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                                 /*
 256                                  * If the input had a space after the ], keep
 257                                  * it.  We don't bother with finding the end of
 258                                  * the space, since we later normalize it
 259                                  * anyway.
 260                                  */
 261                                 if (isspace(subject->buf[at]))
 262                                         at += 1;
 263                         }
 264                         continue;
 265                 }
 266                 break;
 267         }
 268         strbuf_trim(subject);
 269 }
 270
 271 static void cleanup_space(struct strbuf *sb)
 272 {
 273         size_t pos, cnt;
 274         for (pos = 0; pos < sb->len; pos++) {
 275                 if (isspace(sb->buf[pos])) {
 276                         sb->buf[pos] = ' ';
 277                         for (cnt = 0; isspace(sb->buf[pos + cnt + 1]); cnt++);
 278                         strbuf_remove(sb, pos + 1, cnt);
 279                 }
 280         }
 281 }
 282
 283 static void decode_header(struct strbuf *line);
 284 static const char *header[MAX_HDR_PARSED] = {
 285         "From","Subject","Date",
 286 };
 287
 288 static inline int cmp_header(const struct strbuf *line, const char *hdr)
 289 {
 290         int len = strlen(hdr);
 291         return !strncasecmp(line->buf, hdr, len) && line->len > len &&
 292                         line->buf[len] == ':' && isspace(line->buf[len + 1]);
 293 }
 294
 295 static int check_header(const struct strbuf *line,
 296                                 struct strbuf *hdr_data[], int overwrite)
 297 {
 298         int i, ret = 0, len;
 299         struct strbuf sb = STRBUF_INIT;
 300         /* search for the interesting parts */
 301         for (i = 0; header[i]; i++) {
 302                 int len = strlen(header[i]);
 303                 if ((!hdr_data[i] || overwrite) && cmp_header(line, header[i])) {
 304                         /* Unwrap inline B and Q encoding, and optionally
 305                          * normalize the meta information to utf8.
 306                          */
 307                         strbuf_add(&sb, line->buf + len + 2, line->len - len - 2);
 308                         decode_header(&sb);
 309                         handle_header(&hdr_data[i], &sb);
 310                         ret = 1;
 311                         goto check_header_out;
 312                 }
 313         }
 314
 315         /* Content stuff */
 316         if (cmp_header(line, "Content-Type")) {
 317                 len = strlen("Content-Type: ");
 318                 strbuf_add(&sb, line->buf + len, line->len - len);
 319                 decode_header(&sb);
 320                 strbuf_insert(&sb, 0, "Content-Type: ", len);
 321                 handle_content_type(&sb);
 322                 ret = 1;
 323                 goto check_header_out;
 324         }
 325         if (cmp_header(line, "Content-Transfer-Encoding")) {
 326                 len = strlen("Content-Transfer-Encoding: ");
 327                 strbuf_add(&sb, line->buf + len, line->len - len);
 328                 decode_header(&sb);
 329                 handle_content_transfer_encoding(&sb);
 330                 ret = 1;
 331                 goto check_header_out;
 332         }
 333
 334         /* for inbody stuff */
 335         if (!prefixcmp(line->buf, ">From") && isspace(line->buf[5])) {
 336                 ret = 1; /* Should this return 0? */
 337                 goto check_header_out;
 338         }
 339         if (!prefixcmp(line->buf, "[PATCH]") && isspace(line->buf[7])) {
 340                 for (i = 0; header[i]; i++) {
 341                         if (!memcmp("Subject", header[i], 7)) {
 342                                 handle_header(&hdr_data[i], line);
 343                                 ret = 1;
 344                                 goto check_header_out;
 345                         }
 346                 }
 347         }
 348
 349 check_header_out:
 350         strbuf_release(&sb);
 351         return ret;
 352 }
 353
 354 static int is_rfc2822_header(const struct strbuf *line)
 355 {
 356         /*
 357          * The section that defines the loosest possible
 358          * field name is "3.6.8 Optional fields".
 359          *
 360          * optional-field = field-name ":" unstructured CRLF
 361          * field-name = 1*ftext
 362          * ftext = %d33-57 / %59-126
 363          */
 364         int ch;
 365         char *cp = line->buf;
 366
 367         /* Count mbox From headers as headers */
 368         if (!prefixcmp(cp, "From ") || !prefixcmp(cp, ">From "))
 369                 return 1;
 370
 371         while ((ch = *cp++)) {
 372                 if (ch == ':')
 373                         return 1;
 374                 if ((33 <= ch && ch <= 57) ||
 375                     (59 <= ch && ch <= 126))
 376                         continue;
 377                 break;
 378         }
 379         return 0;
 380 }
 381
 382 static int read_one_header_line(struct strbuf *line, FILE *in)
 383 {
 384         /* Get the first part of the line. */
 385         if (strbuf_getline(line, in, '\n'))
 386                 return 0;
 387
 388         /*
 389          * Is it an empty line or not a valid rfc2822 header?
 390          * If so, stop here, and return false ("not a header")
 391          */
 392         strbuf_rtrim(line);
 393         if (!line->len || !is_rfc2822_header(line)) {
 394                 /* Re-add the newline */
 395                 strbuf_addch(line, '\n');
 396                 return 0;
 397         }
 398
 399         /*
 400          * Now we need to eat all the continuation lines..
 401          * Yuck, 2822 header "folding"
 402          */
 403         for (;;) {
 404                 int peek;
 405                 struct strbuf continuation = STRBUF_INIT;
 406
 407                 peek = fgetc(in); ungetc(peek, in);
 408                 if (peek != ' ' && peek != '\t')
 409                         break;
 410                 if (strbuf_getline(&continuation, in, '\n'))
 411                         break;
 412                 continuation.buf[0] = ' ';
 413                 strbuf_rtrim(&continuation);
 414                 strbuf_addbuf(line, &continuation);
 415         }
 416
 417         return 1;
 418 }
 419
 420 static struct strbuf *decode_q_segment(const struct strbuf *q_seg, int rfc2047)
 421 {
 422         const char *in = q_seg->buf;
 423         int c;
 424         struct strbuf *out = xmalloc(sizeof(struct strbuf));
 425         strbuf_init(out, q_seg->len);
 426
 427         while ((c = *in++) != 0) {
 428                 if (c == '=') {
 429                         int d = *in++;
 430                         if (d == '\n' || !d)
 431                                 break; /* drop trailing newline */
 432                         strbuf_addch(out, (hexval(d) << 4) | hexval(*in++));
 433                         continue;
 434                 }
 435                 if (rfc2047 && c == '_') /* rfc2047 4.2 (2) */
 436                         c = 0x20;
 437                 strbuf_addch(out, c);
 438         }
 439         return out;
 440 }
 441
 442 static struct strbuf *decode_b_segment(const struct strbuf *b_seg)
 443 {
 444         /* Decode in..ep, possibly in-place to ot */
 445         int c, pos = 0, acc = 0;
 446         const char *in = b_seg->buf;
 447         struct strbuf *out = xmalloc(sizeof(struct strbuf));
 448         strbuf_init(out, b_seg->len);
 449
 450         while ((c = *in++) != 0) {
 451                 if (c == '+')
 452                         c = 62;
 453                 else if (c == '/')
 454                         c = 63;
 455                 else if ('A' <= c && c <= 'Z')
 456                         c -= 'A';
 457                 else if ('a' <= c && c <= 'z')
 458                         c -= 'a' - 26;
 459                 else if ('0' <= c && c <= '9')
 460                         c -= '0' - 52;
 461                 else
 462                         continue; /* garbage */
 463                 switch (pos++) {
 464                 case 0:
 465                         acc = (c << 2);
 466                         break;
 467                 case 1:
 468                         strbuf_addch(out, (acc | (c >> 4)));
 469                         acc = (c & 15) << 4;
 470                         break;
 471                 case 2:
 472                         strbuf_addch(out, (acc | (c >> 2)));
 473                         acc = (c & 3) << 6;
 474                         break;
 475                 case 3:
 476                         strbuf_addch(out, (acc | c));
 477                         acc = pos = 0;
 478                         break;
 479                 }
 480         }
 481         return out;
 482 }
 483
 484 static void convert_to_utf8(struct strbuf *line, const char *charset)
 485 {
 486         char *out;
 487
 488         if (!charset || !*charset)
 489                 return;
 490         if (!strcasecmp(metainfo_charset, charset))
 491                 return;
 492         out = reencode_string(line->buf, metainfo_charset, charset);
 493         if (!out)
 494                 die("cannot convert from %s to %s",
 495                     charset, metainfo_charset);
 496         strbuf_attach(line, out, strlen(out), strlen(out));
 497 }
 498
 499 static int decode_header_bq(struct strbuf *it)
 500 {
 501         char *in, *ep, *cp;
 502         struct strbuf outbuf = STRBUF_INIT, *dec;
 503         struct strbuf charset_q = STRBUF_INIT, piecebuf = STRBUF_INIT;
 504         int rfc2047 = 0;
 505
 506         in = it->buf;
 507         while (in - it->buf <= it->len && (ep = strstr(in, "=?")) != NULL) {
 508                 int encoding;
 509                 strbuf_reset(&charset_q);
 510                 strbuf_reset(&piecebuf);
 511                 rfc2047 = 1;
 512
 513                 if (in != ep) {
 514                         /*
 515                          * We are about to process an encoded-word
 516                          * that begins at ep, but there is something
 517                          * before the encoded word.
 518                          */
 519                         char *scan;
 520                         for (scan = in; scan < ep; scan++)
 521                                 if (!isspace(*scan))
 522                                         break;
 523
 524                         if (scan != ep || in == it->buf) {
 525                                 /*
 526                                  * We should not lose that "something",
 527                                  * unless we have just processed an
 528                                  * encoded-word, and there is only LWS
 529                                  * before the one we are about to process.
 530                                  */
 531                                 strbuf_add(&outbuf, in, ep - in);
 532                         }
 533                 }
 534                 /* E.g.
 535                  * ep : "=?iso-2022-jp?B?GyR...?= foo"
 536                  * ep : "=?ISO-8859-1?Q?Foo=FCbar?= baz"
 537                  */
 538                 ep += 2;
 539
 540                 if (ep - it->buf >= it->len || !(cp = strchr(ep, '?')))
 541                         goto decode_header_bq_out;
 542
 543                 if (cp + 3 - it->buf > it->len)
 544                         goto decode_header_bq_out;
 545                 strbuf_add(&charset_q, ep, cp - ep);
 546
 547                 encoding = cp[1];
 548                 if (!encoding || cp[2] != '?')
 549                         goto decode_header_bq_out;
 550                 ep = strstr(cp + 3, "?=");
 551                 if (!ep)
 552                         goto decode_header_bq_out;
 553                 strbuf_add(&piecebuf, cp + 3, ep - cp - 3);
 554                 switch (tolower(encoding)) {
 555                 default:
 556                         goto decode_header_bq_out;
 557                 case 'b':
 558                         dec = decode_b_segment(&piecebuf);
 559                         break;
 560                 case 'q':
 561                         dec = decode_q_segment(&piecebuf, 1);
 562                         break;
 563                 }
 564                 if (metainfo_charset)
 565                         convert_to_utf8(dec, charset_q.buf);
 566
 567                 strbuf_addbuf(&outbuf, dec);
 568                 strbuf_release(dec);
 569                 free(dec);
 570                 in = ep + 2;
 571         }
 572         strbuf_addstr(&outbuf, in);
 573         strbuf_reset(it);
 574         strbuf_addbuf(it, &outbuf);
 575 decode_header_bq_out:
 576         strbuf_release(&outbuf);
 577         strbuf_release(&charset_q);
 578         strbuf_release(&piecebuf);
 579         return rfc2047;
 580 }
 581
 582 static void decode_header(struct strbuf *it)
 583 {
 584         if (decode_header_bq(it))
 585                 return;
 586         /* otherwise "it" is a straight copy of the input.
 587          * This can be binary guck but there is no charset specified.
 588          */
 589         if (metainfo_charset)
 590                 convert_to_utf8(it, "");
 591 }
 592
 593 static void decode_transfer_encoding(struct strbuf *line)
 594 {
 595         struct strbuf *ret;
 596
 597         switch (transfer_encoding) {
 598         case TE_QP:
 599                 ret = decode_q_segment(line, 0);
 600                 break;
 601         case TE_BASE64:
 602                 ret = decode_b_segment(line);
 603                 break;
 604         case TE_DONTCARE:
 605         default:
 606                 return;
 607         }
 608         strbuf_reset(line);
 609         strbuf_addbuf(line, ret);
 610         strbuf_release(ret);
 611         free(ret);
 612 }
 613
 614 static void handle_filter(struct strbuf *line);
 615
 616 static int find_boundary(void)
 617 {
 618         while (!strbuf_getline(&line, fin, '\n')) {
 619                 if (*content_top && is_multipart_boundary(&line))
 620                         return 1;
 621         }
 622         return 0;
 623 }
 624
 625 static int handle_boundary(void)
 626 {
 627         struct strbuf newline = STRBUF_INIT;
 628
 629         strbuf_addch(&newline, '\n');
 630 again:
 631         if (line.len >= (*content_top)->len + 2 &&
 632             !memcmp(line.buf + (*content_top)->len, "--", 2)) {
 633                 /* we hit an end boundary */
 634                 /* pop the current boundary off the stack */
 635                 strbuf_release(*content_top);
 636                 free(*content_top);
 637                 *content_top = NULL;
 638
 639                 /* technically won't happen as is_multipart_boundary()
 640                    will fail first.  But just in case..
 641                  */
 642                 if (--content_top < content) {
 643                         fprintf(stderr, "Detected mismatched boundaries, "
 644                                         "can't recover\n");
 645                         exit(1);
 646                 }
 647                 handle_filter(&newline);
 648                 strbuf_release(&newline);
 649
 650                 /* skip to the next boundary */
 651                 if (!find_boundary())
 652                         return 0;
 653                 goto again;
 654         }
 655
 656         /* set some defaults */
 657         transfer_encoding = TE_DONTCARE;
 658         strbuf_reset(&charset);
 659         message_type = TYPE_TEXT;
 660
 661         /* slurp in this section's info */
 662         while (read_one_header_line(&line, fin))
 663                 check_header(&line, p_hdr_data, 0);
 664
 665         strbuf_release(&newline);
 666         /* replenish line */
 667         if (strbuf_getline(&line, fin, '\n'))
 668                 return 0;
 669         strbuf_addch(&line, '\n');
 670         return 1;
 671 }
 672
 673 static inline int patchbreak(const struct strbuf *line)
 674 {
 675         size_t i;
 676
 677         /* Beginning of a "diff -" header? */
 678         if (!prefixcmp(line->buf, "diff -"))
 679                 return 1;
 680
 681         /* CVS "Index: " line? */
 682         if (!prefixcmp(line->buf, "Index: "))
 683                 return 1;
 684
 685         /*
 686          * "--- <filename>" starts patches without headers
 687          * "---<sp>*" is a manual separator
 688          */
 689         if (line->len < 4)
 690                 return 0;
 691
 692         if (!prefixcmp(line->buf, "---")) {
 693                 /* space followed by a filename? */
 694                 if (line->buf[3] == ' ' && !isspace(line->buf[4]))
 695                         return 1;
 696                 /* Just whitespace? */
 697                 for (i = 3; i < line->len; i++) {
 698                         unsigned char c = line->buf[i];
 699                         if (c == '\n')
 700                                 return 1;
 701                         if (!isspace(c))
 702                                 break;
 703                 }
 704                 return 0;
 705         }
 706         return 0;
 707 }
 708
 709 static int is_scissors_line(const struct strbuf *line)
 710 {
 711         size_t i, len = line->len;
 712         int scissors = 0, gap = 0;
 713         int first_nonblank = -1;
 714         int last_nonblank = 0, visible, perforation = 0, in_perforation = 0;
 715         const char *buf = line->buf;
 716
 717         for (i = 0; i < len; i++) {
 718                 if (isspace(buf[i])) {
 719                         if (in_perforation) {
 720                                 perforation++;
 721                                 gap++;
 722                         }
 723                         continue;
 724                 }
 725                 last_nonblank = i;
 726                 if (first_nonblank < 0)
 727                         first_nonblank = i;
 728                 if (buf[i] == '-') {
 729                         in_perforation = 1;
 730                         perforation++;
 731                         continue;
 732                 }
 733                 if (i + 1 < len &&
 734                     (!memcmp(buf + i, ">8", 2) || !memcmp(buf + i, "8<", 2) ||
 735                      !memcmp(buf + i, ">%", 2) || !memcmp(buf + i, "%<", 2))) {
 736                         in_perforation = 1;
 737                         perforation += 2;
 738                         scissors += 2;
 739                         i++;
 740                         continue;
 741                 }
 742                 in_perforation = 0;
 743         }
 744
 745         /*
 746          * The mark must be at least 8 bytes long (e.g. "-- >8 --").
 747          * Even though there can be arbitrary cruft on the same line
 748          * (e.g. "cut here"), in order to avoid misidentification, the
 749          * perforation must occupy more than a third of the visible
 750          * width of the line, and dashes and scissors must occupy more
 751          * than half of the perforation.
 752          */
 753
 754         visible = last_nonblank - first_nonblank + 1;
 755         return (scissors && 8 <= visible &&
 756                 visible < perforation * 3 &&
 757                 gap * 2 < perforation);
 758 }
 759
 760 static int handle_commit_msg(struct strbuf *line)
 761 {
 762         static int still_looking = 1;
 763
 764         if (!cmitmsg)
 765                 return 0;
 766
 767         if (still_looking) {
 768                 if (!line->len || (line->len == 1 && line->buf[0] == '\n'))
 769                         return 0;
 770         }
 771
 772         if (use_inbody_headers && still_looking) {
 773                 still_looking = check_header(line, s_hdr_data, 0);
 774                 if (still_looking)
 775                         return 0;
 776         } else
 777                 /* Only trim the first (blank) line of the commit message
 778                  * when ignoring in-body headers.
 779                  */
 780                 still_looking = 0;
 781
 782         /* normalize the log message to UTF-8. */
 783         if (metainfo_charset)
 784                 convert_to_utf8(line, charset.buf);
 785
 786         if (use_scissors && is_scissors_line(line)) {
 787                 int i;
 788                 if (fseek(cmitmsg, 0L, SEEK_SET))
 789                         die_errno("Could not rewind output message file");
 790                 if (ftruncate(fileno(cmitmsg), 0))
 791                         die_errno("Could not truncate output message file at scissors");
 792                 still_looking = 1;
 793
 794                 /*
 795                  * We may have already read "secondary headers"; purge
 796                  * them to give ourselves a clean restart.
 797                  */
 798                 for (i = 0; header[i]; i++) {
 799                         if (s_hdr_data[i])
 800                                 strbuf_release(s_hdr_data[i]);
 801                         s_hdr_data[i] = NULL;
 802                 }
 803                 return 0;
 804         }
 805
 806         if (patchbreak(line)) {
 807                 fclose(cmitmsg);
 808                 cmitmsg = NULL;
 809                 return 1;
 810         }
 811
 812         fputs(line->buf, cmitmsg);
 813         return 0;
 814 }
 815
 816 static void handle_patch(const struct strbuf *line)
 817 {
 818         fwrite(line->buf, 1, line->len, patchfile);
 819         patch_lines++;
 820 }
 821
 822 static void handle_filter(struct strbuf *line)
 823 {
 824         static int filter = 0;
 825
 826         /* filter tells us which part we left off on */
 827         switch (filter) {
 828         case 0:
 829                 if (!handle_commit_msg(line))
 830                         break;
 831                 filter++;
 832         case 1:
 833                 handle_patch(line);
 834                 break;
 835         }
 836 }
 837
 838 static void handle_body(void)
 839 {
 840         struct strbuf prev = STRBUF_INIT;
 841
 842         /* Skip up to the first boundary */
 843         if (*content_top) {
 844                 if (!find_boundary())
 845                         goto handle_body_out;
 846         }
 847
 848         do {
 849                 /* process any boundary lines */
 850                 if (*content_top && is_multipart_boundary(&line)) {
 851                         /* flush any leftover */
 852                         if (prev.len) {
 853                                 handle_filter(&prev);
 854                                 strbuf_reset(&prev);
 855                         }
 856                         if (!handle_boundary())
 857                                 goto handle_body_out;
 858                 }
 859
 860                 /* Unwrap transfer encoding */
 861                 decode_transfer_encoding(&line);
 862
 863                 switch (transfer_encoding) {
 864                 case TE_BASE64:
 865                 case TE_QP:
 866                 {
 867                         struct strbuf **lines, **it, *sb;
 868
 869                         /* Prepend any previous partial lines */
 870                         strbuf_insert(&line, 0, prev.buf, prev.len);
 871                         strbuf_reset(&prev);
 872
 873                         /* binary data most likely doesn't have newlines */
 874                         if (message_type != TYPE_TEXT) {
 875                                 handle_filter(&line);
 876                                 break;
 877                         }
 878                         /*
 879                          * This is a decoded line that may contain
 880                          * multiple new lines.  Pass only one chunk
 881                          * at a time to handle_filter()
 882                          */
 883                         lines = strbuf_split(&line, '\n');
 884                         for (it = lines; (sb = *it); it++) {
 885                                 if (*(it + 1) == NULL) /* The last line */
 886                                         if (sb->buf[sb->len - 1] != '\n') {
 887                                                 /* Partial line, save it for later. */
 888                                                 strbuf_addbuf(&prev, sb);
 889                                                 break;
 890                                         }
 891                                 handle_filter(sb);
 892                         }
 893                         /*
 894                          * The partial chunk is saved in "prev" and will be
 895                          * appended by the next iteration of read_line_with_nul().
 896                          */
 897                         strbuf_list_free(lines);
 898                         break;
 899                 }
 900                 default:
 901                         handle_filter(&line);
 902                 }
 903
 904         } while (!strbuf_getwholeline(&line, fin, '\n'));
 905
 906 handle_body_out:
 907         strbuf_release(&prev);
 908 }
 909
 910 static void output_header_lines(FILE *fout, const char *hdr, const struct strbuf *data)
 911 {
 912         const char *sp = data->buf;
 913         while (1) {
 914                 char *ep = strchr(sp, '\n');
 915                 int len;
 916                 if (!ep)
 917                         len = strlen(sp);
 918                 else
 919                         len = ep - sp;
 920                 fprintf(fout, "%s: %.*s\n", hdr, len, sp);
 921                 if (!ep)
 922                         break;
 923                 sp = ep + 1;
 924         }
 925 }
 926
 927 static void handle_info(void)
 928 {
 929         struct strbuf *hdr;
 930         int i;
 931
 932         for (i = 0; header[i]; i++) {
 933                 /* only print inbody headers if we output a patch file */
 934                 if (patch_lines && s_hdr_data[i])
 935                         hdr = s_hdr_data[i];
 936                 else if (p_hdr_data[i])
 937                         hdr = p_hdr_data[i];
 938                 else
 939                         continue;
 940
 941                 if (!memcmp(header[i], "Subject", 7)) {
 942                         if (!keep_subject) {
 943                                 cleanup_subject(hdr);
 944                                 cleanup_space(hdr);
 945                         }
 946                         output_header_lines(fout, "Subject", hdr);
 947                 } else if (!memcmp(header[i], "From", 4)) {
 948                         cleanup_space(hdr);
 949                         handle_from(hdr);
 950                         fprintf(fout, "Author: %s\n", name.buf);
 951                         fprintf(fout, "Email: %s\n", email.buf);
 952                 } else {
 953                         cleanup_space(hdr);
 954                         fprintf(fout, "%s: %s\n", header[i], hdr->buf);
 955                 }
 956         }
 957         fprintf(fout, "\n");
 958 }
 959
 960 static int mailinfo(FILE *in, FILE *out, const char *msg, const char *patch)
 961 {
 962         int peek;
 963         fin = in;
 964         fout = out;
 965
 966         cmitmsg = fopen(msg, "w");
 967         if (!cmitmsg) {
 968                 perror(msg);
 969                 return -1;
 970         }
 971         patchfile = fopen(patch, "w");
 972         if (!patchfile) {
 973                 perror(patch);
 974                 fclose(cmitmsg);
 975                 return -1;
 976         }
 977
 978         p_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*p_hdr_data));
 979         s_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*s_hdr_data));
 980
 981         do {
 982                 peek = fgetc(in);
 983         } while (isspace(peek));
 984         ungetc(peek, in);
 985
 986         /* process the email header */
 987         while (read_one_header_line(&line, fin))
 988                 check_header(&line, p_hdr_data, 1);
 989
 990         handle_body();
 991         handle_info();
 992
 993         return 0;
 994 }
 995
 996 static int git_mailinfo_config(const char *var, const char *value, void *unused)
 997 {
 998         if (prefixcmp(var, "mailinfo."))
 999                 return git_default_config(var, value, unused);
1000         if (!strcmp(var, "mailinfo.scissors")) {
1001                 use_scissors = git_config_bool(var, value);
1002                 return 0;
1003         }
1004         /* perhaps others here */
1005         return 0;
1006 }
1007
1008 static const char mailinfo_usage[] =
1009         "git mailinfo [-k|-b] [-u | --encoding=<encoding> | -n] [--scissors | --no-scissors] msg patch < mail >info";
1010
1011 int cmd_mailinfo(int argc, const char **argv, const char *prefix)
1012 {
1013         const char *def_charset;
1014
1015         /* NEEDSWORK: might want to do the optional .git/ directory
1016          * discovery
1017          */
1018         git_config(git_mailinfo_config, NULL);
1019
1020         def_charset = get_commit_output_encoding();
1021         metainfo_charset = def_charset;
1022
1023         while (1 < argc && argv[1][0] == '-') {
1024                 if (!strcmp(argv[1], "-k"))
1025                         keep_subject = 1;
1026                 else if (!strcmp(argv[1], "-b"))
1027                         keep_non_patch_brackets_in_subject = 1;
1028                 else if (!strcmp(argv[1], "-u"))
1029                         metainfo_charset = def_charset;
1030                 else if (!strcmp(argv[1], "-n"))
1031                         metainfo_charset = NULL;
1032                 else if (!prefixcmp(argv[1], "--encoding="))
1033                         metainfo_charset = argv[1] + 11;
1034                 else if (!strcmp(argv[1], "--scissors"))
1035                         use_scissors = 1;
1036                 else if (!strcmp(argv[1], "--no-scissors"))
1037                         use_scissors = 0;
1038                 else if (!strcmp(argv[1], "--no-inbody-headers"))
1039                         use_inbody_headers = 0;
1040                 else
1041                         usage(mailinfo_usage);
1042                 argc--; argv++;
1043         }
1044
1045         if (argc != 3)
1046                 usage(mailinfo_usage);
1047
1048         return !!mailinfo(stdin, stdout, argv[1], argv[2]);
1049 }