1/*
2 * apply.c
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This applies patches on top of some (arbitrary) version of the SCM.
7 *
8 */
9#include <fnmatch.h>
10#include "cache.h"
11#include "quote.h"
12#include "blob.h"
13#include "delta.h"
14#include "builtin.h"
15
16// --check turns on checking that the working tree matches the
17// files that are being modified, but doesn't apply the patch
18// --stat does just a diffstat, and doesn't actually apply
19// --numstat does numeric diffstat, and doesn't actually apply
20// --index-info shows the old and new index info for paths if available.
21// --index updates the cache as well.
22// --cached updates only the cache without ever touching the working tree.
23//
24static const char *prefix;
25static int prefix_length = -1;
26static int newfd = -1;
27
28static int p_value = 1;
29static int allow_binary_replacement = 0;
30static int check_index = 0;
31static int write_index = 0;
32static int cached = 0;
33static int diffstat = 0;
34static int numstat = 0;
35static int summary = 0;
36static int check = 0;
37static int apply = 1;
38static int no_add = 0;
39static int show_index_info = 0;
40static int line_termination = '\n';
41static unsigned long p_context = -1;
42static const char apply_usage[] =
43"git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|error|error-all|strip>] <patch>...";
44
45static enum whitespace_eol {
46 nowarn_whitespace,
47 warn_on_whitespace,
48 error_on_whitespace,
49 strip_whitespace,
50} new_whitespace = warn_on_whitespace;
51static int whitespace_error = 0;
52static int squelch_whitespace_errors = 5;
53static int applied_after_stripping = 0;
54static const char *patch_input_file = NULL;
55
56static void parse_whitespace_option(const char *option)
57{
58 if (!option) {
59 new_whitespace = warn_on_whitespace;
60 return;
61 }
62 if (!strcmp(option, "warn")) {
63 new_whitespace = warn_on_whitespace;
64 return;
65 }
66 if (!strcmp(option, "nowarn")) {
67 new_whitespace = nowarn_whitespace;
68 return;
69 }
70 if (!strcmp(option, "error")) {
71 new_whitespace = error_on_whitespace;
72 return;
73 }
74 if (!strcmp(option, "error-all")) {
75 new_whitespace = error_on_whitespace;
76 squelch_whitespace_errors = 0;
77 return;
78 }
79 if (!strcmp(option, "strip")) {
80 new_whitespace = strip_whitespace;
81 return;
82 }
83 die("unrecognized whitespace option '%s'", option);
84}
85
86static void set_default_whitespace_mode(const char *whitespace_option)
87{
88 if (!whitespace_option && !apply_default_whitespace) {
89 new_whitespace = (apply
90 ? warn_on_whitespace
91 : nowarn_whitespace);
92 }
93}
94
95/*
96 * For "diff-stat" like behaviour, we keep track of the biggest change
97 * we've seen, and the longest filename. That allows us to do simple
98 * scaling.
99 */
100static int max_change, max_len;
101
102/*
103 * Various "current state", notably line numbers and what
104 * file (and how) we're patching right now.. The "is_xxxx"
105 * things are flags, where -1 means "don't know yet".
106 */
107static int linenr = 1;
108
109struct fragment {
110 unsigned long leading, trailing;
111 unsigned long oldpos, oldlines;
112 unsigned long newpos, newlines;
113 const char *patch;
114 int size;
115 struct fragment *next;
116};
117
118struct patch {
119 char *new_name, *old_name, *def_name;
120 unsigned int old_mode, new_mode;
121 int is_rename, is_copy, is_new, is_delete, is_binary;
122#define BINARY_DELTA_DEFLATED 1
123#define BINARY_LITERAL_DEFLATED 2
124 unsigned long deflate_origlen;
125 int lines_added, lines_deleted;
126 int score;
127 struct fragment *fragments;
128 char *result;
129 unsigned long resultsize;
130 char old_sha1_prefix[41];
131 char new_sha1_prefix[41];
132 struct patch *next;
133};
134
135#define CHUNKSIZE (8192)
136#define SLOP (16)
137
138static void *read_patch_file(int fd, unsigned long *sizep)
139{
140 unsigned long size = 0, alloc = CHUNKSIZE;
141 void *buffer = xmalloc(alloc);
142
143 for (;;) {
144 int nr = alloc - size;
145 if (nr < 1024) {
146 alloc += CHUNKSIZE;
147 buffer = xrealloc(buffer, alloc);
148 nr = alloc - size;
149 }
150 nr = xread(fd, buffer + size, nr);
151 if (!nr)
152 break;
153 if (nr < 0)
154 die("git-apply: read returned %s", strerror(errno));
155 size += nr;
156 }
157 *sizep = size;
158
159 /*
160 * Make sure that we have some slop in the buffer
161 * so that we can do speculative "memcmp" etc, and
162 * see to it that it is NUL-filled.
163 */
164 if (alloc < size + SLOP)
165 buffer = xrealloc(buffer, size + SLOP);
166 memset(buffer + size, 0, SLOP);
167 return buffer;
168}
169
170static unsigned long linelen(const char *buffer, unsigned long size)
171{
172 unsigned long len = 0;
173 while (size--) {
174 len++;
175 if (*buffer++ == '\n')
176 break;
177 }
178 return len;
179}
180
181static int is_dev_null(const char *str)
182{
183 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
184}
185
186#define TERM_SPACE 1
187#define TERM_TAB 2
188
189static int name_terminate(const char *name, int namelen, int c, int terminate)
190{
191 if (c == ' ' && !(terminate & TERM_SPACE))
192 return 0;
193 if (c == '\t' && !(terminate & TERM_TAB))
194 return 0;
195
196 return 1;
197}
198
199static char * find_name(const char *line, char *def, int p_value, int terminate)
200{
201 int len;
202 const char *start = line;
203 char *name;
204
205 if (*line == '"') {
206 /* Proposed "new-style" GNU patch/diff format; see
207 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
208 */
209 name = unquote_c_style(line, NULL);
210 if (name) {
211 char *cp = name;
212 while (p_value) {
213 cp = strchr(name, '/');
214 if (!cp)
215 break;
216 cp++;
217 p_value--;
218 }
219 if (cp) {
220 /* name can later be freed, so we need
221 * to memmove, not just return cp
222 */
223 memmove(name, cp, strlen(cp) + 1);
224 free(def);
225 return name;
226 }
227 else {
228 free(name);
229 name = NULL;
230 }
231 }
232 }
233
234 for (;;) {
235 char c = *line;
236
237 if (isspace(c)) {
238 if (c == '\n')
239 break;
240 if (name_terminate(start, line-start, c, terminate))
241 break;
242 }
243 line++;
244 if (c == '/' && !--p_value)
245 start = line;
246 }
247 if (!start)
248 return def;
249 len = line - start;
250 if (!len)
251 return def;
252
253 /*
254 * Generally we prefer the shorter name, especially
255 * if the other one is just a variation of that with
256 * something else tacked on to the end (ie "file.orig"
257 * or "file~").
258 */
259 if (def) {
260 int deflen = strlen(def);
261 if (deflen < len && !strncmp(start, def, deflen))
262 return def;
263 }
264
265 name = xmalloc(len + 1);
266 memcpy(name, start, len);
267 name[len] = 0;
268 free(def);
269 return name;
270}
271
272/*
273 * Get the name etc info from the --/+++ lines of a traditional patch header
274 *
275 * NOTE! This hardcodes "-p1" behaviour in filename detection.
276 *
277 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
278 * files, we can happily check the index for a match, but for creating a
279 * new file we should try to match whatever "patch" does. I have no idea.
280 */
281static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
282{
283 char *name;
284
285 first += 4; // skip "--- "
286 second += 4; // skip "+++ "
287 if (is_dev_null(first)) {
288 patch->is_new = 1;
289 patch->is_delete = 0;
290 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
291 patch->new_name = name;
292 } else if (is_dev_null(second)) {
293 patch->is_new = 0;
294 patch->is_delete = 1;
295 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
296 patch->old_name = name;
297 } else {
298 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
299 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
300 patch->old_name = patch->new_name = name;
301 }
302 if (!name)
303 die("unable to find filename in patch at line %d", linenr);
304}
305
306static int gitdiff_hdrend(const char *line, struct patch *patch)
307{
308 return -1;
309}
310
311/*
312 * We're anal about diff header consistency, to make
313 * sure that we don't end up having strange ambiguous
314 * patches floating around.
315 *
316 * As a result, gitdiff_{old|new}name() will check
317 * their names against any previous information, just
318 * to make sure..
319 */
320static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
321{
322 if (!orig_name && !isnull)
323 return find_name(line, NULL, 1, 0);
324
325 if (orig_name) {
326 int len;
327 const char *name;
328 char *another;
329 name = orig_name;
330 len = strlen(name);
331 if (isnull)
332 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
333 another = find_name(line, NULL, 1, 0);
334 if (!another || memcmp(another, name, len))
335 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
336 free(another);
337 return orig_name;
338 }
339 else {
340 /* expect "/dev/null" */
341 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
342 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
343 return NULL;
344 }
345}
346
347static int gitdiff_oldname(const char *line, struct patch *patch)
348{
349 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
350 return 0;
351}
352
353static int gitdiff_newname(const char *line, struct patch *patch)
354{
355 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
356 return 0;
357}
358
359static int gitdiff_oldmode(const char *line, struct patch *patch)
360{
361 patch->old_mode = strtoul(line, NULL, 8);
362 return 0;
363}
364
365static int gitdiff_newmode(const char *line, struct patch *patch)
366{
367 patch->new_mode = strtoul(line, NULL, 8);
368 return 0;
369}
370
371static int gitdiff_delete(const char *line, struct patch *patch)
372{
373 patch->is_delete = 1;
374 patch->old_name = patch->def_name;
375 return gitdiff_oldmode(line, patch);
376}
377
378static int gitdiff_newfile(const char *line, struct patch *patch)
379{
380 patch->is_new = 1;
381 patch->new_name = patch->def_name;
382 return gitdiff_newmode(line, patch);
383}
384
385static int gitdiff_copysrc(const char *line, struct patch *patch)
386{
387 patch->is_copy = 1;
388 patch->old_name = find_name(line, NULL, 0, 0);
389 return 0;
390}
391
392static int gitdiff_copydst(const char *line, struct patch *patch)
393{
394 patch->is_copy = 1;
395 patch->new_name = find_name(line, NULL, 0, 0);
396 return 0;
397}
398
399static int gitdiff_renamesrc(const char *line, struct patch *patch)
400{
401 patch->is_rename = 1;
402 patch->old_name = find_name(line, NULL, 0, 0);
403 return 0;
404}
405
406static int gitdiff_renamedst(const char *line, struct patch *patch)
407{
408 patch->is_rename = 1;
409 patch->new_name = find_name(line, NULL, 0, 0);
410 return 0;
411}
412
413static int gitdiff_similarity(const char *line, struct patch *patch)
414{
415 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
416 patch->score = 0;
417 return 0;
418}
419
420static int gitdiff_dissimilarity(const char *line, struct patch *patch)
421{
422 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
423 patch->score = 0;
424 return 0;
425}
426
427static int gitdiff_index(const char *line, struct patch *patch)
428{
429 /* index line is N hexadecimal, "..", N hexadecimal,
430 * and optional space with octal mode.
431 */
432 const char *ptr, *eol;
433 int len;
434
435 ptr = strchr(line, '.');
436 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
437 return 0;
438 len = ptr - line;
439 memcpy(patch->old_sha1_prefix, line, len);
440 patch->old_sha1_prefix[len] = 0;
441
442 line = ptr + 2;
443 ptr = strchr(line, ' ');
444 eol = strchr(line, '\n');
445
446 if (!ptr || eol < ptr)
447 ptr = eol;
448 len = ptr - line;
449
450 if (40 < len)
451 return 0;
452 memcpy(patch->new_sha1_prefix, line, len);
453 patch->new_sha1_prefix[len] = 0;
454 if (*ptr == ' ')
455 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
456 return 0;
457}
458
459/*
460 * This is normal for a diff that doesn't change anything: we'll fall through
461 * into the next diff. Tell the parser to break out.
462 */
463static int gitdiff_unrecognized(const char *line, struct patch *patch)
464{
465 return -1;
466}
467
468static const char *stop_at_slash(const char *line, int llen)
469{
470 int i;
471
472 for (i = 0; i < llen; i++) {
473 int ch = line[i];
474 if (ch == '/')
475 return line + i;
476 }
477 return NULL;
478}
479
480/* This is to extract the same name that appears on "diff --git"
481 * line. We do not find and return anything if it is a rename
482 * patch, and it is OK because we will find the name elsewhere.
483 * We need to reliably find name only when it is mode-change only,
484 * creation or deletion of an empty file. In any of these cases,
485 * both sides are the same name under a/ and b/ respectively.
486 */
487static char *git_header_name(char *line, int llen)
488{
489 int len;
490 const char *name;
491 const char *second = NULL;
492
493 line += strlen("diff --git ");
494 llen -= strlen("diff --git ");
495
496 if (*line == '"') {
497 const char *cp;
498 char *first = unquote_c_style(line, &second);
499 if (!first)
500 return NULL;
501
502 /* advance to the first slash */
503 cp = stop_at_slash(first, strlen(first));
504 if (!cp || cp == first) {
505 /* we do not accept absolute paths */
506 free_first_and_fail:
507 free(first);
508 return NULL;
509 }
510 len = strlen(cp+1);
511 memmove(first, cp+1, len+1); /* including NUL */
512
513 /* second points at one past closing dq of name.
514 * find the second name.
515 */
516 while ((second < line + llen) && isspace(*second))
517 second++;
518
519 if (line + llen <= second)
520 goto free_first_and_fail;
521 if (*second == '"') {
522 char *sp = unquote_c_style(second, NULL);
523 if (!sp)
524 goto free_first_and_fail;
525 cp = stop_at_slash(sp, strlen(sp));
526 if (!cp || cp == sp) {
527 free_both_and_fail:
528 free(sp);
529 goto free_first_and_fail;
530 }
531 /* They must match, otherwise ignore */
532 if (strcmp(cp+1, first))
533 goto free_both_and_fail;
534 free(sp);
535 return first;
536 }
537
538 /* unquoted second */
539 cp = stop_at_slash(second, line + llen - second);
540 if (!cp || cp == second)
541 goto free_first_and_fail;
542 cp++;
543 if (line + llen - cp != len + 1 ||
544 memcmp(first, cp, len))
545 goto free_first_and_fail;
546 return first;
547 }
548
549 /* unquoted first name */
550 name = stop_at_slash(line, llen);
551 if (!name || name == line)
552 return NULL;
553
554 name++;
555
556 /* since the first name is unquoted, a dq if exists must be
557 * the beginning of the second name.
558 */
559 for (second = name; second < line + llen; second++) {
560 if (*second == '"') {
561 const char *cp = second;
562 const char *np;
563 char *sp = unquote_c_style(second, NULL);
564
565 if (!sp)
566 return NULL;
567 np = stop_at_slash(sp, strlen(sp));
568 if (!np || np == sp) {
569 free_second_and_fail:
570 free(sp);
571 return NULL;
572 }
573 np++;
574 len = strlen(np);
575 if (len < cp - name &&
576 !strncmp(np, name, len) &&
577 isspace(name[len])) {
578 /* Good */
579 memmove(sp, np, len + 1);
580 return sp;
581 }
582 goto free_second_and_fail;
583 }
584 }
585
586 /*
587 * Accept a name only if it shows up twice, exactly the same
588 * form.
589 */
590 for (len = 0 ; ; len++) {
591 char c = name[len];
592
593 switch (c) {
594 default:
595 continue;
596 case '\n':
597 return NULL;
598 case '\t': case ' ':
599 second = name+len;
600 for (;;) {
601 char c = *second++;
602 if (c == '\n')
603 return NULL;
604 if (c == '/')
605 break;
606 }
607 if (second[len] == '\n' && !memcmp(name, second, len)) {
608 char *ret = xmalloc(len + 1);
609 memcpy(ret, name, len);
610 ret[len] = 0;
611 return ret;
612 }
613 }
614 }
615 return NULL;
616}
617
618/* Verify that we recognize the lines following a git header */
619static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
620{
621 unsigned long offset;
622
623 /* A git diff has explicit new/delete information, so we don't guess */
624 patch->is_new = 0;
625 patch->is_delete = 0;
626
627 /*
628 * Some things may not have the old name in the
629 * rest of the headers anywhere (pure mode changes,
630 * or removing or adding empty files), so we get
631 * the default name from the header.
632 */
633 patch->def_name = git_header_name(line, len);
634
635 line += len;
636 size -= len;
637 linenr++;
638 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
639 static const struct opentry {
640 const char *str;
641 int (*fn)(const char *, struct patch *);
642 } optable[] = {
643 { "@@ -", gitdiff_hdrend },
644 { "--- ", gitdiff_oldname },
645 { "+++ ", gitdiff_newname },
646 { "old mode ", gitdiff_oldmode },
647 { "new mode ", gitdiff_newmode },
648 { "deleted file mode ", gitdiff_delete },
649 { "new file mode ", gitdiff_newfile },
650 { "copy from ", gitdiff_copysrc },
651 { "copy to ", gitdiff_copydst },
652 { "rename old ", gitdiff_renamesrc },
653 { "rename new ", gitdiff_renamedst },
654 { "rename from ", gitdiff_renamesrc },
655 { "rename to ", gitdiff_renamedst },
656 { "similarity index ", gitdiff_similarity },
657 { "dissimilarity index ", gitdiff_dissimilarity },
658 { "index ", gitdiff_index },
659 { "", gitdiff_unrecognized },
660 };
661 int i;
662
663 len = linelen(line, size);
664 if (!len || line[len-1] != '\n')
665 break;
666 for (i = 0; i < ARRAY_SIZE(optable); i++) {
667 const struct opentry *p = optable + i;
668 int oplen = strlen(p->str);
669 if (len < oplen || memcmp(p->str, line, oplen))
670 continue;
671 if (p->fn(line + oplen, patch) < 0)
672 return offset;
673 break;
674 }
675 }
676
677 return offset;
678}
679
680static int parse_num(const char *line, unsigned long *p)
681{
682 char *ptr;
683
684 if (!isdigit(*line))
685 return 0;
686 *p = strtoul(line, &ptr, 10);
687 return ptr - line;
688}
689
690static int parse_range(const char *line, int len, int offset, const char *expect,
691 unsigned long *p1, unsigned long *p2)
692{
693 int digits, ex;
694
695 if (offset < 0 || offset >= len)
696 return -1;
697 line += offset;
698 len -= offset;
699
700 digits = parse_num(line, p1);
701 if (!digits)
702 return -1;
703
704 offset += digits;
705 line += digits;
706 len -= digits;
707
708 *p2 = 1;
709 if (*line == ',') {
710 digits = parse_num(line+1, p2);
711 if (!digits)
712 return -1;
713
714 offset += digits+1;
715 line += digits+1;
716 len -= digits+1;
717 }
718
719 ex = strlen(expect);
720 if (ex > len)
721 return -1;
722 if (memcmp(line, expect, ex))
723 return -1;
724
725 return offset + ex;
726}
727
728/*
729 * Parse a unified diff fragment header of the
730 * form "@@ -a,b +c,d @@"
731 */
732static int parse_fragment_header(char *line, int len, struct fragment *fragment)
733{
734 int offset;
735
736 if (!len || line[len-1] != '\n')
737 return -1;
738
739 /* Figure out the number of lines in a fragment */
740 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
741 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
742
743 return offset;
744}
745
746static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
747{
748 unsigned long offset, len;
749
750 patch->is_rename = patch->is_copy = 0;
751 patch->is_new = patch->is_delete = -1;
752 patch->old_mode = patch->new_mode = 0;
753 patch->old_name = patch->new_name = NULL;
754 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
755 unsigned long nextlen;
756
757 len = linelen(line, size);
758 if (!len)
759 break;
760
761 /* Testing this early allows us to take a few shortcuts.. */
762 if (len < 6)
763 continue;
764
765 /*
766 * Make sure we don't find any unconnected patch fragmants.
767 * That's a sign that we didn't find a header, and that a
768 * patch has become corrupted/broken up.
769 */
770 if (!memcmp("@@ -", line, 4)) {
771 struct fragment dummy;
772 if (parse_fragment_header(line, len, &dummy) < 0)
773 continue;
774 error("patch fragment without header at line %d: %.*s", linenr, (int)len-1, line);
775 }
776
777 if (size < len + 6)
778 break;
779
780 /*
781 * Git patch? It might not have a real patch, just a rename
782 * or mode change, so we handle that specially
783 */
784 if (!memcmp("diff --git ", line, 11)) {
785 int git_hdr_len = parse_git_header(line, len, size, patch);
786 if (git_hdr_len <= len)
787 continue;
788 if (!patch->old_name && !patch->new_name) {
789 if (!patch->def_name)
790 die("git diff header lacks filename information (line %d)", linenr);
791 patch->old_name = patch->new_name = patch->def_name;
792 }
793 *hdrsize = git_hdr_len;
794 return offset;
795 }
796
797 /** --- followed by +++ ? */
798 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
799 continue;
800
801 /*
802 * We only accept unified patches, so we want it to
803 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
804 * minimum
805 */
806 nextlen = linelen(line + len, size - len);
807 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
808 continue;
809
810 /* Ok, we'll consider it a patch */
811 parse_traditional_patch(line, line+len, patch);
812 *hdrsize = len + nextlen;
813 linenr += 2;
814 return offset;
815 }
816 return -1;
817}
818
819/*
820 * Parse a unified diff. Note that this really needs
821 * to parse each fragment separately, since the only
822 * way to know the difference between a "---" that is
823 * part of a patch, and a "---" that starts the next
824 * patch is to look at the line counts..
825 */
826static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
827{
828 int added, deleted;
829 int len = linelen(line, size), offset;
830 unsigned long oldlines, newlines;
831 unsigned long leading, trailing;
832
833 offset = parse_fragment_header(line, len, fragment);
834 if (offset < 0)
835 return -1;
836 oldlines = fragment->oldlines;
837 newlines = fragment->newlines;
838 leading = 0;
839 trailing = 0;
840
841 if (patch->is_new < 0) {
842 patch->is_new = !oldlines;
843 if (!oldlines)
844 patch->old_name = NULL;
845 }
846 if (patch->is_delete < 0) {
847 patch->is_delete = !newlines;
848 if (!newlines)
849 patch->new_name = NULL;
850 }
851
852 if (patch->is_new && oldlines)
853 return error("new file depends on old contents");
854 if (patch->is_delete != !newlines) {
855 if (newlines)
856 return error("deleted file still has contents");
857 fprintf(stderr, "** warning: file %s becomes empty but is not deleted\n", patch->new_name);
858 }
859
860 /* Parse the thing.. */
861 line += len;
862 size -= len;
863 linenr++;
864 added = deleted = 0;
865 for (offset = len; size > 0; offset += len, size -= len, line += len, linenr++) {
866 if (!oldlines && !newlines)
867 break;
868 len = linelen(line, size);
869 if (!len || line[len-1] != '\n')
870 return -1;
871 switch (*line) {
872 default:
873 return -1;
874 case ' ':
875 oldlines--;
876 newlines--;
877 if (!deleted && !added)
878 leading++;
879 trailing++;
880 break;
881 case '-':
882 deleted++;
883 oldlines--;
884 trailing = 0;
885 break;
886 case '+':
887 /*
888 * We know len is at least two, since we have a '+' and
889 * we checked that the last character was a '\n' above.
890 * That is, an addition of an empty line would check
891 * the '+' here. Sneaky...
892 */
893 if ((new_whitespace != nowarn_whitespace) &&
894 isspace(line[len-2])) {
895 whitespace_error++;
896 if (squelch_whitespace_errors &&
897 squelch_whitespace_errors <
898 whitespace_error)
899 ;
900 else {
901 fprintf(stderr, "Adds trailing whitespace.\n%s:%d:%.*s\n",
902 patch_input_file,
903 linenr, len-2, line+1);
904 }
905 }
906 added++;
907 newlines--;
908 trailing = 0;
909 break;
910
911 /* We allow "\ No newline at end of file". Depending
912 * on locale settings when the patch was produced we
913 * don't know what this line looks like. The only
914 * thing we do know is that it begins with "\ ".
915 * Checking for 12 is just for sanity check -- any
916 * l10n of "\ No newline..." is at least that long.
917 */
918 case '\\':
919 if (len < 12 || memcmp(line, "\\ ", 2))
920 return -1;
921 break;
922 }
923 }
924 if (oldlines || newlines)
925 return -1;
926 fragment->leading = leading;
927 fragment->trailing = trailing;
928
929 /* If a fragment ends with an incomplete line, we failed to include
930 * it in the above loop because we hit oldlines == newlines == 0
931 * before seeing it.
932 */
933 if (12 < size && !memcmp(line, "\\ ", 2))
934 offset += linelen(line, size);
935
936 patch->lines_added += added;
937 patch->lines_deleted += deleted;
938 return offset;
939}
940
941static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
942{
943 unsigned long offset = 0;
944 struct fragment **fragp = &patch->fragments;
945
946 while (size > 4 && !memcmp(line, "@@ -", 4)) {
947 struct fragment *fragment;
948 int len;
949
950 fragment = xcalloc(1, sizeof(*fragment));
951 len = parse_fragment(line, size, patch, fragment);
952 if (len <= 0)
953 die("corrupt patch at line %d", linenr);
954
955 fragment->patch = line;
956 fragment->size = len;
957
958 *fragp = fragment;
959 fragp = &fragment->next;
960
961 offset += len;
962 line += len;
963 size -= len;
964 }
965 return offset;
966}
967
968static inline int metadata_changes(struct patch *patch)
969{
970 return patch->is_rename > 0 ||
971 patch->is_copy > 0 ||
972 patch->is_new > 0 ||
973 patch->is_delete ||
974 (patch->old_mode && patch->new_mode &&
975 patch->old_mode != patch->new_mode);
976}
977
978static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
979{
980 /* We have read "GIT binary patch\n"; what follows is a line
981 * that says the patch method (currently, either "deflated
982 * literal" or "deflated delta") and the length of data before
983 * deflating; a sequence of 'length-byte' followed by base-85
984 * encoded data follows.
985 *
986 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
987 * and we would limit the patch line to 66 characters,
988 * so one line can fit up to 13 groups that would decode
989 * to 52 bytes max. The length byte 'A'-'Z' corresponds
990 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
991 * The end of binary is signalled with an empty line.
992 */
993 int llen, used;
994 struct fragment *fragment;
995 char *data = NULL;
996
997 patch->fragments = fragment = xcalloc(1, sizeof(*fragment));
998
999 /* Grab the type of patch */
1000 llen = linelen(buffer, size);
1001 used = llen;
1002 linenr++;
1003
1004 if (!strncmp(buffer, "delta ", 6)) {
1005 patch->is_binary = BINARY_DELTA_DEFLATED;
1006 patch->deflate_origlen = strtoul(buffer + 6, NULL, 10);
1007 }
1008 else if (!strncmp(buffer, "literal ", 8)) {
1009 patch->is_binary = BINARY_LITERAL_DEFLATED;
1010 patch->deflate_origlen = strtoul(buffer + 8, NULL, 10);
1011 }
1012 else
1013 return error("unrecognized binary patch at line %d: %.*s",
1014 linenr-1, llen-1, buffer);
1015 buffer += llen;
1016 while (1) {
1017 int byte_length, max_byte_length, newsize;
1018 llen = linelen(buffer, size);
1019 used += llen;
1020 linenr++;
1021 if (llen == 1)
1022 break;
1023 /* Minimum line is "A00000\n" which is 7-byte long,
1024 * and the line length must be multiple of 5 plus 2.
1025 */
1026 if ((llen < 7) || (llen-2) % 5)
1027 goto corrupt;
1028 max_byte_length = (llen - 2) / 5 * 4;
1029 byte_length = *buffer;
1030 if ('A' <= byte_length && byte_length <= 'Z')
1031 byte_length = byte_length - 'A' + 1;
1032 else if ('a' <= byte_length && byte_length <= 'z')
1033 byte_length = byte_length - 'a' + 27;
1034 else
1035 goto corrupt;
1036 /* if the input length was not multiple of 4, we would
1037 * have filler at the end but the filler should never
1038 * exceed 3 bytes
1039 */
1040 if (max_byte_length < byte_length ||
1041 byte_length <= max_byte_length - 4)
1042 goto corrupt;
1043 newsize = fragment->size + byte_length;
1044 data = xrealloc(data, newsize);
1045 if (decode_85(data + fragment->size,
1046 buffer + 1,
1047 byte_length))
1048 goto corrupt;
1049 fragment->size = newsize;
1050 buffer += llen;
1051 size -= llen;
1052 }
1053 fragment->patch = data;
1054 return used;
1055 corrupt:
1056 return error("corrupt binary patch at line %d: %.*s",
1057 linenr-1, llen-1, buffer);
1058}
1059
1060static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1061{
1062 int hdrsize, patchsize;
1063 int offset = find_header(buffer, size, &hdrsize, patch);
1064
1065 if (offset < 0)
1066 return offset;
1067
1068 patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
1069
1070 if (!patchsize) {
1071 static const char *binhdr[] = {
1072 "Binary files ",
1073 "Files ",
1074 NULL,
1075 };
1076 static const char git_binary[] = "GIT binary patch\n";
1077 int i;
1078 int hd = hdrsize + offset;
1079 unsigned long llen = linelen(buffer + hd, size - hd);
1080
1081 if (llen == sizeof(git_binary) - 1 &&
1082 !memcmp(git_binary, buffer + hd, llen)) {
1083 int used;
1084 linenr++;
1085 used = parse_binary(buffer + hd + llen,
1086 size - hd - llen, patch);
1087 if (used)
1088 patchsize = used + llen;
1089 else
1090 patchsize = 0;
1091 }
1092 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1093 for (i = 0; binhdr[i]; i++) {
1094 int len = strlen(binhdr[i]);
1095 if (len < size - hd &&
1096 !memcmp(binhdr[i], buffer + hd, len)) {
1097 linenr++;
1098 patch->is_binary = 1;
1099 patchsize = llen;
1100 break;
1101 }
1102 }
1103 }
1104
1105 /* Empty patch cannot be applied if:
1106 * - it is a binary patch and we do not do binary_replace, or
1107 * - text patch without metadata change
1108 */
1109 if ((apply || check) &&
1110 (patch->is_binary
1111 ? !allow_binary_replacement
1112 : !metadata_changes(patch)))
1113 die("patch with only garbage at line %d", linenr);
1114 }
1115
1116 return offset + hdrsize + patchsize;
1117}
1118
1119static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1120static const char minuses[]= "----------------------------------------------------------------------";
1121
1122static void show_stats(struct patch *patch)
1123{
1124 const char *prefix = "";
1125 char *name = patch->new_name;
1126 char *qname = NULL;
1127 int len, max, add, del, total;
1128
1129 if (!name)
1130 name = patch->old_name;
1131
1132 if (0 < (len = quote_c_style(name, NULL, NULL, 0))) {
1133 qname = xmalloc(len + 1);
1134 quote_c_style(name, qname, NULL, 0);
1135 name = qname;
1136 }
1137
1138 /*
1139 * "scale" the filename
1140 */
1141 len = strlen(name);
1142 max = max_len;
1143 if (max > 50)
1144 max = 50;
1145 if (len > max) {
1146 char *slash;
1147 prefix = "...";
1148 max -= 3;
1149 name += len - max;
1150 slash = strchr(name, '/');
1151 if (slash)
1152 name = slash;
1153 }
1154 len = max;
1155
1156 /*
1157 * scale the add/delete
1158 */
1159 max = max_change;
1160 if (max + len > 70)
1161 max = 70 - len;
1162
1163 add = patch->lines_added;
1164 del = patch->lines_deleted;
1165 total = add + del;
1166
1167 if (max_change > 0) {
1168 total = (total * max + max_change / 2) / max_change;
1169 add = (add * max + max_change / 2) / max_change;
1170 del = total - add;
1171 }
1172 if (patch->is_binary)
1173 printf(" %s%-*s | Bin\n", prefix, len, name);
1174 else
1175 printf(" %s%-*s |%5d %.*s%.*s\n", prefix,
1176 len, name, patch->lines_added + patch->lines_deleted,
1177 add, pluses, del, minuses);
1178 if (qname)
1179 free(qname);
1180}
1181
1182static int read_old_data(struct stat *st, const char *path, void *buf, unsigned long size)
1183{
1184 int fd;
1185 unsigned long got;
1186
1187 switch (st->st_mode & S_IFMT) {
1188 case S_IFLNK:
1189 return readlink(path, buf, size);
1190 case S_IFREG:
1191 fd = open(path, O_RDONLY);
1192 if (fd < 0)
1193 return error("unable to open %s", path);
1194 got = 0;
1195 for (;;) {
1196 int ret = xread(fd, buf + got, size - got);
1197 if (ret <= 0)
1198 break;
1199 got += ret;
1200 }
1201 close(fd);
1202 return got;
1203
1204 default:
1205 return -1;
1206 }
1207}
1208
1209static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line, int *lines)
1210{
1211 int i;
1212 unsigned long start, backwards, forwards;
1213
1214 if (fragsize > size)
1215 return -1;
1216
1217 start = 0;
1218 if (line > 1) {
1219 unsigned long offset = 0;
1220 i = line-1;
1221 while (offset + fragsize <= size) {
1222 if (buf[offset++] == '\n') {
1223 start = offset;
1224 if (!--i)
1225 break;
1226 }
1227 }
1228 }
1229
1230 /* Exact line number? */
1231 if (!memcmp(buf + start, fragment, fragsize))
1232 return start;
1233
1234 /*
1235 * There's probably some smart way to do this, but I'll leave
1236 * that to the smart and beautiful people. I'm simple and stupid.
1237 */
1238 backwards = start;
1239 forwards = start;
1240 for (i = 0; ; i++) {
1241 unsigned long try;
1242 int n;
1243
1244 /* "backward" */
1245 if (i & 1) {
1246 if (!backwards) {
1247 if (forwards + fragsize > size)
1248 break;
1249 continue;
1250 }
1251 do {
1252 --backwards;
1253 } while (backwards && buf[backwards-1] != '\n');
1254 try = backwards;
1255 } else {
1256 while (forwards + fragsize <= size) {
1257 if (buf[forwards++] == '\n')
1258 break;
1259 }
1260 try = forwards;
1261 }
1262
1263 if (try + fragsize > size)
1264 continue;
1265 if (memcmp(buf + try, fragment, fragsize))
1266 continue;
1267 n = (i >> 1)+1;
1268 if (i & 1)
1269 n = -n;
1270 *lines = n;
1271 return try;
1272 }
1273
1274 /*
1275 * We should start searching forward and backward.
1276 */
1277 return -1;
1278}
1279
1280static void remove_first_line(const char **rbuf, int *rsize)
1281{
1282 const char *buf = *rbuf;
1283 int size = *rsize;
1284 unsigned long offset;
1285 offset = 0;
1286 while (offset <= size) {
1287 if (buf[offset++] == '\n')
1288 break;
1289 }
1290 *rsize = size - offset;
1291 *rbuf = buf + offset;
1292}
1293
1294static void remove_last_line(const char **rbuf, int *rsize)
1295{
1296 const char *buf = *rbuf;
1297 int size = *rsize;
1298 unsigned long offset;
1299 offset = size - 1;
1300 while (offset > 0) {
1301 if (buf[--offset] == '\n')
1302 break;
1303 }
1304 *rsize = offset + 1;
1305}
1306
1307struct buffer_desc {
1308 char *buffer;
1309 unsigned long size;
1310 unsigned long alloc;
1311};
1312
1313static int apply_line(char *output, const char *patch, int plen)
1314{
1315 /* plen is number of bytes to be copied from patch,
1316 * starting at patch+1 (patch[0] is '+'). Typically
1317 * patch[plen] is '\n'.
1318 */
1319 int add_nl_to_tail = 0;
1320 if ((new_whitespace == strip_whitespace) &&
1321 1 < plen && isspace(patch[plen-1])) {
1322 if (patch[plen] == '\n')
1323 add_nl_to_tail = 1;
1324 plen--;
1325 while (0 < plen && isspace(patch[plen]))
1326 plen--;
1327 applied_after_stripping++;
1328 }
1329 memcpy(output, patch + 1, plen);
1330 if (add_nl_to_tail)
1331 output[plen++] = '\n';
1332 return plen;
1333}
1334
1335static int apply_one_fragment(struct buffer_desc *desc, struct fragment *frag)
1336{
1337 char *buf = desc->buffer;
1338 const char *patch = frag->patch;
1339 int offset, size = frag->size;
1340 char *old = xmalloc(size);
1341 char *new = xmalloc(size);
1342 const char *oldlines, *newlines;
1343 int oldsize = 0, newsize = 0;
1344 unsigned long leading, trailing;
1345 int pos, lines;
1346
1347 while (size > 0) {
1348 int len = linelen(patch, size);
1349 int plen;
1350
1351 if (!len)
1352 break;
1353
1354 /*
1355 * "plen" is how much of the line we should use for
1356 * the actual patch data. Normally we just remove the
1357 * first character on the line, but if the line is
1358 * followed by "\ No newline", then we also remove the
1359 * last one (which is the newline, of course).
1360 */
1361 plen = len-1;
1362 if (len < size && patch[len] == '\\')
1363 plen--;
1364 switch (*patch) {
1365 case ' ':
1366 case '-':
1367 memcpy(old + oldsize, patch + 1, plen);
1368 oldsize += plen;
1369 if (*patch == '-')
1370 break;
1371 /* Fall-through for ' ' */
1372 case '+':
1373 if (*patch != '+' || !no_add)
1374 newsize += apply_line(new + newsize, patch,
1375 plen);
1376 break;
1377 case '@': case '\\':
1378 /* Ignore it, we already handled it */
1379 break;
1380 default:
1381 return -1;
1382 }
1383 patch += len;
1384 size -= len;
1385 }
1386
1387#ifdef NO_ACCURATE_DIFF
1388 if (oldsize > 0 && old[oldsize - 1] == '\n' &&
1389 newsize > 0 && new[newsize - 1] == '\n') {
1390 oldsize--;
1391 newsize--;
1392 }
1393#endif
1394
1395 oldlines = old;
1396 newlines = new;
1397 leading = frag->leading;
1398 trailing = frag->trailing;
1399 lines = 0;
1400 pos = frag->newpos;
1401 for (;;) {
1402 offset = find_offset(buf, desc->size, oldlines, oldsize, pos, &lines);
1403 if (offset >= 0) {
1404 int diff = newsize - oldsize;
1405 unsigned long size = desc->size + diff;
1406 unsigned long alloc = desc->alloc;
1407
1408 /* Warn if it was necessary to reduce the number
1409 * of context lines.
1410 */
1411 if ((leading != frag->leading) || (trailing != frag->trailing))
1412 fprintf(stderr, "Context reduced to (%ld/%ld) to apply fragment at %d\n",
1413 leading, trailing, pos + lines);
1414
1415 if (size > alloc) {
1416 alloc = size + 8192;
1417 desc->alloc = alloc;
1418 buf = xrealloc(buf, alloc);
1419 desc->buffer = buf;
1420 }
1421 desc->size = size;
1422 memmove(buf + offset + newsize, buf + offset + oldsize, size - offset - newsize);
1423 memcpy(buf + offset, newlines, newsize);
1424 offset = 0;
1425
1426 break;
1427 }
1428
1429 /* Am I at my context limits? */
1430 if ((leading <= p_context) && (trailing <= p_context))
1431 break;
1432 /* Reduce the number of context lines
1433 * Reduce both leading and trailing if they are equal
1434 * otherwise just reduce the larger context.
1435 */
1436 if (leading >= trailing) {
1437 remove_first_line(&oldlines, &oldsize);
1438 remove_first_line(&newlines, &newsize);
1439 pos--;
1440 leading--;
1441 }
1442 if (trailing > leading) {
1443 remove_last_line(&oldlines, &oldsize);
1444 remove_last_line(&newlines, &newsize);
1445 trailing--;
1446 }
1447 }
1448
1449 free(old);
1450 free(new);
1451 return offset;
1452}
1453
1454static char *inflate_it(const void *data, unsigned long size,
1455 unsigned long inflated_size)
1456{
1457 z_stream stream;
1458 void *out;
1459 int st;
1460
1461 memset(&stream, 0, sizeof(stream));
1462
1463 stream.next_in = (unsigned char *)data;
1464 stream.avail_in = size;
1465 stream.next_out = out = xmalloc(inflated_size);
1466 stream.avail_out = inflated_size;
1467 inflateInit(&stream);
1468 st = inflate(&stream, Z_FINISH);
1469 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1470 free(out);
1471 return NULL;
1472 }
1473 return out;
1474}
1475
1476static int apply_binary_fragment(struct buffer_desc *desc, struct patch *patch)
1477{
1478 unsigned long dst_size;
1479 struct fragment *fragment = patch->fragments;
1480 void *data;
1481 void *result;
1482
1483 data = inflate_it(fragment->patch, fragment->size,
1484 patch->deflate_origlen);
1485 if (!data)
1486 return error("corrupt patch data");
1487 switch (patch->is_binary) {
1488 case BINARY_DELTA_DEFLATED:
1489 result = patch_delta(desc->buffer, desc->size,
1490 data,
1491 patch->deflate_origlen,
1492 &dst_size);
1493 free(desc->buffer);
1494 desc->buffer = result;
1495 free(data);
1496 break;
1497 case BINARY_LITERAL_DEFLATED:
1498 free(desc->buffer);
1499 desc->buffer = data;
1500 dst_size = patch->deflate_origlen;
1501 break;
1502 }
1503 if (!desc->buffer)
1504 return -1;
1505 desc->size = desc->alloc = dst_size;
1506 return 0;
1507}
1508
1509static int apply_binary(struct buffer_desc *desc, struct patch *patch)
1510{
1511 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1512 unsigned char sha1[20];
1513 unsigned char hdr[50];
1514 int hdrlen;
1515
1516 if (!allow_binary_replacement)
1517 return error("cannot apply binary patch to '%s' "
1518 "without --allow-binary-replacement",
1519 name);
1520
1521 /* For safety, we require patch index line to contain
1522 * full 40-byte textual SHA1 for old and new, at least for now.
1523 */
1524 if (strlen(patch->old_sha1_prefix) != 40 ||
1525 strlen(patch->new_sha1_prefix) != 40 ||
1526 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1527 get_sha1_hex(patch->new_sha1_prefix, sha1))
1528 return error("cannot apply binary patch to '%s' "
1529 "without full index line", name);
1530
1531 if (patch->old_name) {
1532 /* See if the old one matches what the patch
1533 * applies to.
1534 */
1535 write_sha1_file_prepare(desc->buffer, desc->size,
1536 blob_type, sha1, hdr, &hdrlen);
1537 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1538 return error("the patch applies to '%s' (%s), "
1539 "which does not match the "
1540 "current contents.",
1541 name, sha1_to_hex(sha1));
1542 }
1543 else {
1544 /* Otherwise, the old one must be empty. */
1545 if (desc->size)
1546 return error("the patch applies to an empty "
1547 "'%s' but it is not empty", name);
1548 }
1549
1550 get_sha1_hex(patch->new_sha1_prefix, sha1);
1551 if (!memcmp(sha1, null_sha1, 20)) {
1552 free(desc->buffer);
1553 desc->alloc = desc->size = 0;
1554 desc->buffer = NULL;
1555 return 0; /* deletion patch */
1556 }
1557
1558 if (has_sha1_file(sha1)) {
1559 /* We already have the postimage */
1560 char type[10];
1561 unsigned long size;
1562
1563 free(desc->buffer);
1564 desc->buffer = read_sha1_file(sha1, type, &size);
1565 if (!desc->buffer)
1566 return error("the necessary postimage %s for "
1567 "'%s' cannot be read",
1568 patch->new_sha1_prefix, name);
1569 desc->alloc = desc->size = size;
1570 }
1571 else {
1572 /* We have verified desc matches the preimage;
1573 * apply the patch data to it, which is stored
1574 * in the patch->fragments->{patch,size}.
1575 */
1576 if (apply_binary_fragment(desc, patch))
1577 return error("binary patch does not apply to '%s'",
1578 name);
1579
1580 /* verify that the result matches */
1581 write_sha1_file_prepare(desc->buffer, desc->size, blob_type,
1582 sha1, hdr, &hdrlen);
1583 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1584 return error("binary patch to '%s' creates incorrect result", name);
1585 }
1586
1587 return 0;
1588}
1589
1590static int apply_fragments(struct buffer_desc *desc, struct patch *patch)
1591{
1592 struct fragment *frag = patch->fragments;
1593 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1594
1595 if (patch->is_binary)
1596 return apply_binary(desc, patch);
1597
1598 while (frag) {
1599 if (apply_one_fragment(desc, frag) < 0)
1600 return error("patch failed: %s:%ld",
1601 name, frag->oldpos);
1602 frag = frag->next;
1603 }
1604 return 0;
1605}
1606
1607static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
1608{
1609 char *buf;
1610 unsigned long size, alloc;
1611 struct buffer_desc desc;
1612
1613 size = 0;
1614 alloc = 0;
1615 buf = NULL;
1616 if (cached) {
1617 if (ce) {
1618 char type[20];
1619 buf = read_sha1_file(ce->sha1, type, &size);
1620 if (!buf)
1621 return error("read of %s failed",
1622 patch->old_name);
1623 alloc = size;
1624 }
1625 }
1626 else if (patch->old_name) {
1627 size = st->st_size;
1628 alloc = size + 8192;
1629 buf = xmalloc(alloc);
1630 if (read_old_data(st, patch->old_name, buf, alloc) != size)
1631 return error("read of %s failed", patch->old_name);
1632 }
1633
1634 desc.size = size;
1635 desc.alloc = alloc;
1636 desc.buffer = buf;
1637 if (apply_fragments(&desc, patch) < 0)
1638 return -1;
1639 patch->result = desc.buffer;
1640 patch->resultsize = desc.size;
1641
1642 if (patch->is_delete && patch->resultsize)
1643 return error("removal patch leaves file contents");
1644
1645 return 0;
1646}
1647
1648static int check_patch(struct patch *patch)
1649{
1650 struct stat st;
1651 const char *old_name = patch->old_name;
1652 const char *new_name = patch->new_name;
1653 const char *name = old_name ? old_name : new_name;
1654 struct cache_entry *ce = NULL;
1655
1656 if (old_name) {
1657 int changed = 0;
1658 int stat_ret = 0;
1659 unsigned st_mode = 0;
1660
1661 if (!cached)
1662 stat_ret = lstat(old_name, &st);
1663 if (check_index) {
1664 int pos = cache_name_pos(old_name, strlen(old_name));
1665 if (pos < 0)
1666 return error("%s: does not exist in index",
1667 old_name);
1668 ce = active_cache[pos];
1669 if (stat_ret < 0) {
1670 struct checkout costate;
1671 if (errno != ENOENT)
1672 return error("%s: %s", old_name,
1673 strerror(errno));
1674 /* checkout */
1675 costate.base_dir = "";
1676 costate.base_dir_len = 0;
1677 costate.force = 0;
1678 costate.quiet = 0;
1679 costate.not_new = 0;
1680 costate.refresh_cache = 1;
1681 if (checkout_entry(ce,
1682 &costate,
1683 NULL) ||
1684 lstat(old_name, &st))
1685 return -1;
1686 }
1687 if (!cached)
1688 changed = ce_match_stat(ce, &st, 1);
1689 if (changed)
1690 return error("%s: does not match index",
1691 old_name);
1692 if (cached)
1693 st_mode = ntohl(ce->ce_mode);
1694 }
1695 else if (stat_ret < 0)
1696 return error("%s: %s", old_name, strerror(errno));
1697
1698 if (!cached)
1699 st_mode = ntohl(create_ce_mode(st.st_mode));
1700
1701 if (patch->is_new < 0)
1702 patch->is_new = 0;
1703 if (!patch->old_mode)
1704 patch->old_mode = st_mode;
1705 if ((st_mode ^ patch->old_mode) & S_IFMT)
1706 return error("%s: wrong type", old_name);
1707 if (st_mode != patch->old_mode)
1708 fprintf(stderr, "warning: %s has type %o, expected %o\n",
1709 old_name, st_mode, patch->old_mode);
1710 }
1711
1712 if (new_name && (patch->is_new | patch->is_rename | patch->is_copy)) {
1713 if (check_index && cache_name_pos(new_name, strlen(new_name)) >= 0)
1714 return error("%s: already exists in index", new_name);
1715 if (!cached) {
1716 if (!lstat(new_name, &st))
1717 return error("%s: already exists in working directory", new_name);
1718 if (errno != ENOENT)
1719 return error("%s: %s", new_name, strerror(errno));
1720 }
1721 if (!patch->new_mode) {
1722 if (patch->is_new)
1723 patch->new_mode = S_IFREG | 0644;
1724 else
1725 patch->new_mode = patch->old_mode;
1726 }
1727 }
1728
1729 if (new_name && old_name) {
1730 int same = !strcmp(old_name, new_name);
1731 if (!patch->new_mode)
1732 patch->new_mode = patch->old_mode;
1733 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
1734 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
1735 patch->new_mode, new_name, patch->old_mode,
1736 same ? "" : " of ", same ? "" : old_name);
1737 }
1738
1739 if (apply_data(patch, &st, ce) < 0)
1740 return error("%s: patch does not apply", name);
1741 return 0;
1742}
1743
1744static int check_patch_list(struct patch *patch)
1745{
1746 int error = 0;
1747
1748 for (;patch ; patch = patch->next)
1749 error |= check_patch(patch);
1750 return error;
1751}
1752
1753static inline int is_null_sha1(const unsigned char *sha1)
1754{
1755 return !memcmp(sha1, null_sha1, 20);
1756}
1757
1758static void show_index_list(struct patch *list)
1759{
1760 struct patch *patch;
1761
1762 /* Once we start supporting the reverse patch, it may be
1763 * worth showing the new sha1 prefix, but until then...
1764 */
1765 for (patch = list; patch; patch = patch->next) {
1766 const unsigned char *sha1_ptr;
1767 unsigned char sha1[20];
1768 const char *name;
1769
1770 name = patch->old_name ? patch->old_name : patch->new_name;
1771 if (patch->is_new)
1772 sha1_ptr = null_sha1;
1773 else if (get_sha1(patch->old_sha1_prefix, sha1))
1774 die("sha1 information is lacking or useless (%s).",
1775 name);
1776 else
1777 sha1_ptr = sha1;
1778
1779 printf("%06o %s ",patch->old_mode, sha1_to_hex(sha1_ptr));
1780 if (line_termination && quote_c_style(name, NULL, NULL, 0))
1781 quote_c_style(name, NULL, stdout, 0);
1782 else
1783 fputs(name, stdout);
1784 putchar(line_termination);
1785 }
1786}
1787
1788static void stat_patch_list(struct patch *patch)
1789{
1790 int files, adds, dels;
1791
1792 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
1793 files++;
1794 adds += patch->lines_added;
1795 dels += patch->lines_deleted;
1796 show_stats(patch);
1797 }
1798
1799 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
1800}
1801
1802static void numstat_patch_list(struct patch *patch)
1803{
1804 for ( ; patch; patch = patch->next) {
1805 const char *name;
1806 name = patch->new_name ? patch->new_name : patch->old_name;
1807 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
1808 if (line_termination && quote_c_style(name, NULL, NULL, 0))
1809 quote_c_style(name, NULL, stdout, 0);
1810 else
1811 fputs(name, stdout);
1812 putchar('\n');
1813 }
1814}
1815
1816static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
1817{
1818 if (mode)
1819 printf(" %s mode %06o %s\n", newdelete, mode, name);
1820 else
1821 printf(" %s %s\n", newdelete, name);
1822}
1823
1824static void show_mode_change(struct patch *p, int show_name)
1825{
1826 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
1827 if (show_name)
1828 printf(" mode change %06o => %06o %s\n",
1829 p->old_mode, p->new_mode, p->new_name);
1830 else
1831 printf(" mode change %06o => %06o\n",
1832 p->old_mode, p->new_mode);
1833 }
1834}
1835
1836static void show_rename_copy(struct patch *p)
1837{
1838 const char *renamecopy = p->is_rename ? "rename" : "copy";
1839 const char *old, *new;
1840
1841 /* Find common prefix */
1842 old = p->old_name;
1843 new = p->new_name;
1844 while (1) {
1845 const char *slash_old, *slash_new;
1846 slash_old = strchr(old, '/');
1847 slash_new = strchr(new, '/');
1848 if (!slash_old ||
1849 !slash_new ||
1850 slash_old - old != slash_new - new ||
1851 memcmp(old, new, slash_new - new))
1852 break;
1853 old = slash_old + 1;
1854 new = slash_new + 1;
1855 }
1856 /* p->old_name thru old is the common prefix, and old and new
1857 * through the end of names are renames
1858 */
1859 if (old != p->old_name)
1860 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
1861 (int)(old - p->old_name), p->old_name,
1862 old, new, p->score);
1863 else
1864 printf(" %s %s => %s (%d%%)\n", renamecopy,
1865 p->old_name, p->new_name, p->score);
1866 show_mode_change(p, 0);
1867}
1868
1869static void summary_patch_list(struct patch *patch)
1870{
1871 struct patch *p;
1872
1873 for (p = patch; p; p = p->next) {
1874 if (p->is_new)
1875 show_file_mode_name("create", p->new_mode, p->new_name);
1876 else if (p->is_delete)
1877 show_file_mode_name("delete", p->old_mode, p->old_name);
1878 else {
1879 if (p->is_rename || p->is_copy)
1880 show_rename_copy(p);
1881 else {
1882 if (p->score) {
1883 printf(" rewrite %s (%d%%)\n",
1884 p->new_name, p->score);
1885 show_mode_change(p, 0);
1886 }
1887 else
1888 show_mode_change(p, 1);
1889 }
1890 }
1891 }
1892}
1893
1894static void patch_stats(struct patch *patch)
1895{
1896 int lines = patch->lines_added + patch->lines_deleted;
1897
1898 if (lines > max_change)
1899 max_change = lines;
1900 if (patch->old_name) {
1901 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
1902 if (!len)
1903 len = strlen(patch->old_name);
1904 if (len > max_len)
1905 max_len = len;
1906 }
1907 if (patch->new_name) {
1908 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
1909 if (!len)
1910 len = strlen(patch->new_name);
1911 if (len > max_len)
1912 max_len = len;
1913 }
1914}
1915
1916static void remove_file(struct patch *patch)
1917{
1918 if (write_index) {
1919 if (remove_file_from_cache(patch->old_name) < 0)
1920 die("unable to remove %s from index", patch->old_name);
1921 }
1922 if (!cached)
1923 unlink(patch->old_name);
1924}
1925
1926static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
1927{
1928 struct stat st;
1929 struct cache_entry *ce;
1930 int namelen = strlen(path);
1931 unsigned ce_size = cache_entry_size(namelen);
1932
1933 if (!write_index)
1934 return;
1935
1936 ce = xcalloc(1, ce_size);
1937 memcpy(ce->name, path, namelen);
1938 ce->ce_mode = create_ce_mode(mode);
1939 ce->ce_flags = htons(namelen);
1940 if (!cached) {
1941 if (lstat(path, &st) < 0)
1942 die("unable to stat newly created file %s", path);
1943 fill_stat_cache_info(ce, &st);
1944 }
1945 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
1946 die("unable to create backing store for newly created file %s", path);
1947 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
1948 die("unable to add cache entry for %s", path);
1949}
1950
1951static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
1952{
1953 int fd;
1954
1955 if (S_ISLNK(mode))
1956 return symlink(buf, path);
1957 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
1958 if (fd < 0)
1959 return -1;
1960 while (size) {
1961 int written = xwrite(fd, buf, size);
1962 if (written < 0)
1963 die("writing file %s: %s", path, strerror(errno));
1964 if (!written)
1965 die("out of space writing file %s", path);
1966 buf += written;
1967 size -= written;
1968 }
1969 if (close(fd) < 0)
1970 die("closing file %s: %s", path, strerror(errno));
1971 return 0;
1972}
1973
1974/*
1975 * We optimistically assume that the directories exist,
1976 * which is true 99% of the time anyway. If they don't,
1977 * we create them and try again.
1978 */
1979static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
1980{
1981 if (cached)
1982 return;
1983 if (!try_create_file(path, mode, buf, size))
1984 return;
1985
1986 if (errno == ENOENT) {
1987 if (safe_create_leading_directories(path))
1988 return;
1989 if (!try_create_file(path, mode, buf, size))
1990 return;
1991 }
1992
1993 if (errno == EEXIST) {
1994 unsigned int nr = getpid();
1995
1996 for (;;) {
1997 const char *newpath;
1998 newpath = mkpath("%s~%u", path, nr);
1999 if (!try_create_file(newpath, mode, buf, size)) {
2000 if (!rename(newpath, path))
2001 return;
2002 unlink(newpath);
2003 break;
2004 }
2005 if (errno != EEXIST)
2006 break;
2007 ++nr;
2008 }
2009 }
2010 die("unable to write file %s mode %o", path, mode);
2011}
2012
2013static void create_file(struct patch *patch)
2014{
2015 char *path = patch->new_name;
2016 unsigned mode = patch->new_mode;
2017 unsigned long size = patch->resultsize;
2018 char *buf = patch->result;
2019
2020 if (!mode)
2021 mode = S_IFREG | 0644;
2022 create_one_file(path, mode, buf, size);
2023 add_index_file(path, mode, buf, size);
2024}
2025
2026static void write_out_one_result(struct patch *patch)
2027{
2028 if (patch->is_delete > 0) {
2029 remove_file(patch);
2030 return;
2031 }
2032 if (patch->is_new > 0 || patch->is_copy) {
2033 create_file(patch);
2034 return;
2035 }
2036 /*
2037 * Rename or modification boils down to the same
2038 * thing: remove the old, write the new
2039 */
2040 remove_file(patch);
2041 create_file(patch);
2042}
2043
2044static void write_out_results(struct patch *list, int skipped_patch)
2045{
2046 if (!list && !skipped_patch)
2047 die("No changes");
2048
2049 while (list) {
2050 write_out_one_result(list);
2051 list = list->next;
2052 }
2053}
2054
2055static struct cache_file cache_file;
2056
2057static struct excludes {
2058 struct excludes *next;
2059 const char *path;
2060} *excludes;
2061
2062static int use_patch(struct patch *p)
2063{
2064 const char *pathname = p->new_name ? p->new_name : p->old_name;
2065 struct excludes *x = excludes;
2066 while (x) {
2067 if (fnmatch(x->path, pathname, 0) == 0)
2068 return 0;
2069 x = x->next;
2070 }
2071 if (0 < prefix_length) {
2072 int pathlen = strlen(pathname);
2073 if (pathlen <= prefix_length ||
2074 memcmp(prefix, pathname, prefix_length))
2075 return 0;
2076 }
2077 return 1;
2078}
2079
2080static int apply_patch(int fd, const char *filename)
2081{
2082 unsigned long offset, size;
2083 char *buffer = read_patch_file(fd, &size);
2084 struct patch *list = NULL, **listp = &list;
2085 int skipped_patch = 0;
2086
2087 patch_input_file = filename;
2088 if (!buffer)
2089 return -1;
2090 offset = 0;
2091 while (size > 0) {
2092 struct patch *patch;
2093 int nr;
2094
2095 patch = xcalloc(1, sizeof(*patch));
2096 nr = parse_chunk(buffer + offset, size, patch);
2097 if (nr < 0)
2098 break;
2099 if (use_patch(patch)) {
2100 patch_stats(patch);
2101 *listp = patch;
2102 listp = &patch->next;
2103 } else {
2104 /* perhaps free it a bit better? */
2105 free(patch);
2106 skipped_patch++;
2107 }
2108 offset += nr;
2109 size -= nr;
2110 }
2111
2112 if (whitespace_error && (new_whitespace == error_on_whitespace))
2113 apply = 0;
2114
2115 write_index = check_index && apply;
2116 if (write_index && newfd < 0)
2117 newfd = hold_index_file_for_update(&cache_file, get_index_file());
2118 if (check_index) {
2119 if (read_cache() < 0)
2120 die("unable to read index file");
2121 }
2122
2123 if ((check || apply) && check_patch_list(list) < 0)
2124 exit(1);
2125
2126 if (apply)
2127 write_out_results(list, skipped_patch);
2128
2129 if (show_index_info)
2130 show_index_list(list);
2131
2132 if (diffstat)
2133 stat_patch_list(list);
2134
2135 if (numstat)
2136 numstat_patch_list(list);
2137
2138 if (summary)
2139 summary_patch_list(list);
2140
2141 free(buffer);
2142 return 0;
2143}
2144
2145static int git_apply_config(const char *var, const char *value)
2146{
2147 if (!strcmp(var, "apply.whitespace")) {
2148 apply_default_whitespace = strdup(value);
2149 return 0;
2150 }
2151 return git_default_config(var, value);
2152}
2153
2154
2155int cmd_apply(int argc, const char **argv, char **envp)
2156{
2157 int i;
2158 int read_stdin = 1;
2159 const char *whitespace_option = NULL;
2160
2161 for (i = 1; i < argc; i++) {
2162 const char *arg = argv[i];
2163 char *end;
2164 int fd;
2165
2166 if (!strcmp(arg, "-")) {
2167 apply_patch(0, "<stdin>");
2168 read_stdin = 0;
2169 continue;
2170 }
2171 if (!strncmp(arg, "--exclude=", 10)) {
2172 struct excludes *x = xmalloc(sizeof(*x));
2173 x->path = arg + 10;
2174 x->next = excludes;
2175 excludes = x;
2176 continue;
2177 }
2178 if (!strncmp(arg, "-p", 2)) {
2179 p_value = atoi(arg + 2);
2180 continue;
2181 }
2182 if (!strcmp(arg, "--no-add")) {
2183 no_add = 1;
2184 continue;
2185 }
2186 if (!strcmp(arg, "--stat")) {
2187 apply = 0;
2188 diffstat = 1;
2189 continue;
2190 }
2191 if (!strcmp(arg, "--allow-binary-replacement") ||
2192 !strcmp(arg, "--binary")) {
2193 allow_binary_replacement = 1;
2194 continue;
2195 }
2196 if (!strcmp(arg, "--numstat")) {
2197 apply = 0;
2198 numstat = 1;
2199 continue;
2200 }
2201 if (!strcmp(arg, "--summary")) {
2202 apply = 0;
2203 summary = 1;
2204 continue;
2205 }
2206 if (!strcmp(arg, "--check")) {
2207 apply = 0;
2208 check = 1;
2209 continue;
2210 }
2211 if (!strcmp(arg, "--index")) {
2212 check_index = 1;
2213 continue;
2214 }
2215 if (!strcmp(arg, "--cached")) {
2216 check_index = 1;
2217 cached = 1;
2218 continue;
2219 }
2220 if (!strcmp(arg, "--apply")) {
2221 apply = 1;
2222 continue;
2223 }
2224 if (!strcmp(arg, "--index-info")) {
2225 apply = 0;
2226 show_index_info = 1;
2227 continue;
2228 }
2229 if (!strcmp(arg, "-z")) {
2230 line_termination = 0;
2231 continue;
2232 }
2233 if (!strncmp(arg, "-C", 2)) {
2234 p_context = strtoul(arg + 2, &end, 0);
2235 if (*end != '\0')
2236 die("unrecognized context count '%s'", arg + 2);
2237 continue;
2238 }
2239 if (!strncmp(arg, "--whitespace=", 13)) {
2240 whitespace_option = arg + 13;
2241 parse_whitespace_option(arg + 13);
2242 continue;
2243 }
2244
2245 if (check_index && prefix_length < 0) {
2246 prefix = setup_git_directory();
2247 prefix_length = prefix ? strlen(prefix) : 0;
2248 git_config(git_apply_config);
2249 if (!whitespace_option && apply_default_whitespace)
2250 parse_whitespace_option(apply_default_whitespace);
2251 }
2252 if (0 < prefix_length)
2253 arg = prefix_filename(prefix, prefix_length, arg);
2254
2255 fd = open(arg, O_RDONLY);
2256 if (fd < 0)
2257 usage(apply_usage);
2258 read_stdin = 0;
2259 set_default_whitespace_mode(whitespace_option);
2260 apply_patch(fd, arg);
2261 close(fd);
2262 }
2263 set_default_whitespace_mode(whitespace_option);
2264 if (read_stdin)
2265 apply_patch(0, "<stdin>");
2266 if (whitespace_error) {
2267 if (squelch_whitespace_errors &&
2268 squelch_whitespace_errors < whitespace_error) {
2269 int squelched =
2270 whitespace_error - squelch_whitespace_errors;
2271 fprintf(stderr, "warning: squelched %d whitespace error%s\n",
2272 squelched,
2273 squelched == 1 ? "" : "s");
2274 }
2275 if (new_whitespace == error_on_whitespace)
2276 die("%d line%s add%s trailing whitespaces.",
2277 whitespace_error,
2278 whitespace_error == 1 ? "" : "s",
2279 whitespace_error == 1 ? "s" : "");
2280 if (applied_after_stripping)
2281 fprintf(stderr, "warning: %d line%s applied after"
2282 " stripping trailing whitespaces.\n",
2283 applied_after_stripping,
2284 applied_after_stripping == 1 ? "" : "s");
2285 else if (whitespace_error)
2286 fprintf(stderr, "warning: %d line%s add%s trailing"
2287 " whitespaces.\n",
2288 whitespace_error,
2289 whitespace_error == 1 ? "" : "s",
2290 whitespace_error == 1 ? "s" : "");
2291 }
2292
2293 if (write_index) {
2294 if (write_cache(newfd, active_cache, active_nr) ||
2295 commit_index_file(&cache_file))
2296 die("Unable to write new cachefile");
2297 }
2298
2299 return 0;
2300}