Merge branch 'tb/commit-graph-fd-exhaustion-fix'
[git] / pretty.c
1 #include "cache.h"
2 #include "config.h"
3 #include "commit.h"
4 #include "utf8.h"
5 #include "diff.h"
6 #include "revision.h"
7 #include "string-list.h"
8 #include "mailmap.h"
9 #include "log-tree.h"
10 #include "notes.h"
11 #include "color.h"
12 #include "reflog-walk.h"
13 #include "gpg-interface.h"
14 #include "trailer.h"
15
16 static char *user_format;
17 static struct cmt_fmt_map {
18         const char *name;
19         enum cmit_fmt format;
20         int is_tformat;
21         int expand_tabs_in_log;
22         int is_alias;
23         enum date_mode_type default_date_mode_type;
24         const char *user_format;
25 } *commit_formats;
26 static size_t builtin_formats_len;
27 static size_t commit_formats_len;
28 static size_t commit_formats_alloc;
29 static struct cmt_fmt_map *find_commit_format(const char *sought);
30
31 int commit_format_is_empty(enum cmit_fmt fmt)
32 {
33         return fmt == CMIT_FMT_USERFORMAT && !*user_format;
34 }
35
36 static void save_user_format(struct rev_info *rev, const char *cp, int is_tformat)
37 {
38         free(user_format);
39         user_format = xstrdup(cp);
40         if (is_tformat)
41                 rev->use_terminator = 1;
42         rev->commit_format = CMIT_FMT_USERFORMAT;
43 }
44
45 static int git_pretty_formats_config(const char *var, const char *value, void *cb)
46 {
47         struct cmt_fmt_map *commit_format = NULL;
48         const char *name;
49         const char *fmt;
50         int i;
51
52         if (!skip_prefix(var, "pretty.", &name))
53                 return 0;
54
55         for (i = 0; i < builtin_formats_len; i++) {
56                 if (!strcmp(commit_formats[i].name, name))
57                         return 0;
58         }
59
60         for (i = builtin_formats_len; i < commit_formats_len; i++) {
61                 if (!strcmp(commit_formats[i].name, name)) {
62                         commit_format = &commit_formats[i];
63                         break;
64                 }
65         }
66
67         if (!commit_format) {
68                 ALLOC_GROW(commit_formats, commit_formats_len+1,
69                            commit_formats_alloc);
70                 commit_format = &commit_formats[commit_formats_len];
71                 memset(commit_format, 0, sizeof(*commit_format));
72                 commit_formats_len++;
73         }
74
75         commit_format->name = xstrdup(name);
76         commit_format->format = CMIT_FMT_USERFORMAT;
77         if (git_config_string(&fmt, var, value))
78                 return -1;
79
80         if (skip_prefix(fmt, "format:", &fmt))
81                 commit_format->is_tformat = 0;
82         else if (skip_prefix(fmt, "tformat:", &fmt) || strchr(fmt, '%'))
83                 commit_format->is_tformat = 1;
84         else
85                 commit_format->is_alias = 1;
86         commit_format->user_format = fmt;
87
88         return 0;
89 }
90
91 static void setup_commit_formats(void)
92 {
93         struct cmt_fmt_map builtin_formats[] = {
94                 { "raw",        CMIT_FMT_RAW,           0,      0 },
95                 { "medium",     CMIT_FMT_MEDIUM,        0,      8 },
96                 { "short",      CMIT_FMT_SHORT,         0,      0 },
97                 { "email",      CMIT_FMT_EMAIL,         0,      0 },
98                 { "mboxrd",     CMIT_FMT_MBOXRD,        0,      0 },
99                 { "fuller",     CMIT_FMT_FULLER,        0,      8 },
100                 { "full",       CMIT_FMT_FULL,          0,      8 },
101                 { "oneline",    CMIT_FMT_ONELINE,       1,      0 },
102                 { "reference",  CMIT_FMT_USERFORMAT,    1,      0,
103                         0, DATE_SHORT, "%C(auto)%h (%s, %ad)" },
104                 /*
105                  * Please update $__git_log_pretty_formats in
106                  * git-completion.bash when you add new formats.
107                  */
108         };
109         commit_formats_len = ARRAY_SIZE(builtin_formats);
110         builtin_formats_len = commit_formats_len;
111         ALLOC_GROW(commit_formats, commit_formats_len, commit_formats_alloc);
112         COPY_ARRAY(commit_formats, builtin_formats,
113                    ARRAY_SIZE(builtin_formats));
114
115         git_config(git_pretty_formats_config, NULL);
116 }
117
118 static struct cmt_fmt_map *find_commit_format_recursive(const char *sought,
119                                                         const char *original,
120                                                         int num_redirections)
121 {
122         struct cmt_fmt_map *found = NULL;
123         size_t found_match_len = 0;
124         int i;
125
126         if (num_redirections >= commit_formats_len)
127                 die("invalid --pretty format: "
128                     "'%s' references an alias which points to itself",
129                     original);
130
131         for (i = 0; i < commit_formats_len; i++) {
132                 size_t match_len;
133
134                 if (!starts_with(commit_formats[i].name, sought))
135                         continue;
136
137                 match_len = strlen(commit_formats[i].name);
138                 if (found == NULL || found_match_len > match_len) {
139                         found = &commit_formats[i];
140                         found_match_len = match_len;
141                 }
142         }
143
144         if (found && found->is_alias) {
145                 found = find_commit_format_recursive(found->user_format,
146                                                      original,
147                                                      num_redirections+1);
148         }
149
150         return found;
151 }
152
153 static struct cmt_fmt_map *find_commit_format(const char *sought)
154 {
155         if (!commit_formats)
156                 setup_commit_formats();
157
158         return find_commit_format_recursive(sought, sought, 0);
159 }
160
161 void get_commit_format(const char *arg, struct rev_info *rev)
162 {
163         struct cmt_fmt_map *commit_format;
164
165         rev->use_terminator = 0;
166         if (!arg) {
167                 rev->commit_format = CMIT_FMT_DEFAULT;
168                 return;
169         }
170         if (skip_prefix(arg, "format:", &arg)) {
171                 save_user_format(rev, arg, 0);
172                 return;
173         }
174
175         if (!*arg || skip_prefix(arg, "tformat:", &arg) || strchr(arg, '%')) {
176                 save_user_format(rev, arg, 1);
177                 return;
178         }
179
180         commit_format = find_commit_format(arg);
181         if (!commit_format)
182                 die("invalid --pretty format: %s", arg);
183
184         rev->commit_format = commit_format->format;
185         rev->use_terminator = commit_format->is_tformat;
186         rev->expand_tabs_in_log_default = commit_format->expand_tabs_in_log;
187         if (!rev->date_mode_explicit && commit_format->default_date_mode_type)
188                 rev->date_mode.type = commit_format->default_date_mode_type;
189         if (commit_format->format == CMIT_FMT_USERFORMAT) {
190                 save_user_format(rev, commit_format->user_format,
191                                  commit_format->is_tformat);
192         }
193 }
194
195 /*
196  * Generic support for pretty-printing the header
197  */
198 static int get_one_line(const char *msg)
199 {
200         int ret = 0;
201
202         for (;;) {
203                 char c = *msg++;
204                 if (!c)
205                         break;
206                 ret++;
207                 if (c == '\n')
208                         break;
209         }
210         return ret;
211 }
212
213 /* High bit set, or ISO-2022-INT */
214 static int non_ascii(int ch)
215 {
216         return !isascii(ch) || ch == '\033';
217 }
218
219 int has_non_ascii(const char *s)
220 {
221         int ch;
222         if (!s)
223                 return 0;
224         while ((ch = *s++) != '\0') {
225                 if (non_ascii(ch))
226                         return 1;
227         }
228         return 0;
229 }
230
231 static int is_rfc822_special(char ch)
232 {
233         switch (ch) {
234         case '(':
235         case ')':
236         case '<':
237         case '>':
238         case '[':
239         case ']':
240         case ':':
241         case ';':
242         case '@':
243         case ',':
244         case '.':
245         case '"':
246         case '\\':
247                 return 1;
248         default:
249                 return 0;
250         }
251 }
252
253 static int needs_rfc822_quoting(const char *s, int len)
254 {
255         int i;
256         for (i = 0; i < len; i++)
257                 if (is_rfc822_special(s[i]))
258                         return 1;
259         return 0;
260 }
261
262 static int last_line_length(struct strbuf *sb)
263 {
264         int i;
265
266         /* How many bytes are already used on the last line? */
267         for (i = sb->len - 1; i >= 0; i--)
268                 if (sb->buf[i] == '\n')
269                         break;
270         return sb->len - (i + 1);
271 }
272
273 static void add_rfc822_quoted(struct strbuf *out, const char *s, int len)
274 {
275         int i;
276
277         /* just a guess, we may have to also backslash-quote */
278         strbuf_grow(out, len + 2);
279
280         strbuf_addch(out, '"');
281         for (i = 0; i < len; i++) {
282                 switch (s[i]) {
283                 case '"':
284                 case '\\':
285                         strbuf_addch(out, '\\');
286                         /* fall through */
287                 default:
288                         strbuf_addch(out, s[i]);
289                 }
290         }
291         strbuf_addch(out, '"');
292 }
293
294 enum rfc2047_type {
295         RFC2047_SUBJECT,
296         RFC2047_ADDRESS
297 };
298
299 static int is_rfc2047_special(char ch, enum rfc2047_type type)
300 {
301         /*
302          * rfc2047, section 4.2:
303          *
304          *    8-bit values which correspond to printable ASCII characters other
305          *    than "=", "?", and "_" (underscore), MAY be represented as those
306          *    characters.  (But see section 5 for restrictions.)  In
307          *    particular, SPACE and TAB MUST NOT be represented as themselves
308          *    within encoded words.
309          */
310
311         /*
312          * rule out non-ASCII characters and non-printable characters (the
313          * non-ASCII check should be redundant as isprint() is not localized
314          * and only knows about ASCII, but be defensive about that)
315          */
316         if (non_ascii(ch) || !isprint(ch))
317                 return 1;
318
319         /*
320          * rule out special printable characters (' ' should be the only
321          * whitespace character considered printable, but be defensive and use
322          * isspace())
323          */
324         if (isspace(ch) || ch == '=' || ch == '?' || ch == '_')
325                 return 1;
326
327         /*
328          * rfc2047, section 5.3:
329          *
330          *    As a replacement for a 'word' entity within a 'phrase', for example,
331          *    one that precedes an address in a From, To, or Cc header.  The ABNF
332          *    definition for 'phrase' from RFC 822 thus becomes:
333          *
334          *    phrase = 1*( encoded-word / word )
335          *
336          *    In this case the set of characters that may be used in a "Q"-encoded
337          *    'encoded-word' is restricted to: <upper and lower case ASCII
338          *    letters, decimal digits, "!", "*", "+", "-", "/", "=", and "_"
339          *    (underscore, ASCII 95.)>.  An 'encoded-word' that appears within a
340          *    'phrase' MUST be separated from any adjacent 'word', 'text' or
341          *    'special' by 'linear-white-space'.
342          */
343
344         if (type != RFC2047_ADDRESS)
345                 return 0;
346
347         /* '=' and '_' are special cases and have been checked above */
348         return !(isalnum(ch) || ch == '!' || ch == '*' || ch == '+' || ch == '-' || ch == '/');
349 }
350
351 static int needs_rfc2047_encoding(const char *line, int len)
352 {
353         int i;
354
355         for (i = 0; i < len; i++) {
356                 int ch = line[i];
357                 if (non_ascii(ch) || ch == '\n')
358                         return 1;
359                 if ((i + 1 < len) && (ch == '=' && line[i+1] == '?'))
360                         return 1;
361         }
362
363         return 0;
364 }
365
366 static void add_rfc2047(struct strbuf *sb, const char *line, size_t len,
367                        const char *encoding, enum rfc2047_type type)
368 {
369         static const int max_encoded_length = 76; /* per rfc2047 */
370         int i;
371         int line_len = last_line_length(sb);
372
373         strbuf_grow(sb, len * 3 + strlen(encoding) + 100);
374         strbuf_addf(sb, "=?%s?q?", encoding);
375         line_len += strlen(encoding) + 5; /* 5 for =??q? */
376
377         while (len) {
378                 /*
379                  * RFC 2047, section 5 (3):
380                  *
381                  * Each 'encoded-word' MUST represent an integral number of
382                  * characters.  A multi-octet character may not be split across
383                  * adjacent 'encoded- word's.
384                  */
385                 const unsigned char *p = (const unsigned char *)line;
386                 int chrlen = mbs_chrlen(&line, &len, encoding);
387                 int is_special = (chrlen > 1) || is_rfc2047_special(*p, type);
388
389                 /* "=%02X" * chrlen, or the byte itself */
390                 const char *encoded_fmt = is_special ? "=%02X"    : "%c";
391                 int         encoded_len = is_special ? 3 * chrlen : 1;
392
393                 /*
394                  * According to RFC 2047, we could encode the special character
395                  * ' ' (space) with '_' (underscore) for readability. But many
396                  * programs do not understand this and just leave the
397                  * underscore in place. Thus, we do nothing special here, which
398                  * causes ' ' to be encoded as '=20', avoiding this problem.
399                  */
400
401                 if (line_len + encoded_len + 2 > max_encoded_length) {
402                         /* It won't fit with trailing "?=" --- break the line */
403                         strbuf_addf(sb, "?=\n =?%s?q?", encoding);
404                         line_len = strlen(encoding) + 5 + 1; /* =??q? plus SP */
405                 }
406
407                 for (i = 0; i < chrlen; i++)
408                         strbuf_addf(sb, encoded_fmt, p[i]);
409                 line_len += encoded_len;
410         }
411         strbuf_addstr(sb, "?=");
412 }
413
414 const char *show_ident_date(const struct ident_split *ident,
415                             const struct date_mode *mode)
416 {
417         timestamp_t date = 0;
418         long tz = 0;
419
420         if (ident->date_begin && ident->date_end)
421                 date = parse_timestamp(ident->date_begin, NULL, 10);
422         if (date_overflows(date))
423                 date = 0;
424         else {
425                 if (ident->tz_begin && ident->tz_end)
426                         tz = strtol(ident->tz_begin, NULL, 10);
427                 if (tz >= INT_MAX || tz <= INT_MIN)
428                         tz = 0;
429         }
430         return show_date(date, tz, mode);
431 }
432
433 void pp_user_info(struct pretty_print_context *pp,
434                   const char *what, struct strbuf *sb,
435                   const char *line, const char *encoding)
436 {
437         struct ident_split ident;
438         char *line_end;
439         const char *mailbuf, *namebuf;
440         size_t namelen, maillen;
441         int max_length = 78; /* per rfc2822 */
442
443         if (pp->fmt == CMIT_FMT_ONELINE)
444                 return;
445
446         line_end = strchrnul(line, '\n');
447         if (split_ident_line(&ident, line, line_end - line))
448                 return;
449
450         mailbuf = ident.mail_begin;
451         maillen = ident.mail_end - ident.mail_begin;
452         namebuf = ident.name_begin;
453         namelen = ident.name_end - ident.name_begin;
454
455         if (pp->mailmap)
456                 map_user(pp->mailmap, &mailbuf, &maillen, &namebuf, &namelen);
457
458         if (cmit_fmt_is_mail(pp->fmt)) {
459                 if (pp->from_ident && ident_cmp(pp->from_ident, &ident)) {
460                         struct strbuf buf = STRBUF_INIT;
461
462                         strbuf_addstr(&buf, "From: ");
463                         strbuf_add(&buf, namebuf, namelen);
464                         strbuf_addstr(&buf, " <");
465                         strbuf_add(&buf, mailbuf, maillen);
466                         strbuf_addstr(&buf, ">\n");
467                         string_list_append(&pp->in_body_headers,
468                                            strbuf_detach(&buf, NULL));
469
470                         mailbuf = pp->from_ident->mail_begin;
471                         maillen = pp->from_ident->mail_end - mailbuf;
472                         namebuf = pp->from_ident->name_begin;
473                         namelen = pp->from_ident->name_end - namebuf;
474                 }
475
476                 strbuf_addstr(sb, "From: ");
477                 if (pp->encode_email_headers &&
478                     needs_rfc2047_encoding(namebuf, namelen)) {
479                         add_rfc2047(sb, namebuf, namelen,
480                                     encoding, RFC2047_ADDRESS);
481                         max_length = 76; /* per rfc2047 */
482                 } else if (needs_rfc822_quoting(namebuf, namelen)) {
483                         struct strbuf quoted = STRBUF_INIT;
484                         add_rfc822_quoted(&quoted, namebuf, namelen);
485                         strbuf_add_wrapped_bytes(sb, quoted.buf, quoted.len,
486                                                         -6, 1, max_length);
487                         strbuf_release(&quoted);
488                 } else {
489                         strbuf_add_wrapped_bytes(sb, namebuf, namelen,
490                                                  -6, 1, max_length);
491                 }
492
493                 if (max_length <
494                     last_line_length(sb) + strlen(" <") + maillen + strlen(">"))
495                         strbuf_addch(sb, '\n');
496                 strbuf_addf(sb, " <%.*s>\n", (int)maillen, mailbuf);
497         } else {
498                 strbuf_addf(sb, "%s: %.*s%.*s <%.*s>\n", what,
499                             (pp->fmt == CMIT_FMT_FULLER) ? 4 : 0, "    ",
500                             (int)namelen, namebuf, (int)maillen, mailbuf);
501         }
502
503         switch (pp->fmt) {
504         case CMIT_FMT_MEDIUM:
505                 strbuf_addf(sb, "Date:   %s\n",
506                             show_ident_date(&ident, &pp->date_mode));
507                 break;
508         case CMIT_FMT_EMAIL:
509         case CMIT_FMT_MBOXRD:
510                 strbuf_addf(sb, "Date: %s\n",
511                             show_ident_date(&ident, DATE_MODE(RFC2822)));
512                 break;
513         case CMIT_FMT_FULLER:
514                 strbuf_addf(sb, "%sDate: %s\n", what,
515                             show_ident_date(&ident, &pp->date_mode));
516                 break;
517         default:
518                 /* notin' */
519                 break;
520         }
521 }
522
523 static int is_blank_line(const char *line, int *len_p)
524 {
525         int len = *len_p;
526         while (len && isspace(line[len - 1]))
527                 len--;
528         *len_p = len;
529         return !len;
530 }
531
532 const char *skip_blank_lines(const char *msg)
533 {
534         for (;;) {
535                 int linelen = get_one_line(msg);
536                 int ll = linelen;
537                 if (!linelen)
538                         break;
539                 if (!is_blank_line(msg, &ll))
540                         break;
541                 msg += linelen;
542         }
543         return msg;
544 }
545
546 static void add_merge_info(const struct pretty_print_context *pp,
547                            struct strbuf *sb, const struct commit *commit)
548 {
549         struct commit_list *parent = commit->parents;
550
551         if ((pp->fmt == CMIT_FMT_ONELINE) || (cmit_fmt_is_mail(pp->fmt)) ||
552             !parent || !parent->next)
553                 return;
554
555         strbuf_addstr(sb, "Merge:");
556
557         while (parent) {
558                 struct object_id *oidp = &parent->item->object.oid;
559                 strbuf_addch(sb, ' ');
560                 if (pp->abbrev)
561                         strbuf_add_unique_abbrev(sb, oidp, pp->abbrev);
562                 else
563                         strbuf_addstr(sb, oid_to_hex(oidp));
564                 parent = parent->next;
565         }
566         strbuf_addch(sb, '\n');
567 }
568
569 static char *get_header(const char *msg, const char *key)
570 {
571         size_t len;
572         const char *v = find_commit_header(msg, key, &len);
573         return v ? xmemdupz(v, len) : NULL;
574 }
575
576 static char *replace_encoding_header(char *buf, const char *encoding)
577 {
578         struct strbuf tmp = STRBUF_INIT;
579         size_t start, len;
580         char *cp = buf;
581
582         /* guess if there is an encoding header before a \n\n */
583         while (!starts_with(cp, "encoding ")) {
584                 cp = strchr(cp, '\n');
585                 if (!cp || *++cp == '\n')
586                         return buf;
587         }
588         start = cp - buf;
589         cp = strchr(cp, '\n');
590         if (!cp)
591                 return buf; /* should not happen but be defensive */
592         len = cp + 1 - (buf + start);
593
594         strbuf_attach(&tmp, buf, strlen(buf), strlen(buf) + 1);
595         if (is_encoding_utf8(encoding)) {
596                 /* we have re-coded to UTF-8; drop the header */
597                 strbuf_remove(&tmp, start, len);
598         } else {
599                 /* just replaces XXXX in 'encoding XXXX\n' */
600                 strbuf_splice(&tmp, start + strlen("encoding "),
601                                           len - strlen("encoding \n"),
602                                           encoding, strlen(encoding));
603         }
604         return strbuf_detach(&tmp, NULL);
605 }
606
607 const char *repo_logmsg_reencode(struct repository *r,
608                                  const struct commit *commit,
609                                  char **commit_encoding,
610                                  const char *output_encoding)
611 {
612         static const char *utf8 = "UTF-8";
613         const char *use_encoding;
614         char *encoding;
615         const char *msg = repo_get_commit_buffer(r, commit, NULL);
616         char *out;
617
618         if (!output_encoding || !*output_encoding) {
619                 if (commit_encoding)
620                         *commit_encoding = get_header(msg, "encoding");
621                 return msg;
622         }
623         encoding = get_header(msg, "encoding");
624         if (commit_encoding)
625                 *commit_encoding = encoding;
626         use_encoding = encoding ? encoding : utf8;
627         if (same_encoding(use_encoding, output_encoding)) {
628                 /*
629                  * No encoding work to be done. If we have no encoding header
630                  * at all, then there's nothing to do, and we can return the
631                  * message verbatim (whether newly allocated or not).
632                  */
633                 if (!encoding)
634                         return msg;
635
636                 /*
637                  * Otherwise, we still want to munge the encoding header in the
638                  * result, which will be done by modifying the buffer. If we
639                  * are using a fresh copy, we can reuse it. But if we are using
640                  * the cached copy from get_commit_buffer, we need to duplicate it
641                  * to avoid munging the cached copy.
642                  */
643                 if (msg == get_cached_commit_buffer(r, commit, NULL))
644                         out = xstrdup(msg);
645                 else
646                         out = (char *)msg;
647         }
648         else {
649                 /*
650                  * There's actual encoding work to do. Do the reencoding, which
651                  * still leaves the header to be replaced in the next step. At
652                  * this point, we are done with msg. If we allocated a fresh
653                  * copy, we can free it.
654                  */
655                 out = reencode_string(msg, output_encoding, use_encoding);
656                 if (out)
657                         repo_unuse_commit_buffer(r, commit, msg);
658         }
659
660         /*
661          * This replacement actually consumes the buffer we hand it, so we do
662          * not have to worry about freeing the old "out" here.
663          */
664         if (out)
665                 out = replace_encoding_header(out, output_encoding);
666
667         if (!commit_encoding)
668                 free(encoding);
669         /*
670          * If the re-encoding failed, out might be NULL here; in that
671          * case we just return the commit message verbatim.
672          */
673         return out ? out : msg;
674 }
675
676 static int mailmap_name(const char **email, size_t *email_len,
677                         const char **name, size_t *name_len)
678 {
679         static struct string_list *mail_map;
680         if (!mail_map) {
681                 mail_map = xcalloc(1, sizeof(*mail_map));
682                 read_mailmap(mail_map, NULL);
683         }
684         return mail_map->nr && map_user(mail_map, email, email_len, name, name_len);
685 }
686
687 static size_t format_person_part(struct strbuf *sb, char part,
688                                  const char *msg, int len,
689                                  const struct date_mode *dmode)
690 {
691         /* currently all placeholders have same length */
692         const int placeholder_len = 2;
693         struct ident_split s;
694         const char *name, *mail;
695         size_t maillen, namelen;
696
697         if (split_ident_line(&s, msg, len) < 0)
698                 goto skip;
699
700         name = s.name_begin;
701         namelen = s.name_end - s.name_begin;
702         mail = s.mail_begin;
703         maillen = s.mail_end - s.mail_begin;
704
705         if (part == 'N' || part == 'E' || part == 'L') /* mailmap lookup */
706                 mailmap_name(&mail, &maillen, &name, &namelen);
707         if (part == 'n' || part == 'N') {       /* name */
708                 strbuf_add(sb, name, namelen);
709                 return placeholder_len;
710         }
711         if (part == 'e' || part == 'E') {       /* email */
712                 strbuf_add(sb, mail, maillen);
713                 return placeholder_len;
714         }
715         if (part == 'l' || part == 'L') {       /* local-part */
716                 const char *at = memchr(mail, '@', maillen);
717                 if (at)
718                         maillen = at - mail;
719                 strbuf_add(sb, mail, maillen);
720                 return placeholder_len;
721         }
722
723         if (!s.date_begin)
724                 goto skip;
725
726         if (part == 't') {      /* date, UNIX timestamp */
727                 strbuf_add(sb, s.date_begin, s.date_end - s.date_begin);
728                 return placeholder_len;
729         }
730
731         switch (part) {
732         case 'd':       /* date */
733                 strbuf_addstr(sb, show_ident_date(&s, dmode));
734                 return placeholder_len;
735         case 'D':       /* date, RFC2822 style */
736                 strbuf_addstr(sb, show_ident_date(&s, DATE_MODE(RFC2822)));
737                 return placeholder_len;
738         case 'r':       /* date, relative */
739                 strbuf_addstr(sb, show_ident_date(&s, DATE_MODE(RELATIVE)));
740                 return placeholder_len;
741         case 'i':       /* date, ISO 8601-like */
742                 strbuf_addstr(sb, show_ident_date(&s, DATE_MODE(ISO8601)));
743                 return placeholder_len;
744         case 'I':       /* date, ISO 8601 strict */
745                 strbuf_addstr(sb, show_ident_date(&s, DATE_MODE(ISO8601_STRICT)));
746                 return placeholder_len;
747         case 's':
748                 strbuf_addstr(sb, show_ident_date(&s, DATE_MODE(SHORT)));
749                 return placeholder_len;
750         }
751
752 skip:
753         /*
754          * reading from either a bogus commit, or a reflog entry with
755          * %gn, %ge, etc.; 'sb' cannot be updated, but we still need
756          * to compute a valid return value.
757          */
758         if (part == 'n' || part == 'e' || part == 't' || part == 'd'
759             || part == 'D' || part == 'r' || part == 'i')
760                 return placeholder_len;
761
762         return 0; /* unknown placeholder */
763 }
764
765 struct chunk {
766         size_t off;
767         size_t len;
768 };
769
770 enum flush_type {
771         no_flush,
772         flush_right,
773         flush_left,
774         flush_left_and_steal,
775         flush_both
776 };
777
778 enum trunc_type {
779         trunc_none,
780         trunc_left,
781         trunc_middle,
782         trunc_right
783 };
784
785 struct format_commit_context {
786         const struct commit *commit;
787         const struct pretty_print_context *pretty_ctx;
788         unsigned commit_header_parsed:1;
789         unsigned commit_message_parsed:1;
790         struct signature_check signature_check;
791         enum flush_type flush_type;
792         enum trunc_type truncate;
793         const char *message;
794         char *commit_encoding;
795         size_t width, indent1, indent2;
796         int auto_color;
797         int padding;
798
799         /* These offsets are relative to the start of the commit message. */
800         struct chunk author;
801         struct chunk committer;
802         size_t message_off;
803         size_t subject_off;
804         size_t body_off;
805
806         /* The following ones are relative to the result struct strbuf. */
807         size_t wrap_start;
808 };
809
810 static void parse_commit_header(struct format_commit_context *context)
811 {
812         const char *msg = context->message;
813         int i;
814
815         for (i = 0; msg[i]; i++) {
816                 const char *name;
817                 int eol;
818                 for (eol = i; msg[eol] && msg[eol] != '\n'; eol++)
819                         ; /* do nothing */
820
821                 if (i == eol) {
822                         break;
823                 } else if (skip_prefix(msg + i, "author ", &name)) {
824                         context->author.off = name - msg;
825                         context->author.len = msg + eol - name;
826                 } else if (skip_prefix(msg + i, "committer ", &name)) {
827                         context->committer.off = name - msg;
828                         context->committer.len = msg + eol - name;
829                 }
830                 i = eol;
831         }
832         context->message_off = i;
833         context->commit_header_parsed = 1;
834 }
835
836 static int istitlechar(char c)
837 {
838         return (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') ||
839                 (c >= '0' && c <= '9') || c == '.' || c == '_';
840 }
841
842 static void format_sanitized_subject(struct strbuf *sb, const char *msg)
843 {
844         size_t trimlen;
845         size_t start_len = sb->len;
846         int space = 2;
847
848         for (; *msg && *msg != '\n'; msg++) {
849                 if (istitlechar(*msg)) {
850                         if (space == 1)
851                                 strbuf_addch(sb, '-');
852                         space = 0;
853                         strbuf_addch(sb, *msg);
854                         if (*msg == '.')
855                                 while (*(msg+1) == '.')
856                                         msg++;
857                 } else
858                         space |= 1;
859         }
860
861         /* trim any trailing '.' or '-' characters */
862         trimlen = 0;
863         while (sb->len - trimlen > start_len &&
864                 (sb->buf[sb->len - 1 - trimlen] == '.'
865                 || sb->buf[sb->len - 1 - trimlen] == '-'))
866                 trimlen++;
867         strbuf_remove(sb, sb->len - trimlen, trimlen);
868 }
869
870 const char *format_subject(struct strbuf *sb, const char *msg,
871                            const char *line_separator)
872 {
873         int first = 1;
874
875         for (;;) {
876                 const char *line = msg;
877                 int linelen = get_one_line(line);
878
879                 msg += linelen;
880                 if (!linelen || is_blank_line(line, &linelen))
881                         break;
882
883                 if (!sb)
884                         continue;
885                 strbuf_grow(sb, linelen + 2);
886                 if (!first)
887                         strbuf_addstr(sb, line_separator);
888                 strbuf_add(sb, line, linelen);
889                 first = 0;
890         }
891         return msg;
892 }
893
894 static void parse_commit_message(struct format_commit_context *c)
895 {
896         const char *msg = c->message + c->message_off;
897         const char *start = c->message;
898
899         msg = skip_blank_lines(msg);
900         c->subject_off = msg - start;
901
902         msg = format_subject(NULL, msg, NULL);
903         msg = skip_blank_lines(msg);
904         c->body_off = msg - start;
905
906         c->commit_message_parsed = 1;
907 }
908
909 static void strbuf_wrap(struct strbuf *sb, size_t pos,
910                         size_t width, size_t indent1, size_t indent2)
911 {
912         struct strbuf tmp = STRBUF_INIT;
913
914         if (pos)
915                 strbuf_add(&tmp, sb->buf, pos);
916         strbuf_add_wrapped_text(&tmp, sb->buf + pos,
917                                 (int) indent1, (int) indent2, (int) width);
918         strbuf_swap(&tmp, sb);
919         strbuf_release(&tmp);
920 }
921
922 static void rewrap_message_tail(struct strbuf *sb,
923                                 struct format_commit_context *c,
924                                 size_t new_width, size_t new_indent1,
925                                 size_t new_indent2)
926 {
927         if (c->width == new_width && c->indent1 == new_indent1 &&
928             c->indent2 == new_indent2)
929                 return;
930         if (c->wrap_start < sb->len)
931                 strbuf_wrap(sb, c->wrap_start, c->width, c->indent1, c->indent2);
932         c->wrap_start = sb->len;
933         c->width = new_width;
934         c->indent1 = new_indent1;
935         c->indent2 = new_indent2;
936 }
937
938 static int format_reflog_person(struct strbuf *sb,
939                                 char part,
940                                 struct reflog_walk_info *log,
941                                 const struct date_mode *dmode)
942 {
943         const char *ident;
944
945         if (!log)
946                 return 2;
947
948         ident = get_reflog_ident(log);
949         if (!ident)
950                 return 2;
951
952         return format_person_part(sb, part, ident, strlen(ident), dmode);
953 }
954
955 static size_t parse_color(struct strbuf *sb, /* in UTF-8 */
956                           const char *placeholder,
957                           struct format_commit_context *c)
958 {
959         const char *rest = placeholder;
960         const char *basic_color = NULL;
961
962         if (placeholder[1] == '(') {
963                 const char *begin = placeholder + 2;
964                 const char *end = strchr(begin, ')');
965                 char color[COLOR_MAXLEN];
966
967                 if (!end)
968                         return 0;
969
970                 if (skip_prefix(begin, "auto,", &begin)) {
971                         if (!want_color(c->pretty_ctx->color))
972                                 return end - placeholder + 1;
973                 } else if (skip_prefix(begin, "always,", &begin)) {
974                         /* nothing to do; we do not respect want_color at all */
975                 } else {
976                         /* the default is the same as "auto" */
977                         if (!want_color(c->pretty_ctx->color))
978                                 return end - placeholder + 1;
979                 }
980
981                 if (color_parse_mem(begin, end - begin, color) < 0)
982                         die(_("unable to parse --pretty format"));
983                 strbuf_addstr(sb, color);
984                 return end - placeholder + 1;
985         }
986
987         /*
988          * We handle things like "%C(red)" above; for historical reasons, there
989          * are a few colors that can be specified without parentheses (and
990          * they cannot support things like "auto" or "always" at all).
991          */
992         if (skip_prefix(placeholder + 1, "red", &rest))
993                 basic_color = GIT_COLOR_RED;
994         else if (skip_prefix(placeholder + 1, "green", &rest))
995                 basic_color = GIT_COLOR_GREEN;
996         else if (skip_prefix(placeholder + 1, "blue", &rest))
997                 basic_color = GIT_COLOR_BLUE;
998         else if (skip_prefix(placeholder + 1, "reset", &rest))
999                 basic_color = GIT_COLOR_RESET;
1000
1001         if (basic_color && want_color(c->pretty_ctx->color))
1002                 strbuf_addstr(sb, basic_color);
1003
1004         return rest - placeholder;
1005 }
1006
1007 static size_t parse_padding_placeholder(const char *placeholder,
1008                                         struct format_commit_context *c)
1009 {
1010         const char *ch = placeholder;
1011         enum flush_type flush_type;
1012         int to_column = 0;
1013
1014         switch (*ch++) {
1015         case '<':
1016                 flush_type = flush_right;
1017                 break;
1018         case '>':
1019                 if (*ch == '<') {
1020                         flush_type = flush_both;
1021                         ch++;
1022                 } else if (*ch == '>') {
1023                         flush_type = flush_left_and_steal;
1024                         ch++;
1025                 } else
1026                         flush_type = flush_left;
1027                 break;
1028         default:
1029                 return 0;
1030         }
1031
1032         /* the next value means "wide enough to that column" */
1033         if (*ch == '|') {
1034                 to_column = 1;
1035                 ch++;
1036         }
1037
1038         if (*ch == '(') {
1039                 const char *start = ch + 1;
1040                 const char *end = start + strcspn(start, ",)");
1041                 char *next;
1042                 int width;
1043                 if (!end || end == start)
1044                         return 0;
1045                 width = strtol(start, &next, 10);
1046                 if (next == start || width == 0)
1047                         return 0;
1048                 if (width < 0) {
1049                         if (to_column)
1050                                 width += term_columns();
1051                         if (width < 0)
1052                                 return 0;
1053                 }
1054                 c->padding = to_column ? -width : width;
1055                 c->flush_type = flush_type;
1056
1057                 if (*end == ',') {
1058                         start = end + 1;
1059                         end = strchr(start, ')');
1060                         if (!end || end == start)
1061                                 return 0;
1062                         if (starts_with(start, "trunc)"))
1063                                 c->truncate = trunc_right;
1064                         else if (starts_with(start, "ltrunc)"))
1065                                 c->truncate = trunc_left;
1066                         else if (starts_with(start, "mtrunc)"))
1067                                 c->truncate = trunc_middle;
1068                         else
1069                                 return 0;
1070                 } else
1071                         c->truncate = trunc_none;
1072
1073                 return end - placeholder + 1;
1074         }
1075         return 0;
1076 }
1077
1078 static int match_placeholder_arg_value(const char *to_parse, const char *candidate,
1079                                        const char **end, const char **valuestart,
1080                                        size_t *valuelen)
1081 {
1082         const char *p;
1083
1084         if (!(skip_prefix(to_parse, candidate, &p)))
1085                 return 0;
1086         if (valuestart) {
1087                 if (*p == '=') {
1088                         *valuestart = p + 1;
1089                         *valuelen = strcspn(*valuestart, ",)");
1090                         p = *valuestart + *valuelen;
1091                 } else {
1092                         if (*p != ',' && *p != ')')
1093                                 return 0;
1094                         *valuestart = NULL;
1095                         *valuelen = 0;
1096                 }
1097         }
1098         if (*p == ',') {
1099                 *end = p + 1;
1100                 return 1;
1101         }
1102         if (*p == ')') {
1103                 *end = p;
1104                 return 1;
1105         }
1106         return 0;
1107 }
1108
1109 static int match_placeholder_bool_arg(const char *to_parse, const char *candidate,
1110                                       const char **end, int *val)
1111 {
1112         const char *argval;
1113         char *strval;
1114         size_t arglen;
1115         int v;
1116
1117         if (!match_placeholder_arg_value(to_parse, candidate, end, &argval, &arglen))
1118                 return 0;
1119
1120         if (!argval) {
1121                 *val = 1;
1122                 return 1;
1123         }
1124
1125         strval = xstrndup(argval, arglen);
1126         v = git_parse_maybe_bool(strval);
1127         free(strval);
1128
1129         if (v == -1)
1130                 return 0;
1131
1132         *val = v;
1133
1134         return 1;
1135 }
1136
1137 static int format_trailer_match_cb(const struct strbuf *key, void *ud)
1138 {
1139         const struct string_list *list = ud;
1140         const struct string_list_item *item;
1141
1142         for_each_string_list_item (item, list) {
1143                 if (key->len == (uintptr_t)item->util &&
1144                     !strncasecmp(item->string, key->buf, key->len))
1145                         return 1;
1146         }
1147         return 0;
1148 }
1149
1150 static size_t format_commit_one(struct strbuf *sb, /* in UTF-8 */
1151                                 const char *placeholder,
1152                                 void *context)
1153 {
1154         struct format_commit_context *c = context;
1155         const struct commit *commit = c->commit;
1156         const char *msg = c->message;
1157         struct commit_list *p;
1158         const char *arg;
1159         size_t res;
1160         char **slot;
1161
1162         /* these are independent of the commit */
1163         res = strbuf_expand_literal_cb(sb, placeholder, NULL);
1164         if (res)
1165                 return res;
1166
1167         switch (placeholder[0]) {
1168         case 'C':
1169                 if (starts_with(placeholder + 1, "(auto)")) {
1170                         c->auto_color = want_color(c->pretty_ctx->color);
1171                         if (c->auto_color && sb->len)
1172                                 strbuf_addstr(sb, GIT_COLOR_RESET);
1173                         return 7; /* consumed 7 bytes, "C(auto)" */
1174                 } else {
1175                         int ret = parse_color(sb, placeholder, c);
1176                         if (ret)
1177                                 c->auto_color = 0;
1178                         /*
1179                          * Otherwise, we decided to treat %C<unknown>
1180                          * as a literal string, and the previous
1181                          * %C(auto) is still valid.
1182                          */
1183                         return ret;
1184                 }
1185         case 'w':
1186                 if (placeholder[1] == '(') {
1187                         unsigned long width = 0, indent1 = 0, indent2 = 0;
1188                         char *next;
1189                         const char *start = placeholder + 2;
1190                         const char *end = strchr(start, ')');
1191                         if (!end)
1192                                 return 0;
1193                         if (end > start) {
1194                                 width = strtoul(start, &next, 10);
1195                                 if (*next == ',') {
1196                                         indent1 = strtoul(next + 1, &next, 10);
1197                                         if (*next == ',') {
1198                                                 indent2 = strtoul(next + 1,
1199                                                                  &next, 10);
1200                                         }
1201                                 }
1202                                 if (*next != ')')
1203                                         return 0;
1204                         }
1205                         rewrap_message_tail(sb, c, width, indent1, indent2);
1206                         return end - placeholder + 1;
1207                 } else
1208                         return 0;
1209
1210         case '<':
1211         case '>':
1212                 return parse_padding_placeholder(placeholder, c);
1213         }
1214
1215         /* these depend on the commit */
1216         if (!commit->object.parsed)
1217                 parse_object(the_repository, &commit->object.oid);
1218
1219         switch (placeholder[0]) {
1220         case 'H':               /* commit hash */
1221                 strbuf_addstr(sb, diff_get_color(c->auto_color, DIFF_COMMIT));
1222                 strbuf_addstr(sb, oid_to_hex(&commit->object.oid));
1223                 strbuf_addstr(sb, diff_get_color(c->auto_color, DIFF_RESET));
1224                 return 1;
1225         case 'h':               /* abbreviated commit hash */
1226                 strbuf_addstr(sb, diff_get_color(c->auto_color, DIFF_COMMIT));
1227                 strbuf_add_unique_abbrev(sb, &commit->object.oid,
1228                                          c->pretty_ctx->abbrev);
1229                 strbuf_addstr(sb, diff_get_color(c->auto_color, DIFF_RESET));
1230                 return 1;
1231         case 'T':               /* tree hash */
1232                 strbuf_addstr(sb, oid_to_hex(get_commit_tree_oid(commit)));
1233                 return 1;
1234         case 't':               /* abbreviated tree hash */
1235                 strbuf_add_unique_abbrev(sb,
1236                                          get_commit_tree_oid(commit),
1237                                          c->pretty_ctx->abbrev);
1238                 return 1;
1239         case 'P':               /* parent hashes */
1240                 for (p = commit->parents; p; p = p->next) {
1241                         if (p != commit->parents)
1242                                 strbuf_addch(sb, ' ');
1243                         strbuf_addstr(sb, oid_to_hex(&p->item->object.oid));
1244                 }
1245                 return 1;
1246         case 'p':               /* abbreviated parent hashes */
1247                 for (p = commit->parents; p; p = p->next) {
1248                         if (p != commit->parents)
1249                                 strbuf_addch(sb, ' ');
1250                         strbuf_add_unique_abbrev(sb, &p->item->object.oid,
1251                                                  c->pretty_ctx->abbrev);
1252                 }
1253                 return 1;
1254         case 'm':               /* left/right/bottom */
1255                 strbuf_addstr(sb, get_revision_mark(NULL, commit));
1256                 return 1;
1257         case 'd':
1258                 format_decorations(sb, commit, c->auto_color);
1259                 return 1;
1260         case 'D':
1261                 format_decorations_extended(sb, commit, c->auto_color, "", ", ", "");
1262                 return 1;
1263         case 'S':               /* tag/branch like --source */
1264                 if (!(c->pretty_ctx->rev && c->pretty_ctx->rev->sources))
1265                         return 0;
1266                 slot = revision_sources_at(c->pretty_ctx->rev->sources, commit);
1267                 if (!(slot && *slot))
1268                         return 0;
1269                 strbuf_addstr(sb, *slot);
1270                 return 1;
1271         case 'g':               /* reflog info */
1272                 switch(placeholder[1]) {
1273                 case 'd':       /* reflog selector */
1274                 case 'D':
1275                         if (c->pretty_ctx->reflog_info)
1276                                 get_reflog_selector(sb,
1277                                                     c->pretty_ctx->reflog_info,
1278                                                     &c->pretty_ctx->date_mode,
1279                                                     c->pretty_ctx->date_mode_explicit,
1280                                                     (placeholder[1] == 'd'));
1281                         return 2;
1282                 case 's':       /* reflog message */
1283                         if (c->pretty_ctx->reflog_info)
1284                                 get_reflog_message(sb, c->pretty_ctx->reflog_info);
1285                         return 2;
1286                 case 'n':
1287                 case 'N':
1288                 case 'e':
1289                 case 'E':
1290                         return format_reflog_person(sb,
1291                                                     placeholder[1],
1292                                                     c->pretty_ctx->reflog_info,
1293                                                     &c->pretty_ctx->date_mode);
1294                 }
1295                 return 0;       /* unknown %g placeholder */
1296         case 'N':
1297                 if (c->pretty_ctx->notes_message) {
1298                         strbuf_addstr(sb, c->pretty_ctx->notes_message);
1299                         return 1;
1300                 }
1301                 return 0;
1302         }
1303
1304         if (placeholder[0] == 'G') {
1305                 if (!c->signature_check.result)
1306                         check_commit_signature(c->commit, &(c->signature_check));
1307                 switch (placeholder[1]) {
1308                 case 'G':
1309                         if (c->signature_check.gpg_output)
1310                                 strbuf_addstr(sb, c->signature_check.gpg_output);
1311                         break;
1312                 case '?':
1313                         switch (c->signature_check.result) {
1314                         case 'G':
1315                                 switch (c->signature_check.trust_level) {
1316                                 case TRUST_UNDEFINED:
1317                                 case TRUST_NEVER:
1318                                         strbuf_addch(sb, 'U');
1319                                         break;
1320                                 default:
1321                                         strbuf_addch(sb, 'G');
1322                                         break;
1323                                 }
1324                                 break;
1325                         case 'B':
1326                         case 'E':
1327                         case 'N':
1328                         case 'X':
1329                         case 'Y':
1330                         case 'R':
1331                                 strbuf_addch(sb, c->signature_check.result);
1332                         }
1333                         break;
1334                 case 'S':
1335                         if (c->signature_check.signer)
1336                                 strbuf_addstr(sb, c->signature_check.signer);
1337                         break;
1338                 case 'K':
1339                         if (c->signature_check.key)
1340                                 strbuf_addstr(sb, c->signature_check.key);
1341                         break;
1342                 case 'F':
1343                         if (c->signature_check.fingerprint)
1344                                 strbuf_addstr(sb, c->signature_check.fingerprint);
1345                         break;
1346                 case 'P':
1347                         if (c->signature_check.primary_key_fingerprint)
1348                                 strbuf_addstr(sb, c->signature_check.primary_key_fingerprint);
1349                         break;
1350                 case 'T':
1351                         switch (c->signature_check.trust_level) {
1352                         case TRUST_UNDEFINED:
1353                                 strbuf_addstr(sb, "undefined");
1354                                 break;
1355                         case TRUST_NEVER:
1356                                 strbuf_addstr(sb, "never");
1357                                 break;
1358                         case TRUST_MARGINAL:
1359                                 strbuf_addstr(sb, "marginal");
1360                                 break;
1361                         case TRUST_FULLY:
1362                                 strbuf_addstr(sb, "fully");
1363                                 break;
1364                         case TRUST_ULTIMATE:
1365                                 strbuf_addstr(sb, "ultimate");
1366                                 break;
1367                         }
1368                         break;
1369                 default:
1370                         return 0;
1371                 }
1372                 return 2;
1373         }
1374
1375
1376         /* For the rest we have to parse the commit header. */
1377         if (!c->commit_header_parsed)
1378                 parse_commit_header(c);
1379
1380         switch (placeholder[0]) {
1381         case 'a':       /* author ... */
1382                 return format_person_part(sb, placeholder[1],
1383                                    msg + c->author.off, c->author.len,
1384                                    &c->pretty_ctx->date_mode);
1385         case 'c':       /* committer ... */
1386                 return format_person_part(sb, placeholder[1],
1387                                    msg + c->committer.off, c->committer.len,
1388                                    &c->pretty_ctx->date_mode);
1389         case 'e':       /* encoding */
1390                 if (c->commit_encoding)
1391                         strbuf_addstr(sb, c->commit_encoding);
1392                 return 1;
1393         case 'B':       /* raw body */
1394                 /* message_off is always left at the initial newline */
1395                 strbuf_addstr(sb, msg + c->message_off + 1);
1396                 return 1;
1397         }
1398
1399         /* Now we need to parse the commit message. */
1400         if (!c->commit_message_parsed)
1401                 parse_commit_message(c);
1402
1403         switch (placeholder[0]) {
1404         case 's':       /* subject */
1405                 format_subject(sb, msg + c->subject_off, " ");
1406                 return 1;
1407         case 'f':       /* sanitized subject */
1408                 format_sanitized_subject(sb, msg + c->subject_off);
1409                 return 1;
1410         case 'b':       /* body */
1411                 strbuf_addstr(sb, msg + c->body_off);
1412                 return 1;
1413         }
1414
1415         if (skip_prefix(placeholder, "(trailers", &arg)) {
1416                 struct process_trailer_options opts = PROCESS_TRAILER_OPTIONS_INIT;
1417                 struct string_list filter_list = STRING_LIST_INIT_NODUP;
1418                 struct strbuf sepbuf = STRBUF_INIT;
1419                 size_t ret = 0;
1420
1421                 opts.no_divider = 1;
1422
1423                 if (*arg == ':') {
1424                         arg++;
1425                         for (;;) {
1426                                 const char *argval;
1427                                 size_t arglen;
1428
1429                                 if (match_placeholder_arg_value(arg, "key", &arg, &argval, &arglen)) {
1430                                         uintptr_t len = arglen;
1431
1432                                         if (!argval)
1433                                                 goto trailer_out;
1434
1435                                         if (len && argval[len - 1] == ':')
1436                                                 len--;
1437                                         string_list_append(&filter_list, argval)->util = (char *)len;
1438
1439                                         opts.filter = format_trailer_match_cb;
1440                                         opts.filter_data = &filter_list;
1441                                         opts.only_trailers = 1;
1442                                 } else if (match_placeholder_arg_value(arg, "separator", &arg, &argval, &arglen)) {
1443                                         char *fmt;
1444
1445                                         strbuf_reset(&sepbuf);
1446                                         fmt = xstrndup(argval, arglen);
1447                                         strbuf_expand(&sepbuf, fmt, strbuf_expand_literal_cb, NULL);
1448                                         free(fmt);
1449                                         opts.separator = &sepbuf;
1450                                 } else if (!match_placeholder_bool_arg(arg, "only", &arg, &opts.only_trailers) &&
1451                                            !match_placeholder_bool_arg(arg, "unfold", &arg, &opts.unfold) &&
1452                                            !match_placeholder_bool_arg(arg, "valueonly", &arg, &opts.value_only))
1453                                         break;
1454                         }
1455                 }
1456                 if (*arg == ')') {
1457                         format_trailers_from_commit(sb, msg + c->subject_off, &opts);
1458                         ret = arg - placeholder + 1;
1459                 }
1460         trailer_out:
1461                 string_list_clear(&filter_list, 0);
1462                 strbuf_release(&sepbuf);
1463                 return ret;
1464         }
1465
1466         return 0;       /* unknown placeholder */
1467 }
1468
1469 static size_t format_and_pad_commit(struct strbuf *sb, /* in UTF-8 */
1470                                     const char *placeholder,
1471                                     struct format_commit_context *c)
1472 {
1473         struct strbuf local_sb = STRBUF_INIT;
1474         int total_consumed = 0, len, padding = c->padding;
1475         if (padding < 0) {
1476                 const char *start = strrchr(sb->buf, '\n');
1477                 int occupied;
1478                 if (!start)
1479                         start = sb->buf;
1480                 occupied = utf8_strnwidth(start, -1, 1);
1481                 occupied += c->pretty_ctx->graph_width;
1482                 padding = (-padding) - occupied;
1483         }
1484         while (1) {
1485                 int modifier = *placeholder == 'C';
1486                 int consumed = format_commit_one(&local_sb, placeholder, c);
1487                 total_consumed += consumed;
1488
1489                 if (!modifier)
1490                         break;
1491
1492                 placeholder += consumed;
1493                 if (*placeholder != '%')
1494                         break;
1495                 placeholder++;
1496                 total_consumed++;
1497         }
1498         len = utf8_strnwidth(local_sb.buf, -1, 1);
1499
1500         if (c->flush_type == flush_left_and_steal) {
1501                 const char *ch = sb->buf + sb->len - 1;
1502                 while (len > padding && ch > sb->buf) {
1503                         const char *p;
1504                         if (*ch == ' ') {
1505                                 ch--;
1506                                 padding++;
1507                                 continue;
1508                         }
1509                         /* check for trailing ansi sequences */
1510                         if (*ch != 'm')
1511                                 break;
1512                         p = ch - 1;
1513                         while (ch - p < 10 && *p != '\033')
1514                                 p--;
1515                         if (*p != '\033' ||
1516                             ch + 1 - p != display_mode_esc_sequence_len(p))
1517                                 break;
1518                         /*
1519                          * got a good ansi sequence, put it back to
1520                          * local_sb as we're cutting sb
1521                          */
1522                         strbuf_insert(&local_sb, 0, p, ch + 1 - p);
1523                         ch = p - 1;
1524                 }
1525                 strbuf_setlen(sb, ch + 1 - sb->buf);
1526                 c->flush_type = flush_left;
1527         }
1528
1529         if (len > padding) {
1530                 switch (c->truncate) {
1531                 case trunc_left:
1532                         strbuf_utf8_replace(&local_sb,
1533                                             0, len - (padding - 2),
1534                                             "..");
1535                         break;
1536                 case trunc_middle:
1537                         strbuf_utf8_replace(&local_sb,
1538                                             padding / 2 - 1,
1539                                             len - (padding - 2),
1540                                             "..");
1541                         break;
1542                 case trunc_right:
1543                         strbuf_utf8_replace(&local_sb,
1544                                             padding - 2, len - (padding - 2),
1545                                             "..");
1546                         break;
1547                 case trunc_none:
1548                         break;
1549                 }
1550                 strbuf_addbuf(sb, &local_sb);
1551         } else {
1552                 int sb_len = sb->len, offset = 0;
1553                 if (c->flush_type == flush_left)
1554                         offset = padding - len;
1555                 else if (c->flush_type == flush_both)
1556                         offset = (padding - len) / 2;
1557                 /*
1558                  * we calculate padding in columns, now
1559                  * convert it back to chars
1560                  */
1561                 padding = padding - len + local_sb.len;
1562                 strbuf_addchars(sb, ' ', padding);
1563                 memcpy(sb->buf + sb_len + offset, local_sb.buf,
1564                        local_sb.len);
1565         }
1566         strbuf_release(&local_sb);
1567         c->flush_type = no_flush;
1568         return total_consumed;
1569 }
1570
1571 static size_t format_commit_item(struct strbuf *sb, /* in UTF-8 */
1572                                  const char *placeholder,
1573                                  void *context)
1574 {
1575         int consumed;
1576         size_t orig_len;
1577         enum {
1578                 NO_MAGIC,
1579                 ADD_LF_BEFORE_NON_EMPTY,
1580                 DEL_LF_BEFORE_EMPTY,
1581                 ADD_SP_BEFORE_NON_EMPTY
1582         } magic = NO_MAGIC;
1583
1584         switch (placeholder[0]) {
1585         case '-':
1586                 magic = DEL_LF_BEFORE_EMPTY;
1587                 break;
1588         case '+':
1589                 magic = ADD_LF_BEFORE_NON_EMPTY;
1590                 break;
1591         case ' ':
1592                 magic = ADD_SP_BEFORE_NON_EMPTY;
1593                 break;
1594         default:
1595                 break;
1596         }
1597         if (magic != NO_MAGIC)
1598                 placeholder++;
1599
1600         orig_len = sb->len;
1601         if (((struct format_commit_context *)context)->flush_type != no_flush)
1602                 consumed = format_and_pad_commit(sb, placeholder, context);
1603         else
1604                 consumed = format_commit_one(sb, placeholder, context);
1605         if (magic == NO_MAGIC)
1606                 return consumed;
1607
1608         if ((orig_len == sb->len) && magic == DEL_LF_BEFORE_EMPTY) {
1609                 while (sb->len && sb->buf[sb->len - 1] == '\n')
1610                         strbuf_setlen(sb, sb->len - 1);
1611         } else if (orig_len != sb->len) {
1612                 if (magic == ADD_LF_BEFORE_NON_EMPTY)
1613                         strbuf_insertstr(sb, orig_len, "\n");
1614                 else if (magic == ADD_SP_BEFORE_NON_EMPTY)
1615                         strbuf_insertstr(sb, orig_len, " ");
1616         }
1617         return consumed + 1;
1618 }
1619
1620 static size_t userformat_want_item(struct strbuf *sb, const char *placeholder,
1621                                    void *context)
1622 {
1623         struct userformat_want *w = context;
1624
1625         if (*placeholder == '+' || *placeholder == '-' || *placeholder == ' ')
1626                 placeholder++;
1627
1628         switch (*placeholder) {
1629         case 'N':
1630                 w->notes = 1;
1631                 break;
1632         case 'S':
1633                 w->source = 1;
1634                 break;
1635         }
1636         return 0;
1637 }
1638
1639 void userformat_find_requirements(const char *fmt, struct userformat_want *w)
1640 {
1641         struct strbuf dummy = STRBUF_INIT;
1642
1643         if (!fmt) {
1644                 if (!user_format)
1645                         return;
1646                 fmt = user_format;
1647         }
1648         strbuf_expand(&dummy, fmt, userformat_want_item, w);
1649         strbuf_release(&dummy);
1650 }
1651
1652 void repo_format_commit_message(struct repository *r,
1653                                 const struct commit *commit,
1654                                 const char *format, struct strbuf *sb,
1655                                 const struct pretty_print_context *pretty_ctx)
1656 {
1657         struct format_commit_context context = {
1658                 .commit = commit,
1659                 .pretty_ctx = pretty_ctx,
1660                 .wrap_start = sb->len
1661         };
1662         const char *output_enc = pretty_ctx->output_encoding;
1663         const char *utf8 = "UTF-8";
1664
1665         /*
1666          * convert a commit message to UTF-8 first
1667          * as far as 'format_commit_item' assumes it in UTF-8
1668          */
1669         context.message = repo_logmsg_reencode(r, commit,
1670                                                &context.commit_encoding,
1671                                                utf8);
1672
1673         strbuf_expand(sb, format, format_commit_item, &context);
1674         rewrap_message_tail(sb, &context, 0, 0, 0);
1675
1676         /* then convert a commit message to an actual output encoding */
1677         if (output_enc) {
1678                 if (same_encoding(utf8, output_enc))
1679                         output_enc = NULL;
1680         } else {
1681                 if (context.commit_encoding &&
1682                     !same_encoding(context.commit_encoding, utf8))
1683                         output_enc = context.commit_encoding;
1684         }
1685
1686         if (output_enc) {
1687                 size_t outsz;
1688                 char *out = reencode_string_len(sb->buf, sb->len,
1689                                                 output_enc, utf8, &outsz);
1690                 if (out)
1691                         strbuf_attach(sb, out, outsz, outsz + 1);
1692         }
1693
1694         free(context.commit_encoding);
1695         repo_unuse_commit_buffer(r, commit, context.message);
1696 }
1697
1698 static void pp_header(struct pretty_print_context *pp,
1699                       const char *encoding,
1700                       const struct commit *commit,
1701                       const char **msg_p,
1702                       struct strbuf *sb)
1703 {
1704         int parents_shown = 0;
1705
1706         for (;;) {
1707                 const char *name, *line = *msg_p;
1708                 int linelen = get_one_line(*msg_p);
1709
1710                 if (!linelen)
1711                         return;
1712                 *msg_p += linelen;
1713
1714                 if (linelen == 1)
1715                         /* End of header */
1716                         return;
1717
1718                 if (pp->fmt == CMIT_FMT_RAW) {
1719                         strbuf_add(sb, line, linelen);
1720                         continue;
1721                 }
1722
1723                 if (starts_with(line, "parent ")) {
1724                         if (linelen != the_hash_algo->hexsz + 8)
1725                                 die("bad parent line in commit");
1726                         continue;
1727                 }
1728
1729                 if (!parents_shown) {
1730                         unsigned num = commit_list_count(commit->parents);
1731                         /* with enough slop */
1732                         strbuf_grow(sb, num * (GIT_MAX_HEXSZ + 10) + 20);
1733                         add_merge_info(pp, sb, commit);
1734                         parents_shown = 1;
1735                 }
1736
1737                 /*
1738                  * MEDIUM == DEFAULT shows only author with dates.
1739                  * FULL shows both authors but not dates.
1740                  * FULLER shows both authors and dates.
1741                  */
1742                 if (skip_prefix(line, "author ", &name)) {
1743                         strbuf_grow(sb, linelen + 80);
1744                         pp_user_info(pp, "Author", sb, name, encoding);
1745                 }
1746                 if (skip_prefix(line, "committer ", &name) &&
1747                     (pp->fmt == CMIT_FMT_FULL || pp->fmt == CMIT_FMT_FULLER)) {
1748                         strbuf_grow(sb, linelen + 80);
1749                         pp_user_info(pp, "Commit", sb, name, encoding);
1750                 }
1751         }
1752 }
1753
1754 void pp_title_line(struct pretty_print_context *pp,
1755                    const char **msg_p,
1756                    struct strbuf *sb,
1757                    const char *encoding,
1758                    int need_8bit_cte)
1759 {
1760         static const int max_length = 78; /* per rfc2047 */
1761         struct strbuf title;
1762
1763         strbuf_init(&title, 80);
1764         *msg_p = format_subject(&title, *msg_p,
1765                                 pp->preserve_subject ? "\n" : " ");
1766
1767         strbuf_grow(sb, title.len + 1024);
1768         if (pp->print_email_subject) {
1769                 if (pp->rev)
1770                         fmt_output_email_subject(sb, pp->rev);
1771                 if (pp->encode_email_headers &&
1772                     needs_rfc2047_encoding(title.buf, title.len))
1773                         add_rfc2047(sb, title.buf, title.len,
1774                                                 encoding, RFC2047_SUBJECT);
1775                 else
1776                         strbuf_add_wrapped_bytes(sb, title.buf, title.len,
1777                                          -last_line_length(sb), 1, max_length);
1778         } else {
1779                 strbuf_addbuf(sb, &title);
1780         }
1781         strbuf_addch(sb, '\n');
1782
1783         if (need_8bit_cte == 0) {
1784                 int i;
1785                 for (i = 0; i < pp->in_body_headers.nr; i++) {
1786                         if (has_non_ascii(pp->in_body_headers.items[i].string)) {
1787                                 need_8bit_cte = 1;
1788                                 break;
1789                         }
1790                 }
1791         }
1792
1793         if (need_8bit_cte > 0) {
1794                 const char *header_fmt =
1795                         "MIME-Version: 1.0\n"
1796                         "Content-Type: text/plain; charset=%s\n"
1797                         "Content-Transfer-Encoding: 8bit\n";
1798                 strbuf_addf(sb, header_fmt, encoding);
1799         }
1800         if (pp->after_subject) {
1801                 strbuf_addstr(sb, pp->after_subject);
1802         }
1803         if (cmit_fmt_is_mail(pp->fmt)) {
1804                 strbuf_addch(sb, '\n');
1805         }
1806
1807         if (pp->in_body_headers.nr) {
1808                 int i;
1809                 for (i = 0; i < pp->in_body_headers.nr; i++) {
1810                         strbuf_addstr(sb, pp->in_body_headers.items[i].string);
1811                         free(pp->in_body_headers.items[i].string);
1812                 }
1813                 string_list_clear(&pp->in_body_headers, 0);
1814                 strbuf_addch(sb, '\n');
1815         }
1816
1817         strbuf_release(&title);
1818 }
1819
1820 static int pp_utf8_width(const char *start, const char *end)
1821 {
1822         int width = 0;
1823         size_t remain = end - start;
1824
1825         while (remain) {
1826                 int n = utf8_width(&start, &remain);
1827                 if (n < 0 || !start)
1828                         return -1;
1829                 width += n;
1830         }
1831         return width;
1832 }
1833
1834 static void strbuf_add_tabexpand(struct strbuf *sb, int tabwidth,
1835                                  const char *line, int linelen)
1836 {
1837         const char *tab;
1838
1839         while ((tab = memchr(line, '\t', linelen)) != NULL) {
1840                 int width = pp_utf8_width(line, tab);
1841
1842                 /*
1843                  * If it wasn't well-formed utf8, or it
1844                  * had characters with badly defined
1845                  * width (control characters etc), just
1846                  * give up on trying to align things.
1847                  */
1848                 if (width < 0)
1849                         break;
1850
1851                 /* Output the data .. */
1852                 strbuf_add(sb, line, tab - line);
1853
1854                 /* .. and the de-tabified tab */
1855                 strbuf_addchars(sb, ' ', tabwidth - (width % tabwidth));
1856
1857                 /* Skip over the printed part .. */
1858                 linelen -= tab + 1 - line;
1859                 line = tab + 1;
1860         }
1861
1862         /*
1863          * Print out everything after the last tab without
1864          * worrying about width - there's nothing more to
1865          * align.
1866          */
1867         strbuf_add(sb, line, linelen);
1868 }
1869
1870 /*
1871  * pp_handle_indent() prints out the intendation, and
1872  * the whole line (without the final newline), after
1873  * de-tabifying.
1874  */
1875 static void pp_handle_indent(struct pretty_print_context *pp,
1876                              struct strbuf *sb, int indent,
1877                              const char *line, int linelen)
1878 {
1879         strbuf_addchars(sb, ' ', indent);
1880         if (pp->expand_tabs_in_log)
1881                 strbuf_add_tabexpand(sb, pp->expand_tabs_in_log, line, linelen);
1882         else
1883                 strbuf_add(sb, line, linelen);
1884 }
1885
1886 static int is_mboxrd_from(const char *line, int len)
1887 {
1888         /*
1889          * a line matching /^From $/ here would only have len == 4
1890          * at this point because is_empty_line would've trimmed all
1891          * trailing space
1892          */
1893         return len > 4 && starts_with(line + strspn(line, ">"), "From ");
1894 }
1895
1896 void pp_remainder(struct pretty_print_context *pp,
1897                   const char **msg_p,
1898                   struct strbuf *sb,
1899                   int indent)
1900 {
1901         int first = 1;
1902         for (;;) {
1903                 const char *line = *msg_p;
1904                 int linelen = get_one_line(line);
1905                 *msg_p += linelen;
1906
1907                 if (!linelen)
1908                         break;
1909
1910                 if (is_blank_line(line, &linelen)) {
1911                         if (first)
1912                                 continue;
1913                         if (pp->fmt == CMIT_FMT_SHORT)
1914                                 break;
1915                 }
1916                 first = 0;
1917
1918                 strbuf_grow(sb, linelen + indent + 20);
1919                 if (indent)
1920                         pp_handle_indent(pp, sb, indent, line, linelen);
1921                 else if (pp->expand_tabs_in_log)
1922                         strbuf_add_tabexpand(sb, pp->expand_tabs_in_log,
1923                                              line, linelen);
1924                 else {
1925                         if (pp->fmt == CMIT_FMT_MBOXRD &&
1926                                         is_mboxrd_from(line, linelen))
1927                                 strbuf_addch(sb, '>');
1928
1929                         strbuf_add(sb, line, linelen);
1930                 }
1931                 strbuf_addch(sb, '\n');
1932         }
1933 }
1934
1935 void pretty_print_commit(struct pretty_print_context *pp,
1936                          const struct commit *commit,
1937                          struct strbuf *sb)
1938 {
1939         unsigned long beginning_of_body;
1940         int indent = 4;
1941         const char *msg;
1942         const char *reencoded;
1943         const char *encoding;
1944         int need_8bit_cte = pp->need_8bit_cte;
1945
1946         if (pp->fmt == CMIT_FMT_USERFORMAT) {
1947                 format_commit_message(commit, user_format, sb, pp);
1948                 return;
1949         }
1950
1951         encoding = get_log_output_encoding();
1952         msg = reencoded = logmsg_reencode(commit, NULL, encoding);
1953
1954         if (pp->fmt == CMIT_FMT_ONELINE || cmit_fmt_is_mail(pp->fmt))
1955                 indent = 0;
1956
1957         /*
1958          * We need to check and emit Content-type: to mark it
1959          * as 8-bit if we haven't done so.
1960          */
1961         if (cmit_fmt_is_mail(pp->fmt) && need_8bit_cte == 0) {
1962                 int i, ch, in_body;
1963
1964                 for (in_body = i = 0; (ch = msg[i]); i++) {
1965                         if (!in_body) {
1966                                 /* author could be non 7-bit ASCII but
1967                                  * the log may be so; skip over the
1968                                  * header part first.
1969                                  */
1970                                 if (ch == '\n' && msg[i+1] == '\n')
1971                                         in_body = 1;
1972                         }
1973                         else if (non_ascii(ch)) {
1974                                 need_8bit_cte = 1;
1975                                 break;
1976                         }
1977                 }
1978         }
1979
1980         pp_header(pp, encoding, commit, &msg, sb);
1981         if (pp->fmt != CMIT_FMT_ONELINE && !pp->print_email_subject) {
1982                 strbuf_addch(sb, '\n');
1983         }
1984
1985         /* Skip excess blank lines at the beginning of body, if any... */
1986         msg = skip_blank_lines(msg);
1987
1988         /* These formats treat the title line specially. */
1989         if (pp->fmt == CMIT_FMT_ONELINE || cmit_fmt_is_mail(pp->fmt))
1990                 pp_title_line(pp, &msg, sb, encoding, need_8bit_cte);
1991
1992         beginning_of_body = sb->len;
1993         if (pp->fmt != CMIT_FMT_ONELINE)
1994                 pp_remainder(pp, &msg, sb, indent);
1995         strbuf_rtrim(sb);
1996
1997         /* Make sure there is an EOLN for the non-oneline case */
1998         if (pp->fmt != CMIT_FMT_ONELINE)
1999                 strbuf_addch(sb, '\n');
2000
2001         /*
2002          * The caller may append additional body text in e-mail
2003          * format.  Make sure we did not strip the blank line
2004          * between the header and the body.
2005          */
2006         if (cmit_fmt_is_mail(pp->fmt) && sb->len <= beginning_of_body)
2007                 strbuf_addch(sb, '\n');
2008
2009         unuse_commit_buffer(commit, reencoded);
2010 }
2011
2012 void pp_commit_easy(enum cmit_fmt fmt, const struct commit *commit,
2013                     struct strbuf *sb)
2014 {
2015         struct pretty_print_context pp = {0};
2016         pp.fmt = fmt;
2017         pretty_print_commit(&pp, commit, sb);
2018 }