git.oblomov.eu Git - git/blob - builtin/mailinfo.c

   1 /*
   2  * Another stupid program, this one parsing the headers of an
   3  * email to figure out authorship and subject
   4  */
   5 #include "cache.h"
   6 #include "builtin.h"
   7 #include "utf8.h"
   8 #include "strbuf.h"
   9
  10 static FILE *cmitmsg, *patchfile, *fin, *fout;
  11
  12 static int keep_subject;
  13 static int keep_non_patch_brackets_in_subject;
  14 static const char *metainfo_charset;
  15 static struct strbuf line = STRBUF_INIT;
  16 static struct strbuf name = STRBUF_INIT;
  17 static struct strbuf email = STRBUF_INIT;
  18
  19 static enum  {
  20         TE_DONTCARE, TE_QP, TE_BASE64
  21 } transfer_encoding;
  22 static enum  {
  23         TYPE_TEXT, TYPE_OTHER
  24 } message_type;
  25
  26 static struct strbuf charset = STRBUF_INIT;
  27 static int patch_lines;
  28 static struct strbuf **p_hdr_data, **s_hdr_data;
  29 static int use_scissors;
  30 static int use_inbody_headers = 1;
  31
  32 #define MAX_HDR_PARSED 10
  33 #define MAX_BOUNDARIES 5
  34
  35 static void cleanup_space(struct strbuf *sb);
  36
  37
  38 static void get_sane_name(struct strbuf *out, struct strbuf *name, struct strbuf *email)
  39 {
  40         struct strbuf *src = name;
  41         if (name->len < 3 || 60 < name->len || strchr(name->buf, '@') ||
  42                 strchr(name->buf, '<') || strchr(name->buf, '>'))
  43                 src = email;
  44         else if (name == out)
  45                 return;
  46         strbuf_reset(out);
  47         strbuf_addbuf(out, src);
  48 }
  49
  50 static void parse_bogus_from(const struct strbuf *line)
  51 {
  52         /* John Doe <johndoe> */
  53
  54         char *bra, *ket;
  55         /* This is fallback, so do not bother if we already have an
  56          * e-mail address.
  57          */
  58         if (email.len)
  59                 return;
  60
  61         bra = strchr(line->buf, '<');
  62         if (!bra)
  63                 return;
  64         ket = strchr(bra, '>');
  65         if (!ket)
  66                 return;
  67
  68         strbuf_reset(&email);
  69         strbuf_add(&email, bra + 1, ket - bra - 1);
  70
  71         strbuf_reset(&name);
  72         strbuf_add(&name, line->buf, bra - line->buf);
  73         strbuf_trim(&name);
  74         get_sane_name(&name, &name, &email);
  75 }
  76
  77 static void handle_from(const struct strbuf *from)
  78 {
  79         char *at;
  80         size_t el;
  81         struct strbuf f;
  82
  83         strbuf_init(&f, from->len);
  84         strbuf_addbuf(&f, from);
  85
  86         at = strchr(f.buf, '@');
  87         if (!at) {
  88                 parse_bogus_from(from);
  89                 return;
  90         }
  91
  92         /*
  93          * If we already have one email, don't take any confusing lines
  94          */
  95         if (email.len && strchr(at + 1, '@')) {
  96                 strbuf_release(&f);
  97                 return;
  98         }
  99
 100         /* Pick up the string around '@', possibly delimited with <>
 101          * pair; that is the email part.
 102          */
 103         while (at > f.buf) {
 104                 char c = at[-1];
 105                 if (isspace(c))
 106                         break;
 107                 if (c == '<') {
 108                         at[-1] = ' ';
 109                         break;
 110                 }
 111                 at--;
 112         }
 113         el = strcspn(at, " \n\t\r\v\f>");
 114         strbuf_reset(&email);
 115         strbuf_add(&email, at, el);
 116         strbuf_remove(&f, at - f.buf, el + (at[el] ? 1 : 0));
 117
 118         /* The remainder is name.  It could be
 119          *
 120          * - "John Doe <john.doe@xz>"                   (a), or
 121          * - "john.doe@xz (John Doe)"                   (b), or
 122          * - "John (zzz) Doe <john.doe@xz> (Comment)"   (c)
 123          *
 124          * but we have removed the email part, so
 125          *
 126          * - remove extra spaces which could stay after email (case 'c'), and
 127          * - trim from both ends, possibly removing the () pair at the end
 128          *   (cases 'a' and 'b').
 129          */
 130         cleanup_space(&f);
 131         strbuf_trim(&f);
 132         if (f.buf[0] == '(' && f.len && f.buf[f.len - 1] == ')') {
 133                 strbuf_remove(&f, 0, 1);
 134                 strbuf_setlen(&f, f.len - 1);
 135         }
 136
 137         get_sane_name(&name, &f, &email);
 138         strbuf_release(&f);
 139 }
 140
 141 static void handle_header(struct strbuf **out, const struct strbuf *line)
 142 {
 143         if (!*out) {
 144                 *out = xmalloc(sizeof(struct strbuf));
 145                 strbuf_init(*out, line->len);
 146         } else
 147                 strbuf_reset(*out);
 148
 149         strbuf_addbuf(*out, line);
 150 }
 151
 152 /* NOTE NOTE NOTE.  We do not claim we do full MIME.  We just attempt
 153  * to have enough heuristics to grok MIME encoded patches often found
 154  * on our mailing lists.  For example, we do not even treat header lines
 155  * case insensitively.
 156  */
 157
 158 static int slurp_attr(const char *line, const char *name, struct strbuf *attr)
 159 {
 160         const char *ends, *ap = strcasestr(line, name);
 161         size_t sz;
 162
 163         if (!ap) {
 164                 strbuf_setlen(attr, 0);
 165                 return 0;
 166         }
 167         ap += strlen(name);
 168         if (*ap == '"') {
 169                 ap++;
 170                 ends = "\"";
 171         }
 172         else
 173                 ends = "; \t";
 174         sz = strcspn(ap, ends);
 175         strbuf_add(attr, ap, sz);
 176         return 1;
 177 }
 178
 179 static struct strbuf *content[MAX_BOUNDARIES];
 180
 181 static struct strbuf **content_top = content;
 182
 183 static void handle_content_type(struct strbuf *line)
 184 {
 185         struct strbuf *boundary = xmalloc(sizeof(struct strbuf));
 186         strbuf_init(boundary, line->len);
 187
 188         if (!strcasestr(line->buf, "text/"))
 189                  message_type = TYPE_OTHER;
 190         if (slurp_attr(line->buf, "boundary=", boundary)) {
 191                 strbuf_insert(boundary, 0, "--", 2);
 192                 if (++content_top > &content[MAX_BOUNDARIES]) {
 193                         fprintf(stderr, "Too many boundaries to handle\n");
 194                         exit(1);
 195                 }
 196                 *content_top = boundary;
 197                 boundary = NULL;
 198         }
 199         slurp_attr(line->buf, "charset=", &charset);
 200
 201         if (boundary) {
 202                 strbuf_release(boundary);
 203                 free(boundary);
 204         }
 205 }
 206
 207 static void handle_content_transfer_encoding(const struct strbuf *line)
 208 {
 209         if (strcasestr(line->buf, "base64"))
 210                 transfer_encoding = TE_BASE64;
 211         else if (strcasestr(line->buf, "quoted-printable"))
 212                 transfer_encoding = TE_QP;
 213         else
 214                 transfer_encoding = TE_DONTCARE;
 215 }
 216
 217 static int is_multipart_boundary(const struct strbuf *line)
 218 {
 219         return (((*content_top)->len <= line->len) &&
 220                 !memcmp(line->buf, (*content_top)->buf, (*content_top)->len));
 221 }
 222
 223 static void cleanup_subject(struct strbuf *subject)
 224 {
 225         size_t at = 0;
 226
 227         while (at < subject->len) {
 228                 char *pos;
 229                 size_t remove;
 230
 231                 switch (subject->buf[at]) {
 232                 case 'r': case 'R':
 233                         if (subject->len <= at + 3)
 234                                 break;
 235                         if (!memcmp(subject->buf + at + 1, "e:", 2)) {
 236                                 strbuf_remove(subject, at, 3);
 237                                 continue;
 238                         }
 239                         at++;
 240                         break;
 241                 case ' ': case '\t': case ':':
 242                         strbuf_remove(subject, at, 1);
 243                         continue;
 244                 case '[':
 245                         pos = strchr(subject->buf + at, ']');
 246                         if (!pos)
 247                                 break;
 248                         remove = pos - subject->buf + at + 1;
 249                         if (!keep_non_patch_brackets_in_subject ||
 250                             (7 <= remove &&
 251                              memmem(subject->buf + at, remove, "PATCH", 5)))
 252                                 strbuf_remove(subject, at, remove);
 253                         else
 254                                 at += remove;
 255                         continue;
 256                 }
 257                 break;
 258         }
 259         strbuf_trim(subject);
 260 }
 261
 262 static void cleanup_space(struct strbuf *sb)
 263 {
 264         size_t pos, cnt;
 265         for (pos = 0; pos < sb->len; pos++) {
 266                 if (isspace(sb->buf[pos])) {
 267                         sb->buf[pos] = ' ';
 268                         for (cnt = 0; isspace(sb->buf[pos + cnt + 1]); cnt++);
 269                         strbuf_remove(sb, pos + 1, cnt);
 270                 }
 271         }
 272 }
 273
 274 static void decode_header(struct strbuf *line);
 275 static const char *header[MAX_HDR_PARSED] = {
 276         "From","Subject","Date",
 277 };
 278
 279 static inline int cmp_header(const struct strbuf *line, const char *hdr)
 280 {
 281         int len = strlen(hdr);
 282         return !strncasecmp(line->buf, hdr, len) && line->len > len &&
 283                         line->buf[len] == ':' && isspace(line->buf[len + 1]);
 284 }
 285
 286 static int check_header(const struct strbuf *line,
 287                                 struct strbuf *hdr_data[], int overwrite)
 288 {
 289         int i, ret = 0, len;
 290         struct strbuf sb = STRBUF_INIT;
 291         /* search for the interesting parts */
 292         for (i = 0; header[i]; i++) {
 293                 int len = strlen(header[i]);
 294                 if ((!hdr_data[i] || overwrite) && cmp_header(line, header[i])) {
 295                         /* Unwrap inline B and Q encoding, and optionally
 296                          * normalize the meta information to utf8.
 297                          */
 298                         strbuf_add(&sb, line->buf + len + 2, line->len - len - 2);
 299                         decode_header(&sb);
 300                         handle_header(&hdr_data[i], &sb);
 301                         ret = 1;
 302                         goto check_header_out;
 303                 }
 304         }
 305
 306         /* Content stuff */
 307         if (cmp_header(line, "Content-Type")) {
 308                 len = strlen("Content-Type: ");
 309                 strbuf_add(&sb, line->buf + len, line->len - len);
 310                 decode_header(&sb);
 311                 strbuf_insert(&sb, 0, "Content-Type: ", len);
 312                 handle_content_type(&sb);
 313                 ret = 1;
 314                 goto check_header_out;
 315         }
 316         if (cmp_header(line, "Content-Transfer-Encoding")) {
 317                 len = strlen("Content-Transfer-Encoding: ");
 318                 strbuf_add(&sb, line->buf + len, line->len - len);
 319                 decode_header(&sb);
 320                 handle_content_transfer_encoding(&sb);
 321                 ret = 1;
 322                 goto check_header_out;
 323         }
 324
 325         /* for inbody stuff */
 326         if (!prefixcmp(line->buf, ">From") && isspace(line->buf[5])) {
 327                 ret = 1; /* Should this return 0? */
 328                 goto check_header_out;
 329         }
 330         if (!prefixcmp(line->buf, "[PATCH]") && isspace(line->buf[7])) {
 331                 for (i = 0; header[i]; i++) {
 332                         if (!memcmp("Subject", header[i], 7)) {
 333                                 handle_header(&hdr_data[i], line);
 334                                 ret = 1;
 335                                 goto check_header_out;
 336                         }
 337                 }
 338         }
 339
 340 check_header_out:
 341         strbuf_release(&sb);
 342         return ret;
 343 }
 344
 345 static int is_rfc2822_header(const struct strbuf *line)
 346 {
 347         /*
 348          * The section that defines the loosest possible
 349          * field name is "3.6.8 Optional fields".
 350          *
 351          * optional-field = field-name ":" unstructured CRLF
 352          * field-name = 1*ftext
 353          * ftext = %d33-57 / %59-126
 354          */
 355         int ch;
 356         char *cp = line->buf;
 357
 358         /* Count mbox From headers as headers */
 359         if (!prefixcmp(cp, "From ") || !prefixcmp(cp, ">From "))
 360                 return 1;
 361
 362         while ((ch = *cp++)) {
 363                 if (ch == ':')
 364                         return 1;
 365                 if ((33 <= ch && ch <= 57) ||
 366                     (59 <= ch && ch <= 126))
 367                         continue;
 368                 break;
 369         }
 370         return 0;
 371 }
 372
 373 static int read_one_header_line(struct strbuf *line, FILE *in)
 374 {
 375         /* Get the first part of the line. */
 376         if (strbuf_getline(line, in, '\n'))
 377                 return 0;
 378
 379         /*
 380          * Is it an empty line or not a valid rfc2822 header?
 381          * If so, stop here, and return false ("not a header")
 382          */
 383         strbuf_rtrim(line);
 384         if (!line->len || !is_rfc2822_header(line)) {
 385                 /* Re-add the newline */
 386                 strbuf_addch(line, '\n');
 387                 return 0;
 388         }
 389
 390         /*
 391          * Now we need to eat all the continuation lines..
 392          * Yuck, 2822 header "folding"
 393          */
 394         for (;;) {
 395                 int peek;
 396                 struct strbuf continuation = STRBUF_INIT;
 397
 398                 peek = fgetc(in); ungetc(peek, in);
 399                 if (peek != ' ' && peek != '\t')
 400                         break;
 401                 if (strbuf_getline(&continuation, in, '\n'))
 402                         break;
 403                 continuation.buf[0] = '\n';
 404                 strbuf_rtrim(&continuation);
 405                 strbuf_addbuf(line, &continuation);
 406         }
 407
 408         return 1;
 409 }
 410
 411 static struct strbuf *decode_q_segment(const struct strbuf *q_seg, int rfc2047)
 412 {
 413         const char *in = q_seg->buf;
 414         int c;
 415         struct strbuf *out = xmalloc(sizeof(struct strbuf));
 416         strbuf_init(out, q_seg->len);
 417
 418         while ((c = *in++) != 0) {
 419                 if (c == '=') {
 420                         int d = *in++;
 421                         if (d == '\n' || !d)
 422                                 break; /* drop trailing newline */
 423                         strbuf_addch(out, (hexval(d) << 4) | hexval(*in++));
 424                         continue;
 425                 }
 426                 if (rfc2047 && c == '_') /* rfc2047 4.2 (2) */
 427                         c = 0x20;
 428                 strbuf_addch(out, c);
 429         }
 430         return out;
 431 }
 432
 433 static struct strbuf *decode_b_segment(const struct strbuf *b_seg)
 434 {
 435         /* Decode in..ep, possibly in-place to ot */
 436         int c, pos = 0, acc = 0;
 437         const char *in = b_seg->buf;
 438         struct strbuf *out = xmalloc(sizeof(struct strbuf));
 439         strbuf_init(out, b_seg->len);
 440
 441         while ((c = *in++) != 0) {
 442                 if (c == '+')
 443                         c = 62;
 444                 else if (c == '/')
 445                         c = 63;
 446                 else if ('A' <= c && c <= 'Z')
 447                         c -= 'A';
 448                 else if ('a' <= c && c <= 'z')
 449                         c -= 'a' - 26;
 450                 else if ('0' <= c && c <= '9')
 451                         c -= '0' - 52;
 452                 else
 453                         continue; /* garbage */
 454                 switch (pos++) {
 455                 case 0:
 456                         acc = (c << 2);
 457                         break;
 458                 case 1:
 459                         strbuf_addch(out, (acc | (c >> 4)));
 460                         acc = (c & 15) << 4;
 461                         break;
 462                 case 2:
 463                         strbuf_addch(out, (acc | (c >> 2)));
 464                         acc = (c & 3) << 6;
 465                         break;
 466                 case 3:
 467                         strbuf_addch(out, (acc | c));
 468                         acc = pos = 0;
 469                         break;
 470                 }
 471         }
 472         return out;
 473 }
 474
 475 /*
 476  * When there is no known charset, guess.
 477  *
 478  * Right now we assume that if the target is UTF-8 (the default),
 479  * and it already looks like UTF-8 (which includes US-ASCII as its
 480  * subset, of course) then that is what it is and there is nothing
 481  * to do.
 482  *
 483  * Otherwise, we default to assuming it is Latin1 for historical
 484  * reasons.
 485  */
 486 static const char *guess_charset(const struct strbuf *line, const char *target_charset)
 487 {
 488         if (is_encoding_utf8(target_charset)) {
 489                 if (is_utf8(line->buf))
 490                         return NULL;
 491         }
 492         return "ISO8859-1";
 493 }
 494
 495 static void convert_to_utf8(struct strbuf *line, const char *charset)
 496 {
 497         char *out;
 498
 499         if (!charset || !*charset) {
 500                 charset = guess_charset(line, metainfo_charset);
 501                 if (!charset)
 502                         return;
 503         }
 504
 505         if (!strcasecmp(metainfo_charset, charset))
 506                 return;
 507         out = reencode_string(line->buf, metainfo_charset, charset);
 508         if (!out)
 509                 die("cannot convert from %s to %s",
 510                     charset, metainfo_charset);
 511         strbuf_attach(line, out, strlen(out), strlen(out));
 512 }
 513
 514 static int decode_header_bq(struct strbuf *it)
 515 {
 516         char *in, *ep, *cp;
 517         struct strbuf outbuf = STRBUF_INIT, *dec;
 518         struct strbuf charset_q = STRBUF_INIT, piecebuf = STRBUF_INIT;
 519         int rfc2047 = 0;
 520
 521         in = it->buf;
 522         while (in - it->buf <= it->len && (ep = strstr(in, "=?")) != NULL) {
 523                 int encoding;
 524                 strbuf_reset(&charset_q);
 525                 strbuf_reset(&piecebuf);
 526                 rfc2047 = 1;
 527
 528                 if (in != ep) {
 529                         /*
 530                          * We are about to process an encoded-word
 531                          * that begins at ep, but there is something
 532                          * before the encoded word.
 533                          */
 534                         char *scan;
 535                         for (scan = in; scan < ep; scan++)
 536                                 if (!isspace(*scan))
 537                                         break;
 538
 539                         if (scan != ep || in == it->buf) {
 540                                 /*
 541                                  * We should not lose that "something",
 542                                  * unless we have just processed an
 543                                  * encoded-word, and there is only LWS
 544                                  * before the one we are about to process.
 545                                  */
 546                                 strbuf_add(&outbuf, in, ep - in);
 547                         }
 548                 }
 549                 /* E.g.
 550                  * ep : "=?iso-2022-jp?B?GyR...?= foo"
 551                  * ep : "=?ISO-8859-1?Q?Foo=FCbar?= baz"
 552                  */
 553                 ep += 2;
 554
 555                 if (ep - it->buf >= it->len || !(cp = strchr(ep, '?')))
 556                         goto decode_header_bq_out;
 557
 558                 if (cp + 3 - it->buf > it->len)
 559                         goto decode_header_bq_out;
 560                 strbuf_add(&charset_q, ep, cp - ep);
 561
 562                 encoding = cp[1];
 563                 if (!encoding || cp[2] != '?')
 564                         goto decode_header_bq_out;
 565                 ep = strstr(cp + 3, "?=");
 566                 if (!ep)
 567                         goto decode_header_bq_out;
 568                 strbuf_add(&piecebuf, cp + 3, ep - cp - 3);
 569                 switch (tolower(encoding)) {
 570                 default:
 571                         goto decode_header_bq_out;
 572                 case 'b':
 573                         dec = decode_b_segment(&piecebuf);
 574                         break;
 575                 case 'q':
 576                         dec = decode_q_segment(&piecebuf, 1);
 577                         break;
 578                 }
 579                 if (metainfo_charset)
 580                         convert_to_utf8(dec, charset_q.buf);
 581
 582                 strbuf_addbuf(&outbuf, dec);
 583                 strbuf_release(dec);
 584                 free(dec);
 585                 in = ep + 2;
 586         }
 587         strbuf_addstr(&outbuf, in);
 588         strbuf_reset(it);
 589         strbuf_addbuf(it, &outbuf);
 590 decode_header_bq_out:
 591         strbuf_release(&outbuf);
 592         strbuf_release(&charset_q);
 593         strbuf_release(&piecebuf);
 594         return rfc2047;
 595 }
 596
 597 static void decode_header(struct strbuf *it)
 598 {
 599         if (decode_header_bq(it))
 600                 return;
 601         /* otherwise "it" is a straight copy of the input.
 602          * This can be binary guck but there is no charset specified.
 603          */
 604         if (metainfo_charset)
 605                 convert_to_utf8(it, "");
 606 }
 607
 608 static void decode_transfer_encoding(struct strbuf *line)
 609 {
 610         struct strbuf *ret;
 611
 612         switch (transfer_encoding) {
 613         case TE_QP:
 614                 ret = decode_q_segment(line, 0);
 615                 break;
 616         case TE_BASE64:
 617                 ret = decode_b_segment(line);
 618                 break;
 619         case TE_DONTCARE:
 620         default:
 621                 return;
 622         }
 623         strbuf_reset(line);
 624         strbuf_addbuf(line, ret);
 625         strbuf_release(ret);
 626         free(ret);
 627 }
 628
 629 static void handle_filter(struct strbuf *line);
 630
 631 static int find_boundary(void)
 632 {
 633         while (!strbuf_getline(&line, fin, '\n')) {
 634                 if (*content_top && is_multipart_boundary(&line))
 635                         return 1;
 636         }
 637         return 0;
 638 }
 639
 640 static int handle_boundary(void)
 641 {
 642         struct strbuf newline = STRBUF_INIT;
 643
 644         strbuf_addch(&newline, '\n');
 645 again:
 646         if (line.len >= (*content_top)->len + 2 &&
 647             !memcmp(line.buf + (*content_top)->len, "--", 2)) {
 648                 /* we hit an end boundary */
 649                 /* pop the current boundary off the stack */
 650                 strbuf_release(*content_top);
 651                 free(*content_top);
 652                 *content_top = NULL;
 653
 654                 /* technically won't happen as is_multipart_boundary()
 655                    will fail first.  But just in case..
 656                  */
 657                 if (--content_top < content) {
 658                         fprintf(stderr, "Detected mismatched boundaries, "
 659                                         "can't recover\n");
 660                         exit(1);
 661                 }
 662                 handle_filter(&newline);
 663                 strbuf_release(&newline);
 664
 665                 /* skip to the next boundary */
 666                 if (!find_boundary())
 667                         return 0;
 668                 goto again;
 669         }
 670
 671         /* set some defaults */
 672         transfer_encoding = TE_DONTCARE;
 673         strbuf_reset(&charset);
 674         message_type = TYPE_TEXT;
 675
 676         /* slurp in this section's info */
 677         while (read_one_header_line(&line, fin))
 678                 check_header(&line, p_hdr_data, 0);
 679
 680         strbuf_release(&newline);
 681         /* replenish line */
 682         if (strbuf_getline(&line, fin, '\n'))
 683                 return 0;
 684         strbuf_addch(&line, '\n');
 685         return 1;
 686 }
 687
 688 static inline int patchbreak(const struct strbuf *line)
 689 {
 690         size_t i;
 691
 692         /* Beginning of a "diff -" header? */
 693         if (!prefixcmp(line->buf, "diff -"))
 694                 return 1;
 695
 696         /* CVS "Index: " line? */
 697         if (!prefixcmp(line->buf, "Index: "))
 698                 return 1;
 699
 700         /*
 701          * "--- <filename>" starts patches without headers
 702          * "---<sp>*" is a manual separator
 703          */
 704         if (line->len < 4)
 705                 return 0;
 706
 707         if (!prefixcmp(line->buf, "---")) {
 708                 /* space followed by a filename? */
 709                 if (line->buf[3] == ' ' && !isspace(line->buf[4]))
 710                         return 1;
 711                 /* Just whitespace? */
 712                 for (i = 3; i < line->len; i++) {
 713                         unsigned char c = line->buf[i];
 714                         if (c == '\n')
 715                                 return 1;
 716                         if (!isspace(c))
 717                                 break;
 718                 }
 719                 return 0;
 720         }
 721         return 0;
 722 }
 723
 724 static int is_scissors_line(const struct strbuf *line)
 725 {
 726         size_t i, len = line->len;
 727         int scissors = 0, gap = 0;
 728         int first_nonblank = -1;
 729         int last_nonblank = 0, visible, perforation = 0, in_perforation = 0;
 730         const char *buf = line->buf;
 731
 732         for (i = 0; i < len; i++) {
 733                 if (isspace(buf[i])) {
 734                         if (in_perforation) {
 735                                 perforation++;
 736                                 gap++;
 737                         }
 738                         continue;
 739                 }
 740                 last_nonblank = i;
 741                 if (first_nonblank < 0)
 742                         first_nonblank = i;
 743                 if (buf[i] == '-') {
 744                         in_perforation = 1;
 745                         perforation++;
 746                         continue;
 747                 }
 748                 if (i + 1 < len &&
 749                     (!memcmp(buf + i, ">8", 2) || !memcmp(buf + i, "8<", 2) ||
 750                      !memcmp(buf + i, ">%", 2) || !memcmp(buf + i, "%<", 2))) {
 751                         in_perforation = 1;
 752                         perforation += 2;
 753                         scissors += 2;
 754                         i++;
 755                         continue;
 756                 }
 757                 in_perforation = 0;
 758         }
 759
 760         /*
 761          * The mark must be at least 8 bytes long (e.g. "-- >8 --").
 762          * Even though there can be arbitrary cruft on the same line
 763          * (e.g. "cut here"), in order to avoid misidentification, the
 764          * perforation must occupy more than a third of the visible
 765          * width of the line, and dashes and scissors must occupy more
 766          * than half of the perforation.
 767          */
 768
 769         visible = last_nonblank - first_nonblank + 1;
 770         return (scissors && 8 <= visible &&
 771                 visible < perforation * 3 &&
 772                 gap * 2 < perforation);
 773 }
 774
 775 static int handle_commit_msg(struct strbuf *line)
 776 {
 777         static int still_looking = 1;
 778
 779         if (!cmitmsg)
 780                 return 0;
 781
 782         if (still_looking) {
 783                 if (!line->len || (line->len == 1 && line->buf[0] == '\n'))
 784                         return 0;
 785         }
 786
 787         if (use_inbody_headers && still_looking) {
 788                 still_looking = check_header(line, s_hdr_data, 0);
 789                 if (still_looking)
 790                         return 0;
 791         } else
 792                 /* Only trim the first (blank) line of the commit message
 793                  * when ignoring in-body headers.
 794                  */
 795                 still_looking = 0;
 796
 797         /* normalize the log message to UTF-8. */
 798         if (metainfo_charset)
 799                 convert_to_utf8(line, charset.buf);
 800
 801         if (use_scissors && is_scissors_line(line)) {
 802                 int i;
 803                 if (fseek(cmitmsg, 0L, SEEK_SET))
 804                         die_errno("Could not rewind output message file");
 805                 if (ftruncate(fileno(cmitmsg), 0))
 806                         die_errno("Could not truncate output message file at scissors");
 807                 still_looking = 1;
 808
 809                 /*
 810                  * We may have already read "secondary headers"; purge
 811                  * them to give ourselves a clean restart.
 812                  */
 813                 for (i = 0; header[i]; i++) {
 814                         if (s_hdr_data[i])
 815                                 strbuf_release(s_hdr_data[i]);
 816                         s_hdr_data[i] = NULL;
 817                 }
 818                 return 0;
 819         }
 820
 821         if (patchbreak(line)) {
 822                 fclose(cmitmsg);
 823                 cmitmsg = NULL;
 824                 return 1;
 825         }
 826
 827         fputs(line->buf, cmitmsg);
 828         return 0;
 829 }
 830
 831 static void handle_patch(const struct strbuf *line)
 832 {
 833         fwrite(line->buf, 1, line->len, patchfile);
 834         patch_lines++;
 835 }
 836
 837 static void handle_filter(struct strbuf *line)
 838 {
 839         static int filter = 0;
 840
 841         /* filter tells us which part we left off on */
 842         switch (filter) {
 843         case 0:
 844                 if (!handle_commit_msg(line))
 845                         break;
 846                 filter++;
 847         case 1:
 848                 handle_patch(line);
 849                 break;
 850         }
 851 }
 852
 853 static void handle_body(void)
 854 {
 855         struct strbuf prev = STRBUF_INIT;
 856
 857         /* Skip up to the first boundary */
 858         if (*content_top) {
 859                 if (!find_boundary())
 860                         goto handle_body_out;
 861         }
 862
 863         do {
 864                 /* process any boundary lines */
 865                 if (*content_top && is_multipart_boundary(&line)) {
 866                         /* flush any leftover */
 867                         if (prev.len) {
 868                                 handle_filter(&prev);
 869                                 strbuf_reset(&prev);
 870                         }
 871                         if (!handle_boundary())
 872                                 goto handle_body_out;
 873                 }
 874
 875                 /* Unwrap transfer encoding */
 876                 decode_transfer_encoding(&line);
 877
 878                 switch (transfer_encoding) {
 879                 case TE_BASE64:
 880                 case TE_QP:
 881                 {
 882                         struct strbuf **lines, **it, *sb;
 883
 884                         /* Prepend any previous partial lines */
 885                         strbuf_insert(&line, 0, prev.buf, prev.len);
 886                         strbuf_reset(&prev);
 887
 888                         /* binary data most likely doesn't have newlines */
 889                         if (message_type != TYPE_TEXT) {
 890                                 handle_filter(&line);
 891                                 break;
 892                         }
 893                         /*
 894                          * This is a decoded line that may contain
 895                          * multiple new lines.  Pass only one chunk
 896                          * at a time to handle_filter()
 897                          */
 898                         lines = strbuf_split(&line, '\n');
 899                         for (it = lines; (sb = *it); it++) {
 900                                 if (*(it + 1) == NULL) /* The last line */
 901                                         if (sb->buf[sb->len - 1] != '\n') {
 902                                                 /* Partial line, save it for later. */
 903                                                 strbuf_addbuf(&prev, sb);
 904                                                 break;
 905                                         }
 906                                 handle_filter(sb);
 907                         }
 908                         /*
 909                          * The partial chunk is saved in "prev" and will be
 910                          * appended by the next iteration of read_line_with_nul().
 911                          */
 912                         strbuf_list_free(lines);
 913                         break;
 914                 }
 915                 default:
 916                         handle_filter(&line);
 917                 }
 918
 919         } while (!strbuf_getwholeline(&line, fin, '\n'));
 920
 921 handle_body_out:
 922         strbuf_release(&prev);
 923 }
 924
 925 static void output_header_lines(FILE *fout, const char *hdr, const struct strbuf *data)
 926 {
 927         const char *sp = data->buf;
 928         while (1) {
 929                 char *ep = strchr(sp, '\n');
 930                 int len;
 931                 if (!ep)
 932                         len = strlen(sp);
 933                 else
 934                         len = ep - sp;
 935                 fprintf(fout, "%s: %.*s\n", hdr, len, sp);
 936                 if (!ep)
 937                         break;
 938                 sp = ep + 1;
 939         }
 940 }
 941
 942 static void handle_info(void)
 943 {
 944         struct strbuf *hdr;
 945         int i;
 946
 947         for (i = 0; header[i]; i++) {
 948                 /* only print inbody headers if we output a patch file */
 949                 if (patch_lines && s_hdr_data[i])
 950                         hdr = s_hdr_data[i];
 951                 else if (p_hdr_data[i])
 952                         hdr = p_hdr_data[i];
 953                 else
 954                         continue;
 955
 956                 if (!memcmp(header[i], "Subject", 7)) {
 957                         if (!keep_subject) {
 958                                 cleanup_subject(hdr);
 959                                 cleanup_space(hdr);
 960                         }
 961                         output_header_lines(fout, "Subject", hdr);
 962                 } else if (!memcmp(header[i], "From", 4)) {
 963                         cleanup_space(hdr);
 964                         handle_from(hdr);
 965                         fprintf(fout, "Author: %s\n", name.buf);
 966                         fprintf(fout, "Email: %s\n", email.buf);
 967                 } else {
 968                         cleanup_space(hdr);
 969                         fprintf(fout, "%s: %s\n", header[i], hdr->buf);
 970                 }
 971         }
 972         fprintf(fout, "\n");
 973 }
 974
 975 static int mailinfo(FILE *in, FILE *out, const char *msg, const char *patch)
 976 {
 977         int peek;
 978         fin = in;
 979         fout = out;
 980
 981         cmitmsg = fopen(msg, "w");
 982         if (!cmitmsg) {
 983                 perror(msg);
 984                 return -1;
 985         }
 986         patchfile = fopen(patch, "w");
 987         if (!patchfile) {
 988                 perror(patch);
 989                 fclose(cmitmsg);
 990                 return -1;
 991         }
 992
 993         p_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*p_hdr_data));
 994         s_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*s_hdr_data));
 995
 996         do {
 997                 peek = fgetc(in);
 998         } while (isspace(peek));
 999         ungetc(peek, in);
1000
1001         /* process the email header */
1002         while (read_one_header_line(&line, fin))
1003                 check_header(&line, p_hdr_data, 1);
1004
1005         handle_body();
1006         handle_info();
1007
1008         return 0;
1009 }
1010
1011 static int git_mailinfo_config(const char *var, const char *value, void *unused)
1012 {
1013         if (prefixcmp(var, "mailinfo."))
1014                 return git_default_config(var, value, unused);
1015         if (!strcmp(var, "mailinfo.scissors")) {
1016                 use_scissors = git_config_bool(var, value);
1017                 return 0;
1018         }
1019         /* perhaps others here */
1020         return 0;
1021 }
1022
1023 static const char mailinfo_usage[] =
1024         "git mailinfo [-k|-b] [-u | --encoding=<encoding> | -n] [--scissors | --no-scissors] msg patch < mail >info";
1025
1026 int cmd_mailinfo(int argc, const char **argv, const char *prefix)
1027 {
1028         const char *def_charset;
1029
1030         /* NEEDSWORK: might want to do the optional .git/ directory
1031          * discovery
1032          */
1033         git_config(git_mailinfo_config, NULL);
1034
1035         def_charset = get_commit_output_encoding();
1036         metainfo_charset = def_charset;
1037
1038         while (1 < argc && argv[1][0] == '-') {
1039                 if (!strcmp(argv[1], "-k"))
1040                         keep_subject = 1;
1041                 else if (!strcmp(argv[1], "-b"))
1042                         keep_non_patch_brackets_in_subject = 1;
1043                 else if (!strcmp(argv[1], "-u"))
1044                         metainfo_charset = def_charset;
1045                 else if (!strcmp(argv[1], "-n"))
1046                         metainfo_charset = NULL;
1047                 else if (!prefixcmp(argv[1], "--encoding="))
1048                         metainfo_charset = argv[1] + 11;
1049                 else if (!strcmp(argv[1], "--scissors"))
1050                         use_scissors = 1;
1051                 else if (!strcmp(argv[1], "--no-scissors"))
1052                         use_scissors = 0;
1053                 else if (!strcmp(argv[1], "--no-inbody-headers"))
1054                         use_inbody_headers = 0;
1055                 else
1056                         usage(mailinfo_usage);
1057                 argc--; argv++;
1058         }
1059
1060         if (argc != 3)
1061                 usage(mailinfo_usage);
1062
1063         return !!mailinfo(stdin, stdout, argv[1], argv[2]);
1064 }