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 /*
 485  * When there is no known charset, guess.
 486  *
 487  * Right now we assume that if the target is UTF-8 (the default),
 488  * and it already looks like UTF-8 (which includes US-ASCII as its
 489  * subset, of course) then that is what it is and there is nothing
 490  * to do.
 491  *
 492  * Otherwise, we default to assuming it is Latin1 for historical
 493  * reasons.
 494  */
 495 static const char *guess_charset(const struct strbuf *line, const char *target_charset)
 496 {
 497         if (is_encoding_utf8(target_charset)) {
 498                 if (is_utf8(line->buf))
 499                         return NULL;
 500         }
 501         return "ISO8859-1";
 502 }
 503
 504 static void convert_to_utf8(struct strbuf *line, const char *charset)
 505 {
 506         char *out;
 507
 508         if (!charset || !*charset) {
 509                 charset = guess_charset(line, metainfo_charset);
 510                 if (!charset)
 511                         return;
 512         }
 513
 514         if (!strcasecmp(metainfo_charset, charset))
 515                 return;
 516         out = reencode_string(line->buf, metainfo_charset, charset);
 517         if (!out)
 518                 die("cannot convert from %s to %s",
 519                     charset, metainfo_charset);
 520         strbuf_attach(line, out, strlen(out), strlen(out));
 521 }
 522
 523 static int decode_header_bq(struct strbuf *it)
 524 {
 525         char *in, *ep, *cp;
 526         struct strbuf outbuf = STRBUF_INIT, *dec;
 527         struct strbuf charset_q = STRBUF_INIT, piecebuf = STRBUF_INIT;
 528         int rfc2047 = 0;
 529
 530         in = it->buf;
 531         while (in - it->buf <= it->len && (ep = strstr(in, "=?")) != NULL) {
 532                 int encoding;
 533                 strbuf_reset(&charset_q);
 534                 strbuf_reset(&piecebuf);
 535                 rfc2047 = 1;
 536
 537                 if (in != ep) {
 538                         /*
 539                          * We are about to process an encoded-word
 540                          * that begins at ep, but there is something
 541                          * before the encoded word.
 542                          */
 543                         char *scan;
 544                         for (scan = in; scan < ep; scan++)
 545                                 if (!isspace(*scan))
 546                                         break;
 547
 548                         if (scan != ep || in == it->buf) {
 549                                 /*
 550                                  * We should not lose that "something",
 551                                  * unless we have just processed an
 552                                  * encoded-word, and there is only LWS
 553                                  * before the one we are about to process.
 554                                  */
 555                                 strbuf_add(&outbuf, in, ep - in);
 556                         }
 557                 }
 558                 /* E.g.
 559                  * ep : "=?iso-2022-jp?B?GyR...?= foo"
 560                  * ep : "=?ISO-8859-1?Q?Foo=FCbar?= baz"
 561                  */
 562                 ep += 2;
 563
 564                 if (ep - it->buf >= it->len || !(cp = strchr(ep, '?')))
 565                         goto decode_header_bq_out;
 566
 567                 if (cp + 3 - it->buf > it->len)
 568                         goto decode_header_bq_out;
 569                 strbuf_add(&charset_q, ep, cp - ep);
 570
 571                 encoding = cp[1];
 572                 if (!encoding || cp[2] != '?')
 573                         goto decode_header_bq_out;
 574                 ep = strstr(cp + 3, "?=");
 575                 if (!ep)
 576                         goto decode_header_bq_out;
 577                 strbuf_add(&piecebuf, cp + 3, ep - cp - 3);
 578                 switch (tolower(encoding)) {
 579                 default:
 580                         goto decode_header_bq_out;
 581                 case 'b':
 582                         dec = decode_b_segment(&piecebuf);
 583                         break;
 584                 case 'q':
 585                         dec = decode_q_segment(&piecebuf, 1);
 586                         break;
 587                 }
 588                 if (metainfo_charset)
 589                         convert_to_utf8(dec, charset_q.buf);
 590
 591                 strbuf_addbuf(&outbuf, dec);
 592                 strbuf_release(dec);
 593                 free(dec);
 594                 in = ep + 2;
 595         }
 596         strbuf_addstr(&outbuf, in);
 597         strbuf_reset(it);
 598         strbuf_addbuf(it, &outbuf);
 599 decode_header_bq_out:
 600         strbuf_release(&outbuf);
 601         strbuf_release(&charset_q);
 602         strbuf_release(&piecebuf);
 603         return rfc2047;
 604 }
 605
 606 static void decode_header(struct strbuf *it)
 607 {
 608         if (decode_header_bq(it))
 609                 return;
 610         /* otherwise "it" is a straight copy of the input.
 611          * This can be binary guck but there is no charset specified.
 612          */
 613         if (metainfo_charset)
 614                 convert_to_utf8(it, "");
 615 }
 616
 617 static void decode_transfer_encoding(struct strbuf *line)
 618 {
 619         struct strbuf *ret;
 620
 621         switch (transfer_encoding) {
 622         case TE_QP:
 623                 ret = decode_q_segment(line, 0);
 624                 break;
 625         case TE_BASE64:
 626                 ret = decode_b_segment(line);
 627                 break;
 628         case TE_DONTCARE:
 629         default:
 630                 return;
 631         }
 632         strbuf_reset(line);
 633         strbuf_addbuf(line, ret);
 634         strbuf_release(ret);
 635         free(ret);
 636 }
 637
 638 static void handle_filter(struct strbuf *line);
 639
 640 static int find_boundary(void)
 641 {
 642         while (!strbuf_getline(&line, fin, '\n')) {
 643                 if (*content_top && is_multipart_boundary(&line))
 644                         return 1;
 645         }
 646         return 0;
 647 }
 648
 649 static int handle_boundary(void)
 650 {
 651         struct strbuf newline = STRBUF_INIT;
 652
 653         strbuf_addch(&newline, '\n');
 654 again:
 655         if (line.len >= (*content_top)->len + 2 &&
 656             !memcmp(line.buf + (*content_top)->len, "--", 2)) {
 657                 /* we hit an end boundary */
 658                 /* pop the current boundary off the stack */
 659                 strbuf_release(*content_top);
 660                 free(*content_top);
 661                 *content_top = NULL;
 662
 663                 /* technically won't happen as is_multipart_boundary()
 664                    will fail first.  But just in case..
 665                  */
 666                 if (--content_top < content) {
 667                         fprintf(stderr, "Detected mismatched boundaries, "
 668                                         "can't recover\n");
 669                         exit(1);
 670                 }
 671                 handle_filter(&newline);
 672                 strbuf_release(&newline);
 673
 674                 /* skip to the next boundary */
 675                 if (!find_boundary())
 676                         return 0;
 677                 goto again;
 678         }
 679
 680         /* set some defaults */
 681         transfer_encoding = TE_DONTCARE;
 682         strbuf_reset(&charset);
 683         message_type = TYPE_TEXT;
 684
 685         /* slurp in this section's info */
 686         while (read_one_header_line(&line, fin))
 687                 check_header(&line, p_hdr_data, 0);
 688
 689         strbuf_release(&newline);
 690         /* replenish line */
 691         if (strbuf_getline(&line, fin, '\n'))
 692                 return 0;
 693         strbuf_addch(&line, '\n');
 694         return 1;
 695 }
 696
 697 static inline int patchbreak(const struct strbuf *line)
 698 {
 699         size_t i;
 700
 701         /* Beginning of a "diff -" header? */
 702         if (!prefixcmp(line->buf, "diff -"))
 703                 return 1;
 704
 705         /* CVS "Index: " line? */
 706         if (!prefixcmp(line->buf, "Index: "))
 707                 return 1;
 708
 709         /*
 710          * "--- <filename>" starts patches without headers
 711          * "---<sp>*" is a manual separator
 712          */
 713         if (line->len < 4)
 714                 return 0;
 715
 716         if (!prefixcmp(line->buf, "---")) {
 717                 /* space followed by a filename? */
 718                 if (line->buf[3] == ' ' && !isspace(line->buf[4]))
 719                         return 1;
 720                 /* Just whitespace? */
 721                 for (i = 3; i < line->len; i++) {
 722                         unsigned char c = line->buf[i];
 723                         if (c == '\n')
 724                                 return 1;
 725                         if (!isspace(c))
 726                                 break;
 727                 }
 728                 return 0;
 729         }
 730         return 0;
 731 }
 732
 733 static int is_scissors_line(const struct strbuf *line)
 734 {
 735         size_t i, len = line->len;
 736         int scissors = 0, gap = 0;
 737         int first_nonblank = -1;
 738         int last_nonblank = 0, visible, perforation = 0, in_perforation = 0;
 739         const char *buf = line->buf;
 740
 741         for (i = 0; i < len; i++) {
 742                 if (isspace(buf[i])) {
 743                         if (in_perforation) {
 744                                 perforation++;
 745                                 gap++;
 746                         }
 747                         continue;
 748                 }
 749                 last_nonblank = i;
 750                 if (first_nonblank < 0)
 751                         first_nonblank = i;
 752                 if (buf[i] == '-') {
 753                         in_perforation = 1;
 754                         perforation++;
 755                         continue;
 756                 }
 757                 if (i + 1 < len &&
 758                     (!memcmp(buf + i, ">8", 2) || !memcmp(buf + i, "8<", 2) ||
 759                      !memcmp(buf + i, ">%", 2) || !memcmp(buf + i, "%<", 2))) {
 760                         in_perforation = 1;
 761                         perforation += 2;
 762                         scissors += 2;
 763                         i++;
 764                         continue;
 765                 }
 766                 in_perforation = 0;
 767         }
 768
 769         /*
 770          * The mark must be at least 8 bytes long (e.g. "-- >8 --").
 771          * Even though there can be arbitrary cruft on the same line
 772          * (e.g. "cut here"), in order to avoid misidentification, the
 773          * perforation must occupy more than a third of the visible
 774          * width of the line, and dashes and scissors must occupy more
 775          * than half of the perforation.
 776          */
 777
 778         visible = last_nonblank - first_nonblank + 1;
 779         return (scissors && 8 <= visible &&
 780                 visible < perforation * 3 &&
 781                 gap * 2 < perforation);
 782 }
 783
 784 static int handle_commit_msg(struct strbuf *line)
 785 {
 786         static int still_looking = 1;
 787
 788         if (!cmitmsg)
 789                 return 0;
 790
 791         if (still_looking) {
 792                 if (!line->len || (line->len == 1 && line->buf[0] == '\n'))
 793                         return 0;
 794         }
 795
 796         if (use_inbody_headers && still_looking) {
 797                 still_looking = check_header(line, s_hdr_data, 0);
 798                 if (still_looking)
 799                         return 0;
 800         } else
 801                 /* Only trim the first (blank) line of the commit message
 802                  * when ignoring in-body headers.
 803                  */
 804                 still_looking = 0;
 805
 806         /* normalize the log message to UTF-8. */
 807         if (metainfo_charset)
 808                 convert_to_utf8(line, charset.buf);
 809
 810         if (use_scissors && is_scissors_line(line)) {
 811                 int i;
 812                 if (fseek(cmitmsg, 0L, SEEK_SET))
 813                         die_errno("Could not rewind output message file");
 814                 if (ftruncate(fileno(cmitmsg), 0))
 815                         die_errno("Could not truncate output message file at scissors");
 816                 still_looking = 1;
 817
 818                 /*
 819                  * We may have already read "secondary headers"; purge
 820                  * them to give ourselves a clean restart.
 821                  */
 822                 for (i = 0; header[i]; i++) {
 823                         if (s_hdr_data[i])
 824                                 strbuf_release(s_hdr_data[i]);
 825                         s_hdr_data[i] = NULL;
 826                 }
 827                 return 0;
 828         }
 829
 830         if (patchbreak(line)) {
 831                 fclose(cmitmsg);
 832                 cmitmsg = NULL;
 833                 return 1;
 834         }
 835
 836         fputs(line->buf, cmitmsg);
 837         return 0;
 838 }
 839
 840 static void handle_patch(const struct strbuf *line)
 841 {
 842         fwrite(line->buf, 1, line->len, patchfile);
 843         patch_lines++;
 844 }
 845
 846 static void handle_filter(struct strbuf *line)
 847 {
 848         static int filter = 0;
 849
 850         /* filter tells us which part we left off on */
 851         switch (filter) {
 852         case 0:
 853                 if (!handle_commit_msg(line))
 854                         break;
 855                 filter++;
 856         case 1:
 857                 handle_patch(line);
 858                 break;
 859         }
 860 }
 861
 862 static void handle_body(void)
 863 {
 864         struct strbuf prev = STRBUF_INIT;
 865
 866         /* Skip up to the first boundary */
 867         if (*content_top) {
 868                 if (!find_boundary())
 869                         goto handle_body_out;
 870         }
 871
 872         do {
 873                 /* process any boundary lines */
 874                 if (*content_top && is_multipart_boundary(&line)) {
 875                         /* flush any leftover */
 876                         if (prev.len) {
 877                                 handle_filter(&prev);
 878                                 strbuf_reset(&prev);
 879                         }
 880                         if (!handle_boundary())
 881                                 goto handle_body_out;
 882                 }
 883
 884                 /* Unwrap transfer encoding */
 885                 decode_transfer_encoding(&line);
 886
 887                 switch (transfer_encoding) {
 888                 case TE_BASE64:
 889                 case TE_QP:
 890                 {
 891                         struct strbuf **lines, **it, *sb;
 892
 893                         /* Prepend any previous partial lines */
 894                         strbuf_insert(&line, 0, prev.buf, prev.len);
 895                         strbuf_reset(&prev);
 896
 897                         /* binary data most likely doesn't have newlines */
 898                         if (message_type != TYPE_TEXT) {
 899                                 handle_filter(&line);
 900                                 break;
 901                         }
 902                         /*
 903                          * This is a decoded line that may contain
 904                          * multiple new lines.  Pass only one chunk
 905                          * at a time to handle_filter()
 906                          */
 907                         lines = strbuf_split(&line, '\n');
 908                         for (it = lines; (sb = *it); it++) {
 909                                 if (*(it + 1) == NULL) /* The last line */
 910                                         if (sb->buf[sb->len - 1] != '\n') {
 911                                                 /* Partial line, save it for later. */
 912                                                 strbuf_addbuf(&prev, sb);
 913                                                 break;
 914                                         }
 915                                 handle_filter(sb);
 916                         }
 917                         /*
 918                          * The partial chunk is saved in "prev" and will be
 919                          * appended by the next iteration of read_line_with_nul().
 920                          */
 921                         strbuf_list_free(lines);
 922                         break;
 923                 }
 924                 default:
 925                         handle_filter(&line);
 926                 }
 927
 928         } while (!strbuf_getwholeline(&line, fin, '\n'));
 929
 930 handle_body_out:
 931         strbuf_release(&prev);
 932 }
 933
 934 static void output_header_lines(FILE *fout, const char *hdr, const struct strbuf *data)
 935 {
 936         const char *sp = data->buf;
 937         while (1) {
 938                 char *ep = strchr(sp, '\n');
 939                 int len;
 940                 if (!ep)
 941                         len = strlen(sp);
 942                 else
 943                         len = ep - sp;
 944                 fprintf(fout, "%s: %.*s\n", hdr, len, sp);
 945                 if (!ep)
 946                         break;
 947                 sp = ep + 1;
 948         }
 949 }
 950
 951 static void handle_info(void)
 952 {
 953         struct strbuf *hdr;
 954         int i;
 955
 956         for (i = 0; header[i]; i++) {
 957                 /* only print inbody headers if we output a patch file */
 958                 if (patch_lines && s_hdr_data[i])
 959                         hdr = s_hdr_data[i];
 960                 else if (p_hdr_data[i])
 961                         hdr = p_hdr_data[i];
 962                 else
 963                         continue;
 964
 965                 if (!memcmp(header[i], "Subject", 7)) {
 966                         if (!keep_subject) {
 967                                 cleanup_subject(hdr);
 968                                 cleanup_space(hdr);
 969                         }
 970                         output_header_lines(fout, "Subject", hdr);
 971                 } else if (!memcmp(header[i], "From", 4)) {
 972                         cleanup_space(hdr);
 973                         handle_from(hdr);
 974                         fprintf(fout, "Author: %s\n", name.buf);
 975                         fprintf(fout, "Email: %s\n", email.buf);
 976                 } else {
 977                         cleanup_space(hdr);
 978                         fprintf(fout, "%s: %s\n", header[i], hdr->buf);
 979                 }
 980         }
 981         fprintf(fout, "\n");
 982 }
 983
 984 static int mailinfo(FILE *in, FILE *out, const char *msg, const char *patch)
 985 {
 986         int peek;
 987         fin = in;
 988         fout = out;
 989
 990         cmitmsg = fopen(msg, "w");
 991         if (!cmitmsg) {
 992                 perror(msg);
 993                 return -1;
 994         }
 995         patchfile = fopen(patch, "w");
 996         if (!patchfile) {
 997                 perror(patch);
 998                 fclose(cmitmsg);
 999                 return -1;
1000         }
1001
1002         p_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*p_hdr_data));
1003         s_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*s_hdr_data));
1004
1005         do {
1006                 peek = fgetc(in);
1007         } while (isspace(peek));
1008         ungetc(peek, in);
1009
1010         /* process the email header */
1011         while (read_one_header_line(&line, fin))
1012                 check_header(&line, p_hdr_data, 1);
1013
1014         handle_body();
1015         handle_info();
1016
1017         return 0;
1018 }
1019
1020 static int git_mailinfo_config(const char *var, const char *value, void *unused)
1021 {
1022         if (prefixcmp(var, "mailinfo."))
1023                 return git_default_config(var, value, unused);
1024         if (!strcmp(var, "mailinfo.scissors")) {
1025                 use_scissors = git_config_bool(var, value);
1026                 return 0;
1027         }
1028         /* perhaps others here */
1029         return 0;
1030 }
1031
1032 static const char mailinfo_usage[] =
1033         "git mailinfo [-k|-b] [-u | --encoding=<encoding> | -n] [--scissors | --no-scissors] msg patch < mail >info";
1034
1035 int cmd_mailinfo(int argc, const char **argv, const char *prefix)
1036 {
1037         const char *def_charset;
1038
1039         /* NEEDSWORK: might want to do the optional .git/ directory
1040          * discovery
1041          */
1042         git_config(git_mailinfo_config, NULL);
1043
1044         def_charset = get_commit_output_encoding();
1045         metainfo_charset = def_charset;
1046
1047         while (1 < argc && argv[1][0] == '-') {
1048                 if (!strcmp(argv[1], "-k"))
1049                         keep_subject = 1;
1050                 else if (!strcmp(argv[1], "-b"))
1051                         keep_non_patch_brackets_in_subject = 1;
1052                 else if (!strcmp(argv[1], "-u"))
1053                         metainfo_charset = def_charset;
1054                 else if (!strcmp(argv[1], "-n"))
1055                         metainfo_charset = NULL;
1056                 else if (!prefixcmp(argv[1], "--encoding="))
1057                         metainfo_charset = argv[1] + 11;
1058                 else if (!strcmp(argv[1], "--scissors"))
1059                         use_scissors = 1;
1060                 else if (!strcmp(argv[1], "--no-scissors"))
1061                         use_scissors = 0;
1062                 else if (!strcmp(argv[1], "--no-inbody-headers"))
1063                         use_inbody_headers = 0;
1064                 else
1065                         usage(mailinfo_usage);
1066                 argc--; argv++;
1067         }
1068
1069         if (argc != 3)
1070                 usage(mailinfo_usage);
1071
1072         return !!mailinfo(stdin, stdout, argv[1], argv[2]);
1073 }