builtin-am: support and auto-detect StGit patches
[git] / builtin / am.c
1 /*
2  * Builtin "git am"
3  *
4  * Based on git-am.sh by Junio C Hamano.
5  */
6 #include "cache.h"
7 #include "builtin.h"
8 #include "exec_cmd.h"
9 #include "parse-options.h"
10 #include "dir.h"
11 #include "run-command.h"
12 #include "quote.h"
13 #include "lockfile.h"
14 #include "cache-tree.h"
15 #include "refs.h"
16 #include "commit.h"
17 #include "diff.h"
18 #include "diffcore.h"
19 #include "unpack-trees.h"
20 #include "branch.h"
21 #include "sequencer.h"
22 #include "revision.h"
23 #include "merge-recursive.h"
24 #include "revision.h"
25 #include "log-tree.h"
26 #include "notes-utils.h"
27 #include "rerere.h"
28
29 /**
30  * Returns 1 if the file is empty or does not exist, 0 otherwise.
31  */
32 static int is_empty_file(const char *filename)
33 {
34         struct stat st;
35
36         if (stat(filename, &st) < 0) {
37                 if (errno == ENOENT)
38                         return 1;
39                 die_errno(_("could not stat %s"), filename);
40         }
41
42         return !st.st_size;
43 }
44
45 /**
46  * Like strbuf_getline(), but treats both '\n' and "\r\n" as line terminators.
47  */
48 static int strbuf_getline_crlf(struct strbuf *sb, FILE *fp)
49 {
50         if (strbuf_getwholeline(sb, fp, '\n'))
51                 return EOF;
52         if (sb->buf[sb->len - 1] == '\n') {
53                 strbuf_setlen(sb, sb->len - 1);
54                 if (sb->len > 0 && sb->buf[sb->len - 1] == '\r')
55                         strbuf_setlen(sb, sb->len - 1);
56         }
57         return 0;
58 }
59
60 /**
61  * Returns the length of the first line of msg.
62  */
63 static int linelen(const char *msg)
64 {
65         return strchrnul(msg, '\n') - msg;
66 }
67
68 /**
69  * Returns true if `str` consists of only whitespace, false otherwise.
70  */
71 static int str_isspace(const char *str)
72 {
73         for (; *str; str++)
74                 if (!isspace(*str))
75                         return 0;
76
77         return 1;
78 }
79
80 enum patch_format {
81         PATCH_FORMAT_UNKNOWN = 0,
82         PATCH_FORMAT_MBOX,
83         PATCH_FORMAT_STGIT
84 };
85
86 enum keep_type {
87         KEEP_FALSE = 0,
88         KEEP_TRUE,      /* pass -k flag to git-mailinfo */
89         KEEP_NON_PATCH  /* pass -b flag to git-mailinfo */
90 };
91
92 enum scissors_type {
93         SCISSORS_UNSET = -1,
94         SCISSORS_FALSE = 0,  /* pass --no-scissors to git-mailinfo */
95         SCISSORS_TRUE        /* pass --scissors to git-mailinfo */
96 };
97
98 struct am_state {
99         /* state directory path */
100         char *dir;
101
102         /* current and last patch numbers, 1-indexed */
103         int cur;
104         int last;
105
106         /* commit metadata and message */
107         char *author_name;
108         char *author_email;
109         char *author_date;
110         char *msg;
111         size_t msg_len;
112
113         /* when --rebasing, records the original commit the patch came from */
114         unsigned char orig_commit[GIT_SHA1_RAWSZ];
115
116         /* number of digits in patch filename */
117         int prec;
118
119         /* various operating modes and command line options */
120         int threeway;
121         int quiet;
122         int signoff;
123         int utf8;
124         int keep; /* enum keep_type */
125         int message_id;
126         int scissors; /* enum scissors_type */
127         struct argv_array git_apply_opts;
128         const char *resolvemsg;
129         int committer_date_is_author_date;
130         int ignore_date;
131         int allow_rerere_autoupdate;
132         const char *sign_commit;
133         int rebasing;
134 };
135
136 /**
137  * Initializes am_state with the default values. The state directory is set to
138  * dir.
139  */
140 static void am_state_init(struct am_state *state, const char *dir)
141 {
142         int gpgsign;
143
144         memset(state, 0, sizeof(*state));
145
146         assert(dir);
147         state->dir = xstrdup(dir);
148
149         state->prec = 4;
150
151         state->utf8 = 1;
152
153         git_config_get_bool("am.messageid", &state->message_id);
154
155         state->scissors = SCISSORS_UNSET;
156
157         argv_array_init(&state->git_apply_opts);
158
159         if (!git_config_get_bool("commit.gpgsign", &gpgsign))
160                 state->sign_commit = gpgsign ? "" : NULL;
161 }
162
163 /**
164  * Releases memory allocated by an am_state.
165  */
166 static void am_state_release(struct am_state *state)
167 {
168         free(state->dir);
169         free(state->author_name);
170         free(state->author_email);
171         free(state->author_date);
172         free(state->msg);
173         argv_array_clear(&state->git_apply_opts);
174 }
175
176 /**
177  * Returns path relative to the am_state directory.
178  */
179 static inline const char *am_path(const struct am_state *state, const char *path)
180 {
181         return mkpath("%s/%s", state->dir, path);
182 }
183
184 /**
185  * If state->quiet is false, calls fprintf(fp, fmt, ...), and appends a newline
186  * at the end.
187  */
188 static void say(const struct am_state *state, FILE *fp, const char *fmt, ...)
189 {
190         va_list ap;
191
192         va_start(ap, fmt);
193         if (!state->quiet) {
194                 vfprintf(fp, fmt, ap);
195                 putc('\n', fp);
196         }
197         va_end(ap);
198 }
199
200 /**
201  * Returns 1 if there is an am session in progress, 0 otherwise.
202  */
203 static int am_in_progress(const struct am_state *state)
204 {
205         struct stat st;
206
207         if (lstat(state->dir, &st) < 0 || !S_ISDIR(st.st_mode))
208                 return 0;
209         if (lstat(am_path(state, "last"), &st) || !S_ISREG(st.st_mode))
210                 return 0;
211         if (lstat(am_path(state, "next"), &st) || !S_ISREG(st.st_mode))
212                 return 0;
213         return 1;
214 }
215
216 /**
217  * Reads the contents of `file` in the `state` directory into `sb`. Returns the
218  * number of bytes read on success, -1 if the file does not exist. If `trim` is
219  * set, trailing whitespace will be removed.
220  */
221 static int read_state_file(struct strbuf *sb, const struct am_state *state,
222                         const char *file, int trim)
223 {
224         strbuf_reset(sb);
225
226         if (strbuf_read_file(sb, am_path(state, file), 0) >= 0) {
227                 if (trim)
228                         strbuf_trim(sb);
229
230                 return sb->len;
231         }
232
233         if (errno == ENOENT)
234                 return -1;
235
236         die_errno(_("could not read '%s'"), am_path(state, file));
237 }
238
239 /**
240  * Reads a KEY=VALUE shell variable assignment from `fp`, returning the VALUE
241  * as a newly-allocated string. VALUE must be a quoted string, and the KEY must
242  * match `key`. Returns NULL on failure.
243  *
244  * This is used by read_author_script() to read the GIT_AUTHOR_* variables from
245  * the author-script.
246  */
247 static char *read_shell_var(FILE *fp, const char *key)
248 {
249         struct strbuf sb = STRBUF_INIT;
250         const char *str;
251
252         if (strbuf_getline(&sb, fp, '\n'))
253                 goto fail;
254
255         if (!skip_prefix(sb.buf, key, &str))
256                 goto fail;
257
258         if (!skip_prefix(str, "=", &str))
259                 goto fail;
260
261         strbuf_remove(&sb, 0, str - sb.buf);
262
263         str = sq_dequote(sb.buf);
264         if (!str)
265                 goto fail;
266
267         return strbuf_detach(&sb, NULL);
268
269 fail:
270         strbuf_release(&sb);
271         return NULL;
272 }
273
274 /**
275  * Reads and parses the state directory's "author-script" file, and sets
276  * state->author_name, state->author_email and state->author_date accordingly.
277  * Returns 0 on success, -1 if the file could not be parsed.
278  *
279  * The author script is of the format:
280  *
281  *      GIT_AUTHOR_NAME='$author_name'
282  *      GIT_AUTHOR_EMAIL='$author_email'
283  *      GIT_AUTHOR_DATE='$author_date'
284  *
285  * where $author_name, $author_email and $author_date are quoted. We are strict
286  * with our parsing, as the file was meant to be eval'd in the old git-am.sh
287  * script, and thus if the file differs from what this function expects, it is
288  * better to bail out than to do something that the user does not expect.
289  */
290 static int read_author_script(struct am_state *state)
291 {
292         const char *filename = am_path(state, "author-script");
293         FILE *fp;
294
295         assert(!state->author_name);
296         assert(!state->author_email);
297         assert(!state->author_date);
298
299         fp = fopen(filename, "r");
300         if (!fp) {
301                 if (errno == ENOENT)
302                         return 0;
303                 die_errno(_("could not open '%s' for reading"), filename);
304         }
305
306         state->author_name = read_shell_var(fp, "GIT_AUTHOR_NAME");
307         if (!state->author_name) {
308                 fclose(fp);
309                 return -1;
310         }
311
312         state->author_email = read_shell_var(fp, "GIT_AUTHOR_EMAIL");
313         if (!state->author_email) {
314                 fclose(fp);
315                 return -1;
316         }
317
318         state->author_date = read_shell_var(fp, "GIT_AUTHOR_DATE");
319         if (!state->author_date) {
320                 fclose(fp);
321                 return -1;
322         }
323
324         if (fgetc(fp) != EOF) {
325                 fclose(fp);
326                 return -1;
327         }
328
329         fclose(fp);
330         return 0;
331 }
332
333 /**
334  * Saves state->author_name, state->author_email and state->author_date in the
335  * state directory's "author-script" file.
336  */
337 static void write_author_script(const struct am_state *state)
338 {
339         struct strbuf sb = STRBUF_INIT;
340
341         strbuf_addstr(&sb, "GIT_AUTHOR_NAME=");
342         sq_quote_buf(&sb, state->author_name);
343         strbuf_addch(&sb, '\n');
344
345         strbuf_addstr(&sb, "GIT_AUTHOR_EMAIL=");
346         sq_quote_buf(&sb, state->author_email);
347         strbuf_addch(&sb, '\n');
348
349         strbuf_addstr(&sb, "GIT_AUTHOR_DATE=");
350         sq_quote_buf(&sb, state->author_date);
351         strbuf_addch(&sb, '\n');
352
353         write_file(am_path(state, "author-script"), 1, "%s", sb.buf);
354
355         strbuf_release(&sb);
356 }
357
358 /**
359  * Reads the commit message from the state directory's "final-commit" file,
360  * setting state->msg to its contents and state->msg_len to the length of its
361  * contents in bytes.
362  *
363  * Returns 0 on success, -1 if the file does not exist.
364  */
365 static int read_commit_msg(struct am_state *state)
366 {
367         struct strbuf sb = STRBUF_INIT;
368
369         assert(!state->msg);
370
371         if (read_state_file(&sb, state, "final-commit", 0) < 0) {
372                 strbuf_release(&sb);
373                 return -1;
374         }
375
376         state->msg = strbuf_detach(&sb, &state->msg_len);
377         return 0;
378 }
379
380 /**
381  * Saves state->msg in the state directory's "final-commit" file.
382  */
383 static void write_commit_msg(const struct am_state *state)
384 {
385         int fd;
386         const char *filename = am_path(state, "final-commit");
387
388         fd = xopen(filename, O_WRONLY | O_CREAT, 0666);
389         if (write_in_full(fd, state->msg, state->msg_len) < 0)
390                 die_errno(_("could not write to %s"), filename);
391         close(fd);
392 }
393
394 /**
395  * Loads state from disk.
396  */
397 static void am_load(struct am_state *state)
398 {
399         struct strbuf sb = STRBUF_INIT;
400
401         if (read_state_file(&sb, state, "next", 1) < 0)
402                 die("BUG: state file 'next' does not exist");
403         state->cur = strtol(sb.buf, NULL, 10);
404
405         if (read_state_file(&sb, state, "last", 1) < 0)
406                 die("BUG: state file 'last' does not exist");
407         state->last = strtol(sb.buf, NULL, 10);
408
409         if (read_author_script(state) < 0)
410                 die(_("could not parse author script"));
411
412         read_commit_msg(state);
413
414         if (read_state_file(&sb, state, "original-commit", 1) < 0)
415                 hashclr(state->orig_commit);
416         else if (get_sha1_hex(sb.buf, state->orig_commit) < 0)
417                 die(_("could not parse %s"), am_path(state, "original-commit"));
418
419         read_state_file(&sb, state, "threeway", 1);
420         state->threeway = !strcmp(sb.buf, "t");
421
422         read_state_file(&sb, state, "quiet", 1);
423         state->quiet = !strcmp(sb.buf, "t");
424
425         read_state_file(&sb, state, "sign", 1);
426         state->signoff = !strcmp(sb.buf, "t");
427
428         read_state_file(&sb, state, "utf8", 1);
429         state->utf8 = !strcmp(sb.buf, "t");
430
431         read_state_file(&sb, state, "keep", 1);
432         if (!strcmp(sb.buf, "t"))
433                 state->keep = KEEP_TRUE;
434         else if (!strcmp(sb.buf, "b"))
435                 state->keep = KEEP_NON_PATCH;
436         else
437                 state->keep = KEEP_FALSE;
438
439         read_state_file(&sb, state, "messageid", 1);
440         state->message_id = !strcmp(sb.buf, "t");
441
442         read_state_file(&sb, state, "scissors", 1);
443         if (!strcmp(sb.buf, "t"))
444                 state->scissors = SCISSORS_TRUE;
445         else if (!strcmp(sb.buf, "f"))
446                 state->scissors = SCISSORS_FALSE;
447         else
448                 state->scissors = SCISSORS_UNSET;
449
450         read_state_file(&sb, state, "apply-opt", 1);
451         argv_array_clear(&state->git_apply_opts);
452         if (sq_dequote_to_argv_array(sb.buf, &state->git_apply_opts) < 0)
453                 die(_("could not parse %s"), am_path(state, "apply-opt"));
454
455         state->rebasing = !!file_exists(am_path(state, "rebasing"));
456
457         strbuf_release(&sb);
458 }
459
460 /**
461  * Removes the am_state directory, forcefully terminating the current am
462  * session.
463  */
464 static void am_destroy(const struct am_state *state)
465 {
466         struct strbuf sb = STRBUF_INIT;
467
468         strbuf_addstr(&sb, state->dir);
469         remove_dir_recursively(&sb, 0);
470         strbuf_release(&sb);
471 }
472
473 /**
474  * Runs applypatch-msg hook. Returns its exit code.
475  */
476 static int run_applypatch_msg_hook(struct am_state *state)
477 {
478         int ret;
479
480         assert(state->msg);
481         ret = run_hook_le(NULL, "applypatch-msg", am_path(state, "final-commit"), NULL);
482
483         if (!ret) {
484                 free(state->msg);
485                 state->msg = NULL;
486                 if (read_commit_msg(state) < 0)
487                         die(_("'%s' was deleted by the applypatch-msg hook"),
488                                 am_path(state, "final-commit"));
489         }
490
491         return ret;
492 }
493
494 /**
495  * Runs post-rewrite hook. Returns it exit code.
496  */
497 static int run_post_rewrite_hook(const struct am_state *state)
498 {
499         struct child_process cp = CHILD_PROCESS_INIT;
500         const char *hook = find_hook("post-rewrite");
501         int ret;
502
503         if (!hook)
504                 return 0;
505
506         argv_array_push(&cp.args, hook);
507         argv_array_push(&cp.args, "rebase");
508
509         cp.in = xopen(am_path(state, "rewritten"), O_RDONLY);
510         cp.stdout_to_stderr = 1;
511
512         ret = run_command(&cp);
513
514         close(cp.in);
515         return ret;
516 }
517
518 /**
519  * Reads the state directory's "rewritten" file, and copies notes from the old
520  * commits listed in the file to their rewritten commits.
521  *
522  * Returns 0 on success, -1 on failure.
523  */
524 static int copy_notes_for_rebase(const struct am_state *state)
525 {
526         struct notes_rewrite_cfg *c;
527         struct strbuf sb = STRBUF_INIT;
528         const char *invalid_line = _("Malformed input line: '%s'.");
529         const char *msg = "Notes added by 'git rebase'";
530         FILE *fp;
531         int ret = 0;
532
533         assert(state->rebasing);
534
535         c = init_copy_notes_for_rewrite("rebase");
536         if (!c)
537                 return 0;
538
539         fp = xfopen(am_path(state, "rewritten"), "r");
540
541         while (!strbuf_getline(&sb, fp, '\n')) {
542                 unsigned char from_obj[GIT_SHA1_RAWSZ], to_obj[GIT_SHA1_RAWSZ];
543
544                 if (sb.len != GIT_SHA1_HEXSZ * 2 + 1) {
545                         ret = error(invalid_line, sb.buf);
546                         goto finish;
547                 }
548
549                 if (get_sha1_hex(sb.buf, from_obj)) {
550                         ret = error(invalid_line, sb.buf);
551                         goto finish;
552                 }
553
554                 if (sb.buf[GIT_SHA1_HEXSZ] != ' ') {
555                         ret = error(invalid_line, sb.buf);
556                         goto finish;
557                 }
558
559                 if (get_sha1_hex(sb.buf + GIT_SHA1_HEXSZ + 1, to_obj)) {
560                         ret = error(invalid_line, sb.buf);
561                         goto finish;
562                 }
563
564                 if (copy_note_for_rewrite(c, from_obj, to_obj))
565                         ret = error(_("Failed to copy notes from '%s' to '%s'"),
566                                         sha1_to_hex(from_obj), sha1_to_hex(to_obj));
567         }
568
569 finish:
570         finish_copy_notes_for_rewrite(c, msg);
571         fclose(fp);
572         strbuf_release(&sb);
573         return ret;
574 }
575
576 /**
577  * Determines if the file looks like a piece of RFC2822 mail by grabbing all
578  * non-indented lines and checking if they look like they begin with valid
579  * header field names.
580  *
581  * Returns 1 if the file looks like a piece of mail, 0 otherwise.
582  */
583 static int is_mail(FILE *fp)
584 {
585         const char *header_regex = "^[!-9;-~]+:";
586         struct strbuf sb = STRBUF_INIT;
587         regex_t regex;
588         int ret = 1;
589
590         if (fseek(fp, 0L, SEEK_SET))
591                 die_errno(_("fseek failed"));
592
593         if (regcomp(&regex, header_regex, REG_NOSUB | REG_EXTENDED))
594                 die("invalid pattern: %s", header_regex);
595
596         while (!strbuf_getline_crlf(&sb, fp)) {
597                 if (!sb.len)
598                         break; /* End of header */
599
600                 /* Ignore indented folded lines */
601                 if (*sb.buf == '\t' || *sb.buf == ' ')
602                         continue;
603
604                 /* It's a header if it matches header_regex */
605                 if (regexec(&regex, sb.buf, 0, NULL, 0)) {
606                         ret = 0;
607                         goto done;
608                 }
609         }
610
611 done:
612         regfree(&regex);
613         strbuf_release(&sb);
614         return ret;
615 }
616
617 /**
618  * Attempts to detect the patch_format of the patches contained in `paths`,
619  * returning the PATCH_FORMAT_* enum value. Returns PATCH_FORMAT_UNKNOWN if
620  * detection fails.
621  */
622 static int detect_patch_format(const char **paths)
623 {
624         enum patch_format ret = PATCH_FORMAT_UNKNOWN;
625         struct strbuf l1 = STRBUF_INIT;
626         struct strbuf l2 = STRBUF_INIT;
627         struct strbuf l3 = STRBUF_INIT;
628         FILE *fp;
629
630         /*
631          * We default to mbox format if input is from stdin and for directories
632          */
633         if (!*paths || !strcmp(*paths, "-") || is_directory(*paths))
634                 return PATCH_FORMAT_MBOX;
635
636         /*
637          * Otherwise, check the first few lines of the first patch, starting
638          * from the first non-blank line, to try to detect its format.
639          */
640
641         fp = xfopen(*paths, "r");
642
643         while (!strbuf_getline_crlf(&l1, fp)) {
644                 if (l1.len)
645                         break;
646         }
647
648         if (starts_with(l1.buf, "From ") || starts_with(l1.buf, "From: ")) {
649                 ret = PATCH_FORMAT_MBOX;
650                 goto done;
651         }
652
653         strbuf_reset(&l2);
654         strbuf_getline_crlf(&l2, fp);
655         strbuf_reset(&l3);
656         strbuf_getline_crlf(&l3, fp);
657
658         /*
659          * If the second line is empty and the third is a From, Author or Date
660          * entry, this is likely an StGit patch.
661          */
662         if (l1.len && !l2.len &&
663                 (starts_with(l3.buf, "From:") ||
664                  starts_with(l3.buf, "Author:") ||
665                  starts_with(l3.buf, "Date:"))) {
666                 ret = PATCH_FORMAT_STGIT;
667                 goto done;
668         }
669
670         if (l1.len && is_mail(fp)) {
671                 ret = PATCH_FORMAT_MBOX;
672                 goto done;
673         }
674
675 done:
676         fclose(fp);
677         strbuf_release(&l1);
678         return ret;
679 }
680
681 /**
682  * Splits out individual email patches from `paths`, where each path is either
683  * a mbox file or a Maildir. Returns 0 on success, -1 on failure.
684  */
685 static int split_mail_mbox(struct am_state *state, const char **paths, int keep_cr)
686 {
687         struct child_process cp = CHILD_PROCESS_INIT;
688         struct strbuf last = STRBUF_INIT;
689
690         cp.git_cmd = 1;
691         argv_array_push(&cp.args, "mailsplit");
692         argv_array_pushf(&cp.args, "-d%d", state->prec);
693         argv_array_pushf(&cp.args, "-o%s", state->dir);
694         argv_array_push(&cp.args, "-b");
695         if (keep_cr)
696                 argv_array_push(&cp.args, "--keep-cr");
697         argv_array_push(&cp.args, "--");
698         argv_array_pushv(&cp.args, paths);
699
700         if (capture_command(&cp, &last, 8))
701                 return -1;
702
703         state->cur = 1;
704         state->last = strtol(last.buf, NULL, 10);
705
706         return 0;
707 }
708
709 /**
710  * Callback signature for split_mail_conv(). The foreign patch should be
711  * read from `in`, and the converted patch (in RFC2822 mail format) should be
712  * written to `out`. Return 0 on success, or -1 on failure.
713  */
714 typedef int (*mail_conv_fn)(FILE *out, FILE *in, int keep_cr);
715
716 /**
717  * Calls `fn` for each file in `paths` to convert the foreign patch to the
718  * RFC2822 mail format suitable for parsing with git-mailinfo.
719  *
720  * Returns 0 on success, -1 on failure.
721  */
722 static int split_mail_conv(mail_conv_fn fn, struct am_state *state,
723                         const char **paths, int keep_cr)
724 {
725         static const char *stdin_only[] = {"-", NULL};
726         int i;
727
728         if (!*paths)
729                 paths = stdin_only;
730
731         for (i = 0; *paths; paths++, i++) {
732                 FILE *in, *out;
733                 const char *mail;
734                 int ret;
735
736                 if (!strcmp(*paths, "-"))
737                         in = stdin;
738                 else
739                         in = fopen(*paths, "r");
740
741                 if (!in)
742                         return error(_("could not open '%s' for reading: %s"),
743                                         *paths, strerror(errno));
744
745                 mail = mkpath("%s/%0*d", state->dir, state->prec, i + 1);
746
747                 out = fopen(mail, "w");
748                 if (!out)
749                         return error(_("could not open '%s' for writing: %s"),
750                                         mail, strerror(errno));
751
752                 ret = fn(out, in, keep_cr);
753
754                 fclose(out);
755                 fclose(in);
756
757                 if (ret)
758                         return error(_("could not parse patch '%s'"), *paths);
759         }
760
761         state->cur = 1;
762         state->last = i;
763         return 0;
764 }
765
766 /**
767  * A split_mail_conv() callback that converts an StGit patch to an RFC2822
768  * message suitable for parsing with git-mailinfo.
769  */
770 static int stgit_patch_to_mail(FILE *out, FILE *in, int keep_cr)
771 {
772         struct strbuf sb = STRBUF_INIT;
773         int subject_printed = 0;
774
775         while (!strbuf_getline(&sb, in, '\n')) {
776                 const char *str;
777
778                 if (str_isspace(sb.buf))
779                         continue;
780                 else if (skip_prefix(sb.buf, "Author:", &str))
781                         fprintf(out, "From:%s\n", str);
782                 else if (starts_with(sb.buf, "From") || starts_with(sb.buf, "Date"))
783                         fprintf(out, "%s\n", sb.buf);
784                 else if (!subject_printed) {
785                         fprintf(out, "Subject: %s\n", sb.buf);
786                         subject_printed = 1;
787                 } else {
788                         fprintf(out, "\n%s\n", sb.buf);
789                         break;
790                 }
791         }
792
793         strbuf_reset(&sb);
794         while (strbuf_fread(&sb, 8192, in) > 0) {
795                 fwrite(sb.buf, 1, sb.len, out);
796                 strbuf_reset(&sb);
797         }
798
799         strbuf_release(&sb);
800         return 0;
801 }
802
803 /**
804  * Splits a list of files/directories into individual email patches. Each path
805  * in `paths` must be a file/directory that is formatted according to
806  * `patch_format`.
807  *
808  * Once split out, the individual email patches will be stored in the state
809  * directory, with each patch's filename being its index, padded to state->prec
810  * digits.
811  *
812  * state->cur will be set to the index of the first mail, and state->last will
813  * be set to the index of the last mail.
814  *
815  * Set keep_cr to 0 to convert all lines ending with \r\n to end with \n, 1
816  * to disable this behavior, -1 to use the default configured setting.
817  *
818  * Returns 0 on success, -1 on failure.
819  */
820 static int split_mail(struct am_state *state, enum patch_format patch_format,
821                         const char **paths, int keep_cr)
822 {
823         if (keep_cr < 0) {
824                 keep_cr = 0;
825                 git_config_get_bool("am.keepcr", &keep_cr);
826         }
827
828         switch (patch_format) {
829         case PATCH_FORMAT_MBOX:
830                 return split_mail_mbox(state, paths, keep_cr);
831         case PATCH_FORMAT_STGIT:
832                 return split_mail_conv(stgit_patch_to_mail, state, paths, keep_cr);
833         default:
834                 die("BUG: invalid patch_format");
835         }
836         return -1;
837 }
838
839 /**
840  * Setup a new am session for applying patches
841  */
842 static void am_setup(struct am_state *state, enum patch_format patch_format,
843                         const char **paths, int keep_cr)
844 {
845         unsigned char curr_head[GIT_SHA1_RAWSZ];
846         const char *str;
847         struct strbuf sb = STRBUF_INIT;
848
849         if (!patch_format)
850                 patch_format = detect_patch_format(paths);
851
852         if (!patch_format) {
853                 fprintf_ln(stderr, _("Patch format detection failed."));
854                 exit(128);
855         }
856
857         if (mkdir(state->dir, 0777) < 0 && errno != EEXIST)
858                 die_errno(_("failed to create directory '%s'"), state->dir);
859
860         if (split_mail(state, patch_format, paths, keep_cr) < 0) {
861                 am_destroy(state);
862                 die(_("Failed to split patches."));
863         }
864
865         if (state->rebasing)
866                 state->threeway = 1;
867
868         write_file(am_path(state, "threeway"), 1, state->threeway ? "t" : "f");
869
870         write_file(am_path(state, "quiet"), 1, state->quiet ? "t" : "f");
871
872         write_file(am_path(state, "sign"), 1, state->signoff ? "t" : "f");
873
874         write_file(am_path(state, "utf8"), 1, state->utf8 ? "t" : "f");
875
876         switch (state->keep) {
877         case KEEP_FALSE:
878                 str = "f";
879                 break;
880         case KEEP_TRUE:
881                 str = "t";
882                 break;
883         case KEEP_NON_PATCH:
884                 str = "b";
885                 break;
886         default:
887                 die("BUG: invalid value for state->keep");
888         }
889
890         write_file(am_path(state, "keep"), 1, "%s", str);
891
892         write_file(am_path(state, "messageid"), 1, state->message_id ? "t" : "f");
893
894         switch (state->scissors) {
895         case SCISSORS_UNSET:
896                 str = "";
897                 break;
898         case SCISSORS_FALSE:
899                 str = "f";
900                 break;
901         case SCISSORS_TRUE:
902                 str = "t";
903                 break;
904         default:
905                 die("BUG: invalid value for state->scissors");
906         }
907
908         write_file(am_path(state, "scissors"), 1, "%s", str);
909
910         sq_quote_argv(&sb, state->git_apply_opts.argv, 0);
911         write_file(am_path(state, "apply-opt"), 1, "%s", sb.buf);
912
913         if (state->rebasing)
914                 write_file(am_path(state, "rebasing"), 1, "%s", "");
915         else
916                 write_file(am_path(state, "applying"), 1, "%s", "");
917
918         if (!get_sha1("HEAD", curr_head)) {
919                 write_file(am_path(state, "abort-safety"), 1, "%s", sha1_to_hex(curr_head));
920                 if (!state->rebasing)
921                         update_ref("am", "ORIG_HEAD", curr_head, NULL, 0,
922                                         UPDATE_REFS_DIE_ON_ERR);
923         } else {
924                 write_file(am_path(state, "abort-safety"), 1, "%s", "");
925                 if (!state->rebasing)
926                         delete_ref("ORIG_HEAD", NULL, 0);
927         }
928
929         /*
930          * NOTE: Since the "next" and "last" files determine if an am_state
931          * session is in progress, they should be written last.
932          */
933
934         write_file(am_path(state, "next"), 1, "%d", state->cur);
935
936         write_file(am_path(state, "last"), 1, "%d", state->last);
937
938         strbuf_release(&sb);
939 }
940
941 /**
942  * Increments the patch pointer, and cleans am_state for the application of the
943  * next patch.
944  */
945 static void am_next(struct am_state *state)
946 {
947         unsigned char head[GIT_SHA1_RAWSZ];
948
949         free(state->author_name);
950         state->author_name = NULL;
951
952         free(state->author_email);
953         state->author_email = NULL;
954
955         free(state->author_date);
956         state->author_date = NULL;
957
958         free(state->msg);
959         state->msg = NULL;
960         state->msg_len = 0;
961
962         unlink(am_path(state, "author-script"));
963         unlink(am_path(state, "final-commit"));
964
965         hashclr(state->orig_commit);
966         unlink(am_path(state, "original-commit"));
967
968         if (!get_sha1("HEAD", head))
969                 write_file(am_path(state, "abort-safety"), 1, "%s", sha1_to_hex(head));
970         else
971                 write_file(am_path(state, "abort-safety"), 1, "%s", "");
972
973         state->cur++;
974         write_file(am_path(state, "next"), 1, "%d", state->cur);
975 }
976
977 /**
978  * Returns the filename of the current patch email.
979  */
980 static const char *msgnum(const struct am_state *state)
981 {
982         static struct strbuf sb = STRBUF_INIT;
983
984         strbuf_reset(&sb);
985         strbuf_addf(&sb, "%0*d", state->prec, state->cur);
986
987         return sb.buf;
988 }
989
990 /**
991  * Refresh and write index.
992  */
993 static void refresh_and_write_cache(void)
994 {
995         struct lock_file *lock_file = xcalloc(1, sizeof(struct lock_file));
996
997         hold_locked_index(lock_file, 1);
998         refresh_cache(REFRESH_QUIET);
999         if (write_locked_index(&the_index, lock_file, COMMIT_LOCK))
1000                 die(_("unable to write index file"));
1001 }
1002
1003 /**
1004  * Returns 1 if the index differs from HEAD, 0 otherwise. When on an unborn
1005  * branch, returns 1 if there are entries in the index, 0 otherwise. If an
1006  * strbuf is provided, the space-separated list of files that differ will be
1007  * appended to it.
1008  */
1009 static int index_has_changes(struct strbuf *sb)
1010 {
1011         unsigned char head[GIT_SHA1_RAWSZ];
1012         int i;
1013
1014         if (!get_sha1_tree("HEAD", head)) {
1015                 struct diff_options opt;
1016
1017                 diff_setup(&opt);
1018                 DIFF_OPT_SET(&opt, EXIT_WITH_STATUS);
1019                 if (!sb)
1020                         DIFF_OPT_SET(&opt, QUICK);
1021                 do_diff_cache(head, &opt);
1022                 diffcore_std(&opt);
1023                 for (i = 0; sb && i < diff_queued_diff.nr; i++) {
1024                         if (i)
1025                                 strbuf_addch(sb, ' ');
1026                         strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
1027                 }
1028                 diff_flush(&opt);
1029                 return DIFF_OPT_TST(&opt, HAS_CHANGES) != 0;
1030         } else {
1031                 for (i = 0; sb && i < active_nr; i++) {
1032                         if (i)
1033                                 strbuf_addch(sb, ' ');
1034                         strbuf_addstr(sb, active_cache[i]->name);
1035                 }
1036                 return !!active_nr;
1037         }
1038 }
1039
1040 /**
1041  * Dies with a user-friendly message on how to proceed after resolving the
1042  * problem. This message can be overridden with state->resolvemsg.
1043  */
1044 static void NORETURN die_user_resolve(const struct am_state *state)
1045 {
1046         if (state->resolvemsg) {
1047                 printf_ln("%s", state->resolvemsg);
1048         } else {
1049                 const char *cmdline = "git am";
1050
1051                 printf_ln(_("When you have resolved this problem, run \"%s --continue\"."), cmdline);
1052                 printf_ln(_("If you prefer to skip this patch, run \"%s --skip\" instead."), cmdline);
1053                 printf_ln(_("To restore the original branch and stop patching, run \"%s --abort\"."), cmdline);
1054         }
1055
1056         exit(128);
1057 }
1058
1059 /**
1060  * Parses `mail` using git-mailinfo, extracting its patch and authorship info.
1061  * state->msg will be set to the patch message. state->author_name,
1062  * state->author_email and state->author_date will be set to the patch author's
1063  * name, email and date respectively. The patch body will be written to the
1064  * state directory's "patch" file.
1065  *
1066  * Returns 1 if the patch should be skipped, 0 otherwise.
1067  */
1068 static int parse_mail(struct am_state *state, const char *mail)
1069 {
1070         FILE *fp;
1071         struct child_process cp = CHILD_PROCESS_INIT;
1072         struct strbuf sb = STRBUF_INIT;
1073         struct strbuf msg = STRBUF_INIT;
1074         struct strbuf author_name = STRBUF_INIT;
1075         struct strbuf author_date = STRBUF_INIT;
1076         struct strbuf author_email = STRBUF_INIT;
1077         int ret = 0;
1078
1079         cp.git_cmd = 1;
1080         cp.in = xopen(mail, O_RDONLY, 0);
1081         cp.out = xopen(am_path(state, "info"), O_WRONLY | O_CREAT, 0777);
1082
1083         argv_array_push(&cp.args, "mailinfo");
1084         argv_array_push(&cp.args, state->utf8 ? "-u" : "-n");
1085
1086         switch (state->keep) {
1087         case KEEP_FALSE:
1088                 break;
1089         case KEEP_TRUE:
1090                 argv_array_push(&cp.args, "-k");
1091                 break;
1092         case KEEP_NON_PATCH:
1093                 argv_array_push(&cp.args, "-b");
1094                 break;
1095         default:
1096                 die("BUG: invalid value for state->keep");
1097         }
1098
1099         if (state->message_id)
1100                 argv_array_push(&cp.args, "-m");
1101
1102         switch (state->scissors) {
1103         case SCISSORS_UNSET:
1104                 break;
1105         case SCISSORS_FALSE:
1106                 argv_array_push(&cp.args, "--no-scissors");
1107                 break;
1108         case SCISSORS_TRUE:
1109                 argv_array_push(&cp.args, "--scissors");
1110                 break;
1111         default:
1112                 die("BUG: invalid value for state->scissors");
1113         }
1114
1115         argv_array_push(&cp.args, am_path(state, "msg"));
1116         argv_array_push(&cp.args, am_path(state, "patch"));
1117
1118         if (run_command(&cp) < 0)
1119                 die("could not parse patch");
1120
1121         close(cp.in);
1122         close(cp.out);
1123
1124         /* Extract message and author information */
1125         fp = xfopen(am_path(state, "info"), "r");
1126         while (!strbuf_getline(&sb, fp, '\n')) {
1127                 const char *x;
1128
1129                 if (skip_prefix(sb.buf, "Subject: ", &x)) {
1130                         if (msg.len)
1131                                 strbuf_addch(&msg, '\n');
1132                         strbuf_addstr(&msg, x);
1133                 } else if (skip_prefix(sb.buf, "Author: ", &x))
1134                         strbuf_addstr(&author_name, x);
1135                 else if (skip_prefix(sb.buf, "Email: ", &x))
1136                         strbuf_addstr(&author_email, x);
1137                 else if (skip_prefix(sb.buf, "Date: ", &x))
1138                         strbuf_addstr(&author_date, x);
1139         }
1140         fclose(fp);
1141
1142         /* Skip pine's internal folder data */
1143         if (!strcmp(author_name.buf, "Mail System Internal Data")) {
1144                 ret = 1;
1145                 goto finish;
1146         }
1147
1148         if (is_empty_file(am_path(state, "patch"))) {
1149                 printf_ln(_("Patch is empty. Was it split wrong?"));
1150                 die_user_resolve(state);
1151         }
1152
1153         strbuf_addstr(&msg, "\n\n");
1154         if (strbuf_read_file(&msg, am_path(state, "msg"), 0) < 0)
1155                 die_errno(_("could not read '%s'"), am_path(state, "msg"));
1156         stripspace(&msg, 0);
1157
1158         if (state->signoff)
1159                 append_signoff(&msg, 0, 0);
1160
1161         assert(!state->author_name);
1162         state->author_name = strbuf_detach(&author_name, NULL);
1163
1164         assert(!state->author_email);
1165         state->author_email = strbuf_detach(&author_email, NULL);
1166
1167         assert(!state->author_date);
1168         state->author_date = strbuf_detach(&author_date, NULL);
1169
1170         assert(!state->msg);
1171         state->msg = strbuf_detach(&msg, &state->msg_len);
1172
1173 finish:
1174         strbuf_release(&msg);
1175         strbuf_release(&author_date);
1176         strbuf_release(&author_email);
1177         strbuf_release(&author_name);
1178         strbuf_release(&sb);
1179         return ret;
1180 }
1181
1182 /**
1183  * Sets commit_id to the commit hash where the mail was generated from.
1184  * Returns 0 on success, -1 on failure.
1185  */
1186 static int get_mail_commit_sha1(unsigned char *commit_id, const char *mail)
1187 {
1188         struct strbuf sb = STRBUF_INIT;
1189         FILE *fp = xfopen(mail, "r");
1190         const char *x;
1191
1192         if (strbuf_getline(&sb, fp, '\n'))
1193                 return -1;
1194
1195         if (!skip_prefix(sb.buf, "From ", &x))
1196                 return -1;
1197
1198         if (get_sha1_hex(x, commit_id) < 0)
1199                 return -1;
1200
1201         strbuf_release(&sb);
1202         fclose(fp);
1203         return 0;
1204 }
1205
1206 /**
1207  * Sets state->msg, state->author_name, state->author_email, state->author_date
1208  * to the commit's respective info.
1209  */
1210 static void get_commit_info(struct am_state *state, struct commit *commit)
1211 {
1212         const char *buffer, *ident_line, *author_date, *msg;
1213         size_t ident_len;
1214         struct ident_split ident_split;
1215         struct strbuf sb = STRBUF_INIT;
1216
1217         buffer = logmsg_reencode(commit, NULL, get_commit_output_encoding());
1218
1219         ident_line = find_commit_header(buffer, "author", &ident_len);
1220
1221         if (split_ident_line(&ident_split, ident_line, ident_len) < 0) {
1222                 strbuf_add(&sb, ident_line, ident_len);
1223                 die(_("invalid ident line: %s"), sb.buf);
1224         }
1225
1226         assert(!state->author_name);
1227         if (ident_split.name_begin) {
1228                 strbuf_add(&sb, ident_split.name_begin,
1229                         ident_split.name_end - ident_split.name_begin);
1230                 state->author_name = strbuf_detach(&sb, NULL);
1231         } else
1232                 state->author_name = xstrdup("");
1233
1234         assert(!state->author_email);
1235         if (ident_split.mail_begin) {
1236                 strbuf_add(&sb, ident_split.mail_begin,
1237                         ident_split.mail_end - ident_split.mail_begin);
1238                 state->author_email = strbuf_detach(&sb, NULL);
1239         } else
1240                 state->author_email = xstrdup("");
1241
1242         author_date = show_ident_date(&ident_split, DATE_MODE(NORMAL));
1243         strbuf_addstr(&sb, author_date);
1244         assert(!state->author_date);
1245         state->author_date = strbuf_detach(&sb, NULL);
1246
1247         assert(!state->msg);
1248         msg = strstr(buffer, "\n\n");
1249         if (!msg)
1250                 die(_("unable to parse commit %s"), sha1_to_hex(commit->object.sha1));
1251         state->msg = xstrdup(msg + 2);
1252         state->msg_len = strlen(state->msg);
1253 }
1254
1255 /**
1256  * Writes `commit` as a patch to the state directory's "patch" file.
1257  */
1258 static void write_commit_patch(const struct am_state *state, struct commit *commit)
1259 {
1260         struct rev_info rev_info;
1261         FILE *fp;
1262
1263         fp = xfopen(am_path(state, "patch"), "w");
1264         init_revisions(&rev_info, NULL);
1265         rev_info.diff = 1;
1266         rev_info.abbrev = 0;
1267         rev_info.disable_stdin = 1;
1268         rev_info.show_root_diff = 1;
1269         rev_info.diffopt.output_format = DIFF_FORMAT_PATCH;
1270         rev_info.no_commit_id = 1;
1271         DIFF_OPT_SET(&rev_info.diffopt, BINARY);
1272         DIFF_OPT_SET(&rev_info.diffopt, FULL_INDEX);
1273         rev_info.diffopt.use_color = 0;
1274         rev_info.diffopt.file = fp;
1275         rev_info.diffopt.close_file = 1;
1276         add_pending_object(&rev_info, &commit->object, "");
1277         diff_setup_done(&rev_info.diffopt);
1278         log_tree_commit(&rev_info, commit);
1279 }
1280
1281 /**
1282  * Like parse_mail(), but parses the mail by looking up its commit ID
1283  * directly. This is used in --rebasing mode to bypass git-mailinfo's munging
1284  * of patches.
1285  *
1286  * state->orig_commit will be set to the original commit ID.
1287  *
1288  * Will always return 0 as the patch should never be skipped.
1289  */
1290 static int parse_mail_rebase(struct am_state *state, const char *mail)
1291 {
1292         struct commit *commit;
1293         unsigned char commit_sha1[GIT_SHA1_RAWSZ];
1294
1295         if (get_mail_commit_sha1(commit_sha1, mail) < 0)
1296                 die(_("could not parse %s"), mail);
1297
1298         commit = lookup_commit_or_die(commit_sha1, mail);
1299
1300         get_commit_info(state, commit);
1301
1302         write_commit_patch(state, commit);
1303
1304         hashcpy(state->orig_commit, commit_sha1);
1305         write_file(am_path(state, "original-commit"), 1, "%s",
1306                         sha1_to_hex(commit_sha1));
1307
1308         return 0;
1309 }
1310
1311 /**
1312  * Applies current patch with git-apply. Returns 0 on success, -1 otherwise. If
1313  * `index_file` is not NULL, the patch will be applied to that index.
1314  */
1315 static int run_apply(const struct am_state *state, const char *index_file)
1316 {
1317         struct child_process cp = CHILD_PROCESS_INIT;
1318
1319         cp.git_cmd = 1;
1320
1321         if (index_file)
1322                 argv_array_pushf(&cp.env_array, "GIT_INDEX_FILE=%s", index_file);
1323
1324         /*
1325          * If we are allowed to fall back on 3-way merge, don't give false
1326          * errors during the initial attempt.
1327          */
1328         if (state->threeway && !index_file) {
1329                 cp.no_stdout = 1;
1330                 cp.no_stderr = 1;
1331         }
1332
1333         argv_array_push(&cp.args, "apply");
1334
1335         argv_array_pushv(&cp.args, state->git_apply_opts.argv);
1336
1337         if (index_file)
1338                 argv_array_push(&cp.args, "--cached");
1339         else
1340                 argv_array_push(&cp.args, "--index");
1341
1342         argv_array_push(&cp.args, am_path(state, "patch"));
1343
1344         if (run_command(&cp))
1345                 return -1;
1346
1347         /* Reload index as git-apply will have modified it. */
1348         discard_cache();
1349         read_cache_from(index_file ? index_file : get_index_file());
1350
1351         return 0;
1352 }
1353
1354 /**
1355  * Builds an index that contains just the blobs needed for a 3way merge.
1356  */
1357 static int build_fake_ancestor(const struct am_state *state, const char *index_file)
1358 {
1359         struct child_process cp = CHILD_PROCESS_INIT;
1360
1361         cp.git_cmd = 1;
1362         argv_array_push(&cp.args, "apply");
1363         argv_array_pushv(&cp.args, state->git_apply_opts.argv);
1364         argv_array_pushf(&cp.args, "--build-fake-ancestor=%s", index_file);
1365         argv_array_push(&cp.args, am_path(state, "patch"));
1366
1367         if (run_command(&cp))
1368                 return -1;
1369
1370         return 0;
1371 }
1372
1373 /**
1374  * Attempt a threeway merge, using index_path as the temporary index.
1375  */
1376 static int fall_back_threeway(const struct am_state *state, const char *index_path)
1377 {
1378         unsigned char orig_tree[GIT_SHA1_RAWSZ], his_tree[GIT_SHA1_RAWSZ],
1379                       our_tree[GIT_SHA1_RAWSZ];
1380         const unsigned char *bases[1] = {orig_tree};
1381         struct merge_options o;
1382         struct commit *result;
1383         char *his_tree_name;
1384
1385         if (get_sha1("HEAD", our_tree) < 0)
1386                 hashcpy(our_tree, EMPTY_TREE_SHA1_BIN);
1387
1388         if (build_fake_ancestor(state, index_path))
1389                 return error("could not build fake ancestor");
1390
1391         discard_cache();
1392         read_cache_from(index_path);
1393
1394         if (write_index_as_tree(orig_tree, &the_index, index_path, 0, NULL))
1395                 return error(_("Repository lacks necessary blobs to fall back on 3-way merge."));
1396
1397         say(state, stdout, _("Using index info to reconstruct a base tree..."));
1398
1399         if (!state->quiet) {
1400                 /*
1401                  * List paths that needed 3-way fallback, so that the user can
1402                  * review them with extra care to spot mismerges.
1403                  */
1404                 struct rev_info rev_info;
1405                 const char *diff_filter_str = "--diff-filter=AM";
1406
1407                 init_revisions(&rev_info, NULL);
1408                 rev_info.diffopt.output_format = DIFF_FORMAT_NAME_STATUS;
1409                 diff_opt_parse(&rev_info.diffopt, &diff_filter_str, 1);
1410                 add_pending_sha1(&rev_info, "HEAD", our_tree, 0);
1411                 diff_setup_done(&rev_info.diffopt);
1412                 run_diff_index(&rev_info, 1);
1413         }
1414
1415         if (run_apply(state, index_path))
1416                 return error(_("Did you hand edit your patch?\n"
1417                                 "It does not apply to blobs recorded in its index."));
1418
1419         if (write_index_as_tree(his_tree, &the_index, index_path, 0, NULL))
1420                 return error("could not write tree");
1421
1422         say(state, stdout, _("Falling back to patching base and 3-way merge..."));
1423
1424         discard_cache();
1425         read_cache();
1426
1427         /*
1428          * This is not so wrong. Depending on which base we picked, orig_tree
1429          * may be wildly different from ours, but his_tree has the same set of
1430          * wildly different changes in parts the patch did not touch, so
1431          * recursive ends up canceling them, saying that we reverted all those
1432          * changes.
1433          */
1434
1435         init_merge_options(&o);
1436
1437         o.branch1 = "HEAD";
1438         his_tree_name = xstrfmt("%.*s", linelen(state->msg), state->msg);
1439         o.branch2 = his_tree_name;
1440
1441         if (state->quiet)
1442                 o.verbosity = 0;
1443
1444         if (merge_recursive_generic(&o, our_tree, his_tree, 1, bases, &result)) {
1445                 rerere(state->allow_rerere_autoupdate);
1446                 free(his_tree_name);
1447                 return error(_("Failed to merge in the changes."));
1448         }
1449
1450         free(his_tree_name);
1451         return 0;
1452 }
1453
1454 /**
1455  * Commits the current index with state->msg as the commit message and
1456  * state->author_name, state->author_email and state->author_date as the author
1457  * information.
1458  */
1459 static void do_commit(const struct am_state *state)
1460 {
1461         unsigned char tree[GIT_SHA1_RAWSZ], parent[GIT_SHA1_RAWSZ],
1462                       commit[GIT_SHA1_RAWSZ];
1463         unsigned char *ptr;
1464         struct commit_list *parents = NULL;
1465         const char *reflog_msg, *author;
1466         struct strbuf sb = STRBUF_INIT;
1467
1468         if (run_hook_le(NULL, "pre-applypatch", NULL))
1469                 exit(1);
1470
1471         if (write_cache_as_tree(tree, 0, NULL))
1472                 die(_("git write-tree failed to write a tree"));
1473
1474         if (!get_sha1_commit("HEAD", parent)) {
1475                 ptr = parent;
1476                 commit_list_insert(lookup_commit(parent), &parents);
1477         } else {
1478                 ptr = NULL;
1479                 say(state, stderr, _("applying to an empty history"));
1480         }
1481
1482         author = fmt_ident(state->author_name, state->author_email,
1483                         state->ignore_date ? NULL : state->author_date,
1484                         IDENT_STRICT);
1485
1486         if (state->committer_date_is_author_date)
1487                 setenv("GIT_COMMITTER_DATE",
1488                         state->ignore_date ? "" : state->author_date, 1);
1489
1490         if (commit_tree(state->msg, state->msg_len, tree, parents, commit,
1491                                 author, state->sign_commit))
1492                 die(_("failed to write commit object"));
1493
1494         reflog_msg = getenv("GIT_REFLOG_ACTION");
1495         if (!reflog_msg)
1496                 reflog_msg = "am";
1497
1498         strbuf_addf(&sb, "%s: %.*s", reflog_msg, linelen(state->msg),
1499                         state->msg);
1500
1501         update_ref(sb.buf, "HEAD", commit, ptr, 0, UPDATE_REFS_DIE_ON_ERR);
1502
1503         if (state->rebasing) {
1504                 FILE *fp = xfopen(am_path(state, "rewritten"), "a");
1505
1506                 assert(!is_null_sha1(state->orig_commit));
1507                 fprintf(fp, "%s ", sha1_to_hex(state->orig_commit));
1508                 fprintf(fp, "%s\n", sha1_to_hex(commit));
1509                 fclose(fp);
1510         }
1511
1512         run_hook_le(NULL, "post-applypatch", NULL);
1513
1514         strbuf_release(&sb);
1515 }
1516
1517 /**
1518  * Validates the am_state for resuming -- the "msg" and authorship fields must
1519  * be filled up.
1520  */
1521 static void validate_resume_state(const struct am_state *state)
1522 {
1523         if (!state->msg)
1524                 die(_("cannot resume: %s does not exist."),
1525                         am_path(state, "final-commit"));
1526
1527         if (!state->author_name || !state->author_email || !state->author_date)
1528                 die(_("cannot resume: %s does not exist."),
1529                         am_path(state, "author-script"));
1530 }
1531
1532 /**
1533  * Applies all queued mail.
1534  *
1535  * If `resume` is true, we are "resuming". The "msg" and authorship fields, as
1536  * well as the state directory's "patch" file is used as-is for applying the
1537  * patch and committing it.
1538  */
1539 static void am_run(struct am_state *state, int resume)
1540 {
1541         const char *argv_gc_auto[] = {"gc", "--auto", NULL};
1542         struct strbuf sb = STRBUF_INIT;
1543
1544         unlink(am_path(state, "dirtyindex"));
1545
1546         refresh_and_write_cache();
1547
1548         if (index_has_changes(&sb)) {
1549                 write_file(am_path(state, "dirtyindex"), 1, "t");
1550                 die(_("Dirty index: cannot apply patches (dirty: %s)"), sb.buf);
1551         }
1552
1553         strbuf_release(&sb);
1554
1555         while (state->cur <= state->last) {
1556                 const char *mail = am_path(state, msgnum(state));
1557                 int apply_status;
1558
1559                 if (!file_exists(mail))
1560                         goto next;
1561
1562                 if (resume) {
1563                         validate_resume_state(state);
1564                         resume = 0;
1565                 } else {
1566                         int skip;
1567
1568                         if (state->rebasing)
1569                                 skip = parse_mail_rebase(state, mail);
1570                         else
1571                                 skip = parse_mail(state, mail);
1572
1573                         if (skip)
1574                                 goto next; /* mail should be skipped */
1575
1576                         write_author_script(state);
1577                         write_commit_msg(state);
1578                 }
1579
1580                 if (run_applypatch_msg_hook(state))
1581                         exit(1);
1582
1583                 say(state, stdout, _("Applying: %.*s"), linelen(state->msg), state->msg);
1584
1585                 apply_status = run_apply(state, NULL);
1586
1587                 if (apply_status && state->threeway) {
1588                         struct strbuf sb = STRBUF_INIT;
1589
1590                         strbuf_addstr(&sb, am_path(state, "patch-merge-index"));
1591                         apply_status = fall_back_threeway(state, sb.buf);
1592                         strbuf_release(&sb);
1593
1594                         /*
1595                          * Applying the patch to an earlier tree and merging
1596                          * the result may have produced the same tree as ours.
1597                          */
1598                         if (!apply_status && !index_has_changes(NULL)) {
1599                                 say(state, stdout, _("No changes -- Patch already applied."));
1600                                 goto next;
1601                         }
1602                 }
1603
1604                 if (apply_status) {
1605                         int advice_amworkdir = 1;
1606
1607                         printf_ln(_("Patch failed at %s %.*s"), msgnum(state),
1608                                 linelen(state->msg), state->msg);
1609
1610                         git_config_get_bool("advice.amworkdir", &advice_amworkdir);
1611
1612                         if (advice_amworkdir)
1613                                 printf_ln(_("The copy of the patch that failed is found in: %s"),
1614                                                 am_path(state, "patch"));
1615
1616                         die_user_resolve(state);
1617                 }
1618
1619                 do_commit(state);
1620
1621 next:
1622                 am_next(state);
1623         }
1624
1625         if (!is_empty_file(am_path(state, "rewritten"))) {
1626                 assert(state->rebasing);
1627                 copy_notes_for_rebase(state);
1628                 run_post_rewrite_hook(state);
1629         }
1630
1631         /*
1632          * In rebasing mode, it's up to the caller to take care of
1633          * housekeeping.
1634          */
1635         if (!state->rebasing) {
1636                 am_destroy(state);
1637                 run_command_v_opt(argv_gc_auto, RUN_GIT_CMD);
1638         }
1639 }
1640
1641 /**
1642  * Resume the current am session after patch application failure. The user did
1643  * all the hard work, and we do not have to do any patch application. Just
1644  * trust and commit what the user has in the index and working tree.
1645  */
1646 static void am_resolve(struct am_state *state)
1647 {
1648         validate_resume_state(state);
1649
1650         say(state, stdout, _("Applying: %.*s"), linelen(state->msg), state->msg);
1651
1652         if (!index_has_changes(NULL)) {
1653                 printf_ln(_("No changes - did you forget to use 'git add'?\n"
1654                         "If there is nothing left to stage, chances are that something else\n"
1655                         "already introduced the same changes; you might want to skip this patch."));
1656                 die_user_resolve(state);
1657         }
1658
1659         if (unmerged_cache()) {
1660                 printf_ln(_("You still have unmerged paths in your index.\n"
1661                         "Did you forget to use 'git add'?"));
1662                 die_user_resolve(state);
1663         }
1664
1665         rerere(0);
1666
1667         do_commit(state);
1668
1669         am_next(state);
1670         am_run(state, 0);
1671 }
1672
1673 /**
1674  * Performs a checkout fast-forward from `head` to `remote`. If `reset` is
1675  * true, any unmerged entries will be discarded. Returns 0 on success, -1 on
1676  * failure.
1677  */
1678 static int fast_forward_to(struct tree *head, struct tree *remote, int reset)
1679 {
1680         struct lock_file *lock_file;
1681         struct unpack_trees_options opts;
1682         struct tree_desc t[2];
1683
1684         if (parse_tree(head) || parse_tree(remote))
1685                 return -1;
1686
1687         lock_file = xcalloc(1, sizeof(struct lock_file));
1688         hold_locked_index(lock_file, 1);
1689
1690         refresh_cache(REFRESH_QUIET);
1691
1692         memset(&opts, 0, sizeof(opts));
1693         opts.head_idx = 1;
1694         opts.src_index = &the_index;
1695         opts.dst_index = &the_index;
1696         opts.update = 1;
1697         opts.merge = 1;
1698         opts.reset = reset;
1699         opts.fn = twoway_merge;
1700         init_tree_desc(&t[0], head->buffer, head->size);
1701         init_tree_desc(&t[1], remote->buffer, remote->size);
1702
1703         if (unpack_trees(2, t, &opts)) {
1704                 rollback_lock_file(lock_file);
1705                 return -1;
1706         }
1707
1708         if (write_locked_index(&the_index, lock_file, COMMIT_LOCK))
1709                 die(_("unable to write new index file"));
1710
1711         return 0;
1712 }
1713
1714 /**
1715  * Clean the index without touching entries that are not modified between
1716  * `head` and `remote`.
1717  */
1718 static int clean_index(const unsigned char *head, const unsigned char *remote)
1719 {
1720         struct lock_file *lock_file;
1721         struct tree *head_tree, *remote_tree, *index_tree;
1722         unsigned char index[GIT_SHA1_RAWSZ];
1723         struct pathspec pathspec;
1724
1725         head_tree = parse_tree_indirect(head);
1726         if (!head_tree)
1727                 return error(_("Could not parse object '%s'."), sha1_to_hex(head));
1728
1729         remote_tree = parse_tree_indirect(remote);
1730         if (!remote_tree)
1731                 return error(_("Could not parse object '%s'."), sha1_to_hex(remote));
1732
1733         read_cache_unmerged();
1734
1735         if (fast_forward_to(head_tree, head_tree, 1))
1736                 return -1;
1737
1738         if (write_cache_as_tree(index, 0, NULL))
1739                 return -1;
1740
1741         index_tree = parse_tree_indirect(index);
1742         if (!index_tree)
1743                 return error(_("Could not parse object '%s'."), sha1_to_hex(index));
1744
1745         if (fast_forward_to(index_tree, remote_tree, 0))
1746                 return -1;
1747
1748         memset(&pathspec, 0, sizeof(pathspec));
1749
1750         lock_file = xcalloc(1, sizeof(struct lock_file));
1751         hold_locked_index(lock_file, 1);
1752
1753         if (read_tree(remote_tree, 0, &pathspec)) {
1754                 rollback_lock_file(lock_file);
1755                 return -1;
1756         }
1757
1758         if (write_locked_index(&the_index, lock_file, COMMIT_LOCK))
1759                 die(_("unable to write new index file"));
1760
1761         remove_branch_state();
1762
1763         return 0;
1764 }
1765
1766 /**
1767  * Resets rerere's merge resolution metadata.
1768  */
1769 static void am_rerere_clear(void)
1770 {
1771         struct string_list merge_rr = STRING_LIST_INIT_DUP;
1772         int fd = setup_rerere(&merge_rr, 0);
1773
1774         if (fd < 0)
1775                 return;
1776
1777         rerere_clear(&merge_rr);
1778         string_list_clear(&merge_rr, 1);
1779 }
1780
1781 /**
1782  * Resume the current am session by skipping the current patch.
1783  */
1784 static void am_skip(struct am_state *state)
1785 {
1786         unsigned char head[GIT_SHA1_RAWSZ];
1787
1788         am_rerere_clear();
1789
1790         if (get_sha1("HEAD", head))
1791                 hashcpy(head, EMPTY_TREE_SHA1_BIN);
1792
1793         if (clean_index(head, head))
1794                 die(_("failed to clean index"));
1795
1796         am_next(state);
1797         am_run(state, 0);
1798 }
1799
1800 /**
1801  * Returns true if it is safe to reset HEAD to the ORIG_HEAD, false otherwise.
1802  *
1803  * It is not safe to reset HEAD when:
1804  * 1. git-am previously failed because the index was dirty.
1805  * 2. HEAD has moved since git-am previously failed.
1806  */
1807 static int safe_to_abort(const struct am_state *state)
1808 {
1809         struct strbuf sb = STRBUF_INIT;
1810         unsigned char abort_safety[GIT_SHA1_RAWSZ], head[GIT_SHA1_RAWSZ];
1811
1812         if (file_exists(am_path(state, "dirtyindex")))
1813                 return 0;
1814
1815         if (read_state_file(&sb, state, "abort-safety", 1) > 0) {
1816                 if (get_sha1_hex(sb.buf, abort_safety))
1817                         die(_("could not parse %s"), am_path(state, "abort_safety"));
1818         } else
1819                 hashclr(abort_safety);
1820
1821         if (get_sha1("HEAD", head))
1822                 hashclr(head);
1823
1824         if (!hashcmp(head, abort_safety))
1825                 return 1;
1826
1827         error(_("You seem to have moved HEAD since the last 'am' failure.\n"
1828                 "Not rewinding to ORIG_HEAD"));
1829
1830         return 0;
1831 }
1832
1833 /**
1834  * Aborts the current am session if it is safe to do so.
1835  */
1836 static void am_abort(struct am_state *state)
1837 {
1838         unsigned char curr_head[GIT_SHA1_RAWSZ], orig_head[GIT_SHA1_RAWSZ];
1839         int has_curr_head, has_orig_head;
1840         char *curr_branch;
1841
1842         if (!safe_to_abort(state)) {
1843                 am_destroy(state);
1844                 return;
1845         }
1846
1847         am_rerere_clear();
1848
1849         curr_branch = resolve_refdup("HEAD", 0, curr_head, NULL);
1850         has_curr_head = !is_null_sha1(curr_head);
1851         if (!has_curr_head)
1852                 hashcpy(curr_head, EMPTY_TREE_SHA1_BIN);
1853
1854         has_orig_head = !get_sha1("ORIG_HEAD", orig_head);
1855         if (!has_orig_head)
1856                 hashcpy(orig_head, EMPTY_TREE_SHA1_BIN);
1857
1858         clean_index(curr_head, orig_head);
1859
1860         if (has_orig_head)
1861                 update_ref("am --abort", "HEAD", orig_head,
1862                                 has_curr_head ? curr_head : NULL, 0,
1863                                 UPDATE_REFS_DIE_ON_ERR);
1864         else if (curr_branch)
1865                 delete_ref(curr_branch, NULL, REF_NODEREF);
1866
1867         free(curr_branch);
1868         am_destroy(state);
1869 }
1870
1871 /**
1872  * parse_options() callback that validates and sets opt->value to the
1873  * PATCH_FORMAT_* enum value corresponding to `arg`.
1874  */
1875 static int parse_opt_patchformat(const struct option *opt, const char *arg, int unset)
1876 {
1877         int *opt_value = opt->value;
1878
1879         if (!strcmp(arg, "mbox"))
1880                 *opt_value = PATCH_FORMAT_MBOX;
1881         else if (!strcmp(arg, "stgit"))
1882                 *opt_value = PATCH_FORMAT_STGIT;
1883         else
1884                 return error(_("Invalid value for --patch-format: %s"), arg);
1885         return 0;
1886 }
1887
1888 enum resume_mode {
1889         RESUME_FALSE = 0,
1890         RESUME_APPLY,
1891         RESUME_RESOLVED,
1892         RESUME_SKIP,
1893         RESUME_ABORT
1894 };
1895
1896 int cmd_am(int argc, const char **argv, const char *prefix)
1897 {
1898         struct am_state state;
1899         int keep_cr = -1;
1900         int patch_format = PATCH_FORMAT_UNKNOWN;
1901         enum resume_mode resume = RESUME_FALSE;
1902
1903         const char * const usage[] = {
1904                 N_("git am [options] [(<mbox>|<Maildir>)...]"),
1905                 N_("git am [options] (--continue | --skip | --abort)"),
1906                 NULL
1907         };
1908
1909         struct option options[] = {
1910                 OPT_BOOL('3', "3way", &state.threeway,
1911                         N_("allow fall back on 3way merging if needed")),
1912                 OPT__QUIET(&state.quiet, N_("be quiet")),
1913                 OPT_BOOL('s', "signoff", &state.signoff,
1914                         N_("add a Signed-off-by line to the commit message")),
1915                 OPT_BOOL('u', "utf8", &state.utf8,
1916                         N_("recode into utf8 (default)")),
1917                 OPT_SET_INT('k', "keep", &state.keep,
1918                         N_("pass -k flag to git-mailinfo"), KEEP_TRUE),
1919                 OPT_SET_INT(0, "keep-non-patch", &state.keep,
1920                         N_("pass -b flag to git-mailinfo"), KEEP_NON_PATCH),
1921                 OPT_BOOL('m', "message-id", &state.message_id,
1922                         N_("pass -m flag to git-mailinfo")),
1923                 { OPTION_SET_INT, 0, "keep-cr", &keep_cr, NULL,
1924                   N_("pass --keep-cr flag to git-mailsplit for mbox format"),
1925                   PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1},
1926                 { OPTION_SET_INT, 0, "no-keep-cr", &keep_cr, NULL,
1927                   N_("do not pass --keep-cr flag to git-mailsplit independent of am.keepcr"),
1928                   PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 0},
1929                 OPT_BOOL('c', "scissors", &state.scissors,
1930                         N_("strip everything before a scissors line")),
1931                 OPT_PASSTHRU_ARGV(0, "whitespace", &state.git_apply_opts, N_("action"),
1932                         N_("pass it through git-apply"),
1933                         0),
1934                 OPT_PASSTHRU_ARGV(0, "ignore-space-change", &state.git_apply_opts, NULL,
1935                         N_("pass it through git-apply"),
1936                         PARSE_OPT_NOARG),
1937                 OPT_PASSTHRU_ARGV(0, "ignore-whitespace", &state.git_apply_opts, NULL,
1938                         N_("pass it through git-apply"),
1939                         PARSE_OPT_NOARG),
1940                 OPT_PASSTHRU_ARGV(0, "directory", &state.git_apply_opts, N_("root"),
1941                         N_("pass it through git-apply"),
1942                         0),
1943                 OPT_PASSTHRU_ARGV(0, "exclude", &state.git_apply_opts, N_("path"),
1944                         N_("pass it through git-apply"),
1945                         0),
1946                 OPT_PASSTHRU_ARGV(0, "include", &state.git_apply_opts, N_("path"),
1947                         N_("pass it through git-apply"),
1948                         0),
1949                 OPT_PASSTHRU_ARGV('C', NULL, &state.git_apply_opts, N_("n"),
1950                         N_("pass it through git-apply"),
1951                         0),
1952                 OPT_PASSTHRU_ARGV('p', NULL, &state.git_apply_opts, N_("num"),
1953                         N_("pass it through git-apply"),
1954                         0),
1955                 OPT_CALLBACK(0, "patch-format", &patch_format, N_("format"),
1956                         N_("format the patch(es) are in"),
1957                         parse_opt_patchformat),
1958                 OPT_PASSTHRU_ARGV(0, "reject", &state.git_apply_opts, NULL,
1959                         N_("pass it through git-apply"),
1960                         PARSE_OPT_NOARG),
1961                 OPT_STRING(0, "resolvemsg", &state.resolvemsg, NULL,
1962                         N_("override error message when patch failure occurs")),
1963                 OPT_CMDMODE(0, "continue", &resume,
1964                         N_("continue applying patches after resolving a conflict"),
1965                         RESUME_RESOLVED),
1966                 OPT_CMDMODE('r', "resolved", &resume,
1967                         N_("synonyms for --continue"),
1968                         RESUME_RESOLVED),
1969                 OPT_CMDMODE(0, "skip", &resume,
1970                         N_("skip the current patch"),
1971                         RESUME_SKIP),
1972                 OPT_CMDMODE(0, "abort", &resume,
1973                         N_("restore the original branch and abort the patching operation."),
1974                         RESUME_ABORT),
1975                 OPT_BOOL(0, "committer-date-is-author-date",
1976                         &state.committer_date_is_author_date,
1977                         N_("lie about committer date")),
1978                 OPT_BOOL(0, "ignore-date", &state.ignore_date,
1979                         N_("use current timestamp for author date")),
1980                 OPT_RERERE_AUTOUPDATE(&state.allow_rerere_autoupdate),
1981                 { OPTION_STRING, 'S', "gpg-sign", &state.sign_commit, N_("key-id"),
1982                   N_("GPG-sign commits"),
1983                   PARSE_OPT_OPTARG, NULL, (intptr_t) "" },
1984                 OPT_HIDDEN_BOOL(0, "rebasing", &state.rebasing,
1985                         N_("(internal use for git-rebase)")),
1986                 OPT_END()
1987         };
1988
1989         /*
1990          * NEEDSWORK: Once all the features of git-am.sh have been
1991          * re-implemented in builtin/am.c, this preamble can be removed.
1992          */
1993         if (!getenv("_GIT_USE_BUILTIN_AM")) {
1994                 const char *path = mkpath("%s/git-am", git_exec_path());
1995
1996                 if (sane_execvp(path, (char **)argv) < 0)
1997                         die_errno("could not exec %s", path);
1998         } else {
1999                 prefix = setup_git_directory();
2000                 trace_repo_setup(prefix);
2001                 setup_work_tree();
2002         }
2003
2004         git_config(git_default_config, NULL);
2005
2006         am_state_init(&state, git_path("rebase-apply"));
2007
2008         argc = parse_options(argc, argv, prefix, options, usage, 0);
2009
2010         if (read_index_preload(&the_index, NULL) < 0)
2011                 die(_("failed to read the index"));
2012
2013         if (am_in_progress(&state)) {
2014                 /*
2015                  * Catch user error to feed us patches when there is a session
2016                  * in progress:
2017                  *
2018                  * 1. mbox path(s) are provided on the command-line.
2019                  * 2. stdin is not a tty: the user is trying to feed us a patch
2020                  *    from standard input. This is somewhat unreliable -- stdin
2021                  *    could be /dev/null for example and the caller did not
2022                  *    intend to feed us a patch but wanted to continue
2023                  *    unattended.
2024                  */
2025                 if (argc || (resume == RESUME_FALSE && !isatty(0)))
2026                         die(_("previous rebase directory %s still exists but mbox given."),
2027                                 state.dir);
2028
2029                 if (resume == RESUME_FALSE)
2030                         resume = RESUME_APPLY;
2031
2032                 am_load(&state);
2033         } else {
2034                 struct argv_array paths = ARGV_ARRAY_INIT;
2035                 int i;
2036
2037                 /*
2038                  * Handle stray state directory in the independent-run case. In
2039                  * the --rebasing case, it is up to the caller to take care of
2040                  * stray directories.
2041                  */
2042                 if (file_exists(state.dir) && !state.rebasing) {
2043                         if (resume == RESUME_ABORT) {
2044                                 am_destroy(&state);
2045                                 am_state_release(&state);
2046                                 return 0;
2047                         }
2048
2049                         die(_("Stray %s directory found.\n"
2050                                 "Use \"git am --abort\" to remove it."),
2051                                 state.dir);
2052                 }
2053
2054                 if (resume)
2055                         die(_("Resolve operation not in progress, we are not resuming."));
2056
2057                 for (i = 0; i < argc; i++) {
2058                         if (is_absolute_path(argv[i]) || !prefix)
2059                                 argv_array_push(&paths, argv[i]);
2060                         else
2061                                 argv_array_push(&paths, mkpath("%s/%s", prefix, argv[i]));
2062                 }
2063
2064                 am_setup(&state, patch_format, paths.argv, keep_cr);
2065
2066                 argv_array_clear(&paths);
2067         }
2068
2069         switch (resume) {
2070         case RESUME_FALSE:
2071                 am_run(&state, 0);
2072                 break;
2073         case RESUME_APPLY:
2074                 am_run(&state, 1);
2075                 break;
2076         case RESUME_RESOLVED:
2077                 am_resolve(&state);
2078                 break;
2079         case RESUME_SKIP:
2080                 am_skip(&state);
2081                 break;
2082         case RESUME_ABORT:
2083                 am_abort(&state);
2084                 break;
2085         default:
2086                 die("BUG: invalid resume value");
2087         }
2088
2089         am_state_release(&state);
2090
2091         return 0;
2092 }