Merge branch 'pm/p4-auto-delete-named-temporary'
[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         const char *user_format;
24 } *commit_formats;
25 static size_t builtin_formats_len;
26 static size_t commit_formats_len;
27 static size_t commit_formats_alloc;
28 static struct cmt_fmt_map *find_commit_format(const char *sought);
29
30 int commit_format_is_empty(enum cmit_fmt fmt)
31 {
32         return fmt == CMIT_FMT_USERFORMAT && !*user_format;
33 }
34
35 static void save_user_format(struct rev_info *rev, const char *cp, int is_tformat)
36 {
37         free(user_format);
38         user_format = xstrdup(cp);
39         if (is_tformat)
40                 rev->use_terminator = 1;
41         rev->commit_format = CMIT_FMT_USERFORMAT;
42 }
43
44 static int git_pretty_formats_config(const char *var, const char *value, void *cb)
45 {
46         struct cmt_fmt_map *commit_format = NULL;
47         const char *name;
48         const char *fmt;
49         int i;
50
51         if (!skip_prefix(var, "pretty.", &name))
52                 return 0;
53
54         for (i = 0; i < builtin_formats_len; i++) {
55                 if (!strcmp(commit_formats[i].name, name))
56                         return 0;
57         }
58
59         for (i = builtin_formats_len; i < commit_formats_len; i++) {
60                 if (!strcmp(commit_formats[i].name, name)) {
61                         commit_format = &commit_formats[i];
62                         break;
63                 }
64         }
65
66         if (!commit_format) {
67                 ALLOC_GROW(commit_formats, commit_formats_len+1,
68                            commit_formats_alloc);
69                 commit_format = &commit_formats[commit_formats_len];
70                 memset(commit_format, 0, sizeof(*commit_format));
71                 commit_formats_len++;
72         }
73
74         commit_format->name = xstrdup(name);
75         commit_format->format = CMIT_FMT_USERFORMAT;
76         if (git_config_string(&fmt, var, value))
77                 return -1;
78
79         if (skip_prefix(fmt, "format:", &fmt))
80                 commit_format->is_tformat = 0;
81         else if (skip_prefix(fmt, "tformat:", &fmt) || strchr(fmt, '%'))
82                 commit_format->is_tformat = 1;
83         else
84                 commit_format->is_alias = 1;
85         commit_format->user_format = fmt;
86
87         return 0;
88 }
89
90 static void setup_commit_formats(void)
91 {
92         struct cmt_fmt_map builtin_formats[] = {
93                 { "raw",        CMIT_FMT_RAW,           0,      0 },
94                 { "medium",     CMIT_FMT_MEDIUM,        0,      8 },
95                 { "short",      CMIT_FMT_SHORT,         0,      0 },
96                 { "email",      CMIT_FMT_EMAIL,         0,      0 },
97                 { "mboxrd",     CMIT_FMT_MBOXRD,        0,      0 },
98                 { "fuller",     CMIT_FMT_FULLER,        0,      8 },
99                 { "full",       CMIT_FMT_FULL,          0,      8 },
100                 { "oneline",    CMIT_FMT_ONELINE,       1,      0 }
101                 /*
102                  * Please update $__git_log_pretty_formats in
103                  * git-completion.bash when you add new formats.
104                  */
105         };
106         commit_formats_len = ARRAY_SIZE(builtin_formats);
107         builtin_formats_len = commit_formats_len;
108         ALLOC_GROW(commit_formats, commit_formats_len, commit_formats_alloc);
109         COPY_ARRAY(commit_formats, builtin_formats,
110                    ARRAY_SIZE(builtin_formats));
111
112         git_config(git_pretty_formats_config, NULL);
113 }
114
115 static struct cmt_fmt_map *find_commit_format_recursive(const char *sought,
116                                                         const char *original,
117                                                         int num_redirections)
118 {
119         struct cmt_fmt_map *found = NULL;
120         size_t found_match_len = 0;
121         int i;
122
123         if (num_redirections >= commit_formats_len)
124                 die("invalid --pretty format: "
125                     "'%s' references an alias which points to itself",
126                     original);
127
128         for (i = 0; i < commit_formats_len; i++) {
129                 size_t match_len;
130
131                 if (!starts_with(commit_formats[i].name, sought))
132                         continue;
133
134                 match_len = strlen(commit_formats[i].name);
135                 if (found == NULL || found_match_len > match_len) {
136                         found = &commit_formats[i];
137                         found_match_len = match_len;
138                 }
139         }
140
141         if (found && found->is_alias) {
142                 found = find_commit_format_recursive(found->user_format,
143                                                      original,
144                                                      num_redirections+1);
145         }
146
147         return found;
148 }
149
150 static struct cmt_fmt_map *find_commit_format(const char *sought)
151 {
152         if (!commit_formats)
153                 setup_commit_formats();
154
155         return find_commit_format_recursive(sought, sought, 0);
156 }
157
158 void get_commit_format(const char *arg, struct rev_info *rev)
159 {
160         struct cmt_fmt_map *commit_format;
161
162         rev->use_terminator = 0;
163         if (!arg) {
164                 rev->commit_format = CMIT_FMT_DEFAULT;
165                 return;
166         }
167         if (skip_prefix(arg, "format:", &arg)) {
168                 save_user_format(rev, arg, 0);
169                 return;
170         }
171
172         if (!*arg || skip_prefix(arg, "tformat:", &arg) || strchr(arg, '%')) {
173                 save_user_format(rev, arg, 1);
174                 return;
175         }
176
177         commit_format = find_commit_format(arg);
178         if (!commit_format)
179                 die("invalid --pretty format: %s", arg);
180
181         rev->commit_format = commit_format->format;
182         rev->use_terminator = commit_format->is_tformat;
183         rev->expand_tabs_in_log_default = commit_format->expand_tabs_in_log;
184         if (commit_format->format == CMIT_FMT_USERFORMAT) {
185                 save_user_format(rev, commit_format->user_format,
186                                  commit_format->is_tformat);
187         }
188 }
189
190 /*
191  * Generic support for pretty-printing the header
192  */
193 static int get_one_line(const char *msg)
194 {
195         int ret = 0;
196
197         for (;;) {
198                 char c = *msg++;
199                 if (!c)
200                         break;
201                 ret++;
202                 if (c == '\n')
203                         break;
204         }
205         return ret;
206 }
207
208 /* High bit set, or ISO-2022-INT */
209 static int non_ascii(int ch)
210 {
211         return !isascii(ch) || ch == '\033';
212 }
213
214 int has_non_ascii(const char *s)
215 {
216         int ch;
217         if (!s)
218                 return 0;
219         while ((ch = *s++) != '\0') {
220                 if (non_ascii(ch))
221                         return 1;
222         }
223         return 0;
224 }
225
226 static int is_rfc822_special(char ch)
227 {
228         switch (ch) {
229         case '(':
230         case ')':
231         case '<':
232         case '>':
233         case '[':
234         case ']':
235         case ':':
236         case ';':
237         case '@':
238         case ',':
239         case '.':
240         case '"':
241         case '\\':
242                 return 1;
243         default:
244                 return 0;
245         }
246 }
247
248 static int needs_rfc822_quoting(const char *s, int len)
249 {
250         int i;
251         for (i = 0; i < len; i++)
252                 if (is_rfc822_special(s[i]))
253                         return 1;
254         return 0;
255 }
256
257 static int last_line_length(struct strbuf *sb)
258 {
259         int i;
260
261         /* How many bytes are already used on the last line? */
262         for (i = sb->len - 1; i >= 0; i--)
263                 if (sb->buf[i] == '\n')
264                         break;
265         return sb->len - (i + 1);
266 }
267
268 static void add_rfc822_quoted(struct strbuf *out, const char *s, int len)
269 {
270         int i;
271
272         /* just a guess, we may have to also backslash-quote */
273         strbuf_grow(out, len + 2);
274
275         strbuf_addch(out, '"');
276         for (i = 0; i < len; i++) {
277                 switch (s[i]) {
278                 case '"':
279                 case '\\':
280                         strbuf_addch(out, '\\');
281                         /* fall through */
282                 default:
283                         strbuf_addch(out, s[i]);
284                 }
285         }
286         strbuf_addch(out, '"');
287 }
288
289 enum rfc2047_type {
290         RFC2047_SUBJECT,
291         RFC2047_ADDRESS
292 };
293
294 static int is_rfc2047_special(char ch, enum rfc2047_type type)
295 {
296         /*
297          * rfc2047, section 4.2:
298          *
299          *    8-bit values which correspond to printable ASCII characters other
300          *    than "=", "?", and "_" (underscore), MAY be represented as those
301          *    characters.  (But see section 5 for restrictions.)  In
302          *    particular, SPACE and TAB MUST NOT be represented as themselves
303          *    within encoded words.
304          */
305
306         /*
307          * rule out non-ASCII characters and non-printable characters (the
308          * non-ASCII check should be redundant as isprint() is not localized
309          * and only knows about ASCII, but be defensive about that)
310          */
311         if (non_ascii(ch) || !isprint(ch))
312                 return 1;
313
314         /*
315          * rule out special printable characters (' ' should be the only
316          * whitespace character considered printable, but be defensive and use
317          * isspace())
318          */
319         if (isspace(ch) || ch == '=' || ch == '?' || ch == '_')
320                 return 1;
321
322         /*
323          * rfc2047, section 5.3:
324          *
325          *    As a replacement for a 'word' entity within a 'phrase', for example,
326          *    one that precedes an address in a From, To, or Cc header.  The ABNF
327          *    definition for 'phrase' from RFC 822 thus becomes:
328          *
329          *    phrase = 1*( encoded-word / word )
330          *
331          *    In this case the set of characters that may be used in a "Q"-encoded
332          *    'encoded-word' is restricted to: <upper and lower case ASCII
333          *    letters, decimal digits, "!", "*", "+", "-", "/", "=", and "_"
334          *    (underscore, ASCII 95.)>.  An 'encoded-word' that appears within a
335          *    'phrase' MUST be separated from any adjacent 'word', 'text' or
336          *    'special' by 'linear-white-space'.
337          */
338
339         if (type != RFC2047_ADDRESS)
340                 return 0;
341
342         /* '=' and '_' are special cases and have been checked above */
343         return !(isalnum(ch) || ch == '!' || ch == '*' || ch == '+' || ch == '-' || ch == '/');
344 }
345
346 static int needs_rfc2047_encoding(const char *line, int len)
347 {
348         int i;
349
350         for (i = 0; i < len; i++) {
351                 int ch = line[i];
352                 if (non_ascii(ch) || ch == '\n')
353                         return 1;
354                 if ((i + 1 < len) && (ch == '=' && line[i+1] == '?'))
355                         return 1;
356         }
357
358         return 0;
359 }
360
361 static void add_rfc2047(struct strbuf *sb, const char *line, size_t len,
362                        const char *encoding, enum rfc2047_type type)
363 {
364         static const int max_encoded_length = 76; /* per rfc2047 */
365         int i;
366         int line_len = last_line_length(sb);
367
368         strbuf_grow(sb, len * 3 + strlen(encoding) + 100);
369         strbuf_addf(sb, "=?%s?q?", encoding);
370         line_len += strlen(encoding) + 5; /* 5 for =??q? */
371
372         while (len) {
373                 /*
374                  * RFC 2047, section 5 (3):
375                  *
376                  * Each 'encoded-word' MUST represent an integral number of
377                  * characters.  A multi-octet character may not be split across
378                  * adjacent 'encoded- word's.
379                  */
380                 const unsigned char *p = (const unsigned char *)line;
381                 int chrlen = mbs_chrlen(&line, &len, encoding);
382                 int is_special = (chrlen > 1) || is_rfc2047_special(*p, type);
383
384                 /* "=%02X" * chrlen, or the byte itself */
385                 const char *encoded_fmt = is_special ? "=%02X"    : "%c";
386                 int         encoded_len = is_special ? 3 * chrlen : 1;
387
388                 /*
389                  * According to RFC 2047, we could encode the special character
390                  * ' ' (space) with '_' (underscore) for readability. But many
391                  * programs do not understand this and just leave the
392                  * underscore in place. Thus, we do nothing special here, which
393                  * causes ' ' to be encoded as '=20', avoiding this problem.
394                  */
395
396                 if (line_len + encoded_len + 2 > max_encoded_length) {
397                         /* It won't fit with trailing "?=" --- break the line */
398                         strbuf_addf(sb, "?=\n =?%s?q?", encoding);
399                         line_len = strlen(encoding) + 5 + 1; /* =??q? plus SP */
400                 }
401
402                 for (i = 0; i < chrlen; i++)
403                         strbuf_addf(sb, encoded_fmt, p[i]);
404                 line_len += encoded_len;
405         }
406         strbuf_addstr(sb, "?=");
407 }
408
409 const char *show_ident_date(const struct ident_split *ident,
410                             const struct date_mode *mode)
411 {
412         timestamp_t date = 0;
413         long tz = 0;
414
415         if (ident->date_begin && ident->date_end)
416                 date = parse_timestamp(ident->date_begin, NULL, 10);
417         if (date_overflows(date))
418                 date = 0;
419         else {
420                 if (ident->tz_begin && ident->tz_end)
421                         tz = strtol(ident->tz_begin, NULL, 10);
422                 if (tz >= INT_MAX || tz <= INT_MIN)
423                         tz = 0;
424         }
425         return show_date(date, tz, mode);
426 }
427
428 void pp_user_info(struct pretty_print_context *pp,
429                   const char *what, struct strbuf *sb,
430                   const char *line, const char *encoding)
431 {
432         struct ident_split ident;
433         char *line_end;
434         const char *mailbuf, *namebuf;
435         size_t namelen, maillen;
436         int max_length = 78; /* per rfc2822 */
437
438         if (pp->fmt == CMIT_FMT_ONELINE)
439                 return;
440
441         line_end = strchrnul(line, '\n');
442         if (split_ident_line(&ident, line, line_end - line))
443                 return;
444
445         mailbuf = ident.mail_begin;
446         maillen = ident.mail_end - ident.mail_begin;
447         namebuf = ident.name_begin;
448         namelen = ident.name_end - ident.name_begin;
449
450         if (pp->mailmap)
451                 map_user(pp->mailmap, &mailbuf, &maillen, &namebuf, &namelen);
452
453         if (cmit_fmt_is_mail(pp->fmt)) {
454                 if (pp->from_ident && ident_cmp(pp->from_ident, &ident)) {
455                         struct strbuf buf = STRBUF_INIT;
456
457                         strbuf_addstr(&buf, "From: ");
458                         strbuf_add(&buf, namebuf, namelen);
459                         strbuf_addstr(&buf, " <");
460                         strbuf_add(&buf, mailbuf, maillen);
461                         strbuf_addstr(&buf, ">\n");
462                         string_list_append(&pp->in_body_headers,
463                                            strbuf_detach(&buf, NULL));
464
465                         mailbuf = pp->from_ident->mail_begin;
466                         maillen = pp->from_ident->mail_end - mailbuf;
467                         namebuf = pp->from_ident->name_begin;
468                         namelen = pp->from_ident->name_end - namebuf;
469                 }
470
471                 strbuf_addstr(sb, "From: ");
472                 if (needs_rfc2047_encoding(namebuf, namelen)) {
473                         add_rfc2047(sb, namebuf, namelen,
474                                     encoding, RFC2047_ADDRESS);
475                         max_length = 76; /* per rfc2047 */
476                 } else if (needs_rfc822_quoting(namebuf, namelen)) {
477                         struct strbuf quoted = STRBUF_INIT;
478                         add_rfc822_quoted(&quoted, namebuf, namelen);
479                         strbuf_add_wrapped_bytes(sb, quoted.buf, quoted.len,
480                                                         -6, 1, max_length);
481                         strbuf_release(&quoted);
482                 } else {
483                         strbuf_add_wrapped_bytes(sb, namebuf, namelen,
484                                                  -6, 1, max_length);
485                 }
486
487                 if (max_length <
488                     last_line_length(sb) + strlen(" <") + maillen + strlen(">"))
489                         strbuf_addch(sb, '\n');
490                 strbuf_addf(sb, " <%.*s>\n", (int)maillen, mailbuf);
491         } else {
492                 strbuf_addf(sb, "%s: %.*s%.*s <%.*s>\n", what,
493                             (pp->fmt == CMIT_FMT_FULLER) ? 4 : 0, "    ",
494                             (int)namelen, namebuf, (int)maillen, mailbuf);
495         }
496
497         switch (pp->fmt) {
498         case CMIT_FMT_MEDIUM:
499                 strbuf_addf(sb, "Date:   %s\n",
500                             show_ident_date(&ident, &pp->date_mode));
501                 break;
502         case CMIT_FMT_EMAIL:
503         case CMIT_FMT_MBOXRD:
504                 strbuf_addf(sb, "Date: %s\n",
505                             show_ident_date(&ident, DATE_MODE(RFC2822)));
506                 break;
507         case CMIT_FMT_FULLER:
508                 strbuf_addf(sb, "%sDate: %s\n", what,
509                             show_ident_date(&ident, &pp->date_mode));
510                 break;
511         default:
512                 /* notin' */
513                 break;
514         }
515 }
516
517 static int is_blank_line(const char *line, int *len_p)
518 {
519         int len = *len_p;
520         while (len && isspace(line[len - 1]))
521                 len--;
522         *len_p = len;
523         return !len;
524 }
525
526 const char *skip_blank_lines(const char *msg)
527 {
528         for (;;) {
529                 int linelen = get_one_line(msg);
530                 int ll = linelen;
531                 if (!linelen)
532                         break;
533                 if (!is_blank_line(msg, &ll))
534                         break;
535                 msg += linelen;
536         }
537         return msg;
538 }
539
540 static void add_merge_info(const struct pretty_print_context *pp,
541                            struct strbuf *sb, const struct commit *commit)
542 {
543         struct commit_list *parent = commit->parents;
544
545         if ((pp->fmt == CMIT_FMT_ONELINE) || (cmit_fmt_is_mail(pp->fmt)) ||
546             !parent || !parent->next)
547                 return;
548
549         strbuf_addstr(sb, "Merge:");
550
551         while (parent) {
552                 struct object_id *oidp = &parent->item->object.oid;
553                 strbuf_addch(sb, ' ');
554                 if (pp->abbrev)
555                         strbuf_add_unique_abbrev(sb, oidp, pp->abbrev);
556                 else
557                         strbuf_addstr(sb, oid_to_hex(oidp));
558                 parent = parent->next;
559         }
560         strbuf_addch(sb, '\n');
561 }
562
563 static char *get_header(const char *msg, const char *key)
564 {
565         size_t len;
566         const char *v = find_commit_header(msg, key, &len);
567         return v ? xmemdupz(v, len) : NULL;
568 }
569
570 static char *replace_encoding_header(char *buf, const char *encoding)
571 {
572         struct strbuf tmp = STRBUF_INIT;
573         size_t start, len;
574         char *cp = buf;
575
576         /* guess if there is an encoding header before a \n\n */
577         while (!starts_with(cp, "encoding ")) {
578                 cp = strchr(cp, '\n');
579                 if (!cp || *++cp == '\n')
580                         return buf;
581         }
582         start = cp - buf;
583         cp = strchr(cp, '\n');
584         if (!cp)
585                 return buf; /* should not happen but be defensive */
586         len = cp + 1 - (buf + start);
587
588         strbuf_attach(&tmp, buf, strlen(buf), strlen(buf) + 1);
589         if (is_encoding_utf8(encoding)) {
590                 /* we have re-coded to UTF-8; drop the header */
591                 strbuf_remove(&tmp, start, len);
592         } else {
593                 /* just replaces XXXX in 'encoding XXXX\n' */
594                 strbuf_splice(&tmp, start + strlen("encoding "),
595                                           len - strlen("encoding \n"),
596                                           encoding, strlen(encoding));
597         }
598         return strbuf_detach(&tmp, NULL);
599 }
600
601 const char *repo_logmsg_reencode(struct repository *r,
602                                  const struct commit *commit,
603                                  char **commit_encoding,
604                                  const char *output_encoding)
605 {
606         static const char *utf8 = "UTF-8";
607         const char *use_encoding;
608         char *encoding;
609         const char *msg = repo_get_commit_buffer(r, commit, NULL);
610         char *out;
611
612         if (!output_encoding || !*output_encoding) {
613                 if (commit_encoding)
614                         *commit_encoding = get_header(msg, "encoding");
615                 return msg;
616         }
617         encoding = get_header(msg, "encoding");
618         if (commit_encoding)
619                 *commit_encoding = encoding;
620         use_encoding = encoding ? encoding : utf8;
621         if (same_encoding(use_encoding, output_encoding)) {
622                 /*
623                  * No encoding work to be done. If we have no encoding header
624                  * at all, then there's nothing to do, and we can return the
625                  * message verbatim (whether newly allocated or not).
626                  */
627                 if (!encoding)
628                         return msg;
629
630                 /*
631                  * Otherwise, we still want to munge the encoding header in the
632                  * result, which will be done by modifying the buffer. If we
633                  * are using a fresh copy, we can reuse it. But if we are using
634                  * the cached copy from get_commit_buffer, we need to duplicate it
635                  * to avoid munging the cached copy.
636                  */
637                 if (msg == get_cached_commit_buffer(r, commit, NULL))
638                         out = xstrdup(msg);
639                 else
640                         out = (char *)msg;
641         }
642         else {
643                 /*
644                  * There's actual encoding work to do. Do the reencoding, which
645                  * still leaves the header to be replaced in the next step. At
646                  * this point, we are done with msg. If we allocated a fresh
647                  * copy, we can free it.
648                  */
649                 out = reencode_string(msg, output_encoding, use_encoding);
650                 if (out)
651                         repo_unuse_commit_buffer(r, commit, msg);
652         }
653
654         /*
655          * This replacement actually consumes the buffer we hand it, so we do
656          * not have to worry about freeing the old "out" here.
657          */
658         if (out)
659                 out = replace_encoding_header(out, output_encoding);
660
661         if (!commit_encoding)
662                 free(encoding);
663         /*
664          * If the re-encoding failed, out might be NULL here; in that
665          * case we just return the commit message verbatim.
666          */
667         return out ? out : msg;
668 }
669
670 static int mailmap_name(const char **email, size_t *email_len,
671                         const char **name, size_t *name_len)
672 {
673         static struct string_list *mail_map;
674         if (!mail_map) {
675                 mail_map = xcalloc(1, sizeof(*mail_map));
676                 read_mailmap(mail_map, NULL);
677         }
678         return mail_map->nr && map_user(mail_map, email, email_len, name, name_len);
679 }
680
681 static size_t format_person_part(struct strbuf *sb, char part,
682                                  const char *msg, int len,
683                                  const struct date_mode *dmode)
684 {
685         /* currently all placeholders have same length */
686         const int placeholder_len = 2;
687         struct ident_split s;
688         const char *name, *mail;
689         size_t maillen, namelen;
690
691         if (split_ident_line(&s, msg, len) < 0)
692                 goto skip;
693
694         name = s.name_begin;
695         namelen = s.name_end - s.name_begin;
696         mail = s.mail_begin;
697         maillen = s.mail_end - s.mail_begin;
698
699         if (part == 'N' || part == 'E') /* mailmap lookup */
700                 mailmap_name(&mail, &maillen, &name, &namelen);
701         if (part == 'n' || part == 'N') {       /* name */
702                 strbuf_add(sb, name, namelen);
703                 return placeholder_len;
704         }
705         if (part == 'e' || part == 'E') {       /* email */
706                 strbuf_add(sb, mail, maillen);
707                 return placeholder_len;
708         }
709
710         if (!s.date_begin)
711                 goto skip;
712
713         if (part == 't') {      /* date, UNIX timestamp */
714                 strbuf_add(sb, s.date_begin, s.date_end - s.date_begin);
715                 return placeholder_len;
716         }
717
718         switch (part) {
719         case 'd':       /* date */
720                 strbuf_addstr(sb, show_ident_date(&s, dmode));
721                 return placeholder_len;
722         case 'D':       /* date, RFC2822 style */
723                 strbuf_addstr(sb, show_ident_date(&s, DATE_MODE(RFC2822)));
724                 return placeholder_len;
725         case 'r':       /* date, relative */
726                 strbuf_addstr(sb, show_ident_date(&s, DATE_MODE(RELATIVE)));
727                 return placeholder_len;
728         case 'i':       /* date, ISO 8601-like */
729                 strbuf_addstr(sb, show_ident_date(&s, DATE_MODE(ISO8601)));
730                 return placeholder_len;
731         case 'I':       /* date, ISO 8601 strict */
732                 strbuf_addstr(sb, show_ident_date(&s, DATE_MODE(ISO8601_STRICT)));
733                 return placeholder_len;
734         }
735
736 skip:
737         /*
738          * reading from either a bogus commit, or a reflog entry with
739          * %gn, %ge, etc.; 'sb' cannot be updated, but we still need
740          * to compute a valid return value.
741          */
742         if (part == 'n' || part == 'e' || part == 't' || part == 'd'
743             || part == 'D' || part == 'r' || part == 'i')
744                 return placeholder_len;
745
746         return 0; /* unknown placeholder */
747 }
748
749 struct chunk {
750         size_t off;
751         size_t len;
752 };
753
754 enum flush_type {
755         no_flush,
756         flush_right,
757         flush_left,
758         flush_left_and_steal,
759         flush_both
760 };
761
762 enum trunc_type {
763         trunc_none,
764         trunc_left,
765         trunc_middle,
766         trunc_right
767 };
768
769 struct format_commit_context {
770         const struct commit *commit;
771         const struct pretty_print_context *pretty_ctx;
772         unsigned commit_header_parsed:1;
773         unsigned commit_message_parsed:1;
774         struct signature_check signature_check;
775         enum flush_type flush_type;
776         enum trunc_type truncate;
777         const char *message;
778         char *commit_encoding;
779         size_t width, indent1, indent2;
780         int auto_color;
781         int padding;
782
783         /* These offsets are relative to the start of the commit message. */
784         struct chunk author;
785         struct chunk committer;
786         size_t message_off;
787         size_t subject_off;
788         size_t body_off;
789
790         /* The following ones are relative to the result struct strbuf. */
791         size_t wrap_start;
792 };
793
794 static void parse_commit_header(struct format_commit_context *context)
795 {
796         const char *msg = context->message;
797         int i;
798
799         for (i = 0; msg[i]; i++) {
800                 const char *name;
801                 int eol;
802                 for (eol = i; msg[eol] && msg[eol] != '\n'; eol++)
803                         ; /* do nothing */
804
805                 if (i == eol) {
806                         break;
807                 } else if (skip_prefix(msg + i, "author ", &name)) {
808                         context->author.off = name - msg;
809                         context->author.len = msg + eol - name;
810                 } else if (skip_prefix(msg + i, "committer ", &name)) {
811                         context->committer.off = name - msg;
812                         context->committer.len = msg + eol - name;
813                 }
814                 i = eol;
815         }
816         context->message_off = i;
817         context->commit_header_parsed = 1;
818 }
819
820 static int istitlechar(char c)
821 {
822         return (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') ||
823                 (c >= '0' && c <= '9') || c == '.' || c == '_';
824 }
825
826 static void format_sanitized_subject(struct strbuf *sb, const char *msg)
827 {
828         size_t trimlen;
829         size_t start_len = sb->len;
830         int space = 2;
831
832         for (; *msg && *msg != '\n'; msg++) {
833                 if (istitlechar(*msg)) {
834                         if (space == 1)
835                                 strbuf_addch(sb, '-');
836                         space = 0;
837                         strbuf_addch(sb, *msg);
838                         if (*msg == '.')
839                                 while (*(msg+1) == '.')
840                                         msg++;
841                 } else
842                         space |= 1;
843         }
844
845         /* trim any trailing '.' or '-' characters */
846         trimlen = 0;
847         while (sb->len - trimlen > start_len &&
848                 (sb->buf[sb->len - 1 - trimlen] == '.'
849                 || sb->buf[sb->len - 1 - trimlen] == '-'))
850                 trimlen++;
851         strbuf_remove(sb, sb->len - trimlen, trimlen);
852 }
853
854 const char *format_subject(struct strbuf *sb, const char *msg,
855                            const char *line_separator)
856 {
857         int first = 1;
858
859         for (;;) {
860                 const char *line = msg;
861                 int linelen = get_one_line(line);
862
863                 msg += linelen;
864                 if (!linelen || is_blank_line(line, &linelen))
865                         break;
866
867                 if (!sb)
868                         continue;
869                 strbuf_grow(sb, linelen + 2);
870                 if (!first)
871                         strbuf_addstr(sb, line_separator);
872                 strbuf_add(sb, line, linelen);
873                 first = 0;
874         }
875         return msg;
876 }
877
878 static void parse_commit_message(struct format_commit_context *c)
879 {
880         const char *msg = c->message + c->message_off;
881         const char *start = c->message;
882
883         msg = skip_blank_lines(msg);
884         c->subject_off = msg - start;
885
886         msg = format_subject(NULL, msg, NULL);
887         msg = skip_blank_lines(msg);
888         c->body_off = msg - start;
889
890         c->commit_message_parsed = 1;
891 }
892
893 static void strbuf_wrap(struct strbuf *sb, size_t pos,
894                         size_t width, size_t indent1, size_t indent2)
895 {
896         struct strbuf tmp = STRBUF_INIT;
897
898         if (pos)
899                 strbuf_add(&tmp, sb->buf, pos);
900         strbuf_add_wrapped_text(&tmp, sb->buf + pos,
901                                 (int) indent1, (int) indent2, (int) width);
902         strbuf_swap(&tmp, sb);
903         strbuf_release(&tmp);
904 }
905
906 static void rewrap_message_tail(struct strbuf *sb,
907                                 struct format_commit_context *c,
908                                 size_t new_width, size_t new_indent1,
909                                 size_t new_indent2)
910 {
911         if (c->width == new_width && c->indent1 == new_indent1 &&
912             c->indent2 == new_indent2)
913                 return;
914         if (c->wrap_start < sb->len)
915                 strbuf_wrap(sb, c->wrap_start, c->width, c->indent1, c->indent2);
916         c->wrap_start = sb->len;
917         c->width = new_width;
918         c->indent1 = new_indent1;
919         c->indent2 = new_indent2;
920 }
921
922 static int format_reflog_person(struct strbuf *sb,
923                                 char part,
924                                 struct reflog_walk_info *log,
925                                 const struct date_mode *dmode)
926 {
927         const char *ident;
928
929         if (!log)
930                 return 2;
931
932         ident = get_reflog_ident(log);
933         if (!ident)
934                 return 2;
935
936         return format_person_part(sb, part, ident, strlen(ident), dmode);
937 }
938
939 static size_t parse_color(struct strbuf *sb, /* in UTF-8 */
940                           const char *placeholder,
941                           struct format_commit_context *c)
942 {
943         const char *rest = placeholder;
944         const char *basic_color = NULL;
945
946         if (placeholder[1] == '(') {
947                 const char *begin = placeholder + 2;
948                 const char *end = strchr(begin, ')');
949                 char color[COLOR_MAXLEN];
950
951                 if (!end)
952                         return 0;
953
954                 if (skip_prefix(begin, "auto,", &begin)) {
955                         if (!want_color(c->pretty_ctx->color))
956                                 return end - placeholder + 1;
957                 } else if (skip_prefix(begin, "always,", &begin)) {
958                         /* nothing to do; we do not respect want_color at all */
959                 } else {
960                         /* the default is the same as "auto" */
961                         if (!want_color(c->pretty_ctx->color))
962                                 return end - placeholder + 1;
963                 }
964
965                 if (color_parse_mem(begin, end - begin, color) < 0)
966                         die(_("unable to parse --pretty format"));
967                 strbuf_addstr(sb, color);
968                 return end - placeholder + 1;
969         }
970
971         /*
972          * We handle things like "%C(red)" above; for historical reasons, there
973          * are a few colors that can be specified without parentheses (and
974          * they cannot support things like "auto" or "always" at all).
975          */
976         if (skip_prefix(placeholder + 1, "red", &rest))
977                 basic_color = GIT_COLOR_RED;
978         else if (skip_prefix(placeholder + 1, "green", &rest))
979                 basic_color = GIT_COLOR_GREEN;
980         else if (skip_prefix(placeholder + 1, "blue", &rest))
981                 basic_color = GIT_COLOR_BLUE;
982         else if (skip_prefix(placeholder + 1, "reset", &rest))
983                 basic_color = GIT_COLOR_RESET;
984
985         if (basic_color && want_color(c->pretty_ctx->color))
986                 strbuf_addstr(sb, basic_color);
987
988         return rest - placeholder;
989 }
990
991 static size_t parse_padding_placeholder(const char *placeholder,
992                                         struct format_commit_context *c)
993 {
994         const char *ch = placeholder;
995         enum flush_type flush_type;
996         int to_column = 0;
997
998         switch (*ch++) {
999         case '<':
1000                 flush_type = flush_right;
1001                 break;
1002         case '>':
1003                 if (*ch == '<') {
1004                         flush_type = flush_both;
1005                         ch++;
1006                 } else if (*ch == '>') {
1007                         flush_type = flush_left_and_steal;
1008                         ch++;
1009                 } else
1010                         flush_type = flush_left;
1011                 break;
1012         default:
1013                 return 0;
1014         }
1015
1016         /* the next value means "wide enough to that column" */
1017         if (*ch == '|') {
1018                 to_column = 1;
1019                 ch++;
1020         }
1021
1022         if (*ch == '(') {
1023                 const char *start = ch + 1;
1024                 const char *end = start + strcspn(start, ",)");
1025                 char *next;
1026                 int width;
1027                 if (!end || end == start)
1028                         return 0;
1029                 width = strtol(start, &next, 10);
1030                 if (next == start || width == 0)
1031                         return 0;
1032                 if (width < 0) {
1033                         if (to_column)
1034                                 width += term_columns();
1035                         if (width < 0)
1036                                 return 0;
1037                 }
1038                 c->padding = to_column ? -width : width;
1039                 c->flush_type = flush_type;
1040
1041                 if (*end == ',') {
1042                         start = end + 1;
1043                         end = strchr(start, ')');
1044                         if (!end || end == start)
1045                                 return 0;
1046                         if (starts_with(start, "trunc)"))
1047                                 c->truncate = trunc_right;
1048                         else if (starts_with(start, "ltrunc)"))
1049                                 c->truncate = trunc_left;
1050                         else if (starts_with(start, "mtrunc)"))
1051                                 c->truncate = trunc_middle;
1052                         else
1053                                 return 0;
1054                 } else
1055                         c->truncate = trunc_none;
1056
1057                 return end - placeholder + 1;
1058         }
1059         return 0;
1060 }
1061
1062 static int match_placeholder_arg_value(const char *to_parse, const char *candidate,
1063                                        const char **end, const char **valuestart,
1064                                        size_t *valuelen)
1065 {
1066         const char *p;
1067
1068         if (!(skip_prefix(to_parse, candidate, &p)))
1069                 return 0;
1070         if (valuestart) {
1071                 if (*p == '=') {
1072                         *valuestart = p + 1;
1073                         *valuelen = strcspn(*valuestart, ",)");
1074                         p = *valuestart + *valuelen;
1075                 } else {
1076                         if (*p != ',' && *p != ')')
1077                                 return 0;
1078                         *valuestart = NULL;
1079                         *valuelen = 0;
1080                 }
1081         }
1082         if (*p == ',') {
1083                 *end = p + 1;
1084                 return 1;
1085         }
1086         if (*p == ')') {
1087                 *end = p;
1088                 return 1;
1089         }
1090         return 0;
1091 }
1092
1093 static int match_placeholder_bool_arg(const char *to_parse, const char *candidate,
1094                                       const char **end, int *val)
1095 {
1096         const char *argval;
1097         char *strval;
1098         size_t arglen;
1099         int v;
1100
1101         if (!match_placeholder_arg_value(to_parse, candidate, end, &argval, &arglen))
1102                 return 0;
1103
1104         if (!argval) {
1105                 *val = 1;
1106                 return 1;
1107         }
1108
1109         strval = xstrndup(argval, arglen);
1110         v = git_parse_maybe_bool(strval);
1111         free(strval);
1112
1113         if (v == -1)
1114                 return 0;
1115
1116         *val = v;
1117
1118         return 1;
1119 }
1120
1121 static int format_trailer_match_cb(const struct strbuf *key, void *ud)
1122 {
1123         const struct string_list *list = ud;
1124         const struct string_list_item *item;
1125
1126         for_each_string_list_item (item, list) {
1127                 if (key->len == (uintptr_t)item->util &&
1128                     !strncasecmp(item->string, key->buf, key->len))
1129                         return 1;
1130         }
1131         return 0;
1132 }
1133
1134 static size_t format_commit_one(struct strbuf *sb, /* in UTF-8 */
1135                                 const char *placeholder,
1136                                 void *context)
1137 {
1138         struct format_commit_context *c = context;
1139         const struct commit *commit = c->commit;
1140         const char *msg = c->message;
1141         struct commit_list *p;
1142         const char *arg;
1143         size_t res;
1144         char **slot;
1145
1146         /* these are independent of the commit */
1147         res = strbuf_expand_literal_cb(sb, placeholder, NULL);
1148         if (res)
1149                 return res;
1150
1151         switch (placeholder[0]) {
1152         case 'C':
1153                 if (starts_with(placeholder + 1, "(auto)")) {
1154                         c->auto_color = want_color(c->pretty_ctx->color);
1155                         if (c->auto_color && sb->len)
1156                                 strbuf_addstr(sb, GIT_COLOR_RESET);
1157                         return 7; /* consumed 7 bytes, "C(auto)" */
1158                 } else {
1159                         int ret = parse_color(sb, placeholder, c);
1160                         if (ret)
1161                                 c->auto_color = 0;
1162                         /*
1163                          * Otherwise, we decided to treat %C<unknown>
1164                          * as a literal string, and the previous
1165                          * %C(auto) is still valid.
1166                          */
1167                         return ret;
1168                 }
1169         case 'w':
1170                 if (placeholder[1] == '(') {
1171                         unsigned long width = 0, indent1 = 0, indent2 = 0;
1172                         char *next;
1173                         const char *start = placeholder + 2;
1174                         const char *end = strchr(start, ')');
1175                         if (!end)
1176                                 return 0;
1177                         if (end > start) {
1178                                 width = strtoul(start, &next, 10);
1179                                 if (*next == ',') {
1180                                         indent1 = strtoul(next + 1, &next, 10);
1181                                         if (*next == ',') {
1182                                                 indent2 = strtoul(next + 1,
1183                                                                  &next, 10);
1184                                         }
1185                                 }
1186                                 if (*next != ')')
1187                                         return 0;
1188                         }
1189                         rewrap_message_tail(sb, c, width, indent1, indent2);
1190                         return end - placeholder + 1;
1191                 } else
1192                         return 0;
1193
1194         case '<':
1195         case '>':
1196                 return parse_padding_placeholder(placeholder, c);
1197         }
1198
1199         /* these depend on the commit */
1200         if (!commit->object.parsed)
1201                 parse_object(the_repository, &commit->object.oid);
1202
1203         switch (placeholder[0]) {
1204         case 'H':               /* commit hash */
1205                 strbuf_addstr(sb, diff_get_color(c->auto_color, DIFF_COMMIT));
1206                 strbuf_addstr(sb, oid_to_hex(&commit->object.oid));
1207                 strbuf_addstr(sb, diff_get_color(c->auto_color, DIFF_RESET));
1208                 return 1;
1209         case 'h':               /* abbreviated commit hash */
1210                 strbuf_addstr(sb, diff_get_color(c->auto_color, DIFF_COMMIT));
1211                 strbuf_add_unique_abbrev(sb, &commit->object.oid,
1212                                          c->pretty_ctx->abbrev);
1213                 strbuf_addstr(sb, diff_get_color(c->auto_color, DIFF_RESET));
1214                 return 1;
1215         case 'T':               /* tree hash */
1216                 strbuf_addstr(sb, oid_to_hex(get_commit_tree_oid(commit)));
1217                 return 1;
1218         case 't':               /* abbreviated tree hash */
1219                 strbuf_add_unique_abbrev(sb,
1220                                          get_commit_tree_oid(commit),
1221                                          c->pretty_ctx->abbrev);
1222                 return 1;
1223         case 'P':               /* parent hashes */
1224                 for (p = commit->parents; p; p = p->next) {
1225                         if (p != commit->parents)
1226                                 strbuf_addch(sb, ' ');
1227                         strbuf_addstr(sb, oid_to_hex(&p->item->object.oid));
1228                 }
1229                 return 1;
1230         case 'p':               /* abbreviated parent hashes */
1231                 for (p = commit->parents; p; p = p->next) {
1232                         if (p != commit->parents)
1233                                 strbuf_addch(sb, ' ');
1234                         strbuf_add_unique_abbrev(sb, &p->item->object.oid,
1235                                                  c->pretty_ctx->abbrev);
1236                 }
1237                 return 1;
1238         case 'm':               /* left/right/bottom */
1239                 strbuf_addstr(sb, get_revision_mark(NULL, commit));
1240                 return 1;
1241         case 'd':
1242                 format_decorations(sb, commit, c->auto_color);
1243                 return 1;
1244         case 'D':
1245                 format_decorations_extended(sb, commit, c->auto_color, "", ", ", "");
1246                 return 1;
1247         case 'S':               /* tag/branch like --source */
1248                 if (!(c->pretty_ctx->rev && c->pretty_ctx->rev->sources))
1249                         return 0;
1250                 slot = revision_sources_at(c->pretty_ctx->rev->sources, commit);
1251                 if (!(slot && *slot))
1252                         return 0;
1253                 strbuf_addstr(sb, *slot);
1254                 return 1;
1255         case 'g':               /* reflog info */
1256                 switch(placeholder[1]) {
1257                 case 'd':       /* reflog selector */
1258                 case 'D':
1259                         if (c->pretty_ctx->reflog_info)
1260                                 get_reflog_selector(sb,
1261                                                     c->pretty_ctx->reflog_info,
1262                                                     &c->pretty_ctx->date_mode,
1263                                                     c->pretty_ctx->date_mode_explicit,
1264                                                     (placeholder[1] == 'd'));
1265                         return 2;
1266                 case 's':       /* reflog message */
1267                         if (c->pretty_ctx->reflog_info)
1268                                 get_reflog_message(sb, c->pretty_ctx->reflog_info);
1269                         return 2;
1270                 case 'n':
1271                 case 'N':
1272                 case 'e':
1273                 case 'E':
1274                         return format_reflog_person(sb,
1275                                                     placeholder[1],
1276                                                     c->pretty_ctx->reflog_info,
1277                                                     &c->pretty_ctx->date_mode);
1278                 }
1279                 return 0;       /* unknown %g placeholder */
1280         case 'N':
1281                 if (c->pretty_ctx->notes_message) {
1282                         strbuf_addstr(sb, c->pretty_ctx->notes_message);
1283                         return 1;
1284                 }
1285                 return 0;
1286         }
1287
1288         if (placeholder[0] == 'G') {
1289                 if (!c->signature_check.result)
1290                         check_commit_signature(c->commit, &(c->signature_check));
1291                 switch (placeholder[1]) {
1292                 case 'G':
1293                         if (c->signature_check.gpg_output)
1294                                 strbuf_addstr(sb, c->signature_check.gpg_output);
1295                         break;
1296                 case '?':
1297                         switch (c->signature_check.result) {
1298                         case 'G':
1299                         case 'B':
1300                         case 'E':
1301                         case 'U':
1302                         case 'N':
1303                         case 'X':
1304                         case 'Y':
1305                         case 'R':
1306                                 strbuf_addch(sb, c->signature_check.result);
1307                         }
1308                         break;
1309                 case 'S':
1310                         if (c->signature_check.signer)
1311                                 strbuf_addstr(sb, c->signature_check.signer);
1312                         break;
1313                 case 'K':
1314                         if (c->signature_check.key)
1315                                 strbuf_addstr(sb, c->signature_check.key);
1316                         break;
1317                 case 'F':
1318                         if (c->signature_check.fingerprint)
1319                                 strbuf_addstr(sb, c->signature_check.fingerprint);
1320                         break;
1321                 case 'P':
1322                         if (c->signature_check.primary_key_fingerprint)
1323                                 strbuf_addstr(sb, c->signature_check.primary_key_fingerprint);
1324                         break;
1325                 default:
1326                         return 0;
1327                 }
1328                 return 2;
1329         }
1330
1331
1332         /* For the rest we have to parse the commit header. */
1333         if (!c->commit_header_parsed)
1334                 parse_commit_header(c);
1335
1336         switch (placeholder[0]) {
1337         case 'a':       /* author ... */
1338                 return format_person_part(sb, placeholder[1],
1339                                    msg + c->author.off, c->author.len,
1340                                    &c->pretty_ctx->date_mode);
1341         case 'c':       /* committer ... */
1342                 return format_person_part(sb, placeholder[1],
1343                                    msg + c->committer.off, c->committer.len,
1344                                    &c->pretty_ctx->date_mode);
1345         case 'e':       /* encoding */
1346                 if (c->commit_encoding)
1347                         strbuf_addstr(sb, c->commit_encoding);
1348                 return 1;
1349         case 'B':       /* raw body */
1350                 /* message_off is always left at the initial newline */
1351                 strbuf_addstr(sb, msg + c->message_off + 1);
1352                 return 1;
1353         }
1354
1355         /* Now we need to parse the commit message. */
1356         if (!c->commit_message_parsed)
1357                 parse_commit_message(c);
1358
1359         switch (placeholder[0]) {
1360         case 's':       /* subject */
1361                 format_subject(sb, msg + c->subject_off, " ");
1362                 return 1;
1363         case 'f':       /* sanitized subject */
1364                 format_sanitized_subject(sb, msg + c->subject_off);
1365                 return 1;
1366         case 'b':       /* body */
1367                 strbuf_addstr(sb, msg + c->body_off);
1368                 return 1;
1369         }
1370
1371         if (skip_prefix(placeholder, "(trailers", &arg)) {
1372                 struct process_trailer_options opts = PROCESS_TRAILER_OPTIONS_INIT;
1373                 struct string_list filter_list = STRING_LIST_INIT_NODUP;
1374                 struct strbuf sepbuf = STRBUF_INIT;
1375                 size_t ret = 0;
1376
1377                 opts.no_divider = 1;
1378
1379                 if (*arg == ':') {
1380                         arg++;
1381                         for (;;) {
1382                                 const char *argval;
1383                                 size_t arglen;
1384
1385                                 if (match_placeholder_arg_value(arg, "key", &arg, &argval, &arglen)) {
1386                                         uintptr_t len = arglen;
1387
1388                                         if (!argval)
1389                                                 goto trailer_out;
1390
1391                                         if (len && argval[len - 1] == ':')
1392                                                 len--;
1393                                         string_list_append(&filter_list, argval)->util = (char *)len;
1394
1395                                         opts.filter = format_trailer_match_cb;
1396                                         opts.filter_data = &filter_list;
1397                                         opts.only_trailers = 1;
1398                                 } else if (match_placeholder_arg_value(arg, "separator", &arg, &argval, &arglen)) {
1399                                         char *fmt;
1400
1401                                         strbuf_reset(&sepbuf);
1402                                         fmt = xstrndup(argval, arglen);
1403                                         strbuf_expand(&sepbuf, fmt, strbuf_expand_literal_cb, NULL);
1404                                         free(fmt);
1405                                         opts.separator = &sepbuf;
1406                                 } else if (!match_placeholder_bool_arg(arg, "only", &arg, &opts.only_trailers) &&
1407                                            !match_placeholder_bool_arg(arg, "unfold", &arg, &opts.unfold) &&
1408                                            !match_placeholder_bool_arg(arg, "valueonly", &arg, &opts.value_only))
1409                                         break;
1410                         }
1411                 }
1412                 if (*arg == ')') {
1413                         format_trailers_from_commit(sb, msg + c->subject_off, &opts);
1414                         ret = arg - placeholder + 1;
1415                 }
1416         trailer_out:
1417                 string_list_clear(&filter_list, 0);
1418                 strbuf_release(&sepbuf);
1419                 return ret;
1420         }
1421
1422         return 0;       /* unknown placeholder */
1423 }
1424
1425 static size_t format_and_pad_commit(struct strbuf *sb, /* in UTF-8 */
1426                                     const char *placeholder,
1427                                     struct format_commit_context *c)
1428 {
1429         struct strbuf local_sb = STRBUF_INIT;
1430         int total_consumed = 0, len, padding = c->padding;
1431         if (padding < 0) {
1432                 const char *start = strrchr(sb->buf, '\n');
1433                 int occupied;
1434                 if (!start)
1435                         start = sb->buf;
1436                 occupied = utf8_strnwidth(start, -1, 1);
1437                 occupied += c->pretty_ctx->graph_width;
1438                 padding = (-padding) - occupied;
1439         }
1440         while (1) {
1441                 int modifier = *placeholder == 'C';
1442                 int consumed = format_commit_one(&local_sb, placeholder, c);
1443                 total_consumed += consumed;
1444
1445                 if (!modifier)
1446                         break;
1447
1448                 placeholder += consumed;
1449                 if (*placeholder != '%')
1450                         break;
1451                 placeholder++;
1452                 total_consumed++;
1453         }
1454         len = utf8_strnwidth(local_sb.buf, -1, 1);
1455
1456         if (c->flush_type == flush_left_and_steal) {
1457                 const char *ch = sb->buf + sb->len - 1;
1458                 while (len > padding && ch > sb->buf) {
1459                         const char *p;
1460                         if (*ch == ' ') {
1461                                 ch--;
1462                                 padding++;
1463                                 continue;
1464                         }
1465                         /* check for trailing ansi sequences */
1466                         if (*ch != 'm')
1467                                 break;
1468                         p = ch - 1;
1469                         while (ch - p < 10 && *p != '\033')
1470                                 p--;
1471                         if (*p != '\033' ||
1472                             ch + 1 - p != display_mode_esc_sequence_len(p))
1473                                 break;
1474                         /*
1475                          * got a good ansi sequence, put it back to
1476                          * local_sb as we're cutting sb
1477                          */
1478                         strbuf_insert(&local_sb, 0, p, ch + 1 - p);
1479                         ch = p - 1;
1480                 }
1481                 strbuf_setlen(sb, ch + 1 - sb->buf);
1482                 c->flush_type = flush_left;
1483         }
1484
1485         if (len > padding) {
1486                 switch (c->truncate) {
1487                 case trunc_left:
1488                         strbuf_utf8_replace(&local_sb,
1489                                             0, len - (padding - 2),
1490                                             "..");
1491                         break;
1492                 case trunc_middle:
1493                         strbuf_utf8_replace(&local_sb,
1494                                             padding / 2 - 1,
1495                                             len - (padding - 2),
1496                                             "..");
1497                         break;
1498                 case trunc_right:
1499                         strbuf_utf8_replace(&local_sb,
1500                                             padding - 2, len - (padding - 2),
1501                                             "..");
1502                         break;
1503                 case trunc_none:
1504                         break;
1505                 }
1506                 strbuf_addbuf(sb, &local_sb);
1507         } else {
1508                 int sb_len = sb->len, offset = 0;
1509                 if (c->flush_type == flush_left)
1510                         offset = padding - len;
1511                 else if (c->flush_type == flush_both)
1512                         offset = (padding - len) / 2;
1513                 /*
1514                  * we calculate padding in columns, now
1515                  * convert it back to chars
1516                  */
1517                 padding = padding - len + local_sb.len;
1518                 strbuf_addchars(sb, ' ', padding);
1519                 memcpy(sb->buf + sb_len + offset, local_sb.buf,
1520                        local_sb.len);
1521         }
1522         strbuf_release(&local_sb);
1523         c->flush_type = no_flush;
1524         return total_consumed;
1525 }
1526
1527 static size_t format_commit_item(struct strbuf *sb, /* in UTF-8 */
1528                                  const char *placeholder,
1529                                  void *context)
1530 {
1531         int consumed;
1532         size_t orig_len;
1533         enum {
1534                 NO_MAGIC,
1535                 ADD_LF_BEFORE_NON_EMPTY,
1536                 DEL_LF_BEFORE_EMPTY,
1537                 ADD_SP_BEFORE_NON_EMPTY
1538         } magic = NO_MAGIC;
1539
1540         switch (placeholder[0]) {
1541         case '-':
1542                 magic = DEL_LF_BEFORE_EMPTY;
1543                 break;
1544         case '+':
1545                 magic = ADD_LF_BEFORE_NON_EMPTY;
1546                 break;
1547         case ' ':
1548                 magic = ADD_SP_BEFORE_NON_EMPTY;
1549                 break;
1550         default:
1551                 break;
1552         }
1553         if (magic != NO_MAGIC)
1554                 placeholder++;
1555
1556         orig_len = sb->len;
1557         if (((struct format_commit_context *)context)->flush_type != no_flush)
1558                 consumed = format_and_pad_commit(sb, placeholder, context);
1559         else
1560                 consumed = format_commit_one(sb, placeholder, context);
1561         if (magic == NO_MAGIC)
1562                 return consumed;
1563
1564         if ((orig_len == sb->len) && magic == DEL_LF_BEFORE_EMPTY) {
1565                 while (sb->len && sb->buf[sb->len - 1] == '\n')
1566                         strbuf_setlen(sb, sb->len - 1);
1567         } else if (orig_len != sb->len) {
1568                 if (magic == ADD_LF_BEFORE_NON_EMPTY)
1569                         strbuf_insert(sb, orig_len, "\n", 1);
1570                 else if (magic == ADD_SP_BEFORE_NON_EMPTY)
1571                         strbuf_insert(sb, orig_len, " ", 1);
1572         }
1573         return consumed + 1;
1574 }
1575
1576 static size_t userformat_want_item(struct strbuf *sb, const char *placeholder,
1577                                    void *context)
1578 {
1579         struct userformat_want *w = context;
1580
1581         if (*placeholder == '+' || *placeholder == '-' || *placeholder == ' ')
1582                 placeholder++;
1583
1584         switch (*placeholder) {
1585         case 'N':
1586                 w->notes = 1;
1587                 break;
1588         case 'S':
1589                 w->source = 1;
1590                 break;
1591         }
1592         return 0;
1593 }
1594
1595 void userformat_find_requirements(const char *fmt, struct userformat_want *w)
1596 {
1597         struct strbuf dummy = STRBUF_INIT;
1598
1599         if (!fmt) {
1600                 if (!user_format)
1601                         return;
1602                 fmt = user_format;
1603         }
1604         strbuf_expand(&dummy, fmt, userformat_want_item, w);
1605         strbuf_release(&dummy);
1606 }
1607
1608 void repo_format_commit_message(struct repository *r,
1609                                 const struct commit *commit,
1610                                 const char *format, struct strbuf *sb,
1611                                 const struct pretty_print_context *pretty_ctx)
1612 {
1613         struct format_commit_context context;
1614         const char *output_enc = pretty_ctx->output_encoding;
1615         const char *utf8 = "UTF-8";
1616
1617         memset(&context, 0, sizeof(context));
1618         context.commit = commit;
1619         context.pretty_ctx = pretty_ctx;
1620         context.wrap_start = sb->len;
1621         /*
1622          * convert a commit message to UTF-8 first
1623          * as far as 'format_commit_item' assumes it in UTF-8
1624          */
1625         context.message = repo_logmsg_reencode(r, commit,
1626                                                &context.commit_encoding,
1627                                                utf8);
1628
1629         strbuf_expand(sb, format, format_commit_item, &context);
1630         rewrap_message_tail(sb, &context, 0, 0, 0);
1631
1632         /* then convert a commit message to an actual output encoding */
1633         if (output_enc) {
1634                 if (same_encoding(utf8, output_enc))
1635                         output_enc = NULL;
1636         } else {
1637                 if (context.commit_encoding &&
1638                     !same_encoding(context.commit_encoding, utf8))
1639                         output_enc = context.commit_encoding;
1640         }
1641
1642         if (output_enc) {
1643                 size_t outsz;
1644                 char *out = reencode_string_len(sb->buf, sb->len,
1645                                                 output_enc, utf8, &outsz);
1646                 if (out)
1647                         strbuf_attach(sb, out, outsz, outsz + 1);
1648         }
1649
1650         free(context.commit_encoding);
1651         repo_unuse_commit_buffer(r, commit, context.message);
1652 }
1653
1654 static void pp_header(struct pretty_print_context *pp,
1655                       const char *encoding,
1656                       const struct commit *commit,
1657                       const char **msg_p,
1658                       struct strbuf *sb)
1659 {
1660         int parents_shown = 0;
1661
1662         for (;;) {
1663                 const char *name, *line = *msg_p;
1664                 int linelen = get_one_line(*msg_p);
1665
1666                 if (!linelen)
1667                         return;
1668                 *msg_p += linelen;
1669
1670                 if (linelen == 1)
1671                         /* End of header */
1672                         return;
1673
1674                 if (pp->fmt == CMIT_FMT_RAW) {
1675                         strbuf_add(sb, line, linelen);
1676                         continue;
1677                 }
1678
1679                 if (starts_with(line, "parent ")) {
1680                         if (linelen != the_hash_algo->hexsz + 8)
1681                                 die("bad parent line in commit");
1682                         continue;
1683                 }
1684
1685                 if (!parents_shown) {
1686                         unsigned num = commit_list_count(commit->parents);
1687                         /* with enough slop */
1688                         strbuf_grow(sb, num * (GIT_MAX_HEXSZ + 10) + 20);
1689                         add_merge_info(pp, sb, commit);
1690                         parents_shown = 1;
1691                 }
1692
1693                 /*
1694                  * MEDIUM == DEFAULT shows only author with dates.
1695                  * FULL shows both authors but not dates.
1696                  * FULLER shows both authors and dates.
1697                  */
1698                 if (skip_prefix(line, "author ", &name)) {
1699                         strbuf_grow(sb, linelen + 80);
1700                         pp_user_info(pp, "Author", sb, name, encoding);
1701                 }
1702                 if (skip_prefix(line, "committer ", &name) &&
1703                     (pp->fmt == CMIT_FMT_FULL || pp->fmt == CMIT_FMT_FULLER)) {
1704                         strbuf_grow(sb, linelen + 80);
1705                         pp_user_info(pp, "Commit", sb, name, encoding);
1706                 }
1707         }
1708 }
1709
1710 void pp_title_line(struct pretty_print_context *pp,
1711                    const char **msg_p,
1712                    struct strbuf *sb,
1713                    const char *encoding,
1714                    int need_8bit_cte)
1715 {
1716         static const int max_length = 78; /* per rfc2047 */
1717         struct strbuf title;
1718
1719         strbuf_init(&title, 80);
1720         *msg_p = format_subject(&title, *msg_p,
1721                                 pp->preserve_subject ? "\n" : " ");
1722
1723         strbuf_grow(sb, title.len + 1024);
1724         if (pp->print_email_subject) {
1725                 if (pp->rev)
1726                         fmt_output_email_subject(sb, pp->rev);
1727                 if (needs_rfc2047_encoding(title.buf, title.len))
1728                         add_rfc2047(sb, title.buf, title.len,
1729                                                 encoding, RFC2047_SUBJECT);
1730                 else
1731                         strbuf_add_wrapped_bytes(sb, title.buf, title.len,
1732                                          -last_line_length(sb), 1, max_length);
1733         } else {
1734                 strbuf_addbuf(sb, &title);
1735         }
1736         strbuf_addch(sb, '\n');
1737
1738         if (need_8bit_cte == 0) {
1739                 int i;
1740                 for (i = 0; i < pp->in_body_headers.nr; i++) {
1741                         if (has_non_ascii(pp->in_body_headers.items[i].string)) {
1742                                 need_8bit_cte = 1;
1743                                 break;
1744                         }
1745                 }
1746         }
1747
1748         if (need_8bit_cte > 0) {
1749                 const char *header_fmt =
1750                         "MIME-Version: 1.0\n"
1751                         "Content-Type: text/plain; charset=%s\n"
1752                         "Content-Transfer-Encoding: 8bit\n";
1753                 strbuf_addf(sb, header_fmt, encoding);
1754         }
1755         if (pp->after_subject) {
1756                 strbuf_addstr(sb, pp->after_subject);
1757         }
1758         if (cmit_fmt_is_mail(pp->fmt)) {
1759                 strbuf_addch(sb, '\n');
1760         }
1761
1762         if (pp->in_body_headers.nr) {
1763                 int i;
1764                 for (i = 0; i < pp->in_body_headers.nr; i++) {
1765                         strbuf_addstr(sb, pp->in_body_headers.items[i].string);
1766                         free(pp->in_body_headers.items[i].string);
1767                 }
1768                 string_list_clear(&pp->in_body_headers, 0);
1769                 strbuf_addch(sb, '\n');
1770         }
1771
1772         strbuf_release(&title);
1773 }
1774
1775 static int pp_utf8_width(const char *start, const char *end)
1776 {
1777         int width = 0;
1778         size_t remain = end - start;
1779
1780         while (remain) {
1781                 int n = utf8_width(&start, &remain);
1782                 if (n < 0 || !start)
1783                         return -1;
1784                 width += n;
1785         }
1786         return width;
1787 }
1788
1789 static void strbuf_add_tabexpand(struct strbuf *sb, int tabwidth,
1790                                  const char *line, int linelen)
1791 {
1792         const char *tab;
1793
1794         while ((tab = memchr(line, '\t', linelen)) != NULL) {
1795                 int width = pp_utf8_width(line, tab);
1796
1797                 /*
1798                  * If it wasn't well-formed utf8, or it
1799                  * had characters with badly defined
1800                  * width (control characters etc), just
1801                  * give up on trying to align things.
1802                  */
1803                 if (width < 0)
1804                         break;
1805
1806                 /* Output the data .. */
1807                 strbuf_add(sb, line, tab - line);
1808
1809                 /* .. and the de-tabified tab */
1810                 strbuf_addchars(sb, ' ', tabwidth - (width % tabwidth));
1811
1812                 /* Skip over the printed part .. */
1813                 linelen -= tab + 1 - line;
1814                 line = tab + 1;
1815         }
1816
1817         /*
1818          * Print out everything after the last tab without
1819          * worrying about width - there's nothing more to
1820          * align.
1821          */
1822         strbuf_add(sb, line, linelen);
1823 }
1824
1825 /*
1826  * pp_handle_indent() prints out the intendation, and
1827  * the whole line (without the final newline), after
1828  * de-tabifying.
1829  */
1830 static void pp_handle_indent(struct pretty_print_context *pp,
1831                              struct strbuf *sb, int indent,
1832                              const char *line, int linelen)
1833 {
1834         strbuf_addchars(sb, ' ', indent);
1835         if (pp->expand_tabs_in_log)
1836                 strbuf_add_tabexpand(sb, pp->expand_tabs_in_log, line, linelen);
1837         else
1838                 strbuf_add(sb, line, linelen);
1839 }
1840
1841 static int is_mboxrd_from(const char *line, int len)
1842 {
1843         /*
1844          * a line matching /^From $/ here would only have len == 4
1845          * at this point because is_empty_line would've trimmed all
1846          * trailing space
1847          */
1848         return len > 4 && starts_with(line + strspn(line, ">"), "From ");
1849 }
1850
1851 void pp_remainder(struct pretty_print_context *pp,
1852                   const char **msg_p,
1853                   struct strbuf *sb,
1854                   int indent)
1855 {
1856         int first = 1;
1857         for (;;) {
1858                 const char *line = *msg_p;
1859                 int linelen = get_one_line(line);
1860                 *msg_p += linelen;
1861
1862                 if (!linelen)
1863                         break;
1864
1865                 if (is_blank_line(line, &linelen)) {
1866                         if (first)
1867                                 continue;
1868                         if (pp->fmt == CMIT_FMT_SHORT)
1869                                 break;
1870                 }
1871                 first = 0;
1872
1873                 strbuf_grow(sb, linelen + indent + 20);
1874                 if (indent)
1875                         pp_handle_indent(pp, sb, indent, line, linelen);
1876                 else if (pp->expand_tabs_in_log)
1877                         strbuf_add_tabexpand(sb, pp->expand_tabs_in_log,
1878                                              line, linelen);
1879                 else {
1880                         if (pp->fmt == CMIT_FMT_MBOXRD &&
1881                                         is_mboxrd_from(line, linelen))
1882                                 strbuf_addch(sb, '>');
1883
1884                         strbuf_add(sb, line, linelen);
1885                 }
1886                 strbuf_addch(sb, '\n');
1887         }
1888 }
1889
1890 void pretty_print_commit(struct pretty_print_context *pp,
1891                          const struct commit *commit,
1892                          struct strbuf *sb)
1893 {
1894         unsigned long beginning_of_body;
1895         int indent = 4;
1896         const char *msg;
1897         const char *reencoded;
1898         const char *encoding;
1899         int need_8bit_cte = pp->need_8bit_cte;
1900
1901         if (pp->fmt == CMIT_FMT_USERFORMAT) {
1902                 format_commit_message(commit, user_format, sb, pp);
1903                 return;
1904         }
1905
1906         encoding = get_log_output_encoding();
1907         msg = reencoded = logmsg_reencode(commit, NULL, encoding);
1908
1909         if (pp->fmt == CMIT_FMT_ONELINE || cmit_fmt_is_mail(pp->fmt))
1910                 indent = 0;
1911
1912         /*
1913          * We need to check and emit Content-type: to mark it
1914          * as 8-bit if we haven't done so.
1915          */
1916         if (cmit_fmt_is_mail(pp->fmt) && need_8bit_cte == 0) {
1917                 int i, ch, in_body;
1918
1919                 for (in_body = i = 0; (ch = msg[i]); i++) {
1920                         if (!in_body) {
1921                                 /* author could be non 7-bit ASCII but
1922                                  * the log may be so; skip over the
1923                                  * header part first.
1924                                  */
1925                                 if (ch == '\n' && msg[i+1] == '\n')
1926                                         in_body = 1;
1927                         }
1928                         else if (non_ascii(ch)) {
1929                                 need_8bit_cte = 1;
1930                                 break;
1931                         }
1932                 }
1933         }
1934
1935         pp_header(pp, encoding, commit, &msg, sb);
1936         if (pp->fmt != CMIT_FMT_ONELINE && !pp->print_email_subject) {
1937                 strbuf_addch(sb, '\n');
1938         }
1939
1940         /* Skip excess blank lines at the beginning of body, if any... */
1941         msg = skip_blank_lines(msg);
1942
1943         /* These formats treat the title line specially. */
1944         if (pp->fmt == CMIT_FMT_ONELINE || cmit_fmt_is_mail(pp->fmt))
1945                 pp_title_line(pp, &msg, sb, encoding, need_8bit_cte);
1946
1947         beginning_of_body = sb->len;
1948         if (pp->fmt != CMIT_FMT_ONELINE)
1949                 pp_remainder(pp, &msg, sb, indent);
1950         strbuf_rtrim(sb);
1951
1952         /* Make sure there is an EOLN for the non-oneline case */
1953         if (pp->fmt != CMIT_FMT_ONELINE)
1954                 strbuf_addch(sb, '\n');
1955
1956         /*
1957          * The caller may append additional body text in e-mail
1958          * format.  Make sure we did not strip the blank line
1959          * between the header and the body.
1960          */
1961         if (cmit_fmt_is_mail(pp->fmt) && sb->len <= beginning_of_body)
1962                 strbuf_addch(sb, '\n');
1963
1964         unuse_commit_buffer(commit, reencoded);
1965 }
1966
1967 void pp_commit_easy(enum cmit_fmt fmt, const struct commit *commit,
1968                     struct strbuf *sb)
1969 {
1970         struct pretty_print_context pp = {0};
1971         pp.fmt = fmt;
1972         pretty_print_commit(&pp, commit, sb);
1973 }