git.oblomov.eu Git - git/blob - builtin/mailinfo.c

   1 /*
   2  * Another stupid program, this one parsing the headers of an
   3  * email to figure out authorship and subject
   4  */
   5 #include "cache.h"
   6 #include "builtin.h"
   7 #include "utf8.h"
   8 #include "strbuf.h"
   9
  10 static FILE *cmitmsg, *patchfile, *fin, *fout;
  11
  12 static int keep_subject;
  13 static int keep_non_patch_brackets_in_subject;
  14 static const char *metainfo_charset;
  15 static struct strbuf line = STRBUF_INIT;
  16 static struct strbuf name = STRBUF_INIT;
  17 static struct strbuf email = STRBUF_INIT;
  18
  19 static enum  {
  20         TE_DONTCARE, TE_QP, TE_BASE64
  21 } transfer_encoding;
  22 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 ((subject->buf[at + 1] == 'e' ||
 235                              subject->buf[at + 1] == 'E') &&
 236                             subject->buf[at + 2] == ':') {
 237                                 strbuf_remove(subject, at, 3);
 238                                 continue;
 239                         }
 240                         at++;
 241                         break;
 242                 case ' ': case '\t': case ':':
 243                         strbuf_remove(subject, at, 1);
 244                         continue;
 245                 case '[':
 246                         pos = strchr(subject->buf + at, ']');
 247                         if (!pos)
 248                                 break;
 249                         remove = pos - subject->buf + at + 1;
 250                         if (!keep_non_patch_brackets_in_subject ||
 251                             (7 <= remove &&
 252                              memmem(subject->buf + at, remove, "PATCH", 5)))
 253                                 strbuf_remove(subject, at, remove);
 254                         else {
 255                                 at += remove;
 256                                 /*
 257                                  * If the input had a space after the ], keep
 258                                  * it.  We don't bother with finding the end of
 259                                  * the space, since we later normalize it
 260                                  * anyway.
 261                                  */
 262                                 if (isspace(subject->buf[at]))
 263                                         at += 1;
 264                         }
 265                         continue;
 266                 }
 267                 break;
 268         }
 269         strbuf_trim(subject);
 270 }
 271
 272 static void cleanup_space(struct strbuf *sb)
 273 {
 274         size_t pos, cnt;
 275         for (pos = 0; pos < sb->len; pos++) {
 276                 if (isspace(sb->buf[pos])) {
 277                         sb->buf[pos] = ' ';
 278                         for (cnt = 0; isspace(sb->buf[pos + cnt + 1]); cnt++);
 279                         strbuf_remove(sb, pos + 1, cnt);
 280                 }
 281         }
 282 }
 283
 284 static void decode_header(struct strbuf *line);
 285 static const char *header[MAX_HDR_PARSED] = {
 286         "From","Subject","Date",
 287 };
 288
 289 static inline int cmp_header(const struct strbuf *line, const char *hdr)
 290 {
 291         int len = strlen(hdr);
 292         return !strncasecmp(line->buf, hdr, len) && line->len > len &&
 293                         line->buf[len] == ':' && isspace(line->buf[len + 1]);
 294 }
 295
 296 static int check_header(const struct strbuf *line,
 297                                 struct strbuf *hdr_data[], int overwrite)
 298 {
 299         int i, ret = 0, len;
 300         struct strbuf sb = STRBUF_INIT;
 301         /* search for the interesting parts */
 302         for (i = 0; header[i]; i++) {
 303                 int len = strlen(header[i]);
 304                 if ((!hdr_data[i] || overwrite) && cmp_header(line, header[i])) {
 305                         /* Unwrap inline B and Q encoding, and optionally
 306                          * normalize the meta information to utf8.
 307                          */
 308                         strbuf_add(&sb, line->buf + len + 2, line->len - len - 2);
 309                         decode_header(&sb);
 310                         handle_header(&hdr_data[i], &sb);
 311                         ret = 1;
 312                         goto check_header_out;
 313                 }
 314         }
 315
 316         /* Content stuff */
 317         if (cmp_header(line, "Content-Type")) {
 318                 len = strlen("Content-Type: ");
 319                 strbuf_add(&sb, line->buf + len, line->len - len);
 320                 decode_header(&sb);
 321                 strbuf_insert(&sb, 0, "Content-Type: ", len);
 322                 handle_content_type(&sb);
 323                 ret = 1;
 324                 goto check_header_out;
 325         }
 326         if (cmp_header(line, "Content-Transfer-Encoding")) {
 327                 len = strlen("Content-Transfer-Encoding: ");
 328                 strbuf_add(&sb, line->buf + len, line->len - len);
 329                 decode_header(&sb);
 330                 handle_content_transfer_encoding(&sb);
 331                 ret = 1;
 332                 goto check_header_out;
 333         }
 334
 335         /* for inbody stuff */
 336         if (!prefixcmp(line->buf, ">From") && isspace(line->buf[5])) {
 337                 ret = 1; /* Should this return 0? */
 338                 goto check_header_out;
 339         }
 340         if (!prefixcmp(line->buf, "[PATCH]") && isspace(line->buf[7])) {
 341                 for (i = 0; header[i]; i++) {
 342                         if (!memcmp("Subject", header[i], 7)) {
 343                                 handle_header(&hdr_data[i], line);
 344                                 ret = 1;
 345                                 goto check_header_out;
 346                         }
 347                 }
 348         }
 349
 350 check_header_out:
 351         strbuf_release(&sb);
 352         return ret;
 353 }
 354
 355 static int is_rfc2822_header(const struct strbuf *line)
 356 {
 357         /*
 358          * The section that defines the loosest possible
 359          * field name is "3.6.8 Optional fields".
 360          *
 361          * optional-field = field-name ":" unstructured CRLF
 362          * field-name = 1*ftext
 363          * ftext = %d33-57 / %59-126
 364          */
 365         int ch;
 366         char *cp = line->buf;
 367
 368         /* Count mbox From headers as headers */
 369         if (!prefixcmp(cp, "From ") || !prefixcmp(cp, ">From "))
 370                 return 1;
 371
 372         while ((ch = *cp++)) {
 373                 if (ch == ':')
 374                         return 1;
 375                 if ((33 <= ch && ch <= 57) ||
 376                     (59 <= ch && ch <= 126))
 377                         continue;
 378                 break;
 379         }
 380         return 0;
 381 }
 382
 383 static int read_one_header_line(struct strbuf *line, FILE *in)
 384 {
 385         /* Get the first part of the line. */
 386         if (strbuf_getline(line, in, '\n'))
 387                 return 0;
 388
 389         /*
 390          * Is it an empty line or not a valid rfc2822 header?
 391          * If so, stop here, and return false ("not a header")
 392          */
 393         strbuf_rtrim(line);
 394         if (!line->len || !is_rfc2822_header(line)) {
 395                 /* Re-add the newline */
 396                 strbuf_addch(line, '\n');
 397                 return 0;
 398         }
 399
 400         /*
 401          * Now we need to eat all the continuation lines..
 402          * Yuck, 2822 header "folding"
 403          */
 404         for (;;) {
 405                 int peek;
 406                 struct strbuf continuation = STRBUF_INIT;
 407
 408                 peek = fgetc(in); ungetc(peek, in);
 409                 if (peek != ' ' && peek != '\t')
 410                         break;
 411                 if (strbuf_getline(&continuation, in, '\n'))
 412                         break;
 413                 continuation.buf[0] = ' ';
 414                 strbuf_rtrim(&continuation);
 415                 strbuf_addbuf(line, &continuation);
 416         }
 417
 418         return 1;
 419 }
 420
 421 static struct strbuf *decode_q_segment(const struct strbuf *q_seg, int rfc2047)
 422 {
 423         const char *in = q_seg->buf;
 424         int c;
 425         struct strbuf *out = xmalloc(sizeof(struct strbuf));
 426         strbuf_init(out, q_seg->len);
 427
 428         while ((c = *in++) != 0) {
 429                 if (c == '=') {
 430                         int d = *in++;
 431                         if (d == '\n' || !d)
 432                                 break; /* drop trailing newline */
 433                         strbuf_addch(out, (hexval(d) << 4) | hexval(*in++));
 434                         continue;
 435                 }
 436                 if (rfc2047 && c == '_') /* rfc2047 4.2 (2) */
 437                         c = 0x20;
 438                 strbuf_addch(out, c);
 439         }
 440         return out;
 441 }
 442
 443 static struct strbuf *decode_b_segment(const struct strbuf *b_seg)
 444 {
 445         /* Decode in..ep, possibly in-place to ot */
 446         int c, pos = 0, acc = 0;
 447         const char *in = b_seg->buf;
 448         struct strbuf *out = xmalloc(sizeof(struct strbuf));
 449         strbuf_init(out, b_seg->len);
 450
 451         while ((c = *in++) != 0) {
 452                 if (c == '+')
 453                         c = 62;
 454                 else if (c == '/')
 455                         c = 63;
 456                 else if ('A' <= c && c <= 'Z')
 457                         c -= 'A';
 458                 else if ('a' <= c && c <= 'z')
 459                         c -= 'a' - 26;
 460                 else if ('0' <= c && c <= '9')
 461                         c -= '0' - 52;
 462                 else
 463                         continue; /* garbage */
 464                 switch (pos++) {
 465                 case 0:
 466                         acc = (c << 2);
 467                         break;
 468                 case 1:
 469                         strbuf_addch(out, (acc | (c >> 4)));
 470                         acc = (c & 15) << 4;
 471                         break;
 472                 case 2:
 473                         strbuf_addch(out, (acc | (c >> 2)));
 474                         acc = (c & 3) << 6;
 475                         break;
 476                 case 3:
 477                         strbuf_addch(out, (acc | c));
 478                         acc = pos = 0;
 479                         break;
 480                 }
 481         }
 482         return out;
 483 }
 484
 485 /*
 486  * When there is no known charset, guess.
 487  *
 488  * Right now we assume that if the target is UTF-8 (the default),
 489  * and it already looks like UTF-8 (which includes US-ASCII as its
 490  * subset, of course) then that is what it is and there is nothing
 491  * to do.
 492  *
 493  * Otherwise, we default to assuming it is Latin1 for historical
 494  * reasons.
 495  */
 496 static const char *guess_charset(const struct strbuf *line, const char *target_charset)
 497 {
 498         if (is_encoding_utf8(target_charset)) {
 499                 if (is_utf8(line->buf))
 500                         return NULL;
 501         }
 502         return "ISO8859-1";
 503 }
 504
 505 static void convert_to_utf8(struct strbuf *line, const char *charset)
 506 {
 507         char *out;
 508
 509         if (!charset || !*charset) {
 510                 charset = guess_charset(line, metainfo_charset);
 511                 if (!charset)
 512                         return;
 513         }
 514
 515         if (!strcasecmp(metainfo_charset, charset))
 516                 return;
 517         out = reencode_string(line->buf, metainfo_charset, charset);
 518         if (!out)
 519                 die("cannot convert from %s to %s",
 520                     charset, metainfo_charset);
 521         strbuf_attach(line, out, strlen(out), strlen(out));
 522 }
 523
 524 static int decode_header_bq(struct strbuf *it)
 525 {
 526         char *in, *ep, *cp;
 527         struct strbuf outbuf = STRBUF_INIT, *dec;
 528         struct strbuf charset_q = STRBUF_INIT, piecebuf = STRBUF_INIT;
 529         int rfc2047 = 0;
 530
 531         in = it->buf;
 532         while (in - it->buf <= it->len && (ep = strstr(in, "=?")) != NULL) {
 533                 int encoding;
 534                 strbuf_reset(&charset_q);
 535                 strbuf_reset(&piecebuf);
 536                 rfc2047 = 1;
 537
 538                 if (in != ep) {
 539                         /*
 540                          * We are about to process an encoded-word
 541                          * that begins at ep, but there is something
 542                          * before the encoded word.
 543                          */
 544                         char *scan;
 545                         for (scan = in; scan < ep; scan++)
 546                                 if (!isspace(*scan))
 547                                         break;
 548
 549                         if (scan != ep || in == it->buf) {
 550                                 /*
 551                                  * We should not lose that "something",
 552                                  * unless we have just processed an
 553                                  * encoded-word, and there is only LWS
 554                                  * before the one we are about to process.
 555                                  */
 556                                 strbuf_add(&outbuf, in, ep - in);
 557                         }
 558                 }
 559                 /* E.g.
 560                  * ep : "=?iso-2022-jp?B?GyR...?= foo"
 561                  * ep : "=?ISO-8859-1?Q?Foo=FCbar?= baz"
 562                  */
 563                 ep += 2;
 564
 565                 if (ep - it->buf >= it->len || !(cp = strchr(ep, '?')))
 566                         goto decode_header_bq_out;
 567
 568                 if (cp + 3 - it->buf > it->len)
 569                         goto decode_header_bq_out;
 570                 strbuf_add(&charset_q, ep, cp - ep);
 571
 572                 encoding = cp[1];
 573                 if (!encoding || cp[2] != '?')
 574                         goto decode_header_bq_out;
 575                 ep = strstr(cp + 3, "?=");
 576                 if (!ep)
 577                         goto decode_header_bq_out;
 578                 strbuf_add(&piecebuf, cp + 3, ep - cp - 3);
 579                 switch (tolower(encoding)) {
 580                 default:
 581                         goto decode_header_bq_out;
 582                 case 'b':
 583                         dec = decode_b_segment(&piecebuf);
 584                         break;
 585                 case 'q':
 586                         dec = decode_q_segment(&piecebuf, 1);
 587                         break;
 588                 }
 589                 if (metainfo_charset)
 590                         convert_to_utf8(dec, charset_q.buf);
 591
 592                 strbuf_addbuf(&outbuf, dec);
 593                 strbuf_release(dec);
 594                 free(dec);
 595                 in = ep + 2;
 596         }
 597         strbuf_addstr(&outbuf, in);
 598         strbuf_reset(it);
 599         strbuf_addbuf(it, &outbuf);
 600 decode_header_bq_out:
 601         strbuf_release(&outbuf);
 602         strbuf_release(&charset_q);
 603         strbuf_release(&piecebuf);
 604         return rfc2047;
 605 }
 606
 607 static void decode_header(struct strbuf *it)
 608 {
 609         if (decode_header_bq(it))
 610                 return;
 611         /* otherwise "it" is a straight copy of the input.
 612          * This can be binary guck but there is no charset specified.
 613          */
 614         if (metainfo_charset)
 615                 convert_to_utf8(it, "");
 616 }
 617
 618 static void decode_transfer_encoding(struct strbuf *line)
 619 {
 620         struct strbuf *ret;
 621
 622         switch (transfer_encoding) {
 623         case TE_QP:
 624                 ret = decode_q_segment(line, 0);
 625                 break;
 626         case TE_BASE64:
 627                 ret = decode_b_segment(line);
 628                 break;
 629         case TE_DONTCARE:
 630         default:
 631                 return;
 632         }
 633         strbuf_reset(line);
 634         strbuf_addbuf(line, ret);
 635         strbuf_release(ret);
 636         free(ret);
 637 }
 638
 639 static void handle_filter(struct strbuf *line);
 640
 641 static int find_boundary(void)
 642 {
 643         while (!strbuf_getline(&line, fin, '\n')) {
 644                 if (*content_top && is_multipart_boundary(&line))
 645                         return 1;
 646         }
 647         return 0;
 648 }
 649
 650 static int handle_boundary(void)
 651 {
 652         struct strbuf newline = STRBUF_INIT;
 653
 654         strbuf_addch(&newline, '\n');
 655 again:
 656         if (line.len >= (*content_top)->len + 2 &&
 657             !memcmp(line.buf + (*content_top)->len, "--", 2)) {
 658                 /* we hit an end boundary */
 659                 /* pop the current boundary off the stack */
 660                 strbuf_release(*content_top);
 661                 free(*content_top);
 662                 *content_top = NULL;
 663
 664                 /* technically won't happen as is_multipart_boundary()
 665                    will fail first.  But just in case..
 666                  */
 667                 if (--content_top < content) {
 668                         fprintf(stderr, "Detected mismatched boundaries, "
 669                                         "can't recover\n");
 670                         exit(1);
 671                 }
 672                 handle_filter(&newline);
 673                 strbuf_release(&newline);
 674
 675                 /* skip to the next boundary */
 676                 if (!find_boundary())
 677                         return 0;
 678                 goto again;
 679         }
 680
 681         /* set some defaults */
 682         transfer_encoding = TE_DONTCARE;
 683         strbuf_reset(&charset);
 684         message_type = TYPE_TEXT;
 685
 686         /* slurp in this section's info */
 687         while (read_one_header_line(&line, fin))
 688                 check_header(&line, p_hdr_data, 0);
 689
 690         strbuf_release(&newline);
 691         /* replenish line */
 692         if (strbuf_getline(&line, fin, '\n'))
 693                 return 0;
 694         strbuf_addch(&line, '\n');
 695         return 1;
 696 }
 697
 698 static inline int patchbreak(const struct strbuf *line)
 699 {
 700         size_t i;
 701
 702         /* Beginning of a "diff -" header? */
 703         if (!prefixcmp(line->buf, "diff -"))
 704                 return 1;
 705
 706         /* CVS "Index: " line? */
 707         if (!prefixcmp(line->buf, "Index: "))
 708                 return 1;
 709
 710         /*
 711          * "--- <filename>" starts patches without headers
 712          * "---<sp>*" is a manual separator
 713          */
 714         if (line->len < 4)
 715                 return 0;
 716
 717         if (!prefixcmp(line->buf, "---")) {
 718                 /* space followed by a filename? */
 719                 if (line->buf[3] == ' ' && !isspace(line->buf[4]))
 720                         return 1;
 721                 /* Just whitespace? */
 722                 for (i = 3; i < line->len; i++) {
 723                         unsigned char c = line->buf[i];
 724                         if (c == '\n')
 725                                 return 1;
 726                         if (!isspace(c))
 727                                 break;
 728                 }
 729                 return 0;
 730         }
 731         return 0;
 732 }
 733
 734 static int is_scissors_line(const struct strbuf *line)
 735 {
 736         size_t i, len = line->len;
 737         int scissors = 0, gap = 0;
 738         int first_nonblank = -1;
 739         int last_nonblank = 0, visible, perforation = 0, in_perforation = 0;
 740         const char *buf = line->buf;
 741
 742         for (i = 0; i < len; i++) {
 743                 if (isspace(buf[i])) {
 744                         if (in_perforation) {
 745                                 perforation++;
 746                                 gap++;
 747                         }
 748                         continue;
 749                 }
 750                 last_nonblank = i;
 751                 if (first_nonblank < 0)
 752                         first_nonblank = i;
 753                 if (buf[i] == '-') {
 754                         in_perforation = 1;
 755                         perforation++;
 756                         continue;
 757                 }
 758                 if (i + 1 < len &&
 759                     (!memcmp(buf + i, ">8", 2) || !memcmp(buf + i, "8<", 2) ||
 760                      !memcmp(buf + i, ">%", 2) || !memcmp(buf + i, "%<", 2))) {
 761                         in_perforation = 1;
 762                         perforation += 2;
 763                         scissors += 2;
 764                         i++;
 765                         continue;
 766                 }
 767                 in_perforation = 0;
 768         }
 769
 770         /*
 771          * The mark must be at least 8 bytes long (e.g. "-- >8 --").
 772          * Even though there can be arbitrary cruft on the same line
 773          * (e.g. "cut here"), in order to avoid misidentification, the
 774          * perforation must occupy more than a third of the visible
 775          * width of the line, and dashes and scissors must occupy more
 776          * than half of the perforation.
 777          */
 778
 779         visible = last_nonblank - first_nonblank + 1;
 780         return (scissors && 8 <= visible &&
 781                 visible < perforation * 3 &&
 782                 gap * 2 < perforation);
 783 }
 784
 785 static int handle_commit_msg(struct strbuf *line)
 786 {
 787         static int still_looking = 1;
 788
 789         if (!cmitmsg)
 790                 return 0;
 791
 792         if (still_looking) {
 793                 if (!line->len || (line->len == 1 && line->buf[0] == '\n'))
 794                         return 0;
 795         }
 796
 797         if (use_inbody_headers && still_looking) {
 798                 still_looking = check_header(line, s_hdr_data, 0);
 799                 if (still_looking)
 800                         return 0;
 801         } else
 802                 /* Only trim the first (blank) line of the commit message
 803                  * when ignoring in-body headers.
 804                  */
 805                 still_looking = 0;
 806
 807         /* normalize the log message to UTF-8. */
 808         if (metainfo_charset)
 809                 convert_to_utf8(line, charset.buf);
 810
 811         if (use_scissors && is_scissors_line(line)) {
 812                 int i;
 813                 if (fseek(cmitmsg, 0L, SEEK_SET))
 814                         die_errno("Could not rewind output message file");
 815                 if (ftruncate(fileno(cmitmsg), 0))
 816                         die_errno("Could not truncate output message file at scissors");
 817                 still_looking = 1;
 818
 819                 /*
 820                  * We may have already read "secondary headers"; purge
 821                  * them to give ourselves a clean restart.
 822                  */
 823                 for (i = 0; header[i]; i++) {
 824                         if (s_hdr_data[i])
 825                                 strbuf_release(s_hdr_data[i]);
 826                         s_hdr_data[i] = NULL;
 827                 }
 828                 return 0;
 829         }
 830
 831         if (patchbreak(line)) {
 832                 fclose(cmitmsg);
 833                 cmitmsg = NULL;
 834                 return 1;
 835         }
 836
 837         fputs(line->buf, cmitmsg);
 838         return 0;
 839 }
 840
 841 static void handle_patch(const struct strbuf *line)
 842 {
 843         fwrite(line->buf, 1, line->len, patchfile);
 844         patch_lines++;
 845 }
 846
 847 static void handle_filter(struct strbuf *line)
 848 {
 849         static int filter = 0;
 850
 851         /* filter tells us which part we left off on */
 852         switch (filter) {
 853         case 0:
 854                 if (!handle_commit_msg(line))
 855                         break;
 856                 filter++;
 857         case 1:
 858                 handle_patch(line);
 859                 break;
 860         }
 861 }
 862
 863 static void handle_body(void)
 864 {
 865         struct strbuf prev = STRBUF_INIT;
 866
 867         /* Skip up to the first boundary */
 868         if (*content_top) {
 869                 if (!find_boundary())
 870                         goto handle_body_out;
 871         }
 872
 873         do {
 874                 /* process any boundary lines */
 875                 if (*content_top && is_multipart_boundary(&line)) {
 876                         /* flush any leftover */
 877                         if (prev.len) {
 878                                 handle_filter(&prev);
 879                                 strbuf_reset(&prev);
 880                         }
 881                         if (!handle_boundary())
 882                                 goto handle_body_out;
 883                 }
 884
 885                 /* Unwrap transfer encoding */
 886                 decode_transfer_encoding(&line);
 887
 888                 switch (transfer_encoding) {
 889                 case TE_BASE64:
 890                 case TE_QP:
 891                 {
 892                         struct strbuf **lines, **it, *sb;
 893
 894                         /* Prepend any previous partial lines */
 895                         strbuf_insert(&line, 0, prev.buf, prev.len);
 896                         strbuf_reset(&prev);
 897
 898                         /* binary data most likely doesn't have newlines */
 899                         if (message_type != TYPE_TEXT) {
 900                                 handle_filter(&line);
 901                                 break;
 902                         }
 903                         /*
 904                          * This is a decoded line that may contain
 905                          * multiple new lines.  Pass only one chunk
 906                          * at a time to handle_filter()
 907                          */
 908                         lines = strbuf_split(&line, '\n');
 909                         for (it = lines; (sb = *it); it++) {
 910                                 if (*(it + 1) == NULL) /* The last line */
 911                                         if (sb->buf[sb->len - 1] != '\n') {
 912                                                 /* Partial line, save it for later. */
 913                                                 strbuf_addbuf(&prev, sb);
 914                                                 break;
 915                                         }
 916                                 handle_filter(sb);
 917                         }
 918                         /*
 919                          * The partial chunk is saved in "prev" and will be
 920                          * appended by the next iteration of read_line_with_nul().
 921                          */
 922                         strbuf_list_free(lines);
 923                         break;
 924                 }
 925                 default:
 926                         handle_filter(&line);
 927                 }
 928
 929         } while (!strbuf_getwholeline(&line, fin, '\n'));
 930
 931 handle_body_out:
 932         strbuf_release(&prev);
 933 }
 934
 935 static void output_header_lines(FILE *fout, const char *hdr, const struct strbuf *data)
 936 {
 937         const char *sp = data->buf;
 938         while (1) {
 939                 char *ep = strchr(sp, '\n');
 940                 int len;
 941                 if (!ep)
 942                         len = strlen(sp);
 943                 else
 944                         len = ep - sp;
 945                 fprintf(fout, "%s: %.*s\n", hdr, len, sp);
 946                 if (!ep)
 947                         break;
 948                 sp = ep + 1;
 949         }
 950 }
 951
 952 static void handle_info(void)
 953 {
 954         struct strbuf *hdr;
 955         int i;
 956
 957         for (i = 0; header[i]; i++) {
 958                 /* only print inbody headers if we output a patch file */
 959                 if (patch_lines && s_hdr_data[i])
 960                         hdr = s_hdr_data[i];
 961                 else if (p_hdr_data[i])
 962                         hdr = p_hdr_data[i];
 963                 else
 964                         continue;
 965
 966                 if (!memcmp(header[i], "Subject", 7)) {
 967                         if (!keep_subject) {
 968                                 cleanup_subject(hdr);
 969                                 cleanup_space(hdr);
 970                         }
 971                         output_header_lines(fout, "Subject", hdr);
 972                 } else if (!memcmp(header[i], "From", 4)) {
 973                         cleanup_space(hdr);
 974                         handle_from(hdr);
 975                         fprintf(fout, "Author: %s\n", name.buf);
 976                         fprintf(fout, "Email: %s\n", email.buf);
 977                 } else {
 978                         cleanup_space(hdr);
 979                         fprintf(fout, "%s: %s\n", header[i], hdr->buf);
 980                 }
 981         }
 982         fprintf(fout, "\n");
 983 }
 984
 985 static int mailinfo(FILE *in, FILE *out, const char *msg, const char *patch)
 986 {
 987         int peek;
 988         fin = in;
 989         fout = out;
 990
 991         cmitmsg = fopen(msg, "w");
 992         if (!cmitmsg) {
 993                 perror(msg);
 994                 return -1;
 995         }
 996         patchfile = fopen(patch, "w");
 997         if (!patchfile) {
 998                 perror(patch);
 999                 fclose(cmitmsg);
1000                 return -1;
1001         }
1002
1003         p_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*p_hdr_data));
1004         s_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*s_hdr_data));
1005
1006         do {
1007                 peek = fgetc(in);
1008         } while (isspace(peek));
1009         ungetc(peek, in);
1010
1011         /* process the email header */
1012         while (read_one_header_line(&line, fin))
1013                 check_header(&line, p_hdr_data, 1);
1014
1015         handle_body();
1016         handle_info();
1017
1018         return 0;
1019 }
1020
1021 static int git_mailinfo_config(const char *var, const char *value, void *unused)
1022 {
1023         if (prefixcmp(var, "mailinfo."))
1024                 return git_default_config(var, value, unused);
1025         if (!strcmp(var, "mailinfo.scissors")) {
1026                 use_scissors = git_config_bool(var, value);
1027                 return 0;
1028         }
1029         /* perhaps others here */
1030         return 0;
1031 }
1032
1033 static const char mailinfo_usage[] =
1034         "git mailinfo [-k|-b] [-u | --encoding=<encoding> | -n] [--scissors | --no-scissors] msg patch < mail >info";
1035
1036 int cmd_mailinfo(int argc, const char **argv, const char *prefix)
1037 {
1038         const char *def_charset;
1039
1040         /* NEEDSWORK: might want to do the optional .git/ directory
1041          * discovery
1042          */
1043         git_config(git_mailinfo_config, NULL);
1044
1045         def_charset = get_commit_output_encoding();
1046         metainfo_charset = def_charset;
1047
1048         while (1 < argc && argv[1][0] == '-') {
1049                 if (!strcmp(argv[1], "-k"))
1050                         keep_subject = 1;
1051                 else if (!strcmp(argv[1], "-b"))
1052                         keep_non_patch_brackets_in_subject = 1;
1053                 else if (!strcmp(argv[1], "-u"))
1054                         metainfo_charset = def_charset;
1055                 else if (!strcmp(argv[1], "-n"))
1056                         metainfo_charset = NULL;
1057                 else if (!prefixcmp(argv[1], "--encoding="))
1058                         metainfo_charset = argv[1] + 11;
1059                 else if (!strcmp(argv[1], "--scissors"))
1060                         use_scissors = 1;
1061                 else if (!strcmp(argv[1], "--no-scissors"))
1062                         use_scissors = 0;
1063                 else if (!strcmp(argv[1], "--no-inbody-headers"))
1064                         use_inbody_headers = 0;
1065                 else
1066                         usage(mailinfo_usage);
1067                 argc--; argv++;
1068         }
1069
1070         if (argc != 3)
1071                 usage(mailinfo_usage);
1072
1073         return !!mailinfo(stdin, stdout, argv[1], argv[2]);
1074 }