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