2 * Copyright (C) 2005 Junio C Hamano
11 static const char *diff_opts
= "-pu";
12 static unsigned char null_sha1
[20] = { 0, };
14 static int reverse_diff
;
16 static const char *external_diff(void)
18 static const char *external_diff_cmd
= NULL
;
19 static int done_preparing
= 0;
22 return external_diff_cmd
;
25 * Default values above are meant to match the
26 * Linux kernel development style. Examples of
27 * alternative styles you can specify via environment
32 if (gitenv("GIT_EXTERNAL_DIFF"))
33 external_diff_cmd
= gitenv("GIT_EXTERNAL_DIFF");
35 /* In case external diff fails... */
36 diff_opts
= gitenv("GIT_DIFF_OPTS") ? : diff_opts
;
39 return external_diff_cmd
;
42 /* Help to copy the thing properly quoted for the shell safety.
43 * any single quote is replaced with '\'', and the caller is
44 * expected to enclose the result within a single quote pair.
47 * original sq_expand result
48 * name ==> name ==> 'name'
49 * a b ==> a b ==> 'a b'
50 * a'b ==> a'\''b ==> 'a'\''b'
52 static char *sq_expand(const char *src
)
54 static char *buf
= NULL
;
59 /* count bytes needed to store the quoted string. */
60 for (cnt
= 1, cp
= src
; *cp
; cnt
++, cp
++)
66 while ((c
= *src
++)) {
70 bp
= strcpy(bp
, "'\\''");
78 static struct diff_tempfile
{
79 const char *name
; /* filename external diff should read from */
85 static void builtin_diff(const char *name_a
,
87 struct diff_tempfile
*temp
,
90 int i
, next_at
, cmd_size
;
91 const char *diff_cmd
= "diff -L'%s%s' -L'%s%s'";
92 const char *diff_arg
= "'%s' '%s'||:"; /* "||:" is to return 0 */
93 const char *input_name_sq
[2];
96 const char *name_sq
[2];
99 name_sq
[0] = sq_expand(name_a
);
100 name_sq
[1] = sq_expand(name_b
);
102 /* diff_cmd and diff_arg have 6 %s in total which makes
103 * the sum of these strings 12 bytes larger than required.
104 * we use 2 spaces around diff-opts, and we need to count
105 * terminating NUL, so we subtract 9 here.
107 cmd_size
= (strlen(diff_cmd
) + strlen(diff_opts
) +
108 strlen(diff_arg
) - 9);
109 for (i
= 0; i
< 2; i
++) {
110 input_name_sq
[i
] = sq_expand(temp
[i
].name
);
111 if (!strcmp(temp
[i
].name
, "/dev/null")) {
112 path0
[i
] = "/dev/null";
115 path0
[i
] = i
? "b/" : "a/";
116 path1
[i
] = name_sq
[i
];
118 cmd_size
+= (strlen(path0
[i
]) + strlen(path1
[i
]) +
119 strlen(input_name_sq
[i
]));
122 cmd
= xmalloc(cmd_size
);
125 next_at
+= snprintf(cmd
+next_at
, cmd_size
-next_at
,
127 path0
[0], path1
[0], path0
[1], path1
[1]);
128 next_at
+= snprintf(cmd
+next_at
, cmd_size
-next_at
,
130 next_at
+= snprintf(cmd
+next_at
, cmd_size
-next_at
,
131 diff_arg
, input_name_sq
[0], input_name_sq
[1]);
133 printf("diff --git a/%s b/%s\n", name_a
, name_b
);
135 printf("new file mode %s\n", temp
[1].mode
);
136 else if (!path1
[1][0])
137 printf("deleted file mode %s\n", temp
[0].mode
);
139 if (strcmp(temp
[0].mode
, temp
[1].mode
)) {
140 printf("old mode %s\n", temp
[0].mode
);
141 printf("new mode %s\n", temp
[1].mode
);
143 if (xfrm_msg
&& xfrm_msg
[0])
144 fputs(xfrm_msg
, stdout
);
146 if (strncmp(temp
[0].mode
, temp
[1].mode
, 3))
147 /* we do not run diff between different kind
153 execlp("/bin/sh","sh", "-c", cmd
, NULL
);
156 struct diff_filespec
*alloc_filespec(const char *path
)
158 int namelen
= strlen(path
);
159 struct diff_filespec
*spec
= xmalloc(sizeof(*spec
) + namelen
+ 1);
160 spec
->path
= (char *)(spec
+ 1);
161 strcpy(spec
->path
, path
);
162 spec
->should_free
= spec
->should_munmap
= 0;
163 spec
->xfrm_flags
= 0;
167 memset(spec
->sha1
, 0, 20);
171 void fill_filespec(struct diff_filespec
*spec
, const unsigned char *sha1
,
174 if (mode
) { /* just playing defensive */
176 memcpy(spec
->sha1
, sha1
, 20);
177 spec
->sha1_valid
= !!memcmp(sha1
, null_sha1
, 20);
182 * Given a name and sha1 pair, if the dircache tells us the file in
183 * the work tree has that object contents, return true, so that
184 * prepare_temp_file() does not have to inflate and extract.
186 static int work_tree_matches(const char *name
, const unsigned char *sha1
)
188 struct cache_entry
*ce
;
192 /* We do not read the cache ourselves here, because the
193 * benchmark with my previous version that always reads cache
194 * shows that it makes things worse for diff-tree comparing
195 * two linux-2.6 kernel trees in an already checked out work
196 * tree. This is because most diff-tree comparisons deal with
197 * only a small number of files, while reading the cache is
198 * expensive for a large project, and its cost outweighs the
199 * savings we get by not inflating the object to a temporary
200 * file. Practically, this code only helps when we are used
201 * by diff-cache --cached, which does read the cache before
208 pos
= cache_name_pos(name
, len
);
211 ce
= active_cache
[pos
];
212 if ((lstat(name
, &st
) < 0) ||
213 !S_ISREG(st
.st_mode
) || /* careful! */
214 ce_match_stat(ce
, &st
) ||
215 memcmp(sha1
, ce
->sha1
, 20))
217 /* we return 1 only when we can stat, it is a regular file,
218 * stat information matches, and sha1 recorded in the cache
219 * matches. I.e. we know the file in the work tree really is
220 * the same as the <name, sha1> pair.
226 * While doing rename detection and pickaxe operation, we may need to
227 * grab the data for the blob (or file) for our own in-core comparison.
228 * diff_filespec has data and size fields for this purpose.
230 int diff_populate_filespec(struct diff_filespec
*s
)
233 if (!DIFF_FILE_VALID(s
))
234 die("internal error: asking to populate invalid file.");
235 if (S_ISDIR(s
->mode
))
240 if (!s
->sha1_valid
||
241 work_tree_matches(s
->path
, s
->sha1
)) {
244 if (lstat(s
->path
, &st
) < 0) {
245 if (errno
== ENOENT
) {
254 s
->size
= st
.st_size
;
257 if (S_ISLNK(st
.st_mode
)) {
259 s
->data
= xmalloc(s
->size
);
261 ret
= readlink(s
->path
, s
->data
, s
->size
);
268 fd
= open(s
->path
, O_RDONLY
);
271 s
->data
= mmap(NULL
, s
->size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
272 s
->should_munmap
= 1;
277 s
->data
= read_sha1_file(s
->sha1
, type
, &s
->size
);
283 void diff_free_filespec_data(struct diff_filespec
*s
)
287 else if (s
->should_munmap
)
288 munmap(s
->data
, s
->size
);
289 s
->should_free
= s
->should_munmap
= 0;
293 static void prep_temp_blob(struct diff_tempfile
*temp
,
301 strcpy(temp
->tmp_path
, ".diff_XXXXXX");
302 fd
= mkstemp(temp
->tmp_path
);
304 die("unable to create temp-file");
305 if (write(fd
, blob
, size
) != size
)
306 die("unable to write temp-file");
308 temp
->name
= temp
->tmp_path
;
309 strcpy(temp
->hex
, sha1_to_hex(sha1
));
311 sprintf(temp
->mode
, "%06o", mode
);
314 static void prepare_temp_file(const char *name
,
315 struct diff_tempfile
*temp
,
316 struct diff_filespec
*one
)
318 if (!DIFF_FILE_VALID(one
)) {
320 /* A '-' entry produces this for file-2, and
321 * a '+' entry produces this for file-1.
323 temp
->name
= "/dev/null";
324 strcpy(temp
->hex
, ".");
325 strcpy(temp
->mode
, ".");
329 if (!one
->sha1_valid
||
330 work_tree_matches(name
, one
->sha1
)) {
332 if (lstat(name
, &st
) < 0) {
334 goto not_a_valid_file
;
335 die("stat(%s): %s", name
, strerror(errno
));
337 if (S_ISLNK(st
.st_mode
)) {
339 char *buf
, buf_
[1024];
340 buf
= ((sizeof(buf_
) < st
.st_size
) ?
341 xmalloc(st
.st_size
) : buf_
);
342 ret
= readlink(name
, buf
, st
.st_size
);
344 die("readlink(%s)", name
);
345 prep_temp_blob(temp
, buf
, st
.st_size
,
347 one
->sha1
: null_sha1
),
349 one
->mode
: S_IFLNK
));
352 /* we can borrow from the file in the work tree */
354 if (!one
->sha1_valid
)
355 strcpy(temp
->hex
, sha1_to_hex(null_sha1
));
357 strcpy(temp
->hex
, sha1_to_hex(one
->sha1
));
358 sprintf(temp
->mode
, "%06o",
359 S_IFREG
|ce_permissions(st
.st_mode
));
364 if (diff_populate_filespec(one
))
365 die("cannot read data blob for %s", one
->path
);
366 prep_temp_blob(temp
, one
->data
, one
->size
,
367 one
->sha1
, one
->mode
);
371 static void remove_tempfile(void)
375 for (i
= 0; i
< 2; i
++)
376 if (diff_temp
[i
].name
== diff_temp
[i
].tmp_path
) {
377 unlink(diff_temp
[i
].name
);
378 diff_temp
[i
].name
= NULL
;
382 static void remove_tempfile_on_signal(int signo
)
387 /* An external diff command takes:
389 * diff-cmd name infile1 infile1-sha1 infile1-mode \
390 * infile2 infile2-sha1 infile2-mode [ rename-to ]
393 static void run_external_diff(const char *name
,
395 struct diff_filespec
*one
,
396 struct diff_filespec
*two
,
397 const char *xfrm_msg
)
399 struct diff_tempfile
*temp
= diff_temp
;
402 static int atexit_asked
= 0;
405 prepare_temp_file(name
, &temp
[0], one
);
406 prepare_temp_file(other
? : name
, &temp
[1], two
);
407 if (! atexit_asked
&&
408 (temp
[0].name
== temp
[0].tmp_path
||
409 temp
[1].name
== temp
[1].tmp_path
)) {
411 atexit(remove_tempfile
);
413 signal(SIGINT
, remove_tempfile_on_signal
);
419 die("unable to fork");
421 const char *pgm
= external_diff();
424 const char *exec_arg
[10];
425 const char **arg
= &exec_arg
[0];
428 *arg
++ = temp
[0].name
;
429 *arg
++ = temp
[0].hex
;
430 *arg
++ = temp
[0].mode
;
431 *arg
++ = temp
[1].name
;
432 *arg
++ = temp
[1].hex
;
433 *arg
++ = temp
[1].mode
;
439 execvp(pgm
, (char *const*) exec_arg
);
442 execlp(pgm
, pgm
, name
, NULL
);
445 * otherwise we use the built-in one.
448 builtin_diff(name
, other
? : name
, temp
, xfrm_msg
);
450 printf("* Unmerged path %s\n", name
);
453 if (waitpid(pid
, &status
, 0) < 0 ||
454 !WIFEXITED(status
) || WEXITSTATUS(status
)) {
455 /* Earlier we did not check the exit status because
456 * diff exits non-zero if files are different, and
457 * we are not interested in knowing that. It was a
458 * mistake which made it harder to quit a diff-*
459 * session that uses the git-apply-patch-script as
460 * the GIT_EXTERNAL_DIFF. A custom GIT_EXTERNAL_DIFF
461 * should also exit non-zero only when it wants to
462 * abort the entire diff-* session.
465 fprintf(stderr
, "external diff died, stopping at %s.\n", name
);
471 void diff_setup(int reverse_diff_
)
473 reverse_diff
= reverse_diff_
;
476 struct diff_queue_struct diff_queued_diff
;
478 void diff_q(struct diff_queue_struct
*queue
, struct diff_filepair
*dp
)
480 if (queue
->alloc
<= queue
->nr
) {
481 queue
->alloc
= alloc_nr(queue
->alloc
);
482 queue
->queue
= xrealloc(queue
->queue
,
483 sizeof(dp
) * queue
->alloc
);
485 queue
->queue
[queue
->nr
++] = dp
;
488 struct diff_filepair
*diff_queue(struct diff_queue_struct
*queue
,
489 struct diff_filespec
*one
,
490 struct diff_filespec
*two
)
492 struct diff_filepair
*dp
= xmalloc(sizeof(*dp
));
500 static void diff_flush_raw(struct diff_filepair
*p
,
501 int line_termination
,
502 int inter_name_termination
)
507 if (line_termination
) {
508 const char *err
= "path %s cannot be expressed without -z";
509 if (strchr(p
->one
->path
, line_termination
) ||
510 strchr(p
->one
->path
, inter_name_termination
))
511 die(err
, p
->one
->path
);
512 if (strchr(p
->two
->path
, line_termination
) ||
513 strchr(p
->two
->path
, inter_name_termination
))
514 die(err
, p
->two
->path
);
520 sprintf(status
, "%c%03d", p
->status
,
521 (int)(0.5 + p
->score
* 100.0/MAX_SCORE
));
525 status
[0] = p
->status
;
529 printf(":%06o %06o %s ",
530 p
->one
->mode
, p
->two
->mode
, sha1_to_hex(p
->one
->sha1
));
532 sha1_to_hex(p
->two
->sha1
),
534 inter_name_termination
,
537 printf("%c%s", inter_name_termination
, p
->two
->path
);
538 putchar(line_termination
);
541 int diff_unmodified_pair(struct diff_filepair
*p
)
543 /* This function is written stricter than necessary to support
544 * the currently implemented transformers, but the idea is to
545 * let transformers to produce diff_filepairs any way they want,
546 * and filter and clean them up here before producing the output.
548 struct diff_filespec
*one
, *two
;
550 if (DIFF_PAIR_UNMERGED(p
))
551 return 0; /* unmerged is interesting */
556 /* deletion, addition, mode change and renames are all interesting. */
557 if (DIFF_FILE_VALID(one
) != DIFF_FILE_VALID(two
) ||
558 (one
->mode
!= two
->mode
) ||
559 strcmp(one
->path
, two
->path
))
562 /* both are valid and point at the same path. that is, we are
563 * dealing with a change.
565 if (one
->sha1_valid
&& two
->sha1_valid
&&
566 !memcmp(one
->sha1
, two
->sha1
, sizeof(one
->sha1
)))
567 return 1; /* no change */
568 if (!one
->sha1_valid
&& !two
->sha1_valid
)
569 return 1; /* both look at the same file on the filesystem. */
573 static void diff_flush_patch(struct diff_filepair
*p
)
575 const char *name
, *other
;
576 char msg_
[PATH_MAX
*2+200], *msg
;
578 if (diff_unmodified_pair(p
))
582 other
= (strcmp(name
, p
->two
->path
) ? p
->two
->path
: NULL
);
583 if ((DIFF_FILE_VALID(p
->one
) && S_ISDIR(p
->one
->mode
)) ||
584 (DIFF_FILE_VALID(p
->two
) && S_ISDIR(p
->two
->mode
)))
585 return; /* no tree diffs in patch format */
590 "similarity index %d%%\n"
593 (int)(0.5 + p
->score
* 100.0/MAX_SCORE
),
594 p
->one
->path
, p
->two
->path
);
599 "similarity index %d%%\n"
602 (int)(0.5 + p
->score
* 100.0/MAX_SCORE
),
603 p
->one
->path
, p
->two
->path
);
610 if (DIFF_PAIR_UNMERGED(p
))
611 run_external_diff(name
, NULL
, NULL
, NULL
, NULL
);
613 run_external_diff(name
, other
, p
->one
, p
->two
, msg
);
616 int diff_needs_to_stay(struct diff_queue_struct
*q
, int i
,
617 struct diff_filespec
*it
)
619 /* If it will be used in later entry (either stay or used
620 * as the source of rename/copy), we need to copy, not rename.
623 struct diff_filepair
*p
= q
->queue
[i
++];
624 if (!DIFF_FILE_VALID(p
->two
))
625 continue; /* removed is fine */
626 if (strcmp(p
->one
->path
, it
->path
))
627 continue; /* not relevant */
629 /* p has its src set to *it and it is not a delete;
630 * it will be used for in-place change, rename/copy,
631 * or just stays there. We cannot rename it out.
638 int diff_queue_is_empty(void)
640 struct diff_queue_struct
*q
= &diff_queued_diff
;
642 for (i
= 0; i
< q
->nr
; i
++)
643 if (!diff_unmodified_pair(q
->queue
[i
]))
649 void diff_debug_filespec(struct diff_filespec
*s
, int x
, const char *one
)
651 fprintf(stderr
, "queue[%d] %s (%s) %s %06o %s\n",
654 DIFF_FILE_VALID(s
) ? "valid" : "invalid",
656 s
->sha1_valid
? sha1_to_hex(s
->sha1
) : "");
657 fprintf(stderr
, "queue[%d] %s size %lu flags %d\n",
659 s
->size
, s
->xfrm_flags
);
662 void diff_debug_filepair(const struct diff_filepair
*p
, int i
)
664 diff_debug_filespec(p
->one
, i
, "one");
665 diff_debug_filespec(p
->two
, i
, "two");
666 fprintf(stderr
, "score %d, status %c\n",
667 p
->score
, p
->status
? : '?');
670 void diff_debug_queue(const char *msg
, struct diff_queue_struct
*q
)
674 fprintf(stderr
, "%s\n", msg
);
675 fprintf(stderr
, "q->nr = %d\n", q
->nr
);
676 for (i
= 0; i
< q
->nr
; i
++) {
677 struct diff_filepair
*p
= q
->queue
[i
];
678 diff_debug_filepair(p
, i
);
683 static void diff_resolve_rename_copy(void)
686 struct diff_filepair
*p
, *pp
;
687 struct diff_queue_struct
*q
= &diff_queued_diff
;
689 /* This should not depend on the ordering of things. */
691 diff_debug_queue("resolve-rename-copy", q
);
693 for (i
= 0; i
< q
->nr
; i
++) {
695 p
->status
= 0; /* undecided */
696 if (DIFF_PAIR_UNMERGED(p
))
698 else if (!DIFF_FILE_VALID((p
)->one
))
700 else if (!DIFF_FILE_VALID((p
)->two
)) {
701 /* Deletion record should be omitted if there
702 * are rename/copy entries using this one as
703 * the source. Then we can say one of them
704 * is a rename and the rest are copies.
707 for (j
= 0; j
< q
->nr
; j
++) {
709 if (!strcmp(pp
->one
->path
, p
->one
->path
) &&
710 strcmp(pp
->one
->path
, pp
->two
->path
)) {
716 else if (DIFF_PAIR_TYPE_CHANGED(p
))
719 /* from this point on, we are dealing with a pair
720 * whose both sides are valid and of the same type, i.e.
721 * either in-place edit or rename/copy edit.
723 else if (strcmp(p
->one
->path
, p
->two
->path
)) {
724 /* See if there is somebody else anywhere that
725 * will keep the path (either modified or
726 * unmodified). If so, we have to be a copy,
727 * not a rename. In addition, if there is
728 * some other rename or copy that comes later
729 * than us that uses the same source, we
730 * have to be a copy, not a rename.
732 for (j
= 0; j
< q
->nr
; j
++) {
734 if (strcmp(pp
->one
->path
, p
->one
->path
))
736 if (!strcmp(pp
->one
->path
, pp
->two
->path
)) {
737 if (DIFF_FILE_VALID(pp
->two
)) {
744 /* pp is a rename/copy ... */
746 /* ... and comes later than us */
754 else if (memcmp(p
->one
->sha1
, p
->two
->sha1
, 20) ||
755 p
->one
->mode
!= p
->two
->mode
)
758 /* this is a "no-change" entry */
761 diff_debug_queue("resolve-rename-copy done", q
);
764 void diff_flush(int diff_output_style
, int resolve_rename_copy
)
766 struct diff_queue_struct
*q
= &diff_queued_diff
;
768 int line_termination
= '\n';
769 int inter_name_termination
= '\t';
771 if (diff_output_style
== DIFF_FORMAT_MACHINE
)
772 line_termination
= inter_name_termination
= 0;
773 if (resolve_rename_copy
)
774 diff_resolve_rename_copy();
776 for (i
= 0; i
< q
->nr
; i
++) {
777 struct diff_filepair
*p
= q
->queue
[i
];
778 if (p
->status
== 'X')
781 die("internal error in diff-resolve-rename-copy");
782 switch (diff_output_style
) {
783 case DIFF_FORMAT_PATCH
:
786 case DIFF_FORMAT_HUMAN
:
787 case DIFF_FORMAT_MACHINE
:
788 diff_flush_raw(p
, line_termination
,
789 inter_name_termination
);
793 for (i
= 0; i
< q
->nr
; i
++) {
794 struct diff_filepair
*p
= q
->queue
[i
];
795 diff_free_filespec_data(p
->one
);
796 diff_free_filespec_data(p
->two
);
801 q
->nr
= q
->alloc
= 0;
804 void diff_addremove(int addremove
, unsigned mode
,
805 const unsigned char *sha1
,
806 const char *base
, const char *path
)
808 char concatpath
[PATH_MAX
];
809 struct diff_filespec
*one
, *two
;
811 /* This may look odd, but it is a preparation for
812 * feeding "there are unchanged files which should
813 * not produce diffs, but when you are doing copy
814 * detection you would need them, so here they are"
815 * entries to the diff-core. They will be prefixed
816 * with something like '=' or '*' (I haven't decided
817 * which but should not make any difference).
818 * Feeding the same new and old to diff_change()
819 * also has the same effect.
820 * Before the final output happens, they are pruned after
821 * merged into rename/copy pairs as appropriate.
824 addremove
= (addremove
== '+' ? '-' :
825 addremove
== '-' ? '+' : addremove
);
827 if (!path
) path
= "";
828 sprintf(concatpath
, "%s%s", base
, path
);
829 one
= alloc_filespec(concatpath
);
830 two
= alloc_filespec(concatpath
);
832 if (addremove
!= '+')
833 fill_filespec(one
, sha1
, mode
);
834 if (addremove
!= '-')
835 fill_filespec(two
, sha1
, mode
);
837 diff_queue(&diff_queued_diff
, one
, two
);
840 void diff_helper_input(unsigned old_mode
,
842 const unsigned char *old_sha1
,
843 const unsigned char *new_sha1
,
844 const char *old_path
,
847 const char *new_path
)
849 struct diff_filespec
*one
, *two
;
850 struct diff_filepair
*dp
;
852 one
= alloc_filespec(old_path
);
853 two
= alloc_filespec(new_path
);
855 fill_filespec(one
, old_sha1
, old_mode
);
857 fill_filespec(two
, new_sha1
, new_mode
);
858 dp
= diff_queue(&diff_queued_diff
, one
, two
);
863 void diff_change(unsigned old_mode
, unsigned new_mode
,
864 const unsigned char *old_sha1
,
865 const unsigned char *new_sha1
,
866 const char *base
, const char *path
)
868 char concatpath
[PATH_MAX
];
869 struct diff_filespec
*one
, *two
;
873 const unsigned char *tmp_c
;
874 tmp
= old_mode
; old_mode
= new_mode
; new_mode
= tmp
;
875 tmp_c
= old_sha1
; old_sha1
= new_sha1
; new_sha1
= tmp_c
;
877 if (!path
) path
= "";
878 sprintf(concatpath
, "%s%s", base
, path
);
879 one
= alloc_filespec(concatpath
);
880 two
= alloc_filespec(concatpath
);
881 fill_filespec(one
, old_sha1
, old_mode
);
882 fill_filespec(two
, new_sha1
, new_mode
);
884 diff_queue(&diff_queued_diff
, one
, two
);
887 void diff_unmerge(const char *path
)
889 struct diff_filespec
*one
, *two
;
890 one
= alloc_filespec(path
);
891 two
= alloc_filespec(path
);
892 diff_queue(&diff_queued_diff
, one
, two
);