git-apply: when validating default names, check the final EOLN too
[git/git-svn.git] / apply.c
blob2fab34754e0c176c5ae97e697b5969fdd361c4f8
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 * NOTE! It does all its work in the index file, and only cares about
9 * the files in the working directory if you tell it to "merge" the
10 * patch apply.
12 * Even when merging it always takes the source from the index, and
13 * uses the working tree as a "branch" for a 3-way merge.
15 #include <ctype.h>
17 #include "cache.h"
19 // We default to the merge behaviour, since that's what most people would
20 // expect
21 static int merge_patch = 1;
22 static int diffstat = 0;
23 static int check = 1;
24 static const char apply_usage[] = "git-apply <patch>";
27 * For "diff-stat" like behaviour, we keep track of the biggest change
28 * we've seen, and the longest filename. That allows us to do simple
29 * scaling.
31 static int max_change, max_len;
34 * Various "current state", notably line numbers and what
35 * file (and how) we're patching right now.. The "is_xxxx"
36 * things are flags, where -1 means "don't know yet".
38 static int linenr = 1;
40 struct fragment {
41 unsigned long oldpos, oldlines;
42 unsigned long newpos, newlines;
43 const char *patch;
44 int size;
45 struct fragment *next;
48 struct patch {
49 char *new_name, *old_name, *def_name;
50 unsigned int old_mode, new_mode;
51 int is_rename, is_copy, is_new, is_delete;
52 int lines_added, lines_deleted;
53 struct fragment *fragments;
54 struct patch *next;
57 #define CHUNKSIZE (8192)
58 #define SLOP (16)
60 static void *read_patch_file(int fd, unsigned long *sizep)
62 unsigned long size = 0, alloc = CHUNKSIZE;
63 void *buffer = xmalloc(alloc);
65 for (;;) {
66 int nr = alloc - size;
67 if (nr < 1024) {
68 alloc += CHUNKSIZE;
69 buffer = xrealloc(buffer, alloc);
70 nr = alloc - size;
72 nr = read(fd, buffer + size, nr);
73 if (!nr)
74 break;
75 if (nr < 0) {
76 if (errno == EAGAIN)
77 continue;
78 die("git-apply: read returned %s", strerror(errno));
80 size += nr;
82 *sizep = size;
85 * Make sure that we have some slop in the buffer
86 * so that we can do speculative "memcmp" etc, and
87 * see to it that it is NUL-filled.
89 if (alloc < size + SLOP)
90 buffer = xrealloc(buffer, size + SLOP);
91 memset(buffer + size, 0, SLOP);
92 return buffer;
95 static unsigned long linelen(char *buffer, unsigned long size)
97 unsigned long len = 0;
98 while (size--) {
99 len++;
100 if (*buffer++ == '\n')
101 break;
103 return len;
106 static int is_dev_null(const char *str)
108 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
111 #define TERM_EXIST 1
112 #define TERM_SPACE 2
113 #define TERM_TAB 4
115 static int name_terminate(const char *name, int namelen, int c, int terminate)
117 if (c == ' ' && !(terminate & TERM_SPACE))
118 return 0;
119 if (c == '\t' && !(terminate & TERM_TAB))
120 return 0;
123 * Do we want an existing name? Return false and
124 * continue if it's not there.
126 if (terminate & TERM_EXIST)
127 return cache_name_pos(name, namelen) >= 0;
129 return 1;
132 static char * find_name(const char *line, char *def, int p_value, int terminate)
134 int len;
135 const char *start = line;
136 char *name;
138 for (;;) {
139 char c = *line;
141 if (isspace(c)) {
142 if (c == '\n')
143 break;
144 if (name_terminate(start, line-start, c, terminate))
145 break;
147 line++;
148 if (c == '/' && !--p_value)
149 start = line;
151 if (!start)
152 return def;
153 len = line - start;
154 if (!len)
155 return def;
158 * Generally we prefer the shorter name, especially
159 * if the other one is just a variation of that with
160 * something else tacked on to the end (ie "file.orig"
161 * or "file~").
163 if (def) {
164 int deflen = strlen(def);
165 if (deflen < len && !strncmp(start, def, deflen))
166 return def;
169 name = xmalloc(len + 1);
170 memcpy(name, start, len);
171 name[len] = 0;
172 free(def);
173 return name;
177 * Get the name etc info from the --/+++ lines of a traditional patch header
179 * NOTE! This hardcodes "-p1" behaviour in filename detection.
181 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
182 * files, we can happily check the index for a match, but for creating a
183 * new file we should try to match whatever "patch" does. I have no idea.
185 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
187 int p_value = 1;
188 char *name;
190 first += 4; // skip "--- "
191 second += 4; // skip "+++ "
192 if (is_dev_null(first)) {
193 patch->is_new = 1;
194 patch->is_delete = 0;
195 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
196 patch->new_name = name;
197 } else if (is_dev_null(second)) {
198 patch->is_new = 0;
199 patch->is_delete = 1;
200 name = find_name(first, NULL, p_value, TERM_EXIST | TERM_SPACE | TERM_TAB);
201 patch->old_name = name;
202 } else {
203 name = find_name(first, NULL, p_value, TERM_EXIST | TERM_SPACE | TERM_TAB);
204 name = find_name(second, name, p_value, TERM_EXIST | TERM_SPACE | TERM_TAB);
205 patch->old_name = patch->new_name = name;
207 if (!name)
208 die("unable to find filename in patch at line %d", linenr);
211 static int gitdiff_hdrend(const char *line, struct patch *patch)
213 return -1;
217 * We're anal about diff header consistency, to make
218 * sure that we don't end up having strange ambiguous
219 * patches floating around.
221 * As a result, gitdiff_{old|new}name() will check
222 * their names against any previous information, just
223 * to make sure..
225 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
227 int len;
228 const char *name;
230 if (!orig_name && !isnull)
231 return find_name(line, NULL, 1, 0);
233 name = "/dev/null";
234 len = 9;
235 if (orig_name) {
236 name = orig_name;
237 len = strlen(name);
238 if (isnull)
239 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
242 if (*name == '/')
243 goto absolute_path;
245 for (;;) {
246 char c = *line++;
247 if (c == '\n')
248 break;
249 if (c != '/')
250 continue;
251 absolute_path:
252 if (memcmp(line, name, len) || line[len] != '\n')
253 break;
254 return orig_name;
256 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
257 return NULL;
260 static int gitdiff_oldname(const char *line, struct patch *patch)
262 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
263 return 0;
266 static int gitdiff_newname(const char *line, struct patch *patch)
268 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
269 return 0;
272 static int gitdiff_oldmode(const char *line, struct patch *patch)
274 patch->old_mode = strtoul(line, NULL, 8);
275 return 0;
278 static int gitdiff_newmode(const char *line, struct patch *patch)
280 patch->new_mode = strtoul(line, NULL, 8);
281 return 0;
284 static int gitdiff_delete(const char *line, struct patch *patch)
286 patch->is_delete = 1;
287 patch->old_name = patch->def_name;
288 return gitdiff_oldmode(line, patch);
291 static int gitdiff_newfile(const char *line, struct patch *patch)
293 patch->is_new = 1;
294 patch->new_name = patch->def_name;
295 return gitdiff_newmode(line, patch);
298 static int gitdiff_copysrc(const char *line, struct patch *patch)
300 patch->is_copy = 1;
301 patch->old_name = find_name(line, NULL, 0, 0);
302 return 0;
305 static int gitdiff_copydst(const char *line, struct patch *patch)
307 patch->is_copy = 1;
308 patch->new_name = find_name(line, NULL, 0, 0);
309 return 0;
312 static int gitdiff_renamesrc(const char *line, struct patch *patch)
314 patch->is_rename = 1;
315 patch->old_name = find_name(line, NULL, 0, 0);
316 return 0;
319 static int gitdiff_renamedst(const char *line, struct patch *patch)
321 patch->is_rename = 1;
322 patch->new_name = find_name(line, NULL, 0, 0);
323 return 0;
326 static int gitdiff_similarity(const char *line, struct patch *patch)
328 return 0;
332 * This is normal for a diff that doesn't change anything: we'll fall through
333 * into the next diff. Tell the parser to break out.
335 static int gitdiff_unrecognized(const char *line, struct patch *patch)
337 return -1;
340 static char *git_header_name(char *line)
342 int len;
343 char *name, *second;
346 * Find the first '/'
348 name = line;
349 for (;;) {
350 char c = *name++;
351 if (c == '\n')
352 return NULL;
353 if (c == '/')
354 break;
358 * We don't accept absolute paths (/dev/null) as possibly valid
360 if (name == line+1)
361 return NULL;
364 * Accept a name only if it shows up twice, exactly the same
365 * form.
367 for (len = 0 ; ; len++) {
368 char c = name[len];
370 switch (c) {
371 default:
372 continue;
373 case '\n':
374 break;
375 case '\t': case ' ':
376 second = name+len;
377 for (;;) {
378 char c = *second++;
379 if (c == '\n')
380 return NULL;
381 if (c == '/')
382 break;
384 if (second[len] == '\n' && !memcmp(name, second, len)) {
385 char *ret = xmalloc(len + 1);
386 memcpy(ret, name, len);
387 ret[len] = 0;
388 return ret;
392 return NULL;
395 /* Verify that we recognize the lines following a git header */
396 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
398 unsigned long offset;
400 /* A git diff has explicit new/delete information, so we don't guess */
401 patch->is_new = 0;
402 patch->is_delete = 0;
405 * Some things may not have the old name in the
406 * rest of the headers anywhere (pure mode changes,
407 * or removing or adding empty files), so we get
408 * the default name from the header.
410 patch->def_name = git_header_name(line + strlen("diff --git "));
412 line += len;
413 size -= len;
414 linenr++;
415 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
416 static const struct opentry {
417 const char *str;
418 int (*fn)(const char *, struct patch *);
419 } optable[] = {
420 { "@@ -", gitdiff_hdrend },
421 { "--- ", gitdiff_oldname },
422 { "+++ ", gitdiff_newname },
423 { "old mode ", gitdiff_oldmode },
424 { "new mode ", gitdiff_newmode },
425 { "deleted file mode ", gitdiff_delete },
426 { "new file mode ", gitdiff_newfile },
427 { "copy from ", gitdiff_copysrc },
428 { "copy to ", gitdiff_copydst },
429 { "rename from ", gitdiff_renamesrc },
430 { "rename to ", gitdiff_renamedst },
431 { "similarity index ", gitdiff_similarity },
432 { "", gitdiff_unrecognized },
434 int i;
436 len = linelen(line, size);
437 if (!len || line[len-1] != '\n')
438 break;
439 for (i = 0; i < sizeof(optable) / sizeof(optable[0]); i++) {
440 const struct opentry *p = optable + i;
441 int oplen = strlen(p->str);
442 if (len < oplen || memcmp(p->str, line, oplen))
443 continue;
444 if (p->fn(line + oplen, patch) < 0)
445 return offset;
446 break;
450 return offset;
453 static int parse_num(const char *line, unsigned long *p)
455 char *ptr;
457 if (!isdigit(*line))
458 return 0;
459 *p = strtoul(line, &ptr, 10);
460 return ptr - line;
463 static int parse_range(const char *line, int len, int offset, const char *expect,
464 unsigned long *p1, unsigned long *p2)
466 int digits, ex;
468 if (offset < 0 || offset >= len)
469 return -1;
470 line += offset;
471 len -= offset;
473 digits = parse_num(line, p1);
474 if (!digits)
475 return -1;
477 offset += digits;
478 line += digits;
479 len -= digits;
481 *p2 = *p1;
482 if (*line == ',') {
483 digits = parse_num(line+1, p2);
484 if (!digits)
485 return -1;
487 offset += digits+1;
488 line += digits+1;
489 len -= digits+1;
492 ex = strlen(expect);
493 if (ex > len)
494 return -1;
495 if (memcmp(line, expect, ex))
496 return -1;
498 return offset + ex;
502 * Parse a unified diff fragment header of the
503 * form "@@ -a,b +c,d @@"
505 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
507 int offset;
509 if (!len || line[len-1] != '\n')
510 return -1;
512 /* Figure out the number of lines in a fragment */
513 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
514 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
516 return offset;
519 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
521 unsigned long offset, len;
523 patch->is_rename = patch->is_copy = 0;
524 patch->is_new = patch->is_delete = -1;
525 patch->old_mode = patch->new_mode = 0;
526 patch->old_name = patch->new_name = NULL;
527 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
528 unsigned long nextlen;
530 len = linelen(line, size);
531 if (!len)
532 break;
534 /* Testing this early allows us to take a few shortcuts.. */
535 if (len < 6)
536 continue;
539 * Make sure we don't find any unconnected patch fragmants.
540 * That's a sign that we didn't find a header, and that a
541 * patch has become corrupted/broken up.
543 if (!memcmp("@@ -", line, 4)) {
544 struct fragment dummy;
545 if (parse_fragment_header(line, len, &dummy) < 0)
546 continue;
547 error("patch fragment without header at line %d: %.*s", linenr, len-1, line);
550 if (size < len + 6)
551 break;
554 * Git patch? It might not have a real patch, just a rename
555 * or mode change, so we handle that specially
557 if (!memcmp("diff --git ", line, 11)) {
558 int git_hdr_len = parse_git_header(line, len, size, patch);
559 if (git_hdr_len < 0)
560 continue;
561 if (!patch->old_name && !patch->new_name)
562 die("git diff header lacks filename information");
563 *hdrsize = git_hdr_len;
564 return offset;
567 /** --- followed by +++ ? */
568 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
569 continue;
572 * We only accept unified patches, so we want it to
573 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
574 * minimum
576 nextlen = linelen(line + len, size - len);
577 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
578 continue;
580 /* Ok, we'll consider it a patch */
581 parse_traditional_patch(line, line+len, patch);
582 *hdrsize = len + nextlen;
583 linenr += 2;
584 return offset;
586 return -1;
590 * Parse a unified diff. Note that this really needs
591 * to parse each fragment separately, since the only
592 * way to know the difference between a "---" that is
593 * part of a patch, and a "---" that starts the next
594 * patch is to look at the line counts..
596 static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
598 int added, deleted;
599 int len = linelen(line, size), offset;
600 unsigned long pos[4], oldlines, newlines;
602 offset = parse_fragment_header(line, len, fragment);
603 if (offset < 0)
604 return -1;
605 oldlines = fragment->oldlines;
606 newlines = fragment->newlines;
608 if (patch->is_new < 0 && (pos[0] || oldlines))
609 patch->is_new = 0;
610 if (patch->is_delete < 0 && (pos[1] || newlines))
611 patch->is_delete = 0;
613 /* Parse the thing.. */
614 line += len;
615 size -= len;
616 linenr++;
617 added = deleted = 0;
618 for (offset = len; size > 0; offset += len, size -= len, line += len, linenr++) {
619 if (!oldlines && !newlines)
620 break;
621 len = linelen(line, size);
622 if (!len || line[len-1] != '\n')
623 return -1;
624 switch (*line) {
625 default:
626 return -1;
627 case ' ':
628 oldlines--;
629 newlines--;
630 break;
631 case '-':
632 deleted++;
633 oldlines--;
634 break;
635 case '+':
636 added++;
637 newlines--;
638 break;
639 /* We allow "\ No newline at end of file" */
640 case '\\':
641 break;
644 patch->lines_added += added;
645 patch->lines_deleted += deleted;
646 return offset;
649 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
651 unsigned long offset = 0;
652 struct fragment **fragp = &patch->fragments;
654 while (size > 4 && !memcmp(line, "@@ -", 4)) {
655 struct fragment *fragment;
656 int len;
658 fragment = xmalloc(sizeof(*fragment));
659 memset(fragment, 0, sizeof(*fragment));
660 len = parse_fragment(line, size, patch, fragment);
661 if (len <= 0)
662 die("corrupt patch at line %d", linenr);
664 fragment->patch = line;
665 fragment->size = len;
667 *fragp = fragment;
668 fragp = &fragment->next;
670 offset += len;
671 line += len;
672 size -= len;
674 return offset;
677 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
679 int hdrsize, patchsize;
680 int offset = find_header(buffer, size, &hdrsize, patch);
682 if (offset < 0)
683 return offset;
685 patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
687 return offset + hdrsize + patchsize;
690 const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
691 const char minuses[]= "----------------------------------------------------------------------";
693 static void show_stats(struct patch *patch)
695 char *name = patch->old_name;
696 int len, max, add, del;
698 if (!name)
699 name = patch->new_name;
702 * "scale" the filename
704 len = strlen(name);
705 max = max_len;
706 if (max > 50)
707 max = 50;
708 if (len > max)
709 name += len - max;
710 len = max;
713 * scale the add/delete
715 max = max_change;
716 if (max + len > 70)
717 max = 70 - len;
719 add = (patch->lines_added * max + max_change/2) / max_change;
720 del = (patch->lines_deleted * max + max_change/2) / max_change;
721 printf(" %-*s |%5d %.*s%.*s\n",
722 len, name, patch->lines_added + patch->lines_deleted,
723 add, pluses, del, minuses);
726 static void check_patch(struct patch *patch)
728 const char *old_name = patch->old_name;
729 const char *new_name = patch->new_name;
731 if (old_name) {
732 if (cache_name_pos(old_name, strlen(old_name)) < 0)
733 die("file %s does not exist", old_name);
734 if (patch->is_new < 0)
735 patch->is_new = 0;
737 if (new_name && (patch->is_new | patch->is_rename | patch->is_copy)) {
738 if (cache_name_pos(new_name, strlen(new_name)) >= 0)
739 die("file %s already exists", new_name);
743 static void apply_patch_list(struct patch *patch)
745 int files, adds, dels;
747 files = adds = dels = 0;
748 if (!patch)
749 die("no patch found");
750 do {
751 if (check)
752 check_patch(patch);
754 if (diffstat) {
755 files++;
756 adds += patch->lines_added;
757 dels += patch->lines_deleted;
758 show_stats(patch);
760 } while ((patch = patch->next) != NULL);
762 if (diffstat)
763 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
766 static void patch_stats(struct patch *patch)
768 int lines = patch->lines_added + patch->lines_deleted;
770 if (lines > max_change)
771 max_change = lines;
772 if (patch->old_name) {
773 int len = strlen(patch->old_name);
774 if (len > max_len)
775 max_len = len;
777 if (patch->new_name) {
778 int len = strlen(patch->new_name);
779 if (len > max_len)
780 max_len = len;
784 static int apply_patch(int fd)
786 unsigned long offset, size;
787 char *buffer = read_patch_file(fd, &size);
788 struct patch *list = NULL, **listp = &list;
790 if (!buffer)
791 return -1;
792 offset = 0;
793 while (size > 0) {
794 struct patch *patch;
795 int nr;
797 patch = xmalloc(sizeof(*patch));
798 memset(patch, 0, sizeof(*patch));
799 nr = parse_chunk(buffer + offset, size, patch);
800 if (nr < 0)
801 break;
802 patch_stats(patch);
803 *listp = patch;
804 listp = &patch->next;
805 offset += nr;
806 size -= nr;
809 apply_patch_list(list);
811 free(buffer);
812 return 0;
815 int main(int argc, char **argv)
817 int i;
818 int read_stdin = 1;
820 if (read_cache() < 0)
821 die("unable to read index file");
823 for (i = 1; i < argc; i++) {
824 const char *arg = argv[i];
825 int fd;
827 if (!strcmp(arg, "-")) {
828 apply_patch(0);
829 read_stdin = 0;
830 continue;
832 if (!strcmp(arg, "--no-merge")) {
833 merge_patch = 0;
834 continue;
836 if (!strcmp(arg, "--stat")) {
837 check = 0;
838 diffstat = 1;
839 continue;
841 fd = open(arg, O_RDONLY);
842 if (fd < 0)
843 usage(apply_usage);
844 read_stdin = 0;
845 apply_patch(fd);
846 close(fd);
848 if (read_stdin)
849 apply_patch(0);
850 return 0;