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         strbuf_setlen(attr, 0);
 164         if (!ap)
 165                 return 0;
 166         ap += strlen(name);
 167         if (*ap == '"') {
 168                 ap++;
 169                 ends = "\"";
 170         }
 171         else
 172                 ends = "; \t";
 173         sz = strcspn(ap, ends);
 174         strbuf_add(attr, ap, sz);
 175         return 1;
 176 }
 177
 178 static struct strbuf *content[MAX_BOUNDARIES];
 179
 180 static struct strbuf **content_top = content;
 181
 182 static void handle_content_type(struct strbuf *line)
 183 {
 184         struct strbuf *boundary = xmalloc(sizeof(struct strbuf));
 185         strbuf_init(boundary, line->len);
 186
 187         if (!strcasestr(line->buf, "text/"))
 188                  message_type = TYPE_OTHER;
 189         if (slurp_attr(line->buf, "boundary=", boundary)) {
 190                 strbuf_insert(boundary, 0, "--", 2);
 191                 if (++content_top > &content[MAX_BOUNDARIES]) {
 192                         fprintf(stderr, "Too many boundaries to handle\n");
 193                         exit(1);
 194                 }
 195                 *content_top = boundary;
 196                 boundary = NULL;
 197         }
 198         slurp_attr(line->buf, "charset=", &charset);
 199
 200         if (boundary) {
 201                 strbuf_release(boundary);
 202                 free(boundary);
 203         }
 204 }
 205
 206 static void handle_content_transfer_encoding(const struct strbuf *line)
 207 {
 208         if (strcasestr(line->buf, "base64"))
 209                 transfer_encoding = TE_BASE64;
 210         else if (strcasestr(line->buf, "quoted-printable"))
 211                 transfer_encoding = TE_QP;
 212         else
 213                 transfer_encoding = TE_DONTCARE;
 214 }
 215
 216 static int is_multipart_boundary(const struct strbuf *line)
 217 {
 218         return (((*content_top)->len <= line->len) &&
 219                 !memcmp(line->buf, (*content_top)->buf, (*content_top)->len));
 220 }
 221
 222 static void cleanup_subject(struct strbuf *subject)
 223 {
 224         size_t at = 0;
 225
 226         while (at < subject->len) {
 227                 char *pos;
 228                 size_t remove;
 229
 230                 switch (subject->buf[at]) {
 231                 case 'r': case 'R':
 232                         if (subject->len <= at + 3)
 233                                 break;
 234                         if (!memcmp(subject->buf + at + 1, "e:", 2)) {
 235                                 strbuf_remove(subject, at, 3);
 236                                 continue;
 237                         }
 238                         at++;
 239                         break;
 240                 case ' ': case '\t': case ':':
 241                         strbuf_remove(subject, at, 1);
 242                         continue;
 243                 case '[':
 244                         pos = strchr(subject->buf + at, ']');
 245                         if (!pos)
 246                                 break;
 247                         remove = pos - subject->buf + at + 1;
 248                         if (!keep_non_patch_brackets_in_subject ||
 249                             (7 <= remove &&
 250                              memmem(subject->buf + at, remove, "PATCH", 5)))
 251                                 strbuf_remove(subject, at, remove);
 252                         else {
 253                                 at += remove;
 254                                 /*
 255                                  * If the input had a space after the ], keep
 256                                  * it.  We don't bother with finding the end of
 257                                  * the space, since we later normalize it
 258                                  * anyway.
 259                                  */
 260                                 if (isspace(subject->buf[at]))
 261                                         at += 1;
 262                         }
 263                         continue;
 264                 }
 265                 break;
 266         }
 267         strbuf_trim(subject);
 268 }
 269
 270 static void cleanup_space(struct strbuf *sb)
 271 {
 272         size_t pos, cnt;
 273         for (pos = 0; pos < sb->len; pos++) {
 274                 if (isspace(sb->buf[pos])) {
 275                         sb->buf[pos] = ' ';
 276                         for (cnt = 0; isspace(sb->buf[pos + cnt + 1]); cnt++);
 277                         strbuf_remove(sb, pos + 1, cnt);
 278                 }
 279         }
 280 }
 281
 282 static void decode_header(struct strbuf *line);
 283 static const char *header[MAX_HDR_PARSED] = {
 284         "From","Subject","Date",
 285 };
 286
 287 static inline int cmp_header(const struct strbuf *line, const char *hdr)
 288 {
 289         int len = strlen(hdr);
 290         return !strncasecmp(line->buf, hdr, len) && line->len > len &&
 291                         line->buf[len] == ':' && isspace(line->buf[len + 1]);
 292 }
 293
 294 static int check_header(const struct strbuf *line,
 295                                 struct strbuf *hdr_data[], int overwrite)
 296 {
 297         int i, ret = 0, len;
 298         struct strbuf sb = STRBUF_INIT;
 299         /* search for the interesting parts */
 300         for (i = 0; header[i]; i++) {
 301                 int len = strlen(header[i]);
 302                 if ((!hdr_data[i] || overwrite) && cmp_header(line, header[i])) {
 303                         /* Unwrap inline B and Q encoding, and optionally
 304                          * normalize the meta information to utf8.
 305                          */
 306                         strbuf_add(&sb, line->buf + len + 2, line->len - len - 2);
 307                         decode_header(&sb);
 308                         handle_header(&hdr_data[i], &sb);
 309                         ret = 1;
 310                         goto check_header_out;
 311                 }
 312         }
 313
 314         /* Content stuff */
 315         if (cmp_header(line, "Content-Type")) {
 316                 len = strlen("Content-Type: ");
 317                 strbuf_add(&sb, line->buf + len, line->len - len);
 318                 decode_header(&sb);
 319                 strbuf_insert(&sb, 0, "Content-Type: ", len);
 320                 handle_content_type(&sb);
 321                 ret = 1;
 322                 goto check_header_out;
 323         }
 324         if (cmp_header(line, "Content-Transfer-Encoding")) {
 325                 len = strlen("Content-Transfer-Encoding: ");
 326                 strbuf_add(&sb, line->buf + len, line->len - len);
 327                 decode_header(&sb);
 328                 handle_content_transfer_encoding(&sb);
 329                 ret = 1;
 330                 goto check_header_out;
 331         }
 332
 333         /* for inbody stuff */
 334         if (!prefixcmp(line->buf, ">From") && isspace(line->buf[5])) {
 335                 ret = 1; /* Should this return 0? */
 336                 goto check_header_out;
 337         }
 338         if (!prefixcmp(line->buf, "[PATCH]") && isspace(line->buf[7])) {
 339                 for (i = 0; header[i]; i++) {
 340                         if (!memcmp("Subject", header[i], 7)) {
 341                                 handle_header(&hdr_data[i], line);
 342                                 ret = 1;
 343                                 goto check_header_out;
 344                         }
 345                 }
 346         }
 347
 348 check_header_out:
 349         strbuf_release(&sb);
 350         return ret;
 351 }
 352
 353 static int is_rfc2822_header(const struct strbuf *line)
 354 {
 355         /*
 356          * The section that defines the loosest possible
 357          * field name is "3.6.8 Optional fields".
 358          *
 359          * optional-field = field-name ":" unstructured CRLF
 360          * field-name = 1*ftext
 361          * ftext = %d33-57 / %59-126
 362          */
 363         int ch;
 364         char *cp = line->buf;
 365
 366         /* Count mbox From headers as headers */
 367         if (!prefixcmp(cp, "From ") || !prefixcmp(cp, ">From "))
 368                 return 1;
 369
 370         while ((ch = *cp++)) {
 371                 if (ch == ':')
 372                         return 1;
 373                 if ((33 <= ch && ch <= 57) ||
 374                     (59 <= ch && ch <= 126))
 375                         continue;
 376                 break;
 377         }
 378         return 0;
 379 }
 380
 381 static int read_one_header_line(struct strbuf *line, FILE *in)
 382 {
 383         /* Get the first part of the line. */
 384         if (strbuf_getline(line, in, '\n'))
 385                 return 0;
 386
 387         /*
 388          * Is it an empty line or not a valid rfc2822 header?
 389          * If so, stop here, and return false ("not a header")
 390          */
 391         strbuf_rtrim(line);
 392         if (!line->len || !is_rfc2822_header(line)) {
 393                 /* Re-add the newline */
 394                 strbuf_addch(line, '\n');
 395                 return 0;
 396         }
 397
 398         /*
 399          * Now we need to eat all the continuation lines..
 400          * Yuck, 2822 header "folding"
 401          */
 402         for (;;) {
 403                 int peek;
 404                 struct strbuf continuation = STRBUF_INIT;
 405
 406                 peek = fgetc(in); ungetc(peek, in);
 407                 if (peek != ' ' && peek != '\t')
 408                         break;
 409                 if (strbuf_getline(&continuation, in, '\n'))
 410                         break;
 411                 continuation.buf[0] = ' ';
 412                 strbuf_rtrim(&continuation);
 413                 strbuf_addbuf(line, &continuation);
 414         }
 415
 416         return 1;
 417 }
 418
 419 static struct strbuf *decode_q_segment(const struct strbuf *q_seg, int rfc2047)
 420 {
 421         const char *in = q_seg->buf;
 422         int c;
 423         struct strbuf *out = xmalloc(sizeof(struct strbuf));
 424         strbuf_init(out, q_seg->len);
 425
 426         while ((c = *in++) != 0) {
 427                 if (c == '=') {
 428                         int d = *in++;
 429                         if (d == '\n' || !d)
 430                                 break; /* drop trailing newline */
 431                         strbuf_addch(out, (hexval(d) << 4) | hexval(*in++));
 432                         continue;
 433                 }
 434                 if (rfc2047 && c == '_') /* rfc2047 4.2 (2) */
 435                         c = 0x20;
 436                 strbuf_addch(out, c);
 437         }
 438         return out;
 439 }
 440
 441 static struct strbuf *decode_b_segment(const struct strbuf *b_seg)
 442 {
 443         /* Decode in..ep, possibly in-place to ot */
 444         int c, pos = 0, acc = 0;
 445         const char *in = b_seg->buf;
 446         struct strbuf *out = xmalloc(sizeof(struct strbuf));
 447         strbuf_init(out, b_seg->len);
 448
 449         while ((c = *in++) != 0) {
 450                 if (c == '+')
 451                         c = 62;
 452                 else if (c == '/')
 453                         c = 63;
 454                 else if ('A' <= c && c <= 'Z')
 455                         c -= 'A';
 456                 else if ('a' <= c && c <= 'z')
 457                         c -= 'a' - 26;
 458                 else if ('0' <= c && c <= '9')
 459                         c -= '0' - 52;
 460                 else
 461                         continue; /* garbage */
 462                 switch (pos++) {
 463                 case 0:
 464                         acc = (c << 2);
 465                         break;
 466                 case 1:
 467                         strbuf_addch(out, (acc | (c >> 4)));
 468                         acc = (c & 15) << 4;
 469                         break;
 470                 case 2:
 471                         strbuf_addch(out, (acc | (c >> 2)));
 472                         acc = (c & 3) << 6;
 473                         break;
 474                 case 3:
 475                         strbuf_addch(out, (acc | c));
 476                         acc = pos = 0;
 477                         break;
 478                 }
 479         }
 480         return out;
 481 }
 482
 483 /*
 484  * When there is no known charset, guess.
 485  *
 486  * Right now we assume that if the target is UTF-8 (the default),
 487  * and it already looks like UTF-8 (which includes US-ASCII as its
 488  * subset, of course) then that is what it is and there is nothing
 489  * to do.
 490  *
 491  * Otherwise, we default to assuming it is Latin1 for historical
 492  * reasons.
 493  */
 494 static const char *guess_charset(const struct strbuf *line, const char *target_charset)
 495 {
 496         if (is_encoding_utf8(target_charset)) {
 497                 if (is_utf8(line->buf))
 498                         return NULL;
 499         }
 500         return "ISO8859-1";
 501 }
 502
 503 static void convert_to_utf8(struct strbuf *line, const char *charset)
 504 {
 505         char *out;
 506
 507         if (!charset || !*charset) {
 508                 charset = guess_charset(line, metainfo_charset);
 509                 if (!charset)
 510                         return;
 511         }
 512
 513         if (!strcasecmp(metainfo_charset, charset))
 514                 return;
 515         out = reencode_string(line->buf, metainfo_charset, charset);
 516         if (!out)
 517                 die("cannot convert from %s to %s",
 518                     charset, metainfo_charset);
 519         strbuf_attach(line, out, strlen(out), strlen(out));
 520 }
 521
 522 static int decode_header_bq(struct strbuf *it)
 523 {
 524         char *in, *ep, *cp;
 525         struct strbuf outbuf = STRBUF_INIT, *dec;
 526         struct strbuf charset_q = STRBUF_INIT, piecebuf = STRBUF_INIT;
 527         int rfc2047 = 0;
 528
 529         in = it->buf;
 530         while (in - it->buf <= it->len && (ep = strstr(in, "=?")) != NULL) {
 531                 int encoding;
 532                 strbuf_reset(&charset_q);
 533                 strbuf_reset(&piecebuf);
 534                 rfc2047 = 1;
 535
 536                 if (in != ep) {
 537                         /*
 538                          * We are about to process an encoded-word
 539                          * that begins at ep, but there is something
 540                          * before the encoded word.
 541                          */
 542                         char *scan;
 543                         for (scan = in; scan < ep; scan++)
 544                                 if (!isspace(*scan))
 545                                         break;
 546
 547                         if (scan != ep || in == it->buf) {
 548                                 /*
 549                                  * We should not lose that "something",
 550                                  * unless we have just processed an
 551                                  * encoded-word, and there is only LWS
 552                                  * before the one we are about to process.
 553                                  */
 554                                 strbuf_add(&outbuf, in, ep - in);
 555                         }
 556                 }
 557                 /* E.g.
 558                  * ep : "=?iso-2022-jp?B?GyR...?= foo"
 559                  * ep : "=?ISO-8859-1?Q?Foo=FCbar?= baz"
 560                  */
 561                 ep += 2;
 562
 563                 if (ep - it->buf >= it->len || !(cp = strchr(ep, '?')))
 564                         goto decode_header_bq_out;
 565
 566                 if (cp + 3 - it->buf > it->len)
 567                         goto decode_header_bq_out;
 568                 strbuf_add(&charset_q, ep, cp - ep);
 569
 570                 encoding = cp[1];
 571                 if (!encoding || cp[2] != '?')
 572                         goto decode_header_bq_out;
 573                 ep = strstr(cp + 3, "?=");
 574                 if (!ep)
 575                         goto decode_header_bq_out;
 576                 strbuf_add(&piecebuf, cp + 3, ep - cp - 3);
 577                 switch (tolower(encoding)) {
 578                 default:
 579                         goto decode_header_bq_out;
 580                 case 'b':
 581                         dec = decode_b_segment(&piecebuf);
 582                         break;
 583                 case 'q':
 584                         dec = decode_q_segment(&piecebuf, 1);
 585                         break;
 586                 }
 587                 if (metainfo_charset)
 588                         convert_to_utf8(dec, charset_q.buf);
 589
 590                 strbuf_addbuf(&outbuf, dec);
 591                 strbuf_release(dec);
 592                 free(dec);
 593                 in = ep + 2;
 594         }
 595         strbuf_addstr(&outbuf, in);
 596         strbuf_reset(it);
 597         strbuf_addbuf(it, &outbuf);
 598 decode_header_bq_out:
 599         strbuf_release(&outbuf);
 600         strbuf_release(&charset_q);
 601         strbuf_release(&piecebuf);
 602         return rfc2047;
 603 }
 604
 605 static void decode_header(struct strbuf *it)
 606 {
 607         if (decode_header_bq(it))
 608                 return;
 609         /* otherwise "it" is a straight copy of the input.
 610          * This can be binary guck but there is no charset specified.
 611          */
 612         if (metainfo_charset)
 613                 convert_to_utf8(it, "");
 614 }
 615
 616 static void decode_transfer_encoding(struct strbuf *line)
 617 {
 618         struct strbuf *ret;
 619
 620         switch (transfer_encoding) {
 621         case TE_QP:
 622                 ret = decode_q_segment(line, 0);
 623                 break;
 624         case TE_BASE64:
 625                 ret = decode_b_segment(line);
 626                 break;
 627         case TE_DONTCARE:
 628         default:
 629                 return;
 630         }
 631         strbuf_reset(line);
 632         strbuf_addbuf(line, ret);
 633         strbuf_release(ret);
 634         free(ret);
 635 }
 636
 637 static void handle_filter(struct strbuf *line);
 638
 639 static int find_boundary(void)
 640 {
 641         while (!strbuf_getline(&line, fin, '\n')) {
 642                 if (*content_top && is_multipart_boundary(&line))
 643                         return 1;
 644         }
 645         return 0;
 646 }
 647
 648 static int handle_boundary(void)
 649 {
 650         struct strbuf newline = STRBUF_INIT;
 651
 652         strbuf_addch(&newline, '\n');
 653 again:
 654         if (line.len >= (*content_top)->len + 2 &&
 655             !memcmp(line.buf + (*content_top)->len, "--", 2)) {
 656                 /* we hit an end boundary */
 657                 /* pop the current boundary off the stack */
 658                 strbuf_release(*content_top);
 659                 free(*content_top);
 660                 *content_top = NULL;
 661
 662                 /* technically won't happen as is_multipart_boundary()
 663                    will fail first.  But just in case..
 664                  */
 665                 if (--content_top < content) {
 666                         fprintf(stderr, "Detected mismatched boundaries, "
 667                                         "can't recover\n");
 668                         exit(1);
 669                 }
 670                 handle_filter(&newline);
 671                 strbuf_release(&newline);
 672
 673                 /* skip to the next boundary */
 674                 if (!find_boundary())
 675                         return 0;
 676                 goto again;
 677         }
 678
 679         /* set some defaults */
 680         transfer_encoding = TE_DONTCARE;
 681         strbuf_reset(&charset);
 682         message_type = TYPE_TEXT;
 683
 684         /* slurp in this section's info */
 685         while (read_one_header_line(&line, fin))
 686                 check_header(&line, p_hdr_data, 0);
 687
 688         strbuf_release(&newline);
 689         /* replenish line */
 690         if (strbuf_getline(&line, fin, '\n'))
 691                 return 0;
 692         strbuf_addch(&line, '\n');
 693         return 1;
 694 }
 695
 696 static inline int patchbreak(const struct strbuf *line)
 697 {
 698         size_t i;
 699
 700         /* Beginning of a "diff -" header? */
 701         if (!prefixcmp(line->buf, "diff -"))
 702                 return 1;
 703
 704         /* CVS "Index: " line? */
 705         if (!prefixcmp(line->buf, "Index: "))
 706                 return 1;
 707
 708         /*
 709          * "--- <filename>" starts patches without headers
 710          * "---<sp>*" is a manual separator
 711          */
 712         if (line->len < 4)
 713                 return 0;
 714
 715         if (!prefixcmp(line->buf, "---")) {
 716                 /* space followed by a filename? */
 717                 if (line->buf[3] == ' ' && !isspace(line->buf[4]))
 718                         return 1;
 719                 /* Just whitespace? */
 720                 for (i = 3; i < line->len; i++) {
 721                         unsigned char c = line->buf[i];
 722                         if (c == '\n')
 723                                 return 1;
 724                         if (!isspace(c))
 725                                 break;
 726                 }
 727                 return 0;
 728         }
 729         return 0;
 730 }
 731
 732 static int is_scissors_line(const struct strbuf *line)
 733 {
 734         size_t i, len = line->len;
 735         int scissors = 0, gap = 0;
 736         int first_nonblank = -1;
 737         int last_nonblank = 0, visible, perforation = 0, in_perforation = 0;
 738         const char *buf = line->buf;
 739
 740         for (i = 0; i < len; i++) {
 741                 if (isspace(buf[i])) {
 742                         if (in_perforation) {
 743                                 perforation++;
 744                                 gap++;
 745                         }
 746                         continue;
 747                 }
 748                 last_nonblank = i;
 749                 if (first_nonblank < 0)
 750                         first_nonblank = i;
 751                 if (buf[i] == '-') {
 752                         in_perforation = 1;
 753                         perforation++;
 754                         continue;
 755                 }
 756                 if (i + 1 < len &&
 757                     (!memcmp(buf + i, ">8", 2) || !memcmp(buf + i, "8<", 2) ||
 758                      !memcmp(buf + i, ">%", 2) || !memcmp(buf + i, "%<", 2))) {
 759                         in_perforation = 1;
 760                         perforation += 2;
 761                         scissors += 2;
 762                         i++;
 763                         continue;
 764                 }
 765                 in_perforation = 0;
 766         }
 767
 768         /*
 769          * The mark must be at least 8 bytes long (e.g. "-- >8 --").
 770          * Even though there can be arbitrary cruft on the same line
 771          * (e.g. "cut here"), in order to avoid misidentification, the
 772          * perforation must occupy more than a third of the visible
 773          * width of the line, and dashes and scissors must occupy more
 774          * than half of the perforation.
 775          */
 776
 777         visible = last_nonblank - first_nonblank + 1;
 778         return (scissors && 8 <= visible &&
 779                 visible < perforation * 3 &&
 780                 gap * 2 < perforation);
 781 }
 782
 783 static int handle_commit_msg(struct strbuf *line)
 784 {
 785         static int still_looking = 1;
 786
 787         if (!cmitmsg)
 788                 return 0;
 789
 790         if (still_looking) {
 791                 if (!line->len || (line->len == 1 && line->buf[0] == '\n'))
 792                         return 0;
 793         }
 794
 795         if (use_inbody_headers && still_looking) {
 796                 still_looking = check_header(line, s_hdr_data, 0);
 797                 if (still_looking)
 798                         return 0;
 799         } else
 800                 /* Only trim the first (blank) line of the commit message
 801                  * when ignoring in-body headers.
 802                  */
 803                 still_looking = 0;
 804
 805         /* normalize the log message to UTF-8. */
 806         if (metainfo_charset)
 807                 convert_to_utf8(line, charset.buf);
 808
 809         if (use_scissors && is_scissors_line(line)) {
 810                 int i;
 811                 if (fseek(cmitmsg, 0L, SEEK_SET))
 812                         die_errno("Could not rewind output message file");
 813                 if (ftruncate(fileno(cmitmsg), 0))
 814                         die_errno("Could not truncate output message file at scissors");
 815                 still_looking = 1;
 816
 817                 /*
 818                  * We may have already read "secondary headers"; purge
 819                  * them to give ourselves a clean restart.
 820                  */
 821                 for (i = 0; header[i]; i++) {
 822                         if (s_hdr_data[i])
 823                                 strbuf_release(s_hdr_data[i]);
 824                         s_hdr_data[i] = NULL;
 825                 }
 826                 return 0;
 827         }
 828
 829         if (patchbreak(line)) {
 830                 fclose(cmitmsg);
 831                 cmitmsg = NULL;
 832                 return 1;
 833         }
 834
 835         fputs(line->buf, cmitmsg);
 836         return 0;
 837 }
 838
 839 static void handle_patch(const struct strbuf *line)
 840 {
 841         fwrite(line->buf, 1, line->len, patchfile);
 842         patch_lines++;
 843 }
 844
 845 static void handle_filter(struct strbuf *line)
 846 {
 847         static int filter = 0;
 848
 849         /* filter tells us which part we left off on */
 850         switch (filter) {
 851         case 0:
 852                 if (!handle_commit_msg(line))
 853                         break;
 854                 filter++;
 855         case 1:
 856                 handle_patch(line);
 857                 break;
 858         }
 859 }
 860
 861 static void handle_body(void)
 862 {
 863         struct strbuf prev = STRBUF_INIT;
 864
 865         /* Skip up to the first boundary */
 866         if (*content_top) {
 867                 if (!find_boundary())
 868                         goto handle_body_out;
 869         }
 870
 871         do {
 872                 /* process any boundary lines */
 873                 if (*content_top && is_multipart_boundary(&line)) {
 874                         /* flush any leftover */
 875                         if (prev.len) {
 876                                 handle_filter(&prev);
 877                                 strbuf_reset(&prev);
 878                         }
 879                         if (!handle_boundary())
 880                                 goto handle_body_out;
 881                 }
 882
 883                 /* Unwrap transfer encoding */
 884                 decode_transfer_encoding(&line);
 885
 886                 switch (transfer_encoding) {
 887                 case TE_BASE64:
 888                 case TE_QP:
 889                 {
 890                         struct strbuf **lines, **it, *sb;
 891
 892                         /* Prepend any previous partial lines */
 893                         strbuf_insert(&line, 0, prev.buf, prev.len);
 894                         strbuf_reset(&prev);
 895
 896                         /* binary data most likely doesn't have newlines */
 897                         if (message_type != TYPE_TEXT) {
 898                                 handle_filter(&line);
 899                                 break;
 900                         }
 901                         /*
 902                          * This is a decoded line that may contain
 903                          * multiple new lines.  Pass only one chunk
 904                          * at a time to handle_filter()
 905                          */
 906                         lines = strbuf_split(&line, '\n');
 907                         for (it = lines; (sb = *it); it++) {
 908                                 if (*(it + 1) == NULL) /* The last line */
 909                                         if (sb->buf[sb->len - 1] != '\n') {
 910                                                 /* Partial line, save it for later. */
 911                                                 strbuf_addbuf(&prev, sb);
 912                                                 break;
 913                                         }
 914                                 handle_filter(sb);
 915                         }
 916                         /*
 917                          * The partial chunk is saved in "prev" and will be
 918                          * appended by the next iteration of read_line_with_nul().
 919                          */
 920                         strbuf_list_free(lines);
 921                         break;
 922                 }
 923                 default:
 924                         handle_filter(&line);
 925                 }
 926
 927         } while (!strbuf_getwholeline(&line, fin, '\n'));
 928
 929 handle_body_out:
 930         strbuf_release(&prev);
 931 }
 932
 933 static void output_header_lines(FILE *fout, const char *hdr, const struct strbuf *data)
 934 {
 935         const char *sp = data->buf;
 936         while (1) {
 937                 char *ep = strchr(sp, '\n');
 938                 int len;
 939                 if (!ep)
 940                         len = strlen(sp);
 941                 else
 942                         len = ep - sp;
 943                 fprintf(fout, "%s: %.*s\n", hdr, len, sp);
 944                 if (!ep)
 945                         break;
 946                 sp = ep + 1;
 947         }
 948 }
 949
 950 static void handle_info(void)
 951 {
 952         struct strbuf *hdr;
 953         int i;
 954
 955         for (i = 0; header[i]; i++) {
 956                 /* only print inbody headers if we output a patch file */
 957                 if (patch_lines && s_hdr_data[i])
 958                         hdr = s_hdr_data[i];
 959                 else if (p_hdr_data[i])
 960                         hdr = p_hdr_data[i];
 961                 else
 962                         continue;
 963
 964                 if (!memcmp(header[i], "Subject", 7)) {
 965                         if (!keep_subject) {
 966                                 cleanup_subject(hdr);
 967                                 cleanup_space(hdr);
 968                         }
 969                         output_header_lines(fout, "Subject", hdr);
 970                 } else if (!memcmp(header[i], "From", 4)) {
 971                         cleanup_space(hdr);
 972                         handle_from(hdr);
 973                         fprintf(fout, "Author: %s\n", name.buf);
 974                         fprintf(fout, "Email: %s\n", email.buf);
 975                 } else {
 976                         cleanup_space(hdr);
 977                         fprintf(fout, "%s: %s\n", header[i], hdr->buf);
 978                 }
 979         }
 980         fprintf(fout, "\n");
 981 }
 982
 983 static int mailinfo(FILE *in, FILE *out, const char *msg, const char *patch)
 984 {
 985         int peek;
 986         fin = in;
 987         fout = out;
 988
 989         cmitmsg = fopen(msg, "w");
 990         if (!cmitmsg) {
 991                 perror(msg);
 992                 return -1;
 993         }
 994         patchfile = fopen(patch, "w");
 995         if (!patchfile) {
 996                 perror(patch);
 997                 fclose(cmitmsg);
 998                 return -1;
 999         }
1000
1001         p_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*p_hdr_data));
1002         s_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*s_hdr_data));
1003
1004         do {
1005                 peek = fgetc(in);
1006         } while (isspace(peek));
1007         ungetc(peek, in);
1008
1009         /* process the email header */
1010         while (read_one_header_line(&line, fin))
1011                 check_header(&line, p_hdr_data, 1);
1012
1013         handle_body();
1014         handle_info();
1015
1016         return 0;
1017 }
1018
1019 static int git_mailinfo_config(const char *var, const char *value, void *unused)
1020 {
1021         if (prefixcmp(var, "mailinfo."))
1022                 return git_default_config(var, value, unused);
1023         if (!strcmp(var, "mailinfo.scissors")) {
1024                 use_scissors = git_config_bool(var, value);
1025                 return 0;
1026         }
1027         /* perhaps others here */
1028         return 0;
1029 }
1030
1031 static const char mailinfo_usage[] =
1032         "git mailinfo [-k|-b] [-u | --encoding=<encoding> | -n] [--scissors | --no-scissors] msg patch < mail >info";
1033
1034 int cmd_mailinfo(int argc, const char **argv, const char *prefix)
1035 {
1036         const char *def_charset;
1037
1038         /* NEEDSWORK: might want to do the optional .git/ directory
1039          * discovery
1040          */
1041         git_config(git_mailinfo_config, NULL);
1042
1043         def_charset = get_commit_output_encoding();
1044         metainfo_charset = def_charset;
1045
1046         while (1 < argc && argv[1][0] == '-') {
1047                 if (!strcmp(argv[1], "-k"))
1048                         keep_subject = 1;
1049                 else if (!strcmp(argv[1], "-b"))
1050                         keep_non_patch_brackets_in_subject = 1;
1051                 else if (!strcmp(argv[1], "-u"))
1052                         metainfo_charset = def_charset;
1053                 else if (!strcmp(argv[1], "-n"))
1054                         metainfo_charset = NULL;
1055                 else if (!prefixcmp(argv[1], "--encoding="))
1056                         metainfo_charset = argv[1] + 11;
1057                 else if (!strcmp(argv[1], "--scissors"))
1058                         use_scissors = 1;
1059                 else if (!strcmp(argv[1], "--no-scissors"))
1060                         use_scissors = 0;
1061                 else if (!strcmp(argv[1], "--no-inbody-headers"))
1062                         use_inbody_headers = 0;
1063                 else
1064                         usage(mailinfo_usage);
1065                 argc--; argv++;
1066         }
1067
1068         if (argc != 3)
1069                 usage(mailinfo_usage);
1070
1071         return !!mailinfo(stdin, stdout, argv[1], argv[2]);
1072 }