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
  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 is_format_patch_separator(const char *line, int len)
 292 {
 293         static const char SAMPLE[] =
 294                 "From e6807f3efca28b30decfecb1732a56c7db1137ee Mon Sep 17 00:00:00 2001\n";
 295         const char *cp;
 296
 297         if (len != strlen(SAMPLE))
 298                 return 0;
 299         if (!skip_prefix(line, "From ", &cp))
 300                 return 0;
 301         if (strspn(cp, "0123456789abcdef") != 40)
 302                 return 0;
 303         cp += 40;
 304         return !memcmp(SAMPLE + (cp - line), cp, strlen(SAMPLE) - (cp - line));
 305 }
 306
 307 static int check_header(const struct strbuf *line,
 308                                 struct strbuf *hdr_data[], int overwrite)
 309 {
 310         int i, ret = 0, len;
 311         struct strbuf sb = STRBUF_INIT;
 312         /* search for the interesting parts */
 313         for (i = 0; header[i]; i++) {
 314                 int len = strlen(header[i]);
 315                 if ((!hdr_data[i] || overwrite) && cmp_header(line, header[i])) {
 316                         /* Unwrap inline B and Q encoding, and optionally
 317                          * normalize the meta information to utf8.
 318                          */
 319                         strbuf_add(&sb, line->buf + len + 2, line->len - len - 2);
 320                         decode_header(&sb);
 321                         handle_header(&hdr_data[i], &sb);
 322                         ret = 1;
 323                         goto check_header_out;
 324                 }
 325         }
 326
 327         /* Content stuff */
 328         if (cmp_header(line, "Content-Type")) {
 329                 len = strlen("Content-Type: ");
 330                 strbuf_add(&sb, line->buf + len, line->len - len);
 331                 decode_header(&sb);
 332                 strbuf_insert(&sb, 0, "Content-Type: ", len);
 333                 handle_content_type(&sb);
 334                 ret = 1;
 335                 goto check_header_out;
 336         }
 337         if (cmp_header(line, "Content-Transfer-Encoding")) {
 338                 len = strlen("Content-Transfer-Encoding: ");
 339                 strbuf_add(&sb, line->buf + len, line->len - len);
 340                 decode_header(&sb);
 341                 handle_content_transfer_encoding(&sb);
 342                 ret = 1;
 343                 goto check_header_out;
 344         }
 345
 346         /* for inbody stuff */
 347         if (starts_with(line->buf, ">From") && isspace(line->buf[5])) {
 348                 ret = is_format_patch_separator(line->buf + 1, line->len - 1);
 349                 goto check_header_out;
 350         }
 351         if (starts_with(line->buf, "[PATCH]") && isspace(line->buf[7])) {
 352                 for (i = 0; header[i]; i++) {
 353                         if (!strcmp("Subject", header[i])) {
 354                                 handle_header(&hdr_data[i], line);
 355                                 ret = 1;
 356                                 goto check_header_out;
 357                         }
 358                 }
 359         }
 360
 361 check_header_out:
 362         strbuf_release(&sb);
 363         return ret;
 364 }
 365
 366 static int is_rfc2822_header(const struct strbuf *line)
 367 {
 368         /*
 369          * The section that defines the loosest possible
 370          * field name is "3.6.8 Optional fields".
 371          *
 372          * optional-field = field-name ":" unstructured CRLF
 373          * field-name = 1*ftext
 374          * ftext = %d33-57 / %59-126
 375          */
 376         int ch;
 377         char *cp = line->buf;
 378
 379         /* Count mbox From headers as headers */
 380         if (starts_with(cp, "From ") || starts_with(cp, ">From "))
 381                 return 1;
 382
 383         while ((ch = *cp++)) {
 384                 if (ch == ':')
 385                         return 1;
 386                 if ((33 <= ch && ch <= 57) ||
 387                     (59 <= ch && ch <= 126))
 388                         continue;
 389                 break;
 390         }
 391         return 0;
 392 }
 393
 394 static int read_one_header_line(struct strbuf *line, FILE *in)
 395 {
 396         /* Get the first part of the line. */
 397         if (strbuf_getline(line, in, '\n'))
 398                 return 0;
 399
 400         /*
 401          * Is it an empty line or not a valid rfc2822 header?
 402          * If so, stop here, and return false ("not a header")
 403          */
 404         strbuf_rtrim(line);
 405         if (!line->len || !is_rfc2822_header(line)) {
 406                 /* Re-add the newline */
 407                 strbuf_addch(line, '\n');
 408                 return 0;
 409         }
 410
 411         /*
 412          * Now we need to eat all the continuation lines..
 413          * Yuck, 2822 header "folding"
 414          */
 415         for (;;) {
 416                 int peek;
 417                 struct strbuf continuation = STRBUF_INIT;
 418
 419                 peek = fgetc(in); ungetc(peek, in);
 420                 if (peek != ' ' && peek != '\t')
 421                         break;
 422                 if (strbuf_getline(&continuation, in, '\n'))
 423                         break;
 424                 continuation.buf[0] = ' ';
 425                 strbuf_rtrim(&continuation);
 426                 strbuf_addbuf(line, &continuation);
 427         }
 428
 429         return 1;
 430 }
 431
 432 static struct strbuf *decode_q_segment(const struct strbuf *q_seg, int rfc2047)
 433 {
 434         const char *in = q_seg->buf;
 435         int c;
 436         struct strbuf *out = xmalloc(sizeof(struct strbuf));
 437         strbuf_init(out, q_seg->len);
 438
 439         while ((c = *in++) != 0) {
 440                 if (c == '=') {
 441                         int d = *in++;
 442                         if (d == '\n' || !d)
 443                                 break; /* drop trailing newline */
 444                         strbuf_addch(out, (hexval(d) << 4) | hexval(*in++));
 445                         continue;
 446                 }
 447                 if (rfc2047 && c == '_') /* rfc2047 4.2 (2) */
 448                         c = 0x20;
 449                 strbuf_addch(out, c);
 450         }
 451         return out;
 452 }
 453
 454 static struct strbuf *decode_b_segment(const struct strbuf *b_seg)
 455 {
 456         /* Decode in..ep, possibly in-place to ot */
 457         int c, pos = 0, acc = 0;
 458         const char *in = b_seg->buf;
 459         struct strbuf *out = xmalloc(sizeof(struct strbuf));
 460         strbuf_init(out, b_seg->len);
 461
 462         while ((c = *in++) != 0) {
 463                 if (c == '+')
 464                         c = 62;
 465                 else if (c == '/')
 466                         c = 63;
 467                 else if ('A' <= c && c <= 'Z')
 468                         c -= 'A';
 469                 else if ('a' <= c && c <= 'z')
 470                         c -= 'a' - 26;
 471                 else if ('0' <= c && c <= '9')
 472                         c -= '0' - 52;
 473                 else
 474                         continue; /* garbage */
 475                 switch (pos++) {
 476                 case 0:
 477                         acc = (c << 2);
 478                         break;
 479                 case 1:
 480                         strbuf_addch(out, (acc | (c >> 4)));
 481                         acc = (c & 15) << 4;
 482                         break;
 483                 case 2:
 484                         strbuf_addch(out, (acc | (c >> 2)));
 485                         acc = (c & 3) << 6;
 486                         break;
 487                 case 3:
 488                         strbuf_addch(out, (acc | c));
 489                         acc = pos = 0;
 490                         break;
 491                 }
 492         }
 493         return out;
 494 }
 495
 496 static void convert_to_utf8(struct strbuf *line, const char *charset)
 497 {
 498         char *out;
 499
 500         if (!charset || !*charset)
 501                 return;
 502
 503         if (same_encoding(metainfo_charset, charset))
 504                 return;
 505         out = reencode_string(line->buf, metainfo_charset, charset);
 506         if (!out)
 507                 die("cannot convert from %s to %s",
 508                     charset, metainfo_charset);
 509         strbuf_attach(line, out, strlen(out), strlen(out));
 510 }
 511
 512 static int decode_header_bq(struct strbuf *it)
 513 {
 514         char *in, *ep, *cp;
 515         struct strbuf outbuf = STRBUF_INIT, *dec;
 516         struct strbuf charset_q = STRBUF_INIT, piecebuf = STRBUF_INIT;
 517         int rfc2047 = 0;
 518
 519         in = it->buf;
 520         while (in - it->buf <= it->len && (ep = strstr(in, "=?")) != NULL) {
 521                 int encoding;
 522                 strbuf_reset(&charset_q);
 523                 strbuf_reset(&piecebuf);
 524                 rfc2047 = 1;
 525
 526                 if (in != ep) {
 527                         /*
 528                          * We are about to process an encoded-word
 529                          * that begins at ep, but there is something
 530                          * before the encoded word.
 531                          */
 532                         char *scan;
 533                         for (scan = in; scan < ep; scan++)
 534                                 if (!isspace(*scan))
 535                                         break;
 536
 537                         if (scan != ep || in == it->buf) {
 538                                 /*
 539                                  * We should not lose that "something",
 540                                  * unless we have just processed an
 541                                  * encoded-word, and there is only LWS
 542                                  * before the one we are about to process.
 543                                  */
 544                                 strbuf_add(&outbuf, in, ep - in);
 545                         }
 546                 }
 547                 /* E.g.
 548                  * ep : "=?iso-2022-jp?B?GyR...?= foo"
 549                  * ep : "=?ISO-8859-1?Q?Foo=FCbar?= baz"
 550                  */
 551                 ep += 2;
 552
 553                 if (ep - it->buf >= it->len || !(cp = strchr(ep, '?')))
 554                         goto decode_header_bq_out;
 555
 556                 if (cp + 3 - it->buf > it->len)
 557                         goto decode_header_bq_out;
 558                 strbuf_add(&charset_q, ep, cp - ep);
 559
 560                 encoding = cp[1];
 561                 if (!encoding || cp[2] != '?')
 562                         goto decode_header_bq_out;
 563                 ep = strstr(cp + 3, "?=");
 564                 if (!ep)
 565                         goto decode_header_bq_out;
 566                 strbuf_add(&piecebuf, cp + 3, ep - cp - 3);
 567                 switch (tolower(encoding)) {
 568                 default:
 569                         goto decode_header_bq_out;
 570                 case 'b':
 571                         dec = decode_b_segment(&piecebuf);
 572                         break;
 573                 case 'q':
 574                         dec = decode_q_segment(&piecebuf, 1);
 575                         break;
 576                 }
 577                 if (metainfo_charset)
 578                         convert_to_utf8(dec, charset_q.buf);
 579
 580                 strbuf_addbuf(&outbuf, dec);
 581                 strbuf_release(dec);
 582                 free(dec);
 583                 in = ep + 2;
 584         }
 585         strbuf_addstr(&outbuf, in);
 586         strbuf_reset(it);
 587         strbuf_addbuf(it, &outbuf);
 588 decode_header_bq_out:
 589         strbuf_release(&outbuf);
 590         strbuf_release(&charset_q);
 591         strbuf_release(&piecebuf);
 592         return rfc2047;
 593 }
 594
 595 static void decode_header(struct strbuf *it)
 596 {
 597         if (decode_header_bq(it))
 598                 return;
 599         /* otherwise "it" is a straight copy of the input.
 600          * This can be binary guck but there is no charset specified.
 601          */
 602         if (metainfo_charset)
 603                 convert_to_utf8(it, "");
 604 }
 605
 606 static void decode_transfer_encoding(struct strbuf *line)
 607 {
 608         struct strbuf *ret;
 609
 610         switch (transfer_encoding) {
 611         case TE_QP:
 612                 ret = decode_q_segment(line, 0);
 613                 break;
 614         case TE_BASE64:
 615                 ret = decode_b_segment(line);
 616                 break;
 617         case TE_DONTCARE:
 618         default:
 619                 return;
 620         }
 621         strbuf_reset(line);
 622         strbuf_addbuf(line, ret);
 623         strbuf_release(ret);
 624         free(ret);
 625 }
 626
 627 static void handle_filter(struct strbuf *line);
 628
 629 static int find_boundary(void)
 630 {
 631         while (!strbuf_getline(&line, fin, '\n')) {
 632                 if (*content_top && is_multipart_boundary(&line))
 633                         return 1;
 634         }
 635         return 0;
 636 }
 637
 638 static int handle_boundary(void)
 639 {
 640         struct strbuf newline = STRBUF_INIT;
 641
 642         strbuf_addch(&newline, '\n');
 643 again:
 644         if (line.len >= (*content_top)->len + 2 &&
 645             !memcmp(line.buf + (*content_top)->len, "--", 2)) {
 646                 /* we hit an end boundary */
 647                 /* pop the current boundary off the stack */
 648                 strbuf_release(*content_top);
 649                 free(*content_top);
 650                 *content_top = NULL;
 651
 652                 /* technically won't happen as is_multipart_boundary()
 653                    will fail first.  But just in case..
 654                  */
 655                 if (--content_top < content) {
 656                         fprintf(stderr, "Detected mismatched boundaries, "
 657                                         "can't recover\n");
 658                         exit(1);
 659                 }
 660                 handle_filter(&newline);
 661                 strbuf_release(&newline);
 662
 663                 /* skip to the next boundary */
 664                 if (!find_boundary())
 665                         return 0;
 666                 goto again;
 667         }
 668
 669         /* set some defaults */
 670         transfer_encoding = TE_DONTCARE;
 671         strbuf_reset(&charset);
 672
 673         /* slurp in this section's info */
 674         while (read_one_header_line(&line, fin))
 675                 check_header(&line, p_hdr_data, 0);
 676
 677         strbuf_release(&newline);
 678         /* replenish line */
 679         if (strbuf_getline(&line, fin, '\n'))
 680                 return 0;
 681         strbuf_addch(&line, '\n');
 682         return 1;
 683 }
 684
 685 static inline int patchbreak(const struct strbuf *line)
 686 {
 687         size_t i;
 688
 689         /* Beginning of a "diff -" header? */
 690         if (starts_with(line->buf, "diff -"))
 691                 return 1;
 692
 693         /* CVS "Index: " line? */
 694         if (starts_with(line->buf, "Index: "))
 695                 return 1;
 696
 697         /*
 698          * "--- <filename>" starts patches without headers
 699          * "---<sp>*" is a manual separator
 700          */
 701         if (line->len < 4)
 702                 return 0;
 703
 704         if (starts_with(line->buf, "---")) {
 705                 /* space followed by a filename? */
 706                 if (line->buf[3] == ' ' && !isspace(line->buf[4]))
 707                         return 1;
 708                 /* Just whitespace? */
 709                 for (i = 3; i < line->len; i++) {
 710                         unsigned char c = line->buf[i];
 711                         if (c == '\n')
 712                                 return 1;
 713                         if (!isspace(c))
 714                                 break;
 715                 }
 716                 return 0;
 717         }
 718         return 0;
 719 }
 720
 721 static int is_scissors_line(const struct strbuf *line)
 722 {
 723         size_t i, len = line->len;
 724         int scissors = 0, gap = 0;
 725         int first_nonblank = -1;
 726         int last_nonblank = 0, visible, perforation = 0, in_perforation = 0;
 727         const char *buf = line->buf;
 728
 729         for (i = 0; i < len; i++) {
 730                 if (isspace(buf[i])) {
 731                         if (in_perforation) {
 732                                 perforation++;
 733                                 gap++;
 734                         }
 735                         continue;
 736                 }
 737                 last_nonblank = i;
 738                 if (first_nonblank < 0)
 739                         first_nonblank = i;
 740                 if (buf[i] == '-') {
 741                         in_perforation = 1;
 742                         perforation++;
 743                         continue;
 744                 }
 745                 if (i + 1 < len &&
 746                     (!memcmp(buf + i, ">8", 2) || !memcmp(buf + i, "8<", 2) ||
 747                      !memcmp(buf + i, ">%", 2) || !memcmp(buf + i, "%<", 2))) {
 748                         in_perforation = 1;
 749                         perforation += 2;
 750                         scissors += 2;
 751                         i++;
 752                         continue;
 753                 }
 754                 in_perforation = 0;
 755         }
 756
 757         /*
 758          * The mark must be at least 8 bytes long (e.g. "-- >8 --").
 759          * Even though there can be arbitrary cruft on the same line
 760          * (e.g. "cut here"), in order to avoid misidentification, the
 761          * perforation must occupy more than a third of the visible
 762          * width of the line, and dashes and scissors must occupy more
 763          * than half of the perforation.
 764          */
 765
 766         visible = last_nonblank - first_nonblank + 1;
 767         return (scissors && 8 <= visible &&
 768                 visible < perforation * 3 &&
 769                 gap * 2 < perforation);
 770 }
 771
 772 static int handle_commit_msg(struct strbuf *line)
 773 {
 774         static int still_looking = 1;
 775
 776         if (!cmitmsg)
 777                 return 0;
 778
 779         if (still_looking) {
 780                 if (!line->len || (line->len == 1 && line->buf[0] == '\n'))
 781                         return 0;
 782         }
 783
 784         if (use_inbody_headers && still_looking) {
 785                 still_looking = check_header(line, s_hdr_data, 0);
 786                 if (still_looking)
 787                         return 0;
 788         } else
 789                 /* Only trim the first (blank) line of the commit message
 790                  * when ignoring in-body headers.
 791                  */
 792                 still_looking = 0;
 793
 794         /* normalize the log message to UTF-8. */
 795         if (metainfo_charset)
 796                 convert_to_utf8(line, charset.buf);
 797
 798         if (use_scissors && is_scissors_line(line)) {
 799                 int i;
 800                 if (fseek(cmitmsg, 0L, SEEK_SET))
 801                         die_errno("Could not rewind output message file");
 802                 if (ftruncate(fileno(cmitmsg), 0))
 803                         die_errno("Could not truncate output message file at scissors");
 804                 still_looking = 1;
 805
 806                 /*
 807                  * We may have already read "secondary headers"; purge
 808                  * them to give ourselves a clean restart.
 809                  */
 810                 for (i = 0; header[i]; i++) {
 811                         if (s_hdr_data[i])
 812                                 strbuf_release(s_hdr_data[i]);
 813                         s_hdr_data[i] = NULL;
 814                 }
 815                 return 0;
 816         }
 817
 818         if (patchbreak(line)) {
 819                 fclose(cmitmsg);
 820                 cmitmsg = NULL;
 821                 return 1;
 822         }
 823
 824         fputs(line->buf, cmitmsg);
 825         return 0;
 826 }
 827
 828 static void handle_patch(const struct strbuf *line)
 829 {
 830         fwrite(line->buf, 1, line->len, patchfile);
 831         patch_lines++;
 832 }
 833
 834 static void handle_filter(struct strbuf *line)
 835 {
 836         static int filter = 0;
 837
 838         /* filter tells us which part we left off on */
 839         switch (filter) {
 840         case 0:
 841                 if (!handle_commit_msg(line))
 842                         break;
 843                 filter++;
 844         case 1:
 845                 handle_patch(line);
 846                 break;
 847         }
 848 }
 849
 850 static void handle_body(void)
 851 {
 852         struct strbuf prev = STRBUF_INIT;
 853
 854         /* Skip up to the first boundary */
 855         if (*content_top) {
 856                 if (!find_boundary())
 857                         goto handle_body_out;
 858         }
 859
 860         do {
 861                 /* process any boundary lines */
 862                 if (*content_top && is_multipart_boundary(&line)) {
 863                         /* flush any leftover */
 864                         if (prev.len) {
 865                                 handle_filter(&prev);
 866                                 strbuf_reset(&prev);
 867                         }
 868                         if (!handle_boundary())
 869                                 goto handle_body_out;
 870                 }
 871
 872                 /* Unwrap transfer encoding */
 873                 decode_transfer_encoding(&line);
 874
 875                 switch (transfer_encoding) {
 876                 case TE_BASE64:
 877                 case TE_QP:
 878                 {
 879                         struct strbuf **lines, **it, *sb;
 880
 881                         /* Prepend any previous partial lines */
 882                         strbuf_insert(&line, 0, prev.buf, prev.len);
 883                         strbuf_reset(&prev);
 884
 885                         /*
 886                          * This is a decoded line that may contain
 887                          * multiple new lines.  Pass only one chunk
 888                          * at a time to handle_filter()
 889                          */
 890                         lines = strbuf_split(&line, '\n');
 891                         for (it = lines; (sb = *it); it++) {
 892                                 if (*(it + 1) == NULL) /* The last line */
 893                                         if (sb->buf[sb->len - 1] != '\n') {
 894                                                 /* Partial line, save it for later. */
 895                                                 strbuf_addbuf(&prev, sb);
 896                                                 break;
 897                                         }
 898                                 handle_filter(sb);
 899                         }
 900                         /*
 901                          * The partial chunk is saved in "prev" and will be
 902                          * appended by the next iteration of read_line_with_nul().
 903                          */
 904                         strbuf_list_free(lines);
 905                         break;
 906                 }
 907                 default:
 908                         handle_filter(&line);
 909                 }
 910
 911         } while (!strbuf_getwholeline(&line, fin, '\n'));
 912
 913 handle_body_out:
 914         strbuf_release(&prev);
 915 }
 916
 917 static void output_header_lines(FILE *fout, const char *hdr, const struct strbuf *data)
 918 {
 919         const char *sp = data->buf;
 920         while (1) {
 921                 char *ep = strchr(sp, '\n');
 922                 int len;
 923                 if (!ep)
 924                         len = strlen(sp);
 925                 else
 926                         len = ep - sp;
 927                 fprintf(fout, "%s: %.*s\n", hdr, len, sp);
 928                 if (!ep)
 929                         break;
 930                 sp = ep + 1;
 931         }
 932 }
 933
 934 static void handle_info(void)
 935 {
 936         struct strbuf *hdr;
 937         int i;
 938
 939         for (i = 0; header[i]; i++) {
 940                 /* only print inbody headers if we output a patch file */
 941                 if (patch_lines && s_hdr_data[i])
 942                         hdr = s_hdr_data[i];
 943                 else if (p_hdr_data[i])
 944                         hdr = p_hdr_data[i];
 945                 else
 946                         continue;
 947
 948                 if (!strcmp(header[i], "Subject")) {
 949                         if (!keep_subject) {
 950                                 cleanup_subject(hdr);
 951                                 cleanup_space(hdr);
 952                         }
 953                         output_header_lines(fout, "Subject", hdr);
 954                 } else if (!strcmp(header[i], "From")) {
 955                         cleanup_space(hdr);
 956                         handle_from(hdr);
 957                         fprintf(fout, "Author: %s\n", name.buf);
 958                         fprintf(fout, "Email: %s\n", email.buf);
 959                 } else {
 960                         cleanup_space(hdr);
 961                         fprintf(fout, "%s: %s\n", header[i], hdr->buf);
 962                 }
 963         }
 964         fprintf(fout, "\n");
 965 }
 966
 967 static int mailinfo(FILE *in, FILE *out, const char *msg, const char *patch)
 968 {
 969         int peek;
 970         fin = in;
 971         fout = out;
 972
 973         cmitmsg = fopen(msg, "w");
 974         if (!cmitmsg) {
 975                 perror(msg);
 976                 return -1;
 977         }
 978         patchfile = fopen(patch, "w");
 979         if (!patchfile) {
 980                 perror(patch);
 981                 fclose(cmitmsg);
 982                 return -1;
 983         }
 984
 985         p_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*p_hdr_data));
 986         s_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*s_hdr_data));
 987
 988         do {
 989                 peek = fgetc(in);
 990         } while (isspace(peek));
 991         ungetc(peek, in);
 992
 993         /* process the email header */
 994         while (read_one_header_line(&line, fin))
 995                 check_header(&line, p_hdr_data, 1);
 996
 997         handle_body();
 998         handle_info();
 999
1000         return 0;
1001 }
1002
1003 static int git_mailinfo_config(const char *var, const char *value, void *unused)
1004 {
1005         if (!starts_with(var, "mailinfo."))
1006                 return git_default_config(var, value, unused);
1007         if (!strcmp(var, "mailinfo.scissors")) {
1008                 use_scissors = git_config_bool(var, value);
1009                 return 0;
1010         }
1011         /* perhaps others here */
1012         return 0;
1013 }
1014
1015 static const char mailinfo_usage[] =
1016         "git mailinfo [-k|-b] [-u | --encoding=<encoding> | -n] [--scissors | --no-scissors] msg patch < mail >info";
1017
1018 int cmd_mailinfo(int argc, const char **argv, const char *prefix)
1019 {
1020         const char *def_charset;
1021
1022         /* NEEDSWORK: might want to do the optional .git/ directory
1023          * discovery
1024          */
1025         git_config(git_mailinfo_config, NULL);
1026
1027         def_charset = get_commit_output_encoding();
1028         metainfo_charset = def_charset;
1029
1030         while (1 < argc && argv[1][0] == '-') {
1031                 if (!strcmp(argv[1], "-k"))
1032                         keep_subject = 1;
1033                 else if (!strcmp(argv[1], "-b"))
1034                         keep_non_patch_brackets_in_subject = 1;
1035                 else if (!strcmp(argv[1], "-u"))
1036                         metainfo_charset = def_charset;
1037                 else if (!strcmp(argv[1], "-n"))
1038                         metainfo_charset = NULL;
1039                 else if (starts_with(argv[1], "--encoding="))
1040                         metainfo_charset = argv[1] + 11;
1041                 else if (!strcmp(argv[1], "--scissors"))
1042                         use_scissors = 1;
1043                 else if (!strcmp(argv[1], "--no-scissors"))
1044                         use_scissors = 0;
1045                 else if (!strcmp(argv[1], "--no-inbody-headers"))
1046                         use_inbody_headers = 0;
1047                 else
1048                         usage(mailinfo_usage);
1049                 argc--; argv++;
1050         }
1051
1052         if (argc != 3)
1053                 usage(mailinfo_usage);
1054
1055         return !!mailinfo(stdin, stdout, argv[1], argv[2]);
1056 }