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