |
877 | printf("File %s is a %s while file %s is a %s\n", fullpath[0], "regular file", fullpath[1], "directory"); |
878 | } else |
879 | print_status(diffreg(path), fullpath); |
880 | |
881 | free(fullpath[0]); |
882 | free(fullpath[1]); |
883 | } |
884 | free(dp[k]); |
885 | list[k].s++; |
886 | if (pos == 0) { |
887 | free(dp[1 - k]); |
888 | list[1 - k].s++; |
889 | } |
890 | } |
891 | if (ENABLE_FEATURE_CLEAN_UP) { |
892 | free(list[0].dl); |
893 | free(list[1].dl); |
894 | } |
895 | } |
896 | #endif |
897 | |
898 | #if ENABLE_FEATURE_DIFF_LONG_OPTIONS |
899 | static const char diff_longopts[] ALIGN1 = |
900 | "ignore-case\0" No_argument "i" |
901 | "ignore-tab-expansion\0" No_argument "E" |
902 | "ignore-space-change\0" No_argument "b" |
903 | "ignore-all-space\0" No_argument "w" |
904 | "ignore-blank-lines\0" No_argument "B" |
905 | "text\0" No_argument "a" |
906 | "unified\0" Required_argument "U" |
907 | "label\0" Required_argument "L" |
908 | "show-c-function\0" No_argument "p" |
909 | "brief\0" No_argument "q" |
910 | "expand-tabs\0" No_argument "t" |
911 | "initial-tab\0" No_argument "T" |
912 | "recursive\0" No_argument "r" |
913 | "new-file\0" No_argument "N" |
914 | "report-identical-files\0" No_argument "s" |
915 | "starting-file\0" Required_argument "S" |
916 | "minimal\0" No_argument "d" |
917 | ; |
918 | #endif |
919 | |
920 | int diff_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE; |
921 | int diff_main(int argc UNUSED_PARAM, char **argv) |
922 | { |
923 | int gotstdin = 0, i; |
924 | char *file[2], *s_start = NULL; |
925 | llist_t *L_arg = NULL; |
926 | |
927 | INIT_G(); |
928 | |
929 | /* exactly 2 params; collect multiple -L <label>; -U N */ |
930 | opt_complementary = "=2:L::U+"; |
931 | #if ENABLE_FEATURE_DIFF_LONG_OPTIONS |
932 | applet_long_options = diff_longopts; |
933 | #endif |
934 | getopt32(argv, "abdiL:NqrsS:tTU:wupBE", |
935 | &L_arg, &s_start, &opt_U_context); |
936 | argv += optind; |
937 | while (L_arg) |
938 | label[!!label[0]] = llist_pop(&L_arg); |
939 | xfunc_error_retval = 2; |
940 | for (i = 0; i < 2; i++) { |
941 | file[i] = argv[i]; |
942 | /* Compat: "diff file name_which_doesnt_exist" exits with 2 */ |
943 | if (LONE_DASH(file[i])) { |
944 | fstat(STDIN_FILENO, &stb[i]); |
945 | gotstdin++; |
946 | } else |
947 | xstat(file[i], &stb[i]); |
948 | } |
949 | xfunc_error_retval = 1; |
950 | if (gotstdin && (S_ISDIR(stb[0].st_mode) || S_ISDIR(stb[1].st_mode))) |
951 | bb_error_msg_and_die("can't compare stdin to a directory"); |
952 | |
953 | if (S_ISDIR(stb[0].st_mode) && S_ISDIR(stb[1].st_mode)) { |
954 | #if ENABLE_FEATURE_DIFF_DIR |
955 | diffdir(file, s_start); |
956 | #else |
957 | bb_error_msg_and_die("no support for directory comparison"); |
958 | #endif |
959 | } else { |
960 | bool dirfile = S_ISDIR(stb[0].st_mode) || S_ISDIR(stb[1].st_mode); |
961 | bool dir = S_ISDIR(stb[1].st_mode); |
962 | if (dirfile) { |
963 | const char *slash = strrchr(file[!dir], '/'); |
964 | file[dir] = concat_path_file(file[dir], slash ? slash + 1 : file[!dir]); |
965 | xstat(file[dir], &stb[dir]); |
966 | } |
967 | /* diffreg can get non-regular files here */ |
968 | print_status(gotstdin > 1 ? STATUS_SAME : diffreg(file), file); |
969 | |
970 | if (dirfile) |
971 | free(file[dir]); |
972 | } |
973 | |
974 | return exit_status; |
975 | } |